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

   1 /*
   2  * Pixel and vertex shaders implementation using ARB_vertex_program
   3  * and ARB_fragment_program GL extensions.
   4  *
   5  * Copyright 2002-2003 Jason Edmeades
   6  * Copyright 2002-2003 Raphael Junqueira
   7  * Copyright 2004 Christian Costa
   8  * Copyright 2005 Oliver Stieber
   9  * Copyright 2006 Ivan Gyurdiev
  10  * Copyright 2006 Jason Green
  11  * Copyright 2006 Henri Verbeet
  12  * Copyright 2007-2008 Stefan Dösinger for CodeWeavers
  13  * Copyright 2009 Henri Verbeet for CodeWeavers
  14  *
  15  * This library is free software; you can redistribute it and/or
  16  * modify it under the terms of the GNU Lesser General Public
  17  * License as published by the Free Software Foundation; either
  18  * version 2.1 of the License, or (at your option) any later version.
  19  *
  20  * This library is distributed in the hope that it will be useful,
  21  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  22  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  23  * Lesser General Public License for more details.
  24  *
  25  * You should have received a copy of the GNU Lesser General Public
  26  * License along with this library; if not, write to the Free Software
  27  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301, USA
  28  */
  29
  30 #include "config.h"
  31 #include "wine/port.h"
  32
  33 #include <math.h>
  34 #include <stdio.h>
  35
  36 #include "wined3d_private.h"
  37
  38 WINE_DEFAULT_DEBUG_CHANNEL(d3d_shader);
  39 WINE_DECLARE_DEBUG_CHANNEL(d3d_constants);
  40 WINE_DECLARE_DEBUG_CHANNEL(d3d);
  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     BOOL swizzle_coord = FALSE;
1918
1919     /* All versions have a destination register */
1920     shader_arb_get_dst_param(ins, dst, reg_dest);
1921
1922     /* 1.0-1.4: Use destination register number as texture code.
1923        2.0+: Use provided sampler number as texure code. */
1924     if (shader_version < WINED3D_SHADER_VERSION(2,0))
1925         reg_sampler_code = dst->reg.idx;
1926     else
1927         reg_sampler_code = ins->src[1].reg.idx;
1928
1929     /* 1.0-1.3: Use the texcoord varying.
1930        1.4+: Use provided coordinate source register. */
1931     if (shader_version < WINED3D_SHADER_VERSION(1,4))
1932         sprintf(reg_coord, "fragment.texcoord[%u]", reg_sampler_code);
1933     else {
1934         /* TEX is the only instruction that can handle DW and DZ natively */
1935         src = ins->src[0];
1936         if(src.modifiers == WINED3DSPSM_DW) src.modifiers = WINED3DSPSM_NONE;
1937         if(src.modifiers == WINED3DSPSM_DZ) src.modifiers = WINED3DSPSM_NONE;
1938         shader_arb_get_src_param(ins, &src, 0, reg_coord);
1939     }
1940
1941     /* projection flag:
1942      * 1.1, 1.2, 1.3: Use WINED3D_TSS_TEXTURETRANSFORMFLAGS
1943      * 1.4: Use WINED3DSPSM_DZ or WINED3DSPSM_DW on src[0]
1944      * 2.0+: Use WINED3DSI_TEXLD_PROJECT on the opcode
1945      */
1946     if (shader_version < WINED3D_SHADER_VERSION(1,4))
1947     {
1948         DWORD flags = 0;
1949         if (reg_sampler_code < MAX_TEXTURES)
1950             flags = priv->cur_ps_args->super.tex_transform >> reg_sampler_code * WINED3D_PSARGS_TEXTRANSFORM_SHIFT;
1951         if (flags & WINED3D_PSARGS_PROJECTED)
1952         {
1953             myflags |= TEX_PROJ;
1954             if ((flags & ~WINED3D_PSARGS_PROJECTED) == WINED3D_TTFF_COUNT3)
1955                 swizzle_coord = TRUE;
1956         }
1957     }
1958     else if (shader_version < WINED3D_SHADER_VERSION(2,0))
1959     {
1960         enum wined3d_shader_src_modifier src_mod = ins->src[0].modifiers;
1961         if (src_mod == WINED3DSPSM_DZ)
1962         {
1963             swizzle_coord = TRUE;
1964             myflags |= TEX_PROJ;
1965         } else if(src_mod == WINED3DSPSM_DW) {
1966             myflags |= TEX_PROJ;
1967         }
1968     } else {
1969         if (ins->flags & WINED3DSI_TEXLD_PROJECT) myflags |= TEX_PROJ;
1970         if (ins->flags & WINED3DSI_TEXLD_BIAS) myflags |= TEX_BIAS;
1971     }
1972
1973     if (swizzle_coord)
1974     {
1975         /* TXP cannot handle DZ natively, so move the z coordinate to .w.
1976          * reg_coord is a read-only varying register, so we need a temp reg */
1977         shader_addline(ins->ctx->buffer, "SWZ TA, %s, x, y, z, z;\n", reg_coord);
1978         strcpy(reg_coord, "TA");
1979     }
1980
1981     shader_hw_sample(ins, reg_sampler_code, reg_dest, reg_coord, myflags, NULL, NULL);
1982 }
1983
1984 static void pshader_hw_texcoord(const struct wined3d_shader_instruction *ins)
1985 {
1986     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1987     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1988     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
1989             ins->ctx->reg_maps->shader_version.minor);
1990     char dst_str[50];
1991
1992     if (shader_version < WINED3D_SHADER_VERSION(1,4))
1993     {
1994         DWORD reg = dst->reg.idx;
1995
1996         shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
1997         shader_addline(buffer, "MOV_SAT %s, fragment.texcoord[%u];\n", dst_str, reg);
1998     } else {
1999         char reg_src[40];
2000
2001         shader_arb_get_src_param(ins, &ins->src[0], 0, reg_src);
2002         shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2003         shader_addline(buffer, "MOV %s, %s;\n", dst_str, reg_src);
2004    }
2005 }
2006
2007 static void pshader_hw_texreg2ar(const struct wined3d_shader_instruction *ins)
2008 {
2009      struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2010      DWORD flags = 0;
2011
2012      DWORD reg1 = ins->dst[0].reg.idx;
2013      char dst_str[50];
2014      char src_str[50];
2015
2016      /* Note that texreg2ar treats Tx as a temporary register, not as a varying */
2017      shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2018      shader_arb_get_src_param(ins, &ins->src[0], 0, src_str);
2019      /* Move .x first in case src_str is "TA" */
2020      shader_addline(buffer, "MOV TA.y, %s.x;\n", src_str);
2021      shader_addline(buffer, "MOV TA.x, %s.w;\n", src_str);
2022      if (reg1 < MAX_TEXTURES)
2023      {
2024          struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2025          flags = priv->cur_ps_args->super.tex_transform >> reg1 * WINED3D_PSARGS_TEXTRANSFORM_SHIFT;
2026      }
2027      shader_hw_sample(ins, reg1, dst_str, "TA", flags & WINED3D_PSARGS_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2028 }
2029
2030 static void pshader_hw_texreg2gb(const struct wined3d_shader_instruction *ins)
2031 {
2032      struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2033
2034      DWORD reg1 = ins->dst[0].reg.idx;
2035      char dst_str[50];
2036      char src_str[50];
2037
2038      /* Note that texreg2gb treats Tx as a temporary register, not as a varying */
2039      shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2040      shader_arb_get_src_param(ins, &ins->src[0], 0, src_str);
2041      shader_addline(buffer, "MOV TA.x, %s.y;\n", src_str);
2042      shader_addline(buffer, "MOV TA.y, %s.z;\n", src_str);
2043      shader_hw_sample(ins, reg1, dst_str, "TA", 0, NULL, NULL);
2044 }
2045
2046 static void pshader_hw_texreg2rgb(const struct wined3d_shader_instruction *ins)
2047 {
2048     DWORD reg1 = ins->dst[0].reg.idx;
2049     char dst_str[50];
2050     char src_str[50];
2051
2052     /* Note that texreg2rg treats Tx as a temporary register, not as a varying */
2053     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2054     shader_arb_get_src_param(ins, &ins->src[0], 0, src_str);
2055     shader_hw_sample(ins, reg1, dst_str, src_str, 0, NULL, NULL);
2056 }
2057
2058 static void pshader_hw_texbem(const struct wined3d_shader_instruction *ins)
2059 {
2060     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2061     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2062     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2063     char reg_coord[40], dst_reg[50], src_reg[50];
2064     DWORD reg_dest_code;
2065
2066     /* All versions have a destination register. The Tx where the texture coordinates come
2067      * from is the varying incarnation of the texture register
2068      */
2069     reg_dest_code = dst->reg.idx;
2070     shader_arb_get_dst_param(ins, &ins->dst[0], dst_reg);
2071     shader_arb_get_src_param(ins, &ins->src[0], 0, src_reg);
2072     sprintf(reg_coord, "fragment.texcoord[%u]", reg_dest_code);
2073
2074     /* Sampling the perturbation map in Tsrc was done already, including the signedness correction if needed
2075      * The Tx in which the perturbation map is stored is the tempreg incarnation of the texture register
2076      *
2077      * GL_NV_fragment_program_option could handle this in one instruction via X2D:
2078      * X2D TA.xy, fragment.texcoord, T%u, bumpenvmat%u.xzyw
2079      *
2080      * However, the NV extensions are never enabled for <= 2.0 shaders because of the performance penalty that
2081      * comes with it, and texbem is an 1.x only instruction. No 1.x instruction forces us to enable the NV
2082      * extension.
2083      */
2084     shader_addline(buffer, "SWZ TB, bumpenvmat%d, x, z, 0, 0;\n", reg_dest_code);
2085     shader_addline(buffer, "DP3 TA.x, TB, %s;\n", src_reg);
2086     shader_addline(buffer, "SWZ TB, bumpenvmat%d, y, w, 0, 0;\n", reg_dest_code);
2087     shader_addline(buffer, "DP3 TA.y, TB, %s;\n", src_reg);
2088
2089     /* with projective textures, texbem only divides the static texture coord, not the displacement,
2090      * so we can't let the GL handle this.
2091      */
2092     if ((priv->cur_ps_args->super.tex_transform >> reg_dest_code * WINED3D_PSARGS_TEXTRANSFORM_SHIFT)
2093             & WINED3D_PSARGS_PROJECTED)
2094     {
2095         shader_addline(buffer, "RCP TB.w, %s.w;\n", reg_coord);
2096         shader_addline(buffer, "MUL TB.xy, %s, TB.w;\n", reg_coord);
2097         shader_addline(buffer, "ADD TA.xy, TA, TB;\n");
2098     } else {
2099         shader_addline(buffer, "ADD TA.xy, TA, %s;\n", reg_coord);
2100     }
2101
2102     shader_hw_sample(ins, reg_dest_code, dst_reg, "TA", 0, NULL, NULL);
2103
2104     if (ins->handler_idx == WINED3DSIH_TEXBEML)
2105     {
2106         /* No src swizzles are allowed, so this is ok */
2107         shader_addline(buffer, "MAD TA, %s.z, luminance%d.x, luminance%d.y;\n",
2108                        src_reg, reg_dest_code, reg_dest_code);
2109         shader_addline(buffer, "MUL %s, %s, TA;\n", dst_reg, dst_reg);
2110     }
2111 }
2112
2113 static void pshader_hw_texm3x2pad(const struct wined3d_shader_instruction *ins)
2114 {
2115     DWORD reg = ins->dst[0].reg.idx;
2116     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2117     char src0_name[50], dst_name[50];
2118     BOOL is_color;
2119     struct wined3d_shader_register tmp_reg = ins->dst[0].reg;
2120
2121     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2122     /* The next instruction will be a texm3x2tex or texm3x2depth that writes to the uninitialized
2123      * T<reg+1> register. Use this register to store the calculated vector
2124      */
2125     tmp_reg.idx = reg + 1;
2126     shader_arb_get_register_name(ins, &tmp_reg, dst_name, &is_color);
2127     shader_addline(buffer, "DP3 %s.x, fragment.texcoord[%u], %s;\n", dst_name, reg, src0_name);
2128 }
2129
2130 static void pshader_hw_texm3x2tex(const struct wined3d_shader_instruction *ins)
2131 {
2132     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2133     DWORD flags;
2134     DWORD reg = ins->dst[0].reg.idx;
2135     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2136     char dst_str[50];
2137     char src0_name[50];
2138     char dst_reg[50];
2139     BOOL is_color;
2140
2141     /* We know that we're writing to the uninitialized T<reg> register, so use it for temporary storage */
2142     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_reg, &is_color);
2143
2144     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2145     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2146     shader_addline(buffer, "DP3 %s.y, fragment.texcoord[%u], %s;\n", dst_reg, reg, src0_name);
2147     flags = reg < MAX_TEXTURES ? priv->cur_ps_args->super.tex_transform >> reg * WINED3D_PSARGS_TEXTRANSFORM_SHIFT : 0;
2148     shader_hw_sample(ins, reg, dst_str, dst_reg, flags & WINED3D_PSARGS_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2149 }
2150
2151 static void pshader_hw_texm3x3pad(const struct wined3d_shader_instruction *ins)
2152 {
2153     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
2154     DWORD reg = ins->dst[0].reg.idx;
2155     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2156     char src0_name[50], dst_name[50];
2157     struct wined3d_shader_register tmp_reg = ins->dst[0].reg;
2158     BOOL is_color;
2159
2160     /* There are always 2 texm3x3pad instructions followed by one texm3x3[tex,vspec, ...] instruction, with
2161      * incrementing ins->dst[0].register_idx numbers. So the pad instruction already knows the final destination
2162      * register, and this register is uninitialized(otherwise the assembler complains that it is 'redeclared')
2163      */
2164     tmp_reg.idx = reg + 2 - tex_mx->current_row;
2165     shader_arb_get_register_name(ins, &tmp_reg, dst_name, &is_color);
2166
2167     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2168     shader_addline(buffer, "DP3 %s.%c, fragment.texcoord[%u], %s;\n",
2169                    dst_name, 'x' + tex_mx->current_row, reg, src0_name);
2170     tex_mx->texcoord_w[tex_mx->current_row++] = reg;
2171 }
2172
2173 static void pshader_hw_texm3x3tex(const struct wined3d_shader_instruction *ins)
2174 {
2175     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2176     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
2177     DWORD flags;
2178     DWORD reg = ins->dst[0].reg.idx;
2179     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2180     char dst_str[50];
2181     char src0_name[50], dst_name[50];
2182     BOOL is_color;
2183
2184     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2185     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2186     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_name, reg, src0_name);
2187
2188     /* Sample the texture using the calculated coordinates */
2189     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2190     flags = reg < MAX_TEXTURES ? priv->cur_ps_args->super.tex_transform >> reg * WINED3D_PSARGS_TEXTRANSFORM_SHIFT : 0;
2191     shader_hw_sample(ins, reg, dst_str, dst_name, flags & WINED3D_PSARGS_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2192     tex_mx->current_row = 0;
2193 }
2194
2195 static void pshader_hw_texm3x3vspec(const struct wined3d_shader_instruction *ins)
2196 {
2197     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2198     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
2199     DWORD flags;
2200     DWORD reg = ins->dst[0].reg.idx;
2201     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2202     char dst_str[50];
2203     char src0_name[50];
2204     char dst_reg[50];
2205     BOOL is_color;
2206
2207     /* Get the dst reg without writemask strings. We know this register is uninitialized, so we can use all
2208      * components for temporary data storage
2209      */
2210     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_reg, &is_color);
2211     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2212     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_reg, reg, src0_name);
2213
2214     /* Construct the eye-ray vector from w coordinates */
2215     shader_addline(buffer, "MOV TB.x, fragment.texcoord[%u].w;\n", tex_mx->texcoord_w[0]);
2216     shader_addline(buffer, "MOV TB.y, fragment.texcoord[%u].w;\n", tex_mx->texcoord_w[1]);
2217     shader_addline(buffer, "MOV TB.z, fragment.texcoord[%u].w;\n", reg);
2218
2219     /* Calculate reflection vector
2220      */
2221     shader_addline(buffer, "DP3 %s.w, %s, TB;\n", dst_reg, dst_reg);
2222     /* The .w is ignored when sampling, so I can use TB.w to calculate dot(N, N) */
2223     shader_addline(buffer, "DP3 TB.w, %s, %s;\n", dst_reg, dst_reg);
2224     shader_addline(buffer, "RCP TB.w, TB.w;\n");
2225     shader_addline(buffer, "MUL %s.w, %s.w, TB.w;\n", dst_reg, dst_reg);
2226     shader_addline(buffer, "MUL %s, %s.w, %s;\n", dst_reg, dst_reg, dst_reg);
2227     shader_addline(buffer, "MAD %s, coefmul.x, %s, -TB;\n", dst_reg, dst_reg);
2228
2229     /* Sample the texture using the calculated coordinates */
2230     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2231     flags = reg < MAX_TEXTURES ? priv->cur_ps_args->super.tex_transform >> reg * WINED3D_PSARGS_TEXTRANSFORM_SHIFT : 0;
2232     shader_hw_sample(ins, reg, dst_str, dst_reg, flags & WINED3D_PSARGS_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2233     tex_mx->current_row = 0;
2234 }
2235
2236 static void pshader_hw_texm3x3spec(const struct wined3d_shader_instruction *ins)
2237 {
2238     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2239     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
2240     DWORD flags;
2241     DWORD reg = ins->dst[0].reg.idx;
2242     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2243     char dst_str[50];
2244     char src0_name[50];
2245     char src1_name[50];
2246     char dst_reg[50];
2247     BOOL is_color;
2248
2249     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2250     shader_arb_get_src_param(ins, &ins->src[0], 1, src1_name);
2251     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_reg, &is_color);
2252     /* Note: dst_reg.xy is input here, generated by two texm3x3pad instructions */
2253     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_reg, reg, src0_name);
2254
2255     /* Calculate reflection vector.
2256      *
2257      *                   dot(N, E)
2258      * dst_reg.xyz = 2 * --------- * N - E
2259      *                   dot(N, N)
2260      *
2261      * Which normalizes the normal vector
2262      */
2263     shader_addline(buffer, "DP3 %s.w, %s, %s;\n", dst_reg, dst_reg, src1_name);
2264     shader_addline(buffer, "DP3 TC.w, %s, %s;\n", dst_reg, dst_reg);
2265     shader_addline(buffer, "RCP TC.w, TC.w;\n");
2266     shader_addline(buffer, "MUL %s.w, %s.w, TC.w;\n", dst_reg, dst_reg);
2267     shader_addline(buffer, "MUL %s, %s.w, %s;\n", dst_reg, dst_reg, dst_reg);
2268     shader_addline(buffer, "MAD %s, coefmul.x, %s, -%s;\n", dst_reg, dst_reg, src1_name);
2269
2270     /* Sample the texture using the calculated coordinates */
2271     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2272     flags = reg < MAX_TEXTURES ? priv->cur_ps_args->super.tex_transform >> reg * WINED3D_PSARGS_TEXTRANSFORM_SHIFT : 0;
2273     shader_hw_sample(ins, reg, dst_str, dst_reg, flags & WINED3D_PSARGS_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2274     tex_mx->current_row = 0;
2275 }
2276
2277 static void pshader_hw_texdepth(const struct wined3d_shader_instruction *ins)
2278 {
2279     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2280     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2281     char dst_name[50];
2282     const char *zero = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ZERO);
2283     const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
2284
2285     /* texdepth has an implicit destination, the fragment depth value. It's only parameter,
2286      * which is essentially an input, is the destination register because it is the first
2287      * parameter. According to the msdn, this must be register r5, but let's keep it more flexible
2288      * here(writemasks/swizzles are not valid on texdepth)
2289      */
2290     shader_arb_get_dst_param(ins, dst, dst_name);
2291
2292     /* According to the msdn, the source register(must be r5) is unusable after
2293      * the texdepth instruction, so we're free to modify it
2294      */
2295     shader_addline(buffer, "MIN %s.y, %s.y, %s;\n", dst_name, dst_name, one);
2296
2297     /* How to deal with the special case dst_name.g == 0? if r != 0, then
2298      * the r * (1 / 0) will give infinity, which is clamped to 1.0, the correct
2299      * result. But if r = 0.0, then 0 * inf = 0, which is incorrect.
2300      */
2301     shader_addline(buffer, "RCP %s.y, %s.y;\n", dst_name, dst_name);
2302     shader_addline(buffer, "MUL TA.x, %s.x, %s.y;\n", dst_name, dst_name);
2303     shader_addline(buffer, "MIN TA.x, TA.x, %s;\n", one);
2304     shader_addline(buffer, "MAX result.depth, TA.x, %s;\n", zero);
2305 }
2306
2307 /** Process the WINED3DSIO_TEXDP3TEX instruction in ARB:
2308  * Take a 3-component dot product of the TexCoord[dstreg] and src,
2309  * then perform a 1D texture lookup from stage dstregnum, place into dst. */
2310 static void pshader_hw_texdp3tex(const struct wined3d_shader_instruction *ins)
2311 {
2312     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2313     DWORD sampler_idx = ins->dst[0].reg.idx;
2314     char src0[50];
2315     char dst_str[50];
2316
2317     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2318     shader_addline(buffer, "MOV TB, 0.0;\n");
2319     shader_addline(buffer, "DP3 TB.x, fragment.texcoord[%u], %s;\n", sampler_idx, src0);
2320
2321     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2322     shader_hw_sample(ins, sampler_idx, dst_str, "TB", 0 /* Only one coord, can't be projected */, NULL, NULL);
2323 }
2324
2325 /** Process the WINED3DSIO_TEXDP3 instruction in ARB:
2326  * Take a 3-component dot product of the TexCoord[dstreg] and src. */
2327 static void pshader_hw_texdp3(const struct wined3d_shader_instruction *ins)
2328 {
2329     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2330     char src0[50];
2331     char dst_str[50];
2332     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2333
2334     /* Handle output register */
2335     shader_arb_get_dst_param(ins, dst, dst_str);
2336     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2337     shader_addline(buffer, "DP3 %s, fragment.texcoord[%u], %s;\n", dst_str, dst->reg.idx, src0);
2338 }
2339
2340 /** Process the WINED3DSIO_TEXM3X3 instruction in ARB
2341  * Perform the 3rd row of a 3x3 matrix multiply */
2342 static void pshader_hw_texm3x3(const struct wined3d_shader_instruction *ins)
2343 {
2344     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2345     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2346     char dst_str[50], dst_name[50];
2347     char src0[50];
2348     BOOL is_color;
2349
2350     shader_arb_get_dst_param(ins, dst, dst_str);
2351     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2352     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2353     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_name, dst->reg.idx, src0);
2354     shader_addline(buffer, "MOV %s, %s;\n", dst_str, dst_name);
2355 }
2356
2357 /** Process the WINED3DSIO_TEXM3X2DEPTH instruction in ARB:
2358  * Last row of a 3x2 matrix multiply, use the result to calculate the depth:
2359  * Calculate tmp0.y = TexCoord[dstreg] . src.xyz;  (tmp0.x has already been calculated)
2360  * depth = (tmp0.y == 0.0) ? 1.0 : tmp0.x / tmp0.y
2361  */
2362 static void pshader_hw_texm3x2depth(const struct wined3d_shader_instruction *ins)
2363 {
2364     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2365     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2366     char src0[50], dst_name[50];
2367     BOOL is_color;
2368     const char *zero = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ZERO);
2369     const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
2370
2371     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2372     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2373     shader_addline(buffer, "DP3 %s.y, fragment.texcoord[%u], %s;\n", dst_name, dst->reg.idx, src0);
2374
2375     /* How to deal with the special case dst_name.g == 0? if r != 0, then
2376      * the r * (1 / 0) will give infinity, which is clamped to 1.0, the correct
2377      * result. But if r = 0.0, then 0 * inf = 0, which is incorrect.
2378      */
2379     shader_addline(buffer, "RCP %s.y, %s.y;\n", dst_name, dst_name);
2380     shader_addline(buffer, "MUL %s.x, %s.x, %s.y;\n", dst_name, dst_name, dst_name);
2381     shader_addline(buffer, "MIN %s.x, %s.x, %s;\n", dst_name, dst_name, one);
2382     shader_addline(buffer, "MAX result.depth, %s.x, %s;\n", dst_name, zero);
2383 }
2384
2385 /** Handles transforming all WINED3DSIO_M?x? opcodes for
2386     Vertex/Pixel shaders to ARB_vertex_program codes */
2387 static void shader_hw_mnxn(const struct wined3d_shader_instruction *ins)
2388 {
2389     int i;
2390     int nComponents = 0;
2391     struct wined3d_shader_dst_param tmp_dst = {{0}};
2392     struct wined3d_shader_src_param tmp_src[2] = {{{0}}};
2393     struct wined3d_shader_instruction tmp_ins;
2394
2395     memset(&tmp_ins, 0, sizeof(tmp_ins));
2396
2397     /* Set constants for the temporary argument */
2398     tmp_ins.ctx = ins->ctx;
2399     tmp_ins.dst_count = 1;
2400     tmp_ins.dst = &tmp_dst;
2401     tmp_ins.src_count = 2;
2402     tmp_ins.src = tmp_src;
2403
2404     switch(ins->handler_idx)
2405     {
2406         case WINED3DSIH_M4x4:
2407             nComponents = 4;
2408             tmp_ins.handler_idx = WINED3DSIH_DP4;
2409             break;
2410         case WINED3DSIH_M4x3:
2411             nComponents = 3;
2412             tmp_ins.handler_idx = WINED3DSIH_DP4;
2413             break;
2414         case WINED3DSIH_M3x4:
2415             nComponents = 4;
2416             tmp_ins.handler_idx = WINED3DSIH_DP3;
2417             break;
2418         case WINED3DSIH_M3x3:
2419             nComponents = 3;
2420             tmp_ins.handler_idx = WINED3DSIH_DP3;
2421             break;
2422         case WINED3DSIH_M3x2:
2423             nComponents = 2;
2424             tmp_ins.handler_idx = WINED3DSIH_DP3;
2425             break;
2426         default:
2427             FIXME("Unhandled opcode %#x\n", ins->handler_idx);
2428             break;
2429     }
2430
2431     tmp_dst = ins->dst[0];
2432     tmp_src[0] = ins->src[0];
2433     tmp_src[1] = ins->src[1];
2434     for (i = 0; i < nComponents; i++) {
2435         tmp_dst.write_mask = WINED3DSP_WRITEMASK_0 << i;
2436         shader_hw_map2gl(&tmp_ins);
2437         ++tmp_src[1].reg.idx;
2438     }
2439 }
2440
2441 static void shader_hw_rcp(const struct wined3d_shader_instruction *ins)
2442 {
2443     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2444
2445     char dst[50];
2446     char src[50];
2447
2448     shader_arb_get_dst_param(ins, &ins->dst[0], dst); /* Destination */
2449     shader_arb_get_src_param(ins, &ins->src[0], 0, src);
2450     if (ins->src[0].swizzle == WINED3DSP_NOSWIZZLE)
2451     {
2452         /* Dx sdk says .x is used if no swizzle is given, but our test shows that
2453          * .w is used
2454          */
2455         strcat(src, ".w");
2456     }
2457
2458     shader_addline(buffer, "RCP%s %s, %s;\n", shader_arb_get_modifier(ins), dst, src);
2459 }
2460
2461 static void shader_hw_scalar_op(const struct wined3d_shader_instruction *ins)
2462 {
2463     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2464     const char *instruction;
2465
2466     char dst[50];
2467     char src[50];
2468
2469     switch(ins->handler_idx)
2470     {
2471         case WINED3DSIH_RSQ:  instruction = "RSQ"; break;
2472         case WINED3DSIH_RCP:  instruction = "RCP"; break;
2473         case WINED3DSIH_EXP:  instruction = "EX2"; break;
2474         case WINED3DSIH_EXPP: instruction = "EXP"; break;
2475         default: instruction = "";
2476             FIXME("Unhandled opcode %#x\n", ins->handler_idx);
2477             break;
2478     }
2479
2480     shader_arb_get_dst_param(ins, &ins->dst[0], dst); /* Destination */
2481     shader_arb_get_src_param(ins, &ins->src[0], 0, src);
2482     if (ins->src[0].swizzle == WINED3DSP_NOSWIZZLE)
2483     {
2484         /* Dx sdk says .x is used if no swizzle is given, but our test shows that
2485          * .w is used
2486          */
2487         strcat(src, ".w");
2488     }
2489
2490     shader_addline(buffer, "%s%s %s, %s;\n", instruction, shader_arb_get_modifier(ins), dst, src);
2491 }
2492
2493 static void shader_hw_nrm(const struct wined3d_shader_instruction *ins)
2494 {
2495     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2496     char dst_name[50];
2497     char src_name[50];
2498     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2499     BOOL pshader = shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type);
2500     const char *zero = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ZERO);
2501
2502     shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2503     shader_arb_get_src_param(ins, &ins->src[0], 1 /* Use TB */, src_name);
2504
2505     /* In D3D, NRM of a vector with length zero returns zero. Catch this situation, as
2506      * otherwise NRM or RSQ would return NaN */
2507     if(pshader && priv->target_version >= NV3)
2508     {
2509         /* GL_NV_fragment_program2's NRM needs protection against length zero vectors too
2510          *
2511          * TODO: Find out if DP3+NRM+MOV is really faster than DP3+RSQ+MUL
2512          */
2513         shader_addline(buffer, "DP3C TA, %s, %s;\n", src_name, src_name);
2514         shader_addline(buffer, "NRM%s %s, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name);
2515         shader_addline(buffer, "MOV %s (EQ), %s;\n", dst_name, zero);
2516     }
2517     else if(priv->target_version >= NV2)
2518     {
2519         shader_addline(buffer, "DP3C TA.x, %s, %s;\n", src_name, src_name);
2520         shader_addline(buffer, "RSQ TA.x (NE), TA.x;\n");
2521         shader_addline(buffer, "MUL%s %s, %s, TA.x;\n", shader_arb_get_modifier(ins), dst_name,
2522                        src_name);
2523     }
2524     else
2525     {
2526         const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
2527
2528         shader_addline(buffer, "DP3 TA.x, %s, %s;\n", src_name, src_name);
2529         /* Pass any non-zero value to RSQ if the input vector has a length of zero. The
2530          * RSQ result doesn't matter, as long as multiplying it by 0 returns 0.
2531          */
2532         shader_addline(buffer, "SGE TA.y, -TA.x, %s;\n", zero);
2533         shader_addline(buffer, "MAD TA.x, %s, TA.y, TA.x;\n", one);
2534
2535         shader_addline(buffer, "RSQ TA.x, TA.x;\n");
2536         /* dst.w = src[0].w * 1 / (src.x^2 + src.y^2 + src.z^2)^(1/2) according to msdn*/
2537         shader_addline(buffer, "MUL%s %s, %s, TA.x;\n", shader_arb_get_modifier(ins), dst_name,
2538                     src_name);
2539     }
2540 }
2541
2542 static void shader_hw_lrp(const struct wined3d_shader_instruction *ins)
2543 {
2544     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2545     char dst_name[50];
2546     char src_name[3][50];
2547
2548     /* ARB_fragment_program has a convenient LRP instruction */
2549     if(shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type)) {
2550         shader_hw_map2gl(ins);
2551         return;
2552     }
2553
2554     shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2555     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
2556     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
2557     shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
2558
2559     shader_addline(buffer, "SUB TA, %s, %s;\n", src_name[1], src_name[2]);
2560     shader_addline(buffer, "MAD%s %s, %s, TA, %s;\n", shader_arb_get_modifier(ins),
2561                    dst_name, src_name[0], src_name[2]);
2562 }
2563
2564 static void shader_hw_sincos(const struct wined3d_shader_instruction *ins)
2565 {
2566     /* This instruction exists in ARB, but the d3d instruction takes two extra parameters which
2567      * must contain fixed constants. So we need a separate function to filter those constants and
2568      * can't use map2gl
2569      */
2570     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2571     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2572     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2573     char dst_name[50];
2574     char src_name0[50], src_name1[50], src_name2[50];
2575     BOOL is_color;
2576
2577     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name0);
2578     if(shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type)) {
2579         shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2580         /* No modifiers are supported on SCS */
2581         shader_addline(buffer, "SCS %s, %s;\n", dst_name, src_name0);
2582
2583         if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE)
2584         {
2585             shader_arb_get_register_name(ins, &dst->reg, src_name0, &is_color);
2586             shader_addline(buffer, "MOV_SAT %s, %s;\n", dst_name, src_name0);
2587         }
2588     } else if(priv->target_version >= NV2) {
2589         shader_arb_get_register_name(ins, &dst->reg, dst_name, &is_color);
2590
2591         /* Sincos writemask must be .x, .y or .xy */
2592         if(dst->write_mask & WINED3DSP_WRITEMASK_0)
2593             shader_addline(buffer, "COS%s %s.x, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name0);
2594         if(dst->write_mask & WINED3DSP_WRITEMASK_1)
2595             shader_addline(buffer, "SIN%s %s.y, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name0);
2596     } else {
2597         /* Approximate sine and cosine with a taylor series, as per math textbook. The application passes 8
2598          * helper constants(D3DSINCOSCONST1 and D3DSINCOSCONST2) in src1 and src2.
