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 2007-2011, 2013 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 static BOOL shader_is_pshader_version(enum wined3d_shader_type type)
  43 {
  44     return type == WINED3D_SHADER_TYPE_PIXEL;
  45 }
  46
  47 static BOOL shader_is_vshader_version(enum wined3d_shader_type type)
  48 {
  49     return type == WINED3D_SHADER_TYPE_VERTEX;
  50 }
  51
  52 /* Extract a line. Note that this modifies the source string. */
  53 static char *get_line(char **ptr)
  54 {
  55     char *p, *q;
  56
  57     p = *ptr;
  58     if (!(q = strstr(p, "\n")))
  59     {
  60         if (!*p) return NULL;
  61         *ptr += strlen(p);
  62         return p;
  63     }
  64     *q = '\0';
  65     *ptr = q + 1;
  66
  67     return p;
  68 }
  69
  70 static void shader_arb_dump_program_source(const char *source)
  71 {
  72     ULONG source_size;
  73     char *ptr, *line, *tmp;
  74
  75     source_size = strlen(source) + 1;
  76     tmp = HeapAlloc(GetProcessHeap(), 0, source_size);
  77     if (!tmp)
  78     {
  79         ERR("Failed to allocate %u bytes for shader source.\n", source_size);
  80         return;
  81     }
  82     memcpy(tmp, source, source_size);
  83
  84     ptr = tmp;
  85     while ((line = get_line(&ptr))) FIXME("    %s\n", line);
  86     FIXME("\n");
  87
  88     HeapFree(GetProcessHeap(), 0, tmp);
  89 }
  90
  91 enum arb_helper_value
  92 {
  93     ARB_ZERO,
  94     ARB_ONE,
  95     ARB_TWO,
  96     ARB_0001,
  97     ARB_EPS,
  98
  99     ARB_VS_REL_OFFSET
 100 };
 101
 102 static const char *arb_get_helper_value(enum wined3d_shader_type shader, enum arb_helper_value value)
 103 {
 104     if (shader == WINED3D_SHADER_TYPE_GEOMETRY)
 105     {
 106         ERR("Geometry shaders are unsupported\n");
 107         return "bad";
 108     }
 109
 110     if (shader == WINED3D_SHADER_TYPE_PIXEL)
 111     {
 112         switch (value)
 113         {
 114             case ARB_ZERO: return "ps_helper_const.x";
 115             case ARB_ONE: return "ps_helper_const.y";
 116             case ARB_TWO: return "coefmul.x";
 117             case ARB_0001: return "ps_helper_const.xxxy";
 118             case ARB_EPS: return "ps_helper_const.z";
 119             default: break;
 120         }
 121     }
 122     else
 123     {
 124         switch (value)
 125         {
 126             case ARB_ZERO: return "helper_const.x";
 127             case ARB_ONE: return "helper_const.y";
 128             case ARB_TWO: return "helper_const.z";
 129             case ARB_EPS: return "helper_const.w";
 130             case ARB_0001: return "helper_const.xxxy";
 131             case ARB_VS_REL_OFFSET: return "rel_addr_const.y";
 132         }
 133     }
 134     FIXME("Unmanaged %s shader helper constant requested: %u\n",
 135           shader == WINED3D_SHADER_TYPE_PIXEL ? "pixel" : "vertex", value);
 136     switch (value)
 137     {
 138         case ARB_ZERO: return "0.0";
 139         case ARB_ONE: return "1.0";
 140         case ARB_TWO: return "2.0";
 141         case ARB_0001: return "{0.0, 0.0, 0.0, 1.0}";
 142         case ARB_EPS: return "1e-8";
 143         default: return "bad";
 144     }
 145 }
 146
 147 static inline BOOL ffp_clip_emul(const struct wined3d_state *state)
 148 {
 149     return state->lowest_disabled_stage < 7;
 150 }
 151
 152 /* ARB_program_shader private data */
 153
 154 struct control_frame
 155 {
 156     struct                          list entry;
 157     enum
 158     {
 159         IF,
 160         IFC,
 161         LOOP,
 162         REP
 163     } type;
 164     BOOL                            muting;
 165     BOOL                            outer_loop;
 166     union
 167     {
 168         unsigned int                loop;
 169         unsigned int                ifc;
 170     } no;
 171     struct wined3d_shader_loop_control loop_control;
 172     BOOL                            had_else;
 173 };
 174
 175 struct arb_ps_np2fixup_info
 176 {
 177     struct ps_np2fixup_info         super;
 178     /* For ARB we need a offset value:
 179      * With both GLSL and ARB mode the NP2 fixup information (the texture dimensions) are stored in a
 180      * consecutive way (GLSL uses a uniform array). Since ARB doesn't know the notion of a "standalone"
 181      * array we need an offset to the index inside the program local parameter array. */
 182     UINT                            offset;
 183 };
 184
 185 struct arb_ps_compile_args
 186 {
 187     struct ps_compile_args          super;
 188     WORD                            bools;
 189     WORD                            clip;  /* only a boolean, use a WORD for alignment */
 190     unsigned char                   loop_ctrl[MAX_CONST_I][3];
 191 };
 192
 193 struct stb_const_desc
 194 {
 195     unsigned char           texunit;
 196     UINT                    const_num;
 197 };
 198
 199 struct arb_ps_compiled_shader
 200 {
 201     struct arb_ps_compile_args      args;
 202     struct arb_ps_np2fixup_info     np2fixup_info;
 203     struct stb_const_desc           bumpenvmatconst[MAX_TEXTURES];
 204     struct stb_const_desc           luminanceconst[MAX_TEXTURES];
 205     UINT                            int_consts[MAX_CONST_I];
 206     GLuint                          prgId;
 207     UINT                            ycorrection;
 208     unsigned char                   numbumpenvmatconsts;
 209     char                            num_int_consts;
 210 };
 211
 212 struct arb_vs_compile_args
 213 {
 214     struct vs_compile_args          super;
 215     union
 216     {
 217         struct
 218         {
 219             WORD                    bools;
 220             unsigned char           clip_texcoord;
 221             unsigned char           clipplane_mask;
 222         }                           boolclip;
 223         DWORD                       boolclip_compare;
 224     } clip;
 225     DWORD                           ps_signature;
 226     union
 227     {
 228         unsigned char               samplers[4];
 229         DWORD                       samplers_compare;
 230     } vertex;
 231     unsigned char                   loop_ctrl[MAX_CONST_I][3];
 232 };
 233
 234 struct arb_vs_compiled_shader
 235 {
 236     struct arb_vs_compile_args      args;
 237     GLuint                          prgId;
 238     UINT                            int_consts[MAX_CONST_I];
 239     char                            num_int_consts;
 240     char                            need_color_unclamp;
 241     UINT                            pos_fixup;
 242 };
 243
 244 struct recorded_instruction
 245 {
 246     struct wined3d_shader_instruction ins;
 247     struct list entry;
 248 };
 249
 250 struct shader_arb_ctx_priv
 251 {
 252     char addr_reg[20];
 253     enum
 254     {
 255         /* plain GL_ARB_vertex_program or GL_ARB_fragment_program */
 256         ARB,
 257         /* GL_NV_vertex_progam2_option or GL_NV_fragment_program_option */
 258         NV2,
 259         /* GL_NV_vertex_program3 or GL_NV_fragment_program2 */
 260         NV3
 261     } target_version;
 262
 263     const struct arb_vs_compile_args    *cur_vs_args;
 264     const struct arb_ps_compile_args    *cur_ps_args;
 265     const struct arb_ps_compiled_shader *compiled_fprog;
 266     const struct arb_vs_compiled_shader *compiled_vprog;
 267     struct arb_ps_np2fixup_info         *cur_np2fixup_info;
 268     struct list                         control_frames;
 269     struct list                         record;
 270     BOOL                                recording;
 271     BOOL                                muted;
 272     unsigned int                        num_loops, loop_depth, num_ifcs;
 273     int                                 aL;
 274
 275     unsigned int                        vs_clipplanes;
 276     BOOL                                footer_written;
 277     BOOL                                in_main_func;
 278
 279     /* For 3.0 vertex shaders */
 280     const char                          *vs_output[MAX_REG_OUTPUT];
 281     /* For 2.x and earlier vertex shaders */
 282     const char                          *texcrd_output[8], *color_output[2], *fog_output;
 283
 284     /* 3.0 pshader input for compatibility with fixed function */
 285     const char                          *ps_input[MAX_REG_INPUT];
 286 };
 287
 288 struct ps_signature
 289 {
 290     struct wined3d_shader_signature_element *sig;
 291     DWORD                               idx;
 292     struct wine_rb_entry                entry;
 293 };
 294
 295 struct arb_pshader_private {
 296     struct arb_ps_compiled_shader   *gl_shaders;
 297     UINT                            num_gl_shaders, shader_array_size;
 298     DWORD                           input_signature_idx;
 299     DWORD                           clipplane_emulation;
 300     BOOL                            clamp_consts;
 301 };
 302
 303 struct arb_vshader_private {
 304     struct arb_vs_compiled_shader   *gl_shaders;
 305     UINT                            num_gl_shaders, shader_array_size;
 306     UINT rel_offset;
 307 };
 308
 309 struct shader_arb_priv
 310 {
 311     GLuint                  current_vprogram_id;
 312     GLuint                  current_fprogram_id;
 313     const struct arb_ps_compiled_shader *compiled_fprog;
 314     const struct arb_vs_compiled_shader *compiled_vprog;
 315     GLuint                  depth_blt_vprogram_id;
 316     GLuint                  depth_blt_fprogram_id_full[tex_type_count];
 317     GLuint                  depth_blt_fprogram_id_masked[tex_type_count];
 318     BOOL                    use_arbfp_fixed_func;
 319     struct wine_rb_tree     fragment_shaders;
 320     BOOL                    last_ps_const_clamped;
 321     BOOL                    last_vs_color_unclamp;
 322
 323     struct wine_rb_tree     signature_tree;
 324     DWORD ps_sig_number;
 325
 326     unsigned int highest_dirty_ps_const, highest_dirty_vs_const;
 327     char *vshader_const_dirty, *pshader_const_dirty;
 328     const struct wined3d_context *last_context;
 329
 330     const struct wined3d_vertex_pipe_ops *vertex_pipe;
 331     const struct fragment_pipeline *fragment_pipe;
 332     BOOL ffp_proj_control;
 333 };
 334
 335 /* Context activation for state handlers is done by the caller. */
 336
 337 static BOOL need_rel_addr_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 (shader_data->rel_offset) return TRUE;
 341     if (!reg_maps->usesmova) return FALSE;
 342     return !gl_info->supported[NV_VERTEX_PROGRAM2_OPTION];
 343 }
 344
 345 /* Returns TRUE if result.clip from GL_NV_vertex_program2 should be used and FALSE otherwise */
 346 static inline BOOL use_nv_clip(const struct wined3d_gl_info *gl_info)
 347 {
 348     return gl_info->supported[NV_VERTEX_PROGRAM2_OPTION]
 349             && !(gl_info->quirks & WINED3D_QUIRK_NV_CLIP_BROKEN);
 350 }
 351
 352 static BOOL need_helper_const(const struct arb_vshader_private *shader_data,
 353         const struct wined3d_shader_reg_maps *reg_maps, const struct wined3d_gl_info *gl_info)
 354 {
 355     if (need_rel_addr_const(shader_data, reg_maps, gl_info)) return TRUE;
 356     if (!gl_info->supported[NV_VERTEX_PROGRAM]) return TRUE; /* Need to init colors. */
 357     if (gl_info->quirks & WINED3D_QUIRK_ARB_VS_OFFSET_LIMIT) return TRUE; /* Load the immval offset. */
 358     if (gl_info->quirks & WINED3D_QUIRK_SET_TEXCOORD_W) return TRUE; /* Have to init texcoords. */
 359     if (!use_nv_clip(gl_info)) return TRUE; /* Init the clip texcoord */
 360     if (reg_maps->usesnrm) return TRUE; /* 0.0 */
 361     if (reg_maps->usespow) return TRUE; /* EPS, 0.0 and 1.0 */
 362     if (reg_maps->fog) return TRUE; /* Clamping fog coord, 0.0 and 1.0 */
 363     return FALSE;
 364 }
 365
 366 static unsigned int reserved_vs_const(const struct arb_vshader_private *shader_data,
 367         const struct wined3d_shader_reg_maps *reg_maps, const struct wined3d_gl_info *gl_info)
 368 {
 369     unsigned int ret = 1;
 370     /* We use one PARAM for the pos fixup, and in some cases one to load
 371      * some immediate values into the shader. */
 372     if (need_helper_const(shader_data, reg_maps, gl_info)) ++ret;
 373     if (need_rel_addr_const(shader_data, reg_maps, gl_info)) ++ret;
 374     return ret;
 375 }
 376
 377 /* Loads floating point constants into the currently set ARB_vertex/fragment_program.
 378  * When constant_list == NULL, it will load all the constants.
 379  *
 380  * @target_type should be either GL_VERTEX_PROGRAM_ARB (for vertex shaders)
 381  *  or GL_FRAGMENT_PROGRAM_ARB (for pixel shaders)
 382  */
 383 /* Context activation is done by the caller. */
 384 static unsigned int shader_arb_load_constantsF(const struct wined3d_shader *shader,
 385         const struct wined3d_gl_info *gl_info, GLuint target_type, unsigned int max_constants,
 386         const float *constants, char *dirty_consts)
 387 {
 388     struct wined3d_shader_lconst *lconst;
 389     DWORD i, j;
 390     unsigned int ret;
 391
 392     if (TRACE_ON(d3d_constants))
 393     {
 394         for(i = 0; i < max_constants; i++) {
 395             if(!dirty_consts[i]) continue;
 396             TRACE_(d3d_constants)("Loading constants %i: %f, %f, %f, %f\n", i,
 397                         constants[i * 4 + 0], constants[i * 4 + 1],
 398                         constants[i * 4 + 2], constants[i * 4 + 3]);
 399         }
 400     }
 401
 402     i = 0;
 403
 404     /* In 1.X pixel shaders constants are implicitly clamped in the range [-1;1] */
 405     if (target_type == GL_FRAGMENT_PROGRAM_ARB && shader->reg_maps.shader_version.major == 1)
 406     {
 407         float lcl_const[4];
 408         /* ps 1.x supports only 8 constants, clamp only those. When switching between 1.x and higher
 409          * shaders, the first 8 constants are marked dirty for reload
 410          */
 411         for(; i < min(8, max_constants); i++) {
 412             if(!dirty_consts[i]) continue;
 413             dirty_consts[i] = 0;
 414
 415             j = 4 * i;
 416             if (constants[j + 0] > 1.0f) lcl_const[0] = 1.0f;
 417             else if (constants[j + 0] < -1.0f) lcl_const[0] = -1.0f;
 418             else lcl_const[0] = constants[j + 0];
 419
 420             if (constants[j + 1] > 1.0f) lcl_const[1] = 1.0f;
 421             else if (constants[j + 1] < -1.0f) lcl_const[1] = -1.0f;
 422             else lcl_const[1] = constants[j + 1];
 423
 424             if (constants[j + 2] > 1.0f) lcl_const[2] = 1.0f;
 425             else if (constants[j + 2] < -1.0f) lcl_const[2] = -1.0f;
 426             else lcl_const[2] = constants[j + 2];
 427
 428             if (constants[j + 3] > 1.0f) lcl_const[3] = 1.0f;
 429             else if (constants[j + 3] < -1.0f) lcl_const[3] = -1.0f;
 430             else lcl_const[3] = constants[j + 3];
 431
 432             GL_EXTCALL(glProgramEnvParameter4fvARB(target_type, i, lcl_const));
 433         }
 434
 435         /* If further constants are dirty, reload them without clamping.
 436          *
 437          * The alternative is not to touch them, but then we cannot reset the dirty constant count
 438          * to zero. That's bad for apps that only use PS 1.x shaders, because in that case the code
 439          * above would always re-check the first 8 constants since max_constant remains at the init
 440          * value
 441          */
 442     }
 443
 444     if (gl_info->supported[EXT_GPU_PROGRAM_PARAMETERS])
 445     {
 446         /* TODO: Benchmark if we're better of with finding the dirty constants ourselves,
 447          * or just reloading *all* constants at once
 448          *
 449         GL_EXTCALL(glProgramEnvParameters4fvEXT(target_type, i, max_constants, constants + (i * 4)));
 450          */
 451         for(; i < max_constants; i++) {
 452             if(!dirty_consts[i]) continue;
 453
 454             /* Find the next block of dirty constants */
 455             dirty_consts[i] = 0;
 456             j = i;
 457             for(i++; (i < max_constants) && dirty_consts[i]; i++) {
 458                 dirty_consts[i] = 0;
 459             }
 460
 461             GL_EXTCALL(glProgramEnvParameters4fvEXT(target_type, j, i - j, constants + (j * 4)));
 462         }
 463     } else {
 464         for(; i < max_constants; i++) {
 465             if(dirty_consts[i]) {
 466                 dirty_consts[i] = 0;
 467                 GL_EXTCALL(glProgramEnvParameter4fvARB(target_type, i, constants + (i * 4)));
 468             }
 469         }
 470     }
 471     checkGLcall("glProgramEnvParameter4fvARB()");
 472
 473     /* Load immediate constants */
 474     if (shader->load_local_constsF)
 475     {
 476         if (TRACE_ON(d3d_shader))
 477         {
 478             LIST_FOR_EACH_ENTRY(lconst, &shader->constantsF, struct wined3d_shader_lconst, entry)
 479             {
 480                 GLfloat* values = (GLfloat*)lconst->value;
 481                 TRACE_(d3d_constants)("Loading local constants %i: %f, %f, %f, %f\n", lconst->idx,
 482                         values[0], values[1], values[2], values[3]);
 483             }
 484         }
 485         /* Immediate constants are clamped for 1.X shaders at loading times */
 486         ret = 0;
 487         LIST_FOR_EACH_ENTRY(lconst, &shader->constantsF, struct wined3d_shader_lconst, entry)
 488         {
 489             dirty_consts[lconst->idx] = 1; /* Dirtify so the non-immediate constant overwrites it next time */
 490             ret = max(ret, lconst->idx + 1);
 491             GL_EXTCALL(glProgramEnvParameter4fvARB(target_type, lconst->idx, (GLfloat*)lconst->value));
 492         }
 493         checkGLcall("glProgramEnvParameter4fvARB()");
 494         return ret; /* The loaded immediate constants need reloading for the next shader */
 495     } else {
 496         return 0; /* No constants are dirty now */
 497     }
 498 }
 499
 500 /**
 501  * Loads the texture dimensions for NP2 fixup into the currently set ARB_[vertex/fragment]_programs.
 502  */
 503 static void shader_arb_load_np2fixup_constants(void *shader_priv,
 504         const struct wined3d_gl_info *gl_info, const struct wined3d_state *state)
 505 {
 506     const struct shader_arb_priv * priv = shader_priv;
 507
 508     /* NP2 texcoord fixup is (currently) only done for pixelshaders. */
 509     if (!use_ps(state)) return;
 510
 511     if (priv->compiled_fprog && priv->compiled_fprog->np2fixup_info.super.active) {
 512         const struct arb_ps_np2fixup_info* const fixup = &priv->compiled_fprog->np2fixup_info;
 513         UINT i;
 514         WORD active = fixup->super.active;
 515         GLfloat np2fixup_constants[4 * MAX_FRAGMENT_SAMPLERS];
 516
 517         for (i = 0; active; active >>= 1, ++i)
 518         {
 519             const struct wined3d_texture *tex = state->textures[i];
 520             const unsigned char idx = fixup->super.idx[i];
 521             GLfloat *tex_dim = &np2fixup_constants[(idx >> 1) * 4];
 522
 523             if (!(active & 1)) continue;
 524
 525             if (!tex) {
 526                 FIXME("Nonexistent texture is flagged for NP2 texcoord fixup\n");
 527                 continue;
 528             }
 529
 530             if (idx % 2)
 531             {
 532                 tex_dim[2] = tex->pow2_matrix[0];
 533                 tex_dim[3] = tex->pow2_matrix[5];
 534             }
 535             else
 536             {
 537                 tex_dim[0] = tex->pow2_matrix[0];
 538                 tex_dim[1] = tex->pow2_matrix[5];
 539             }
 540         }
 541
 542         for (i = 0; i < fixup->super.num_consts; ++i) {
 543             GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB,
 544                                                    fixup->offset + i, &np2fixup_constants[i * 4]));
 545         }
 546     }
 547 }
 548
 549 /* Context activation is done by the caller. */
 550 static void shader_arb_ps_local_constants(const struct arb_ps_compiled_shader *gl_shader,
 551         const struct wined3d_context *context, const struct wined3d_state *state, UINT rt_height)
 552 {
 553     const struct wined3d_gl_info *gl_info = context->gl_info;
 554     unsigned char i;
 555
 556     for(i = 0; i < gl_shader->numbumpenvmatconsts; i++)
 557     {
 558         int texunit = gl_shader->bumpenvmatconst[i].texunit;
 559
 560         /* The state manager takes care that this function is always called if the bump env matrix changes */
 561         const float *data = (const float *)&state->texture_states[texunit][WINED3D_TSS_BUMPENV_MAT00];
 562         GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB,
 563                 gl_shader->bumpenvmatconst[i].const_num, data));
 564
 565         if (gl_shader->luminanceconst[i].const_num != WINED3D_CONST_NUM_UNUSED)
 566         {
 567             /* WINED3D_TSS_BUMPENVLSCALE and WINED3D_TSS_BUMPENVLOFFSET are next to each other.
 568              * point gl to the scale, and load 4 floats. x = scale, y = offset, z and w are junk, we
 569              * don't care about them. The pointers are valid for sure because the stateblock is bigger.
 570              * (they're WINED3D_TSS_TEXTURETRANSFORMFLAGS and WINED3D_TSS_ADDRESSW, so most likely 0 or NaN
 571             */
 572             const float *scale = (const float *)&state->texture_states[texunit][WINED3D_TSS_BUMPENV_LSCALE];
 573             GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB,
 574                     gl_shader->luminanceconst[i].const_num, scale));
 575         }
 576     }
 577     checkGLcall("Load bumpmap consts");
 578
 579     if(gl_shader->ycorrection != WINED3D_CONST_NUM_UNUSED)
 580     {
 581         /* ycorrection.x: Backbuffer height(onscreen) or 0(offscreen).
 582         * ycorrection.y: -1.0(onscreen), 1.0(offscreen)
 583         * ycorrection.z: 1.0
 584         * ycorrection.w: 0.0
 585         */
 586         float val[4];
 587         val[0] = context->render_offscreen ? 0.0f : (float) rt_height;
 588         val[1] = context->render_offscreen ? 1.0f : -1.0f;
 589         val[2] = 1.0f;
 590         val[3] = 0.0f;
 591         GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, gl_shader->ycorrection, val));
 592         checkGLcall("y correction loading");
 593     }
 594
 595     if (!gl_shader->num_int_consts) return;
 596
 597     for(i = 0; i < MAX_CONST_I; i++)
 598     {
 599         if(gl_shader->int_consts[i] != WINED3D_CONST_NUM_UNUSED)
 600         {
 601             float val[4];
 602             val[0] = (float)state->ps_consts_i[4 * i];
 603             val[1] = (float)state->ps_consts_i[4 * i + 1];
 604             val[2] = (float)state->ps_consts_i[4 * i + 2];
 605             val[3] = -1.0f;
 606
 607             GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, gl_shader->int_consts[i], val));
 608         }
 609     }
 610     checkGLcall("Load ps int consts");
 611 }
 612
 613 /* Context activation is done by the caller. */
 614 static void shader_arb_vs_local_constants(const struct arb_vs_compiled_shader *gl_shader,
 615         const struct wined3d_context *context, const struct wined3d_state *state)
 616 {
 617     const struct wined3d_gl_info *gl_info = context->gl_info;
 618     float position_fixup[4];
 619     unsigned char i;
 620
 621     /* Upload the position fixup */
 622     shader_get_position_fixup(context, state, position_fixup);
 623     GL_EXTCALL(glProgramLocalParameter4fvARB(GL_VERTEX_PROGRAM_ARB, gl_shader->pos_fixup, position_fixup));
 624
 625     if (!gl_shader->num_int_consts) return;
 626
 627     for(i = 0; i < MAX_CONST_I; i++)
 628     {
 629         if(gl_shader->int_consts[i] != WINED3D_CONST_NUM_UNUSED)
 630         {
 631             float val[4];
 632             val[0] = (float)state->vs_consts_i[4 * i];
 633             val[1] = (float)state->vs_consts_i[4 * i + 1];
 634             val[2] = (float)state->vs_consts_i[4 * i + 2];
 635             val[3] = -1.0f;
 636
 637             GL_EXTCALL(glProgramLocalParameter4fvARB(GL_VERTEX_PROGRAM_ARB, gl_shader->int_consts[i], val));
 638         }
 639     }
 640     checkGLcall("Load vs int consts");
 641 }
 642
 643 static void shader_arb_select(const struct wined3d_context *context, enum wined3d_shader_mode vertex_mode,
 644         enum wined3d_shader_mode fragment_mode);
 645
 646 /**
 647  * Loads the app-supplied constants into the currently set ARB_[vertex/fragment]_programs.
 648  *
 649  * We only support float constants in ARB at the moment, so don't
 650  * worry about the Integers or Booleans
 651  */
 652 /* Context activation is done by the caller (state handler). */
 653 static void shader_arb_load_constants_internal(const struct wined3d_context *context,
 654         BOOL usePixelShader, BOOL useVertexShader, BOOL from_shader_select)
 655 {
 656     struct wined3d_device *device = context->swapchain->device;
 657     const struct wined3d_stateblock *stateblock = device->stateBlock;
 658     const struct wined3d_state *state = &stateblock->state;
 659     const struct wined3d_gl_info *gl_info = context->gl_info;
 660     struct shader_arb_priv *priv = device->shader_priv;
 661
 662     if (!from_shader_select)
 663     {
 664         const struct wined3d_shader *vshader = state->vertex_shader, *pshader = state->pixel_shader;
 665         if (vshader
 666                 && (stateblock->changed.vertexShaderConstantsB & vshader->reg_maps.boolean_constants
 667                 || (!gl_info->supported[NV_VERTEX_PROGRAM2_OPTION]
 668                 && (stateblock->changed.vertexShaderConstantsI
 669                 & vshader->reg_maps.integer_constants & ~vshader->reg_maps.local_int_consts))))
 670         {
 671             TRACE("bool/integer vertex shader constants potentially modified, forcing shader reselection.\n");
 672             shader_arb_select(context,
 673                     useVertexShader ? WINED3D_SHADER_MODE_SHADER : WINED3D_SHADER_MODE_FFP,
 674                     usePixelShader ? WINED3D_SHADER_MODE_SHADER : WINED3D_SHADER_MODE_FFP);
 675         }
 676         else if (pshader
 677                 && (stateblock->changed.pixelShaderConstantsB & pshader->reg_maps.boolean_constants
 678                 || (!gl_info->supported[NV_FRAGMENT_PROGRAM_OPTION]
 679                 && (stateblock->changed.pixelShaderConstantsI
 680                 & pshader->reg_maps.integer_constants & ~pshader->reg_maps.local_int_consts))))
 681         {
 682             TRACE("bool/integer pixel shader constants potentially modified, forcing shader reselection.\n");
 683             shader_arb_select(context,
 684                     useVertexShader ? WINED3D_SHADER_MODE_SHADER : WINED3D_SHADER_MODE_FFP,
 685                     usePixelShader ? WINED3D_SHADER_MODE_SHADER : WINED3D_SHADER_MODE_FFP);
 686         }
 687     }
 688
 689     if (context != priv->last_context)
 690     {
 691         memset(priv->vshader_const_dirty, 1,
 692                 sizeof(*priv->vshader_const_dirty) * device->d3d_vshader_constantF);
 693         priv->highest_dirty_vs_const = device->d3d_vshader_constantF;
 694
 695         memset(priv->pshader_const_dirty, 1,
 696                 sizeof(*priv->pshader_const_dirty) * device->d3d_pshader_constantF);
 697         priv->highest_dirty_ps_const = device->d3d_pshader_constantF;
 698
 699         priv->last_context = context;
 700     }
 701
 702     if (useVertexShader)
 703     {
 704         struct wined3d_shader *vshader = state->vertex_shader;
 705         const struct arb_vs_compiled_shader *gl_shader = priv->compiled_vprog;
 706
 707         /* Load DirectX 9 float constants for vertex shader */
 708         priv->highest_dirty_vs_const = shader_arb_load_constantsF(vshader, gl_info, GL_VERTEX_PROGRAM_ARB,
 709                 priv->highest_dirty_vs_const, state->vs_consts_f, priv->vshader_const_dirty);
 710         shader_arb_vs_local_constants(gl_shader, context, state);
 711     }
 712
 713     if (usePixelShader)
 714     {
 715         struct wined3d_shader *pshader = state->pixel_shader;
 716         const struct arb_ps_compiled_shader *gl_shader = priv->compiled_fprog;
 717         UINT rt_height = state->fb->render_targets[0]->resource.height;
 718
 719         /* Load DirectX 9 float constants for pixel shader */
 720         priv->highest_dirty_ps_const = shader_arb_load_constantsF(pshader, gl_info, GL_FRAGMENT_PROGRAM_ARB,
 721                 priv->highest_dirty_ps_const, state->ps_consts_f, priv->pshader_const_dirty);
 722         shader_arb_ps_local_constants(gl_shader, context, state, rt_height);
 723     }
 724 }
 725
 726 static void shader_arb_load_constants(const struct wined3d_context *context, BOOL ps, BOOL vs)
 727 {
 728     shader_arb_load_constants_internal(context, ps, vs, FALSE);
 729 }
 730
 731 static void shader_arb_update_float_vertex_constants(struct wined3d_device *device, UINT start, UINT count)
 732 {
 733     struct wined3d_context *context = context_get_current();
 734     struct shader_arb_priv *priv = device->shader_priv;
 735
 736     /* We don't want shader constant dirtification to be an O(contexts), so just dirtify the active
 737      * context. On a context switch the old context will be fully dirtified */
 738     if (!context || context->swapchain->device != device) return;
 739
 740     memset(priv->vshader_const_dirty + start, 1, sizeof(*priv->vshader_const_dirty) * count);
 741     priv->highest_dirty_vs_const = max(priv->highest_dirty_vs_const, start + count);
 742 }
 743
 744 static void shader_arb_update_float_pixel_constants(struct wined3d_device *device, UINT start, UINT count)
 745 {
 746     struct wined3d_context *context = context_get_current();
 747     struct shader_arb_priv *priv = device->shader_priv;
 748
 749     /* We don't want shader constant dirtification to be an O(contexts), so just dirtify the active
 750      * context. On a context switch the old context will be fully dirtified */
 751     if (!context || context->swapchain->device != device) return;
 752
 753     memset(priv->pshader_const_dirty + start, 1, sizeof(*priv->pshader_const_dirty) * count);
 754     priv->highest_dirty_ps_const = max(priv->highest_dirty_ps_const, start + count);
 755 }
 756
 757 /* Generate the variable & register declarations for the ARB_vertex_program output target */
 758 static void shader_generate_arb_declarations(const struct wined3d_shader *shader,
 759         const struct wined3d_shader_reg_maps *reg_maps, struct wined3d_shader_buffer *buffer,
 760         const struct wined3d_gl_info *gl_info, DWORD *num_clipplanes,
 761         const struct shader_arb_ctx_priv *ctx)
 762 {
 763     DWORD i;
 764     char pshader = shader_is_pshader_version(reg_maps->shader_version.type);
 765     const struct wined3d_shader_lconst *lconst;
 766     unsigned max_constantsF;
 767     DWORD map;
 768
 769     /* In pixel shaders, all private constants are program local, we don't need anything
 770      * from program.env. Thus we can advertise the full set of constants in pixel shaders.
 771      * If we need a private constant the GL implementation will squeeze it in somewhere
 772      *
 773      * With vertex shaders we need the posFixup and on some GL implementations 4 helper
 774      * immediate values. The posFixup is loaded using program.env for now, so always
 775      * subtract one from the number of constants. If the shader uses indirect addressing,
 776      * account for the helper const too because we have to declare all available d3d constants
 777      * and don't know which are actually used.
 778      */
 779     if (pshader)
 780     {
 781         max_constantsF = gl_info->limits.arb_ps_native_constants;
 782         /* 24 is the minimum MAX_PROGRAM_ENV_PARAMETERS_ARB value. */
 783         if (max_constantsF < 24)
 784             max_constantsF = gl_info->limits.arb_ps_float_constants;
 785     }
 786     else
 787     {
 788         const struct arb_vshader_private *shader_data = shader->backend_data;
 789         max_constantsF = gl_info->limits.arb_vs_native_constants;
 790         /* 96 is the minimum MAX_PROGRAM_ENV_PARAMETERS_ARB value.
 791          * Also prevents max_constantsF from becoming less than 0 and
 792          * wrapping . */
 793         if (max_constantsF < 96)
 794             max_constantsF = gl_info->limits.arb_vs_float_constants;
 795
 796         if (reg_maps->usesrelconstF)
 797         {
 798             DWORD highest_constf = 0, clip_limit;
 799
 800             max_constantsF -= reserved_vs_const(shader_data, reg_maps, gl_info);
 801             max_constantsF -= count_bits(reg_maps->integer_constants);
 802             max_constantsF -= gl_info->reserved_arb_constants;
 803
 804             for (i = 0; i < shader->limits.constant_float; ++i)
 805             {
 806                 DWORD idx = i >> 5;
 807                 DWORD shift = i & 0x1f;
 808                 if(reg_maps->constf[idx] & (1 << shift)) highest_constf = i;
 809             }
 810
 811             if(use_nv_clip(gl_info) && ctx->target_version >= NV2)
 812             {
 813                 if(ctx->cur_vs_args->super.clip_enabled)
 814                     clip_limit = gl_info->limits.clipplanes;
 815                 else
 816                     clip_limit = 0;
 817             }
 818             else
 819             {
 820                 unsigned int mask = ctx->cur_vs_args->clip.boolclip.clipplane_mask;
 821                 clip_limit = min(count_bits(mask), 4);
 822             }
 823             *num_clipplanes = min(clip_limit, max_constantsF - highest_constf - 1);
 824             max_constantsF -= *num_clipplanes;
 825             if(*num_clipplanes < clip_limit)
 826             {
 827                 WARN("Only %u clipplanes out of %u enabled\n", *num_clipplanes, gl_info->limits.clipplanes);
 828             }
 829         }
 830         else
 831         {
 832             if (ctx->target_version >= NV2) *num_clipplanes = gl_info->limits.clipplanes;
 833             else *num_clipplanes = min(gl_info->limits.clipplanes, 4);
 834         }
 835     }
 836
 837     for (i = 0, map = reg_maps->temporary; map; map >>= 1, ++i)
 838     {
 839         if (map & 1) shader_addline(buffer, "TEMP R%u;\n", i);
 840     }
 841
 842     for (i = 0, map = reg_maps->address; map; map >>= 1, ++i)
 843     {
 844         if (map & 1) shader_addline(buffer, "ADDRESS A%u;\n", i);
 845     }
 846
 847     if (pshader && reg_maps->shader_version.major == 1 && reg_maps->shader_version.minor <= 3)
 848     {
 849         for (i = 0, map = reg_maps->texcoord; map; map >>= 1, ++i)
 850         {
 851             if (map & 1) shader_addline(buffer, "TEMP T%u;\n", i);
 852         }
 853     }
 854
 855     if (!shader->load_local_constsF)
 856     {
 857         LIST_FOR_EACH_ENTRY(lconst, &shader->constantsF, struct wined3d_shader_lconst, entry)
 858         {
 859             const float *value;
 860             value = (const float *)lconst->value;
 861             shader_addline(buffer, "PARAM C%u = {%.8e, %.8e, %.8e, %.8e};\n", lconst->idx,
 862                            value[0], value[1], value[2], value[3]);
 863         }
 864     }
 865
 866     /* After subtracting privately used constants from the hardware limit(they are loaded as
 867      * local constants), make sure the shader doesn't violate the env constant limit
 868      */
 869     if(pshader)
 870     {
 871         max_constantsF = min(max_constantsF, gl_info->limits.arb_ps_float_constants);
 872     }
 873     else
 874     {
 875         max_constantsF = min(max_constantsF, gl_info->limits.arb_vs_float_constants);
 876     }
 877
 878     /* Avoid declaring more constants than needed */
 879     max_constantsF = min(max_constantsF, shader->limits.constant_float);
 880
 881     /* we use the array-based constants array if the local constants are marked for loading,
 882      * because then we use indirect addressing, or when the local constant list is empty,
 883      * because then we don't know if we're using indirect addressing or not. If we're hardcoding
 884      * local constants do not declare the loaded constants as an array because ARB compilers usually
 885      * do not optimize unused constants away
 886      */
 887     if (reg_maps->usesrelconstF)
 888     {
 889         /* Need to PARAM the environment parameters (constants) so we can use relative addressing */
 890         shader_addline(buffer, "PARAM C[%d] = { program.env[0..%d] };\n",
 891                     max_constantsF, max_constantsF - 1);
 892     } else {
 893         for(i = 0; i < max_constantsF; i++) {
 894             DWORD idx, mask;
 895             idx = i >> 5;
 896             mask = 1 << (i & 0x1f);
 897             if (!shader_constant_is_local(shader, i) && (reg_maps->constf[idx] & mask))
 898             {
 899                 shader_addline(buffer, "PARAM C%d = program.env[%d];\n",i, i);
 900             }
 901         }
 902     }
 903 }
 904
 905 static const char * const shift_tab[] = {
 906     "dummy",     /*  0 (none) */
 907     "coefmul.x", /*  1 (x2)   */
 908     "coefmul.y", /*  2 (x4)   */
 909     "coefmul.z", /*  3 (x8)   */
 910     "coefmul.w", /*  4 (x16)  */
 911     "dummy",     /*  5 (x32)  */
 912     "dummy",     /*  6 (x64)  */
 913     "dummy",     /*  7 (x128) */
 914     "dummy",     /*  8 (d256) */
 915     "dummy",     /*  9 (d128) */
 916     "dummy",     /* 10 (d64)  */
 917     "dummy",     /* 11 (d32)  */
 918     "coefdiv.w", /* 12 (d16)  */
 919     "coefdiv.z", /* 13 (d8)   */
 920     "coefdiv.y", /* 14 (d4)   */
 921     "coefdiv.x"  /* 15 (d2)   */
 922 };
 923
 924 static void shader_arb_get_write_mask(const struct wined3d_shader_instruction *ins,
 925         const struct wined3d_shader_dst_param *dst, char *write_mask)
 926 {
 927     char *ptr = write_mask;
 928
 929     if (dst->write_mask != WINED3DSP_WRITEMASK_ALL)
 930     {
 931         *ptr++ = '.';
 932         if (dst->write_mask & WINED3DSP_WRITEMASK_0) *ptr++ = 'x';
 933         if (dst->write_mask & WINED3DSP_WRITEMASK_1) *ptr++ = 'y';
 934         if (dst->write_mask & WINED3DSP_WRITEMASK_2) *ptr++ = 'z';
 935         if (dst->write_mask & WINED3DSP_WRITEMASK_3) *ptr++ = 'w';
 936     }
 937
 938     *ptr = '\0';
 939 }
 940
 941 static void shader_arb_get_swizzle(const struct wined3d_shader_src_param *param, BOOL fixup, char *swizzle_str)
 942 {
 943     /* For registers of type WINED3DDECLTYPE_D3DCOLOR, data is stored as "bgra",
 944      * but addressed as "rgba". To fix this we need to swap the register's x
 945      * and z components. */
 946     const char *swizzle_chars = fixup ? "zyxw" : "xyzw";
 947     char *ptr = swizzle_str;
 948
 949     /* swizzle bits fields: wwzzyyxx */
 950     DWORD swizzle = param->swizzle;
 951     DWORD swizzle_x = swizzle & 0x03;
 952     DWORD swizzle_y = (swizzle >> 2) & 0x03;
 953     DWORD swizzle_z = (swizzle >> 4) & 0x03;
 954     DWORD swizzle_w = (swizzle >> 6) & 0x03;
 955
 956     /* If the swizzle is the default swizzle (ie, "xyzw"), we don't need to
 957      * generate a swizzle string. Unless we need to our own swizzling. */
 958     if (swizzle != WINED3DSP_NOSWIZZLE || fixup)
 959     {
 960         *ptr++ = '.';
 961         if (swizzle_x == swizzle_y && swizzle_x == swizzle_z && swizzle_x == swizzle_w) {
 962             *ptr++ = swizzle_chars[swizzle_x];
 963         } else {
 964             *ptr++ = swizzle_chars[swizzle_x];
 965             *ptr++ = swizzle_chars[swizzle_y];
 966             *ptr++ = swizzle_chars[swizzle_z];
 967             *ptr++ = swizzle_chars[swizzle_w];
 968         }
 969     }
 970
 971     *ptr = '\0';
 972 }
 973
 974 static void shader_arb_request_a0(const struct wined3d_shader_instruction *ins, const char *src)
 975 {
 976     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
 977     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
 978
 979     if (!strcmp(priv->addr_reg, src)) return;
 980
 981     strcpy(priv->addr_reg, src);
 982     shader_addline(buffer, "ARL A0.x, %s;\n", src);
 983 }
 984
 985 static void shader_arb_get_src_param(const struct wined3d_shader_instruction *ins,
 986         const struct wined3d_shader_src_param *src, unsigned int tmpreg, char *outregstr);
 987
 988 static void shader_arb_get_register_name(const struct wined3d_shader_instruction *ins,
 989         const struct wined3d_shader_register *reg, char *register_name, BOOL *is_color)
 990 {
 991     /* oPos, oFog and oPts in D3D */
 992     static const char * const rastout_reg_names[] = {"TMP_OUT", "TMP_FOGCOORD", "result.pointsize"};
 993     const struct wined3d_shader *shader = ins->ctx->shader;
 994     const struct wined3d_shader_reg_maps *reg_maps = ins->ctx->reg_maps;
 995     BOOL pshader = shader_is_pshader_version(reg_maps->shader_version.type);
 996     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
 997
 998     *is_color = FALSE;
 999
1000     switch (reg->type)
1001     {
1002         case WINED3DSPR_TEMP:
1003             sprintf(register_name, "R%u", reg->idx[0].offset);
1004             break;
1005
1006         case WINED3DSPR_INPUT:
1007             if (pshader)
1008             {
1009                 if (reg_maps->shader_version.major < 3)
1010                 {
1011                     if (!reg->idx[0].offset)
1012                         strcpy(register_name, "fragment.color.primary");
1013                     else
1014                         strcpy(register_name, "fragment.color.secondary");
1015                 }
1016                 else
1017                 {
1018                     if (reg->idx[0].rel_addr)
1019                     {
1020                         char rel_reg[50];
1021                         shader_arb_get_src_param(ins, reg->idx[0].rel_addr, 0, rel_reg);
1022
1023                         if (!strcmp(rel_reg, "**aL_emul**"))
1024                         {
1025                             DWORD idx = ctx->aL + reg->idx[0].offset;
1026                             if(idx < MAX_REG_INPUT)
1027                             {
1028                                 strcpy(register_name, ctx->ps_input[idx]);
1029                             }
1030                             else
1031                             {
1032                                 ERR("Pixel shader input register out of bounds: %u\n", idx);
1033                                 sprintf(register_name, "out_of_bounds_%u", idx);
1034                             }
1035                         }
1036                         else if (reg_maps->input_registers & 0x0300)
1037                         {
1038                             /* There are two ways basically:
1039                              *
1040                              * 1) Use the unrolling code that is used for loop emulation and unroll the loop.
1041                              *    That means trouble if the loop also contains a breakc or if the control values
1042                              *    aren't local constants.
1043                              * 2) Generate an if block that checks if aL.y < 8, == 8 or == 9 and selects the
1044                              *    source dynamically. The trouble is that we cannot simply read aL.y because it
1045                              *    is an ADDRESS register. We could however push it, load .zw with a value and use
1046                              *    ADAC to load the condition code register and pop it again afterwards
1047                              */
1048                             FIXME("Relative input register addressing with more than 8 registers\n");
1049
1050                             /* This is better than nothing for now */
1051                             sprintf(register_name, "fragment.texcoord[%s + %u]", rel_reg, reg->idx[0].offset);
1052                         }
1053                         else if(ctx->cur_ps_args->super.vp_mode != vertexshader)
1054                         {
1055                             /* This is problematic because we'd have to consult the ctx->ps_input strings
1056                              * for where to find the varying. Some may be "0.0", others can be texcoords or
1057                              * colors. This needs either a pipeline replacement to make the vertex shader feed
1058                              * proper varyings, or loop unrolling
1059                              *
1060                              * For now use the texcoords and hope for the best
1061                              */
1062                             FIXME("Non-vertex shader varying input with indirect addressing\n");
1063                             sprintf(register_name, "fragment.texcoord[%s + %u]", rel_reg, reg->idx[0].offset);
1064                         }
1065                         else
1066                         {
1067                             /* D3D supports indirect addressing only with aL in loop registers. The loop instruction
1068                              * pulls GL_NV_fragment_program2 in
1069                              */
1070                             sprintf(register_name, "fragment.texcoord[%s + %u]", rel_reg, reg->idx[0].offset);
1071                         }
1072                     }
1073                     else
1074                     {
1075                         if (reg->idx[0].offset < MAX_REG_INPUT)
1076                         {
1077                             strcpy(register_name, ctx->ps_input[reg->idx[0].offset]);
1078                         }
1079                         else
1080                         {
1081                             ERR("Pixel shader input register out of bounds: %u\n", reg->idx[0].offset);
1082                             sprintf(register_name, "out_of_bounds_%u", reg->idx[0].offset);
1083                         }
1084                     }
1085                 }
1086             }
1087             else
1088             {
1089                 if (ctx->cur_vs_args->super.swizzle_map & (1 << reg->idx[0].offset))
1090                     *is_color = TRUE;
1091                 sprintf(register_name, "vertex.attrib[%u]", reg->idx[0].offset);
1092             }
1093             break;
1094
1095         case WINED3DSPR_CONST:
1096             if (!pshader && reg->idx[0].rel_addr)
1097             {
1098                 const struct arb_vshader_private *shader_data = shader->backend_data;
1099                 UINT rel_offset = shader_data->rel_offset;
1100                 BOOL aL = FALSE;
1101                 char rel_reg[50];
1102                 if (reg_maps->shader_version.major < 2)
1103                 {
1104                     sprintf(rel_reg, "A0.x");
1105                 }
1106                 else
1107                 {
1108                     shader_arb_get_src_param(ins, reg->idx[0].rel_addr, 0, rel_reg);
1109                     if (ctx->target_version == ARB)
1110                     {
1111                         if (!strcmp(rel_reg, "**aL_emul**"))
1112                         {
1113                             aL = TRUE;
1114                         } else {
1115                             shader_arb_request_a0(ins, rel_reg);
1116                             sprintf(rel_reg, "A0.x");
1117                         }
1118                     }
1119                 }
1120                 if (aL)
1121                     sprintf(register_name, "C[%u]", ctx->aL + reg->idx[0].offset);
1122                 else if (reg->idx[0].offset >= rel_offset)
1123                     sprintf(register_name, "C[%s + %u]", rel_reg, reg->idx[0].offset - rel_offset);
1124                 else
1125                     sprintf(register_name, "C[%s - %u]", rel_reg, rel_offset - reg->idx[0].offset);
1126             }
1127             else
1128             {
1129                 if (reg_maps->usesrelconstF)
1130                     sprintf(register_name, "C[%u]", reg->idx[0].offset);
1131                 else
1132                     sprintf(register_name, "C%u", reg->idx[0].offset);
1133             }
1134             break;
1135
1136         case WINED3DSPR_TEXTURE: /* case WINED3DSPR_ADDR: */
1137             if (pshader)
1138             {
1139                 if (reg_maps->shader_version.major == 1
1140                         && reg_maps->shader_version.minor <= 3)
1141                     /* In ps <= 1.3, Tx is a temporary register as destination
1142                      * to all instructions, and as source to most instructions.