2599          *
2600          * sin(x) = x - x^3/3! + x^5/5! - x^7/7! + ...
2601          * cos(x) = 1 - x^2/2! + x^4/4! - x^6/6! + ...
2602          *
2603          * The constants we get are:
2604          *
2605          *  +1   +1,     -1     -1     +1      +1      -1       -1
2606          *      ---- ,  ---- , ---- , ----- , ----- , ----- , ------
2607          *      1!*2    2!*4   3!*8   4!*16   5!*32   6!*64   7!*128
2608          *
2609          * If used with x^2, x^3, x^4 etc they calculate sin(x/2) and cos(x/2):
2610          *
2611          * (x/2)^2 = x^2 / 4
2612          * (x/2)^3 = x^3 / 8
2613          * (x/2)^4 = x^4 / 16
2614          * (x/2)^5 = x^5 / 32
2615          * etc
2616          *
2617          * To get the final result:
2618          * sin(x) = 2 * sin(x/2) * cos(x/2)
2619          * cos(x) = cos(x/2)^2 - sin(x/2)^2
2620          * (from sin(x+y) and cos(x+y) rules)
2621          *
2622          * As per MSDN, dst.z is undefined after the operation, and so is
2623          * dst.x and dst.y if they're masked out by the writemask. Ie
2624          * sincos dst.y, src1, c0, c1
2625          * returns the sine in dst.y. dst.x and dst.z are undefined, dst.w is not touched. The assembler
2626          * vsa.exe also stops with an error if the dest register is the same register as the source
2627          * register. This means we can use dest.xyz as temporary storage. The assembler vsa.exe output also
2628          * indicates that sincos consumes 8 instruction slots in vs_2_0(and, strangely, in vs_3_0).
2629          */
2630         shader_arb_get_src_param(ins, &ins->src[1], 1, src_name1);
2631         shader_arb_get_src_param(ins, &ins->src[2], 2, src_name2);
2632         shader_arb_get_register_name(ins, &dst->reg, dst_name, &is_color);
2633
2634         shader_addline(buffer, "MUL %s.x, %s, %s;\n", dst_name, src_name0, src_name0);  /* x ^ 2 */
2635         shader_addline(buffer, "MUL TA.y, %s.x, %s;\n", dst_name, src_name0);           /* x ^ 3 */
2636         shader_addline(buffer, "MUL %s.y, TA.y, %s;\n", dst_name, src_name0);           /* x ^ 4 */
2637         shader_addline(buffer, "MUL TA.z, %s.y, %s;\n", dst_name, src_name0);           /* x ^ 5 */
2638         shader_addline(buffer, "MUL %s.z, TA.z, %s;\n", dst_name, src_name0);           /* x ^ 6 */
2639         shader_addline(buffer, "MUL TA.w, %s.z, %s;\n", dst_name, src_name0);           /* x ^ 7 */
2640
2641         /* sin(x/2)
2642          *
2643          * Unfortunately we don't get the constants in a DP4-capable form. Is there a way to
2644          * properly merge that with MULs in the code above?
2645          * The swizzles .yz and xw however fit into the .yzxw swizzle added to ps_2_0. Maybe
2646          * we can merge the sine and cosine MAD rows to calculate them together.
2647          */
2648         shader_addline(buffer, "MUL TA.x, %s, %s.w;\n", src_name0, src_name2); /* x^1, +1/(1!*2) */
2649         shader_addline(buffer, "MAD TA.x, TA.y, %s.x, TA.x;\n", src_name2); /* -1/(3!*8) */
2650         shader_addline(buffer, "MAD TA.x, TA.z, %s.w, TA.x;\n", src_name1); /* +1/(5!*32) */
2651         shader_addline(buffer, "MAD TA.x, TA.w, %s.x, TA.x;\n", src_name1); /* -1/(7!*128) */
2652
2653         /* cos(x/2) */
2654         shader_addline(buffer, "MAD TA.y, %s.x, %s.y, %s.z;\n", dst_name, src_name2, src_name2); /* -1/(2!*4), +1.0 */
2655         shader_addline(buffer, "MAD TA.y, %s.y, %s.z, TA.y;\n", dst_name, src_name1); /* +1/(4!*16) */
2656         shader_addline(buffer, "MAD TA.y, %s.z, %s.y, TA.y;\n", dst_name, src_name1); /* -1/(6!*64) */
2657
2658         if(dst->write_mask & WINED3DSP_WRITEMASK_0) {
2659             /* cos x */
2660             shader_addline(buffer, "MUL TA.z, TA.y, TA.y;\n");
2661             shader_addline(buffer, "MAD %s.x, -TA.x, TA.x, TA.z;\n", dst_name);
2662         }
2663         if(dst->write_mask & WINED3DSP_WRITEMASK_1) {
2664             /* sin x */
2665             shader_addline(buffer, "MUL %s.y, TA.x, TA.y;\n", dst_name);
2666             shader_addline(buffer, "ADD %s.y, %s.y, %s.y;\n", dst_name, dst_name, dst_name);
2667         }
2668     }
2669 }
2670
2671 static void shader_hw_sgn(const struct wined3d_shader_instruction *ins)
2672 {
2673     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2674     char dst_name[50];
2675     char src_name[50];
2676     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
2677
2678     shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2679     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name);
2680
2681     /* SGN is only valid in vertex shaders */
2682     if(ctx->target_version >= NV2) {
2683         shader_addline(buffer, "SSG%s %s, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name);
2684         return;
2685     }
2686
2687     /* If SRC > 0.0, -SRC < SRC = TRUE, otherwise false.
2688      * if SRC < 0.0,  SRC < -SRC = TRUE. If neither is true, src = 0.0
2689      */
2690     if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE) {
2691         shader_addline(buffer, "SLT %s, -%s, %s;\n", dst_name, src_name, src_name);
2692     } else {
2693         /* src contains TA? Write to the dest first. This won't overwrite our destination.
2694          * Then use TA, and calculate the final result
2695          *
2696          * Not reading from TA? Store the first result in TA to avoid overwriting the
2697          * destination if src reg = dst reg
2698          */
2699         if(strstr(src_name, "TA"))
2700         {
2701             shader_addline(buffer, "SLT %s,  %s, -%s;\n", dst_name, src_name, src_name);
2702             shader_addline(buffer, "SLT TA, -%s, %s;\n", src_name, src_name);
2703             shader_addline(buffer, "ADD %s, %s, -TA;\n", dst_name, dst_name);
2704         }
2705         else
2706         {
2707             shader_addline(buffer, "SLT TA, -%s, %s;\n", src_name, src_name);
2708             shader_addline(buffer, "SLT %s,  %s, -%s;\n", dst_name, src_name, src_name);
2709             shader_addline(buffer, "ADD %s, TA, -%s;\n", dst_name, dst_name);
2710         }
2711     }
2712 }
2713
2714 static void shader_hw_dsy(const struct wined3d_shader_instruction *ins)
2715 {
2716     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2717     char src[50];
2718     char dst[50];
2719     char dst_name[50];
2720     BOOL is_color;
2721
2722     shader_arb_get_dst_param(ins, &ins->dst[0], dst);
2723     shader_arb_get_src_param(ins, &ins->src[0], 0, src);
2724     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2725
2726     shader_addline(buffer, "DDY %s, %s;\n", dst, src);
2727     shader_addline(buffer, "MUL%s %s, %s, ycorrection.y;\n", shader_arb_get_modifier(ins), dst, dst_name);
2728 }
2729
2730 static DWORD abs_modifier(DWORD mod, BOOL *need_abs)
2731 {
2732     *need_abs = FALSE;
2733
2734     switch(mod)
2735     {
2736         case WINED3DSPSM_NONE:      return WINED3DSPSM_ABS;
2737         case WINED3DSPSM_NEG:       return WINED3DSPSM_ABS;
2738         case WINED3DSPSM_BIAS:      *need_abs = TRUE; return WINED3DSPSM_BIAS;
2739         case WINED3DSPSM_BIASNEG:   *need_abs = TRUE; return WINED3DSPSM_BIASNEG;
2740         case WINED3DSPSM_SIGN:      *need_abs = TRUE; return WINED3DSPSM_SIGN;
2741         case WINED3DSPSM_SIGNNEG:   *need_abs = TRUE; return WINED3DSPSM_SIGNNEG;
2742         case WINED3DSPSM_COMP:      *need_abs = TRUE; return WINED3DSPSM_COMP;
2743         case WINED3DSPSM_X2:        *need_abs = TRUE; return WINED3DSPSM_X2;
2744         case WINED3DSPSM_X2NEG:     *need_abs = TRUE; return WINED3DSPSM_X2NEG;
2745         case WINED3DSPSM_DZ:        *need_abs = TRUE; return WINED3DSPSM_DZ;
2746         case WINED3DSPSM_DW:        *need_abs = TRUE; return WINED3DSPSM_DW;
2747         case WINED3DSPSM_ABS:       return WINED3DSPSM_ABS;
2748         case WINED3DSPSM_ABSNEG:    return WINED3DSPSM_ABS;
2749     }
2750     FIXME("Unknown modifier %u\n", mod);
2751     return mod;
2752 }
2753
2754 static void shader_hw_log(const struct wined3d_shader_instruction *ins)
2755 {
2756     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2757     char src0[50], dst[50];
2758     struct wined3d_shader_src_param src0_copy = ins->src[0];
2759     BOOL need_abs = FALSE;
2760     const char *instr;
2761
2762     switch(ins->handler_idx)
2763     {
2764         case WINED3DSIH_LOG:  instr = "LG2"; break;
2765         case WINED3DSIH_LOGP: instr = "LOG"; break;
2766         default:
2767             ERR("Unexpected instruction %d\n", ins->handler_idx);
2768             return;
2769     }
2770
2771     /* LOG and LOGP operate on the absolute value of the input */
2772     src0_copy.modifiers = abs_modifier(src0_copy.modifiers, &need_abs);
2773
2774     shader_arb_get_dst_param(ins, &ins->dst[0], dst);
2775     shader_arb_get_src_param(ins, &src0_copy, 0, src0);
2776
2777     if(need_abs)
2778     {
2779         shader_addline(buffer, "ABS TA, %s;\n", src0);
2780         shader_addline(buffer, "%s%s %s, TA;\n", instr, shader_arb_get_modifier(ins), dst);
2781     }
2782     else
2783     {
2784         shader_addline(buffer, "%s%s %s, %s;\n", instr, shader_arb_get_modifier(ins), dst, src0);
2785     }
2786 }
2787
2788 static void shader_hw_pow(const struct wined3d_shader_instruction *ins)
2789 {
2790     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2791     char src0[50], src1[50], dst[50];
2792     struct wined3d_shader_src_param src0_copy = ins->src[0];
2793     BOOL need_abs = FALSE;
2794     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2795     const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
2796
2797     /* POW operates on the absolute value of the input */
2798     src0_copy.modifiers = abs_modifier(src0_copy.modifiers, &need_abs);
2799
2800     shader_arb_get_dst_param(ins, &ins->dst[0], dst);
2801     shader_arb_get_src_param(ins, &src0_copy, 0, src0);
2802     shader_arb_get_src_param(ins, &ins->src[1], 1, src1);
2803
2804     if (need_abs)
2805         shader_addline(buffer, "ABS TA.x, %s;\n", src0);
2806     else
2807         shader_addline(buffer, "MOV TA.x, %s;\n", src0);
2808
2809     if (priv->target_version >= NV2)
2810     {
2811         shader_addline(buffer, "MOVC TA.y, %s;\n", src1);
2812         shader_addline(buffer, "POW%s %s, TA.x, TA.y;\n", shader_arb_get_modifier(ins), dst);
2813         shader_addline(buffer, "MOV %s (EQ.y), %s;\n", dst, one);
2814     }
2815     else
2816     {
2817         const char *zero = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ZERO);
2818         const char *flt_eps = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_EPS);
2819
2820         shader_addline(buffer, "ABS TA.y, %s;\n", src1);
2821         shader_addline(buffer, "SGE TA.y, -TA.y, %s;\n", zero);
2822         /* Possibly add flt_eps to avoid getting float special values */
2823         shader_addline(buffer, "MAD TA.z, TA.y, %s, %s;\n", flt_eps, src1);
2824         shader_addline(buffer, "POW%s TA.x, TA.x, TA.z;\n", shader_arb_get_modifier(ins));
2825         shader_addline(buffer, "MAD TA.x, -TA.x, TA.y, TA.x;\n");
2826         shader_addline(buffer, "MAD %s, TA.y, %s, TA.x;\n", dst, one);
2827     }
2828 }
2829
2830 static void shader_hw_loop(const struct wined3d_shader_instruction *ins)
2831 {
2832     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2833     char src_name[50];
2834     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2835
2836     /* src0 is aL */
2837     shader_arb_get_src_param(ins, &ins->src[1], 0, src_name);
2838
2839     if(vshader)
2840     {
2841         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2842         struct list *e = list_head(&priv->control_frames);
2843         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2844
2845         if(priv->loop_depth > 1) shader_addline(buffer, "PUSHA aL;\n");
2846         /* The constant loader makes sure to load -1 into iX.w */
2847         shader_addline(buffer, "ARLC aL, %s.xywz;\n", src_name);
2848         shader_addline(buffer, "BRA loop_%u_end (LE.x);\n", control_frame->no.loop);
2849         shader_addline(buffer, "loop_%u_start:\n", control_frame->no.loop);
2850     }
2851     else
2852     {
2853         shader_addline(buffer, "LOOP %s;\n", src_name);
2854     }
2855 }
2856
2857 static void shader_hw_rep(const struct wined3d_shader_instruction *ins)
2858 {
2859     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2860     char src_name[50];
2861     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2862
2863     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name);
2864
2865     /* The constant loader makes sure to load -1 into iX.w */
2866     if(vshader)
2867     {
2868         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2869         struct list *e = list_head(&priv->control_frames);
2870         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2871
2872         if(priv->loop_depth > 1) shader_addline(buffer, "PUSHA aL;\n");
2873
2874         shader_addline(buffer, "ARLC aL, %s.xywz;\n", src_name);
2875         shader_addline(buffer, "BRA loop_%u_end (LE.x);\n", control_frame->no.loop);
2876         shader_addline(buffer, "loop_%u_start:\n", control_frame->no.loop);
2877     }
2878     else
2879     {
2880         shader_addline(buffer, "REP %s;\n", src_name);
2881     }
2882 }
2883
2884 static void shader_hw_endloop(const struct wined3d_shader_instruction *ins)
2885 {
2886     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2887     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2888
2889     if(vshader)
2890     {
2891         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2892         struct list *e = list_head(&priv->control_frames);
2893         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2894
2895         shader_addline(buffer, "ARAC aL.xy, aL;\n");
2896         shader_addline(buffer, "BRA loop_%u_start (GT.x);\n", control_frame->no.loop);
2897         shader_addline(buffer, "loop_%u_end:\n", control_frame->no.loop);
2898
2899         if(priv->loop_depth > 1) shader_addline(buffer, "POPA aL;\n");
2900     }
2901     else
2902     {
2903         shader_addline(buffer, "ENDLOOP;\n");
2904     }
2905 }
2906
2907 static void shader_hw_endrep(const struct wined3d_shader_instruction *ins)
2908 {
2909     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2910     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2911
2912     if(vshader)
2913     {
2914         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2915         struct list *e = list_head(&priv->control_frames);
2916         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2917
2918         shader_addline(buffer, "ARAC aL.xy, aL;\n");
2919         shader_addline(buffer, "BRA loop_%u_start (GT.x);\n", control_frame->no.loop);
2920         shader_addline(buffer, "loop_%u_end:\n", control_frame->no.loop);
2921
2922         if(priv->loop_depth > 1) shader_addline(buffer, "POPA aL;\n");
2923     }
2924     else
2925     {
2926         shader_addline(buffer, "ENDREP;\n");
2927     }
2928 }
2929
2930 static const struct control_frame *find_last_loop(const struct shader_arb_ctx_priv *priv)
2931 {
2932     struct control_frame *control_frame;
2933
2934     LIST_FOR_EACH_ENTRY(control_frame, &priv->control_frames, struct control_frame, entry)
2935     {
2936         if(control_frame->type == LOOP || control_frame->type == REP) return control_frame;
2937     }
2938     ERR("Could not find loop for break\n");
2939     return NULL;
2940 }
2941
2942 static void shader_hw_break(const struct wined3d_shader_instruction *ins)
2943 {
2944     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2945     const struct control_frame *control_frame = find_last_loop(ins->ctx->backend_data);
2946     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2947
2948     if(vshader)
2949     {
2950         shader_addline(buffer, "BRA loop_%u_end;\n", control_frame->no.loop);
2951     }
2952     else
2953     {
2954         shader_addline(buffer, "BRK;\n");
2955     }
2956 }
2957
2958 static const char *get_compare(enum wined3d_shader_rel_op op)
2959 {
2960     switch (op)
2961     {
2962         case WINED3D_SHADER_REL_OP_GT: return "GT";
2963         case WINED3D_SHADER_REL_OP_EQ: return "EQ";
2964         case WINED3D_SHADER_REL_OP_GE: return "GE";
2965         case WINED3D_SHADER_REL_OP_LT: return "LT";
2966         case WINED3D_SHADER_REL_OP_NE: return "NE";
2967         case WINED3D_SHADER_REL_OP_LE: return "LE";
2968         default:
2969             FIXME("Unrecognized operator %#x.\n", op);
2970             return "(\?\?)";
2971     }
2972 }
2973
2974 static enum wined3d_shader_rel_op invert_compare(enum wined3d_shader_rel_op op)
2975 {
2976     switch (op)
2977     {
2978         case WINED3D_SHADER_REL_OP_GT: return WINED3D_SHADER_REL_OP_LE;
2979         case WINED3D_SHADER_REL_OP_EQ: return WINED3D_SHADER_REL_OP_NE;
2980         case WINED3D_SHADER_REL_OP_GE: return WINED3D_SHADER_REL_OP_LT;
2981         case WINED3D_SHADER_REL_OP_LT: return WINED3D_SHADER_REL_OP_GE;
2982         case WINED3D_SHADER_REL_OP_NE: return WINED3D_SHADER_REL_OP_EQ;
2983         case WINED3D_SHADER_REL_OP_LE: return WINED3D_SHADER_REL_OP_GT;
2984         default:
2985             FIXME("Unrecognized operator %#x.\n", op);
2986             return -1;
2987     }
2988 }
2989
2990 static void shader_hw_breakc(const struct wined3d_shader_instruction *ins)
2991 {
2992     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2993     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2994     const struct control_frame *control_frame = find_last_loop(ins->ctx->backend_data);
2995     char src_name0[50];
2996     char src_name1[50];
2997     const char *comp = get_compare(ins->flags);
2998
2999     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name0);
3000     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name1);
3001
3002     if(vshader)
3003     {
3004         /* SUBC CC, src0, src1" works only in pixel shaders, so use TA to throw
3005          * away the subtraction result
3006          */
3007         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
3008         shader_addline(buffer, "BRA loop_%u_end (%s.x);\n", control_frame->no.loop, comp);
3009     }
3010     else
3011     {
3012         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
3013         shader_addline(buffer, "BRK (%s.x);\n", comp);
3014     }
3015 }
3016
3017 static void shader_hw_ifc(const struct wined3d_shader_instruction *ins)
3018 {
3019     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3020     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3021     struct list *e = list_head(&priv->control_frames);
3022     struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
3023     const char *comp;
3024     char src_name0[50];
3025     char src_name1[50];
3026     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
3027
3028     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name0);
3029     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name1);
3030
3031     if(vshader)
3032     {
3033         /* Invert the flag. We jump to the else label if the condition is NOT true */
3034         comp = get_compare(invert_compare(ins->flags));
3035         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
3036         shader_addline(buffer, "BRA ifc_%u_else (%s.x);\n", control_frame->no.ifc, comp);
3037     }
3038     else
3039     {
3040         comp = get_compare(ins->flags);
3041         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
3042         shader_addline(buffer, "IF %s.x;\n", comp);
3043     }
3044 }
3045
3046 static void shader_hw_else(const struct wined3d_shader_instruction *ins)
3047 {
3048     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3049     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3050     struct list *e = list_head(&priv->control_frames);
3051     struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
3052     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
3053
3054     if(vshader)
3055     {
3056         shader_addline(buffer, "BRA ifc_%u_endif;\n", control_frame->no.ifc);
3057         shader_addline(buffer, "ifc_%u_else:\n", control_frame->no.ifc);
3058         control_frame->had_else = TRUE;
3059     }
3060     else
3061     {
3062         shader_addline(buffer, "ELSE;\n");
3063     }
3064 }
3065
3066 static void shader_hw_endif(const struct wined3d_shader_instruction *ins)
3067 {
3068     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3069     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3070     struct list *e = list_head(&priv->control_frames);
3071     struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
3072     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
3073
3074     if(vshader)
3075     {
3076         if(control_frame->had_else)
3077         {
3078             shader_addline(buffer, "ifc_%u_endif:\n", control_frame->no.ifc);
3079         }
3080         else
3081         {
3082             shader_addline(buffer, "#No else branch. else is endif\n");
3083             shader_addline(buffer, "ifc_%u_else:\n", control_frame->no.ifc);
3084         }
3085     }
3086     else
3087     {
3088         shader_addline(buffer, "ENDIF;\n");
3089     }
3090 }
3091
3092 static void shader_hw_texldd(const struct wined3d_shader_instruction *ins)
3093 {
3094     DWORD sampler_idx = ins->src[1].reg.idx;
3095     char reg_dest[40];
3096     char reg_src[3][40];
3097     WORD flags = TEX_DERIV;
3098
3099     shader_arb_get_dst_param(ins, &ins->dst[0], reg_dest);
3100     shader_arb_get_src_param(ins, &ins->src[0], 0, reg_src[0]);
3101     shader_arb_get_src_param(ins, &ins->src[2], 1, reg_src[1]);
3102     shader_arb_get_src_param(ins, &ins->src[3], 2, reg_src[2]);
3103
3104     if (ins->flags & WINED3DSI_TEXLD_PROJECT) flags |= TEX_PROJ;
3105     if (ins->flags & WINED3DSI_TEXLD_BIAS) flags |= TEX_BIAS;
3106
3107     shader_hw_sample(ins, sampler_idx, reg_dest, reg_src[0], flags, reg_src[1], reg_src[2]);
3108 }
3109
3110 static void shader_hw_texldl(const struct wined3d_shader_instruction *ins)
3111 {
3112     DWORD sampler_idx = ins->src[1].reg.idx;
3113     char reg_dest[40];
3114     char reg_coord[40];
3115     WORD flags = TEX_LOD;
3116
3117     shader_arb_get_dst_param(ins, &ins->dst[0], reg_dest);
3118     shader_arb_get_src_param(ins, &ins->src[0], 0, reg_coord);
3119
3120     if (ins->flags & WINED3DSI_TEXLD_PROJECT) flags |= TEX_PROJ;
3121     if (ins->flags & WINED3DSI_TEXLD_BIAS) flags |= TEX_BIAS;
3122
3123     shader_hw_sample(ins, sampler_idx, reg_dest, reg_coord, flags, NULL, NULL);
3124 }
3125
3126 static void shader_hw_label(const struct wined3d_shader_instruction *ins)
3127 {
3128     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3129     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3130
3131     priv->in_main_func = FALSE;
3132     /* Call instructions activate the NV extensions, not labels and rets. If there is an uncalled
3133      * subroutine, don't generate a label that will make GL complain
3134      */
3135     if(priv->target_version == ARB) return;
3136
3137     shader_addline(buffer, "l%u:\n", ins->src[0].reg.idx);
3138 }
3139
3140 static void vshader_add_footer(struct shader_arb_ctx_priv *priv_ctx,
3141         const struct arb_vshader_private *shader_data, const struct arb_vs_compile_args *args,
3142         const struct wined3d_shader_reg_maps *reg_maps, const struct wined3d_gl_info *gl_info,
3143         struct wined3d_shader_buffer *buffer)
3144 {
3145     unsigned int i;
3146
3147     /* The D3DRS_FOGTABLEMODE render state defines if the shader-generated fog coord is used
3148      * or if the fragment depth is used. If the fragment depth is used(FOGTABLEMODE != NONE),
3149      * the fog frag coord is thrown away. If the fog frag coord is used, but not written by
3150      * the shader, it is set to 0.0(fully fogged, since start = 1.0, end = 0.0)
3151      */
3152     if (args->super.fog_src == VS_FOG_Z)
3153     {
3154         shader_addline(buffer, "MOV result.fogcoord, TMP_OUT.z;\n");
3155     }
3156     else
3157     {
3158         if (!reg_maps->fog)
3159         {
3160             /* posFixup.x is always 1.0, so we can safely use it */
3161             shader_addline(buffer, "ADD result.fogcoord, posFixup.x, -posFixup.x;\n");
3162         }
3163         else
3164         {
3165             /* Clamp fogcoord */
3166             const char *zero = arb_get_helper_value(reg_maps->shader_version.type, ARB_ZERO);
3167             const char *one = arb_get_helper_value(reg_maps->shader_version.type, ARB_ONE);
3168
3169             shader_addline(buffer, "MIN TMP_FOGCOORD.x, TMP_FOGCOORD.x, %s;\n", one);
3170             shader_addline(buffer, "MAX result.fogcoord.x, TMP_FOGCOORD.x, %s;\n", zero);
3171         }
3172     }
3173
3174     /* Clipplanes are always stored without y inversion */
3175     if (use_nv_clip(gl_info) && priv_ctx->target_version >= NV2)
3176     {
3177         if (args->super.clip_enabled)
3178         {
3179             for (i = 0; i < priv_ctx->vs_clipplanes; i++)
3180             {
3181                 shader_addline(buffer, "DP4 result.clip[%u].x, TMP_OUT, state.clip[%u].plane;\n", i, i);
3182             }
3183         }
3184     }
3185     else if (args->clip.boolclip.clip_texcoord)
3186     {
3187         unsigned int cur_clip = 0;
3188         char component[4] = {'x', 'y', 'z', 'w'};
3189         const char *zero = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_ZERO);
3190
3191         for (i = 0; i < gl_info->limits.clipplanes; ++i)
3192         {
3193             if (args->clip.boolclip.clipplane_mask & (1 << i))
3194             {
3195                 shader_addline(buffer, "DP4 TA.%c, TMP_OUT, state.clip[%u].plane;\n",
3196                                component[cur_clip++], i);
3197             }
3198         }
3199         switch (cur_clip)
3200         {
3201             case 0:
3202                 shader_addline(buffer, "MOV TA, %s;\n", zero);
3203                 break;
3204             case 1:
3205                 shader_addline(buffer, "MOV TA.yzw, %s;\n", zero);
3206                 break;
3207             case 2:
3208                 shader_addline(buffer, "MOV TA.zw, %s;\n", zero);
3209                 break;
3210             case 3:
3211                 shader_addline(buffer, "MOV TA.w, %s;\n", zero);
3212                 break;
3213         }
3214         shader_addline(buffer, "MOV result.texcoord[%u], TA;\n",
3215                        args->clip.boolclip.clip_texcoord - 1);
3216     }
3217
3218     /* Write the final position.
3219      *
3220      * OpenGL coordinates specify the center of the pixel while d3d coords specify
3221      * the corner. The offsets are stored in z and w in posFixup. posFixup.y contains
3222      * 1.0 or -1.0 to turn the rendering upside down for offscreen rendering. PosFixup.x
3223      * contains 1.0 to allow a mad, but arb vs swizzles are too restricted for that.
3224      */
3225     shader_addline(buffer, "MUL TA, posFixup, TMP_OUT.w;\n");
3226     shader_addline(buffer, "ADD TMP_OUT.x, TMP_OUT.x, TA.z;\n");
3227     shader_addline(buffer, "MAD TMP_OUT.y, TMP_OUT.y, posFixup.y, TA.w;\n");
3228
3229     /* Z coord [0;1]->[-1;1] mapping, see comment in transform_projection in state.c
3230      * and the glsl equivalent
3231      */
3232     if (need_helper_const(shader_data, reg_maps, gl_info))
3233     {
3234         const char *two = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_TWO);
3235         shader_addline(buffer, "MAD TMP_OUT.z, TMP_OUT.z, %s, -TMP_OUT.w;\n", two);
3236     }
3237     else
3238     {
3239         shader_addline(buffer, "ADD TMP_OUT.z, TMP_OUT.z, TMP_OUT.z;\n");
3240         shader_addline(buffer, "ADD TMP_OUT.z, TMP_OUT.z, -TMP_OUT.w;\n");
3241     }
3242
3243     shader_addline(buffer, "MOV result.position, TMP_OUT;\n");
3244
3245     priv_ctx->footer_written = TRUE;
3246 }
3247
3248 static void shader_hw_ret(const struct wined3d_shader_instruction *ins)
3249 {
3250     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3251     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3252     const struct wined3d_shader *shader = ins->ctx->shader;
3253     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
3254
3255     if(priv->target_version == ARB) return;
3256
3257     if(vshader)
3258     {
3259         if (priv->in_main_func) vshader_add_footer(priv, shader->backend_data,
3260                 priv->cur_vs_args, ins->ctx->reg_maps, ins->ctx->gl_info, buffer);
3261     }
3262
3263     shader_addline(buffer, "RET;\n");
3264 }
3265
3266 static void shader_hw_call(const struct wined3d_shader_instruction *ins)
3267 {
3268     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3269     shader_addline(buffer, "CAL l%u;\n", ins->src[0].reg.idx);
3270 }
3271
3272 /* GL locking is done by the caller */
3273 static GLuint create_arb_blt_vertex_program(const struct wined3d_gl_info *gl_info)
3274 {
3275     GLuint program_id = 0;
3276     GLint pos;
3277
3278     const char *blt_vprogram =
3279         "!!ARBvp1.0\n"
3280         "PARAM c[1] = { { 1, 0.5 } };\n"
3281         "MOV result.position, vertex.position;\n"
3282         "MOV result.color, c[0].x;\n"
3283         "MOV result.texcoord[0], vertex.texcoord[0];\n"
3284         "END\n";
3285
3286     GL_EXTCALL(glGenProgramsARB(1, &program_id));
3287     GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, program_id));
3288     GL_EXTCALL(glProgramStringARB(GL_VERTEX_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
3289             strlen(blt_vprogram), blt_vprogram));
3290     checkGLcall("glProgramStringARB()");
3291
3292     gl_info->gl_ops.gl.p_glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
3293     if (pos != -1)
3294     {
3295         FIXME("Vertex program error at position %d: %s\n\n", pos,
3296             debugstr_a((const char *)gl_info->gl_ops.gl.p_glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
3297         shader_arb_dump_program_source(blt_vprogram);
3298     }
3299     else
3300     {
3301         GLint native;
3302
3303         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
3304         checkGLcall("glGetProgramivARB()");
3305         if (!native) WARN("Program exceeds native resource limits.\n");
3306     }
3307
3308     return program_id;
3309 }
3310
3311 /* GL locking is done by the caller */
3312 static GLuint create_arb_blt_fragment_program(const struct wined3d_gl_info *gl_info,
3313         enum tex_types tex_type, BOOL masked)
3314 {
3315     GLuint program_id = 0;
3316     const char *fprogram;
3317     GLint pos;
3318
3319     static const char * const blt_fprograms_full[tex_type_count] =
3320     {
3321         /* tex_1d */
3322         NULL,
3323         /* tex_2d */
3324         "!!ARBfp1.0\n"
3325         "TEMP R0;\n"
3326         "TEX R0.x, fragment.texcoord[0], texture[0], 2D;\n"
3327         "MOV result.depth.z, R0.x;\n"
3328         "END\n",
3329         /* tex_3d */
3330         NULL,
3331         /* tex_cube */
3332         "!!ARBfp1.0\n"
3333         "TEMP R0;\n"
3334         "TEX R0.x, fragment.texcoord[0], texture[0], CUBE;\n"
3335         "MOV result.depth.z, R0.x;\n"
3336         "END\n",
3337         /* tex_rect */
3338         "!!ARBfp1.0\n"
3339         "TEMP R0;\n"
3340         "TEX R0.x, fragment.texcoord[0], texture[0], RECT;\n"
3341         "MOV result.depth.z, R0.x;\n"
3342         "END\n",
3343     };
3344
3345     static const char * const blt_fprograms_masked[tex_type_count] =
3346     {
3347         /* tex_1d */
3348         NULL,
3349         /* tex_2d */
3350         "!!ARBfp1.0\n"
3351         "PARAM mask = program.local[0];\n"
3352         "TEMP R0;\n"
3353         "SLT R0.xy, fragment.position, mask.zwzw;\n"
3354         "MUL R0.x, R0.x, R0.y;\n"
3355         "KIL -R0.x;\n"
3356         "TEX R0.x, fragment.texcoord[0], texture[0], 2D;\n"
3357         "MOV result.depth.z, R0.x;\n"
3358         "END\n",
3359         /* tex_3d */
3360         NULL,
3361         /* tex_cube */
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], CUBE;\n"
3369         "MOV result.depth.z, R0.x;\n"
3370         "END\n",
3371         /* tex_rect */
3372         "!!ARBfp1.0\n"
3373         "PARAM mask = program.local[0];\n"
3374         "TEMP R0;\n"
3375         "SLT R0.xy, fragment.position, mask.zwzw;\n"
3376         "MUL R0.x, R0.x, R0.y;\n"
3377         "KIL -R0.x;\n"
3378         "TEX R0.x, fragment.texcoord[0], texture[0], RECT;\n"
3379         "MOV result.depth.z, R0.x;\n"
3380         "END\n",
3381     };
3382
3383     fprogram = masked ? blt_fprograms_masked[tex_type] : blt_fprograms_full[tex_type];
3384     if (!fprogram)
3385     {
3386         FIXME("tex_type %#x not supported, falling back to tex_2d\n", tex_type);
3387         tex_type = tex_2d;
3388         fprogram = masked ? blt_fprograms_masked[tex_type] : blt_fprograms_full[tex_type];
3389     }
3390
3391     GL_EXTCALL(glGenProgramsARB(1, &program_id));
3392     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, program_id));
3393     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB, strlen(fprogram), fprogram));
3394     checkGLcall("glProgramStringARB()");
3395
3396     gl_info->gl_ops.gl.p_glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
3397     if (pos != -1)
3398     {
3399         FIXME("Fragment program error at position %d: %s\n\n", pos,
3400             debugstr_a((const char *)gl_info->gl_ops.gl.p_glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
3401         shader_arb_dump_program_source(fprogram);
3402     }
3403     else
3404     {
3405         GLint native;
3406
3407         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
3408         checkGLcall("glGetProgramivARB()");
3409         if (!native) WARN("Program exceeds native resource limits.\n");
3410     }
3411
3412     return program_id;
3413 }
3414
3415 static void arbfp_add_sRGB_correction(struct wined3d_shader_buffer *buffer, const char *fragcolor,
3416         const char *tmp1, const char *tmp2, const char *tmp3, const char *tmp4, BOOL condcode)
3417 {
3418     /* Perform sRGB write correction. See GLX_EXT_framebuffer_sRGB */
3419
3420     if(condcode)
3421     {
3422         /* Sigh. MOVC CC doesn't work, so use one of the temps as dummy dest */
3423         shader_addline(buffer, "SUBC %s, %s.x, srgb_consts1.y;\n", tmp1, fragcolor);
3424         /* Calculate the > 0.0031308 case */
3425         shader_addline(buffer, "POW %s.x (GE), %s.x, srgb_consts1.z;\n", fragcolor, fragcolor);
3426         shader_addline(buffer, "POW %s.y (GE), %s.y, srgb_consts1.z;\n", fragcolor, fragcolor);
3427         shader_addline(buffer, "POW %s.z (GE), %s.z, srgb_consts1.z;\n", fragcolor, fragcolor);
3428         shader_addline(buffer, "MUL %s.xyz (GE), %s, srgb_consts1.w;\n", fragcolor, fragcolor);
3429         shader_addline(buffer, "SUB %s.xyz (GE), %s, srgb_consts2.x;\n", fragcolor, fragcolor);
3430         /* Calculate the < case */
3431         shader_addline(buffer, "MUL %s.xyz (LT), srgb_consts1.x, %s;\n", fragcolor, fragcolor);
3432     }
3433     else
3434     {
3435         /* Calculate the > 0.0031308 case */
3436         shader_addline(buffer, "POW %s.x, %s.x, srgb_consts1.z;\n", tmp1, fragcolor);
3437         shader_addline(buffer, "POW %s.y, %s.y, srgb_consts1.z;\n", tmp1, fragcolor);
3438         shader_addline(buffer, "POW %s.z, %s.z, srgb_consts1.z;\n", tmp1, fragcolor);
3439         shader_addline(buffer, "MUL %s, %s, srgb_consts1.w;\n", tmp1, tmp1);
3440         shader_addline(buffer, "SUB %s, %s, srgb_consts2.x;\n", tmp1, tmp1);
3441         /* Calculate the < case */
3442         shader_addline(buffer, "MUL %s, srgb_consts1.x, %s;\n", tmp2, fragcolor);
3443         /* Get 1.0 / 0.0 masks for > 0.0031308 and < 0.0031308 */
3444         shader_addline(buffer, "SLT %s, srgb_consts1.y, %s;\n", tmp3, fragcolor);
3445         shader_addline(buffer, "SGE %s, srgb_consts1.y, %s;\n", tmp4, fragcolor);
3446         /* Store the components > 0.0031308 in the destination */
3447         shader_addline(buffer, "MUL %s.xyz, %s, %s;\n", fragcolor, tmp1, tmp3);
3448         /* Add the components that are < 0.0031308 */
3449         shader_addline(buffer, "MAD %s.xyz, %s, %s, %s;\n", fragcolor, tmp2, tmp4, fragcolor);
3450         /* Move everything into result.color at once. Nvidia hardware cannot handle partial
3451         * result.color writes(.rgb first, then .a), or handle overwriting already written
3452         * components. The assembler uses a temporary register in this case, which is usually
3453         * not allocated from one of our registers that were used earlier.
3454         */
3455     }
3456     /* [0.0;1.0] clamping. Not needed, this is done implicitly */
3457 }
3458
3459 static const DWORD *find_loop_control_values(const struct wined3d_shader *shader, DWORD idx)
3460 {
3461     const struct wined3d_shader_lconst *constant;
3462
3463     LIST_FOR_EACH_ENTRY(constant, &shader->constantsI, struct wined3d_shader_lconst, entry)
3464     {
3465         if (constant->idx == idx)
3466         {
3467             return constant->value;
3468         }
3469     }
3470     return NULL;
3471 }
3472
3473 static void init_ps_input(const struct wined3d_shader *shader,
3474         const struct arb_ps_compile_args *args, struct shader_arb_ctx_priv *priv)
3475 {
3476     static const char * const texcoords[8] =
3477     {
3478         "fragment.texcoord[0]", "fragment.texcoord[1]", "fragment.texcoord[2]", "fragment.texcoord[3]",
3479         "fragment.texcoord[4]", "fragment.texcoord[5]", "fragment.texcoord[6]", "fragment.texcoord[7]"
3480     };
3481     unsigned int i;
3482     const struct wined3d_shader_signature_element *sig = shader->input_signature;
3483     const char *semantic_name;
3484     DWORD semantic_idx;
3485
3486     switch(args->super.vp_mode)
3487     {
3488         case pretransformed:
3489         case fixedfunction:
3490             /* The pixelshader has to collect the varyings on its own. In any case properly load
3491              * color0 and color1. In the case of pretransformed vertices also load texcoords. Set
3492              * other attribs to 0.0.
3493              *
3494              * For fixedfunction this behavior is correct, according to the tests. For pretransformed
3495              * we'd either need a replacement shader that can load other attribs like BINORMAL, or
3496              * load the texcoord attrib pointers to match the pixel shader signature
3497              */
3498             for(i = 0; i < MAX_REG_INPUT; i++)
3499             {
3500                 semantic_name = sig[i].semantic_name;
3501                 semantic_idx = sig[i].semantic_idx;
3502                 if (!semantic_name) continue;
3503
3504                 if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_COLOR))
3505                 {
3506                     if (!semantic_idx) priv->ps_input[i] = "fragment.color.primary";
3507                     else if(semantic_idx == 1) priv->ps_input[i] = "fragment.color.secondary";
3508                     else priv->ps_input[i] = "0.0";
3509                 }
3510                 else if(args->super.vp_mode == fixedfunction)
3511                 {
3512                     priv->ps_input[i] = "0.0";
3513                 }
3514                 else if(shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_TEXCOORD))
3515                 {
3516                     if(semantic_idx < 8) priv->ps_input[i] = texcoords[semantic_idx];
3517                     else priv->ps_input[i] = "0.0";
3518                 }
3519                 else if(shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_FOG))
3520                 {
3521                     if (!semantic_idx) priv->ps_input[i] = "fragment.fogcoord";
3522                     else priv->ps_input[i] = "0.0";
3523                 }
3524                 else
3525                 {
3526                     priv->ps_input[i] = "0.0";
3527                 }
3528
3529                 TRACE("v%u, semantic %s%u is %s\n", i, semantic_name, semantic_idx, priv->ps_input[i]);
3530             }
3531             break;
3532
3533         case vertexshader:
3534             /* That one is easy. The vertex shaders provide v0-v7 in fragment.texcoord and v8 and v9 in
3535              * fragment.color
3536              */
3537             for(i = 0; i < 8; i++)
3538             {
3539                 priv->ps_input[i] = texcoords[i];
3540             }
3541             priv->ps_input[8] = "fragment.color.primary";
3542             priv->ps_input[9] = "fragment.color.secondary";
3543             break;
3544     }
3545 }
3546
3547 /* GL locking is done by the caller */
3548 static GLuint shader_arb_generate_pshader(const struct wined3d_shader *shader,
3549         const struct wined3d_gl_info *gl_info, struct wined3d_shader_buffer *buffer,
3550         const struct arb_ps_compile_args *args, struct arb_ps_compiled_shader *compiled)
3551 {
3552     const struct wined3d_shader_reg_maps *reg_maps = &shader->reg_maps;
3553     const struct wined3d_shader_lconst *lconst;
3554     const DWORD *function = shader->function;
3555     GLuint retval;
3556     char fragcolor[16];
3557     DWORD *lconst_map = local_const_mapping(shader), next_local;
3558     struct shader_arb_ctx_priv priv_ctx;
3559     BOOL dcl_td = FALSE;
3560     BOOL want_nv_prog = FALSE;
3561     struct arb_pshader_private *shader_priv = shader->backend_data;
3562     GLint errPos;
3563     DWORD map;
3564
3565     char srgbtmp[4][4];
3566     unsigned int i, found = 0;
3567
3568     for (i = 0, map = reg_maps->temporary; map; map >>= 1, ++i)
3569     {
3570         if (!(map & 1)
3571                 || (shader->u.ps.color0_mov && i == shader->u.ps.color0_reg)
3572                 || (reg_maps->shader_version.major < 2 && !i))
3573             continue;
3574
3575         sprintf(srgbtmp[found], "R%u", i);
3576         ++found;
3577         if (found == 4) break;
3578     }
3579
3580     switch(found) {
3581         case 0:
3582             sprintf(srgbtmp[0], "TA");
3583             sprintf(srgbtmp[1], "TB");
3584             sprintf(srgbtmp[2], "TC");
3585             sprintf(srgbtmp[3], "TD");
3586             dcl_td = TRUE;
3587             break;
3588         case 1:
3589             sprintf(srgbtmp[1], "TA");
3590             sprintf(srgbtmp[2], "TB");
3591             sprintf(srgbtmp[3], "TC");
3592             break;
3593         case 2:
3594             sprintf(srgbtmp[2], "TA");
3595             sprintf(srgbtmp[3], "TB");
3596             break;
3597         case 3:
3598             sprintf(srgbtmp[3], "TA");
3599             break;
3600         case 4:
3601             break;
3602     }
3603
3604     /*  Create the hw ARB shader */
3605     memset(&priv_ctx, 0, sizeof(priv_ctx));
3606     priv_ctx.cur_ps_args = args;
3607     priv_ctx.compiled_fprog = compiled;
3608     priv_ctx.cur_np2fixup_info = &compiled->np2fixup_info;
3609     init_ps_input(shader, args, &priv_ctx);
3610     list_init(&priv_ctx.control_frames);
3611
3612     /* Avoid enabling NV_fragment_program* if we do not need it.
3613      *
3614      * Enabling GL_NV_fragment_program_option causes the driver to occupy a temporary register,
3615      * and it slows down the shader execution noticeably(about 5%). Usually our instruction emulation
3616      * is faster than what we gain from using higher native instructions. There are some things though
3617      * that cannot be emulated. In that case enable the extensions.
3618      * If the extension is enabled, instruction handlers that support both ways will use it.
3619      *
3620      * Testing shows no performance difference between OPTION NV_fragment_program2 and NV_fragment_program.
3621      * So enable the best we can get.
3622      */
3623     if(reg_maps->usesdsx || reg_maps->usesdsy || reg_maps->loop_depth > 0 || reg_maps->usestexldd ||
3624        reg_maps->usestexldl || reg_maps->usesfacing || reg_maps->usesifc || reg_maps->usescall)
3625     {
3626         want_nv_prog = TRUE;
3627     }
3628
3629     shader_addline(buffer, "!!ARBfp1.0\n");
3630     if (want_nv_prog && gl_info->supported[NV_FRAGMENT_PROGRAM2])
3631     {
3632         shader_addline(buffer, "OPTION NV_fragment_program2;\n");
3633         priv_ctx.target_version = NV3;
3634     }
3635     else if (want_nv_prog && gl_info->supported[NV_FRAGMENT_PROGRAM_OPTION])
3636     {
3637         shader_addline(buffer, "OPTION NV_fragment_program;\n");
3638         priv_ctx.target_version = NV2;
3639     } else {
3640         if(want_nv_prog)
3641         {
3642             /* This is an error - either we're advertising the wrong shader version, or aren't enforcing some
3643              * limits properly
3644              */
3645             ERR("The shader requires instructions that are not available in plain GL_ARB_fragment_program\n");
3646             ERR("Try GLSL\n");
3647         }
3648         priv_ctx.target_version = ARB;
3649     }
3650
3651     if (reg_maps->rt_mask > 1)
3652     {
3653         shader_addline(buffer, "OPTION ARB_draw_buffers;\n");
3654     }
3655
3656     if (reg_maps->shader_version.major < 3)
3657     {
3658         switch(args->super.fog) {
3659             case FOG_OFF:
3660                 break;
3661             case FOG_LINEAR:
3662                 shader_addline(buffer, "OPTION ARB_fog_linear;\n");
3663                 break;
3664             case FOG_EXP:
3665                 shader_addline(buffer, "OPTION ARB_fog_exp;\n");
3666                 break;
3667             case FOG_EXP2:
3668                 shader_addline(buffer, "OPTION ARB_fog_exp2;\n");
3669                 break;
3670         }
3671     }
3672
3673     /* For now always declare the temps. At least the Nvidia assembler optimizes completely
3674      * unused temps away(but occupies them for the whole shader if they're used once). Always
3675      * declaring them avoids tricky bookkeeping work
3676      */
3677     shader_addline(buffer, "TEMP TA;\n");      /* Used for modifiers */
3678     shader_addline(buffer, "TEMP TB;\n");      /* Used for modifiers */
3679     shader_addline(buffer, "TEMP TC;\n");      /* Used for modifiers */
3680     if(dcl_td) shader_addline(buffer, "TEMP TD;\n"); /* Used for sRGB writing */
3681     shader_addline(buffer, "PARAM coefdiv = { 0.5, 0.25, 0.125, 0.0625 };\n");
3682     shader_addline(buffer, "PARAM coefmul = { 2, 4, 8, 16 };\n");
3683     shader_addline(buffer, "PARAM ps_helper_const = { 0.0, 1.0, %1.10f, 0.0 };\n", eps);
3684
3685     if (reg_maps->shader_version.major < 2)
3686     {
3687         strcpy(fragcolor, "R0");
3688     }
3689     else
3690     {
3691         if (args->super.srgb_correction)
3692         {
3693             if (shader->u.ps.color0_mov)
3694             {
3695                 sprintf(fragcolor, "R%u", shader->u.ps.color0_reg);
3696             }
3697             else
3698             {
3699                 shader_addline(buffer, "TEMP TMP_COLOR;\n");
3700                 strcpy(fragcolor, "TMP_COLOR");
3701             }
3702         } else {
3703             strcpy(fragcolor, "result.color");
3704         }
3705     }
3706
3707     if(args->super.srgb_correction) {
3708         shader_addline(buffer, "PARAM srgb_consts1 = {%f, %f, %f, %f};\n",
3709                        srgb_mul_low, srgb_cmp, srgb_pow, srgb_mul_high);
3710         shader_addline(buffer, "PARAM srgb_consts2 = {%f, %f, %f, %f};\n",
3711                        srgb_sub_high, 0.0, 0.0, 0.0);
3712     }
3713
3714     /* Base Declarations */
3715     next_local = shader_generate_arb_declarations(shader, reg_maps,
3716             buffer, gl_info, lconst_map, NULL, &priv_ctx);
3717
3718     for (i = 0, map = reg_maps->bumpmat; map; map >>= 1, ++i)
3719     {
3720         unsigned char bump_const;
3721
3722         if (!(map & 1)) continue;
3723
3724         bump_const = compiled->numbumpenvmatconsts;
3725         compiled->bumpenvmatconst[bump_const].const_num = WINED3D_CONST_NUM_UNUSED;
3726         compiled->bumpenvmatconst[bump_const].texunit = i;
3727         compiled->luminanceconst[bump_const].const_num = WINED3D_CONST_NUM_UNUSED;
3728         compiled->luminanceconst[bump_const].texunit = i;
3729
3730         /* We can fit the constants into the constant limit for sure because texbem, texbeml, bem and beml are only supported
3731          * in 1.x shaders, and GL_ARB_fragment_program has a constant limit of 24 constants. So in the worst case we're loading
3732          * 8 shader constants, 8 bump matrices and 8 luminance parameters and are perfectly fine. (No NP2 fixup on bumpmapped
3733          * textures due to conditional NP2 restrictions)
3734          *
3735          * Use local constants to load the bump env parameters, not program.env. This avoids collisions with d3d constants of
3736          * shaders in newer shader models. Since the bump env parameters have to share their space with NP2 fixup constants,
3737          * their location is shader dependent anyway and they cannot be loaded globally.
3738          */
3739         compiled->bumpenvmatconst[bump_const].const_num = next_local++;
3740         shader_addline(buffer, "PARAM bumpenvmat%d = program.local[%d];\n",
3741                        i, compiled->bumpenvmatconst[bump_const].const_num);
3742         compiled->numbumpenvmatconsts = bump_const + 1;
3743
3744         if (!(reg_maps->luminanceparams & (1 << i))) continue;
3745
3746         compiled->luminanceconst[bump_const].const_num = next_local++;
3747         shader_addline(buffer, "PARAM luminance%d = program.local[%d];\n",
3748                        i, compiled->luminanceconst[bump_const].const_num);
3749     }
3750
3751     for(i = 0; i < MAX_CONST_I; i++)
3752     {
3753         compiled->int_consts[i] = WINED3D_CONST_NUM_UNUSED;
3754         if (reg_maps->integer_constants & (1 << i) && priv_ctx.target_version >= NV2)
3755         {
3756             const DWORD *control_values = find_loop_control_values(shader, i);
3757
3758             if(control_values)
3759             {
3760                 shader_addline(buffer, "PARAM I%u = {%u, %u, %u, -1};\n", i,
3761                                 control_values[0], control_values[1], control_values[2]);
3762             }
3763             else
3764             {
3765                 compiled->int_consts[i] = next_local;
3766                 compiled->num_int_consts++;
3767                 shader_addline(buffer, "PARAM I%u = program.local[%u];\n", i, next_local++);
3768             }
3769         }
3770     }
3771
3772     if(reg_maps->vpos || reg_maps->usesdsy)
3773     {
3774         compiled->ycorrection = next_local;
3775         shader_addline(buffer, "PARAM ycorrection = program.local[%u];\n", next_local++);
3776
3777         if(reg_maps->vpos)
3778         {
3779             shader_addline(buffer, "TEMP vpos;\n");
3780             /* ycorrection.x: Backbuffer height(onscreen) or 0(offscreen).
3781              * ycorrection.y: -1.0(onscreen), 1.0(offscreen)
3782              * ycorrection.z: 1.0
3783              * ycorrection.w: 0.0
3784              */
3785             shader_addline(buffer, "MAD vpos, fragment.position, ycorrection.zyww, ycorrection.wxww;\n");
3786             shader_addline(buffer, "FLR vpos.xy, vpos;\n");
3787         }
3788     }
3789     else
3790     {
3791         compiled->ycorrection = WINED3D_CONST_NUM_UNUSED;
3792     }
3793
3794     /* Load constants to fixup NP2 texcoords if there are still free constants left:
3795      * Constants (texture dimensions) for the NP2 fixup are loaded as local program parameters. This will consume
3796      * at most 8 (MAX_FRAGMENT_SAMPLERS / 2) parameters, which is highly unlikely, since the application had to
3797      * use 16 NP2 textures at the same time. In case that we run out of constants the fixup is simply not
3798      * applied / activated. This will probably result in wrong rendering of the texture, but will save us from
3799      * shader compilation errors and the subsequent errors when drawing with this shader. */
3800     if (priv_ctx.cur_ps_args->super.np2_fixup) {
3801         unsigned char cur_fixup_sampler = 0;
3802
3803         struct arb_ps_np2fixup_info* const fixup = priv_ctx.cur_np2fixup_info;
3804         const WORD map = priv_ctx.cur_ps_args->super.np2_fixup;
3805         const UINT max_lconsts = gl_info->limits.arb_ps_local_constants;
3806
3807         fixup->offset = next_local;
3808         fixup->super.active = 0;
3809
3810         for (i = 0; i < MAX_FRAGMENT_SAMPLERS; ++i) {
3811             if (!(map & (1 << i))) continue;
3812
3813             if (fixup->offset + (cur_fixup_sampler >> 1) < max_lconsts) {
3814                 fixup->super.active |= (1 << i);
3815                 fixup->super.idx[i] = cur_fixup_sampler++;
3816             } else {
3817                 FIXME("No free constant found to load NP2 fixup data into shader. "
3818                       "Sampling from this texture will probably look wrong.\n");
3819                 break;
3820             }
3821         }
3822
3823         fixup->super.num_consts = (cur_fixup_sampler + 1) >> 1;
3824         if (fixup->super.num_consts) {
3825             shader_addline(buffer, "PARAM np2fixup[%u] = { program.env[%u..%u] };\n",
3826                            fixup->super.num_consts, fixup->offset, fixup->super.num_consts + fixup->offset - 1);
3827         }
3828     }
3829
3830     if (shader_priv->clipplane_emulation != ~0U && args->clip)
3831     {
3832         shader_addline(buffer, "KIL fragment.texcoord[%u];\n", shader_priv->clipplane_emulation);
3833     }
3834
3835     /* Base Shader Body */
3836     shader_generate_main(shader, buffer, reg_maps, function, &priv_ctx);
3837
3838     if(args->super.srgb_correction) {
3839         arbfp_add_sRGB_correction(buffer, fragcolor, srgbtmp[0], srgbtmp[1], srgbtmp[2], srgbtmp[3],
3840                                   priv_ctx.target_version >= NV2);
3841     }
3842
3843     if(strcmp(fragcolor, "result.color")) {
3844         shader_addline(buffer, "MOV result.color, %s;\n", fragcolor);
3845     }
3846     shader_addline(buffer, "END\n");
3847
3848     /* TODO: change to resource.glObjectHandle or something like that */
3849     GL_EXTCALL(glGenProgramsARB(1, &retval));
3850
3851     TRACE("Creating a hw pixel shader, prg=%d\n", retval);
3852     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, retval));
3853
3854     TRACE("Created hw pixel shader, prg=%d\n", retval);
3855     /* Create the program and check for errors */
3856     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
3857                buffer->bsize, buffer->buffer));
3858     checkGLcall("glProgramStringARB()");
3859
3860     gl_info->gl_ops.gl.p_glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &errPos);
3861     if (errPos != -1)
3862     {
3863         FIXME("HW PixelShader Error at position %d: %s\n\n",
3864               errPos, debugstr_a((const char *)gl_info->gl_ops.gl.p_glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
3865         shader_arb_dump_program_source(buffer->buffer);
3866         retval = 0;
3867     }
3868     else
3869     {
3870         GLint native;
3871
3872         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
3873         checkGLcall("glGetProgramivARB()");
3874         if (!native) WARN("Program exceeds native resource limits.\n");
3875     }
3876
3877     /* Load immediate constants */
3878     if (lconst_map)
3879     {
3880         LIST_FOR_EACH_ENTRY(lconst, &shader->constantsF, struct wined3d_shader_lconst, entry)
3881         {
3882             const float *value = (const float *)lconst->value;
3883             GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, lconst_map[lconst->idx], value));
3884             checkGLcall("glProgramLocalParameter4fvARB");
3885         }
3886         HeapFree(GetProcessHeap(), 0, lconst_map);
3887     }
3888
3889     return retval;
3890 }
3891
3892 static int compare_sig(const struct wined3d_shader_signature_element *sig1, const struct wined3d_shader_signature_element *sig2)
3893 {
3894     unsigned int i;
3895     int ret;
3896
3897     for(i = 0; i < MAX_REG_INPUT; i++)
3898     {
3899         if (!sig1[i].semantic_name || !sig2[i].semantic_name)
3900         {
3901             /* Compare pointers, not contents. One string is NULL(element does not exist), the other one is not NULL */
3902             if(sig1[i].semantic_name != sig2[i].semantic_name) return sig1[i].semantic_name < sig2[i].semantic_name ? -1 : 1;
3903             continue;
3904         }
3905
3906         if ((ret = strcmp(sig1[i].semantic_name, sig2[i].semantic_name))) return ret;
3907         if(sig1[i].semantic_idx    != sig2[i].semantic_idx)    return sig1[i].semantic_idx    < sig2[i].semantic_idx    ? -1 : 1;
3908         if(sig1[i].sysval_semantic != sig2[i].sysval_semantic) return sig1[i].sysval_semantic < sig2[i].sysval_semantic ? -1 : 1;
3909         if(sig1[i].component_type  != sig2[i].component_type)  return sig1[i].component_type  < sig2[i].component_type  ? -1 : 1;
3910         if(sig1[i].register_idx    != sig2[i].register_idx)    return sig1[i].register_idx    < sig2[i].register_idx    ? -1 : 1;
3911         if(sig1[i].mask            != sig2[i].mask)            return sig1[i].mask            < sig2[i].mask            ? -1 : 1;
3912     }
3913     return 0;
3914 }
3915
3916 static struct wined3d_shader_signature_element *clone_sig(const struct wined3d_shader_signature_element *sig)
3917 {
3918     struct wined3d_shader_signature_element *new;
3919     int i;
3920     char *name;
3921
3922     new = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*new) * MAX_REG_INPUT);
3923     for(i = 0; i < MAX_REG_INPUT; i++)
3924     {
3925         if (!sig[i].semantic_name) continue;
3926
3927         new[i] = sig[i];
3928         /* Clone the semantic string */
3929         name = HeapAlloc(GetProcessHeap(), 0, strlen(sig[i].semantic_name) + 1);
3930         strcpy(name, sig[i].semantic_name);
3931         new[i].semantic_name = name;
3932     }
3933     return new;
3934 }
3935
3936 static DWORD find_input_signature(struct shader_arb_priv *priv, const struct wined3d_shader_signature_element *sig)
3937 {
3938     struct wine_rb_entry *entry = wine_rb_get(&priv->signature_tree, sig);
3939     struct ps_signature *found_sig;
3940
3941     if (entry)
3942     {
3943         found_sig = WINE_RB_ENTRY_VALUE(entry, struct ps_signature, entry);
3944         TRACE("Found existing signature %u\n", found_sig->idx);
3945         return found_sig->idx;
3946     }
3947     found_sig = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*sig));
3948     found_sig->sig = clone_sig(sig);
3949     found_sig->idx = priv->ps_sig_number++;
3950     TRACE("New signature stored and assigned number %u\n", found_sig->idx);
3951     if(wine_rb_put(&priv->signature_tree, sig, &found_sig->entry) == -1)
3952     {
3953         ERR("Failed to insert program entry.\n");
3954     }
3955     return found_sig->idx;
3956 }
3957
3958 static void init_output_registers(const struct wined3d_shader *shader, DWORD sig_num,
3959         struct shader_arb_ctx_priv *priv_ctx, struct arb_vs_compiled_shader *compiled)
3960 {
3961     unsigned int i, j;
3962     static const char * const texcoords[8] =
3963     {
3964         "result.texcoord[0]", "result.texcoord[1]", "result.texcoord[2]", "result.texcoord[3]",
3965         "result.texcoord[4]", "result.texcoord[5]", "result.texcoord[6]", "result.texcoord[7]"
3966     };
3967     struct wined3d_device *device = shader->device;
3968     const struct wined3d_shader_signature_element *sig;
3969     const char *semantic_name;
3970     DWORD semantic_idx, reg_idx;
3971
3972     /* Write generic input varyings 0 to 7 to result.texcoord[], varying 8 to result.color.primary
3973      * and varying 9 to result.color.secondary
3974      */
3975     static const char * const decl_idx_to_string[MAX_REG_INPUT] =
3976     {
3977         "result.texcoord[0]", "result.texcoord[1]", "result.texcoord[2]", "result.texcoord[3]",
3978         "result.texcoord[4]", "result.texcoord[5]", "result.texcoord[6]", "result.texcoord[7]",
3979         "result.color.primary", "result.color.secondary"
3980     };
3981
3982     if(sig_num == ~0)
3983     {
3984         TRACE("Pixel shader uses builtin varyings\n");
3985         /* Map builtins to builtins */
3986         for(i = 0; i < 8; i++)
3987         {
3988             priv_ctx->texcrd_output[i] = texcoords[i];
3989         }
3990         priv_ctx->color_output[0] = "result.color.primary";
3991         priv_ctx->color_output[1] = "result.color.secondary";
3992         priv_ctx->fog_output = "TMP_FOGCOORD";
3993
3994         /* Map declared regs to builtins. Use "TA" to /dev/null unread output */
3995         for (i = 0; i < (sizeof(shader->output_signature) / sizeof(*shader->output_signature)); ++i)
3996         {
3997             semantic_name = shader->output_signature[i].semantic_name;
3998             if (!semantic_name) continue;
3999
4000             if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_POSITION))
4001             {
4002                 TRACE("o%u is TMP_OUT\n", i);
4003                 if (!shader->output_signature[i].semantic_idx) priv_ctx->vs_output[i] = "TMP_OUT";
4004                 else priv_ctx->vs_output[i] = "TA";
4005             }
4006             else if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_PSIZE))
4007             {
4008                 TRACE("o%u is result.pointsize\n", i);
4009                 if (!shader->output_signature[i].semantic_idx) priv_ctx->vs_output[i] = "result.pointsize";
4010                 else priv_ctx->vs_output[i] = "TA";
4011             }
4012             else if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_COLOR))
4013             {
4014                 TRACE("o%u is result.color.?, idx %u\n", i, shader->output_signature[i].semantic_idx);
4015                 if (!shader->output_signature[i].semantic_idx)
4016                     priv_ctx->vs_output[i] = "result.color.primary";
4017                 else if (shader->output_signature[i].semantic_idx == 1)
4018                     priv_ctx->vs_output[i] = "result.color.secondary";
4019                 else priv_ctx->vs_output[i] = "TA";
4020             }
4021             else if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_TEXCOORD))
4022             {
4023                 TRACE("o%u is %s\n", i, texcoords[shader->output_signature[i].semantic_idx]);
4024                 if (shader->output_signature[i].semantic_idx >= 8) priv_ctx->vs_output[i] = "TA";
4025                 else priv_ctx->vs_output[i] = texcoords[shader->output_signature[i].semantic_idx];
4026             }
4027             else if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_FOG))
4028             {
4029                 TRACE("o%u is result.fogcoord\n", i);
4030                 if (shader->output_signature[i].semantic_idx > 0) priv_ctx->vs_output[i] = "TA";
4031                 else priv_ctx->vs_output[i] = "result.fogcoord";
4032             }
4033             else
4034             {
4035                 priv_ctx->vs_output[i] = "TA";
4036             }
4037         }
4038         return;
4039     }
4040
4041     /* Instead of searching for the signature in the signature list, read the one from the current pixel shader.