1143                      * For some instructions it is the texcoord input. Those
1144                      * instructions know about the special use. */
1145                     sprintf(register_name, "T%u", reg->idx[0].offset);
1146                 else
1147                     /* In ps 1.4 and 2.x Tx is always a (read-only) varying. */
1148                     sprintf(register_name, "fragment.texcoord[%u]", reg->idx[0].offset);
1149             }
1150             else
1151             {
1152                 if (reg_maps->shader_version.major == 1 || ctx->target_version >= NV2)
1153                     sprintf(register_name, "A%u", reg->idx[0].offset);
1154                 else
1155                     sprintf(register_name, "A%u_SHADOW", reg->idx[0].offset);
1156             }
1157             break;
1158
1159         case WINED3DSPR_COLOROUT:
1160             if (ctx->cur_ps_args->super.srgb_correction && !reg->idx[0].offset)
1161             {
1162                 strcpy(register_name, "TMP_COLOR");
1163             }
1164             else
1165             {
1166                 if (ctx->cur_ps_args->super.srgb_correction)
1167                     FIXME("sRGB correction on higher render targets.\n");
1168                 if (reg_maps->rt_mask > 1)
1169                     sprintf(register_name, "result.color[%u]", reg->idx[0].offset);
1170                 else
1171                     strcpy(register_name, "result.color");
1172             }
1173             break;
1174
1175         case WINED3DSPR_RASTOUT:
1176             if (reg->idx[0].offset == 1)
1177                 sprintf(register_name, "%s", ctx->fog_output);
1178             else
1179                 sprintf(register_name, "%s", rastout_reg_names[reg->idx[0].offset]);
1180             break;
1181
1182         case WINED3DSPR_DEPTHOUT:
1183             strcpy(register_name, "result.depth");
1184             break;
1185
1186         case WINED3DSPR_ATTROUT:
1187         /* case WINED3DSPR_OUTPUT: */
1188             if (pshader)
1189                 sprintf(register_name, "oD[%u]", reg->idx[0].offset);
1190             else
1191                 strcpy(register_name, ctx->color_output[reg->idx[0].offset]);
1192             break;
1193
1194         case WINED3DSPR_TEXCRDOUT:
1195             if (pshader)
1196                 sprintf(register_name, "oT[%u]", reg->idx[0].offset);
1197             else if (reg_maps->shader_version.major < 3)
1198                 strcpy(register_name, ctx->texcrd_output[reg->idx[0].offset]);
1199             else
1200                 strcpy(register_name, ctx->vs_output[reg->idx[0].offset]);
1201             break;
1202
1203         case WINED3DSPR_LOOP:
1204             if(ctx->target_version >= NV2)
1205             {
1206                 /* Pshader has an implicitly declared loop index counter A0.x that cannot be renamed */
1207                 if(pshader) sprintf(register_name, "A0.x");
1208                 else sprintf(register_name, "aL.y");
1209             }
1210             else
1211             {
1212                 /* Unfortunately this code cannot return the value of ctx->aL here. An immediate value
1213                  * would be valid, but if aL is used for indexing(its only use), there's likely an offset,
1214                  * thus the result would be something like C[15 + 30], which is not valid in the ARB program
1215                  * grammar. So return a marker for the emulated aL and intercept it in constant and varying
1216                  * indexing
1217                  */
1218                 sprintf(register_name, "**aL_emul**");
1219             }
1220
1221             break;
1222
1223         case WINED3DSPR_CONSTINT:
1224             sprintf(register_name, "I%u", reg->idx[0].offset);
1225             break;
1226
1227         case WINED3DSPR_MISCTYPE:
1228             if (!reg->idx[0].offset)
1229                 sprintf(register_name, "vpos");
1230             else if (reg->idx[0].offset == 1)
1231                 sprintf(register_name, "fragment.facing.x");
1232             else
1233                 FIXME("Unknown MISCTYPE register index %u.\n", reg->idx[0].offset);
1234             break;
1235
1236         default:
1237             FIXME("Unhandled register type %#x[%u].\n", reg->type, reg->idx[0].offset);
1238             sprintf(register_name, "unrecognized_register[%u]", reg->idx[0].offset);
1239             break;
1240     }
1241 }
1242
1243 static void shader_arb_get_dst_param(const struct wined3d_shader_instruction *ins,
1244         const struct wined3d_shader_dst_param *wined3d_dst, char *str)
1245 {
1246     char register_name[255];
1247     char write_mask[6];
1248     BOOL is_color;
1249
1250     shader_arb_get_register_name(ins, &wined3d_dst->reg, register_name, &is_color);
1251     strcpy(str, register_name);
1252
1253     shader_arb_get_write_mask(ins, wined3d_dst, write_mask);
1254     strcat(str, write_mask);
1255 }
1256
1257 static const char *shader_arb_get_fixup_swizzle(enum fixup_channel_source channel_source)
1258 {
1259     switch(channel_source)
1260     {
1261         case CHANNEL_SOURCE_ZERO: return "0";
1262         case CHANNEL_SOURCE_ONE: return "1";
1263         case CHANNEL_SOURCE_X: return "x";
1264         case CHANNEL_SOURCE_Y: return "y";
1265         case CHANNEL_SOURCE_Z: return "z";
1266         case CHANNEL_SOURCE_W: return "w";
1267         default:
1268             FIXME("Unhandled channel source %#x\n", channel_source);
1269             return "undefined";
1270     }
1271 }
1272
1273 static void gen_color_correction(struct wined3d_shader_buffer *buffer, const char *reg,
1274         DWORD dst_mask, const char *one, const char *two, struct color_fixup_desc fixup)
1275 {
1276     DWORD mask;
1277
1278     if (is_complex_fixup(fixup))
1279     {
1280         enum complex_fixup complex_fixup = get_complex_fixup(fixup);
1281         FIXME("Complex fixup (%#x) not supported\n", complex_fixup);
1282         return;
1283     }
1284
1285     mask = 0;
1286     if (fixup.x_source != CHANNEL_SOURCE_X) mask |= WINED3DSP_WRITEMASK_0;
1287     if (fixup.y_source != CHANNEL_SOURCE_Y) mask |= WINED3DSP_WRITEMASK_1;
1288     if (fixup.z_source != CHANNEL_SOURCE_Z) mask |= WINED3DSP_WRITEMASK_2;
1289     if (fixup.w_source != CHANNEL_SOURCE_W) mask |= WINED3DSP_WRITEMASK_3;
1290     mask &= dst_mask;
1291
1292     if (mask)
1293     {
1294         shader_addline(buffer, "SWZ %s, %s, %s, %s, %s, %s;\n", reg, reg,
1295                 shader_arb_get_fixup_swizzle(fixup.x_source), shader_arb_get_fixup_swizzle(fixup.y_source),
1296                 shader_arb_get_fixup_swizzle(fixup.z_source), shader_arb_get_fixup_swizzle(fixup.w_source));
1297     }
1298
1299     mask = 0;
1300     if (fixup.x_sign_fixup) mask |= WINED3DSP_WRITEMASK_0;
1301     if (fixup.y_sign_fixup) mask |= WINED3DSP_WRITEMASK_1;
1302     if (fixup.z_sign_fixup) mask |= WINED3DSP_WRITEMASK_2;
1303     if (fixup.w_sign_fixup) mask |= WINED3DSP_WRITEMASK_3;
1304     mask &= dst_mask;
1305
1306     if (mask)
1307     {
1308         char reg_mask[6];
1309         char *ptr = reg_mask;
1310
1311         if (mask != WINED3DSP_WRITEMASK_ALL)
1312         {
1313             *ptr++ = '.';
1314             if (mask & WINED3DSP_WRITEMASK_0) *ptr++ = 'x';
1315             if (mask & WINED3DSP_WRITEMASK_1) *ptr++ = 'y';
1316             if (mask & WINED3DSP_WRITEMASK_2) *ptr++ = 'z';
1317             if (mask & WINED3DSP_WRITEMASK_3) *ptr++ = 'w';
1318         }
1319         *ptr = '\0';
1320
1321         shader_addline(buffer, "MAD %s%s, %s, %s, -%s;\n", reg, reg_mask, reg, two, one);
1322     }
1323 }
1324
1325 static const char *shader_arb_get_modifier(const struct wined3d_shader_instruction *ins)
1326 {
1327     DWORD mod;
1328     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
1329     if (!ins->dst_count) return "";
1330
1331     mod = ins->dst[0].modifiers;
1332
1333     /* Silently ignore PARTIALPRECISION if its not supported */
1334     if(priv->target_version == ARB) mod &= ~WINED3DSPDM_PARTIALPRECISION;
1335
1336     if(mod & WINED3DSPDM_MSAMPCENTROID)
1337     {
1338         FIXME("Unhandled modifier WINED3DSPDM_MSAMPCENTROID\n");
1339         mod &= ~WINED3DSPDM_MSAMPCENTROID;
1340     }
1341
1342     switch(mod)
1343     {
1344         case WINED3DSPDM_SATURATE | WINED3DSPDM_PARTIALPRECISION:
1345             return "H_SAT";
1346
1347         case WINED3DSPDM_SATURATE:
1348             return "_SAT";
1349
1350         case WINED3DSPDM_PARTIALPRECISION:
1351             return "H";
1352
1353         case 0:
1354             return "";
1355
1356         default:
1357             FIXME("Unknown modifiers 0x%08x\n", mod);
1358             return "";
1359     }
1360 }
1361
1362 #define TEX_PROJ        0x1
1363 #define TEX_BIAS        0x2
1364 #define TEX_LOD         0x4
1365 #define TEX_DERIV       0x10
1366
1367 static void shader_hw_sample(const struct wined3d_shader_instruction *ins, DWORD sampler_idx,
1368         const char *dst_str, const char *coord_reg, WORD flags, const char *dsx, const char *dsy)
1369 {
1370     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1371     DWORD sampler_type = ins->ctx->reg_maps->sampler_type[sampler_idx];
1372     const char *tex_type;
1373     BOOL np2_fixup = FALSE;
1374     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
1375     const char *mod;
1376     BOOL pshader = shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type);
1377     const struct wined3d_shader *shader;
1378     const struct wined3d_device *device;
1379     const struct wined3d_gl_info *gl_info;
1380
1381     /* D3D vertex shader sampler IDs are vertex samplers(0-3), not global d3d samplers */
1382     if(!pshader) sampler_idx += MAX_FRAGMENT_SAMPLERS;
1383
1384     switch(sampler_type) {
1385         case WINED3DSTT_1D:
1386             tex_type = "1D";
1387             break;
1388
1389         case WINED3DSTT_2D:
1390             shader = ins->ctx->shader;
1391             device = shader->device;
1392             gl_info = &device->adapter->gl_info;
1393
1394             if (pshader && priv->cur_ps_args->super.np2_fixup & (1 << sampler_idx)
1395                     && gl_info->supported[ARB_TEXTURE_RECTANGLE])
1396                 tex_type = "RECT";
1397             else
1398                 tex_type = "2D";
1399             if (shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type))
1400             {
1401                 if (priv->cur_np2fixup_info->super.active & (1 << sampler_idx))
1402                 {
1403                     if (flags) FIXME("Only ordinary sampling from NP2 textures is supported.\n");
1404                     else np2_fixup = TRUE;
1405                 }
1406             }
1407             break;
1408
1409         case WINED3DSTT_VOLUME:
1410             tex_type = "3D";
1411             break;
1412
1413         case WINED3DSTT_CUBE:
1414             tex_type = "CUBE";
1415             break;
1416
1417         default:
1418             ERR("Unexpected texture type %d\n", sampler_type);
1419             tex_type = "";
1420     }
1421
1422     /* TEX, TXL, TXD and TXP do not support the "H" modifier,
1423      * so don't use shader_arb_get_modifier
1424      */
1425     if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE) mod = "_SAT";
1426     else mod = "";
1427
1428     /* Fragment samplers always have indentity mapping */
1429     if(sampler_idx >= MAX_FRAGMENT_SAMPLERS)
1430     {
1431         sampler_idx = priv->cur_vs_args->vertex.samplers[sampler_idx - MAX_FRAGMENT_SAMPLERS];
1432     }
1433
1434     if (flags & TEX_DERIV)
1435     {
1436         if(flags & TEX_PROJ) FIXME("Projected texture sampling with custom derivatives\n");
1437         if(flags & TEX_BIAS) FIXME("Biased texture sampling with custom derivatives\n");
1438         shader_addline(buffer, "TXD%s %s, %s, %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg,
1439                        dsx, dsy,sampler_idx, tex_type);
1440     }
1441     else if(flags & TEX_LOD)
1442     {
1443         if(flags & TEX_PROJ) FIXME("Projected texture sampling with explicit lod\n");
1444         if(flags & TEX_BIAS) FIXME("Biased texture sampling with explicit lod\n");
1445         shader_addline(buffer, "TXL%s %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg,
1446                        sampler_idx, tex_type);
1447     }
1448     else if (flags & TEX_BIAS)
1449     {
1450         /* Shouldn't be possible, but let's check for it */
1451         if(flags & TEX_PROJ) FIXME("Biased and Projected texture sampling\n");
1452         /* TXB takes the 4th component of the source vector automatically, as d3d. Nothing more to do */
1453         shader_addline(buffer, "TXB%s %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg, sampler_idx, tex_type);
1454     }
1455     else if (flags & TEX_PROJ)
1456     {
1457         shader_addline(buffer, "TXP%s %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg, sampler_idx, tex_type);
1458     }
1459     else
1460     {
1461         if (np2_fixup)
1462         {
1463             const unsigned char idx = priv->cur_np2fixup_info->super.idx[sampler_idx];
1464             shader_addline(buffer, "MUL TA, np2fixup[%u].%s, %s;\n", idx >> 1,
1465                            (idx % 2) ? "zwxy" : "xyzw", coord_reg);
1466
1467             shader_addline(buffer, "TEX%s %s, TA, texture[%u], %s;\n", mod, dst_str, sampler_idx, tex_type);
1468         }
1469         else
1470             shader_addline(buffer, "TEX%s %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg, sampler_idx, tex_type);
1471     }
1472
1473     if (pshader)
1474     {
1475         gen_color_correction(buffer, dst_str, ins->dst[0].write_mask,
1476                 arb_get_helper_value(WINED3D_SHADER_TYPE_PIXEL, ARB_ONE),
1477                 arb_get_helper_value(WINED3D_SHADER_TYPE_PIXEL, ARB_TWO),
1478                 priv->cur_ps_args->super.color_fixup[sampler_idx]);
1479     }
1480 }
1481
1482 static void shader_arb_get_src_param(const struct wined3d_shader_instruction *ins,
1483         const struct wined3d_shader_src_param *src, unsigned int tmpreg, char *outregstr)
1484 {
1485     /* Generate a line that does the input modifier computation and return the input register to use */
1486     BOOL is_color = FALSE;
1487     char regstr[256];
1488     char swzstr[20];
1489     int insert_line;
1490     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1491     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
1492     const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
1493     const char *two = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_TWO);
1494
1495     /* Assume a new line will be added */
1496     insert_line = 1;
1497
1498     /* Get register name */
1499     shader_arb_get_register_name(ins, &src->reg, regstr, &is_color);
1500     shader_arb_get_swizzle(src, is_color, swzstr);
1501
1502     switch (src->modifiers)
1503     {
1504     case WINED3DSPSM_NONE:
1505         sprintf(outregstr, "%s%s", regstr, swzstr);
1506         insert_line = 0;
1507         break;
1508     case WINED3DSPSM_NEG:
1509         sprintf(outregstr, "-%s%s", regstr, swzstr);
1510         insert_line = 0;
1511         break;
1512     case WINED3DSPSM_BIAS:
1513         shader_addline(buffer, "ADD T%c, %s, -coefdiv.x;\n", 'A' + tmpreg, regstr);
1514         break;
1515     case WINED3DSPSM_BIASNEG:
1516         shader_addline(buffer, "ADD T%c, -%s, coefdiv.x;\n", 'A' + tmpreg, regstr);
1517         break;
1518     case WINED3DSPSM_SIGN:
1519         shader_addline(buffer, "MAD T%c, %s, %s, -%s;\n", 'A' + tmpreg, regstr, two, one);
1520         break;
1521     case WINED3DSPSM_SIGNNEG:
1522         shader_addline(buffer, "MAD T%c, %s, -%s, %s;\n", 'A' + tmpreg, regstr, two, one);
1523         break;
1524     case WINED3DSPSM_COMP:
1525         shader_addline(buffer, "SUB T%c, %s, %s;\n", 'A' + tmpreg, one, regstr);
1526         break;
1527     case WINED3DSPSM_X2:
1528         shader_addline(buffer, "ADD T%c, %s, %s;\n", 'A' + tmpreg, regstr, regstr);
1529         break;
1530     case WINED3DSPSM_X2NEG:
1531         shader_addline(buffer, "ADD T%c, -%s, -%s;\n", 'A' + tmpreg, regstr, regstr);
1532         break;
1533     case WINED3DSPSM_DZ:
1534         shader_addline(buffer, "RCP T%c, %s.z;\n", 'A' + tmpreg, regstr);
1535         shader_addline(buffer, "MUL T%c, %s, T%c;\n", 'A' + tmpreg, regstr, 'A' + tmpreg);
1536         break;
1537     case WINED3DSPSM_DW:
1538         shader_addline(buffer, "RCP T%c, %s.w;\n", 'A' + tmpreg, regstr);
1539         shader_addline(buffer, "MUL T%c, %s, T%c;\n", 'A' + tmpreg, regstr, 'A' + tmpreg);
1540         break;
1541     case WINED3DSPSM_ABS:
1542         if(ctx->target_version >= NV2) {
1543             sprintf(outregstr, "|%s%s|", regstr, swzstr);
1544             insert_line = 0;
1545         } else {
1546             shader_addline(buffer, "ABS T%c, %s;\n", 'A' + tmpreg, regstr);
1547         }
1548         break;
1549     case WINED3DSPSM_ABSNEG:
1550         if(ctx->target_version >= NV2) {
1551             sprintf(outregstr, "-|%s%s|", regstr, swzstr);
1552         } else {
1553             shader_addline(buffer, "ABS T%c, %s;\n", 'A' + tmpreg, regstr);
1554             sprintf(outregstr, "-T%c%s", 'A' + tmpreg, swzstr);
1555         }
1556         insert_line = 0;
1557         break;
1558     default:
1559         sprintf(outregstr, "%s%s", regstr, swzstr);
1560         insert_line = 0;
1561     }
1562
1563     /* Return modified or original register, with swizzle */
1564     if (insert_line)
1565         sprintf(outregstr, "T%c%s", 'A' + tmpreg, swzstr);
1566 }
1567
1568 static void pshader_hw_bem(const struct wined3d_shader_instruction *ins)
1569 {
1570     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1571     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1572     DWORD sampler_code = dst->reg.idx[0].offset;
1573     char dst_name[50];
1574     char src_name[2][50];
1575
1576     shader_arb_get_dst_param(ins, dst, dst_name);
1577
1578     /* Sampling the perturbation map in Tsrc was done already, including the signedness correction if needed
1579      *
1580      * Keep in mind that src_name[1] can be "TB" and src_name[0] can be "TA" because modifiers like _x2 are valid
1581      * with bem. So delay loading the first parameter until after the perturbation calculation which needs two
1582      * temps is done.
1583      */
1584     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1585     shader_addline(buffer, "SWZ TA, bumpenvmat%d, x, z, 0, 0;\n", sampler_code);
1586     shader_addline(buffer, "DP3 TC.r, TA, %s;\n", src_name[1]);
1587     shader_addline(buffer, "SWZ TA, bumpenvmat%d, y, w, 0, 0;\n", sampler_code);
1588     shader_addline(buffer, "DP3 TC.g, TA, %s;\n", src_name[1]);
1589
1590     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
1591     shader_addline(buffer, "ADD %s, %s, TC;\n", dst_name, src_name[0]);
1592 }
1593
1594 static DWORD negate_modifiers(DWORD mod, char *extra_char)
1595 {
1596     *extra_char = ' ';
1597     switch(mod)
1598     {
1599         case WINED3DSPSM_NONE:      return WINED3DSPSM_NEG;
1600         case WINED3DSPSM_NEG:       return WINED3DSPSM_NONE;
1601         case WINED3DSPSM_BIAS:      return WINED3DSPSM_BIASNEG;
1602         case WINED3DSPSM_BIASNEG:   return WINED3DSPSM_BIAS;
1603         case WINED3DSPSM_SIGN:      return WINED3DSPSM_SIGNNEG;
1604         case WINED3DSPSM_SIGNNEG:   return WINED3DSPSM_SIGN;
1605         case WINED3DSPSM_COMP:      *extra_char = '-'; return WINED3DSPSM_COMP;
1606         case WINED3DSPSM_X2:        return WINED3DSPSM_X2NEG;
1607         case WINED3DSPSM_X2NEG:     return WINED3DSPSM_X2;
1608         case WINED3DSPSM_DZ:        *extra_char = '-'; return WINED3DSPSM_DZ;
1609         case WINED3DSPSM_DW:        *extra_char = '-'; return WINED3DSPSM_DW;
1610         case WINED3DSPSM_ABS:       return WINED3DSPSM_ABSNEG;
1611         case WINED3DSPSM_ABSNEG:    return WINED3DSPSM_ABS;
1612     }
1613     FIXME("Unknown modifier %u\n", mod);
1614     return mod;
1615 }
1616
1617 static void pshader_hw_cnd(const struct wined3d_shader_instruction *ins)
1618 {
1619     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1620     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1621     char dst_name[50];
1622     char src_name[3][50];
1623     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
1624             ins->ctx->reg_maps->shader_version.minor);
1625
1626     shader_arb_get_dst_param(ins, dst, dst_name);
1627     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1628
1629     /* The coissue flag changes the semantic of the cnd instruction in <= 1.3 shaders */
1630     if (shader_version <= WINED3D_SHADER_VERSION(1, 3) && ins->coissue)
1631     {
1632         shader_addline(buffer, "MOV%s %s, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name[1]);
1633     }
1634     else
1635     {
1636         struct wined3d_shader_src_param src0_copy = ins->src[0];
1637         char extra_neg;
1638
1639         /* src0 may have a negate srcmod set, so we can't blindly add "-" to the name */
1640         src0_copy.modifiers = negate_modifiers(src0_copy.modifiers, &extra_neg);
1641
1642         shader_arb_get_src_param(ins, &src0_copy, 0, src_name[0]);
1643         shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
1644         shader_addline(buffer, "ADD TA, %c%s, coefdiv.x;\n", extra_neg, src_name[0]);
1645         shader_addline(buffer, "CMP%s %s, TA, %s, %s;\n", shader_arb_get_modifier(ins),
1646                 dst_name, src_name[1], src_name[2]);
1647     }
1648 }
1649
1650 static void pshader_hw_cmp(const struct wined3d_shader_instruction *ins)
1651 {
1652     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1653     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1654     char dst_name[50];
1655     char src_name[3][50];
1656
1657     shader_arb_get_dst_param(ins, dst, dst_name);
1658
1659     /* Generate input register names (with modifiers) */
1660     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
1661     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1662     shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
1663
1664     shader_addline(buffer, "CMP%s %s, %s, %s, %s;\n", shader_arb_get_modifier(ins),
1665             dst_name, src_name[0], src_name[2], src_name[1]);
1666 }
1667
1668 /** Process the WINED3DSIO_DP2ADD instruction in ARB.
1669  * dst = dot2(src0, src1) + src2 */
1670 static void pshader_hw_dp2add(const struct wined3d_shader_instruction *ins)
1671 {
1672     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1673     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1674     char dst_name[50];
1675     char src_name[3][50];
1676     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
1677
1678     shader_arb_get_dst_param(ins, dst, dst_name);
1679     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
1680     shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
1681
1682     if(ctx->target_version >= NV3)
1683     {
1684         /* GL_NV_fragment_program2 has a 1:1 matching instruction */
1685         shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1686         shader_addline(buffer, "DP2A%s %s, %s, %s, %s;\n", shader_arb_get_modifier(ins),
1687                        dst_name, src_name[0], src_name[1], src_name[2]);
1688     }
1689     else if(ctx->target_version >= NV2)
1690     {
1691         /* dst.x = src2.?, src0.x, src1.x + src0.y * src1.y
1692          * dst.y = src2.?, src0.x, src1.z + src0.y * src1.w
1693          * dst.z = src2.?, src0.x, src1.x + src0.y * src1.y
1694          * dst.z = src2.?, src0.x, src1.z + src0.y * src1.w
1695          *
1696          * Make sure that src1.zw = src1.xy, then we get a classic dp2add
1697          *
1698          * .xyxy and other swizzles that we could get with this are not valid in
1699          * plain ARBfp, but luckily the NV extension grammar lifts this limitation.
1700          */
1701         struct wined3d_shader_src_param tmp_param = ins->src[1];
1702         DWORD swizzle = tmp_param.swizzle & 0xf; /* Selects .xy */
1703         tmp_param.swizzle = swizzle | (swizzle << 4); /* Creates .xyxy */
1704
1705         shader_arb_get_src_param(ins, &tmp_param, 1, src_name[1]);
1706
1707         shader_addline(buffer, "X2D%s %s, %s, %s, %s;\n", shader_arb_get_modifier(ins),
1708                        dst_name, src_name[2], src_name[0], src_name[1]);
1709     }
1710     else
1711     {
1712         shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1713         /* 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]
1714         * src_name[0] can be TA, but TA is a private temp for modifiers, so it is save to overwrite
1715         */
1716         shader_addline(buffer, "MOV TA, %s;\n", src_name[0]);
1717         shader_addline(buffer, "MOV TA.z, 0.0;\n");
1718         shader_addline(buffer, "DP3 TA, TA, %s;\n", src_name[1]);
1719         shader_addline(buffer, "ADD%s %s, TA, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name[2]);
1720     }
1721 }
1722
1723 /* Map the opcode 1-to-1 to the GL code */
1724 static void shader_hw_map2gl(const struct wined3d_shader_instruction *ins)
1725 {
1726     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1727     const char *instruction;
1728     char arguments[256], dst_str[50];
1729     unsigned int i;
1730     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1731
1732     switch (ins->handler_idx)
1733     {
1734         case WINED3DSIH_ABS: instruction = "ABS"; break;
1735         case WINED3DSIH_ADD: instruction = "ADD"; break;
1736         case WINED3DSIH_CRS: instruction = "XPD"; break;
1737         case WINED3DSIH_DP3: instruction = "DP3"; break;
1738         case WINED3DSIH_DP4: instruction = "DP4"; break;
1739         case WINED3DSIH_DST: instruction = "DST"; break;
1740         case WINED3DSIH_FRC: instruction = "FRC"; break;
1741         case WINED3DSIH_LIT: instruction = "LIT"; break;
1742         case WINED3DSIH_LRP: instruction = "LRP"; break;
1743         case WINED3DSIH_MAD: instruction = "MAD"; break;
1744         case WINED3DSIH_MAX: instruction = "MAX"; break;
1745         case WINED3DSIH_MIN: instruction = "MIN"; break;
1746         case WINED3DSIH_MOV: instruction = "MOV"; break;
1747         case WINED3DSIH_MUL: instruction = "MUL"; break;
1748         case WINED3DSIH_SGE: instruction = "SGE"; break;
1749         case WINED3DSIH_SLT: instruction = "SLT"; break;
1750         case WINED3DSIH_SUB: instruction = "SUB"; break;
1751         case WINED3DSIH_MOVA:instruction = "ARR"; break;
1752         case WINED3DSIH_DSX: instruction = "DDX"; break;
1753         default: instruction = "";
1754             FIXME("Unhandled opcode %#x\n", ins->handler_idx);
1755             break;
1756     }
1757
1758     /* Note that shader_arb_add_dst_param() adds spaces. */
1759     arguments[0] = '\0';
1760     shader_arb_get_dst_param(ins, dst, dst_str);
1761     for (i = 0; i < ins->src_count; ++i)
1762     {
1763         char operand[100];
1764         strcat(arguments, ", ");
1765         shader_arb_get_src_param(ins, &ins->src[i], i, operand);
1766         strcat(arguments, operand);
1767     }
1768     shader_addline(buffer, "%s%s %s%s;\n", instruction, shader_arb_get_modifier(ins), dst_str, arguments);
1769 }
1770
1771 static void shader_hw_nop(const struct wined3d_shader_instruction *ins) {}
1772
1773 static void shader_hw_mov(const struct wined3d_shader_instruction *ins)
1774 {
1775     const struct wined3d_shader *shader = ins->ctx->shader;
1776     const struct wined3d_shader_reg_maps *reg_maps = ins->ctx->reg_maps;
1777     BOOL pshader = shader_is_pshader_version(reg_maps->shader_version.type);
1778     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
1779     const char *zero = arb_get_helper_value(reg_maps->shader_version.type, ARB_ZERO);
1780     const char *one = arb_get_helper_value(reg_maps->shader_version.type, ARB_ONE);
1781     const char *two = arb_get_helper_value(reg_maps->shader_version.type, ARB_TWO);
1782
1783     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1784     char src0_param[256];
1785
1786     if (ins->handler_idx == WINED3DSIH_MOVA)
1787     {
1788         const struct arb_vshader_private *shader_data = shader->backend_data;
1789         char write_mask[6];
1790         const char *offset = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_VS_REL_OFFSET);
1791
1792         if(ctx->target_version >= NV2) {
1793             shader_hw_map2gl(ins);
1794             return;
1795         }
1796         shader_arb_get_src_param(ins, &ins->src[0], 0, src0_param);
1797         shader_arb_get_write_mask(ins, &ins->dst[0], write_mask);
1798
1799         /* This implements the mova formula used in GLSL. The first two instructions
1800          * prepare the sign() part. Note that it is fine to have my_sign(0.0) = 1.0
1801          * in this case:
1802          * mova A0.x, 0.0
1803          *
1804          * A0.x = arl(floor(abs(0.0) + 0.5) * 1.0) = floor(0.5) = 0.0 since arl does a floor
1805          *
1806          * The ARL is performed when A0 is used - the requested component is read from A0_SHADOW into
1807          * A0.x. We can use the overwritten component of A0_shadow as temporary storage for the sign.