4042      * Its maybe not the shader where the signature came from, but it is the same signature and faster to find
4043      */
4044     sig = device->stateBlock->state.pixel_shader->input_signature;
4045     TRACE("Pixel shader uses declared varyings\n");
4046
4047     /* Map builtin to declared. /dev/null the results by default to the TA temp reg */
4048     for(i = 0; i < 8; i++)
4049     {
4050         priv_ctx->texcrd_output[i] = "TA";
4051     }
4052     priv_ctx->color_output[0] = "TA";
4053     priv_ctx->color_output[1] = "TA";
4054     priv_ctx->fog_output = "TA";
4055
4056     for(i = 0; i < MAX_REG_INPUT; i++)
4057     {
4058         semantic_name = sig[i].semantic_name;
4059         semantic_idx = sig[i].semantic_idx;
4060         reg_idx = sig[i].register_idx;
4061         if (!semantic_name) continue;
4062
4063         /* If a declared input register is not written by builtin arguments, don't write to it.
4064          * GL_NV_vertex_program makes sure the input defaults to 0.0, which is correct with D3D
4065          *
4066          * Don't care about POSITION and PSIZE here - this is a builtin vertex shader, position goes
4067          * to TMP_OUT in any case
4068          */
4069         if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_TEXCOORD))
4070         {
4071             if (semantic_idx < 8)
4072                 priv_ctx->texcrd_output[semantic_idx] = decl_idx_to_string[reg_idx];
4073         }
4074         else if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_COLOR))
4075         {
4076             if (semantic_idx < 2)
4077                 priv_ctx->color_output[semantic_idx] = decl_idx_to_string[reg_idx];
4078         }
4079         else if(shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_FOG))
4080         {
4081             if (!semantic_idx)
4082                 priv_ctx->fog_output = decl_idx_to_string[reg_idx];
4083         }
4084         else
4085         {
4086             continue;
4087         }
4088
4089         if (!strcmp(decl_idx_to_string[reg_idx], "result.color.primary")
4090                 || !strcmp(decl_idx_to_string[reg_idx], "result.color.secondary"))
4091         {
4092             compiled->need_color_unclamp = TRUE;
4093         }
4094     }
4095
4096     /* Map declared to declared */
4097     for (i = 0; i < (sizeof(shader->output_signature) / sizeof(*shader->output_signature)); ++i)
4098     {
4099         /* Write unread output to TA to throw them away */
4100         priv_ctx->vs_output[i] = "TA";
4101         semantic_name = shader->output_signature[i].semantic_name;
4102         if (!semantic_name) continue;
4103
4104         if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_POSITION)
4105                 && !shader->output_signature[i].semantic_idx)
4106         {
4107             priv_ctx->vs_output[i] = "TMP_OUT";
4108             continue;
4109         }
4110         else if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_PSIZE)
4111                 && !shader->output_signature[i].semantic_idx)
4112         {
4113             priv_ctx->vs_output[i] = "result.pointsize";
4114             continue;
4115         }
4116
4117         for(j = 0; j < MAX_REG_INPUT; j++)
4118         {
4119             if (!sig[j].semantic_name) continue;
4120
4121             if (!strcmp(sig[j].semantic_name, semantic_name)
4122                     && sig[j].semantic_idx == shader->output_signature[i].semantic_idx)
4123             {
4124                 priv_ctx->vs_output[i] = decl_idx_to_string[sig[j].register_idx];
4125
4126                 if (!strcmp(priv_ctx->vs_output[i], "result.color.primary")
4127                         || !strcmp(priv_ctx->vs_output[i], "result.color.secondary"))
4128                 {
4129                     compiled->need_color_unclamp = TRUE;
4130                 }
4131             }
4132         }
4133     }
4134 }
4135
4136 /* GL locking is done by the caller */
4137 static GLuint shader_arb_generate_vshader(const struct wined3d_shader *shader,
4138         const struct wined3d_gl_info *gl_info, struct wined3d_shader_buffer *buffer,
4139         const struct arb_vs_compile_args *args, struct arb_vs_compiled_shader *compiled)
4140 {
4141     const struct arb_vshader_private *shader_data = shader->backend_data;
4142     const struct wined3d_shader_reg_maps *reg_maps = &shader->reg_maps;
4143     const struct wined3d_shader_lconst *lconst;
4144     const DWORD *function = shader->function;
4145     GLuint ret;
4146     DWORD next_local, *lconst_map = local_const_mapping(shader);
4147     struct shader_arb_ctx_priv priv_ctx;
4148     unsigned int i;
4149     GLint errPos;
4150
4151     memset(&priv_ctx, 0, sizeof(priv_ctx));
4152     priv_ctx.cur_vs_args = args;
4153     list_init(&priv_ctx.control_frames);
4154     init_output_registers(shader, args->ps_signature, &priv_ctx, compiled);
4155
4156     /*  Create the hw ARB shader */
4157     shader_addline(buffer, "!!ARBvp1.0\n");
4158
4159     /* Always enable the NV extension if available. Unlike fragment shaders, there is no
4160      * mesurable performance penalty, and we can always make use of it for clipplanes.
4161      */
4162     if (gl_info->supported[NV_VERTEX_PROGRAM3])
4163     {
4164         shader_addline(buffer, "OPTION NV_vertex_program3;\n");
4165         priv_ctx.target_version = NV3;
4166         shader_addline(buffer, "ADDRESS aL;\n");
4167     }
4168     else if (gl_info->supported[NV_VERTEX_PROGRAM2_OPTION])
4169     {
4170         shader_addline(buffer, "OPTION NV_vertex_program2;\n");
4171         priv_ctx.target_version = NV2;
4172         shader_addline(buffer, "ADDRESS aL;\n");
4173     } else {
4174         priv_ctx.target_version = ARB;
4175     }
4176
4177     shader_addline(buffer, "TEMP TMP_OUT;\n");
4178     if (reg_maps->fog)
4179         shader_addline(buffer, "TEMP TMP_FOGCOORD;\n");
4180     if (need_helper_const(shader_data, reg_maps, gl_info))
4181     {
4182         shader_addline(buffer, "PARAM helper_const = { 0.0, 1.0, 2.0, %1.10f};\n", eps);
4183     }
4184     if (need_rel_addr_const(shader_data, reg_maps, gl_info))
4185     {
4186         shader_addline(buffer, "PARAM rel_addr_const = { 0.5, %d.0, 0.0, 0.0 };\n", shader_data->rel_offset);
4187         shader_addline(buffer, "TEMP A0_SHADOW;\n");
4188     }
4189
4190     shader_addline(buffer, "TEMP TA;\n");
4191     shader_addline(buffer, "TEMP TB;\n");
4192
4193     /* Base Declarations */
4194     next_local = shader_generate_arb_declarations(shader, reg_maps, buffer,
4195             gl_info, lconst_map, &priv_ctx.vs_clipplanes, &priv_ctx);
4196
4197     for(i = 0; i < MAX_CONST_I; i++)
4198     {
4199         compiled->int_consts[i] = WINED3D_CONST_NUM_UNUSED;
4200         if(reg_maps->integer_constants & (1 << i) && priv_ctx.target_version >= NV2)
4201         {
4202             const DWORD *control_values = find_loop_control_values(shader, i);
4203
4204             if(control_values)
4205             {
4206                 shader_addline(buffer, "PARAM I%u = {%u, %u, %u, -1};\n", i,
4207                                 control_values[0], control_values[1], control_values[2]);
4208             }
4209             else
4210             {
4211                 compiled->int_consts[i] = next_local;
4212                 compiled->num_int_consts++;
4213                 shader_addline(buffer, "PARAM I%u = program.local[%u];\n", i, next_local++);
4214             }
4215         }
4216     }
4217
4218     /* We need a constant to fixup the final position */
4219     shader_addline(buffer, "PARAM posFixup = program.local[%u];\n", next_local);
4220     compiled->pos_fixup = next_local++;
4221
4222     /* Initialize output parameters. GL_ARB_vertex_program does not require special initialization values
4223      * for output parameters. D3D in theory does not do that either, but some applications depend on a
4224      * proper initialization of the secondary color, and programs using the fixed function pipeline without
4225      * a replacement shader depend on the texcoord.w being set properly.
4226      *
4227      * GL_NV_vertex_program defines that all output values are initialized to {0.0, 0.0, 0.0, 1.0}. This
4228      * assertion is in effect even when using GL_ARB_vertex_program without any NV specific additions. So
4229      * skip this if NV_vertex_program is supported. Otherwise, initialize the secondary color. For the tex-
4230      * coords, we have a flag in the opengl caps. Many cards do not require the texcoord being set, and
4231      * this can eat a number of instructions, so skip it unless this cap is set as well
4232      */
4233     if (!gl_info->supported[NV_VERTEX_PROGRAM])
4234     {
4235         struct wined3d_device *device = shader->device;
4236         const char *color_init = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_0001);
4237         shader_addline(buffer, "MOV result.color.secondary, %s;\n", color_init);
4238
4239         if (gl_info->quirks & WINED3D_QUIRK_SET_TEXCOORD_W && !device->frag_pipe->ffp_proj_control)
4240         {
4241             int i;
4242             const char *one = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_ONE);
4243             for(i = 0; i < min(8, MAX_REG_TEXCRD); i++)
4244             {
4245                 if (reg_maps->texcoord_mask[i] && reg_maps->texcoord_mask[i] != WINED3DSP_WRITEMASK_ALL)
4246                     shader_addline(buffer, "MOV result.texcoord[%u].w, %s\n", i, one);
4247             }
4248         }
4249     }
4250
4251     /* The shader starts with the main function */
4252     priv_ctx.in_main_func = TRUE;
4253     /* Base Shader Body */
4254     shader_generate_main(shader, buffer, reg_maps, function, &priv_ctx);
4255
4256     if (!priv_ctx.footer_written) vshader_add_footer(&priv_ctx,
4257             shader_data, args, reg_maps, gl_info, buffer);
4258
4259     shader_addline(buffer, "END\n");
4260
4261     /* TODO: change to resource.glObjectHandle or something like that */
4262     GL_EXTCALL(glGenProgramsARB(1, &ret));
4263
4264     TRACE("Creating a hw vertex shader, prg=%d\n", ret);
4265     GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, ret));
4266
4267     TRACE("Created hw vertex shader, prg=%d\n", ret);
4268     /* Create the program and check for errors */
4269     GL_EXTCALL(glProgramStringARB(GL_VERTEX_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
4270                buffer->bsize, buffer->buffer));
4271     checkGLcall("glProgramStringARB()");
4272
4273     gl_info->gl_ops.gl.p_glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &errPos);
4274     if (errPos != -1)
4275     {
4276         FIXME("HW VertexShader Error at position %d: %s\n\n",
4277               errPos, debugstr_a((const char *)gl_info->gl_ops.gl.p_glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
4278         shader_arb_dump_program_source(buffer->buffer);
4279         ret = -1;
4280     }
4281     else
4282     {
4283         GLint native;
4284
4285         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
4286         checkGLcall("glGetProgramivARB()");
4287         if (!native) WARN("Program exceeds native resource limits.\n");
4288
4289         /* Load immediate constants */
4290         if (lconst_map)
4291         {
4292             LIST_FOR_EACH_ENTRY(lconst, &shader->constantsF, struct wined3d_shader_lconst, entry)
4293             {
4294                 const float *value = (const float *)lconst->value;
4295                 GL_EXTCALL(glProgramLocalParameter4fvARB(GL_VERTEX_PROGRAM_ARB, lconst_map[lconst->idx], value));
4296             }
4297         }
4298     }
4299     HeapFree(GetProcessHeap(), 0, lconst_map);
4300
4301     return ret;
4302 }
4303
4304 /* GL locking is done by the caller */
4305 static struct arb_ps_compiled_shader *find_arb_pshader(struct wined3d_shader *shader,
4306         const struct arb_ps_compile_args *args)
4307 {
4308     struct wined3d_device *device = shader->device;
4309     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4310     UINT i;
4311     DWORD new_size;
4312     struct arb_ps_compiled_shader *new_array;
4313     struct wined3d_shader_buffer buffer;
4314     struct arb_pshader_private *shader_data;
4315     GLuint ret;
4316
4317     if (!shader->backend_data)
4318     {
4319         struct shader_arb_priv *priv = device->shader_priv;
4320
4321         shader->backend_data = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data));
4322         shader_data = shader->backend_data;
4323         shader_data->clamp_consts = shader->reg_maps.shader_version.major == 1;
4324
4325         if (shader->reg_maps.shader_version.major < 3)
4326             shader_data->input_signature_idx = ~0;
4327         else
4328             shader_data->input_signature_idx = find_input_signature(priv, shader->input_signature);
4329
4330         TRACE("Shader got assigned input signature index %u\n", shader_data->input_signature_idx);
4331
4332         if (!device->vs_clipping)
4333             shader_data->clipplane_emulation = shader_find_free_input_register(&shader->reg_maps,
4334                     gl_info->limits.texture_stages - 1);
4335         else
4336             shader_data->clipplane_emulation = ~0U;
4337     }
4338     shader_data = shader->backend_data;
4339
4340     /* Usually we have very few GL shaders for each d3d shader(just 1 or maybe 2),
4341      * so a linear search is more performant than a hashmap or a binary search
4342      * (cache coherency etc)
4343      */
4344     for (i = 0; i < shader_data->num_gl_shaders; ++i)
4345     {
4346         if (!memcmp(&shader_data->gl_shaders[i].args, args, sizeof(*args)))
4347             return &shader_data->gl_shaders[i];
4348     }
4349
4350     TRACE("No matching GL shader found, compiling a new shader\n");
4351     if(shader_data->shader_array_size == shader_data->num_gl_shaders) {
4352         if (shader_data->num_gl_shaders)
4353         {
4354             new_size = shader_data->shader_array_size + max(1, shader_data->shader_array_size / 2);
4355             new_array = HeapReAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, shader_data->gl_shaders,
4356                                     new_size * sizeof(*shader_data->gl_shaders));
4357         } else {
4358             new_array = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data->gl_shaders));
4359             new_size = 1;
4360         }
4361
4362         if(!new_array) {
4363             ERR("Out of memory\n");
4364             return 0;
4365         }
4366         shader_data->gl_shaders = new_array;
4367         shader_data->shader_array_size = new_size;
4368     }
4369
4370     shader_data->gl_shaders[shader_data->num_gl_shaders].args = *args;
4371
4372     pixelshader_update_samplers(&shader->reg_maps, device->stateBlock->state.textures);
4373
4374     if (!shader_buffer_init(&buffer))
4375     {
4376         ERR("Failed to initialize shader buffer.\n");
4377         return 0;
4378     }
4379
4380     ret = shader_arb_generate_pshader(shader, gl_info, &buffer, args,
4381             &shader_data->gl_shaders[shader_data->num_gl_shaders]);
4382     shader_buffer_free(&buffer);
4383     shader_data->gl_shaders[shader_data->num_gl_shaders].prgId = ret;
4384
4385     return &shader_data->gl_shaders[shader_data->num_gl_shaders++];
4386 }
4387
4388 static inline BOOL vs_args_equal(const struct arb_vs_compile_args *stored, const struct arb_vs_compile_args *new,
4389                                  const DWORD use_map, BOOL skip_int) {
4390     if((stored->super.swizzle_map & use_map) != new->super.swizzle_map) return FALSE;
4391     if(stored->super.clip_enabled != new->super.clip_enabled) return FALSE;
4392     if(stored->super.fog_src != new->super.fog_src) return FALSE;
4393     if(stored->clip.boolclip_compare != new->clip.boolclip_compare) return FALSE;
4394     if(stored->ps_signature != new->ps_signature) return FALSE;
4395     if(stored->vertex.samplers_compare != new->vertex.samplers_compare) return FALSE;
4396     if(skip_int) return TRUE;
4397
4398     return !memcmp(stored->loop_ctrl, new->loop_ctrl, sizeof(stored->loop_ctrl));
4399 }
4400
4401 static struct arb_vs_compiled_shader *find_arb_vshader(struct wined3d_shader *shader,
4402         const struct arb_vs_compile_args *args)
4403 {
4404     struct wined3d_device *device = shader->device;
4405     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4406     DWORD use_map = device->strided_streams.use_map;
4407     UINT i;
4408     DWORD new_size;
4409     struct arb_vs_compiled_shader *new_array;
4410     struct wined3d_shader_buffer buffer;
4411     struct arb_vshader_private *shader_data;
4412     GLuint ret;
4413
4414     if (!shader->backend_data)
4415     {
4416         const struct wined3d_shader_reg_maps *reg_maps = &shader->reg_maps;
4417
4418         shader->backend_data = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data));
4419         shader_data = shader->backend_data;
4420
4421         if ((gl_info->quirks & WINED3D_QUIRK_ARB_VS_OFFSET_LIMIT)
4422                 && reg_maps->min_rel_offset <= reg_maps->max_rel_offset)
4423         {
4424             if (reg_maps->max_rel_offset - reg_maps->min_rel_offset > 127)
4425             {
4426                 FIXME("The difference between the minimum and maximum relative offset is > 127.\n");
4427                 FIXME("Which this OpenGL implementation does not support. Try using GLSL.\n");
4428                 FIXME("Min: %u, Max: %u.\n", reg_maps->min_rel_offset, reg_maps->max_rel_offset);
4429             }
4430             else if (reg_maps->max_rel_offset - reg_maps->min_rel_offset > 63)
4431                 shader_data->rel_offset = reg_maps->min_rel_offset + 63;
4432             else if (reg_maps->max_rel_offset > 63)
4433                 shader_data->rel_offset = reg_maps->min_rel_offset;
4434         }
4435     }
4436     shader_data = shader->backend_data;
4437
4438     /* Usually we have very few GL shaders for each d3d shader(just 1 or maybe 2),
4439      * so a linear search is more performant than a hashmap or a binary search
4440      * (cache coherency etc)
4441      */
4442     for(i = 0; i < shader_data->num_gl_shaders; i++) {
4443         if (vs_args_equal(&shader_data->gl_shaders[i].args, args,
4444                 use_map, gl_info->supported[NV_VERTEX_PROGRAM2_OPTION]))
4445         {
4446             return &shader_data->gl_shaders[i];
4447         }
4448     }
4449
4450     TRACE("No matching GL shader found, compiling a new shader\n");
4451
4452     if(shader_data->shader_array_size == shader_data->num_gl_shaders) {
4453         if (shader_data->num_gl_shaders)
4454         {
4455             new_size = shader_data->shader_array_size + max(1, shader_data->shader_array_size / 2);
4456             new_array = HeapReAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, shader_data->gl_shaders,
4457                                     new_size * sizeof(*shader_data->gl_shaders));
4458         } else {
4459             new_array = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data->gl_shaders));
4460             new_size = 1;
4461         }
4462
4463         if(!new_array) {
4464             ERR("Out of memory\n");
4465             return 0;
4466         }
4467         shader_data->gl_shaders = new_array;
4468         shader_data->shader_array_size = new_size;
4469     }
4470
4471     shader_data->gl_shaders[shader_data->num_gl_shaders].args = *args;
4472
4473     if (!shader_buffer_init(&buffer))
4474     {
4475         ERR("Failed to initialize shader buffer.\n");
4476         return 0;
4477     }
4478
4479     ret = shader_arb_generate_vshader(shader, gl_info, &buffer, args,
4480             &shader_data->gl_shaders[shader_data->num_gl_shaders]);
4481     shader_buffer_free(&buffer);
4482     shader_data->gl_shaders[shader_data->num_gl_shaders].prgId = ret;
4483
4484     return &shader_data->gl_shaders[shader_data->num_gl_shaders++];
4485 }
4486
4487 static void find_arb_ps_compile_args(const struct wined3d_state *state,
4488         const struct wined3d_shader *shader, struct arb_ps_compile_args *args)
4489 {
4490     struct wined3d_device *device = shader->device;
4491     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4492     int i;
4493     WORD int_skip;
4494
4495     find_ps_compile_args(state, shader, &args->super);
4496
4497     /* This forces all local boolean constants to 1 to make them stateblock independent */
4498     args->bools = shader->reg_maps.local_bool_consts;
4499
4500     for(i = 0; i < MAX_CONST_B; i++)
4501     {
4502         if (state->ps_consts_b[i])
4503             args->bools |= ( 1 << i);
4504     }
4505
4506     /* Only enable the clip plane emulation KIL if at least one clipplane is enabled. The KIL instruction
4507      * is quite expensive because it forces the driver to disable early Z discards. It is cheaper to
4508      * duplicate the shader than have a no-op KIL instruction in every shader
4509      */
4510     if (!device->vs_clipping && use_vs(state)
4511             && state->render_states[WINED3D_RS_CLIPPING]
4512             && state->render_states[WINED3D_RS_CLIPPLANEENABLE])
4513         args->clip = 1;
4514     else
4515         args->clip = 0;
4516
4517     /* Skip if unused or local, or supported natively */
4518     int_skip = ~shader->reg_maps.integer_constants | shader->reg_maps.local_int_consts;
4519     if (int_skip == 0xffff || gl_info->supported[NV_FRAGMENT_PROGRAM_OPTION])
4520     {
4521         memset(args->loop_ctrl, 0, sizeof(args->loop_ctrl));
4522         return;
4523     }
4524
4525     for(i = 0; i < MAX_CONST_I; i++)
4526     {
4527         if(int_skip & (1 << i))
4528         {
4529             args->loop_ctrl[i][0] = 0;
4530             args->loop_ctrl[i][1] = 0;
4531             args->loop_ctrl[i][2] = 0;
4532         }
4533         else
4534         {
4535             args->loop_ctrl[i][0] = state->ps_consts_i[i * 4];
4536             args->loop_ctrl[i][1] = state->ps_consts_i[i * 4 + 1];
4537             args->loop_ctrl[i][2] = state->ps_consts_i[i * 4 + 2];
4538         }
4539     }
4540 }
4541
4542 static void find_arb_vs_compile_args(const struct wined3d_state *state,
4543         const struct wined3d_shader *shader, struct arb_vs_compile_args *args)
4544 {
4545     struct wined3d_device *device = shader->device;
4546     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4547     int i;
4548     WORD int_skip;
4549
4550     find_vs_compile_args(state, shader, &args->super);
4551
4552     args->clip.boolclip_compare = 0;
4553     if (use_ps(state))
4554     {
4555         const struct wined3d_shader *ps = state->pixel_shader;
4556         const struct arb_pshader_private *shader_priv = ps->backend_data;
4557         args->ps_signature = shader_priv->input_signature_idx;
4558
4559         args->clip.boolclip.clip_texcoord = shader_priv->clipplane_emulation + 1;
4560     }
4561     else
4562     {
4563         args->ps_signature = ~0;
4564         if (!device->vs_clipping && device->adapter->fragment_pipe == &arbfp_fragment_pipeline)
4565         {
4566             args->clip.boolclip.clip_texcoord = ffp_clip_emul(state) ? gl_info->limits.texture_stages : 0;
4567         }
4568         /* Otherwise: Setting boolclip_compare set clip_texcoord to 0 */
4569     }
4570
4571     if (args->clip.boolclip.clip_texcoord)
4572     {
4573         if (state->render_states[WINED3D_RS_CLIPPING])
4574             args->clip.boolclip.clipplane_mask = (unsigned char)state->render_states[WINED3D_RS_CLIPPLANEENABLE];
4575         /* clipplane_mask was set to 0 by setting boolclip_compare to 0 */
4576     }
4577
4578     /* This forces all local boolean constants to 1 to make them stateblock independent */
4579     args->clip.boolclip.bools = shader->reg_maps.local_bool_consts;
4580     /* TODO: Figure out if it would be better to store bool constants as bitmasks in the stateblock */
4581     for(i = 0; i < MAX_CONST_B; i++)
4582     {
4583         if (state->vs_consts_b[i])
4584             args->clip.boolclip.bools |= ( 1 << i);
4585     }
4586
4587     args->vertex.samplers[0] = device->texUnitMap[MAX_FRAGMENT_SAMPLERS + 0];
4588     args->vertex.samplers[1] = device->texUnitMap[MAX_FRAGMENT_SAMPLERS + 1];
4589     args->vertex.samplers[2] = device->texUnitMap[MAX_FRAGMENT_SAMPLERS + 2];
4590     args->vertex.samplers[3] = 0;
4591
4592     /* Skip if unused or local */
4593     int_skip = ~shader->reg_maps.integer_constants | shader->reg_maps.local_int_consts;
4594     /* This is about flow control, not clipping. */
4595     if (int_skip == 0xffff || gl_info->supported[NV_VERTEX_PROGRAM2_OPTION])
4596     {
4597         memset(args->loop_ctrl, 0, sizeof(args->loop_ctrl));
4598         return;
4599     }
4600
4601     for(i = 0; i < MAX_CONST_I; i++)
4602     {
4603         if(int_skip & (1 << i))
4604         {
4605             args->loop_ctrl[i][0] = 0;
4606             args->loop_ctrl[i][1] = 0;
4607             args->loop_ctrl[i][2] = 0;
4608         }
4609         else
4610         {
4611             args->loop_ctrl[i][0] = state->vs_consts_i[i * 4];
4612             args->loop_ctrl[i][1] = state->vs_consts_i[i * 4 + 1];
4613             args->loop_ctrl[i][2] = state->vs_consts_i[i * 4 + 2];
4614         }
4615     }
4616 }
4617
4618 /* GL locking is done by the caller */
4619 static void shader_arb_select(const struct wined3d_context *context, BOOL usePS, BOOL useVS)
4620 {
4621     struct wined3d_device *device = context->swapchain->device;
4622     struct shader_arb_priv *priv = device->shader_priv;
4623     const struct wined3d_gl_info *gl_info = context->gl_info;
4624     const struct wined3d_state *state = &device->stateBlock->state;
4625     int i;
4626
4627     /* Deal with pixel shaders first so the vertex shader arg function has the input signature ready */
4628     if (usePS)
4629     {
4630         struct wined3d_shader *ps = state->pixel_shader;
4631         struct arb_ps_compile_args compile_args;
4632         struct arb_ps_compiled_shader *compiled;
4633
4634         TRACE("Using pixel shader %p.\n", ps);
4635         find_arb_ps_compile_args(state, ps, &compile_args);
4636         compiled = find_arb_pshader(ps, &compile_args);
4637         priv->current_fprogram_id = compiled->prgId;
4638         priv->compiled_fprog = compiled;
4639
4640         /* Bind the fragment program */
4641         GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, priv->current_fprogram_id));
4642         checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, priv->current_fprogram_id);");
4643
4644         if (!priv->use_arbfp_fixed_func)
4645         {
4646             /* Enable OpenGL fragment programs. */
4647             gl_info->gl_ops.gl.p_glEnable(GL_FRAGMENT_PROGRAM_ARB);
4648             checkGLcall("glEnable(GL_FRAGMENT_PROGRAM_ARB);");
4649         }
4650         TRACE("(%p) : Bound fragment program %u and enabled GL_FRAGMENT_PROGRAM_ARB\n",
4651                 device, priv->current_fprogram_id);
4652
4653         /* Pixel Shader 1.x constants are clamped to [-1;1], Pixel Shader 2.0 constants are not. If switching between
4654          * a 1.x and newer shader, reload the first 8 constants
4655          */
4656         if (priv->last_ps_const_clamped != ((struct arb_pshader_private *)ps->backend_data)->clamp_consts)
4657         {
4658             priv->last_ps_const_clamped = ((struct arb_pshader_private *)ps->backend_data)->clamp_consts;
4659             priv->highest_dirty_ps_const = max(priv->highest_dirty_ps_const, 8);
4660             for(i = 0; i < 8; i++)
4661             {
4662                 priv->pshader_const_dirty[i] = 1;
4663             }
4664             /* Also takes care of loading local constants */
4665             shader_arb_load_constants(context, TRUE, FALSE);
4666         }
4667         else
4668         {
4669             UINT rt_height = state->fb->render_targets[0]->resource.height;
4670             shader_arb_ps_local_constants(compiled, context, state, rt_height);
4671         }
4672
4673         /* Force constant reloading for the NP2 fixup (see comment in shader_glsl_select for more info) */
4674         if (compiled->np2fixup_info.super.active)
4675             shader_arb_load_np2fixup_constants(priv, gl_info, state);
4676     }
4677     else if (gl_info->supported[ARB_FRAGMENT_PROGRAM] && !priv->use_arbfp_fixed_func)
4678     {
4679         /* Disable only if we're not using arbfp fixed function fragment processing. If this is used,
4680         * keep GL_FRAGMENT_PROGRAM_ARB enabled, and the fixed function pipeline will bind the fixed function
4681         * replacement shader
4682         */
4683         gl_info->gl_ops.gl.p_glDisable(GL_FRAGMENT_PROGRAM_ARB);
4684         checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
4685         priv->current_fprogram_id = 0;
4686     }
4687
4688     if (useVS)
4689     {
4690         struct wined3d_shader *vs = state->vertex_shader;
4691         struct arb_vs_compile_args compile_args;
4692         struct arb_vs_compiled_shader *compiled;
4693
4694         TRACE("Using vertex shader %p\n", vs);
4695         find_arb_vs_compile_args(state, vs, &compile_args);
4696         compiled = find_arb_vshader(vs, &compile_args);
4697         priv->current_vprogram_id = compiled->prgId;
4698         priv->compiled_vprog = compiled;
4699
4700         /* Bind the vertex program */
4701         GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->current_vprogram_id));
4702         checkGLcall("glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->current_vprogram_id);");
4703
4704         /* Enable OpenGL vertex programs */
4705         gl_info->gl_ops.gl.p_glEnable(GL_VERTEX_PROGRAM_ARB);
4706         checkGLcall("glEnable(GL_VERTEX_PROGRAM_ARB);");
4707         TRACE("(%p) : Bound vertex program %u and enabled GL_VERTEX_PROGRAM_ARB\n", device, priv->current_vprogram_id);
4708         shader_arb_vs_local_constants(compiled, context, state);
4709
4710         if(priv->last_vs_color_unclamp != compiled->need_color_unclamp) {
4711             priv->last_vs_color_unclamp = compiled->need_color_unclamp;
4712
4713             if (gl_info->supported[ARB_COLOR_BUFFER_FLOAT])
4714             {
4715                 GL_EXTCALL(glClampColorARB(GL_CLAMP_VERTEX_COLOR_ARB, !compiled->need_color_unclamp));
4716                 checkGLcall("glClampColorARB");
4717             } else {
4718                 FIXME("vertex color clamp needs to be changed, but extension not supported.\n");
4719             }
4720         }
4721     }
4722     else if (gl_info->supported[ARB_VERTEX_PROGRAM])
4723     {
4724         priv->current_vprogram_id = 0;
4725         gl_info->gl_ops.gl.p_glDisable(GL_VERTEX_PROGRAM_ARB);
4726         checkGLcall("glDisable(GL_VERTEX_PROGRAM_ARB)");
4727     }
4728 }
4729
4730 /* GL locking is done by the caller */
4731 static void shader_arb_select_depth_blt(void *shader_priv, const struct wined3d_gl_info *gl_info,
4732         enum tex_types tex_type, const SIZE *ds_mask_size)
4733 {
4734     const float mask[] = {0.0f, 0.0f, (float)ds_mask_size->cx, (float)ds_mask_size->cy};
4735     BOOL masked = ds_mask_size->cx && ds_mask_size->cy;
4736     struct shader_arb_priv *priv = shader_priv;
4737     GLuint *blt_fprogram;
4738
4739     if (!priv->depth_blt_vprogram_id) priv->depth_blt_vprogram_id = create_arb_blt_vertex_program(gl_info);
4740     GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->depth_blt_vprogram_id));
4741     gl_info->gl_ops.gl.p_glEnable(GL_VERTEX_PROGRAM_ARB);
4742
4743     blt_fprogram = masked ? &priv->depth_blt_fprogram_id_masked[tex_type] : &priv->depth_blt_fprogram_id_full[tex_type];
4744     if (!*blt_fprogram) *blt_fprogram = create_arb_blt_fragment_program(gl_info, tex_type, masked);
4745     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, *blt_fprogram));
4746     if (masked) GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, 0, mask));
4747     gl_info->gl_ops.gl.p_glEnable(GL_FRAGMENT_PROGRAM_ARB);
4748 }
4749
4750 /* GL locking is done by the caller */
4751 static void shader_arb_deselect_depth_blt(void *shader_priv, const struct wined3d_gl_info *gl_info)
4752 {
4753     struct shader_arb_priv *priv = shader_priv;
4754
4755     if (priv->current_vprogram_id) {
4756         GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->current_vprogram_id));
4757         checkGLcall("glBindProgramARB(GL_VERTEX_PROGRAM_ARB, vertexShader->prgId);");
4758
4759         TRACE("Bound vertex program %u and enabled GL_VERTEX_PROGRAM_ARB.\n", priv->current_vprogram_id);
4760     }
4761     else
4762     {
4763         gl_info->gl_ops.gl.p_glDisable(GL_VERTEX_PROGRAM_ARB);
4764         checkGLcall("glDisable(GL_VERTEX_PROGRAM_ARB)");
4765     }
4766
4767     if (priv->current_fprogram_id) {
4768         GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, priv->current_fprogram_id));
4769         checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, pixelShader->prgId);");
4770
4771         TRACE("Bound fragment program %u and enabled GL_FRAGMENT_PROGRAM_ARB.\n", priv->current_fprogram_id);
4772     }
4773     else if(!priv->use_arbfp_fixed_func)
4774     {
4775         gl_info->gl_ops.gl.p_glDisable(GL_FRAGMENT_PROGRAM_ARB);
4776         checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
4777     }
4778 }
4779
4780 static void shader_arb_destroy(struct wined3d_shader *shader)
4781 {
4782     struct wined3d_device *device = shader->device;
4783     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4784
4785     if (shader_is_pshader_version(shader->reg_maps.shader_version.type))
4786     {
4787         struct arb_pshader_private *shader_data = shader->backend_data;
4788         UINT i;
4789
4790         if(!shader_data) return; /* This can happen if a shader was never compiled */
4791
4792         if (shader_data->num_gl_shaders)
4793         {
4794             struct wined3d_context *context = context_acquire(device, NULL);
4795
4796             ENTER_GL();
4797             for (i = 0; i < shader_data->num_gl_shaders; ++i)
4798             {
4799                 GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId));
4800                 checkGLcall("GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId))");
4801             }
4802             LEAVE_GL();
4803
4804             context_release(context);
4805         }
4806
4807         HeapFree(GetProcessHeap(), 0, shader_data->gl_shaders);
4808         HeapFree(GetProcessHeap(), 0, shader_data);
4809         shader->backend_data = NULL;
4810     }
4811     else
4812     {
4813         struct arb_vshader_private *shader_data = shader->backend_data;
4814         UINT i;
4815
4816         if(!shader_data) return; /* This can happen if a shader was never compiled */
4817
4818         if (shader_data->num_gl_shaders)
4819         {
4820             struct wined3d_context *context = context_acquire(device, NULL);
4821
4822             ENTER_GL();
4823             for (i = 0; i < shader_data->num_gl_shaders; ++i)
4824             {
4825                 GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId));
4826                 checkGLcall("GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId))");
4827             }
4828             LEAVE_GL();
4829
4830             context_release(context);
4831         }
4832
4833         HeapFree(GetProcessHeap(), 0, shader_data->gl_shaders);
4834         HeapFree(GetProcessHeap(), 0, shader_data);
4835         shader->backend_data = NULL;
4836     }
4837 }
4838
4839 static int sig_tree_compare(const void *key, const struct wine_rb_entry *entry)
4840 {
4841     struct ps_signature *e = WINE_RB_ENTRY_VALUE(entry, struct ps_signature, entry);
4842     return compare_sig(key, e->sig);
4843 }
4844
4845 static const struct wine_rb_functions sig_tree_functions =
4846 {
4847     wined3d_rb_alloc,
4848     wined3d_rb_realloc,
4849     wined3d_rb_free,
4850     sig_tree_compare
4851 };
4852
4853 static HRESULT shader_arb_alloc(struct wined3d_device *device)
4854 {
4855     struct shader_arb_priv *priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*priv));
4856
4857     priv->vshader_const_dirty = HeapAlloc(GetProcessHeap(), 0,
4858             sizeof(*priv->vshader_const_dirty) * device->d3d_vshader_constantF);
4859     if (!priv->vshader_const_dirty)
4860         goto fail;
4861     memset(priv->vshader_const_dirty, 1,
4862            sizeof(*priv->vshader_const_dirty) * device->d3d_vshader_constantF);
4863
4864     priv->pshader_const_dirty = HeapAlloc(GetProcessHeap(), 0,
4865             sizeof(*priv->pshader_const_dirty) * device->d3d_pshader_constantF);
4866     if (!priv->pshader_const_dirty)
4867         goto fail;
4868     memset(priv->pshader_const_dirty, 1,
4869             sizeof(*priv->pshader_const_dirty) * device->d3d_pshader_constantF);
4870
4871     if(wine_rb_init(&priv->signature_tree, &sig_tree_functions) == -1)
4872     {
4873         ERR("RB tree init failed\n");
4874         goto fail;
4875     }
4876     device->shader_priv = priv;
4877     return WINED3D_OK;
4878
4879 fail:
4880     HeapFree(GetProcessHeap(), 0, priv->pshader_const_dirty);
4881     HeapFree(GetProcessHeap(), 0, priv->vshader_const_dirty);
4882     HeapFree(GetProcessHeap(), 0, priv);
4883     return E_OUTOFMEMORY;
4884 }
4885
4886 static void release_signature(struct wine_rb_entry *entry, void *context)
4887 {
4888     struct ps_signature *sig = WINE_RB_ENTRY_VALUE(entry, struct ps_signature, entry);
4889     int i;
4890     for(i = 0; i < MAX_REG_INPUT; i++)
4891     {
4892         HeapFree(GetProcessHeap(), 0, (char *) sig->sig[i].semantic_name);
4893     }
4894     HeapFree(GetProcessHeap(), 0, sig->sig);
4895     HeapFree(GetProcessHeap(), 0, sig);
4896 }
4897
4898 /* Context activation is done by the caller. */
4899 static void shader_arb_free(struct wined3d_device *device)
4900 {
4901     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4902     struct shader_arb_priv *priv = device->shader_priv;
4903     int i;
4904
4905     ENTER_GL();
4906     if(priv->depth_blt_vprogram_id) {
4907         GL_EXTCALL(glDeleteProgramsARB(1, &priv->depth_blt_vprogram_id));
4908     }
4909     for (i = 0; i < tex_type_count; ++i)
4910     {
4911         if (priv->depth_blt_fprogram_id_full[i])
4912         {
4913             GL_EXTCALL(glDeleteProgramsARB(1, &priv->depth_blt_fprogram_id_full[i]));
4914         }
4915         if (priv->depth_blt_fprogram_id_masked[i])
4916         {
4917             GL_EXTCALL(glDeleteProgramsARB(1, &priv->depth_blt_fprogram_id_masked[i]));
4918         }
4919     }
4920     LEAVE_GL();
4921
4922     wine_rb_destroy(&priv->signature_tree, release_signature, NULL);
4923     HeapFree(GetProcessHeap(), 0, priv->pshader_const_dirty);
4924     HeapFree(GetProcessHeap(), 0, priv->vshader_const_dirty);
4925     HeapFree(GetProcessHeap(), 0, device->shader_priv);
4926 }
4927
4928 static void shader_arb_context_destroyed(void *shader_priv, const struct wined3d_context *context)
4929 {
4930     struct shader_arb_priv *priv = shader_priv;
4931
4932     if (priv->last_context == context)
4933         priv->last_context = NULL;
4934 }
4935
4936 static void shader_arb_get_caps(const struct wined3d_gl_info *gl_info, struct shader_caps *caps)
4937 {
4938     if (gl_info->supported[ARB_VERTEX_PROGRAM])
4939     {
4940         DWORD vs_consts;
4941
4942         /* 96 is the minimum allowed value of MAX_PROGRAM_ENV_PARAMETERS_ARB
4943          * for vertex programs. If the native limit is less than that it's
4944          * not very useful, and e.g. Mesa swrast returns 0, probably to
4945          * indicate it's a software implementation. */
4946         if (gl_info->limits.arb_vs_native_constants < 96)
4947             vs_consts = gl_info->limits.arb_vs_float_constants;
4948         else
4949             vs_consts = min(gl_info->limits.arb_vs_float_constants, gl_info->limits.arb_vs_native_constants);
4950
4951         if (gl_info->supported[NV_VERTEX_PROGRAM3])
4952         {
4953             caps->VertexShaderVersion = 3;
4954             TRACE("Hardware vertex shader version 3.0 enabled (NV_VERTEX_PROGRAM3)\n");
4955         }
4956         else if (vs_consts >= 256)
4957         {
4958             /* Shader Model 2.0 requires at least 256 vertex shader constants */
4959             caps->VertexShaderVersion = 2;
4960             TRACE("Hardware vertex shader version 2.0 enabled (ARB_PROGRAM)\n");
4961         }
4962         else
4963         {
4964             caps->VertexShaderVersion = 1;
4965             TRACE("Hardware vertex shader version 1.1 enabled (ARB_PROGRAM)\n");
4966         }
4967         caps->MaxVertexShaderConst = vs_consts;
4968     }
4969     else
4970     {
4971         caps->VertexShaderVersion = 0;
4972         caps->MaxVertexShaderConst = 0;
4973     }
4974
4975     if (gl_info->supported[ARB_FRAGMENT_PROGRAM])
4976     {
4977         DWORD ps_consts;
4978
4979         /* Similar as above for vertex programs, but the minimum for fragment
4980          * programs is 24. */
4981         if (gl_info->limits.arb_ps_native_constants < 24)
4982             ps_consts = gl_info->limits.arb_ps_float_constants;
4983         else
4984             ps_consts = min(gl_info->limits.arb_ps_float_constants, gl_info->limits.arb_ps_native_constants);
4985
4986         if (gl_info->supported[NV_FRAGMENT_PROGRAM2])
4987         {
4988             caps->PixelShaderVersion = 3;
4989             TRACE("Hardware pixel shader version 3.0 enabled (NV_FRAGMENT_PROGRAM2)\n");
4990         }
4991         else if (ps_consts >= 32)
4992         {
4993             /* Shader Model 2.0 requires at least 32 pixel shader constants */
4994             caps->PixelShaderVersion = 2;
4995             TRACE("Hardware pixel shader version 2.0 enabled (ARB_PROGRAM)\n");
4996         }
4997         else
4998         {
4999             caps->PixelShaderVersion = 1;
5000             TRACE("Hardware pixel shader version 1.4 enabled (ARB_PROGRAM)\n");
5001         }
5002         caps->PixelShader1xMaxValue = 8.0f;
5003         caps->MaxPixelShaderConst = ps_consts;
5004     }
5005     else
5006     {
5007         caps->PixelShaderVersion = 0;
5008         caps->PixelShader1xMaxValue = 0.0f;
5009         caps->MaxPixelShaderConst = 0;
5010     }
5011
5012     caps->VSClipping = use_nv_clip(gl_info);
5013 }
5014
5015 static BOOL shader_arb_color_fixup_supported(struct color_fixup_desc fixup)
5016 {
5017     if (TRACE_ON(d3d_shader) && TRACE_ON(d3d))
5018     {
5019         TRACE("Checking support for color_fixup:\n");
5020         dump_color_fixup_desc(fixup);
5021     }
5022
5023     /* We support everything except complex conversions. */
5024     if (!is_complex_fixup(fixup))
5025     {
5026         TRACE("[OK]\n");
5027         return TRUE;
5028     }
5029
5030     TRACE("[FAILED]\n");
5031     return FALSE;
5032 }
5033
5034 static void shader_arb_add_instruction_modifiers(const struct wined3d_shader_instruction *ins) {
5035     DWORD shift;
5036     char write_mask[20], regstr[50];
5037     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
5038     BOOL is_color = FALSE;
5039     const struct wined3d_shader_dst_param *dst;
5040
5041     if (!ins->dst_count) return;
5042
5043     dst = &ins->dst[0];
5044     shift = dst->shift;
5045     if (!shift) return; /* Saturate alone is handled by the instructions */
5046
5047     shader_arb_get_write_mask(ins, dst, write_mask);
5048     shader_arb_get_register_name(ins, &dst->reg, regstr, &is_color);
5049
5050     /* Generate a line that does the output modifier computation
5051      * FIXME: _SAT vs shift? _SAT alone is already handled in the instructions, if this
5052      * maps problems in e.g. _d4_sat modify shader_arb_get_modifier
5053      */
5054     shader_addline(buffer, "MUL%s %s%s, %s, %s;\n", shader_arb_get_modifier(ins),
5055                    regstr, write_mask, regstr, shift_tab[shift]);
5056 }
5057
5058 static const SHADER_HANDLER shader_arb_instruction_handler_table[WINED3DSIH_TABLE_SIZE] =
5059 {
5060     /* WINED3DSIH_ABS           */ shader_hw_map2gl,
5061     /* WINED3DSIH_ADD           */ shader_hw_map2gl,
5062     /* WINED3DSIH_AND           */ NULL,
5063     /* WINED3DSIH_BEM           */ pshader_hw_bem,
5064     /* WINED3DSIH_BREAK         */ shader_hw_break,
5065     /* WINED3DSIH_BREAKC        */ shader_hw_breakc,
5066     /* WINED3DSIH_BREAKP        */ NULL,
5067     /* WINED3DSIH_CALL          */ shader_hw_call,
5068     /* WINED3DSIH_CALLNZ        */ NULL,
5069     /* WINED3DSIH_CMP           */ pshader_hw_cmp,
5070     /* WINED3DSIH_CND           */ pshader_hw_cnd,
5071     /* WINED3DSIH_CRS           */ shader_hw_map2gl,
5072     /* WINED3DSIH_CUT           */ NULL,
5073     /* WINED3DSIH_DCL           */ NULL,
5074     /* WINED3DSIH_DEF           */ NULL,
5075     /* WINED3DSIH_DEFB          */ NULL,
5076     /* WINED3DSIH_DEFI          */ NULL,
5077     /* WINED3DSIH_DIV           */ NULL,
5078     /* WINED3DSIH_DP2ADD        */ pshader_hw_dp2add,
5079     /* WINED3DSIH_DP3           */ shader_hw_map2gl,
5080     /* WINED3DSIH_DP4           */ shader_hw_map2gl,
5081     /* WINED3DSIH_DST           */ shader_hw_map2gl,
5082     /* WINED3DSIH_DSX           */ shader_hw_map2gl,
5083     /* WINED3DSIH_DSY           */ shader_hw_dsy,
5084     /* WINED3DSIH_ELSE          */ shader_hw_else,
5085     /* WINED3DSIH_EMIT          */ NULL,
5086     /* WINED3DSIH_ENDIF         */ shader_hw_endif,
5087     /* WINED3DSIH_ENDLOOP       */ shader_hw_endloop,
5088     /* WINED3DSIH_ENDREP        */ shader_hw_endrep,
5089     /* WINED3DSIH_EQ            */ NULL,
5090     /* WINED3DSIH_EXP           */ shader_hw_scalar_op,
5091     /* WINED3DSIH_EXPP          */ shader_hw_scalar_op,
5092     /* WINED3DSIH_FRC           */ shader_hw_map2gl,
5093     /* WINED3DSIH_FTOI          */ NULL,
5094     /* WINED3DSIH_GE            */ NULL,
5095     /* WINED3DSIH_IADD          */ NULL,
5096     /* WINED3DSIH_IEQ           */ NULL,
5097     /* WINED3DSIH_IF            */ NULL /* Hardcoded into the shader */,
5098     /* WINED3DSIH_IFC           */ shader_hw_ifc,
5099     /* WINED3DSIH_IGE           */ NULL,
5100     /* WINED3DSIH_IMUL          */ NULL,
5101     /* WINED3DSIH_ITOF          */ NULL,
5102     /* WINED3DSIH_LABEL         */ shader_hw_label,
5103     /* WINED3DSIH_LD            */ NULL,
5104     /* WINED3DSIH_LIT           */ shader_hw_map2gl,
5105     /* WINED3DSIH_LOG           */ shader_hw_log,
5106     /* WINED3DSIH_LOGP          */ shader_hw_log,
5107     /* WINED3DSIH_LOOP          */ shader_hw_loop,
5108     /* WINED3DSIH_LRP           */ shader_hw_lrp,
5109     /* WINED3DSIH_LT            */ NULL,
5110     /* WINED3DSIH_M3x2          */ shader_hw_mnxn,
5111     /* WINED3DSIH_M3x3          */ shader_hw_mnxn,
5112     /* WINED3DSIH_M3x4          */ shader_hw_mnxn,
5113     /* WINED3DSIH_M4x3          */ shader_hw_mnxn,
5114     /* WINED3DSIH_M4x4          */ shader_hw_mnxn,
5115     /* WINED3DSIH_MAD           */ shader_hw_map2gl,
5116     /* WINED3DSIH_MAX           */ shader_hw_map2gl,
5117     /* WINED3DSIH_MIN           */ shader_hw_map2gl,
5118     /* WINED3DSIH_MOV           */ shader_hw_mov,
5119     /* WINED3DSIH_MOVA          */ shader_hw_mov,
5120     /* WINED3DSIH_MOVC          */ NULL,
5121     /* WINED3DSIH_MUL           */ shader_hw_map2gl,
5122     /* WINED3DSIH_NOP           */ shader_hw_nop,
5123     /* WINED3DSIH_NRM           */ shader_hw_nrm,
5124     /* WINED3DSIH_PHASE         */ NULL,
5125     /* WINED3DSIH_POW           */ shader_hw_pow,
5126     /* WINED3DSIH_RCP           */ shader_hw_rcp,
5127     /* WINED3DSIH_REP           */ shader_hw_rep,
5128     /* WINED3DSIH_RET           */ shader_hw_ret,
5129     /* WINED3DSIH_ROUND_NI      */ NULL,
5130     /* WINED3DSIH_RSQ           */ shader_hw_scalar_op,
5131     /* WINED3DSIH_SAMPLE        */ NULL,
5132     /* WINED3DSIH_SAMPLE_GRAD   */ NULL,
5133     /* WINED3DSIH_SAMPLE_LOD    */ NULL,
5134     /* WINED3DSIH_SETP          */ NULL,
5135     /* WINED3DSIH_SGE           */ shader_hw_map2gl,
5136     /* WINED3DSIH_SGN           */ shader_hw_sgn,
5137     /* WINED3DSIH_SINCOS        */ shader_hw_sincos,
5138     /* WINED3DSIH_SLT           */ shader_hw_map2gl,
5139     /* WINED3DSIH_SQRT          */ NULL,
5140     /* WINED3DSIH_SUB           */ shader_hw_map2gl,
5141     /* WINED3DSIH_TEX           */ pshader_hw_tex,
5142     /* WINED3DSIH_TEXBEM        */ pshader_hw_texbem,
5143     /* WINED3DSIH_TEXBEML       */ pshader_hw_texbem,
5144     /* WINED3DSIH_TEXCOORD      */ pshader_hw_texcoord,
5145     /* WINED3DSIH_TEXDEPTH      */ pshader_hw_texdepth,
5146     /* WINED3DSIH_TEXDP3        */ pshader_hw_texdp3,
5147     /* WINED3DSIH_TEXDP3TEX     */ pshader_hw_texdp3tex,
5148     /* WINED3DSIH_TEXKILL       */ pshader_hw_texkill,
5149     /* WINED3DSIH_TEXLDD        */ shader_hw_texldd,
5150     /* WINED3DSIH_TEXLDL        */ shader_hw_texldl,
5151     /* WINED3DSIH_TEXM3x2DEPTH  */ pshader_hw_texm3x2depth,
5152     /* WINED3DSIH_TEXM3x2PAD    */ pshader_hw_texm3x2pad,
5153     /* WINED3DSIH_TEXM3x2TEX    */ pshader_hw_texm3x2tex,
5154     /* WINED3DSIH_TEXM3x3       */ pshader_hw_texm3x3,
5155     /* WINED3DSIH_TEXM3x3DIFF   */ NULL,
5156     /* WINED3DSIH_TEXM3x3PAD    */ pshader_hw_texm3x3pad,
5157     /* WINED3DSIH_TEXM3x3SPEC   */ pshader_hw_texm3x3spec,
5158     /* WINED3DSIH_TEXM3x3TEX    */ pshader_hw_texm3x3tex,
5159     /* WINED3DSIH_TEXM3x3VSPEC  */ pshader_hw_texm3x3vspec,
5160     /* WINED3DSIH_TEXREG2AR     */ pshader_hw_texreg2ar,
5161     /* WINED3DSIH_TEXREG2GB     */ pshader_hw_texreg2gb,
5162     /* WINED3DSIH_TEXREG2RGB    */ pshader_hw_texreg2rgb,
5163     /* WINED3DSIH_UDIV          */ NULL,
5164     /* WINED3DSIH_USHR          */ NULL,
5165     /* WINED3DSIH_UTOF          */ NULL,
5166     /* WINED3DSIH_XOR           */ NULL,
5167 };
5168
5169 static BOOL get_bool_const(const struct wined3d_shader_instruction *ins,
5170         const struct wined3d_shader *shader, DWORD idx)
5171 {
5172     const struct wined3d_shader_reg_maps *reg_maps = ins->ctx->reg_maps;
5173     BOOL vshader = shader_is_vshader_version(reg_maps->shader_version.type);
5174     const struct wined3d_shader_lconst *constant;
5175     WORD bools = 0;
5176     WORD flag = (1 << idx);
5177     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
5178
5179     if (reg_maps->local_bool_consts & flag)
5180     {
5181         /* What good is a if(bool) with a hardcoded local constant? I don't know, but handle it */
5182         LIST_FOR_EACH_ENTRY(constant, &shader->constantsB, struct wined3d_shader_lconst, entry)
5183         {
5184             if (constant->idx == idx)
5185             {
5186                 return constant->value[0];
5187             }
5188         }
5189         ERR("Local constant not found\n");
5190         return FALSE;
5191     }
5192     else
5193     {
5194         if(vshader) bools = priv->cur_vs_args->clip.boolclip.bools;
5195         else bools = priv->cur_ps_args->bools;
5196         return bools & flag;
5197     }
5198 }
5199
5200 static void get_loop_control_const(const struct wined3d_shader_instruction *ins,
5201         const struct wined3d_shader *shader, UINT idx, struct wined3d_shader_loop_control *loop_control)
5202 {
5203     const struct wined3d_shader_reg_maps *reg_maps = ins->ctx->reg_maps;
5204     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
5205
5206     /* Integer constants can either be a local constant, or they can be stored in the shader
5207      * type specific compile args. */
5208     if (reg_maps->local_int_consts & (1 << idx))
5209     {
5210         const struct wined3d_shader_lconst *constant;
5211
5212         LIST_FOR_EACH_ENTRY(constant, &shader->constantsI, struct wined3d_shader_lconst, entry)
5213         {
5214             if (constant->idx == idx)
5215             {
5216                 loop_control->count = constant->value[0];
5217                 loop_control->start = constant->value[1];
5218                 /* Step is signed. */
5219                 loop_control->step = (int)constant->value[2];
5220                 return;
5221             }
5222         }
5223         /* If this happens the flag was set incorrectly */
5224         ERR("Local constant not found\n");
5225         loop_control->count = 0;
5226         loop_control->start = 0;
5227         loop_control->step = 0;
5228         return;
5229     }
5230
5231     switch (reg_maps->shader_version.type)
5232     {
5233         case WINED3D_SHADER_TYPE_VERTEX:
5234             /* Count and aL start value are unsigned */
5235             loop_control->count = priv->cur_vs_args->loop_ctrl[idx][0];
5236             loop_control->start = priv->cur_vs_args->loop_ctrl[idx][1];
5237             /* Step is signed. */
5238             loop_control->step = ((char)priv->cur_vs_args->loop_ctrl[idx][2]);
5239             break;
5240
5241         case WINED3D_SHADER_TYPE_PIXEL:
5242             loop_control->count = priv->cur_ps_args->loop_ctrl[idx][0];
5243             loop_control->start = priv->cur_ps_args->loop_ctrl[idx][1];
5244             loop_control->step = ((char)priv->cur_ps_args->loop_ctrl[idx][2]);
5245             break;
5246
5247         default:
5248             FIXME("Unhandled shader type %#x.\n", reg_maps->shader_version.type);
5249             break;
5250     }
5251 }
5252
5253 static void record_instruction(struct list *list, const struct wined3d_shader_instruction *ins)
5254 {
5255     unsigned int i;
5256     struct wined3d_shader_dst_param *dst_param = NULL;
5257     struct wined3d_shader_src_param *src_param = NULL, *rel_addr = NULL;
5258     struct recorded_instruction *rec = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*rec));
5259     if(!rec)
5260     {
5261         ERR("Out of memory\n");
5262         return;
5263     }
5264
5265     rec->ins = *ins;
5266     dst_param = HeapAlloc(GetProcessHeap(), 0, sizeof(*dst_param));
5267     if(!dst_param) goto free;
5268     *dst_param = *ins->dst;
5269     if(ins->dst->reg.rel_addr)
5270     {
5271         rel_addr = HeapAlloc(GetProcessHeap(), 0, sizeof(*dst_param->reg.rel_addr));
5272         if(!rel_addr) goto free;
5273         *rel_addr = *ins->dst->reg.rel_addr;
5274         dst_param->reg.rel_addr = rel_addr;
5275     }
5276     rec->ins.dst = dst_param;
5277
5278     src_param = HeapAlloc(GetProcessHeap(), 0, sizeof(*src_param) * ins->src_count);
5279     if(!src_param) goto free;
5280     for(i = 0; i < ins->src_count; i++)
5281     {
5282         src_param[i] = ins->src[i];
5283         if(ins->src[i].reg.rel_addr)
5284         {
5285             rel_addr = HeapAlloc(GetProcessHeap(), 0, sizeof(*rel_addr));
5286             if(!rel_addr) goto free;
5287             *rel_addr = *ins->src[i].reg.rel_addr;
5288             src_param[i].reg.rel_addr = rel_addr;
5289         }
5290     }
5291     rec->ins.src = src_param;
5292     list_add_tail(list, &rec->entry);
5293     return;
5294
5295 free:
5296     ERR("Out of memory\n");
5297     if(dst_param)
5298     {
5299         HeapFree(GetProcessHeap(), 0, (void *) dst_param->reg.rel_addr);
5300         HeapFree(GetProcessHeap(), 0, dst_param);
5301     }
5302     if(src_param)
5303     {
5304         for(i = 0; i < ins->src_count; i++)
5305         {
5306             HeapFree(GetProcessHeap(), 0, (void *) src_param[i].reg.rel_addr);
5307         }
5308         HeapFree(GetProcessHeap(), 0, src_param);
5309     }
5310     HeapFree(GetProcessHeap(), 0, rec);
5311 }
5312
5313 static void free_recorded_instruction(struct list *list)
5314 {
5315     struct recorded_instruction *rec_ins, *entry2;
5316     unsigned int i;
5317
5318     LIST_FOR_EACH_ENTRY_SAFE(rec_ins, entry2, list, struct recorded_instruction, entry)
5319     {
5320         list_remove(&rec_ins->entry);
5321         if(rec_ins->ins.dst)
5322         {
5323             HeapFree(GetProcessHeap(), 0, (void *) rec_ins->ins.dst->reg.rel_addr);
5324             HeapFree(GetProcessHeap(), 0, (void *) rec_ins->ins.dst);
5325         }
5326         if(rec_ins->ins.src)
5327         {
5328             for(i = 0; i < rec_ins->ins.src_count; i++)
5329             {
5330                 HeapFree(GetProcessHeap(), 0, (void *) rec_ins->ins.src[i].reg.rel_addr);
5331             }
5332             HeapFree(GetProcessHeap(), 0, (void *) rec_ins->ins.src);
5333         }
5334         HeapFree(GetProcessHeap(), 0, rec_ins);
5335     }
5336 }
5337
5338 static void shader_arb_handle_instruction(const struct wined3d_shader_instruction *ins) {
5339     SHADER_HANDLER hw_fct;
5340     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
5341     const struct wined3d_shader *shader = ins->ctx->shader;
5342     struct control_frame *control_frame;
5343     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
5344     BOOL bool_const;
5345
5346     if(ins->handler_idx == WINED3DSIH_LOOP || ins->handler_idx == WINED3DSIH_REP)
5347     {
5348         control_frame = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*control_frame));
5349         list_add_head(&priv->control_frames, &control_frame->entry);
5350
5351         if(ins->handler_idx == WINED3DSIH_LOOP) control_frame->type = LOOP;
5352         if(ins->handler_idx == WINED3DSIH_REP) control_frame->type = REP;
5353
5354         if(priv->target_version >= NV2)
5355         {
5356             control_frame->no.loop = priv->num_loops++;
5357             priv->loop_depth++;
5358         }
5359         else
5360         {
5361             /* Don't bother recording when we're in a not used if branch */
5362             if(priv->muted)
5363             {
5364                 return;
5365             }
5366
5367             if(!priv->recording)
5368             {
5369                 list_init(&priv->record);
5370                 priv->recording = TRUE;
5371                 control_frame->outer_loop = TRUE;
5372                 get_loop_control_const(ins, shader, ins->src[0].reg.idx, &control_frame->loop_control);
5373                 return; /* Instruction is handled */
5374             }
5375             /* Record this loop in the outer loop's recording */
5376         }
5377     }
5378     else if(ins->handler_idx == WINED3DSIH_ENDLOOP || ins->handler_idx == WINED3DSIH_ENDREP)
5379     {
5380         if(priv->target_version >= NV2)
5381         {
5382             /* Nothing to do. The control frame is popped after the HW instr handler */
5383         }
5384         else
5385         {
5386             struct list *e = list_head(&priv->control_frames);
5387             control_frame = LIST_ENTRY(e, struct control_frame, entry);
5388             list_remove(&control_frame->entry);
5389
5390             if(control_frame->outer_loop)
5391             {
5392                 unsigned int iteration;
5393                 int aL = 0;
5394                 struct list copy;
5395
5396                 /* Turn off recording before playback */
5397                 priv->recording = FALSE;
5398
5399                 /* Move the recorded instructions to a separate list and get them out of the private data
5400                  * structure. If there are nested loops, the shader_arb_handle_instruction below will
5401                  * be recorded again, thus priv->record might be overwritten
5402                  */
5403                 list_init(&copy);
5404                 list_move_tail(&copy, &priv->record);
5405                 list_init(&priv->record);
5406
5407                 if(ins->handler_idx == WINED3DSIH_ENDLOOP)
5408                 {
5409                     shader_addline(buffer, "#unrolling loop: %u iterations, aL=%u, inc %d\n",
5410                                    control_frame->loop_control.count, control_frame->loop_control.start,
5411                                    control_frame->loop_control.step);
5412                     aL = control_frame->loop_control.start;
5413                 }
5414                 else
5415                 {
5416                     shader_addline(buffer, "#unrolling rep: %u iterations\n", control_frame->loop_control.count);
5417                 }
5418
5419                 for (iteration = 0; iteration < control_frame->loop_control.count; ++iteration)
5420                 {
5421                     struct recorded_instruction *rec_ins;
5422                     if(ins->handler_idx == WINED3DSIH_ENDLOOP)
5423                     {
5424                         priv->aL = aL;
5425                         shader_addline(buffer, "#Iteration %u, aL=%d\n", iteration, aL);
5426                     }
5427                     else
5428                     {
5429                         shader_addline(buffer, "#Iteration %u\n", iteration);
5430                     }
5431
5432                     LIST_FOR_EACH_ENTRY(rec_ins, &copy, struct recorded_instruction, entry)
5433                     {
5434                         shader_arb_handle_instruction(&rec_ins->ins);
5435                     }
5436
5437                     if(ins->handler_idx == WINED3DSIH_ENDLOOP)
5438                     {
5439                         aL += control_frame->loop_control.step;
5440                     }
5441                 }
5442                 shader_addline(buffer, "#end loop/rep\n");
5443
5444                 free_recorded_instruction(&copy);
5445                 HeapFree(GetProcessHeap(), 0, control_frame);
5446                 return; /* Instruction is handled */
5447             }
5448             else
5449             {
5450                 /* This is a nested loop. Proceed to the normal recording function */
5451                 HeapFree(GetProcessHeap(), 0, control_frame);
5452             }
5453         }
5454     }
5455
5456     if(priv->recording)
5457     {
5458         record_instruction(&priv->record, ins);
5459         return;
5460     }
5461
5462     /* boolean if */
5463     if(ins->handler_idx == WINED3DSIH_IF)
5464     {
5465         control_frame = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*control_frame));
5466         list_add_head(&priv->control_frames, &control_frame->entry);
5467         control_frame->type = IF;
5468
5469         bool_const = get_bool_const(ins, shader, ins->src[0].reg.idx);
5470         if(ins->src[0].modifiers == WINED3DSPSM_NOT) bool_const = !bool_const;
5471         if (!priv->muted && !bool_const)
5472         {
5473             shader_addline(buffer, "#if(FALSE){\n");
5474             priv->muted = TRUE;
5475             control_frame->muting = TRUE;
5476         }
5477         else shader_addline(buffer, "#if(TRUE) {\n");
5478
5479         return; /* Instruction is handled */
5480     }
5481     else if(ins->handler_idx == WINED3DSIH_IFC)
5482     {
5483         /* IF(bool) and if_cond(a, b) use the same ELSE and ENDIF tokens */
5484         control_frame = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*control_frame));
5485         control_frame->type = IFC;
5486         control_frame->no.ifc = priv->num_ifcs++;
5487         list_add_head(&priv->control_frames, &control_frame->entry);
5488     }
5489     else if(ins->handler_idx == WINED3DSIH_ELSE)
5490     {
5491         struct list *e = list_head(&priv->control_frames);
5492         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5493
5494         if(control_frame->type == IF)
5495         {
5496             shader_addline(buffer, "#} else {\n");
5497             if(!priv->muted && !control_frame->muting)
5498             {
5499                 priv->muted = TRUE;
5500                 control_frame->muting = TRUE;
5501             }
5502             else if(control_frame->muting) priv->muted = FALSE;
5503             return; /* Instruction is handled. */
5504         }
5505         /* In case of an ifc, generate a HW shader instruction */
5506     }
5507     else if(ins->handler_idx == WINED3DSIH_ENDIF)
5508     {
5509         struct list *e = list_head(&priv->control_frames);
5510         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5511
5512         if(control_frame->type == IF)
5513         {
5514             shader_addline(buffer, "#} endif\n");
5515             if(control_frame->muting) priv->muted = FALSE;
5516             list_remove(&control_frame->entry);
5517             HeapFree(GetProcessHeap(), 0, control_frame);
5518             return; /* Instruction is handled */
5519         }
5520     }
5521
5522     if(priv->muted) return;
5523
5524     /* Select handler */
5525     hw_fct = shader_arb_instruction_handler_table[ins->handler_idx];
5526
5527     /* Unhandled opcode */
5528     if (!hw_fct)
5529     {
5530         FIXME("Backend can't handle opcode %#x\n", ins->handler_idx);
5531         return;
5532     }
5533     hw_fct(ins);
5534
5535     if(ins->handler_idx == WINED3DSIH_ENDLOOP || ins->handler_idx == WINED3DSIH_ENDREP)
5536     {