1808          */
1809         shader_addline(buffer, "SGE A0_SHADOW%s, %s, %s;\n", write_mask, src0_param, zero);
1810         shader_addline(buffer, "MAD A0_SHADOW%s, A0_SHADOW, %s, -%s;\n", write_mask, two, one);
1811
1812         shader_addline(buffer, "ABS TA%s, %s;\n", write_mask, src0_param);
1813         shader_addline(buffer, "ADD TA%s, TA, rel_addr_const.x;\n", write_mask);
1814         shader_addline(buffer, "FLR TA%s, TA;\n", write_mask);
1815         if (shader_data->rel_offset)
1816         {
1817             shader_addline(buffer, "ADD TA%s, TA, %s;\n", write_mask, offset);
1818         }
1819         shader_addline(buffer, "MUL A0_SHADOW%s, TA, A0_SHADOW;\n", write_mask);
1820
1821         ((struct shader_arb_ctx_priv *)ins->ctx->backend_data)->addr_reg[0] = '\0';
1822     }
1823     else if (reg_maps->shader_version.major == 1
1824           && !shader_is_pshader_version(reg_maps->shader_version.type)
1825           && ins->dst[0].reg.type == WINED3DSPR_ADDR)
1826     {
1827         const struct arb_vshader_private *shader_data = shader->backend_data;
1828         src0_param[0] = '\0';
1829
1830         if (shader_data->rel_offset)
1831         {
1832             const char *offset = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_VS_REL_OFFSET);
1833             shader_arb_get_src_param(ins, &ins->src[0], 0, src0_param);
1834             shader_addline(buffer, "ADD TA.x, %s, %s;\n", src0_param, offset);
1835             shader_addline(buffer, "ARL A0.x, TA.x;\n");
1836         }
1837         else
1838         {
1839             /* Apple's ARB_vertex_program implementation does not accept an ARL source argument
1840              * with more than one component. Thus replicate the first source argument over all
1841              * 4 components. For example, .xyzw -> .x (or better: .xxxx), .zwxy -> .z, etc) */
1842             struct wined3d_shader_src_param tmp_src = ins->src[0];
1843             tmp_src.swizzle = (tmp_src.swizzle & 0x3) * 0x55;
1844             shader_arb_get_src_param(ins, &tmp_src, 0, src0_param);
1845             shader_addline(buffer, "ARL A0.x, %s;\n", src0_param);
1846         }
1847     }
1848     else if (ins->dst[0].reg.type == WINED3DSPR_COLOROUT && !ins->dst[0].reg.idx[0].offset && pshader)
1849     {
1850         if (ctx->cur_ps_args->super.srgb_correction && shader->u.ps.color0_mov)
1851         {
1852             shader_addline(buffer, "#mov handled in srgb write code\n");
1853             return;
1854         }
1855         shader_hw_map2gl(ins);
1856     }
1857     else
1858     {
1859         shader_hw_map2gl(ins);
1860     }
1861 }
1862
1863 static void pshader_hw_texkill(const struct wined3d_shader_instruction *ins)
1864 {
1865     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1866     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1867     char reg_dest[40];
1868
1869     /* No swizzles are allowed in d3d's texkill. PS 1.x ignores the 4th component as documented,
1870      * but >= 2.0 honors it (undocumented, but tested by the d3d9 testsuite)
1871      */
1872     shader_arb_get_dst_param(ins, dst, reg_dest);
1873
1874     if (ins->ctx->reg_maps->shader_version.major >= 2)
1875     {
1876         const char *kilsrc = "TA";
1877         BOOL is_color;
1878
1879         shader_arb_get_register_name(ins, &dst->reg, reg_dest, &is_color);
1880         if(dst->write_mask == WINED3DSP_WRITEMASK_ALL)
1881         {
1882             kilsrc = reg_dest;
1883         }
1884         else
1885         {
1886             /* Sigh. KIL doesn't support swizzles/writemasks. KIL passes a writemask, but ".xy" for example
1887              * is not valid as a swizzle in ARB (needs ".xyyy"). Use SWZ to load the register properly, and set
1888              * masked out components to 0(won't kill)
1889              */
1890             char x = '0', y = '0', z = '0', w = '0';
1891             if(dst->write_mask & WINED3DSP_WRITEMASK_0) x = 'x';
1892             if(dst->write_mask & WINED3DSP_WRITEMASK_1) y = 'y';
1893             if(dst->write_mask & WINED3DSP_WRITEMASK_2) z = 'z';
1894             if(dst->write_mask & WINED3DSP_WRITEMASK_3) w = 'w';
1895             shader_addline(buffer, "SWZ TA, %s, %c, %c, %c, %c;\n", reg_dest, x, y, z, w);
1896         }
1897         shader_addline(buffer, "KIL %s;\n", kilsrc);
1898     }
1899     else
1900     {
1901         /* ARB fp doesn't like swizzles on the parameter of the KIL instruction. To mask the 4th component,
1902          * copy the register into our general purpose TMP variable, overwrite .w and pass TMP to KIL
1903          *
1904          * ps_1_3 shaders use the texcoord incarnation of the Tx register. ps_1_4 shaders can use the same,
1905          * or pass in any temporary register(in shader phase 2)
1906          */
1907         if (ins->ctx->reg_maps->shader_version.minor <= 3)
1908             sprintf(reg_dest, "fragment.texcoord[%u]", dst->reg.idx[0].offset);
1909         else
1910             shader_arb_get_dst_param(ins, dst, reg_dest);
1911         shader_addline(buffer, "SWZ TA, %s, x, y, z, 1;\n", reg_dest);
1912         shader_addline(buffer, "KIL TA;\n");
1913     }
1914 }
1915
1916 static void pshader_hw_tex(const struct wined3d_shader_instruction *ins)
1917 {
1918     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
1919     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1920     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
1921             ins->ctx->reg_maps->shader_version.minor);
1922     struct wined3d_shader_src_param src;
1923
1924     char reg_dest[40];
1925     char reg_coord[40];
1926     DWORD reg_sampler_code;
1927     WORD myflags = 0;
1928     BOOL swizzle_coord = FALSE;
1929
1930     /* All versions have a destination register */
1931     shader_arb_get_dst_param(ins, dst, reg_dest);
1932
1933     /* 1.0-1.4: Use destination register number as texture code.
1934        2.0+: Use provided sampler number as texure code. */
1935     if (shader_version < WINED3D_SHADER_VERSION(2,0))
1936         reg_sampler_code = dst->reg.idx[0].offset;
1937     else
1938         reg_sampler_code = ins->src[1].reg.idx[0].offset;
1939
1940     /* 1.0-1.3: Use the texcoord varying.
1941        1.4+: Use provided coordinate source register. */
1942     if (shader_version < WINED3D_SHADER_VERSION(1,4))
1943         sprintf(reg_coord, "fragment.texcoord[%u]", reg_sampler_code);
1944     else {
1945         /* TEX is the only instruction that can handle DW and DZ natively */
1946         src = ins->src[0];
1947         if(src.modifiers == WINED3DSPSM_DW) src.modifiers = WINED3DSPSM_NONE;
1948         if(src.modifiers == WINED3DSPSM_DZ) src.modifiers = WINED3DSPSM_NONE;
1949         shader_arb_get_src_param(ins, &src, 0, reg_coord);
1950     }
1951
1952     /* projection flag:
1953      * 1.1, 1.2, 1.3: Use WINED3D_TSS_TEXTURETRANSFORMFLAGS
1954      * 1.4: Use WINED3DSPSM_DZ or WINED3DSPSM_DW on src[0]
1955      * 2.0+: Use WINED3DSI_TEXLD_PROJECT on the opcode
1956      */
1957     if (shader_version < WINED3D_SHADER_VERSION(1,4))
1958     {
1959         DWORD flags = 0;
1960         if (reg_sampler_code < MAX_TEXTURES)
1961             flags = priv->cur_ps_args->super.tex_transform >> reg_sampler_code * WINED3D_PSARGS_TEXTRANSFORM_SHIFT;
1962         if (flags & WINED3D_PSARGS_PROJECTED)
1963         {
1964             myflags |= TEX_PROJ;
1965             if ((flags & ~WINED3D_PSARGS_PROJECTED) == WINED3D_TTFF_COUNT3)
1966                 swizzle_coord = TRUE;
1967         }
1968     }
1969     else if (shader_version < WINED3D_SHADER_VERSION(2,0))
1970     {
1971         enum wined3d_shader_src_modifier src_mod = ins->src[0].modifiers;
1972         if (src_mod == WINED3DSPSM_DZ)
1973         {
1974             swizzle_coord = TRUE;
1975             myflags |= TEX_PROJ;
1976         } else if(src_mod == WINED3DSPSM_DW) {
1977             myflags |= TEX_PROJ;
1978         }
1979     } else {
1980         if (ins->flags & WINED3DSI_TEXLD_PROJECT) myflags |= TEX_PROJ;
1981         if (ins->flags & WINED3DSI_TEXLD_BIAS) myflags |= TEX_BIAS;
1982     }
1983
1984     if (swizzle_coord)
1985     {
1986         /* TXP cannot handle DZ natively, so move the z coordinate to .w.
1987          * reg_coord is a read-only varying register, so we need a temp reg */
1988         shader_addline(ins->ctx->buffer, "SWZ TA, %s, x, y, z, z;\n", reg_coord);
1989         strcpy(reg_coord, "TA");
1990     }
1991
1992     shader_hw_sample(ins, reg_sampler_code, reg_dest, reg_coord, myflags, NULL, NULL);
1993 }
1994
1995 static void pshader_hw_texcoord(const struct wined3d_shader_instruction *ins)
1996 {
1997     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1998     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1999     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
2000             ins->ctx->reg_maps->shader_version.minor);
2001     char dst_str[50];
2002
2003     if (shader_version < WINED3D_SHADER_VERSION(1,4))
2004     {
2005         DWORD reg = dst->reg.idx[0].offset;
2006
2007         shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2008         shader_addline(buffer, "MOV_SAT %s, fragment.texcoord[%u];\n", dst_str, reg);
2009     } else {
2010         char reg_src[40];
2011
2012         shader_arb_get_src_param(ins, &ins->src[0], 0, reg_src);
2013         shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2014         shader_addline(buffer, "MOV %s, %s;\n", dst_str, reg_src);
2015     }
2016 }
2017
2018 static void pshader_hw_texreg2ar(const struct wined3d_shader_instruction *ins)
2019 {
2020      struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2021      DWORD flags = 0;
2022
2023      DWORD reg1 = ins->dst[0].reg.idx[0].offset;
2024      char dst_str[50];
2025      char src_str[50];
2026
2027      /* Note that texreg2ar treats Tx as a temporary register, not as a varying */
2028      shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2029      shader_arb_get_src_param(ins, &ins->src[0], 0, src_str);
2030      /* Move .x first in case src_str is "TA" */
2031      shader_addline(buffer, "MOV TA.y, %s.x;\n", src_str);
2032      shader_addline(buffer, "MOV TA.x, %s.w;\n", src_str);
2033      if (reg1 < MAX_TEXTURES)
2034      {
2035          struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2036          flags = priv->cur_ps_args->super.tex_transform >> reg1 * WINED3D_PSARGS_TEXTRANSFORM_SHIFT;
2037      }
2038      shader_hw_sample(ins, reg1, dst_str, "TA", flags & WINED3D_PSARGS_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2039 }
2040
2041 static void pshader_hw_texreg2gb(const struct wined3d_shader_instruction *ins)
2042 {
2043      struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2044
2045      DWORD reg1 = ins->dst[0].reg.idx[0].offset;
2046      char dst_str[50];
2047      char src_str[50];
2048
2049      /* Note that texreg2gb treats Tx as a temporary register, not as a varying */
2050      shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2051      shader_arb_get_src_param(ins, &ins->src[0], 0, src_str);
2052      shader_addline(buffer, "MOV TA.x, %s.y;\n", src_str);
2053      shader_addline(buffer, "MOV TA.y, %s.z;\n", src_str);
2054      shader_hw_sample(ins, reg1, dst_str, "TA", 0, NULL, NULL);
2055 }
2056
2057 static void pshader_hw_texreg2rgb(const struct wined3d_shader_instruction *ins)
2058 {
2059     DWORD reg1 = ins->dst[0].reg.idx[0].offset;
2060     char dst_str[50];
2061     char src_str[50];
2062
2063     /* Note that texreg2rg treats Tx as a temporary register, not as a varying */
2064     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2065     shader_arb_get_src_param(ins, &ins->src[0], 0, src_str);
2066     shader_hw_sample(ins, reg1, dst_str, src_str, 0, NULL, NULL);
2067 }
2068
2069 static void pshader_hw_texbem(const struct wined3d_shader_instruction *ins)
2070 {
2071     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2072     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2073     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2074     char reg_coord[40], dst_reg[50], src_reg[50];
2075     DWORD reg_dest_code;
2076
2077     /* All versions have a destination register. The Tx where the texture coordinates come
2078      * from is the varying incarnation of the texture register
2079      */
2080     reg_dest_code = dst->reg.idx[0].offset;
2081     shader_arb_get_dst_param(ins, &ins->dst[0], dst_reg);
2082     shader_arb_get_src_param(ins, &ins->src[0], 0, src_reg);
2083     sprintf(reg_coord, "fragment.texcoord[%u]", reg_dest_code);
2084
2085     /* Sampling the perturbation map in Tsrc was done already, including the signedness correction if needed
2086      * The Tx in which the perturbation map is stored is the tempreg incarnation of the texture register
2087      *
2088      * GL_NV_fragment_program_option could handle this in one instruction via X2D:
2089      * X2D TA.xy, fragment.texcoord, T%u, bumpenvmat%u.xzyw
2090      *
2091      * However, the NV extensions are never enabled for <= 2.0 shaders because of the performance penalty that
2092      * comes with it, and texbem is an 1.x only instruction. No 1.x instruction forces us to enable the NV
2093      * extension.
2094      */
2095     shader_addline(buffer, "SWZ TB, bumpenvmat%d, x, z, 0, 0;\n", reg_dest_code);
2096     shader_addline(buffer, "DP3 TA.x, TB, %s;\n", src_reg);
2097     shader_addline(buffer, "SWZ TB, bumpenvmat%d, y, w, 0, 0;\n", reg_dest_code);
2098     shader_addline(buffer, "DP3 TA.y, TB, %s;\n", src_reg);
2099
2100     /* with projective textures, texbem only divides the static texture coord, not the displacement,
2101      * so we can't let the GL handle this.
2102      */
2103     if ((priv->cur_ps_args->super.tex_transform >> reg_dest_code * WINED3D_PSARGS_TEXTRANSFORM_SHIFT)
2104             & WINED3D_PSARGS_PROJECTED)
2105     {
2106         shader_addline(buffer, "RCP TB.w, %s.w;\n", reg_coord);
2107         shader_addline(buffer, "MUL TB.xy, %s, TB.w;\n", reg_coord);
2108         shader_addline(buffer, "ADD TA.xy, TA, TB;\n");
2109     } else {
2110         shader_addline(buffer, "ADD TA.xy, TA, %s;\n", reg_coord);
2111     }
2112
2113     shader_hw_sample(ins, reg_dest_code, dst_reg, "TA", 0, NULL, NULL);
2114
2115     if (ins->handler_idx == WINED3DSIH_TEXBEML)
2116     {
2117         /* No src swizzles are allowed, so this is ok */
2118         shader_addline(buffer, "MAD TA, %s.z, luminance%d.x, luminance%d.y;\n",
2119                        src_reg, reg_dest_code, reg_dest_code);
2120         shader_addline(buffer, "MUL %s, %s, TA;\n", dst_reg, dst_reg);
2121     }
2122 }
2123
2124 static void pshader_hw_texm3x2pad(const struct wined3d_shader_instruction *ins)
2125 {
2126     DWORD reg = ins->dst[0].reg.idx[0].offset;
2127     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2128     char src0_name[50], dst_name[50];
2129     BOOL is_color;
2130     struct wined3d_shader_register tmp_reg = ins->dst[0].reg;
2131
2132     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2133     /* The next instruction will be a texm3x2tex or texm3x2depth that writes to the uninitialized
2134      * T<reg+1> register. Use this register to store the calculated vector
2135      */
2136     tmp_reg.idx[0].offset = reg + 1;
2137     shader_arb_get_register_name(ins, &tmp_reg, dst_name, &is_color);
2138     shader_addline(buffer, "DP3 %s.x, fragment.texcoord[%u], %s;\n", dst_name, reg, src0_name);
2139 }
2140
2141 static void pshader_hw_texm3x2tex(const struct wined3d_shader_instruction *ins)
2142 {
2143     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2144     DWORD flags;
2145     DWORD reg = ins->dst[0].reg.idx[0].offset;
2146     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2147     char dst_str[50];
2148     char src0_name[50];
2149     char dst_reg[50];
2150     BOOL is_color;
2151
2152     /* We know that we're writing to the uninitialized T<reg> register, so use it for temporary storage */
2153     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_reg, &is_color);
2154
2155     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2156     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2157     shader_addline(buffer, "DP3 %s.y, fragment.texcoord[%u], %s;\n", dst_reg, reg, src0_name);
2158     flags = reg < MAX_TEXTURES ? priv->cur_ps_args->super.tex_transform >> reg * WINED3D_PSARGS_TEXTRANSFORM_SHIFT : 0;
2159     shader_hw_sample(ins, reg, dst_str, dst_reg, flags & WINED3D_PSARGS_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2160 }
2161
2162 static void pshader_hw_texm3x3pad(const struct wined3d_shader_instruction *ins)
2163 {
2164     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
2165     DWORD reg = ins->dst[0].reg.idx[0].offset;
2166     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2167     char src0_name[50], dst_name[50];
2168     struct wined3d_shader_register tmp_reg = ins->dst[0].reg;
2169     BOOL is_color;
2170
2171     /* There are always 2 texm3x3pad instructions followed by one texm3x3[tex,vspec, ...] instruction, with
2172      * incrementing ins->dst[0].register_idx numbers. So the pad instruction already knows the final destination
2173      * register, and this register is uninitialized(otherwise the assembler complains that it is 'redeclared')
2174      */
2175     tmp_reg.idx[0].offset = reg + 2 - tex_mx->current_row;
2176     shader_arb_get_register_name(ins, &tmp_reg, dst_name, &is_color);
2177
2178     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2179     shader_addline(buffer, "DP3 %s.%c, fragment.texcoord[%u], %s;\n",
2180                    dst_name, 'x' + tex_mx->current_row, reg, src0_name);
2181     tex_mx->texcoord_w[tex_mx->current_row++] = reg;
2182 }
2183
2184 static void pshader_hw_texm3x3tex(const struct wined3d_shader_instruction *ins)
2185 {
2186     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2187     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
2188     DWORD flags;
2189     DWORD reg = ins->dst[0].reg.idx[0].offset;
2190     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2191     char dst_str[50];
2192     char src0_name[50], dst_name[50];
2193     BOOL is_color;
2194
2195     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2196     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2197     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_name, reg, src0_name);
2198
2199     /* Sample the texture using the calculated coordinates */
2200     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2201     flags = reg < MAX_TEXTURES ? priv->cur_ps_args->super.tex_transform >> reg * WINED3D_PSARGS_TEXTRANSFORM_SHIFT : 0;
2202     shader_hw_sample(ins, reg, dst_str, dst_name, flags & WINED3D_PSARGS_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2203     tex_mx->current_row = 0;
2204 }
2205
2206 static void pshader_hw_texm3x3vspec(const struct wined3d_shader_instruction *ins)
2207 {
2208     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2209     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
2210     DWORD flags;
2211     DWORD reg = ins->dst[0].reg.idx[0].offset;
2212     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2213     char dst_str[50];
2214     char src0_name[50];
2215     char dst_reg[50];
2216     BOOL is_color;
2217
2218     /* Get the dst reg without writemask strings. We know this register is uninitialized, so we can use all
2219      * components for temporary data storage
2220      */
2221     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_reg, &is_color);
2222     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2223     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_reg, reg, src0_name);
2224
2225     /* Construct the eye-ray vector from w coordinates */
2226     shader_addline(buffer, "MOV TB.x, fragment.texcoord[%u].w;\n", tex_mx->texcoord_w[0]);
2227     shader_addline(buffer, "MOV TB.y, fragment.texcoord[%u].w;\n", tex_mx->texcoord_w[1]);
2228     shader_addline(buffer, "MOV TB.z, fragment.texcoord[%u].w;\n", reg);
2229
2230     /* Calculate reflection vector
2231      */
2232     shader_addline(buffer, "DP3 %s.w, %s, TB;\n", dst_reg, dst_reg);
2233     /* The .w is ignored when sampling, so I can use TB.w to calculate dot(N, N) */
2234     shader_addline(buffer, "DP3 TB.w, %s, %s;\n", dst_reg, dst_reg);
2235     shader_addline(buffer, "RCP TB.w, TB.w;\n");
2236     shader_addline(buffer, "MUL %s.w, %s.w, TB.w;\n", dst_reg, dst_reg);
2237     shader_addline(buffer, "MUL %s, %s.w, %s;\n", dst_reg, dst_reg, dst_reg);
2238     shader_addline(buffer, "MAD %s, coefmul.x, %s, -TB;\n", dst_reg, dst_reg);
2239
2240     /* Sample the texture using the calculated coordinates */
2241     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2242     flags = reg < MAX_TEXTURES ? priv->cur_ps_args->super.tex_transform >> reg * WINED3D_PSARGS_TEXTRANSFORM_SHIFT : 0;
2243     shader_hw_sample(ins, reg, dst_str, dst_reg, flags & WINED3D_PSARGS_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2244     tex_mx->current_row = 0;
2245 }
2246
2247 static void pshader_hw_texm3x3spec(const struct wined3d_shader_instruction *ins)
2248 {
2249     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2250     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
2251     DWORD flags;
2252     DWORD reg = ins->dst[0].reg.idx[0].offset;
2253     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2254     char dst_str[50];
2255     char src0_name[50];
2256     char src1_name[50];
2257     char dst_reg[50];
2258     BOOL is_color;
2259
2260     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2261     shader_arb_get_src_param(ins, &ins->src[0], 1, src1_name);
2262     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_reg, &is_color);
2263     /* Note: dst_reg.xy is input here, generated by two texm3x3pad instructions */
2264     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_reg, reg, src0_name);
2265
2266     /* Calculate reflection vector.