5537         struct list *e = list_head(&priv->control_frames);
5538         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5539         list_remove(&control_frame->entry);
5540         HeapFree(GetProcessHeap(), 0, control_frame);
5541         priv->loop_depth--;
5542     }
5543     else if(ins->handler_idx == WINED3DSIH_ENDIF)
5544     {
5545         /* Non-ifc ENDIFs don't reach that place because of the return in the if block above */
5546         struct list *e = list_head(&priv->control_frames);
5547         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5548         list_remove(&control_frame->entry);
5549         HeapFree(GetProcessHeap(), 0, control_frame);
5550     }
5551
5552
5553     shader_arb_add_instruction_modifiers(ins);
5554 }
5555
5556 const struct wined3d_shader_backend_ops arb_program_shader_backend =
5557 {
5558     shader_arb_handle_instruction,
5559     shader_arb_select,
5560     shader_arb_select_depth_blt,
5561     shader_arb_deselect_depth_blt,
5562     shader_arb_update_float_vertex_constants,
5563     shader_arb_update_float_pixel_constants,
5564     shader_arb_load_constants,
5565     shader_arb_load_np2fixup_constants,
5566     shader_arb_destroy,
5567     shader_arb_alloc,
5568     shader_arb_free,
5569     shader_arb_context_destroyed,
5570     shader_arb_get_caps,
5571     shader_arb_color_fixup_supported,
5572 };
5573
5574 /* ARB_fragment_program fixed function pipeline replacement definitions */
5575 #define ARB_FFP_CONST_TFACTOR           0
5576 #define ARB_FFP_CONST_SPECULAR_ENABLE   ((ARB_FFP_CONST_TFACTOR) + 1)
5577 #define ARB_FFP_CONST_CONSTANT(i)       ((ARB_FFP_CONST_SPECULAR_ENABLE) + 1 + i)
5578 #define ARB_FFP_CONST_BUMPMAT(i)        ((ARB_FFP_CONST_CONSTANT(7)) + 1 + i)
5579 #define ARB_FFP_CONST_LUMINANCE(i)      ((ARB_FFP_CONST_BUMPMAT(7)) + 1 + i)
5580
5581 struct arbfp_ffp_desc
5582 {
5583     struct ffp_frag_desc parent;
5584     GLuint shader;
5585     unsigned int num_textures_used;
5586 };
5587
5588 /* Context activation and GL locking are done by the caller. */
5589 static void arbfp_enable(const struct wined3d_gl_info *gl_info, BOOL enable)
5590 {
5591     if (enable)
5592     {
5593         gl_info->gl_ops.gl.p_glEnable(GL_FRAGMENT_PROGRAM_ARB);
5594         checkGLcall("glEnable(GL_FRAGMENT_PROGRAM_ARB)");
5595     }
5596     else
5597     {
5598         gl_info->gl_ops.gl.p_glDisable(GL_FRAGMENT_PROGRAM_ARB);
5599         checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
5600     }
5601 }
5602
5603 static HRESULT arbfp_alloc(struct wined3d_device *device)
5604 {
5605     struct shader_arb_priv *priv;
5606     /* Share private data between the shader backend and the pipeline replacement, if both
5607      * are the arb implementation. This is needed to figure out whether ARBfp should be disabled
5608      * if no pixel shader is bound or not
5609      */
5610     if (device->shader_backend == &arb_program_shader_backend)
5611     {
5612         device->fragment_priv = device->shader_priv;
5613     }
5614     else
5615     {
5616         device->fragment_priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(struct shader_arb_priv));
5617         if (!device->fragment_priv) return E_OUTOFMEMORY;
5618     }
5619     priv = device->fragment_priv;
5620     if (wine_rb_init(&priv->fragment_shaders, &wined3d_ffp_frag_program_rb_functions) == -1)
5621     {
5622         ERR("Failed to initialize rbtree.\n");
5623         HeapFree(GetProcessHeap(), 0, device->fragment_priv);
5624         return E_OUTOFMEMORY;
5625     }
5626     priv->use_arbfp_fixed_func = TRUE;
5627     return WINED3D_OK;
5628 }
5629
5630 /* Context activation is done by the caller. */
5631 static void arbfp_free_ffpshader(struct wine_rb_entry *entry, void *context)
5632 {
5633     const struct wined3d_gl_info *gl_info = context;
5634     struct arbfp_ffp_desc *entry_arb = WINE_RB_ENTRY_VALUE(entry, struct arbfp_ffp_desc, parent.entry);
5635
5636     ENTER_GL();
5637     GL_EXTCALL(glDeleteProgramsARB(1, &entry_arb->shader));
5638     checkGLcall("glDeleteProgramsARB(1, &entry_arb->shader)");
5639     HeapFree(GetProcessHeap(), 0, entry_arb);
5640     LEAVE_GL();
5641 }
5642
5643 /* Context activation is done by the caller. */
5644 static void arbfp_free(struct wined3d_device *device)
5645 {
5646     struct shader_arb_priv *priv = device->fragment_priv;
5647
5648     wine_rb_destroy(&priv->fragment_shaders, arbfp_free_ffpshader, &device->adapter->gl_info);
5649     priv->use_arbfp_fixed_func = FALSE;
5650
5651     if (device->shader_backend != &arb_program_shader_backend)
5652     {
5653         HeapFree(GetProcessHeap(), 0, device->fragment_priv);
5654     }
5655 }
5656
5657 static void arbfp_get_caps(const struct wined3d_gl_info *gl_info, struct fragment_caps *caps)
5658 {
5659     caps->PrimitiveMiscCaps = WINED3DPMISCCAPS_TSSARGTEMP;
5660     caps->TextureOpCaps =  WINED3DTEXOPCAPS_DISABLE                     |
5661                            WINED3DTEXOPCAPS_SELECTARG1                  |
5662                            WINED3DTEXOPCAPS_SELECTARG2                  |
5663                            WINED3DTEXOPCAPS_MODULATE4X                  |
5664                            WINED3DTEXOPCAPS_MODULATE2X                  |
5665                            WINED3DTEXOPCAPS_MODULATE                    |
5666                            WINED3DTEXOPCAPS_ADDSIGNED2X                 |
5667                            WINED3DTEXOPCAPS_ADDSIGNED                   |
5668                            WINED3DTEXOPCAPS_ADD                         |
5669                            WINED3DTEXOPCAPS_SUBTRACT                    |
5670                            WINED3DTEXOPCAPS_ADDSMOOTH                   |
5671                            WINED3DTEXOPCAPS_BLENDCURRENTALPHA           |
5672                            WINED3DTEXOPCAPS_BLENDFACTORALPHA            |
5673                            WINED3DTEXOPCAPS_BLENDTEXTUREALPHA           |
5674                            WINED3DTEXOPCAPS_BLENDDIFFUSEALPHA           |
5675                            WINED3DTEXOPCAPS_BLENDTEXTUREALPHAPM         |
5676                            WINED3DTEXOPCAPS_MODULATEALPHA_ADDCOLOR      |
5677                            WINED3DTEXOPCAPS_MODULATECOLOR_ADDALPHA      |
5678                            WINED3DTEXOPCAPS_MODULATEINVCOLOR_ADDALPHA   |
5679                            WINED3DTEXOPCAPS_MODULATEINVALPHA_ADDCOLOR   |
5680                            WINED3DTEXOPCAPS_DOTPRODUCT3                 |
5681                            WINED3DTEXOPCAPS_MULTIPLYADD                 |
5682                            WINED3DTEXOPCAPS_LERP                        |
5683                            WINED3DTEXOPCAPS_BUMPENVMAP                  |
5684                            WINED3DTEXOPCAPS_BUMPENVMAPLUMINANCE;
5685
5686     /* TODO: Implement WINED3DTEXOPCAPS_PREMODULATE */
5687
5688     caps->MaxTextureBlendStages   = 8;
5689     caps->MaxSimultaneousTextures = min(gl_info->limits.fragment_samplers, 8);
5690 }
5691
5692 static void state_texfactor_arbfp(struct wined3d_context *context,
5693         const struct wined3d_state *state, DWORD state_id)
5694 {
5695     struct wined3d_device *device = context->swapchain->device;
5696     const struct wined3d_gl_info *gl_info = context->gl_info;
5697     float col[4];
5698
5699     /* Don't load the parameter if we're using an arbfp pixel shader,
5700      * otherwise we'll overwrite application provided constants. */
5701     if (device->shader_backend == &arb_program_shader_backend)
5702     {
5703         struct shader_arb_priv *priv;
5704
5705         if (use_ps(state)) return;
5706
5707         priv = device->shader_priv;
5708         priv->pshader_const_dirty[ARB_FFP_CONST_TFACTOR] = 1;
5709         priv->highest_dirty_ps_const = max(priv->highest_dirty_ps_const, ARB_FFP_CONST_TFACTOR + 1);
5710     }
5711
5712     D3DCOLORTOGLFLOAT4(state->render_states[WINED3D_RS_TEXTUREFACTOR], col);
5713     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_TFACTOR, col));
5714     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_TFACTOR, col)");
5715 }
5716
5717 static void state_arb_specularenable(struct wined3d_context *context,
5718         const struct wined3d_state *state, DWORD state_id)
5719 {
5720     struct wined3d_device *device = context->swapchain->device;
5721     const struct wined3d_gl_info *gl_info = context->gl_info;
5722     float col[4];
5723
5724     /* Don't load the parameter if we're using an arbfp pixel shader, otherwise we'll overwrite
5725      * application provided constants
5726      */
5727     if (device->shader_backend == &arb_program_shader_backend)
5728     {
5729         struct shader_arb_priv *priv;
5730
5731         if (use_ps(state)) return;
5732
5733         priv = device->shader_priv;
5734         priv->pshader_const_dirty[ARB_FFP_CONST_SPECULAR_ENABLE] = 1;
5735         priv->highest_dirty_ps_const = max(priv->highest_dirty_ps_const, ARB_FFP_CONST_SPECULAR_ENABLE + 1);
5736     }
5737
5738     if (state->render_states[WINED3D_RS_SPECULARENABLE])
5739     {
5740         /* The specular color has no alpha */
5741         col[0] = 1.0f; col[1] = 1.0f;
5742         col[2] = 1.0f; col[3] = 0.0f;
5743     } else {
5744         col[0] = 0.0f; col[1] = 0.0f;
5745         col[2] = 0.0f; col[3] = 0.0f;
5746     }
5747     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_SPECULAR_ENABLE, col));
5748     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_SPECULAR_ENABLE, col)");
5749 }
5750
5751 static void set_bumpmat_arbfp(struct wined3d_context *context, const struct wined3d_state *state, DWORD state_id)
5752 {
5753     DWORD stage = (state_id - STATE_TEXTURESTAGE(0, 0)) / (WINED3D_HIGHEST_TEXTURE_STATE + 1);
5754     struct wined3d_device *device = context->swapchain->device;
5755     const struct wined3d_gl_info *gl_info = context->gl_info;
5756     float mat[2][2];
5757
5758     if (use_ps(state))
5759     {
5760         if (stage && (state->pixel_shader->reg_maps.bumpmat & (1 << stage)))
5761         {
5762             /* The pixel shader has to know the bump env matrix. Do a constants update if it isn't scheduled
5763              * anyway
5764              */
5765             if (!isStateDirty(context, STATE_PIXELSHADERCONSTANT))
5766                 context_apply_state(context, state, STATE_PIXELSHADERCONSTANT);
5767         }
5768
5769         if(device->shader_backend == &arb_program_shader_backend) {
5770             /* Exit now, don't set the bumpmat below, otherwise we may overwrite pixel shader constants */
5771             return;
5772         }
5773     }
5774     else if (device->shader_backend == &arb_program_shader_backend)
5775     {
5776         struct shader_arb_priv *priv = device->shader_priv;
5777         priv->pshader_const_dirty[ARB_FFP_CONST_BUMPMAT(stage)] = 1;
5778         priv->highest_dirty_ps_const = max(priv->highest_dirty_ps_const, ARB_FFP_CONST_BUMPMAT(stage) + 1);
5779     }
5780
5781     mat[0][0] = *((float *)&state->texture_states[stage][WINED3D_TSS_BUMPENV_MAT00]);
5782     mat[0][1] = *((float *)&state->texture_states[stage][WINED3D_TSS_BUMPENV_MAT01]);
5783     mat[1][0] = *((float *)&state->texture_states[stage][WINED3D_TSS_BUMPENV_MAT10]);
5784     mat[1][1] = *((float *)&state->texture_states[stage][WINED3D_TSS_BUMPENV_MAT11]);
5785
5786     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_BUMPMAT(stage), &mat[0][0]));
5787     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_BUMPMAT(stage), &mat[0][0])");
5788 }
5789
5790 static void tex_bumpenvlum_arbfp(struct wined3d_context *context,
5791         const struct wined3d_state *state, DWORD state_id)
5792 {
5793     DWORD stage = (state_id - STATE_TEXTURESTAGE(0, 0)) / (WINED3D_HIGHEST_TEXTURE_STATE + 1);
5794     struct wined3d_device *device = context->swapchain->device;
5795     const struct wined3d_gl_info *gl_info = context->gl_info;
5796     float param[4];
5797
5798     if (use_ps(state))
5799     {
5800         if (stage && (state->pixel_shader->reg_maps.luminanceparams & (1 << stage)))
5801         {
5802             /* The pixel shader has to know the luminance offset. Do a constants update if it
5803              * isn't scheduled anyway
5804              */
5805             if (!isStateDirty(context, STATE_PIXELSHADERCONSTANT))
5806                 context_apply_state(context, state, STATE_PIXELSHADERCONSTANT);
5807         }
5808
5809         if(device->shader_backend == &arb_program_shader_backend) {
5810             /* Exit now, don't set the bumpmat below, otherwise we may overwrite pixel shader constants */
5811             return;
5812         }
5813     }
5814     else if (device->shader_backend == &arb_program_shader_backend)
5815     {
5816         struct shader_arb_priv *priv = device->shader_priv;
5817         priv->pshader_const_dirty[ARB_FFP_CONST_LUMINANCE(stage)] = 1;
5818         priv->highest_dirty_ps_const = max(priv->highest_dirty_ps_const, ARB_FFP_CONST_LUMINANCE(stage) + 1);
5819     }
5820
5821     param[0] = *((float *)&state->texture_states[stage][WINED3D_TSS_BUMPENV_LSCALE]);
5822     param[1] = *((float *)&state->texture_states[stage][WINED3D_TSS_BUMPENV_LOFFSET]);
5823     param[2] = 0.0f;
5824     param[3] = 0.0f;
5825
5826     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_LUMINANCE(stage), param));
5827     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_LUMINANCE(stage), param)");
5828 }
5829
5830 static const char *get_argreg(struct wined3d_shader_buffer *buffer, DWORD argnum, unsigned int stage, DWORD arg)
5831 {
5832     const char *ret;
5833
5834     if(arg == ARG_UNUSED) return "unused"; /* This is the marker for unused registers */
5835
5836     switch(arg & WINED3DTA_SELECTMASK) {
5837         case WINED3DTA_DIFFUSE:
5838             ret = "fragment.color.primary"; break;
5839
5840         case WINED3DTA_CURRENT:
5841             if (!stage) ret = "fragment.color.primary";
5842             else ret = "ret";
5843             break;
5844
5845         case WINED3DTA_TEXTURE:
5846             switch(stage) {
5847                 case 0: ret = "tex0"; break;
5848                 case 1: ret = "tex1"; break;
5849                 case 2: ret = "tex2"; break;
5850                 case 3: ret = "tex3"; break;
5851                 case 4: ret = "tex4"; break;
5852                 case 5: ret = "tex5"; break;
5853                 case 6: ret = "tex6"; break;
5854                 case 7: ret = "tex7"; break;
5855                 default: ret = "unknown texture";
5856             }
5857             break;
5858
5859         case WINED3DTA_TFACTOR:
5860             ret = "tfactor"; break;
5861
5862         case WINED3DTA_SPECULAR:
5863             ret = "fragment.color.secondary"; break;
5864
5865         case WINED3DTA_TEMP:
5866             ret = "tempreg"; break;
5867
5868         case WINED3DTA_CONSTANT:
5869             FIXME("Implement perstage constants\n");
5870             switch(stage) {
5871                 case 0: ret = "const0"; break;
5872                 case 1: ret = "const1"; break;
5873                 case 2: ret = "const2"; break;
5874                 case 3: ret = "const3"; break;
5875                 case 4: ret = "const4"; break;
5876                 case 5: ret = "const5"; break;
5877                 case 6: ret = "const6"; break;
5878                 case 7: ret = "const7"; break;
5879                 default: ret = "unknown constant";
5880             }
5881             break;
5882
5883         default:
5884             return "unknown";
5885     }
5886
5887     if(arg & WINED3DTA_COMPLEMENT) {
5888         shader_addline(buffer, "SUB arg%u, const.x, %s;\n", argnum, ret);
5889         if(argnum == 0) ret = "arg0";
5890         if(argnum == 1) ret = "arg1";
5891         if(argnum == 2) ret = "arg2";
5892     }
5893     if(arg & WINED3DTA_ALPHAREPLICATE) {
5894         shader_addline(buffer, "MOV arg%u, %s.w;\n", argnum, ret);
5895         if(argnum == 0) ret = "arg0";
5896         if(argnum == 1) ret = "arg1";
5897         if(argnum == 2) ret = "arg2";
5898     }
5899     return ret;
5900 }
5901
5902 static void gen_ffp_instr(struct wined3d_shader_buffer *buffer, unsigned int stage, BOOL color,
5903         BOOL alpha, DWORD dst, DWORD op, DWORD dw_arg0, DWORD dw_arg1, DWORD dw_arg2)
5904 {
5905     const char *dstmask, *dstreg, *arg0, *arg1, *arg2;
5906     unsigned int mul = 1;
5907     BOOL mul_final_dest = FALSE;
5908
5909     if(color && alpha) dstmask = "";
5910     else if(color) dstmask = ".xyz";
5911     else dstmask = ".w";
5912
5913     if(dst == tempreg) dstreg = "tempreg";
5914     else dstreg = "ret";
5915
5916     arg0 = get_argreg(buffer, 0, stage, dw_arg0);
5917     arg1 = get_argreg(buffer, 1, stage, dw_arg1);
5918     arg2 = get_argreg(buffer, 2, stage, dw_arg2);
5919
5920     switch (op)
5921     {
5922         case WINED3D_TOP_DISABLE:
5923             if (!stage)
5924                 shader_addline(buffer, "MOV %s%s, fragment.color.primary;\n", dstreg, dstmask);
5925             break;
5926
5927         case WINED3D_TOP_SELECT_ARG2:
5928             arg1 = arg2;
5929             /* FALLTHROUGH */
5930         case WINED3D_TOP_SELECT_ARG1:
5931             shader_addline(buffer, "MOV %s%s, %s;\n", dstreg, dstmask, arg1);
5932             break;
5933
5934         case WINED3D_TOP_MODULATE_4X:
5935             mul = 2;
5936             /* FALLTHROUGH */
5937         case WINED3D_TOP_MODULATE_2X:
5938             mul *= 2;
5939             if (!strcmp(dstreg, "result.color"))
5940             {
5941                 dstreg = "ret";
5942                 mul_final_dest = TRUE;
5943             }
5944             /* FALLTHROUGH */
5945         case WINED3D_TOP_MODULATE:
5946             shader_addline(buffer, "MUL %s%s, %s, %s;\n", dstreg, dstmask, arg1, arg2);
5947             break;
5948
5949         case WINED3D_TOP_ADD_SIGNED_2X:
5950             mul = 2;
5951             if (!strcmp(dstreg, "result.color"))
5952             {
5953                 dstreg = "ret";
5954                 mul_final_dest = TRUE;
5955             }
5956             /* FALLTHROUGH */
5957         case WINED3D_TOP_ADD_SIGNED:
5958             shader_addline(buffer, "SUB arg2, %s, const.w;\n", arg2);
5959             arg2 = "arg2";
5960             /* FALLTHROUGH */
5961         case WINED3D_TOP_ADD:
5962             shader_addline(buffer, "ADD_SAT %s%s, %s, %s;\n", dstreg, dstmask, arg1, arg2);
5963             break;
5964
5965         case WINED3D_TOP_SUBTRACT:
5966             shader_addline(buffer, "SUB_SAT %s%s, %s, %s;\n", dstreg, dstmask, arg1, arg2);
5967             break;
5968
5969         case WINED3D_TOP_ADD_SMOOTH:
5970             shader_addline(buffer, "SUB arg1, const.x, %s;\n", arg1);
5971             shader_addline(buffer, "MAD_SAT %s%s, arg1, %s, %s;\n", dstreg, dstmask, arg2, arg1);
5972             break;
5973
5974         case WINED3D_TOP_BLEND_CURRENT_ALPHA:
5975             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_CURRENT);
5976             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
5977             break;
5978         case WINED3D_TOP_BLEND_FACTOR_ALPHA:
5979             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_TFACTOR);
5980             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
5981             break;
5982         case WINED3D_TOP_BLEND_TEXTURE_ALPHA:
5983             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_TEXTURE);
5984             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
5985             break;
5986         case WINED3D_TOP_BLEND_DIFFUSE_ALPHA:
5987             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_DIFFUSE);
5988             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
5989             break;
5990
5991         case WINED3D_TOP_BLEND_TEXTURE_ALPHA_PM:
5992             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_TEXTURE);
5993             shader_addline(buffer, "SUB arg0.w, const.x, %s.w;\n", arg0);
5994             shader_addline(buffer, "MAD_SAT %s%s, %s, arg0.w, %s;\n", dstreg, dstmask, arg2, arg1);
5995             break;
5996
5997         /* D3DTOP_PREMODULATE ???? */
5998
5999         case WINED3D_TOP_MODULATE_INVALPHA_ADD_COLOR:
6000             shader_addline(buffer, "SUB arg0.w, const.x, %s;\n", arg1);
6001             shader_addline(buffer, "MAD_SAT %s%s, arg0.w, %s, %s;\n", dstreg, dstmask, arg2, arg1);
6002             break;
6003         case WINED3D_TOP_MODULATE_ALPHA_ADD_COLOR:
6004             shader_addline(buffer, "MAD_SAT %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg1, arg2, arg1);
6005             break;
6006         case WINED3D_TOP_MODULATE_INVCOLOR_ADD_ALPHA:
6007             shader_addline(buffer, "SUB arg0, const.x, %s;\n", arg1);
6008             shader_addline(buffer, "MAD_SAT %s%s, arg0, %s, %s.w;\n", dstreg, dstmask, arg2, arg1);
6009             break;
6010         case WINED3D_TOP_MODULATE_COLOR_ADD_ALPHA:
6011             shader_addline(buffer, "MAD_SAT %s%s, %s, %s, %s.w;\n", dstreg, dstmask, arg1, arg2, arg1);
6012             break;
6013
6014         case WINED3D_TOP_DOTPRODUCT3:
6015             mul = 4;
6016             if (!strcmp(dstreg, "result.color"))
6017             {
6018                 dstreg = "ret";
6019                 mul_final_dest = TRUE;
6020             }
6021             shader_addline(buffer, "SUB arg1, %s, const.w;\n", arg1);
6022             shader_addline(buffer, "SUB arg2, %s, const.w;\n", arg2);
6023             shader_addline(buffer, "DP3_SAT %s%s, arg1, arg2;\n", dstreg, dstmask);
6024             break;
6025
6026         case WINED3D_TOP_MULTIPLY_ADD:
6027             shader_addline(buffer, "MAD_SAT %s%s, %s, %s, %s;\n", dstreg, dstmask, arg1, arg2, arg0);
6028             break;
6029
6030         case WINED3D_TOP_LERP:
6031             /* The msdn is not quite right here */
6032             shader_addline(buffer, "LRP %s%s, %s, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
6033             break;
6034
6035         case WINED3D_TOP_BUMPENVMAP:
6036         case WINED3D_TOP_BUMPENVMAP_LUMINANCE:
6037             /* Those are handled in the first pass of the shader(generation pass 1 and 2) already */
6038             break;
6039
6040         default:
6041             FIXME("Unhandled texture op %08x\n", op);
6042     }
6043
6044     if(mul == 2) {
6045         shader_addline(buffer, "MUL_SAT %s%s, %s, const.y;\n", mul_final_dest ? "result.color" : dstreg, dstmask, dstreg);
6046     } else if(mul == 4) {
6047         shader_addline(buffer, "MUL_SAT %s%s, %s, const.z;\n", mul_final_dest ? "result.color" : dstreg, dstmask, dstreg);
6048     }
6049 }
6050
6051 static GLuint gen_arbfp_ffp_shader(const struct ffp_frag_settings *settings, const struct wined3d_gl_info *gl_info)
6052 {
6053     unsigned int stage;
6054     struct wined3d_shader_buffer buffer;
6055     BOOL tex_read[MAX_TEXTURES] = {FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE};
6056     BOOL bump_used[MAX_TEXTURES] = {FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE};
6057     BOOL luminance_used[MAX_TEXTURES] = {FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE};
6058     UINT lowest_disabled_stage;
6059     const char *textype;
6060     const char *instr, *sat;
6061     char colorcor_dst[8];
6062     GLuint ret;
6063     DWORD arg0, arg1, arg2;
6064     BOOL tempreg_used = FALSE, tfactor_used = FALSE;
6065     BOOL op_equal;
6066     const char *final_combiner_src = "ret";
6067     GLint pos;
6068
6069     /* Find out which textures are read */
6070     for (stage = 0; stage < MAX_TEXTURES; ++stage)
6071     {
6072         if (settings->op[stage].cop == WINED3D_TOP_DISABLE)
6073             break;
6074         arg0 = settings->op[stage].carg0 & WINED3DTA_SELECTMASK;
6075         arg1 = settings->op[stage].carg1 & WINED3DTA_SELECTMASK;
6076         arg2 = settings->op[stage].carg2 & WINED3DTA_SELECTMASK;
6077         if(arg0 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
6078         if(arg1 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
6079         if(arg2 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
6080
6081         if (settings->op[stage].cop == WINED3D_TOP_BLEND_TEXTURE_ALPHA)
6082             tex_read[stage] = TRUE;
6083         if (settings->op[stage].cop == WINED3D_TOP_BLEND_TEXTURE_ALPHA_PM)
6084             tex_read[stage] = TRUE;
6085         if (settings->op[stage].cop == WINED3D_TOP_BUMPENVMAP)
6086         {
6087             bump_used[stage] = TRUE;
6088             tex_read[stage] = TRUE;
6089         }
6090         if (settings->op[stage].cop == WINED3D_TOP_BUMPENVMAP_LUMINANCE)
6091         {
6092             bump_used[stage] = TRUE;
6093             tex_read[stage] = TRUE;
6094             luminance_used[stage] = TRUE;
6095         }
6096         else if (settings->op[stage].cop == WINED3D_TOP_BLEND_FACTOR_ALPHA)
6097         {
6098             tfactor_used = TRUE;
6099         }
6100
6101         if(arg0 == WINED3DTA_TFACTOR || arg1 == WINED3DTA_TFACTOR || arg2 == WINED3DTA_TFACTOR) {
6102             tfactor_used = TRUE;
6103         }
6104
6105         if(settings->op[stage].dst == tempreg) tempreg_used = TRUE;
6106         if(arg0 == WINED3DTA_TEMP || arg1 == WINED3DTA_TEMP || arg2 == WINED3DTA_TEMP) {
6107             tempreg_used = TRUE;
6108         }
6109
6110         if (settings->op[stage].aop == WINED3D_TOP_DISABLE)
6111             continue;
6112         arg0 = settings->op[stage].aarg0 & WINED3DTA_SELECTMASK;
6113         arg1 = settings->op[stage].aarg1 & WINED3DTA_SELECTMASK;
6114         arg2 = settings->op[stage].aarg2 & WINED3DTA_SELECTMASK;
6115         if(arg0 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
6116         if(arg1 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
6117         if(arg2 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
6118
6119         if(arg0 == WINED3DTA_TEMP || arg1 == WINED3DTA_TEMP || arg2 == WINED3DTA_TEMP) {
6120             tempreg_used = TRUE;
6121         }
6122         if(arg0 == WINED3DTA_TFACTOR || arg1 == WINED3DTA_TFACTOR || arg2 == WINED3DTA_TFACTOR) {
6123             tfactor_used = TRUE;
6124         }
6125     }
6126     lowest_disabled_stage = stage;
6127
6128     /* Shader header */
6129     if (!shader_buffer_init(&buffer))
6130     {
6131         ERR("Failed to initialize shader buffer.\n");
6132         return 0;
6133     }
6134
6135     shader_addline(&buffer, "!!ARBfp1.0\n");
6136
6137     switch(settings->fog) {
6138         case FOG_OFF:                                                         break;
6139         case FOG_LINEAR: shader_addline(&buffer, "OPTION ARB_fog_linear;\n"); break;
6140         case FOG_EXP:    shader_addline(&buffer, "OPTION ARB_fog_exp;\n");    break;
6141         case FOG_EXP2:   shader_addline(&buffer, "OPTION ARB_fog_exp2;\n");   break;
6142         default: FIXME("Unexpected fog setting %d\n", settings->fog);
6143     }
6144
6145     shader_addline(&buffer, "PARAM const = {1, 2, 4, 0.5};\n");
6146     shader_addline(&buffer, "TEMP TMP;\n");
6147     shader_addline(&buffer, "TEMP ret;\n");
6148     if(tempreg_used || settings->sRGB_write) shader_addline(&buffer, "TEMP tempreg;\n");
6149     shader_addline(&buffer, "TEMP arg0;\n");
6150     shader_addline(&buffer, "TEMP arg1;\n");
6151     shader_addline(&buffer, "TEMP arg2;\n");
6152     for(stage = 0; stage < MAX_TEXTURES; stage++) {
6153         if(!tex_read[stage]) continue;
6154         shader_addline(&buffer, "TEMP tex%u;\n", stage);
6155         if(!bump_used[stage]) continue;
6156         shader_addline(&buffer, "PARAM bumpmat%u = program.env[%u];\n", stage, ARB_FFP_CONST_BUMPMAT(stage));
6157         if(!luminance_used[stage]) continue;
6158         shader_addline(&buffer, "PARAM luminance%u = program.env[%u];\n", stage, ARB_FFP_CONST_LUMINANCE(stage));
6159     }
6160     if(tfactor_used) {
6161         shader_addline(&buffer, "PARAM tfactor = program.env[%u];\n", ARB_FFP_CONST_TFACTOR);
6162     }
6163         shader_addline(&buffer, "PARAM specular_enable = program.env[%u];\n", ARB_FFP_CONST_SPECULAR_ENABLE);
6164
6165     if(settings->sRGB_write) {
6166         shader_addline(&buffer, "PARAM srgb_consts1 = {%f, %f, %f, %f};\n",
6167                        srgb_mul_low, srgb_cmp, srgb_pow, srgb_mul_high);
6168         shader_addline(&buffer, "PARAM srgb_consts2 = {%f, %f, %f, %f};\n",
6169                        srgb_sub_high, 0.0, 0.0, 0.0);
6170     }
6171
6172     if (lowest_disabled_stage < 7 && settings->emul_clipplanes)
6173         shader_addline(&buffer, "KIL fragment.texcoord[7];\n");
6174
6175     /* Generate texture sampling instructions) */
6176     for (stage = 0; stage < MAX_TEXTURES && settings->op[stage].cop != WINED3D_TOP_DISABLE; ++stage)
6177     {
6178         if (!tex_read[stage])
6179             continue;
6180
6181         switch(settings->op[stage].tex_type) {
6182             case tex_1d:                    textype = "1D";     break;
6183             case tex_2d:                    textype = "2D";     break;
6184             case tex_3d:                    textype = "3D";     break;
6185             case tex_cube:                  textype = "CUBE";   break;
6186             case tex_rect:                  textype = "RECT";   break;
6187             default: textype = "unexpected_textype";   break;
6188         }
6189
6190         if (settings->op[stage].cop == WINED3D_TOP_BUMPENVMAP
6191                 || settings->op[stage].cop == WINED3D_TOP_BUMPENVMAP_LUMINANCE)
6192             sat = "";
6193         else
6194             sat = "_SAT";
6195
6196         if(settings->op[stage].projected == proj_none) {
6197             instr = "TEX";
6198         } else if(settings->op[stage].projected == proj_count4 ||
6199                   settings->op[stage].projected == proj_count3) {
6200             instr = "TXP";
6201         } else {
6202             FIXME("Unexpected projection mode %d\n", settings->op[stage].projected);
6203             instr = "TXP";
6204         }
6205
6206         if (stage > 0
6207                 && (settings->op[stage - 1].cop == WINED3D_TOP_BUMPENVMAP
6208                 || settings->op[stage - 1].cop == WINED3D_TOP_BUMPENVMAP_LUMINANCE))
6209         {
6210             shader_addline(&buffer, "SWZ arg1, bumpmat%u, x, z, 0, 0;\n", stage - 1);
6211             shader_addline(&buffer, "DP3 ret.x, arg1, tex%u;\n", stage - 1);
6212             shader_addline(&buffer, "SWZ arg1, bumpmat%u, y, w, 0, 0;\n", stage - 1);
6213             shader_addline(&buffer, "DP3 ret.y, arg1, tex%u;\n", stage - 1);
6214
6215             /* with projective textures, texbem only divides the static texture coord, not the displacement,
6216              * so multiply the displacement with the dividing parameter before passing it to TXP
6217              */
6218             if (settings->op[stage].projected != proj_none) {
6219                 if(settings->op[stage].projected == proj_count4) {
6220                     shader_addline(&buffer, "MOV ret.w, fragment.texcoord[%u].w;\n", stage);
6221                     shader_addline(&buffer, "MUL ret.xyz, ret, fragment.texcoord[%u].w, fragment.texcoord[%u];\n", stage, stage);
6222                 } else {
6223                     shader_addline(&buffer, "MOV ret.w, fragment.texcoord[%u].z;\n", stage);
6224                     shader_addline(&buffer, "MAD ret.xyz, ret, fragment.texcoord[%u].z, fragment.texcoord[%u];\n", stage, stage);
6225                 }
6226             } else {
6227                 shader_addline(&buffer, "ADD ret, ret, fragment.texcoord[%u];\n", stage);
6228             }
6229
6230             shader_addline(&buffer, "%s%s tex%u, ret, texture[%u], %s;\n",
6231                     instr, sat, stage, stage, textype);
6232             if (settings->op[stage - 1].cop == WINED3D_TOP_BUMPENVMAP_LUMINANCE)
6233             {
6234                 shader_addline(&buffer, "MAD_SAT ret.x, tex%u.z, luminance%u.x, luminance%u.y;\n",
6235                                stage - 1, stage - 1, stage - 1);
6236                 shader_addline(&buffer, "MUL tex%u, tex%u, ret.x;\n", stage, stage);
6237             }
6238         } else if(settings->op[stage].projected == proj_count3) {
6239             shader_addline(&buffer, "MOV ret, fragment.texcoord[%u];\n", stage);
6240             shader_addline(&buffer, "MOV ret.w, ret.z;\n");
6241             shader_addline(&buffer, "%s%s tex%u, ret, texture[%u], %s;\n",
6242                             instr, sat, stage, stage, textype);
6243         } else {
6244             shader_addline(&buffer, "%s%s tex%u, fragment.texcoord[%u], texture[%u], %s;\n",
6245                             instr, sat, stage, stage, stage, textype);
6246         }
6247
6248         sprintf(colorcor_dst, "tex%u", stage);
6249         gen_color_correction(&buffer, colorcor_dst, WINED3DSP_WRITEMASK_ALL, "const.x", "const.y",
6250                 settings->op[stage].color_fixup);
6251     }
6252
6253     /* Generate the main shader */
6254     for (stage = 0; stage < MAX_TEXTURES; ++stage)
6255     {
6256         if (settings->op[stage].cop == WINED3D_TOP_DISABLE)
6257         {
6258             if (!stage)
6259                 final_combiner_src = "fragment.color.primary";
6260             break;
6261         }
6262
6263         if (settings->op[stage].cop == WINED3D_TOP_SELECT_ARG1
6264                 && settings->op[stage].aop == WINED3D_TOP_SELECT_ARG1)
6265             op_equal = settings->op[stage].carg1 == settings->op[stage].aarg1;
6266         else if (settings->op[stage].cop == WINED3D_TOP_SELECT_ARG1
6267                 && settings->op[stage].aop == WINED3D_TOP_SELECT_ARG2)
6268             op_equal = settings->op[stage].carg1 == settings->op[stage].aarg2;
6269         else if (settings->op[stage].cop == WINED3D_TOP_SELECT_ARG2
6270                 && settings->op[stage].aop == WINED3D_TOP_SELECT_ARG1)
6271             op_equal = settings->op[stage].carg2 == settings->op[stage].aarg1;
6272         else if (settings->op[stage].cop == WINED3D_TOP_SELECT_ARG2
6273                 && settings->op[stage].aop == WINED3D_TOP_SELECT_ARG2)
6274             op_equal = settings->op[stage].carg2 == settings->op[stage].aarg2;
6275         else
6276             op_equal = settings->op[stage].aop   == settings->op[stage].cop
6277                     && settings->op[stage].carg0 == settings->op[stage].aarg0
6278                     && settings->op[stage].carg1 == settings->op[stage].aarg1
6279                     && settings->op[stage].carg2 == settings->op[stage].aarg2;
6280
6281         if (settings->op[stage].aop == WINED3D_TOP_DISABLE)
6282         {
6283             gen_ffp_instr(&buffer, stage, TRUE, FALSE, settings->op[stage].dst,
6284                           settings->op[stage].cop, settings->op[stage].carg0,
6285                           settings->op[stage].carg1, settings->op[stage].carg2);
6286             if (!stage)
6287                 shader_addline(&buffer, "MOV ret.w, fragment.color.primary.w;\n");
6288         }
6289         else if (op_equal)
6290         {
6291             gen_ffp_instr(&buffer, stage, TRUE, TRUE, settings->op[stage].dst,
6292                           settings->op[stage].cop, settings->op[stage].carg0,
6293                           settings->op[stage].carg1, settings->op[stage].carg2);
6294         } else {
6295             gen_ffp_instr(&buffer, stage, TRUE, FALSE, settings->op[stage].dst,
6296                           settings->op[stage].cop, settings->op[stage].carg0,
6297                           settings->op[stage].carg1, settings->op[stage].carg2);
6298             gen_ffp_instr(&buffer, stage, FALSE, TRUE, settings->op[stage].dst,
6299                           settings->op[stage].aop, settings->op[stage].aarg0,
6300                           settings->op[stage].aarg1, settings->op[stage].aarg2);
6301         }
6302     }
6303
6304     if(settings->sRGB_write) {
6305         shader_addline(&buffer, "MAD ret, fragment.color.secondary, specular_enable, %s;\n", final_combiner_src);
6306         arbfp_add_sRGB_correction(&buffer, "ret", "arg0", "arg1", "arg2", "tempreg", FALSE);
6307         shader_addline(&buffer, "MOV result.color, ret;\n");
6308     } else {
6309         shader_addline(&buffer, "MAD result.color, fragment.color.secondary, specular_enable, %s;\n", final_combiner_src);
6310     }
6311
6312     /* Footer */
6313     shader_addline(&buffer, "END\n");
6314
6315     /* Generate the shader */
6316     GL_EXTCALL(glGenProgramsARB(1, &ret));
6317     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, ret));
6318     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
6319             strlen(buffer.buffer), buffer.buffer));
6320     checkGLcall("glProgramStringARB()");
6321
6322     gl_info->gl_ops.gl.p_glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
6323     if (pos != -1)
6324     {
6325         FIXME("Fragment program error at position %d: %s\n\n", pos,
6326               debugstr_a((const char *)gl_info->gl_ops.gl.p_glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
6327         shader_arb_dump_program_source(buffer.buffer);
6328     }
6329     else
6330     {
6331         GLint native;
6332
6333         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
6334         checkGLcall("glGetProgramivARB()");
6335         if (!native) WARN("Program exceeds native resource limits.\n");
6336     }
6337
6338     shader_buffer_free(&buffer);
6339     return ret;
6340 }
6341
6342 static void fragment_prog_arbfp(struct wined3d_context *context, const struct wined3d_state *state, DWORD state_id)
6343 {
6344     const struct wined3d_device *device = context->swapchain->device;
6345     const struct wined3d_gl_info *gl_info = context->gl_info;
6346     struct shader_arb_priv *priv = device->fragment_priv;
6347     BOOL use_vshader = use_vs(state);
6348     BOOL use_pshader = use_ps(state);
6349     struct ffp_frag_settings settings;
6350     const struct arbfp_ffp_desc *desc;
6351     unsigned int i;
6352
6353     TRACE("context %p, state %p, state_id %#x.\n", context, state, state_id);
6354
6355     if (isStateDirty(context, STATE_RENDER(WINED3D_RS_FOGENABLE)))
6356     {
6357         if (!use_pshader && device->shader_backend == &arb_program_shader_backend && context->last_was_pshader)
6358         {
6359             /* Reload fixed function constants since they collide with the
6360              * pixel shader constants. */
6361             for (i = 0; i < MAX_TEXTURES; ++i)
6362             {
6363                 set_bumpmat_arbfp(context, state, STATE_TEXTURESTAGE(i, WINED3D_TSS_BUMPENV_MAT00));
6364             }
6365             state_texfactor_arbfp(context, state, STATE_RENDER(WINED3D_RS_TEXTUREFACTOR));
6366             state_arb_specularenable(context, state, STATE_RENDER(WINED3D_RS_SPECULARENABLE));
6367         }
6368         else if (use_pshader && !isStateDirty(context, context->state_table[STATE_VSHADER].representative))
6369         {
6370             device->shader_backend->shader_select(context, use_pshader, use_vshader);
6371         }
6372         return;
6373     }
6374
6375     if (!use_pshader)
6376     {
6377         /* Find or create a shader implementing the fixed function pipeline
6378          * settings, then activate it. */
6379         gen_ffp_frag_op(device, state, &settings, FALSE);
6380         desc = (const struct arbfp_ffp_desc *)find_ffp_frag_shader(&priv->fragment_shaders, &settings);
6381         if(!desc) {
6382             struct arbfp_ffp_desc *new_desc = HeapAlloc(GetProcessHeap(), 0, sizeof(*new_desc));
6383             if (!new_desc)
6384             {
6385                 ERR("Out of memory\n");
6386                 return;
6387             }
6388             new_desc->num_textures_used = 0;
6389             for (i = 0; i < gl_info->limits.texture_stages; ++i)
6390             {
6391                 if (settings.op[i].cop == WINED3D_TOP_DISABLE)
6392                     break;
6393                 new_desc->num_textures_used = i;
6394             }
6395
6396             memcpy(&new_desc->parent.settings, &settings, sizeof(settings));
6397             new_desc->shader = gen_arbfp_ffp_shader(&settings, gl_info);
6398             add_ffp_frag_shader(&priv->fragment_shaders, &new_desc->parent);
6399             TRACE("Allocated fixed function replacement shader descriptor %p\n", new_desc);
6400             desc = new_desc;
6401         }
6402
6403         /* Now activate the replacement program. GL_FRAGMENT_PROGRAM_ARB is already active (however, note the
6404          * comment above the shader_select call below). If e.g. GLSL is active, the shader_select call will
6405          * deactivate it.
6406          */
6407         GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, desc->shader));
6408         checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, desc->shader)");
6409         priv->current_fprogram_id = desc->shader;
6410
6411         if (device->shader_backend == &arb_program_shader_backend && context->last_was_pshader)
6412         {
6413             /* Reload fixed function constants since they collide with the
6414              * pixel shader constants. */
6415             for (i = 0; i < MAX_TEXTURES; ++i)
6416             {
6417                 set_bumpmat_arbfp(context, state, STATE_TEXTURESTAGE(i, WINED3D_TSS_BUMPENV_MAT00));
6418             }
6419             state_texfactor_arbfp(context, state, STATE_RENDER(WINED3D_RS_TEXTUREFACTOR));
6420             state_arb_specularenable(context, state, STATE_RENDER(WINED3D_RS_SPECULARENABLE));
6421         }
6422         context->last_was_pshader = FALSE;
6423     } else {
6424         context->last_was_pshader = TRUE;
6425     }
6426
6427     /* Finally, select the shader. If a pixel shader is used, it will be set and enabled by the shader backend.
6428      * If this shader backend is arbfp(most likely), then it will simply overwrite the last fixed function
6429      * replacement shader. If the shader backend is not ARB, it currently is important that the opengl implementation
6430      * type overwrites GL_ARB_fragment_program. This is currently the case with GLSL. If we really want to use
6431      * atifs or nvrc pixel shaders with arb fragment programs we'd have to disable GL_FRAGMENT_PROGRAM_ARB here
6432      *
6433      * Don't call shader_select if the vertex shader is dirty, because it will be called later on by the vertex
6434      * shader handler.
6435      */
6436     if (!isStateDirty(context, context->state_table[STATE_VSHADER].representative))
6437     {
6438         device->shader_backend->shader_select(context, use_pshader, use_vshader);
6439
6440         if (!isStateDirty(context, STATE_VERTEXSHADERCONSTANT) && (use_vshader || use_pshader))
6441             context_apply_state(context, state, STATE_VERTEXSHADERCONSTANT);
6442     }
6443     if (use_pshader)
6444         context_apply_state(context, state, STATE_PIXELSHADERCONSTANT);
6445 }
6446
6447 /* We can't link the fog states to the fragment state directly since the
6448  * vertex pipeline links them to FOGENABLE. A different linking in different
6449  * pipeline parts can't be expressed in the combined state table, so we need
6450  * to handle that with a forwarding function. The other invisible side effect
6451  * is that changing the fog start and fog end (which links to FOGENABLE in
6452  * vertex) results in the fragment_prog_arbfp function being called because
6453  * FOGENABLE is dirty, which calls this function here. */
6454 static void state_arbfp_fog(struct wined3d_context *context, const struct wined3d_state *state, DWORD state_id)
6455 {
6456     enum fogsource new_source;
6457
6458     TRACE("context %p, state %p, state_id %#x.\n", context, state, state_id);
6459
6460     if (!isStateDirty(context, STATE_PIXELSHADER))
6461         fragment_prog_arbfp(context, state, state_id);
6462
6463     if (!state->render_states[WINED3D_RS_FOGENABLE])
6464         return;
6465
6466     if (state->render_states[WINED3D_RS_FOGTABLEMODE] == WINED3D_FOG_NONE)
6467     {
6468         if (use_vs(state))
6469         {
6470             new_source = FOGSOURCE_VS;
6471         }
6472         else
6473         {
6474             if (state->render_states[WINED3D_RS_FOGVERTEXMODE] == WINED3D_FOG_NONE || context->last_was_rhw)
6475                 new_source = FOGSOURCE_COORD;
6476             else
6477                 new_source = FOGSOURCE_FFP;
6478         }
6479     }
6480     else
6481     {
6482         new_source = FOGSOURCE_FFP;
6483     }
6484
6485     if (new_source != context->fog_source)
6486     {
6487         context->fog_source = new_source;
6488         state_fogstartend(context, state, STATE_RENDER(WINED3D_RS_FOGSTART));
6489     }
6490 }
6491
6492 static void textransform(struct wined3d_context *context, const struct wined3d_state *state, DWORD state_id)
6493 {
6494     if (!isStateDirty(context, STATE_PIXELSHADER))
6495         fragment_prog_arbfp(context, state, state_id);
6496 }
6497
6498 static void state_srgbwrite(struct wined3d_context *context, const struct wined3d_state *state, DWORD state_id)
6499 {
6500     const struct wined3d_gl_info *gl_info = context->gl_info;
6501     const struct wined3d_surface *rt = state->fb->render_targets[0];
6502
6503     TRACE("context %p, state %p, state_id %#x.\n", context, state, state_id);
6504
6505     if (state->render_states[WINED3D_RS_SRGBWRITEENABLE]
6506             && rt->resource.format->flags & WINED3DFMT_FLAG_SRGB_WRITE)
6507         gl_info->gl_ops.gl.p_glEnable(GL_FRAMEBUFFER_SRGB);
6508     else
6509         gl_info->gl_ops.gl.p_glDisable(GL_FRAMEBUFFER_SRGB);
6510 }
6511
6512 static const struct StateEntryTemplate arbfp_fragmentstate_template[] =
6513 {
6514     {STATE_RENDER(WINED3D_RS_TEXTUREFACTOR),              { STATE_RENDER(WINED3D_RS_TEXTUREFACTOR),             state_texfactor_arbfp   }, WINED3D_GL_EXT_NONE             },
6515     {STATE_TEXTURESTAGE(0, WINED3D_TSS_COLOR_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6516     {STATE_TEXTURESTAGE(0, WINED3D_TSS_COLOR_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6517     {STATE_TEXTURESTAGE(0, WINED3D_TSS_COLOR_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6518     {STATE_TEXTURESTAGE(0, WINED3D_TSS_COLOR_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6519     {STATE_TEXTURESTAGE(0, WINED3D_TSS_ALPHA_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6520     {STATE_TEXTURESTAGE(0, WINED3D_TSS_ALPHA_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6521     {STATE_TEXTURESTAGE(0, WINED3D_TSS_ALPHA_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6522     {STATE_TEXTURESTAGE(0, WINED3D_TSS_ALPHA_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6523     {STATE_TEXTURESTAGE(0, WINED3D_TSS_RESULT_ARG),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6524     {STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6525     {STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6526     {STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6527     {STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6528     {STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6529     {STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6530     {STATE_TEXTURESTAGE(1, WINED3D_TSS_COLOR_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6531     {STATE_TEXTURESTAGE(1, WINED3D_TSS_COLOR_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6532     {STATE_TEXTURESTAGE(1, WINED3D_TSS_COLOR_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6533     {STATE_TEXTURESTAGE(1, WINED3D_TSS_COLOR_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6534     {STATE_TEXTURESTAGE(1, WINED3D_TSS_ALPHA_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6535     {STATE_TEXTURESTAGE(1, WINED3D_TSS_ALPHA_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6536     {STATE_TEXTURESTAGE(1, WINED3D_TSS_ALPHA_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6537     {STATE_TEXTURESTAGE(1, WINED3D_TSS_ALPHA_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6538     {STATE_TEXTURESTAGE(1, WINED3D_TSS_RESULT_ARG),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6539     {STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6540     {STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6541     {STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6542     {STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6543     {STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6544     {STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6545     {STATE_TEXTURESTAGE(2, WINED3D_TSS_COLOR_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6546     {STATE_TEXTURESTAGE(2, WINED3D_TSS_COLOR_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6547     {STATE_TEXTURESTAGE(2, WINED3D_TSS_COLOR_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6548     {STATE_TEXTURESTAGE(2, WINED3D_TSS_COLOR_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6549     {STATE_TEXTURESTAGE(2, WINED3D_TSS_ALPHA_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6550     {STATE_TEXTURESTAGE(2, WINED3D_TSS_ALPHA_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6551     {STATE_TEXTURESTAGE(2, WINED3D_TSS_ALPHA_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6552     {STATE_TEXTURESTAGE(2, WINED3D_TSS_ALPHA_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6553     {STATE_TEXTURESTAGE(2, WINED3D_TSS_RESULT_ARG),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6554     {STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6555     {STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6556     {STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6557     {STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6558     {STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6559     {STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6560     {STATE_TEXTURESTAGE(3, WINED3D_TSS_COLOR_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6561     {STATE_TEXTURESTAGE(3, WINED3D_TSS_COLOR_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6562     {STATE_TEXTURESTAGE(3, WINED3D_TSS_COLOR_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6563     {STATE_TEXTURESTAGE(3, WINED3D_TSS_COLOR_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6564     {STATE_TEXTURESTAGE(3, WINED3D_TSS_ALPHA_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6565     {STATE_TEXTURESTAGE(3, WINED3D_TSS_ALPHA_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6566     {STATE_TEXTURESTAGE(3, WINED3D_TSS_ALPHA_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6567     {STATE_TEXTURESTAGE(3, WINED3D_TSS_ALPHA_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6568     {STATE_TEXTURESTAGE(3, WINED3D_TSS_RESULT_ARG),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6569     {STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6570     {STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6571     {STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6572     {STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6573     {STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6574     {STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6575     {STATE_TEXTURESTAGE(4, WINED3D_TSS_COLOR_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6576     {STATE_TEXTURESTAGE(4, WINED3D_TSS_COLOR_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6577     {STATE_TEXTURESTAGE(4, WINED3D_TSS_COLOR_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6578     {STATE_TEXTURESTAGE(4, WINED3D_TSS_COLOR_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6579     {STATE_TEXTURESTAGE(4, WINED3D_TSS_ALPHA_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6580     {STATE_TEXTURESTAGE(4, WINED3D_TSS_ALPHA_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6581     {STATE_TEXTURESTAGE(4, WINED3D_TSS_ALPHA_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6582     {STATE_TEXTURESTAGE(4, WINED3D_TSS_ALPHA_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6583     {STATE_TEXTURESTAGE(4, WINED3D_TSS_RESULT_ARG),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6584     {STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6585     {STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6586     {STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6587     {STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6588     {STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6589     {STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6590     {STATE_TEXTURESTAGE(5, WINED3D_TSS_COLOR_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6591     {STATE_TEXTURESTAGE(5, WINED3D_TSS_COLOR_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6592     {STATE_TEXTURESTAGE(5, WINED3D_TSS_COLOR_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6593     {STATE_TEXTURESTAGE(5, WINED3D_TSS_COLOR_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6594     {STATE_TEXTURESTAGE(5, WINED3D_TSS_ALPHA_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6595     {STATE_TEXTURESTAGE(5, WINED3D_TSS_ALPHA_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6596     {STATE_TEXTURESTAGE(5, WINED3D_TSS_ALPHA_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6597     {STATE_TEXTURESTAGE(5, WINED3D_TSS_ALPHA_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6598     {STATE_TEXTURESTAGE(5, WINED3D_TSS_RESULT_ARG),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6599     {STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6600     {STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6601     {STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6602     {STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6603     {STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6604     {STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6605     {STATE_TEXTURESTAGE(6, WINED3D_TSS_COLOR_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6606     {STATE_TEXTURESTAGE(6, WINED3D_TSS_COLOR_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6607     {STATE_TEXTURESTAGE(6, WINED3D_TSS_COLOR_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6608     {STATE_TEXTURESTAGE(6, WINED3D_TSS_COLOR_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6609     {STATE_TEXTURESTAGE(6, WINED3D_TSS_ALPHA_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6610     {STATE_TEXTURESTAGE(6, WINED3D_TSS_ALPHA_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6611     {STATE_TEXTURESTAGE(6, WINED3D_TSS_ALPHA_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6612     {STATE_TEXTURESTAGE(6, WINED3D_TSS_ALPHA_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6613     {STATE_TEXTURESTAGE(6, WINED3D_TSS_RESULT_ARG),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6614     {STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6615     {STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6616     {STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6617     {STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6618     {STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6619     {STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6620     {STATE_TEXTURESTAGE(7, WINED3D_TSS_COLOR_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6621     {STATE_TEXTURESTAGE(7, WINED3D_TSS_COLOR_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6622     {STATE_TEXTURESTAGE(7, WINED3D_TSS_COLOR_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6623     {STATE_TEXTURESTAGE(7, WINED3D_TSS_COLOR_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6624     {STATE_TEXTURESTAGE(7, WINED3D_TSS_ALPHA_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6625     {STATE_TEXTURESTAGE(7, WINED3D_TSS_ALPHA_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6626     {STATE_TEXTURESTAGE(7, WINED3D_TSS_ALPHA_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6627     {STATE_TEXTURESTAGE(7, WINED3D_TSS_ALPHA_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6628     {STATE_TEXTURESTAGE(7, WINED3D_TSS_RESULT_ARG),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6629     {STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6630     {STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6631     {STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6632     {STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6633     {STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6634     {STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6635     {STATE_PIXELSHADER,                                   { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6636     {STATE_RENDER(WINED3D_RS_FOGENABLE),                  { STATE_RENDER(WINED3D_RS_FOGENABLE),                 state_arbfp_fog         }, WINED3D_GL_EXT_NONE             },
6637     {STATE_RENDER(WINED3D_RS_FOGTABLEMODE),               { STATE_RENDER(WINED3D_RS_FOGENABLE),                 NULL                    }, WINED3D_GL_EXT_NONE             },
6638     {STATE_RENDER(WINED3D_RS_FOGVERTEXMODE),              { STATE_RENDER(WINED3D_RS_FOGENABLE),                 NULL                    }, WINED3D_GL_EXT_NONE             },
6639     {STATE_RENDER(WINED3D_RS_FOGSTART),                   { STATE_RENDER(WINED3D_RS_FOGSTART),                  state_fogstartend       }, WINED3D_GL_EXT_NONE             },
6640     {STATE_RENDER(WINED3D_RS_FOGEND),                     { STATE_RENDER(WINED3D_RS_FOGSTART),                  NULL                    }, WINED3D_GL_EXT_NONE             },
6641     {STATE_RENDER(WINED3D_RS_SRGBWRITEENABLE),            { STATE_RENDER(WINED3D_RS_SRGBWRITEENABLE),           state_srgbwrite         }, ARB_FRAMEBUFFER_SRGB            },
6642     {STATE_RENDER(WINED3D_RS_SRGBWRITEENABLE),            { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6643     {STATE_RENDER(WINED3D_RS_FOGCOLOR),                   { STATE_RENDER(WINED3D_RS_FOGCOLOR),                  state_fogcolor          }, WINED3D_GL_EXT_NONE             },
6644     {STATE_RENDER(WINED3D_RS_FOGDENSITY),                 { STATE_RENDER(WINED3D_RS_FOGDENSITY),                state_fogdensity        }, WINED3D_GL_EXT_NONE             },
6645     {STATE_TEXTURESTAGE(0,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(0, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6646     {STATE_TEXTURESTAGE(1,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(1, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6647     {STATE_TEXTURESTAGE(2,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(2, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6648     {STATE_TEXTURESTAGE(3,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(3, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6649     {STATE_TEXTURESTAGE(4,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(4, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6650     {STATE_TEXTURESTAGE(5,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(5, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6651     {STATE_TEXTURESTAGE(6,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(6, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6652     {STATE_TEXTURESTAGE(7,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(7, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6653     {STATE_RENDER(WINED3D_RS_SPECULARENABLE),             { STATE_RENDER(WINED3D_RS_SPECULARENABLE),            state_arb_specularenable}, WINED3D_GL_EXT_NONE             },
6654     {0 /* Terminate */,                                   { 0,                                                  0                       }, WINED3D_GL_EXT_NONE             },
6655 };
6656
6657 const struct fragment_pipeline arbfp_fragment_pipeline = {
6658     arbfp_enable,
6659     arbfp_get_caps,
6660     arbfp_alloc,
6661     arbfp_free,
6662     shader_arb_color_fixup_supported,
6663     arbfp_fragmentstate_template,
6664     TRUE /* We can disable projected textures */
6665 };
6666
6667 struct arbfp_blit_priv {
6668     GLenum yuy2_rect_shader, yuy2_2d_shader;
6669     GLenum uyvy_rect_shader, uyvy_2d_shader;
6670     GLenum yv12_rect_shader, yv12_2d_shader;
6671     GLenum p8_rect_shader, p8_2d_shader;
6672     GLuint palette_texture;
6673 };
6674
6675 static HRESULT arbfp_blit_alloc(struct wined3d_device *device)
6676 {
6677     device->blit_priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(struct arbfp_blit_priv));
6678     if(!device->blit_priv) {
6679         ERR("Out of memory\n");
6680         return E_OUTOFMEMORY;
6681     }
6682     return WINED3D_OK;
6683 }
6684
6685 /* Context activation is done by the caller. */
6686 static void arbfp_blit_free(struct wined3d_device *device)
6687 {
6688     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
6689     struct arbfp_blit_priv *priv = device->blit_priv;
6690
6691     ENTER_GL();
6692     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yuy2_rect_shader));
6693     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yuy2_2d_shader));
6694     GL_EXTCALL(glDeleteProgramsARB(1, &priv->uyvy_rect_shader));
6695     GL_EXTCALL(glDeleteProgramsARB(1, &priv->uyvy_2d_shader));
6696     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yv12_rect_shader));
6697     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yv12_2d_shader));
6698     GL_EXTCALL(glDeleteProgramsARB(1, &priv->p8_rect_shader));
6699     GL_EXTCALL(glDeleteProgramsARB(1, &priv->p8_2d_shader));
6700     checkGLcall("Delete yuv and p8 programs");
6701
6702     if (priv->palette_texture)
6703         gl_info->gl_ops.gl.p_glDeleteTextures(1, &priv->palette_texture);
6704     LEAVE_GL();
6705
6706     HeapFree(GetProcessHeap(), 0, device->blit_priv);
6707     device->blit_priv = NULL;
6708 }
6709
6710 static BOOL gen_planar_yuv_read(struct wined3d_shader_buffer *buffer, enum complex_fixup fixup,
6711         GLenum textype, char *luminance)
6712 {
6713     char chroma;
6714     const char *tex, *texinstr;
6715
6716     if (fixup == COMPLEX_FIXUP_UYVY) {
6717         chroma = 'x';
6718         *luminance = 'w';
6719     } else {
6720         chroma = 'w';
6721         *luminance = 'x';
6722     }
6723     switch(textype) {
6724         case GL_TEXTURE_2D:             tex = "2D";     texinstr = "TXP"; break;
6725         case GL_TEXTURE_RECTANGLE_ARB:  tex = "RECT";   texinstr = "TEX"; break;
6726         default:
6727             /* This is more tricky than just replacing the texture type - we have to navigate
6728              * properly in the texture to find the correct chroma values
6729              */
6730             FIXME("Implement yuv correction for non-2d, non-rect textures\n");
6731             return FALSE;
6732     }
6733
6734     /* First we have to read the chroma values. This means we need at least two pixels(no filtering),
6735      * or 4 pixels(with filtering). To get the unmodified chromas, we have to rid ourselves of the
6736      * filtering when we sample the texture.
6737      *
6738      * These are the rules for reading the chroma:
6739      *
6740      * Even pixel: Cr
6741      * Even pixel: U
6742      * Odd pixel: V
6743      *
6744      * So we have to get the sampling x position in non-normalized coordinates in integers
6745      */
6746     if(textype != GL_TEXTURE_RECTANGLE_ARB) {
6747         shader_addline(buffer, "MUL texcrd.xy, fragment.texcoord[0], size.x;\n");
6748         shader_addline(buffer, "MOV texcrd.w, size.x;\n");
6749     } else {
6750         shader_addline(buffer, "MOV texcrd, fragment.texcoord[0];\n");
6751     }
6752     /* We must not allow filtering between pixel x and x+1, this would mix U and V
6753      * Vertical filtering is ok. However, bear in mind that the pixel center is at
6754      * 0.5, so add 0.5.
6755      */
6756     shader_addline(buffer, "FLR texcrd.x, texcrd.x;\n");
6757     shader_addline(buffer, "ADD texcrd.x, texcrd.x, coef.y;\n");
6758
6759     /* Divide the x coordinate by 0.5 and get the fraction. This gives 0.25 and 0.75 for the
6760      * even and odd pixels respectively
6761      */
6762     shader_addline(buffer, "MUL texcrd2, texcrd, coef.y;\n");
6763     shader_addline(buffer, "FRC texcrd2, texcrd2;\n");
6764
6765     /* Sample Pixel 1 */
6766     shader_addline(buffer, "%s luminance, texcrd, texture[0], %s;\n", texinstr, tex);
6767
6768     /* Put the value into either of the chroma values */
6769     shader_addline(buffer, "SGE temp.x, texcrd2.x, coef.y;\n");
6770     shader_addline(buffer, "MUL chroma.x, luminance.%c, temp.x;\n", chroma);
6771     shader_addline(buffer, "SLT temp.x, texcrd2.x, coef.y;\n");
6772     shader_addline(buffer, "MUL chroma.y, luminance.%c, temp.x;\n", chroma);
6773
6774     /* Sample pixel 2. If we read an even pixel(SLT above returned 1), sample
6775      * the pixel right to the current one. Otherwise, sample the left pixel.