2267      *
2268      *                   dot(N, E)
2269      * dst_reg.xyz = 2 * --------- * N - E
2270      *                   dot(N, N)
2271      *
2272      * Which normalizes the normal vector
2273      */
2274     shader_addline(buffer, "DP3 %s.w, %s, %s;\n", dst_reg, dst_reg, src1_name);
2275     shader_addline(buffer, "DP3 TC.w, %s, %s;\n", dst_reg, dst_reg);
2276     shader_addline(buffer, "RCP TC.w, TC.w;\n");
2277     shader_addline(buffer, "MUL %s.w, %s.w, TC.w;\n", dst_reg, dst_reg);
2278     shader_addline(buffer, "MUL %s, %s.w, %s;\n", dst_reg, dst_reg, dst_reg);
2279     shader_addline(buffer, "MAD %s, coefmul.x, %s, -%s;\n", dst_reg, dst_reg, src1_name);
2280
2281     /* Sample the texture using the calculated coordinates */
2282     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2283     flags = reg < MAX_TEXTURES ? priv->cur_ps_args->super.tex_transform >> reg * WINED3D_PSARGS_TEXTRANSFORM_SHIFT : 0;
2284     shader_hw_sample(ins, reg, dst_str, dst_reg, flags & WINED3D_PSARGS_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2285     tex_mx->current_row = 0;
2286 }
2287
2288 static void pshader_hw_texdepth(const struct wined3d_shader_instruction *ins)
2289 {
2290     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2291     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2292     char dst_name[50];
2293     const char *zero = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ZERO);
2294     const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
2295
2296     /* texdepth has an implicit destination, the fragment depth value. It's only parameter,
2297      * which is essentially an input, is the destination register because it is the first
2298      * parameter. According to the msdn, this must be register r5, but let's keep it more flexible
2299      * here(writemasks/swizzles are not valid on texdepth)
2300      */
2301     shader_arb_get_dst_param(ins, dst, dst_name);
2302
2303     /* According to the msdn, the source register(must be r5) is unusable after
2304      * the texdepth instruction, so we're free to modify it
2305      */
2306     shader_addline(buffer, "MIN %s.y, %s.y, %s;\n", dst_name, dst_name, one);
2307
2308     /* How to deal with the special case dst_name.g == 0? if r != 0, then
2309      * the r * (1 / 0) will give infinity, which is clamped to 1.0, the correct
2310      * result. But if r = 0.0, then 0 * inf = 0, which is incorrect.
2311      */
2312     shader_addline(buffer, "RCP %s.y, %s.y;\n", dst_name, dst_name);
2313     shader_addline(buffer, "MUL TA.x, %s.x, %s.y;\n", dst_name, dst_name);
2314     shader_addline(buffer, "MIN TA.x, TA.x, %s;\n", one);
2315     shader_addline(buffer, "MAX result.depth, TA.x, %s;\n", zero);
2316 }
2317
2318 /** Process the WINED3DSIO_TEXDP3TEX instruction in ARB:
2319  * Take a 3-component dot product of the TexCoord[dstreg] and src,
2320  * then perform a 1D texture lookup from stage dstregnum, place into dst. */
2321 static void pshader_hw_texdp3tex(const struct wined3d_shader_instruction *ins)
2322 {
2323     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2324     DWORD sampler_idx = ins->dst[0].reg.idx[0].offset;
2325     char src0[50];
2326     char dst_str[50];
2327
2328     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2329     shader_addline(buffer, "MOV TB, 0.0;\n");
2330     shader_addline(buffer, "DP3 TB.x, fragment.texcoord[%u], %s;\n", sampler_idx, src0);
2331
2332     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2333     shader_hw_sample(ins, sampler_idx, dst_str, "TB", 0 /* Only one coord, can't be projected */, NULL, NULL);
2334 }
2335
2336 /** Process the WINED3DSIO_TEXDP3 instruction in ARB:
2337  * Take a 3-component dot product of the TexCoord[dstreg] and src. */
2338 static void pshader_hw_texdp3(const struct wined3d_shader_instruction *ins)
2339 {
2340     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2341     char src0[50];
2342     char dst_str[50];
2343     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2344
2345     /* Handle output register */
2346     shader_arb_get_dst_param(ins, dst, dst_str);
2347     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2348     shader_addline(buffer, "DP3 %s, fragment.texcoord[%u], %s;\n", dst_str, dst->reg.idx[0].offset, src0);
2349 }
2350
2351 /** Process the WINED3DSIO_TEXM3X3 instruction in ARB
2352  * Perform the 3rd row of a 3x3 matrix multiply */
2353 static void pshader_hw_texm3x3(const struct wined3d_shader_instruction *ins)
2354 {
2355     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2356     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2357     char dst_str[50], dst_name[50];
2358     char src0[50];
2359     BOOL is_color;
2360
2361     shader_arb_get_dst_param(ins, dst, dst_str);
2362     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2363     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2364     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_name, dst->reg.idx[0].offset, src0);
2365     shader_addline(buffer, "MOV %s, %s;\n", dst_str, dst_name);
2366 }
2367
2368 /** Process the WINED3DSIO_TEXM3X2DEPTH instruction in ARB:
2369  * Last row of a 3x2 matrix multiply, use the result to calculate the depth:
2370  * Calculate tmp0.y = TexCoord[dstreg] . src.xyz;  (tmp0.x has already been calculated)
2371  * depth = (tmp0.y == 0.0) ? 1.0 : tmp0.x / tmp0.y
2372  */
2373 static void pshader_hw_texm3x2depth(const struct wined3d_shader_instruction *ins)
2374 {
2375     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2376     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2377     char src0[50], dst_name[50];
2378     BOOL is_color;
2379     const char *zero = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ZERO);
2380     const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
2381
2382     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2383     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2384     shader_addline(buffer, "DP3 %s.y, fragment.texcoord[%u], %s;\n", dst_name, dst->reg.idx[0].offset, src0);
2385
2386     /* How to deal with the special case dst_name.g == 0? if r != 0, then
2387      * the r * (1 / 0) will give infinity, which is clamped to 1.0, the correct
2388      * result. But if r = 0.0, then 0 * inf = 0, which is incorrect.
2389      */
2390     shader_addline(buffer, "RCP %s.y, %s.y;\n", dst_name, dst_name);
2391     shader_addline(buffer, "MUL %s.x, %s.x, %s.y;\n", dst_name, dst_name, dst_name);
2392     shader_addline(buffer, "MIN %s.x, %s.x, %s;\n", dst_name, dst_name, one);
2393     shader_addline(buffer, "MAX result.depth, %s.x, %s;\n", dst_name, zero);
2394 }
2395
2396 /** Handles transforming all WINED3DSIO_M?x? opcodes for
2397     Vertex/Pixel shaders to ARB_vertex_program codes */
2398 static void shader_hw_mnxn(const struct wined3d_shader_instruction *ins)
2399 {
2400     int i;
2401     int nComponents = 0;
2402     struct wined3d_shader_dst_param tmp_dst = {{0}};
2403     struct wined3d_shader_src_param tmp_src[2] = {{{0}}};
2404     struct wined3d_shader_instruction tmp_ins;
2405
2406     memset(&tmp_ins, 0, sizeof(tmp_ins));
2407
2408     /* Set constants for the temporary argument */
2409     tmp_ins.ctx = ins->ctx;
2410     tmp_ins.dst_count = 1;
2411     tmp_ins.dst = &tmp_dst;
2412     tmp_ins.src_count = 2;
2413     tmp_ins.src = tmp_src;
2414
2415     switch(ins->handler_idx)
2416     {
2417         case WINED3DSIH_M4x4:
2418             nComponents = 4;
2419             tmp_ins.handler_idx = WINED3DSIH_DP4;
2420             break;
2421         case WINED3DSIH_M4x3:
2422             nComponents = 3;
2423             tmp_ins.handler_idx = WINED3DSIH_DP4;
2424             break;
2425         case WINED3DSIH_M3x4:
2426             nComponents = 4;
2427             tmp_ins.handler_idx = WINED3DSIH_DP3;
2428             break;
2429         case WINED3DSIH_M3x3:
2430             nComponents = 3;
2431             tmp_ins.handler_idx = WINED3DSIH_DP3;
2432             break;
2433         case WINED3DSIH_M3x2:
2434             nComponents = 2;
2435             tmp_ins.handler_idx = WINED3DSIH_DP3;
2436             break;
2437         default:
2438             FIXME("Unhandled opcode %#x\n", ins->handler_idx);
2439             break;
2440     }
2441
2442     tmp_dst = ins->dst[0];
2443     tmp_src[0] = ins->src[0];
2444     tmp_src[1] = ins->src[1];
2445     for (i = 0; i < nComponents; ++i)
2446     {
2447         tmp_dst.write_mask = WINED3DSP_WRITEMASK_0 << i;
2448         shader_hw_map2gl(&tmp_ins);
2449         ++tmp_src[1].reg.idx[0].offset;
2450     }
2451 }
2452
2453 static void shader_hw_rcp(const struct wined3d_shader_instruction *ins)
2454 {
2455     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2456
2457     char dst[50];
2458     char src[50];
2459
2460     shader_arb_get_dst_param(ins, &ins->dst[0], dst); /* Destination */
2461     shader_arb_get_src_param(ins, &ins->src[0], 0, src);
2462     if (ins->src[0].swizzle == WINED3DSP_NOSWIZZLE)
2463     {
2464         /* Dx sdk says .x is used if no swizzle is given, but our test shows that
2465          * .w is used
2466          */
2467         strcat(src, ".w");
2468     }
2469
2470     shader_addline(buffer, "RCP%s %s, %s;\n", shader_arb_get_modifier(ins), dst, src);
2471 }
2472
2473 static void shader_hw_scalar_op(const struct wined3d_shader_instruction *ins)
2474 {
2475     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2476     const char *instruction;
2477
2478     char dst[50];
2479     char src[50];
2480
2481     switch(ins->handler_idx)
2482     {
2483         case WINED3DSIH_RSQ:  instruction = "RSQ"; break;
2484         case WINED3DSIH_RCP:  instruction = "RCP"; break;
2485         case WINED3DSIH_EXP:  instruction = "EX2"; break;
2486         case WINED3DSIH_EXPP: instruction = "EXP"; break;
2487         default: instruction = "";
2488             FIXME("Unhandled opcode %#x\n", ins->handler_idx);
2489             break;
2490     }
2491
2492     shader_arb_get_dst_param(ins, &ins->dst[0], dst); /* Destination */
2493     shader_arb_get_src_param(ins, &ins->src[0], 0, src);
2494     if (ins->src[0].swizzle == WINED3DSP_NOSWIZZLE)
2495     {
2496         /* Dx sdk says .x is used if no swizzle is given, but our test shows that
2497          * .w is used
2498          */
2499         strcat(src, ".w");
2500     }
2501
2502     shader_addline(buffer, "%s%s %s, %s;\n", instruction, shader_arb_get_modifier(ins), dst, src);
2503 }
2504
2505 static void shader_hw_nrm(const struct wined3d_shader_instruction *ins)
2506 {
2507     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2508     char dst_name[50];
2509     char src_name[50];
2510     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2511     BOOL pshader = shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type);
2512     const char *zero = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ZERO);
2513
2514     shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2515     shader_arb_get_src_param(ins, &ins->src[0], 1 /* Use TB */, src_name);
2516
2517     /* In D3D, NRM of a vector with length zero returns zero. Catch this situation, as
2518      * otherwise NRM or RSQ would return NaN */
2519     if(pshader && priv->target_version >= NV3)
2520     {
2521         /* GL_NV_fragment_program2's NRM needs protection against length zero vectors too
2522          *
2523          * TODO: Find out if DP3+NRM+MOV is really faster than DP3+RSQ+MUL
2524          */
2525         shader_addline(buffer, "DP3C TA, %s, %s;\n", src_name, src_name);
2526         shader_addline(buffer, "NRM%s %s, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name);
2527         shader_addline(buffer, "MOV %s (EQ), %s;\n", dst_name, zero);
2528     }
2529     else if(priv->target_version >= NV2)
2530     {
2531         shader_addline(buffer, "DP3C TA.x, %s, %s;\n", src_name, src_name);
2532         shader_addline(buffer, "RSQ TA.x (NE), TA.x;\n");
2533         shader_addline(buffer, "MUL%s %s, %s, TA.x;\n", shader_arb_get_modifier(ins), dst_name,
2534                        src_name);
2535     }
2536     else
2537     {
2538         const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
2539
2540         shader_addline(buffer, "DP3 TA.x, %s, %s;\n", src_name, src_name);
2541         /* Pass any non-zero value to RSQ if the input vector has a length of zero. The
2542          * RSQ result doesn't matter, as long as multiplying it by 0 returns 0.
2543          */
2544         shader_addline(buffer, "SGE TA.y, -TA.x, %s;\n", zero);
2545         shader_addline(buffer, "MAD TA.x, %s, TA.y, TA.x;\n", one);
2546
2547         shader_addline(buffer, "RSQ TA.x, TA.x;\n");
2548         /* dst.w = src[0].w * 1 / (src.x^2 + src.y^2 + src.z^2)^(1/2) according to msdn*/
2549         shader_addline(buffer, "MUL%s %s, %s, TA.x;\n", shader_arb_get_modifier(ins), dst_name,
2550                     src_name);
2551     }
2552 }
2553
2554 static void shader_hw_lrp(const struct wined3d_shader_instruction *ins)
2555 {
2556     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2557     char dst_name[50];
2558     char src_name[3][50];
2559
2560     /* ARB_fragment_program has a convenient LRP instruction */
2561     if(shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type)) {
2562         shader_hw_map2gl(ins);
2563         return;
2564     }
2565
2566     shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2567     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
2568     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
2569     shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
2570
2571     shader_addline(buffer, "SUB TA, %s, %s;\n", src_name[1], src_name[2]);
2572     shader_addline(buffer, "MAD%s %s, %s, TA, %s;\n", shader_arb_get_modifier(ins),
2573                    dst_name, src_name[0], src_name[2]);
2574 }
2575
2576 static void shader_hw_sincos(const struct wined3d_shader_instruction *ins)
2577 {
2578     /* This instruction exists in ARB, but the d3d instruction takes two extra parameters which
2579      * must contain fixed constants. So we need a separate function to filter those constants and
2580      * can't use map2gl
2581      */
2582     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2583     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2584     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2585     char dst_name[50];
2586     char src_name0[50], src_name1[50], src_name2[50];
2587     BOOL is_color;
2588
2589     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name0);
2590     if(shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type)) {
2591         shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2592         /* No modifiers are supported on SCS */
2593         shader_addline(buffer, "SCS %s, %s;\n", dst_name, src_name0);
2594
2595         if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE)
2596         {
2597             shader_arb_get_register_name(ins, &dst->reg, src_name0, &is_color);
2598             shader_addline(buffer, "MOV_SAT %s, %s;\n", dst_name, src_name0);
2599         }
2600     } else if(priv->target_version >= NV2) {
2601         shader_arb_get_register_name(ins, &dst->reg, dst_name, &is_color);
2602
2603         /* Sincos writemask must be .x, .y or .xy */
2604         if(dst->write_mask & WINED3DSP_WRITEMASK_0)
2605             shader_addline(buffer, "COS%s %s.x, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name0);
2606         if(dst->write_mask & WINED3DSP_WRITEMASK_1)
2607             shader_addline(buffer, "SIN%s %s.y, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name0);
2608     } else {
2609         /* Approximate sine and cosine with a taylor series, as per math textbook. The application passes 8
2610          * helper constants(D3DSINCOSCONST1 and D3DSINCOSCONST2) in src1 and src2.
2611          *
2612          * sin(x) = x - x^3/3! + x^5/5! - x^7/7! + ...
2613          * cos(x) = 1 - x^2/2! + x^4/4! - x^6/6! + ...
2614          *
2615          * The constants we get are:
2616          *
2617          *  +1   +1,     -1     -1     +1      +1      -1       -1
2618          *      ---- ,  ---- , ---- , ----- , ----- , ----- , ------
2619          *      1!*2    2!*4   3!*8   4!*16   5!*32   6!*64   7!*128
2620          *
2621          * If used with x^2, x^3, x^4 etc they calculate sin(x/2) and cos(x/2):
2622          *
2623          * (x/2)^2 = x^2 / 4
2624          * (x/2)^3 = x^3 / 8
2625          * (x/2)^4 = x^4 / 16
2626          * (x/2)^5 = x^5 / 32
2627          * etc
2628          *
2629          * To get the final result:
2630          * sin(x) = 2 * sin(x/2) * cos(x/2)
2631          * cos(x) = cos(x/2)^2 - sin(x/2)^2
2632          * (from sin(x+y) and cos(x+y) rules)
2633          *
2634          * As per MSDN, dst.z is undefined after the operation, and so is
2635          * dst.x and dst.y if they're masked out by the writemask. Ie
2636          * sincos dst.y, src1, c0, c1
2637          * returns the sine in dst.y. dst.x and dst.z are undefined, dst.w is not touched. The assembler
2638          * vsa.exe also stops with an error if the dest register is the same register as the source
2639          * register. This means we can use dest.xyz as temporary storage. The assembler vsa.exe output also
2640          * indicates that sincos consumes 8 instruction slots in vs_2_0(and, strangely, in vs_3_0).
2641          */
2642         shader_arb_get_src_param(ins, &ins->src[1], 1, src_name1);
2643         shader_arb_get_src_param(ins, &ins->src[2], 2, src_name2);
2644         shader_arb_get_register_name(ins, &dst->reg, dst_name, &is_color);
2645
2646         shader_addline(buffer, "MUL %s.x, %s, %s;\n", dst_name, src_name0, src_name0);  /* x ^ 2 */
2647         shader_addline(buffer, "MUL TA.y, %s.x, %s;\n", dst_name, src_name0);           /* x ^ 3 */
2648         shader_addline(buffer, "MUL %s.y, TA.y, %s;\n", dst_name, src_name0);           /* x ^ 4 */
2649         shader_addline(buffer, "MUL TA.z, %s.y, %s;\n", dst_name, src_name0);           /* x ^ 5 */
2650         shader_addline(buffer, "MUL %s.z, TA.z, %s;\n", dst_name, src_name0);           /* x ^ 6 */
2651         shader_addline(buffer, "MUL TA.w, %s.z, %s;\n", dst_name, src_name0);           /* x ^ 7 */
2652
2653         /* sin(x/2)
2654          *
2655          * Unfortunately we don't get the constants in a DP4-capable form. Is there a way to
2656          * properly merge that with MULs in the code above?
2657          * The swizzles .yz and xw however fit into the .yzxw swizzle added to ps_2_0. Maybe
2658          * we can merge the sine and cosine MAD rows to calculate them together.
2659          */
2660         shader_addline(buffer, "MUL TA.x, %s, %s.w;\n", src_name0, src_name2); /* x^1, +1/(1!*2) */
2661         shader_addline(buffer, "MAD TA.x, TA.y, %s.x, TA.x;\n", src_name2); /* -1/(3!*8) */
2662         shader_addline(buffer, "MAD TA.x, TA.z, %s.w, TA.x;\n", src_name1); /* +1/(5!*32) */
2663         shader_addline(buffer, "MAD TA.x, TA.w, %s.x, TA.x;\n", src_name1); /* -1/(7!*128) */
2664
2665         /* cos(x/2) */
2666         shader_addline(buffer, "MAD TA.y, %s.x, %s.y, %s.z;\n", dst_name, src_name2, src_name2); /* -1/(2!*4), +1.0 */
2667         shader_addline(buffer, "MAD TA.y, %s.y, %s.z, TA.y;\n", dst_name, src_name1); /* +1/(4!*16) */
2668         shader_addline(buffer, "MAD TA.y, %s.z, %s.y, TA.y;\n", dst_name, src_name1); /* -1/(6!*64) */
2669
2670         if(dst->write_mask & WINED3DSP_WRITEMASK_0) {
2671             /* cos x */
2672             shader_addline(buffer, "MUL TA.z, TA.y, TA.y;\n");
2673             shader_addline(buffer, "MAD %s.x, -TA.x, TA.x, TA.z;\n", dst_name);
2674         }
2675         if(dst->write_mask & WINED3DSP_WRITEMASK_1) {
2676             /* sin x */
2677             shader_addline(buffer, "MUL %s.y, TA.x, TA.y;\n", dst_name);
2678             shader_addline(buffer, "ADD %s.y, %s.y, %s.y;\n", dst_name, dst_name, dst_name);
2679         }
2680     }
2681 }
2682
2683 static void shader_hw_sgn(const struct wined3d_shader_instruction *ins)
2684 {
2685     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2686     char dst_name[50];
2687     char src_name[50];
2688     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
2689
2690     shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2691     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name);
2692
2693     /* SGN is only valid in vertex shaders */
2694     if(ctx->target_version >= NV2) {
2695         shader_addline(buffer, "SSG%s %s, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name);
2696         return;
2697     }
2698
2699     /* If SRC > 0.0, -SRC < SRC = TRUE, otherwise false.
2700      * if SRC < 0.0,  SRC < -SRC = TRUE. If neither is true, src = 0.0
2701      */
2702     if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE) {
2703         shader_addline(buffer, "SLT %s, -%s, %s;\n", dst_name, src_name, src_name);
2704     } else {
2705         /* src contains TA? Write to the dest first. This won't overwrite our destination.
2706          * Then use TA, and calculate the final result
2707          *
2708          * Not reading from TA? Store the first result in TA to avoid overwriting the
2709          * destination if src reg = dst reg
2710          */
2711         if(strstr(src_name, "TA"))
2712         {
2713             shader_addline(buffer, "SLT %s,  %s, -%s;\n", dst_name, src_name, src_name);
2714             shader_addline(buffer, "SLT TA, -%s, %s;\n", src_name, src_name);
2715             shader_addline(buffer, "ADD %s, %s, -TA;\n", dst_name, dst_name);
2716         }
2717         else
2718         {
2719             shader_addline(buffer, "SLT TA, -%s, %s;\n", src_name, src_name);
2720             shader_addline(buffer, "SLT %s,  %s, -%s;\n", dst_name, src_name, src_name);
2721             shader_addline(buffer, "ADD %s, TA, -%s;\n", dst_name, dst_name);
2722         }
2723     }
2724 }
2725
2726 static void shader_hw_dsy(const struct wined3d_shader_instruction *ins)
2727 {
2728     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2729     char src[50];
2730     char dst[50];
2731     char dst_name[50];
2732     BOOL is_color;
2733
2734     shader_arb_get_dst_param(ins, &ins->dst[0], dst);
2735     shader_arb_get_src_param(ins, &ins->src[0], 0, src);
2736     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2737
2738     shader_addline(buffer, "DDY %s, %s;\n", dst, src);
2739     shader_addline(buffer, "MUL%s %s, %s, ycorrection.y;\n", shader_arb_get_modifier(ins), dst, dst_name);
2740 }
2741
2742 static DWORD abs_modifier(DWORD mod, BOOL *need_abs)
2743 {
2744     *need_abs = FALSE;
2745
2746     switch(mod)
2747     {
2748         case WINED3DSPSM_NONE:      return WINED3DSPSM_ABS;
2749         case WINED3DSPSM_NEG:       return WINED3DSPSM_ABS;
2750         case WINED3DSPSM_BIAS:      *need_abs = TRUE; return WINED3DSPSM_BIAS;
2751         case WINED3DSPSM_BIASNEG:   *need_abs = TRUE; return WINED3DSPSM_BIASNEG;
2752         case WINED3DSPSM_SIGN:      *need_abs = TRUE; return WINED3DSPSM_SIGN;
2753         case WINED3DSPSM_SIGNNEG:   *need_abs = TRUE; return WINED3DSPSM_SIGNNEG;
2754         case WINED3DSPSM_COMP:      *need_abs = TRUE; return WINED3DSPSM_COMP;
2755         case WINED3DSPSM_X2:        *need_abs = TRUE; return WINED3DSPSM_X2;
2756         case WINED3DSPSM_X2NEG:     *need_abs = TRUE; return WINED3DSPSM_X2NEG;
2757         case WINED3DSPSM_DZ:        *need_abs = TRUE; return WINED3DSPSM_DZ;
2758         case WINED3DSPSM_DW:        *need_abs = TRUE; return WINED3DSPSM_DW;
2759         case WINED3DSPSM_ABS:       return WINED3DSPSM_ABS;
2760         case WINED3DSPSM_ABSNEG:    return WINED3DSPSM_ABS;
2761     }
2762     FIXME("Unknown modifier %u\n", mod);
2763     return mod;
2764 }
2765
2766 static void shader_hw_log(const struct wined3d_shader_instruction *ins)
2767 {
2768     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2769     char src0[50], dst[50];
2770     struct wined3d_shader_src_param src0_copy = ins->src[0];
2771     BOOL need_abs = FALSE;
2772     const char *instr;
2773
2774     switch(ins->handler_idx)
2775     {
2776         case WINED3DSIH_LOG:  instr = "LG2"; break;
2777         case WINED3DSIH_LOGP: instr = "LOG"; break;
2778         default:
2779             ERR("Unexpected instruction %d\n", ins->handler_idx);
2780             return;
2781     }
2782
2783     /* LOG and LOGP operate on the absolute value of the input */
2784     src0_copy.modifiers = abs_modifier(src0_copy.modifiers, &need_abs);
2785
2786     shader_arb_get_dst_param(ins, &ins->dst[0], dst);
2787     shader_arb_get_src_param(ins, &src0_copy, 0, src0);
2788
2789     if(need_abs)
2790     {
2791         shader_addline(buffer, "ABS TA, %s;\n", src0);
2792         shader_addline(buffer, "%s%s %s, TA;\n", instr, shader_arb_get_modifier(ins), dst);
2793     }
2794     else
2795     {
2796         shader_addline(buffer, "%s%s %s, %s;\n", instr, shader_arb_get_modifier(ins), dst, src0);
2797     }
2798 }
2799
2800 static void shader_hw_pow(const struct wined3d_shader_instruction *ins)
2801 {
2802     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2803     char src0[50], src1[50], dst[50];
2804     struct wined3d_shader_src_param src0_copy = ins->src[0];
2805     BOOL need_abs = FALSE;
2806     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2807     const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
2808
2809     /* POW operates on the absolute value of the input */
2810     src0_copy.modifiers = abs_modifier(src0_copy.modifiers, &need_abs);
2811
2812     shader_arb_get_dst_param(ins, &ins->dst[0], dst);
2813     shader_arb_get_src_param(ins, &src0_copy, 0, src0);
2814     shader_arb_get_src_param(ins, &ins->src[1], 1, src1);
2815
2816     if (need_abs)
2817         shader_addline(buffer, "ABS TA.x, %s;\n", src0);
2818     else
2819         shader_addline(buffer, "MOV TA.x, %s;\n", src0);
2820
2821     if (priv->target_version >= NV2)
2822     {
2823         shader_addline(buffer, "MOVC TA.y, %s;\n", src1);
2824         shader_addline(buffer, "POW%s %s, TA.x, TA.y;\n", shader_arb_get_modifier(ins), dst);
2825         shader_addline(buffer, "MOV %s (EQ.y), %s;\n", dst, one);
2826     }
2827     else
2828     {
2829         const char *zero = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ZERO);
2830         const char *flt_eps = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_EPS);
2831
2832         shader_addline(buffer, "ABS TA.y, %s;\n", src1);
2833         shader_addline(buffer, "SGE TA.y, -TA.y, %s;\n", zero);
2834         /* Possibly add flt_eps to avoid getting float special values */
2835         shader_addline(buffer, "MAD TA.z, TA.y, %s, %s;\n", flt_eps, src1);
2836         shader_addline(buffer, "POW%s TA.x, TA.x, TA.z;\n", shader_arb_get_modifier(ins));
2837         shader_addline(buffer, "MAD TA.x, -TA.x, TA.y, TA.x;\n");
2838         shader_addline(buffer, "MAD %s, TA.y, %s, TA.x;\n", dst, one);
2839     }
2840 }
2841
2842 static void shader_hw_loop(const struct wined3d_shader_instruction *ins)
2843 {
2844     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2845     char src_name[50];
2846     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2847
2848     /* src0 is aL */
2849     shader_arb_get_src_param(ins, &ins->src[1], 0, src_name);
2850
2851     if(vshader)
2852     {
2853         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2854         struct list *e = list_head(&priv->control_frames);
2855         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2856
2857         if(priv->loop_depth > 1) shader_addline(buffer, "PUSHA aL;\n");
2858         /* The constant loader makes sure to load -1 into iX.w */
2859         shader_addline(buffer, "ARLC aL, %s.xywz;\n", src_name);
2860         shader_addline(buffer, "BRA loop_%u_end (LE.x);\n", control_frame->no.loop);
2861         shader_addline(buffer, "loop_%u_start:\n", control_frame->no.loop);
2862     }
2863     else
2864     {
2865         shader_addline(buffer, "LOOP %s;\n", src_name);
2866     }
2867 }
2868
2869 static void shader_hw_rep(const struct wined3d_shader_instruction *ins)
2870 {
2871     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2872     char src_name[50];
2873     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2874
2875     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name);
2876
2877     /* The constant loader makes sure to load -1 into iX.w */
2878     if(vshader)
2879     {
2880         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2881         struct list *e = list_head(&priv->control_frames);
2882         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2883
2884         if(priv->loop_depth > 1) shader_addline(buffer, "PUSHA aL;\n");
2885
2886         shader_addline(buffer, "ARLC aL, %s.xywz;\n", src_name);
2887         shader_addline(buffer, "BRA loop_%u_end (LE.x);\n", control_frame->no.loop);
2888         shader_addline(buffer, "loop_%u_start:\n", control_frame->no.loop);
2889     }
2890     else
2891     {
2892         shader_addline(buffer, "REP %s;\n", src_name);
2893     }
2894 }
2895
2896 static void shader_hw_endloop(const struct wined3d_shader_instruction *ins)
2897 {
2898     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2899     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2900
2901     if(vshader)
2902     {
2903         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2904         struct list *e = list_head(&priv->control_frames);
2905         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2906
2907         shader_addline(buffer, "ARAC aL.xy, aL;\n");
2908         shader_addline(buffer, "BRA loop_%u_start (GT.x);\n", control_frame->no.loop);
2909         shader_addline(buffer, "loop_%u_end:\n", control_frame->no.loop);
2910
2911         if(priv->loop_depth > 1) shader_addline(buffer, "POPA aL;\n");
2912     }
2913     else
2914     {
2915         shader_addline(buffer, "ENDLOOP;\n");
2916     }
2917 }
2918
2919 static void shader_hw_endrep(const struct wined3d_shader_instruction *ins)
2920 {
2921     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2922     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2923
2924     if(vshader)
2925     {
2926         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2927         struct list *e = list_head(&priv->control_frames);
2928         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2929
2930         shader_addline(buffer, "ARAC aL.xy, aL;\n");
2931         shader_addline(buffer, "BRA loop_%u_start (GT.x);\n", control_frame->no.loop);
2932         shader_addline(buffer, "loop_%u_end:\n", control_frame->no.loop);
2933
2934         if(priv->loop_depth > 1) shader_addline(buffer, "POPA aL;\n");
2935     }
2936     else
2937     {
2938         shader_addline(buffer, "ENDREP;\n");
2939     }
2940 }
2941
2942 static const struct control_frame *find_last_loop(const struct shader_arb_ctx_priv *priv)
2943 {
2944     struct control_frame *control_frame;
2945
2946     LIST_FOR_EACH_ENTRY(control_frame, &priv->control_frames, struct control_frame, entry)
2947     {
2948         if(control_frame->type == LOOP || control_frame->type == REP) return control_frame;
2949     }
2950     ERR("Could not find loop for break\n");
2951     return NULL;
2952 }
2953
2954 static void shader_hw_break(const struct wined3d_shader_instruction *ins)
2955 {
2956     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2957     const struct control_frame *control_frame = find_last_loop(ins->ctx->backend_data);
2958     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2959
2960     if(vshader)
2961     {
2962         shader_addline(buffer, "BRA loop_%u_end;\n", control_frame->no.loop);
2963     }
2964     else
2965     {
2966         shader_addline(buffer, "BRK;\n");
2967     }
2968 }
2969
2970 static const char *get_compare(enum wined3d_shader_rel_op op)
2971 {
2972     switch (op)
2973     {
2974         case WINED3D_SHADER_REL_OP_GT: return "GT";
2975         case WINED3D_SHADER_REL_OP_EQ: return "EQ";
2976         case WINED3D_SHADER_REL_OP_GE: return "GE";
2977         case WINED3D_SHADER_REL_OP_LT: return "LT";
2978         case WINED3D_SHADER_REL_OP_NE: return "NE";
2979         case WINED3D_SHADER_REL_OP_LE: return "LE";
2980         default:
2981             FIXME("Unrecognized operator %#x.\n", op);
2982             return "(\?\?)";
2983     }
2984 }
2985
2986 static enum wined3d_shader_rel_op invert_compare(enum wined3d_shader_rel_op op)
2987 {
2988     switch (op)
2989     {
2990         case WINED3D_SHADER_REL_OP_GT: return WINED3D_SHADER_REL_OP_LE;
2991         case WINED3D_SHADER_REL_OP_EQ: return WINED3D_SHADER_REL_OP_NE;
2992         case WINED3D_SHADER_REL_OP_GE: return WINED3D_SHADER_REL_OP_LT;
2993         case WINED3D_SHADER_REL_OP_LT: return WINED3D_SHADER_REL_OP_GE;
2994         case WINED3D_SHADER_REL_OP_NE: return WINED3D_SHADER_REL_OP_EQ;
2995         case WINED3D_SHADER_REL_OP_LE: return WINED3D_SHADER_REL_OP_GT;
2996         default:
2997             FIXME("Unrecognized operator %#x.\n", op);
2998             return -1;
2999     }
3000 }
3001
3002 static void shader_hw_breakc(const struct wined3d_shader_instruction *ins)
3003 {
3004     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3005     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
3006     const struct control_frame *control_frame = find_last_loop(ins->ctx->backend_data);
3007     char src_name0[50];
3008     char src_name1[50];
3009     const char *comp = get_compare(ins->flags);
3010
3011     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name0);
3012     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name1);
3013
3014     if(vshader)
3015     {
3016         /* SUBC CC, src0, src1" works only in pixel shaders, so use TA to throw
3017          * away the subtraction result
3018          */
3019         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
3020         shader_addline(buffer, "BRA loop_%u_end (%s.x);\n", control_frame->no.loop, comp);
3021     }
3022     else
3023     {
3024         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
3025         shader_addline(buffer, "BRK (%s.x);\n", comp);
3026     }
3027 }
3028
3029 static void shader_hw_ifc(const struct wined3d_shader_instruction *ins)
3030 {
3031     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3032     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3033     struct list *e = list_head(&priv->control_frames);
3034     struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
3035     const char *comp;
3036     char src_name0[50];
3037     char src_name1[50];
3038     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
3039
3040     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name0);
3041     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name1);
3042
3043     if(vshader)
3044     {
3045         /* Invert the flag. We jump to the else label if the condition is NOT true */
3046         comp = get_compare(invert_compare(ins->flags));
3047         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
3048         shader_addline(buffer, "BRA ifc_%u_else (%s.x);\n", control_frame->no.ifc, comp);
3049     }
3050     else
3051     {
3052         comp = get_compare(ins->flags);
3053         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
3054         shader_addline(buffer, "IF %s.x;\n", comp);
3055     }
3056 }
3057
3058 static void shader_hw_else(const struct wined3d_shader_instruction *ins)
3059 {
3060     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3061     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3062     struct list *e = list_head(&priv->control_frames);
3063     struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
3064     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
3065
3066     if(vshader)
3067     {
3068         shader_addline(buffer, "BRA ifc_%u_endif;\n", control_frame->no.ifc);
3069         shader_addline(buffer, "ifc_%u_else:\n", control_frame->no.ifc);
3070         control_frame->had_else = TRUE;
3071     }
3072     else
3073     {
3074         shader_addline(buffer, "ELSE;\n");
3075     }
3076 }
3077
3078 static void shader_hw_endif(const struct wined3d_shader_instruction *ins)
3079 {
3080     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3081     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3082     struct list *e = list_head(&priv->control_frames);
3083     struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
3084     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
3085
3086     if(vshader)
3087     {
3088         if(control_frame->had_else)
3089         {
3090             shader_addline(buffer, "ifc_%u_endif:\n", control_frame->no.ifc);
3091         }
3092         else
3093         {
3094             shader_addline(buffer, "#No else branch. else is endif\n");
3095             shader_addline(buffer, "ifc_%u_else:\n", control_frame->no.ifc);
3096         }
3097     }
3098     else
3099     {
3100         shader_addline(buffer, "ENDIF;\n");
3101     }
3102 }
3103
3104 static void shader_hw_texldd(const struct wined3d_shader_instruction *ins)
3105 {
3106     DWORD sampler_idx = ins->src[1].reg.idx[0].offset;
3107     char reg_dest[40];
3108     char reg_src[3][40];
3109     WORD flags = TEX_DERIV;
3110
3111     shader_arb_get_dst_param(ins, &ins->dst[0], reg_dest);
3112     shader_arb_get_src_param(ins, &ins->src[0], 0, reg_src[0]);
3113     shader_arb_get_src_param(ins, &ins->src[2], 1, reg_src[1]);
3114     shader_arb_get_src_param(ins, &ins->src[3], 2, reg_src[2]);
3115
3116     if (ins->flags & WINED3DSI_TEXLD_PROJECT) flags |= TEX_PROJ;
3117     if (ins->flags & WINED3DSI_TEXLD_BIAS) flags |= TEX_BIAS;
3118
3119     shader_hw_sample(ins, sampler_idx, reg_dest, reg_src[0], flags, reg_src[1], reg_src[2]);
3120 }
3121
3122 static void shader_hw_texldl(const struct wined3d_shader_instruction *ins)
3123 {
3124     DWORD sampler_idx = ins->src[1].reg.idx[0].offset;
3125     char reg_dest[40];
3126     char reg_coord[40];
3127     WORD flags = TEX_LOD;
3128
3129     shader_arb_get_dst_param(ins, &ins->dst[0], reg_dest);
3130     shader_arb_get_src_param(ins, &ins->src[0], 0, reg_coord);
3131
3132     if (ins->flags & WINED3DSI_TEXLD_PROJECT) flags |= TEX_PROJ;
3133     if (ins->flags & WINED3DSI_TEXLD_BIAS) flags |= TEX_BIAS;
3134
3135     shader_hw_sample(ins, sampler_idx, reg_dest, reg_coord, flags, NULL, NULL);
3136 }
3137
3138 static void shader_hw_label(const struct wined3d_shader_instruction *ins)
3139 {
3140     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3141     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3142
3143     priv->in_main_func = FALSE;
3144     /* Call instructions activate the NV extensions, not labels and rets. If there is an uncalled
3145      * subroutine, don't generate a label that will make GL complain
3146      */
3147     if(priv->target_version == ARB) return;
3148
3149     shader_addline(buffer, "l%u:\n", ins->src[0].reg.idx[0].offset);
3150 }
3151
3152 static void vshader_add_footer(struct shader_arb_ctx_priv *priv_ctx,
3153         const struct arb_vshader_private *shader_data, const struct arb_vs_compile_args *args,
3154         const struct wined3d_shader_reg_maps *reg_maps, const struct wined3d_gl_info *gl_info,
3155         struct wined3d_shader_buffer *buffer)
3156 {
3157     unsigned int i;
3158
3159     /* The D3DRS_FOGTABLEMODE render state defines if the shader-generated fog coord is used
3160      * or if the fragment depth is used. If the fragment depth is used(FOGTABLEMODE != NONE),
3161      * the fog frag coord is thrown away. If the fog frag coord is used, but not written by
3162      * the shader, it is set to 0.0(fully fogged, since start = 1.0, end = 0.0)
3163      */
3164     if (args->super.fog_src == VS_FOG_Z)
3165     {
3166         shader_addline(buffer, "MOV result.fogcoord, TMP_OUT.z;\n");
3167     }
3168     else
3169     {
3170         if (!reg_maps->fog)
3171         {
3172             /* posFixup.x is always 1.0, so we can safely use it */
3173             shader_addline(buffer, "ADD result.fogcoord, posFixup.x, -posFixup.x;\n");
3174         }
3175         else
3176         {
3177             /* Clamp fogcoord */
3178             const char *zero = arb_get_helper_value(reg_maps->shader_version.type, ARB_ZERO);
3179             const char *one = arb_get_helper_value(reg_maps->shader_version.type, ARB_ONE);
3180
3181             shader_addline(buffer, "MIN TMP_FOGCOORD.x, TMP_FOGCOORD.x, %s;\n", one);
3182             shader_addline(buffer, "MAX result.fogcoord.x, TMP_FOGCOORD.x, %s;\n", zero);
3183         }
3184     }
3185
3186     /* Clipplanes are always stored without y inversion */
3187     if (use_nv_clip(gl_info) && priv_ctx->target_version >= NV2)
3188     {
3189         if (args->super.clip_enabled)
3190         {
3191             for (i = 0; i < priv_ctx->vs_clipplanes; i++)
3192             {
3193                 shader_addline(buffer, "DP4 result.clip[%u].x, TMP_OUT, state.clip[%u].plane;\n", i, i);
3194             }
3195         }
3196     }
3197     else if (args->clip.boolclip.clip_texcoord)
3198     {
3199         unsigned int cur_clip = 0;
3200         char component[4] = {'x', 'y', 'z', 'w'};
3201         const char *zero = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_ZERO);
3202
3203         for (i = 0; i < gl_info->limits.clipplanes; ++i)
3204         {
3205             if (args->clip.boolclip.clipplane_mask & (1 << i))
3206             {
3207                 shader_addline(buffer, "DP4 TA.%c, TMP_OUT, state.clip[%u].plane;\n",
3208                                component[cur_clip++], i);
3209             }
3210         }
3211         switch (cur_clip)
3212         {
3213             case 0:
3214                 shader_addline(buffer, "MOV TA, %s;\n", zero);
3215                 break;
3216             case 1:
3217                 shader_addline(buffer, "MOV TA.yzw, %s;\n", zero);
3218                 break;
3219             case 2:
3220                 shader_addline(buffer, "MOV TA.zw, %s;\n", zero);
3221                 break;
3222             case 3:
3223                 shader_addline(buffer, "MOV TA.w, %s;\n", zero);
3224                 break;
3225         }
3226         shader_addline(buffer, "MOV result.texcoord[%u], TA;\n",
3227                        args->clip.boolclip.clip_texcoord - 1);
3228     }
3229
3230     /* Write the final position.
3231      *
3232      * OpenGL coordinates specify the center of the pixel while d3d coords specify
3233      * the corner. The offsets are stored in z and w in posFixup. posFixup.y contains
3234      * 1.0 or -1.0 to turn the rendering upside down for offscreen rendering. PosFixup.x
3235      * contains 1.0 to allow a mad, but arb vs swizzles are too restricted for that.
3236      */
3237     shader_addline(buffer, "MUL TA, posFixup, TMP_OUT.w;\n");
3238     shader_addline(buffer, "ADD TMP_OUT.x, TMP_OUT.x, TA.z;\n");
3239     shader_addline(buffer, "MAD TMP_OUT.y, TMP_OUT.y, posFixup.y, TA.w;\n");
3240
3241     /* Z coord [0;1]->[-1;1] mapping, see comment in transform_projection in state.c
3242      * and the glsl equivalent
3243      */
3244     if (need_helper_const(shader_data, reg_maps, gl_info))
3245     {
3246         const char *two = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_TWO);
3247         shader_addline(buffer, "MAD TMP_OUT.z, TMP_OUT.z, %s, -TMP_OUT.w;\n", two);
3248     }
3249     else
3250     {
3251         shader_addline(buffer, "ADD TMP_OUT.z, TMP_OUT.z, TMP_OUT.z;\n");
3252         shader_addline(buffer, "ADD TMP_OUT.z, TMP_OUT.z, -TMP_OUT.w;\n");
3253     }
3254
3255     shader_addline(buffer, "MOV result.position, TMP_OUT;\n");
3256
3257     priv_ctx->footer_written = TRUE;
3258 }
3259
3260 static void shader_hw_ret(const struct wined3d_shader_instruction *ins)
3261 {
3262     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3263     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3264     const struct wined3d_shader *shader = ins->ctx->shader;
3265     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
3266
3267     if(priv->target_version == ARB) return;
3268
3269     if(vshader)
3270     {
3271         if (priv->in_main_func) vshader_add_footer(priv, shader->backend_data,
3272                 priv->cur_vs_args, ins->ctx->reg_maps, ins->ctx->gl_info, buffer);
3273     }
3274
3275     shader_addline(buffer, "RET;\n");
3276 }
3277
3278 static void shader_hw_call(const struct wined3d_shader_instruction *ins)
3279 {
3280     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3281     shader_addline(buffer, "CAL l%u;\n", ins->src[0].reg.idx[0].offset);
3282 }
3283
3284 /* Context activation is done by the caller. */
3285 static GLuint create_arb_blt_vertex_program(const struct wined3d_gl_info *gl_info)
3286 {
3287     GLuint program_id = 0;
3288     GLint pos;
3289
3290     const char *blt_vprogram =
3291         "!!ARBvp1.0\n"
3292         "PARAM c[1] = { { 1, 0.5 } };\n"
3293         "MOV result.position, vertex.position;\n"
3294         "MOV result.color, c[0].x;\n"
3295         "MOV result.texcoord[0], vertex.texcoord[0];\n"
3296         "END\n";
3297
3298     GL_EXTCALL(glGenProgramsARB(1, &program_id));
3299     GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, program_id));
3300     GL_EXTCALL(glProgramStringARB(GL_VERTEX_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
3301             strlen(blt_vprogram), blt_vprogram));
3302     checkGLcall("glProgramStringARB()");
3303
3304     gl_info->gl_ops.gl.p_glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
3305     if (pos != -1)
3306     {
3307         FIXME("Vertex program error at position %d: %s\n\n", pos,
3308             debugstr_a((const char *)gl_info->gl_ops.gl.p_glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
3309         shader_arb_dump_program_source(blt_vprogram);
3310     }
3311     else
3312     {
3313         GLint native;
3314
3315         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
3316         checkGLcall("glGetProgramivARB()");
3317         if (!native) WARN("Program exceeds native resource limits.\n");
3318     }
3319
3320     return program_id;
3321 }
3322
3323 /* Context activation is done by the caller. */
3324 static GLuint create_arb_blt_fragment_program(const struct wined3d_gl_info *gl_info,
3325         enum tex_types tex_type, BOOL masked)
3326 {
3327     GLuint program_id = 0;
3328     const char *fprogram;
3329     GLint pos;
3330
3331     static const char * const blt_fprograms_full[tex_type_count] =
3332     {
3333         /* tex_1d */
3334         NULL,
3335         /* tex_2d */
3336         "!!ARBfp1.0\n"
3337         "TEMP R0;\n"
3338         "TEX R0.x, fragment.texcoord[0], texture[0], 2D;\n"
3339         "MOV result.depth.z, R0.x;\n"
3340         "END\n",
3341         /* tex_3d */
3342         NULL,
3343         /* tex_cube */
3344         "!!ARBfp1.0\n"
3345         "TEMP R0;\n"
3346         "TEX R0.x, fragment.texcoord[0], texture[0], CUBE;\n"
3347         "MOV result.depth.z, R0.x;\n"
3348         "END\n",
3349         /* tex_rect */
3350         "!!ARBfp1.0\n"
3351         "TEMP R0;\n"
3352         "TEX R0.x, fragment.texcoord[0], texture[0], RECT;\n"
3353         "MOV result.depth.z, R0.x;\n"
3354         "END\n",
3355     };
3356
3357     static const char * const blt_fprograms_masked[tex_type_count] =
3358     {
3359         /* tex_1d */
3360         NULL,
3361         /* tex_2d */
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], 2D;\n"
3369         "MOV result.depth.z, R0.x;\n"
3370         "END\n",
3371         /* tex_3d */
3372         NULL,
3373         /* tex_cube */
3374         "!!ARBfp1.0\n"
3375         "PARAM mask = program.local[0];\n"
3376         "TEMP R0;\n"
3377         "SLT R0.xy, fragment.position, mask.zwzw;\n"
3378         "MUL R0.x, R0.x, R0.y;\n"
3379         "KIL -R0.x;\n"
3380         "TEX R0.x, fragment.texcoord[0], texture[0], CUBE;\n"
3381         "MOV result.depth.z, R0.x;\n"
3382         "END\n",
3383         /* tex_rect */
3384         "!!ARBfp1.0\n"
3385         "PARAM mask = program.local[0];\n"
3386         "TEMP R0;\n"
3387         "SLT R0.xy, fragment.position, mask.zwzw;\n"
3388         "MUL R0.x, R0.x, R0.y;\n"
3389         "KIL -R0.x;\n"
3390         "TEX R0.x, fragment.texcoord[0], texture[0], RECT;\n"
3391         "MOV result.depth.z, R0.x;\n"
3392         "END\n",
3393     };
3394
3395     fprogram = masked ? blt_fprograms_masked[tex_type] : blt_fprograms_full[tex_type];
3396     if (!fprogram)
3397     {
3398         FIXME("tex_type %#x not supported, falling back to tex_2d\n", tex_type);
3399         tex_type = tex_2d;
3400         fprogram = masked ? blt_fprograms_masked[tex_type] : blt_fprograms_full[tex_type];
3401     }
3402
3403     GL_EXTCALL(glGenProgramsARB(1, &program_id));
3404     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, program_id));
3405     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB, strlen(fprogram), fprogram));
3406     checkGLcall("glProgramStringARB()");
3407
3408     gl_info->gl_ops.gl.p_glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
3409     if (pos != -1)
3410     {
3411         FIXME("Fragment program error at position %d: %s\n\n", pos,
3412             debugstr_a((const char *)gl_info->gl_ops.gl.p_glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
3413         shader_arb_dump_program_source(fprogram);
3414     }
3415     else
3416     {
3417         GLint native;
3418
3419         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
3420         checkGLcall("glGetProgramivARB()");
3421         if (!native) WARN("Program exceeds native resource limits.\n");
3422     }
3423
3424     return program_id;
3425 }
3426
3427 static void arbfp_add_sRGB_correction(struct wined3d_shader_buffer *buffer, const char *fragcolor,
3428         const char *tmp1, const char *tmp2, const char *tmp3, const char *tmp4, BOOL condcode)
3429 {
3430     /* Perform sRGB write correction. See GLX_EXT_framebuffer_sRGB */
3431
3432     if(condcode)
3433     {
3434         /* Sigh. MOVC CC doesn't work, so use one of the temps as dummy dest */
3435         shader_addline(buffer, "SUBC %s, %s.x, srgb_consts1.y;\n", tmp1, fragcolor);
3436         /* Calculate the > 0.0031308 case */
3437         shader_addline(buffer, "POW %s.x (GE), %s.x, srgb_consts1.z;\n", fragcolor, fragcolor);
3438         shader_addline(buffer, "POW %s.y (GE), %s.y, srgb_consts1.z;\n", fragcolor, fragcolor);
3439         shader_addline(buffer, "POW %s.z (GE), %s.z, srgb_consts1.z;\n", fragcolor, fragcolor);
3440         shader_addline(buffer, "MUL %s.xyz (GE), %s, srgb_consts1.w;\n", fragcolor, fragcolor);
3441         shader_addline(buffer, "SUB %s.xyz (GE), %s, srgb_consts2.x;\n", fragcolor, fragcolor);
3442         /* Calculate the < case */
3443         shader_addline(buffer, "MUL %s.xyz (LT), srgb_consts1.x, %s;\n", fragcolor, fragcolor);
3444     }
3445     else
3446     {
3447         /* Calculate the > 0.0031308 case */
3448         shader_addline(buffer, "POW %s.x, %s.x, srgb_consts1.z;\n", tmp1, fragcolor);
3449         shader_addline(buffer, "POW %s.y, %s.y, srgb_consts1.z;\n", tmp1, fragcolor);
3450         shader_addline(buffer, "POW %s.z, %s.z, srgb_consts1.z;\n", tmp1, fragcolor);
3451         shader_addline(buffer, "MUL %s, %s, srgb_consts1.w;\n", tmp1, tmp1);
3452         shader_addline(buffer, "SUB %s, %s, srgb_consts2.x;\n", tmp1, tmp1);
3453         /* Calculate the < case */
3454         shader_addline(buffer, "MUL %s, srgb_consts1.x, %s;\n", tmp2, fragcolor);
3455         /* Get 1.0 / 0.0 masks for > 0.0031308 and < 0.0031308 */
3456         shader_addline(buffer, "SLT %s, srgb_consts1.y, %s;\n", tmp3, fragcolor);
3457         shader_addline(buffer, "SGE %s, srgb_consts1.y, %s;\n", tmp4, fragcolor);
3458         /* Store the components > 0.0031308 in the destination */
3459         shader_addline(buffer, "MUL %s.xyz, %s, %s;\n", fragcolor, tmp1, tmp3);
3460         /* Add the components that are < 0.0031308 */
3461         shader_addline(buffer, "MAD %s.xyz, %s, %s, %s;\n", fragcolor, tmp2, tmp4, fragcolor);
3462         /* Move everything into result.color at once. Nvidia hardware cannot handle partial
3463         * result.color writes(.rgb first, then .a), or handle overwriting already written
3464         * components. The assembler uses a temporary register in this case, which is usually
3465         * not allocated from one of our registers that were used earlier.