6776      * Bias and scale the SLT result to -1;1 and add it to the texcrd.x.
6777      */
6778     shader_addline(buffer, "MAD temp.x, temp.x, coef.z, -coef.x;\n");
6779     shader_addline(buffer, "ADD texcrd.x, texcrd, temp.x;\n");
6780     shader_addline(buffer, "%s luminance, texcrd, texture[0], %s;\n", texinstr, tex);
6781
6782     /* Put the value into the other chroma */
6783     shader_addline(buffer, "SGE temp.x, texcrd2.x, coef.y;\n");
6784     shader_addline(buffer, "MAD chroma.y, luminance.%c, temp.x, chroma.y;\n", chroma);
6785     shader_addline(buffer, "SLT temp.x, texcrd2.x, coef.y;\n");
6786     shader_addline(buffer, "MAD chroma.x, luminance.%c, temp.x, chroma.x;\n", chroma);
6787
6788     /* TODO: If filtering is enabled, sample a 2nd pair of pixels left or right of
6789      * the current one and lerp the two U and V values
6790      */
6791
6792     /* This gives the correctly filtered luminance value */
6793     shader_addline(buffer, "TEX luminance, fragment.texcoord[0], texture[0], %s;\n", tex);
6794
6795     return TRUE;
6796 }
6797
6798 static BOOL gen_yv12_read(struct wined3d_shader_buffer *buffer, GLenum textype, char *luminance)
6799 {
6800     const char *tex;
6801
6802     switch(textype) {
6803         case GL_TEXTURE_2D:             tex = "2D";     break;
6804         case GL_TEXTURE_RECTANGLE_ARB:  tex = "RECT";   break;
6805         default:
6806             FIXME("Implement yv12 correction for non-2d, non-rect textures\n");
6807             return FALSE;
6808     }
6809
6810     /* YV12 surfaces contain a WxH sized luminance plane, followed by a (W/2)x(H/2)
6811      * V and a (W/2)x(H/2) U plane, each with 8 bit per pixel. So the effective
6812      * bitdepth is 12 bits per pixel. Since the U and V planes have only half the
6813      * pitch of the luminance plane, the packing into the gl texture is a bit
6814      * unfortunate. If the whole texture is interpreted as luminance data it looks
6815      * approximately like this:
6816      *
6817      *        +----------------------------------+----
6818      *        |                                  |
6819      *        |                                  |
6820      *        |                                  |
6821      *        |                                  |
6822      *        |                                  |   2
6823      *        |            LUMINANCE             |   -
6824      *        |                                  |   3
6825      *        |                                  |
6826      *        |                                  |
6827      *        |                                  |
6828      *        |                                  |
6829      *        +----------------+-----------------+----
6830      *        |                |                 |
6831      *        |  U even rows   |  U odd rows     |
6832      *        |                |                 |   1
6833      *        +----------------+------------------   -
6834      *        |                |                 |   3
6835      *        |  V even rows   |  V odd rows     |
6836      *        |                |                 |
6837      *        +----------------+-----------------+----
6838      *        |                |                 |
6839      *        |     0.5        |       0.5       |
6840      *
6841      * So it appears as if there are 4 chroma images, but in fact the odd rows
6842      * in the chroma images are in the same row as the even ones. So its is
6843      * kinda tricky to read
6844      *
6845      * When reading from rectangle textures, keep in mind that the input y coordinates
6846      * go from 0 to d3d_height, whereas the opengl texture height is 1.5 * d3d_height
6847      */
6848     shader_addline(buffer, "PARAM yv12_coef = {%f, %f, %f, %f};\n",
6849             2.0f / 3.0f, 1.0f / 6.0f, (2.0f / 3.0f) + (1.0f / 6.0f), 1.0f / 3.0f);
6850
6851     shader_addline(buffer, "MOV texcrd, fragment.texcoord[0];\n");
6852     /* the chroma planes have only half the width */
6853     shader_addline(buffer, "MUL texcrd.x, texcrd.x, coef.y;\n");
6854
6855     /* The first value is between 2/3 and 5/6th of the texture's height, so scale+bias
6856      * the coordinate. Also read the right side of the image when reading odd lines
6857      *
6858      * Don't forget to clamp the y values in into the range, otherwise we'll get filtering
6859      * bleeding
6860      */
6861     if(textype == GL_TEXTURE_2D) {
6862
6863         shader_addline(buffer, "RCP chroma.w, size.y;\n");
6864
6865         shader_addline(buffer, "MUL texcrd2.y, texcrd.y, size.y;\n");
6866
6867         shader_addline(buffer, "FLR texcrd2.y, texcrd2.y;\n");
6868         shader_addline(buffer, "MAD texcrd.y, texcrd.y, yv12_coef.y, yv12_coef.x;\n");
6869
6870         /* Read odd lines from the right side(add size * 0.5 to the x coordinate */
6871         shader_addline(buffer, "ADD texcrd2.x, texcrd2.y, yv12_coef.y;\n"); /* To avoid 0.5 == 0.5 comparisons */
6872         shader_addline(buffer, "FRC texcrd2.x, texcrd2.x;\n");
6873         shader_addline(buffer, "SGE texcrd2.x, texcrd2.x, coef.y;\n");
6874         shader_addline(buffer, "MAD texcrd.x, texcrd2.x, coef.y, texcrd.x;\n");
6875
6876         /* clamp, keep the half pixel origin in mind */
6877         shader_addline(buffer, "MAD temp.y, coef.y, chroma.w, yv12_coef.x;\n");
6878         shader_addline(buffer, "MAX texcrd.y, temp.y, texcrd.y;\n");
6879         shader_addline(buffer, "MAD temp.y, -coef.y, chroma.w, yv12_coef.z;\n");
6880         shader_addline(buffer, "MIN texcrd.y, temp.y, texcrd.y;\n");
6881     } else {
6882         /* Read from [size - size+size/4] */
6883         shader_addline(buffer, "FLR texcrd.y, texcrd.y;\n");
6884         shader_addline(buffer, "MAD texcrd.y, texcrd.y, coef.w, size.y;\n");
6885
6886         /* Read odd lines from the right side(add size * 0.5 to the x coordinate */
6887         shader_addline(buffer, "ADD texcrd2.x, texcrd.y, yv12_coef.y;\n"); /* To avoid 0.5 == 0.5 comparisons */
6888         shader_addline(buffer, "FRC texcrd2.x, texcrd2.x;\n");
6889         shader_addline(buffer, "SGE texcrd2.x, texcrd2.x, coef.y;\n");
6890         shader_addline(buffer, "MUL texcrd2.x, texcrd2.x, size.x;\n");
6891         shader_addline(buffer, "MAD texcrd.x, texcrd2.x, coef.y, texcrd.x;\n");
6892
6893         /* Make sure to read exactly from the pixel center */
6894         shader_addline(buffer, "FLR texcrd.y, texcrd.y;\n");
6895         shader_addline(buffer, "ADD texcrd.y, texcrd.y, coef.y;\n");
6896
6897         /* Clamp */
6898         shader_addline(buffer, "MAD temp.y, size.y, coef.w, size.y;\n");
6899         shader_addline(buffer, "ADD temp.y, temp.y, -coef.y;\n");
6900         shader_addline(buffer, "MIN texcrd.y, temp.y, texcrd.y;\n");
6901         shader_addline(buffer, "ADD temp.y, size.y, -coef.y;\n");
6902         shader_addline(buffer, "MAX texcrd.y, temp.y, texcrd.y;\n");
6903     }
6904     /* Read the texture, put the result into the output register */
6905     shader_addline(buffer, "TEX temp, texcrd, texture[0], %s;\n", tex);
6906     shader_addline(buffer, "MOV chroma.x, temp.w;\n");
6907
6908     /* The other chroma value is 1/6th of the texture lower, from 5/6th to 6/6th
6909      * No need to clamp because we're just reusing the already clamped value from above
6910      */
6911     if(textype == GL_TEXTURE_2D) {
6912         shader_addline(buffer, "ADD texcrd.y, texcrd.y, yv12_coef.y;\n");
6913     } else {
6914         shader_addline(buffer, "MAD texcrd.y, size.y, coef.w, texcrd.y;\n");
6915     }
6916     shader_addline(buffer, "TEX temp, texcrd, texture[0], %s;\n", tex);
6917     shader_addline(buffer, "MOV chroma.y, temp.w;\n");
6918
6919     /* Sample the luminance value. It is in the top 2/3rd of the texture, so scale the y coordinate.
6920      * Clamp the y coordinate to prevent the chroma values from bleeding into the sampled luminance
6921      * values due to filtering
6922      */
6923     shader_addline(buffer, "MOV texcrd, fragment.texcoord[0];\n");
6924     if(textype == GL_TEXTURE_2D) {
6925         /* Multiply the y coordinate by 2/3 and clamp it */
6926         shader_addline(buffer, "MUL texcrd.y, texcrd.y, yv12_coef.x;\n");
6927         shader_addline(buffer, "MAD temp.y, -coef.y, chroma.w, yv12_coef.x;\n");
6928         shader_addline(buffer, "MIN texcrd.y, temp.y, texcrd.y;\n");
6929         shader_addline(buffer, "TEX luminance, texcrd, texture[0], %s;\n", tex);
6930     } else {
6931         /* Reading from texture_rectangles is pretty straightforward, just use the unmodified
6932          * texture coordinate. It is still a good idea to clamp it though, since the opengl texture
6933          * is bigger
6934          */
6935         shader_addline(buffer, "ADD temp.x, size.y, -coef.y;\n");
6936         shader_addline(buffer, "MIN texcrd.y, texcrd.y, size.x;\n");
6937         shader_addline(buffer, "TEX luminance, texcrd, texture[0], %s;\n", tex);
6938     }
6939     *luminance = 'a';
6940
6941     return TRUE;
6942 }
6943
6944 static GLuint gen_p8_shader(struct arbfp_blit_priv *priv,
6945         const struct wined3d_gl_info *gl_info, GLenum textype)
6946 {
6947     GLenum shader;
6948     struct wined3d_shader_buffer buffer;
6949     GLint pos;
6950
6951     /* Shader header */
6952     if (!shader_buffer_init(&buffer))
6953     {
6954         ERR("Failed to initialize shader buffer.\n");
6955         return 0;
6956     }
6957
6958     ENTER_GL();
6959     GL_EXTCALL(glGenProgramsARB(1, &shader));
6960     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader));
6961     LEAVE_GL();
6962     if(!shader) {
6963         shader_buffer_free(&buffer);
6964         return 0;
6965     }
6966
6967     shader_addline(&buffer, "!!ARBfp1.0\n");
6968     shader_addline(&buffer, "TEMP index;\n");
6969
6970     /* { 255/256, 0.5/255*255/256, 0, 0 } */
6971     shader_addline(&buffer, "PARAM constants = { 0.996, 0.00195, 0, 0 };\n");
6972
6973     /* The alpha-component contains the palette index */
6974     if(textype == GL_TEXTURE_RECTANGLE_ARB)
6975         shader_addline(&buffer, "TXP index, fragment.texcoord[0], texture[0], RECT;\n");
6976     else
6977         shader_addline(&buffer, "TEX index, fragment.texcoord[0], texture[0], 2D;\n");
6978
6979     /* Scale the index by 255/256 and add a bias of '0.5' in order to sample in the middle */
6980     shader_addline(&buffer, "MAD index.a, index.a, constants.x, constants.y;\n");
6981
6982     /* Use the alpha-component as an index in the palette to get the final color */
6983     shader_addline(&buffer, "TEX result.color, index.a, texture[1], 1D;\n");
6984     shader_addline(&buffer, "END\n");
6985
6986     ENTER_GL();
6987     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
6988             strlen(buffer.buffer), buffer.buffer));
6989     checkGLcall("glProgramStringARB()");
6990
6991     gl_info->gl_ops.gl.p_glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
6992     if (pos != -1)
6993     {
6994         FIXME("Fragment program error at position %d: %s\n\n", pos,
6995               debugstr_a((const char *)gl_info->gl_ops.gl.p_glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
6996         shader_arb_dump_program_source(buffer.buffer);
6997     }
6998
6999     if (textype == GL_TEXTURE_RECTANGLE_ARB)
7000         priv->p8_rect_shader = shader;
7001     else
7002         priv->p8_2d_shader = shader;
7003
7004     shader_buffer_free(&buffer);
7005     LEAVE_GL();
7006
7007     return shader;
7008 }
7009
7010 /* Context activation is done by the caller. */
7011 static void upload_palette(const struct wined3d_surface *surface, struct wined3d_context *context)
7012 {
7013     BYTE table[256][4];
7014     struct wined3d_device *device = surface->resource.device;
7015     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
7016     struct arbfp_blit_priv *priv = device->blit_priv;
7017     BOOL colorkey = (surface->CKeyFlags & WINEDDSD_CKSRCBLT) != 0;
7018
7019     d3dfmt_p8_init_palette(surface, table, colorkey);
7020
7021     ENTER_GL();
7022
7023     if (gl_info->supported[APPLE_CLIENT_STORAGE])
7024     {
7025         gl_info->gl_ops.gl.p_glPixelStorei(GL_UNPACK_CLIENT_STORAGE_APPLE, GL_FALSE);
7026         checkGLcall("glPixelStorei(GL_UNPACK_CLIENT_STORAGE_APPLE, GL_FALSE)");
7027     }
7028
7029     if (!priv->palette_texture)
7030         gl_info->gl_ops.gl.p_glGenTextures(1, &priv->palette_texture);
7031
7032     GL_EXTCALL(glActiveTextureARB(GL_TEXTURE1));
7033     gl_info->gl_ops.gl.p_glBindTexture(GL_TEXTURE_1D, priv->palette_texture);
7034
7035     gl_info->gl_ops.gl.p_glTexEnvi(GL_TEXTURE_ENV, GL_TEXTURE_ENV_MODE, GL_REPLACE);
7036
7037     gl_info->gl_ops.gl.p_glTexParameteri(GL_TEXTURE_1D, GL_TEXTURE_MIN_FILTER, GL_NEAREST);
7038     /* Make sure we have discrete color levels. */
7039     gl_info->gl_ops.gl.p_glTexParameteri(GL_TEXTURE_1D, GL_TEXTURE_MAG_FILTER, GL_NEAREST);
7040     gl_info->gl_ops.gl.p_glTexParameteri(GL_TEXTURE_1D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
7041     /* Upload the palette */
7042     /* TODO: avoid unneeded uploads in the future by adding some SFLAG_PALETTE_DIRTY mechanism */
7043     gl_info->gl_ops.gl.p_glTexImage1D(GL_TEXTURE_1D, 0, GL_RGBA, 256, 0, GL_RGBA, GL_UNSIGNED_BYTE, table);
7044
7045     if (gl_info->supported[APPLE_CLIENT_STORAGE])
7046     {
7047         gl_info->gl_ops.gl.p_glPixelStorei(GL_UNPACK_CLIENT_STORAGE_APPLE, GL_TRUE);
7048         checkGLcall("glPixelStorei(GL_UNPACK_CLIENT_STORAGE_APPLE, GL_TRUE)");
7049     }
7050
7051     /* Switch back to unit 0 in which the 2D texture will be stored. */
7052     context_active_texture(context, gl_info, 0);
7053     LEAVE_GL();
7054 }
7055
7056 /* Context activation is done by the caller. */
7057 static GLuint gen_yuv_shader(struct arbfp_blit_priv *priv, const struct wined3d_gl_info *gl_info,
7058         enum complex_fixup yuv_fixup, GLenum textype)
7059 {
7060     GLenum shader;
7061     struct wined3d_shader_buffer buffer;
7062     char luminance_component;
7063     GLint pos;
7064
7065     /* Shader header */
7066     if (!shader_buffer_init(&buffer))
7067     {
7068         ERR("Failed to initialize shader buffer.\n");
7069         return 0;
7070     }
7071
7072     ENTER_GL();
7073     GL_EXTCALL(glGenProgramsARB(1, &shader));
7074     checkGLcall("GL_EXTCALL(glGenProgramsARB(1, &shader))");
7075     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader));
7076     checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader)");
7077     LEAVE_GL();
7078     if(!shader) {
7079         shader_buffer_free(&buffer);
7080         return 0;
7081     }
7082
7083     /* The YUY2 and UYVY formats contain two pixels packed into a 32 bit macropixel,
7084      * giving effectively 16 bit per pixel. The color consists of a luminance(Y) and
7085      * two chroma(U and V) values. Each macropixel has two luminance values, one for
7086      * each single pixel it contains, and one U and one V value shared between both
7087      * pixels.
7088      *
7089      * The data is loaded into an A8L8 texture. With YUY2, the luminance component
7090      * contains the luminance and alpha the chroma. With UYVY it is vice versa. Thus
7091      * take the format into account when generating the read swizzles
7092      *
7093      * Reading the Y value is straightforward - just sample the texture. The hardware
7094      * takes care of filtering in the horizontal and vertical direction.
7095      *
7096      * Reading the U and V values is harder. We have to avoid filtering horizontally,
7097      * because that would mix the U and V values of one pixel or two adjacent pixels.
7098      * Thus floor the texture coordinate and add 0.5 to get an unfiltered read,
7099      * regardless of the filtering setting. Vertical filtering works automatically
7100      * though - the U and V values of two rows are mixed nicely.
7101      *
7102      * Apart of avoiding filtering issues, the code has to know which value it just
7103      * read, and where it can find the other one. To determine this, it checks if
7104      * it sampled an even or odd pixel, and shifts the 2nd read accordingly.
7105      *
7106      * Handling horizontal filtering of U and V values requires reading a 2nd pair
7107      * of pixels, extracting U and V and mixing them. This is not implemented yet.
7108      *
7109      * An alternative implementation idea is to load the texture as A8R8G8B8 texture,
7110      * with width / 2. This way one read gives all 3 values, finding U and V is easy
7111      * in an unfiltered situation. Finding the luminance on the other hand requires
7112      * finding out if it is an odd or even pixel. The real drawback of this approach
7113      * is filtering. This would have to be emulated completely in the shader, reading
7114      * up two 2 packed pixels in up to 2 rows and interpolating both horizontally and
7115      * vertically. Beyond that it would require adjustments to the texture handling
7116      * code to deal with the width scaling
7117      */
7118     shader_addline(&buffer, "!!ARBfp1.0\n");
7119     shader_addline(&buffer, "TEMP luminance;\n");
7120     shader_addline(&buffer, "TEMP temp;\n");
7121     shader_addline(&buffer, "TEMP chroma;\n");
7122     shader_addline(&buffer, "TEMP texcrd;\n");
7123     shader_addline(&buffer, "TEMP texcrd2;\n");
7124     shader_addline(&buffer, "PARAM coef = {1.0, 0.5, 2.0, 0.25};\n");
7125     shader_addline(&buffer, "PARAM yuv_coef = {1.403, 0.344, 0.714, 1.770};\n");
7126     shader_addline(&buffer, "PARAM size = program.local[0];\n");
7127
7128     switch (yuv_fixup)
7129     {
7130         case COMPLEX_FIXUP_UYVY:
7131         case COMPLEX_FIXUP_YUY2:
7132             if (!gen_planar_yuv_read(&buffer, yuv_fixup, textype, &luminance_component))
7133             {
7134                 shader_buffer_free(&buffer);
7135                 return 0;
7136             }
7137             break;
7138
7139         case COMPLEX_FIXUP_YV12:
7140             if (!gen_yv12_read(&buffer, textype, &luminance_component))
7141             {
7142                 shader_buffer_free(&buffer);
7143                 return 0;
7144             }
7145             break;
7146
7147         default:
7148             FIXME("Unsupported YUV fixup %#x\n", yuv_fixup);
7149             shader_buffer_free(&buffer);
7150             return 0;
7151     }
7152
7153     /* Calculate the final result. Formula is taken from
7154      * http://www.fourcc.org/fccyvrgb.php. Note that the chroma
7155      * ranges from -0.5 to 0.5
7156      */
7157     shader_addline(&buffer, "SUB chroma.xy, chroma, coef.y;\n");
7158
7159     shader_addline(&buffer, "MAD result.color.x, chroma.x, yuv_coef.x, luminance.%c;\n", luminance_component);
7160     shader_addline(&buffer, "MAD temp.x, -chroma.y, yuv_coef.y, luminance.%c;\n", luminance_component);
7161     shader_addline(&buffer, "MAD result.color.y, -chroma.x, yuv_coef.z, temp.x;\n");
7162     shader_addline(&buffer, "MAD result.color.z, chroma.y, yuv_coef.w, luminance.%c;\n", luminance_component);
7163     shader_addline(&buffer, "END\n");
7164
7165     ENTER_GL();
7166     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
7167             strlen(buffer.buffer), buffer.buffer));
7168     checkGLcall("glProgramStringARB()");
7169
7170     gl_info->gl_ops.gl.p_glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
7171     if (pos != -1)
7172     {
7173         FIXME("Fragment program error at position %d: %s\n\n", pos,
7174               debugstr_a((const char *)gl_info->gl_ops.gl.p_glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
7175         shader_arb_dump_program_source(buffer.buffer);
7176     }
7177     else
7178     {
7179         GLint native;
7180
7181         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
7182         checkGLcall("glGetProgramivARB()");
7183         if (!native) WARN("Program exceeds native resource limits.\n");
7184     }
7185
7186     shader_buffer_free(&buffer);
7187     LEAVE_GL();
7188
7189     switch (yuv_fixup)
7190     {
7191         case COMPLEX_FIXUP_YUY2:
7192             if (textype == GL_TEXTURE_RECTANGLE_ARB) priv->yuy2_rect_shader = shader;
7193             else priv->yuy2_2d_shader = shader;
7194             break;
7195
7196         case COMPLEX_FIXUP_UYVY:
7197             if (textype == GL_TEXTURE_RECTANGLE_ARB) priv->uyvy_rect_shader = shader;
7198             else priv->uyvy_2d_shader = shader;
7199             break;
7200
7201         case COMPLEX_FIXUP_YV12:
7202             if (textype == GL_TEXTURE_RECTANGLE_ARB) priv->yv12_rect_shader = shader;
7203             else priv->yv12_2d_shader = shader;
7204             break;
7205         default:
7206             ERR("Unsupported complex fixup: %d\n", yuv_fixup);
7207     }
7208
7209     return shader;
7210 }
7211
7212 /* Context activation is done by the caller. */
7213 static HRESULT arbfp_blit_set(void *blit_priv, struct wined3d_context *context, const struct wined3d_surface *surface)
7214 {
7215     GLenum shader;
7216     float size[4] = {(float) surface->pow2Width, (float) surface->pow2Height, 1.0f, 1.0f};
7217     struct arbfp_blit_priv *priv = blit_priv;
7218     enum complex_fixup fixup;
7219     const struct wined3d_gl_info *gl_info = context->gl_info;
7220     GLenum textype;
7221
7222     if (surface->container.type == WINED3D_CONTAINER_TEXTURE)
7223         textype = surface->container.u.texture->target;
7224     else
7225         textype = surface->texture_target;
7226
7227     if (surface->flags & SFLAG_CONVERTED)
7228     {
7229         ENTER_GL();
7230         gl_info->gl_ops.gl.p_glEnable(textype);
7231         checkGLcall("glEnable(textype)");
7232         LEAVE_GL();
7233         return WINED3D_OK;
7234     }
7235
7236     if (!is_complex_fixup(surface->resource.format->color_fixup))
7237     {
7238         TRACE("Fixup:\n");
7239         dump_color_fixup_desc(surface->resource.format->color_fixup);
7240         /* Don't bother setting up a shader for unconverted formats */
7241         ENTER_GL();
7242         gl_info->gl_ops.gl.p_glEnable(textype);
7243         checkGLcall("glEnable(textype)");
7244         LEAVE_GL();
7245         return WINED3D_OK;
7246     }
7247
7248     fixup = get_complex_fixup(surface->resource.format->color_fixup);
7249
7250     switch(fixup)
7251     {
7252         case COMPLEX_FIXUP_YUY2:
7253             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->yuy2_rect_shader : priv->yuy2_2d_shader;
7254             break;
7255
7256         case COMPLEX_FIXUP_UYVY:
7257             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->uyvy_rect_shader : priv->uyvy_2d_shader;
7258             break;
7259
7260         case COMPLEX_FIXUP_YV12:
7261             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->yv12_rect_shader : priv->yv12_2d_shader;
7262             break;
7263
7264         case COMPLEX_FIXUP_P8:
7265             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->p8_rect_shader : priv->p8_2d_shader;
7266             if (!shader) shader = gen_p8_shader(priv, gl_info, textype);
7267
7268             upload_palette(surface, context);
7269             break;
7270
7271         default:
7272             FIXME("Unsupported complex fixup %#x, not setting a shader\n", fixup);
7273             ENTER_GL();
7274             gl_info->gl_ops.gl.p_glEnable(textype);
7275             checkGLcall("glEnable(textype)");
7276             LEAVE_GL();
7277             return E_NOTIMPL;
7278     }
7279
7280     if (!shader) shader = gen_yuv_shader(priv, gl_info, fixup, textype);
7281
7282     ENTER_GL();
7283     gl_info->gl_ops.gl.p_glEnable(GL_FRAGMENT_PROGRAM_ARB);
7284     checkGLcall("glEnable(GL_FRAGMENT_PROGRAM_ARB)");
7285     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader));
7286     checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader)");
7287     GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, 0, size));
7288     checkGLcall("glProgramLocalParameter4fvARB");
7289     LEAVE_GL();
7290
7291     return WINED3D_OK;
7292 }
7293
7294 /* Context activation is done by the caller. */
7295 static void arbfp_blit_unset(const struct wined3d_gl_info *gl_info)
7296 {
7297     ENTER_GL();
7298     gl_info->gl_ops.gl.p_glDisable(GL_FRAGMENT_PROGRAM_ARB);
7299     checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
7300     gl_info->gl_ops.gl.p_glDisable(GL_TEXTURE_2D);
7301     checkGLcall("glDisable(GL_TEXTURE_2D)");
7302     if (gl_info->supported[ARB_TEXTURE_CUBE_MAP])
7303     {
7304         gl_info->gl_ops.gl.p_glDisable(GL_TEXTURE_CUBE_MAP_ARB);
7305         checkGLcall("glDisable(GL_TEXTURE_CUBE_MAP_ARB)");
7306     }
7307     if (gl_info->supported[ARB_TEXTURE_RECTANGLE])
7308     {
7309         gl_info->gl_ops.gl.p_glDisable(GL_TEXTURE_RECTANGLE_ARB);
7310         checkGLcall("glDisable(GL_TEXTURE_RECTANGLE_ARB)");
7311     }
7312     LEAVE_GL();
7313 }
7314
7315 static BOOL arbfp_blit_supported(const struct wined3d_gl_info *gl_info, enum wined3d_blit_op blit_op,
7316         const RECT *src_rect, DWORD src_usage, enum wined3d_pool src_pool, const struct wined3d_format *src_format,
7317         const RECT *dst_rect, DWORD dst_usage, enum wined3d_pool dst_pool, const struct wined3d_format *dst_format)
7318 {
7319     enum complex_fixup src_fixup;
7320
7321     if (!gl_info->supported[ARB_FRAGMENT_PROGRAM])
7322         return FALSE;
7323
7324     if (blit_op != WINED3D_BLIT_OP_COLOR_BLIT)
7325     {
7326         TRACE("Unsupported blit_op=%d\n", blit_op);
7327         return FALSE;
7328     }
7329
7330     if (src_pool == WINED3D_POOL_SYSTEM_MEM || dst_pool == WINED3D_POOL_SYSTEM_MEM)
7331         return FALSE;
7332
7333     src_fixup = get_complex_fixup(src_format->color_fixup);
7334     if (TRACE_ON(d3d_shader) && TRACE_ON(d3d))
7335     {
7336         TRACE("Checking support for fixup:\n");
7337         dump_color_fixup_desc(src_format->color_fixup);
7338     }
7339
7340     if (!is_identity_fixup(dst_format->color_fixup))
7341     {
7342         TRACE("Destination fixups are not supported\n");
7343         return FALSE;
7344     }
7345
7346     if (is_identity_fixup(src_format->color_fixup))
7347     {
7348         TRACE("[OK]\n");
7349         return TRUE;
7350     }
7351
7352      /* We only support YUV conversions. */
7353     if (!is_complex_fixup(src_format->color_fixup))
7354     {
7355         TRACE("[FAILED]\n");
7356         return FALSE;
7357     }
7358
7359     switch(src_fixup)
7360     {
7361         case COMPLEX_FIXUP_YUY2:
7362         case COMPLEX_FIXUP_UYVY:
7363         case COMPLEX_FIXUP_YV12:
7364         case COMPLEX_FIXUP_P8:
7365             TRACE("[OK]\n");
7366             return TRUE;
7367
7368         default:
7369             FIXME("Unsupported YUV fixup %#x\n", src_fixup);
7370             TRACE("[FAILED]\n");
7371             return FALSE;
7372     }
7373 }
7374
7375 HRESULT arbfp_blit_surface(struct wined3d_device *device, DWORD filter,
7376         struct wined3d_surface *src_surface, const RECT *src_rect_in,
7377         struct wined3d_surface *dst_surface, const RECT *dst_rect_in)
7378 {
7379     struct wined3d_context *context;
7380     RECT src_rect = *src_rect_in;
7381     RECT dst_rect = *dst_rect_in;
7382
7383     /* Now load the surface */
7384     if (wined3d_settings.offscreen_rendering_mode != ORM_FBO
7385             && (src_surface->flags & (SFLAG_INTEXTURE | SFLAG_INDRAWABLE)) == SFLAG_INDRAWABLE)
7386     {
7387         /* Without FBO blits transferring from the drawable to the texture is
7388          * expensive, because we have to flip the data in sysmem. Since we can
7389          * flip in the blitter, we don't actually need that flip anyway. So we
7390          * use the surface's texture as scratch texture, and flip the source
7391          * rectangle instead. */
7392         surface_load_fb_texture(src_surface, FALSE);
7393
7394         src_rect.top = src_surface->resource.height - src_rect.top;
7395         src_rect.bottom = src_surface->resource.height - src_rect.bottom;
7396     }
7397     else
7398         surface_internal_preload(src_surface, SRGB_RGB);
7399
7400     /* Activate the destination context, set it up for blitting */
7401     context = context_acquire(device, dst_surface);
7402     context_apply_blit_state(context, device);
7403
7404     if (!surface_is_offscreen(dst_surface))
7405         surface_translate_drawable_coords(dst_surface, context->win_handle, &dst_rect);
7406
7407     arbfp_blit_set(device->blit_priv, context, src_surface);
7408
7409     ENTER_GL();
7410
7411     /* Draw a textured quad */
7412     draw_textured_quad(src_surface, context, &src_rect, &dst_rect, filter);
7413
7414     LEAVE_GL();
7415
7416     /* Leave the opengl state valid for blitting */
7417     arbfp_blit_unset(context->gl_info);
7418
7419     if (wined3d_settings.strict_draw_ordering
7420             || (dst_surface->container.type == WINED3D_CONTAINER_SWAPCHAIN
7421             && (dst_surface->container.u.swapchain->front_buffer == dst_surface)))
7422         context->gl_info->gl_ops.gl.p_glFlush(); /* Flush to ensure ordering across contexts. */
7423
7424     context_release(context);
7425
7426     surface_modify_location(dst_surface, dst_surface->draw_binding, TRUE);
7427     return WINED3D_OK;
7428 }
7429
7430 /* Do not call while under the GL lock. */
7431 static HRESULT arbfp_blit_color_fill(struct wined3d_device *device, struct wined3d_surface *dst_surface,
7432         const RECT *dst_rect, const struct wined3d_color *color)
7433 {
7434     FIXME("Color filling not implemented by arbfp_blit\n");
7435     return WINED3DERR_INVALIDCALL;
7436 }
7437
7438 /* Do not call while under the GL lock. */
7439 static HRESULT arbfp_blit_depth_fill(struct wined3d_device *device,
7440         struct wined3d_surface *surface, const RECT *rect, float depth)
7441 {
7442     FIXME("Depth filling not implemented by arbfp_blit.\n");
7443     return WINED3DERR_INVALIDCALL;
7444 }
7445
7446 const struct blit_shader arbfp_blit = {
7447     arbfp_blit_alloc,
7448     arbfp_blit_free,
7449     arbfp_blit_set,
7450     arbfp_blit_unset,
7451     arbfp_blit_supported,
7452     arbfp_blit_color_fill,
7453     arbfp_blit_depth_fill,
7454 };