3466         */
3467     }
3468     /* [0.0;1.0] clamping. Not needed, this is done implicitly */
3469 }
3470
3471 static const DWORD *find_loop_control_values(const struct wined3d_shader *shader, DWORD idx)
3472 {
3473     const struct wined3d_shader_lconst *constant;
3474
3475     LIST_FOR_EACH_ENTRY(constant, &shader->constantsI, struct wined3d_shader_lconst, entry)
3476     {
3477         if (constant->idx == idx)
3478         {
3479             return constant->value;
3480         }
3481     }
3482     return NULL;
3483 }
3484
3485 static void init_ps_input(const struct wined3d_shader *shader,
3486         const struct arb_ps_compile_args *args, struct shader_arb_ctx_priv *priv)
3487 {
3488     static const char * const texcoords[8] =
3489     {
3490         "fragment.texcoord[0]", "fragment.texcoord[1]", "fragment.texcoord[2]", "fragment.texcoord[3]",
3491         "fragment.texcoord[4]", "fragment.texcoord[5]", "fragment.texcoord[6]", "fragment.texcoord[7]"
3492     };
3493     unsigned int i;
3494     const struct wined3d_shader_signature_element *sig = shader->input_signature;
3495     const char *semantic_name;
3496     DWORD semantic_idx;
3497
3498     switch(args->super.vp_mode)
3499     {
3500         case pretransformed:
3501         case fixedfunction:
3502             /* The pixelshader has to collect the varyings on its own. In any case properly load
3503              * color0 and color1. In the case of pretransformed vertices also load texcoords. Set
3504              * other attribs to 0.0.
3505              *
3506              * For fixedfunction this behavior is correct, according to the tests. For pretransformed
3507              * we'd either need a replacement shader that can load other attribs like BINORMAL, or
3508              * load the texcoord attrib pointers to match the pixel shader signature
3509              */
3510             for(i = 0; i < MAX_REG_INPUT; i++)
3511             {
3512                 semantic_name = sig[i].semantic_name;
3513                 semantic_idx = sig[i].semantic_idx;
3514                 if (!semantic_name) continue;
3515
3516                 if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_COLOR))
3517                 {
3518                     if (!semantic_idx) priv->ps_input[i] = "fragment.color.primary";
3519                     else if(semantic_idx == 1) priv->ps_input[i] = "fragment.color.secondary";
3520                     else priv->ps_input[i] = "0.0";
3521                 }
3522                 else if(args->super.vp_mode == fixedfunction)
3523                 {
3524                     priv->ps_input[i] = "0.0";
3525                 }
3526                 else if(shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_TEXCOORD))
3527                 {
3528                     if(semantic_idx < 8) priv->ps_input[i] = texcoords[semantic_idx];
3529                     else priv->ps_input[i] = "0.0";
3530                 }
3531                 else if(shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_FOG))
3532                 {
3533                     if (!semantic_idx) priv->ps_input[i] = "fragment.fogcoord";
3534                     else priv->ps_input[i] = "0.0";
3535                 }
3536                 else
3537                 {
3538                     priv->ps_input[i] = "0.0";
3539                 }
3540
3541                 TRACE("v%u, semantic %s%u is %s\n", i, semantic_name, semantic_idx, priv->ps_input[i]);
3542             }
3543             break;
3544
3545         case vertexshader:
3546             /* That one is easy. The vertex shaders provide v0-v7 in fragment.texcoord and v8 and v9 in
3547              * fragment.color
3548              */
3549             for(i = 0; i < 8; i++)
3550             {
3551                 priv->ps_input[i] = texcoords[i];
3552             }
3553             priv->ps_input[8] = "fragment.color.primary";
3554             priv->ps_input[9] = "fragment.color.secondary";
3555             break;
3556     }
3557 }
3558
3559 /* Context activation is done by the caller. */
3560 static GLuint shader_arb_generate_pshader(const struct wined3d_shader *shader,
3561         const struct wined3d_gl_info *gl_info, struct wined3d_shader_buffer *buffer,
3562         const struct arb_ps_compile_args *args, struct arb_ps_compiled_shader *compiled)
3563 {
3564     const struct wined3d_shader_reg_maps *reg_maps = &shader->reg_maps;
3565     const DWORD *function = shader->function;
3566     GLuint retval;
3567     char fragcolor[16];
3568     DWORD next_local = 0;
3569     struct shader_arb_ctx_priv priv_ctx;
3570     BOOL dcl_td = FALSE;
3571     BOOL want_nv_prog = FALSE;
3572     struct arb_pshader_private *shader_priv = shader->backend_data;
3573     GLint errPos;
3574     DWORD map;
3575
3576     char srgbtmp[4][4];
3577     unsigned int i, found = 0;
3578
3579     for (i = 0, map = reg_maps->temporary; map; map >>= 1, ++i)
3580     {
3581         if (!(map & 1)
3582                 || (shader->u.ps.color0_mov && i == shader->u.ps.color0_reg)
3583                 || (reg_maps->shader_version.major < 2 && !i))
3584             continue;
3585
3586         sprintf(srgbtmp[found], "R%u", i);
3587         ++found;
3588         if (found == 4) break;
3589     }
3590
3591     switch(found) {
3592         case 0:
3593             sprintf(srgbtmp[0], "TA");
3594             sprintf(srgbtmp[1], "TB");
3595             sprintf(srgbtmp[2], "TC");
3596             sprintf(srgbtmp[3], "TD");
3597             dcl_td = TRUE;
3598             break;
3599         case 1:
3600             sprintf(srgbtmp[1], "TA");
3601             sprintf(srgbtmp[2], "TB");
3602             sprintf(srgbtmp[3], "TC");
3603             break;
3604         case 2:
3605             sprintf(srgbtmp[2], "TA");
3606             sprintf(srgbtmp[3], "TB");
3607             break;
3608         case 3:
3609             sprintf(srgbtmp[3], "TA");
3610             break;
3611         case 4:
3612             break;
3613     }
3614
3615     /*  Create the hw ARB shader */
3616     memset(&priv_ctx, 0, sizeof(priv_ctx));
3617     priv_ctx.cur_ps_args = args;
3618     priv_ctx.compiled_fprog = compiled;
3619     priv_ctx.cur_np2fixup_info = &compiled->np2fixup_info;
3620     init_ps_input(shader, args, &priv_ctx);
3621     list_init(&priv_ctx.control_frames);
3622
3623     /* Avoid enabling NV_fragment_program* if we do not need it.
3624      *
3625      * Enabling GL_NV_fragment_program_option causes the driver to occupy a temporary register,
3626      * and it slows down the shader execution noticeably(about 5%). Usually our instruction emulation
3627      * is faster than what we gain from using higher native instructions. There are some things though
3628      * that cannot be emulated. In that case enable the extensions.
3629      * If the extension is enabled, instruction handlers that support both ways will use it.
3630      *
3631      * Testing shows no performance difference between OPTION NV_fragment_program2 and NV_fragment_program.
3632      * So enable the best we can get.
3633      */
3634     if(reg_maps->usesdsx || reg_maps->usesdsy || reg_maps->loop_depth > 0 || reg_maps->usestexldd ||
3635        reg_maps->usestexldl || reg_maps->usesfacing || reg_maps->usesifc || reg_maps->usescall)
3636     {
3637         want_nv_prog = TRUE;
3638     }
3639
3640     shader_addline(buffer, "!!ARBfp1.0\n");
3641     if (want_nv_prog && gl_info->supported[NV_FRAGMENT_PROGRAM2])
3642     {
3643         shader_addline(buffer, "OPTION NV_fragment_program2;\n");
3644         priv_ctx.target_version = NV3;
3645     }
3646     else if (want_nv_prog && gl_info->supported[NV_FRAGMENT_PROGRAM_OPTION])
3647     {
3648         shader_addline(buffer, "OPTION NV_fragment_program;\n");
3649         priv_ctx.target_version = NV2;
3650     } else {
3651         if(want_nv_prog)
3652         {
3653             /* This is an error - either we're advertising the wrong shader version, or aren't enforcing some
3654              * limits properly
3655              */
3656             ERR("The shader requires instructions that are not available in plain GL_ARB_fragment_program\n");
3657             ERR("Try GLSL\n");
3658         }
3659         priv_ctx.target_version = ARB;
3660     }
3661
3662     if (reg_maps->rt_mask > 1)
3663     {
3664         shader_addline(buffer, "OPTION ARB_draw_buffers;\n");
3665     }
3666
3667     if (reg_maps->shader_version.major < 3)
3668     {
3669         switch(args->super.fog) {
3670             case FOG_OFF:
3671                 break;
3672             case FOG_LINEAR:
3673                 shader_addline(buffer, "OPTION ARB_fog_linear;\n");
3674                 break;
3675             case FOG_EXP:
3676                 shader_addline(buffer, "OPTION ARB_fog_exp;\n");
3677                 break;
3678             case FOG_EXP2:
3679                 shader_addline(buffer, "OPTION ARB_fog_exp2;\n");
3680                 break;
3681         }
3682     }
3683
3684     /* For now always declare the temps. At least the Nvidia assembler optimizes completely
3685      * unused temps away(but occupies them for the whole shader if they're used once). Always
3686      * declaring them avoids tricky bookkeeping work
3687      */
3688     shader_addline(buffer, "TEMP TA;\n");      /* Used for modifiers */
3689     shader_addline(buffer, "TEMP TB;\n");      /* Used for modifiers */
3690     shader_addline(buffer, "TEMP TC;\n");      /* Used for modifiers */
3691     if(dcl_td) shader_addline(buffer, "TEMP TD;\n"); /* Used for sRGB writing */
3692     shader_addline(buffer, "PARAM coefdiv = { 0.5, 0.25, 0.125, 0.0625 };\n");
3693     shader_addline(buffer, "PARAM coefmul = { 2, 4, 8, 16 };\n");
3694     shader_addline(buffer, "PARAM ps_helper_const = { 0.0, 1.0, %1.10f, 0.0 };\n", eps);
3695
3696     if (reg_maps->shader_version.major < 2)
3697     {
3698         strcpy(fragcolor, "R0");
3699     }
3700     else
3701     {
3702         if (args->super.srgb_correction)
3703         {
3704             if (shader->u.ps.color0_mov)
3705             {
3706                 sprintf(fragcolor, "R%u", shader->u.ps.color0_reg);
3707             }
3708             else
3709             {
3710                 shader_addline(buffer, "TEMP TMP_COLOR;\n");
3711                 strcpy(fragcolor, "TMP_COLOR");
3712             }
3713         } else {
3714             strcpy(fragcolor, "result.color");
3715         }
3716     }
3717
3718     if(args->super.srgb_correction) {
3719         shader_addline(buffer, "PARAM srgb_consts1 = {%f, %f, %f, %f};\n",
3720                        srgb_mul_low, srgb_cmp, srgb_pow, srgb_mul_high);
3721         shader_addline(buffer, "PARAM srgb_consts2 = {%f, %f, %f, %f};\n",
3722                        srgb_sub_high, 0.0, 0.0, 0.0);
3723     }
3724
3725     /* Base Declarations */
3726     shader_generate_arb_declarations(shader, reg_maps, buffer, gl_info, NULL, &priv_ctx);
3727
3728     for (i = 0, map = reg_maps->bumpmat; map; map >>= 1, ++i)
3729     {
3730         unsigned char bump_const;
3731
3732         if (!(map & 1)) continue;
3733
3734         bump_const = compiled->numbumpenvmatconsts;
3735         compiled->bumpenvmatconst[bump_const].const_num = WINED3D_CONST_NUM_UNUSED;
3736         compiled->bumpenvmatconst[bump_const].texunit = i;
3737         compiled->luminanceconst[bump_const].const_num = WINED3D_CONST_NUM_UNUSED;
3738         compiled->luminanceconst[bump_const].texunit = i;
3739
3740         /* We can fit the constants into the constant limit for sure because texbem, texbeml, bem and beml are only supported
3741          * in 1.x shaders, and GL_ARB_fragment_program has a constant limit of 24 constants. So in the worst case we're loading
3742          * 8 shader constants, 8 bump matrices and 8 luminance parameters and are perfectly fine. (No NP2 fixup on bumpmapped
3743          * textures due to conditional NP2 restrictions)
3744          *
3745          * Use local constants to load the bump env parameters, not program.env. This avoids collisions with d3d constants of
3746          * shaders in newer shader models. Since the bump env parameters have to share their space with NP2 fixup constants,
3747          * their location is shader dependent anyway and they cannot be loaded globally.
3748          */
3749         compiled->bumpenvmatconst[bump_const].const_num = next_local++;
3750         shader_addline(buffer, "PARAM bumpenvmat%d = program.local[%d];\n",
3751                        i, compiled->bumpenvmatconst[bump_const].const_num);
3752         compiled->numbumpenvmatconsts = bump_const + 1;
3753
3754         if (!(reg_maps->luminanceparams & (1 << i))) continue;
3755
3756         compiled->luminanceconst[bump_const].const_num = next_local++;
3757         shader_addline(buffer, "PARAM luminance%d = program.local[%d];\n",
3758                        i, compiled->luminanceconst[bump_const].const_num);
3759     }
3760
3761     for(i = 0; i < MAX_CONST_I; i++)
3762     {
3763         compiled->int_consts[i] = WINED3D_CONST_NUM_UNUSED;
3764         if (reg_maps->integer_constants & (1 << i) && priv_ctx.target_version >= NV2)
3765         {
3766             const DWORD *control_values = find_loop_control_values(shader, i);
3767
3768             if(control_values)
3769             {
3770                 shader_addline(buffer, "PARAM I%u = {%u, %u, %u, -1};\n", i,
3771                                 control_values[0], control_values[1], control_values[2]);
3772             }
3773             else
3774             {
3775                 compiled->int_consts[i] = next_local;
3776                 compiled->num_int_consts++;
3777                 shader_addline(buffer, "PARAM I%u = program.local[%u];\n", i, next_local++);
3778             }
3779         }
3780     }
3781
3782     if(reg_maps->vpos || reg_maps->usesdsy)
3783     {
3784         compiled->ycorrection = next_local;
3785         shader_addline(buffer, "PARAM ycorrection = program.local[%u];\n", next_local++);
3786
3787         if(reg_maps->vpos)
3788         {
3789             shader_addline(buffer, "TEMP vpos;\n");
3790             /* ycorrection.x: Backbuffer height(onscreen) or 0(offscreen).
3791              * ycorrection.y: -1.0(onscreen), 1.0(offscreen)
3792              * ycorrection.z: 1.0
3793              * ycorrection.w: 0.0
3794              */
3795             shader_addline(buffer, "MAD vpos, fragment.position, ycorrection.zyww, ycorrection.wxww;\n");
3796             shader_addline(buffer, "FLR vpos.xy, vpos;\n");
3797         }
3798     }
3799     else
3800     {
3801         compiled->ycorrection = WINED3D_CONST_NUM_UNUSED;
3802     }
3803
3804     /* Load constants to fixup NP2 texcoords if there are still free constants left:
3805      * Constants (texture dimensions) for the NP2 fixup are loaded as local program parameters. This will consume
3806      * at most 8 (MAX_FRAGMENT_SAMPLERS / 2) parameters, which is highly unlikely, since the application had to
3807      * use 16 NP2 textures at the same time. In case that we run out of constants the fixup is simply not
3808      * applied / activated. This will probably result in wrong rendering of the texture, but will save us from
3809      * shader compilation errors and the subsequent errors when drawing with this shader. */
3810     if (priv_ctx.cur_ps_args->super.np2_fixup) {
3811         unsigned char cur_fixup_sampler = 0;
3812
3813         struct arb_ps_np2fixup_info* const fixup = priv_ctx.cur_np2fixup_info;
3814         const WORD map = priv_ctx.cur_ps_args->super.np2_fixup;
3815         const UINT max_lconsts = gl_info->limits.arb_ps_local_constants;
3816
3817         fixup->offset = next_local;
3818         fixup->super.active = 0;
3819
3820         for (i = 0; i < MAX_FRAGMENT_SAMPLERS; ++i) {
3821             if (!(map & (1 << i))) continue;
3822
3823             if (fixup->offset + (cur_fixup_sampler >> 1) < max_lconsts) {
3824                 fixup->super.active |= (1 << i);
3825                 fixup->super.idx[i] = cur_fixup_sampler++;
3826             } else {
3827                 FIXME("No free constant found to load NP2 fixup data into shader. "
3828                       "Sampling from this texture will probably look wrong.\n");
3829                 break;
3830             }
3831         }
3832
3833         fixup->super.num_consts = (cur_fixup_sampler + 1) >> 1;
3834         if (fixup->super.num_consts) {
3835             shader_addline(buffer, "PARAM np2fixup[%u] = { program.env[%u..%u] };\n",
3836                            fixup->super.num_consts, fixup->offset, fixup->super.num_consts + fixup->offset - 1);
3837         }
3838     }
3839
3840     if (shader_priv->clipplane_emulation != ~0U && args->clip)
3841     {
3842         shader_addline(buffer, "KIL fragment.texcoord[%u];\n", shader_priv->clipplane_emulation);
3843     }
3844
3845     /* Base Shader Body */
3846     shader_generate_main(shader, buffer, reg_maps, function, &priv_ctx);
3847
3848     if(args->super.srgb_correction) {
3849         arbfp_add_sRGB_correction(buffer, fragcolor, srgbtmp[0], srgbtmp[1], srgbtmp[2], srgbtmp[3],
3850                                   priv_ctx.target_version >= NV2);
3851     }
3852
3853     if(strcmp(fragcolor, "result.color")) {
3854         shader_addline(buffer, "MOV result.color, %s;\n", fragcolor);
3855     }
3856     shader_addline(buffer, "END\n");
3857
3858     /* TODO: change to resource.glObjectHandle or something like that */
3859     GL_EXTCALL(glGenProgramsARB(1, &retval));
3860
3861     TRACE("Creating a hw pixel shader, prg=%d\n", retval);
3862     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, retval));
3863
3864     TRACE("Created hw pixel shader, prg=%d\n", retval);
3865     /* Create the program and check for errors */
3866     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
3867                buffer->content_size, buffer->buffer));
3868     checkGLcall("glProgramStringARB()");
3869
3870     gl_info->gl_ops.gl.p_glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &errPos);
3871     if (errPos != -1)
3872     {
3873         FIXME("HW PixelShader Error at position %d: %s\n\n",
3874               errPos, debugstr_a((const char *)gl_info->gl_ops.gl.p_glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
3875         shader_arb_dump_program_source(buffer->buffer);
3876         retval = 0;
3877     }
3878     else
3879     {
3880         GLint native;
3881
3882         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
3883         checkGLcall("glGetProgramivARB()");
3884         if (!native) WARN("Program exceeds native resource limits.\n");
3885     }
3886
3887     return retval;
3888 }
3889
3890 static int compare_sig(const struct wined3d_shader_signature_element *sig1, const struct wined3d_shader_signature_element *sig2)
3891 {
3892     unsigned int i;
3893     int ret;
3894
3895     for(i = 0; i < MAX_REG_INPUT; i++)
3896     {
3897         if (!sig1[i].semantic_name || !sig2[i].semantic_name)
3898         {
3899             /* Compare pointers, not contents. One string is NULL(element does not exist), the other one is not NULL */
3900             if(sig1[i].semantic_name != sig2[i].semantic_name) return sig1[i].semantic_name < sig2[i].semantic_name ? -1 : 1;
3901             continue;
3902         }
3903
3904         if ((ret = strcmp(sig1[i].semantic_name, sig2[i].semantic_name))) return ret;
3905         if(sig1[i].semantic_idx    != sig2[i].semantic_idx)    return sig1[i].semantic_idx    < sig2[i].semantic_idx    ? -1 : 1;
3906         if(sig1[i].sysval_semantic != sig2[i].sysval_semantic) return sig1[i].sysval_semantic < sig2[i].sysval_semantic ? -1 : 1;
3907         if(sig1[i].component_type  != sig2[i].component_type)  return sig1[i].component_type  < sig2[i].component_type  ? -1 : 1;
3908         if(sig1[i].register_idx    != sig2[i].register_idx)    return sig1[i].register_idx    < sig2[i].register_idx    ? -1 : 1;
3909         if(sig1[i].mask            != sig2[i].mask)            return sig1[i].mask            < sig2[i].mask            ? -1 : 1;
3910     }
3911     return 0;
3912 }
3913
3914 static struct wined3d_shader_signature_element *clone_sig(const struct wined3d_shader_signature_element *sig)
3915 {
3916     struct wined3d_shader_signature_element *new;
3917     int i;
3918     char *name;
3919
3920     new = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*new) * MAX_REG_INPUT);
3921     for(i = 0; i < MAX_REG_INPUT; i++)
3922     {
3923         if (!sig[i].semantic_name) continue;
3924
3925         new[i] = sig[i];
3926         /* Clone the semantic string */
3927         name = HeapAlloc(GetProcessHeap(), 0, strlen(sig[i].semantic_name) + 1);
3928         strcpy(name, sig[i].semantic_name);
3929         new[i].semantic_name = name;
3930     }
3931     return new;
3932 }
3933
3934 static DWORD find_input_signature(struct shader_arb_priv *priv, const struct wined3d_shader_signature_element *sig)
3935 {
3936     struct wine_rb_entry *entry = wine_rb_get(&priv->signature_tree, sig);
3937     struct ps_signature *found_sig;
3938
3939     if (entry)
3940     {
3941         found_sig = WINE_RB_ENTRY_VALUE(entry, struct ps_signature, entry);
3942         TRACE("Found existing signature %u\n", found_sig->idx);
3943         return found_sig->idx;
3944     }
3945     found_sig = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*sig));
3946     found_sig->sig = clone_sig(sig);
3947     found_sig->idx = priv->ps_sig_number++;
3948     TRACE("New signature stored and assigned number %u\n", found_sig->idx);
3949     if(wine_rb_put(&priv->signature_tree, sig, &found_sig->entry) == -1)
3950     {
3951         ERR("Failed to insert program entry.\n");
3952     }
3953     return found_sig->idx;
3954 }
3955
3956 static void init_output_registers(const struct wined3d_shader *shader, DWORD sig_num,
3957         struct shader_arb_ctx_priv *priv_ctx, struct arb_vs_compiled_shader *compiled)
3958 {
3959     unsigned int i, j;
3960     static const char * const texcoords[8] =
3961     {
3962         "result.texcoord[0]", "result.texcoord[1]", "result.texcoord[2]", "result.texcoord[3]",
3963         "result.texcoord[4]", "result.texcoord[5]", "result.texcoord[6]", "result.texcoord[7]"
3964     };
3965     struct wined3d_device *device = shader->device;
3966     const struct wined3d_shader_signature_element *sig;
3967     const char *semantic_name;
3968     DWORD semantic_idx, reg_idx;
3969
3970     /* Write generic input varyings 0 to 7 to result.texcoord[], varying 8 to result.color.primary
3971      * and varying 9 to result.color.secondary
3972      */
3973     static const char * const decl_idx_to_string[MAX_REG_INPUT] =
3974     {
3975         "result.texcoord[0]", "result.texcoord[1]", "result.texcoord[2]", "result.texcoord[3]",
3976         "result.texcoord[4]", "result.texcoord[5]", "result.texcoord[6]", "result.texcoord[7]",
3977         "result.color.primary", "result.color.secondary"
3978     };
3979
3980     if(sig_num == ~0)
3981     {
3982         TRACE("Pixel shader uses builtin varyings\n");
3983         /* Map builtins to builtins */
3984         for(i = 0; i < 8; i++)
3985         {
3986             priv_ctx->texcrd_output[i] = texcoords[i];
3987         }
3988         priv_ctx->color_output[0] = "result.color.primary";
3989         priv_ctx->color_output[1] = "result.color.secondary";
3990         priv_ctx->fog_output = "TMP_FOGCOORD";
3991
3992         /* Map declared regs to builtins. Use "TA" to /dev/null unread output */
3993         for (i = 0; i < (sizeof(shader->output_signature) / sizeof(*shader->output_signature)); ++i)
3994         {
3995             semantic_name = shader->output_signature[i].semantic_name;
3996             if (!semantic_name) continue;
3997
3998             if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_POSITION))
3999             {
4000                 TRACE("o%u is TMP_OUT\n", i);
4001                 if (!shader->output_signature[i].semantic_idx) priv_ctx->vs_output[i] = "TMP_OUT";
4002                 else priv_ctx->vs_output[i] = "TA";
4003             }
4004             else if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_PSIZE))
4005             {
4006                 TRACE("o%u is result.pointsize\n", i);
4007                 if (!shader->output_signature[i].semantic_idx) priv_ctx->vs_output[i] = "result.pointsize";
4008                 else priv_ctx->vs_output[i] = "TA";
4009             }
4010             else if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_COLOR))
4011             {
4012                 TRACE("o%u is result.color.?, idx %u\n", i, shader->output_signature[i].semantic_idx);
4013                 if (!shader->output_signature[i].semantic_idx)
4014                     priv_ctx->vs_output[i] = "result.color.primary";
4015                 else if (shader->output_signature[i].semantic_idx == 1)
4016                     priv_ctx->vs_output[i] = "result.color.secondary";
4017                 else priv_ctx->vs_output[i] = "TA";
4018             }
4019             else if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_TEXCOORD))
4020             {
4021                 TRACE("o%u is %s\n", i, texcoords[shader->output_signature[i].semantic_idx]);
4022                 if (shader->output_signature[i].semantic_idx >= 8) priv_ctx->vs_output[i] = "TA";
4023                 else priv_ctx->vs_output[i] = texcoords[shader->output_signature[i].semantic_idx];
4024             }
4025             else if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_FOG))
4026             {
4027                 TRACE("o%u is result.fogcoord\n", i);
4028                 if (shader->output_signature[i].semantic_idx > 0) priv_ctx->vs_output[i] = "TA";
4029                 else priv_ctx->vs_output[i] = "result.fogcoord";
4030             }
4031             else
4032             {
4033                 priv_ctx->vs_output[i] = "TA";
4034             }
4035         }
4036         return;
4037     }
4038
4039     /* Instead of searching for the signature in the signature list, read the one from the current pixel shader.
4040      * Its maybe not the shader where the signature came from, but it is the same signature and faster to find
4041      */
4042     sig = device->stateBlock->state.pixel_shader->input_signature;
4043     TRACE("Pixel shader uses declared varyings\n");
4044
4045     /* Map builtin to declared. /dev/null the results by default to the TA temp reg */
4046     for(i = 0; i < 8; i++)
4047     {
4048         priv_ctx->texcrd_output[i] = "TA";
4049     }
4050     priv_ctx->color_output[0] = "TA";
4051     priv_ctx->color_output[1] = "TA";
4052     priv_ctx->fog_output = "TA";
4053
4054     for(i = 0; i < MAX_REG_INPUT; i++)
4055     {
4056         semantic_name = sig[i].semantic_name;
4057         semantic_idx = sig[i].semantic_idx;
4058         reg_idx = sig[i].register_idx;
4059         if (!semantic_name) continue;
4060
4061         /* If a declared input register is not written by builtin arguments, don't write to it.
4062          * GL_NV_vertex_program makes sure the input defaults to 0.0, which is correct with D3D
4063          *
4064          * Don't care about POSITION and PSIZE here - this is a builtin vertex shader, position goes
4065          * to TMP_OUT in any case
4066          */
4067         if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_TEXCOORD))
4068         {
4069             if (semantic_idx < 8)
4070                 priv_ctx->texcrd_output[semantic_idx] = decl_idx_to_string[reg_idx];
4071         }
4072         else if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_COLOR))
4073         {
4074             if (semantic_idx < 2)
4075                 priv_ctx->color_output[semantic_idx] = decl_idx_to_string[reg_idx];
4076         }
4077         else if(shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_FOG))
4078         {
4079             if (!semantic_idx)
4080                 priv_ctx->fog_output = decl_idx_to_string[reg_idx];
4081         }
4082         else
4083         {
4084             continue;
4085         }
4086
4087         if (!strcmp(decl_idx_to_string[reg_idx], "result.color.primary")
4088                 || !strcmp(decl_idx_to_string[reg_idx], "result.color.secondary"))
4089         {
4090             compiled->need_color_unclamp = TRUE;
4091         }
4092     }
4093
4094     /* Map declared to declared */
4095     for (i = 0; i < (sizeof(shader->output_signature) / sizeof(*shader->output_signature)); ++i)
4096     {
4097         /* Write unread output to TA to throw them away */
4098         priv_ctx->vs_output[i] = "TA";
4099         semantic_name = shader->output_signature[i].semantic_name;
4100         if (!semantic_name) continue;
4101
4102         if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_POSITION)
4103                 && !shader->output_signature[i].semantic_idx)
4104         {
4105             priv_ctx->vs_output[i] = "TMP_OUT";
4106             continue;
4107         }
4108         else if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_PSIZE)
4109                 && !shader->output_signature[i].semantic_idx)
4110         {
4111             priv_ctx->vs_output[i] = "result.pointsize";
4112             continue;
4113         }
4114
4115         for(j = 0; j < MAX_REG_INPUT; j++)
4116         {
4117             if (!sig[j].semantic_name) continue;
4118
4119             if (!strcmp(sig[j].semantic_name, semantic_name)
4120                     && sig[j].semantic_idx == shader->output_signature[i].semantic_idx)
4121             {
4122                 priv_ctx->vs_output[i] = decl_idx_to_string[sig[j].register_idx];
4123
4124                 if (!strcmp(priv_ctx->vs_output[i], "result.color.primary")
4125                         || !strcmp(priv_ctx->vs_output[i], "result.color.secondary"))
4126                 {
4127                     compiled->need_color_unclamp = TRUE;
4128                 }
4129             }
4130         }
4131     }
4132 }
4133
4134 /* Context activation is done by the caller. */
4135 static GLuint shader_arb_generate_vshader(const struct wined3d_shader *shader,
4136         const struct wined3d_gl_info *gl_info, struct wined3d_shader_buffer *buffer,
4137         const struct arb_vs_compile_args *args, struct arb_vs_compiled_shader *compiled)
4138 {
4139     const struct arb_vshader_private *shader_data = shader->backend_data;
4140     const struct wined3d_shader_reg_maps *reg_maps = &shader->reg_maps;
4141     struct shader_arb_priv *priv = shader->device->shader_priv;
4142     const DWORD *function = shader->function;
4143     GLuint ret;
4144     DWORD next_local = 0;
4145     struct shader_arb_ctx_priv priv_ctx;
4146     unsigned int i;
4147     GLint errPos;
4148
4149     memset(&priv_ctx, 0, sizeof(priv_ctx));
4150     priv_ctx.cur_vs_args = args;
4151     list_init(&priv_ctx.control_frames);
4152     init_output_registers(shader, args->ps_signature, &priv_ctx, compiled);
4153
4154     /*  Create the hw ARB shader */
4155     shader_addline(buffer, "!!ARBvp1.0\n");
4156
4157     /* Always enable the NV extension if available. Unlike fragment shaders, there is no
4158      * mesurable performance penalty, and we can always make use of it for clipplanes.
4159      */
4160     if (gl_info->supported[NV_VERTEX_PROGRAM3])
4161     {
4162         shader_addline(buffer, "OPTION NV_vertex_program3;\n");
4163         priv_ctx.target_version = NV3;
4164         shader_addline(buffer, "ADDRESS aL;\n");
4165     }
4166     else if (gl_info->supported[NV_VERTEX_PROGRAM2_OPTION])
4167     {
4168         shader_addline(buffer, "OPTION NV_vertex_program2;\n");
4169         priv_ctx.target_version = NV2;
4170         shader_addline(buffer, "ADDRESS aL;\n");
4171     } else {
4172         priv_ctx.target_version = ARB;
4173     }
4174
4175     shader_addline(buffer, "TEMP TMP_OUT;\n");
4176     if (reg_maps->fog)
4177         shader_addline(buffer, "TEMP TMP_FOGCOORD;\n");
4178     if (need_helper_const(shader_data, reg_maps, gl_info))
4179     {
4180         shader_addline(buffer, "PARAM helper_const = { 0.0, 1.0, 2.0, %1.10f};\n", eps);
4181     }
4182     if (need_rel_addr_const(shader_data, reg_maps, gl_info))
4183     {
4184         shader_addline(buffer, "PARAM rel_addr_const = { 0.5, %d.0, 0.0, 0.0 };\n", shader_data->rel_offset);
4185         shader_addline(buffer, "TEMP A0_SHADOW;\n");
4186     }
4187
4188     shader_addline(buffer, "TEMP TA;\n");
4189     shader_addline(buffer, "TEMP TB;\n");
4190
4191     /* Base Declarations */
4192     shader_generate_arb_declarations(shader, reg_maps, buffer, gl_info,
4193             &priv_ctx.vs_clipplanes, &priv_ctx);
4194
4195     for(i = 0; i < MAX_CONST_I; i++)
4196     {
4197         compiled->int_consts[i] = WINED3D_CONST_NUM_UNUSED;
4198         if(reg_maps->integer_constants & (1 << i) && priv_ctx.target_version >= NV2)
4199         {
4200             const DWORD *control_values = find_loop_control_values(shader, i);
4201
4202             if(control_values)
4203             {
4204                 shader_addline(buffer, "PARAM I%u = {%u, %u, %u, -1};\n", i,
4205                                 control_values[0], control_values[1], control_values[2]);
4206             }
4207             else
4208             {
4209                 compiled->int_consts[i] = next_local;
4210                 compiled->num_int_consts++;
4211                 shader_addline(buffer, "PARAM I%u = program.local[%u];\n", i, next_local++);
4212             }
4213         }
4214     }
4215
4216     /* We need a constant to fixup the final position */
4217     shader_addline(buffer, "PARAM posFixup = program.local[%u];\n", next_local);
4218     compiled->pos_fixup = next_local++;
4219
4220     /* Initialize output parameters. GL_ARB_vertex_program does not require special initialization values
4221      * for output parameters. D3D in theory does not do that either, but some applications depend on a
4222      * proper initialization of the secondary color, and programs using the fixed function pipeline without
4223      * a replacement shader depend on the texcoord.w being set properly.
4224      *
4225      * GL_NV_vertex_program defines that all output values are initialized to {0.0, 0.0, 0.0, 1.0}. This
4226      * assertion is in effect even when using GL_ARB_vertex_program without any NV specific additions. So
4227      * skip this if NV_vertex_program is supported. Otherwise, initialize the secondary color. For the tex-
4228      * coords, we have a flag in the opengl caps. Many cards do not require the texcoord being set, and
4229      * this can eat a number of instructions, so skip it unless this cap is set as well
4230      */
4231     if (!gl_info->supported[NV_VERTEX_PROGRAM])
4232     {
4233         const char *color_init = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_0001);
4234         shader_addline(buffer, "MOV result.color.secondary, %s;\n", color_init);
4235
4236         if (gl_info->quirks & WINED3D_QUIRK_SET_TEXCOORD_W && !priv->ffp_proj_control)
4237         {
4238             int i;
4239             const char *one = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_ONE);
4240             for(i = 0; i < min(8, MAX_REG_TEXCRD); i++)
4241             {
4242                 if (reg_maps->texcoord_mask[i] && reg_maps->texcoord_mask[i] != WINED3DSP_WRITEMASK_ALL)
4243                     shader_addline(buffer, "MOV result.texcoord[%u].w, %s\n", i, one);
4244             }
4245         }
4246     }
4247
4248     /* The shader starts with the main function */
4249     priv_ctx.in_main_func = TRUE;
4250     /* Base Shader Body */
4251     shader_generate_main(shader, buffer, reg_maps, function, &priv_ctx);
4252
4253     if (!priv_ctx.footer_written) vshader_add_footer(&priv_ctx,
4254             shader_data, args, reg_maps, gl_info, buffer);
4255
4256     shader_addline(buffer, "END\n");
4257
4258     /* TODO: change to resource.glObjectHandle or something like that */
4259     GL_EXTCALL(glGenProgramsARB(1, &ret));
4260
4261     TRACE("Creating a hw vertex shader, prg=%d\n", ret);
4262     GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, ret));
4263
4264     TRACE("Created hw vertex shader, prg=%d\n", ret);
4265     /* Create the program and check for errors */
4266     GL_EXTCALL(glProgramStringARB(GL_VERTEX_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
4267                buffer->content_size, buffer->buffer));
4268     checkGLcall("glProgramStringARB()");
4269
4270     gl_info->gl_ops.gl.p_glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &errPos);
4271     if (errPos != -1)
4272     {
4273         FIXME("HW VertexShader Error at position %d: %s\n\n",
4274               errPos, debugstr_a((const char *)gl_info->gl_ops.gl.p_glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
4275         shader_arb_dump_program_source(buffer->buffer);
4276         ret = -1;
4277     }
4278     else
4279     {
4280         GLint native;
4281
4282         GL_EXTCALL(glGetProgramivARB(GL_VERTEX_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
4283         checkGLcall("glGetProgramivARB()");
4284         if (!native) WARN("Program exceeds native resource limits.\n");
4285     }
4286
4287     return ret;
4288 }
4289
4290 /* Context activation is done by the caller. */
4291 static struct arb_ps_compiled_shader *find_arb_pshader(struct wined3d_shader *shader,
4292         const struct arb_ps_compile_args *args)
4293 {
4294     struct wined3d_device *device = shader->device;
4295     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4296     UINT i;
4297     DWORD new_size;
4298     struct arb_ps_compiled_shader *new_array;
4299     struct wined3d_shader_buffer buffer;
4300     struct arb_pshader_private *shader_data;
4301     GLuint ret;
4302
4303     if (!shader->backend_data)
4304     {
4305         struct shader_arb_priv *priv = device->shader_priv;
4306
4307         shader->backend_data = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data));
4308         shader_data = shader->backend_data;
4309         shader_data->clamp_consts = shader->reg_maps.shader_version.major == 1;
4310
4311         if (shader->reg_maps.shader_version.major < 3)
4312             shader_data->input_signature_idx = ~0;
4313         else
4314             shader_data->input_signature_idx = find_input_signature(priv, shader->input_signature);
4315
4316         TRACE("Shader got assigned input signature index %u\n", shader_data->input_signature_idx);
4317
4318         if (!device->vs_clipping)
4319             shader_data->clipplane_emulation = shader_find_free_input_register(&shader->reg_maps,
4320                     gl_info->limits.texture_stages - 1);
4321         else
4322             shader_data->clipplane_emulation = ~0U;
4323     }
4324     shader_data = shader->backend_data;
4325
4326     /* Usually we have very few GL shaders for each d3d shader(just 1 or maybe 2),
4327      * so a linear search is more performant than a hashmap or a binary search
4328      * (cache coherency etc)
4329      */
4330     for (i = 0; i < shader_data->num_gl_shaders; ++i)
4331     {
4332         if (!memcmp(&shader_data->gl_shaders[i].args, args, sizeof(*args)))
4333             return &shader_data->gl_shaders[i];
4334     }
4335
4336     TRACE("No matching GL shader found, compiling a new shader\n");
4337     if(shader_data->shader_array_size == shader_data->num_gl_shaders) {
4338         if (shader_data->num_gl_shaders)
4339         {
4340             new_size = shader_data->shader_array_size + max(1, shader_data->shader_array_size / 2);
4341             new_array = HeapReAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, shader_data->gl_shaders,
4342                                     new_size * sizeof(*shader_data->gl_shaders));
4343         } else {
4344             new_array = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data->gl_shaders));
4345             new_size = 1;
4346         }
4347
4348         if(!new_array) {
4349             ERR("Out of memory\n");
4350             return 0;
4351         }
4352         shader_data->gl_shaders = new_array;
4353         shader_data->shader_array_size = new_size;
4354     }
4355
4356     shader_data->gl_shaders[shader_data->num_gl_shaders].args = *args;
4357
4358     pixelshader_update_samplers(shader, args->super.tex_types);
4359
4360     if (!shader_buffer_init(&buffer))
4361     {
4362         ERR("Failed to initialize shader buffer.\n");
4363         return 0;
4364     }
4365
4366     ret = shader_arb_generate_pshader(shader, gl_info, &buffer, args,
4367             &shader_data->gl_shaders[shader_data->num_gl_shaders]);
4368     shader_buffer_free(&buffer);
4369     shader_data->gl_shaders[shader_data->num_gl_shaders].prgId = ret;
4370
4371     return &shader_data->gl_shaders[shader_data->num_gl_shaders++];
4372 }
4373
4374 static inline BOOL vs_args_equal(const struct arb_vs_compile_args *stored, const struct arb_vs_compile_args *new,
4375                                  const DWORD use_map, BOOL skip_int) {
4376     if((stored->super.swizzle_map & use_map) != new->super.swizzle_map) return FALSE;
4377     if(stored->super.clip_enabled != new->super.clip_enabled) return FALSE;
4378     if(stored->super.fog_src != new->super.fog_src) return FALSE;
4379     if(stored->clip.boolclip_compare != new->clip.boolclip_compare) return FALSE;
4380     if(stored->ps_signature != new->ps_signature) return FALSE;
4381     if(stored->vertex.samplers_compare != new->vertex.samplers_compare) return FALSE;
4382     if(skip_int) return TRUE;
4383
4384     return !memcmp(stored->loop_ctrl, new->loop_ctrl, sizeof(stored->loop_ctrl));
4385 }
4386
4387 static struct arb_vs_compiled_shader *find_arb_vshader(struct wined3d_shader *shader,
4388         const struct arb_vs_compile_args *args)
4389 {
4390     struct wined3d_device *device = shader->device;
4391     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4392     DWORD use_map = device->stream_info.use_map;
4393     UINT i;
4394     DWORD new_size;
4395     struct arb_vs_compiled_shader *new_array;
4396     struct wined3d_shader_buffer buffer;
4397     struct arb_vshader_private *shader_data;
4398     GLuint ret;
4399
4400     if (!shader->backend_data)
4401     {
4402         const struct wined3d_shader_reg_maps *reg_maps = &shader->reg_maps;
4403
4404         shader->backend_data = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data));
4405         shader_data = shader->backend_data;
4406
4407         if ((gl_info->quirks & WINED3D_QUIRK_ARB_VS_OFFSET_LIMIT)
4408                 && reg_maps->min_rel_offset <= reg_maps->max_rel_offset)
4409         {
4410             if (reg_maps->max_rel_offset - reg_maps->min_rel_offset > 127)
4411             {
4412                 FIXME("The difference between the minimum and maximum relative offset is > 127.\n");
4413                 FIXME("Which this OpenGL implementation does not support. Try using GLSL.\n");
4414                 FIXME("Min: %u, Max: %u.\n", reg_maps->min_rel_offset, reg_maps->max_rel_offset);
4415             }
4416             else if (reg_maps->max_rel_offset - reg_maps->min_rel_offset > 63)
4417                 shader_data->rel_offset = reg_maps->min_rel_offset + 63;
4418             else if (reg_maps->max_rel_offset > 63)
4419                 shader_data->rel_offset = reg_maps->min_rel_offset;
4420         }
4421     }
4422     shader_data = shader->backend_data;
4423
4424     /* Usually we have very few GL shaders for each d3d shader(just 1 or maybe 2),
4425      * so a linear search is more performant than a hashmap or a binary search
4426      * (cache coherency etc)
4427      */
4428     for(i = 0; i < shader_data->num_gl_shaders; i++) {
4429         if (vs_args_equal(&shader_data->gl_shaders[i].args, args,
4430                 use_map, gl_info->supported[NV_VERTEX_PROGRAM2_OPTION]))
4431         {
4432             return &shader_data->gl_shaders[i];
4433         }
4434     }
4435
4436     TRACE("No matching GL shader found, compiling a new shader\n");
4437
4438     if(shader_data->shader_array_size == shader_data->num_gl_shaders) {
4439         if (shader_data->num_gl_shaders)
4440         {
4441             new_size = shader_data->shader_array_size + max(1, shader_data->shader_array_size / 2);
4442             new_array = HeapReAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, shader_data->gl_shaders,
4443                                     new_size * sizeof(*shader_data->gl_shaders));
4444         } else {
4445             new_array = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data->gl_shaders));
4446             new_size = 1;
4447         }
4448
4449         if(!new_array) {
4450             ERR("Out of memory\n");
4451             return 0;
4452         }
4453         shader_data->gl_shaders = new_array;
4454         shader_data->shader_array_size = new_size;
4455     }
4456
4457     shader_data->gl_shaders[shader_data->num_gl_shaders].args = *args;
4458
4459     if (!shader_buffer_init(&buffer))
4460     {
4461         ERR("Failed to initialize shader buffer.\n");
4462         return 0;
4463     }
4464
4465     ret = shader_arb_generate_vshader(shader, gl_info, &buffer, args,
4466             &shader_data->gl_shaders[shader_data->num_gl_shaders]);
4467     shader_buffer_free(&buffer);
4468     shader_data->gl_shaders[shader_data->num_gl_shaders].prgId = ret;
4469
4470     return &shader_data->gl_shaders[shader_data->num_gl_shaders++];
4471 }
4472
4473 static void find_arb_ps_compile_args(const struct wined3d_state *state,
4474         const struct wined3d_shader *shader, struct arb_ps_compile_args *args)
4475 {
4476     struct wined3d_device *device = shader->device;
4477     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4478     int i;
4479     WORD int_skip;
4480
4481     find_ps_compile_args(state, shader, &args->super);
4482
4483     /* This forces all local boolean constants to 1 to make them stateblock independent */
4484     args->bools = shader->reg_maps.local_bool_consts;
4485
4486     for(i = 0; i < MAX_CONST_B; i++)
4487     {
4488         if (state->ps_consts_b[i])
4489             args->bools |= ( 1 << i);
4490     }
4491
4492     /* Only enable the clip plane emulation KIL if at least one clipplane is enabled. The KIL instruction
4493      * is quite expensive because it forces the driver to disable early Z discards. It is cheaper to
4494      * duplicate the shader than have a no-op KIL instruction in every shader
4495      */
4496     if (!device->vs_clipping && use_vs(state)
4497             && state->render_states[WINED3D_RS_CLIPPING]
4498             && state->render_states[WINED3D_RS_CLIPPLANEENABLE])
4499         args->clip = 1;
4500     else
4501         args->clip = 0;
4502
4503     /* Skip if unused or local, or supported natively */
4504     int_skip = ~shader->reg_maps.integer_constants | shader->reg_maps.local_int_consts;
4505     if (int_skip == 0xffff || gl_info->supported[NV_FRAGMENT_PROGRAM_OPTION])
4506     {
4507         memset(args->loop_ctrl, 0, sizeof(args->loop_ctrl));
4508         return;
4509     }
4510
4511     for(i = 0; i < MAX_CONST_I; i++)
4512     {
4513         if(int_skip & (1 << i))
4514         {
4515             args->loop_ctrl[i][0] = 0;
4516             args->loop_ctrl[i][1] = 0;
4517             args->loop_ctrl[i][2] = 0;
4518         }
4519         else
4520         {
4521             args->loop_ctrl[i][0] = state->ps_consts_i[i * 4];
4522             args->loop_ctrl[i][1] = state->ps_consts_i[i * 4 + 1];
4523             args->loop_ctrl[i][2] = state->ps_consts_i[i * 4 + 2];
4524         }
4525     }
4526 }
4527
4528 static void find_arb_vs_compile_args(const struct wined3d_state *state,
4529         const struct wined3d_shader *shader, struct arb_vs_compile_args *args)
4530 {
4531     struct wined3d_device *device = shader->device;
4532     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4533     int i;
4534     WORD int_skip;
4535
4536     find_vs_compile_args(state, shader, &args->super);
4537
4538     args->clip.boolclip_compare = 0;
4539     if (use_ps(state))
4540     {
4541         const struct wined3d_shader *ps = state->pixel_shader;
4542         const struct arb_pshader_private *shader_priv = ps->backend_data;
4543         args->ps_signature = shader_priv->input_signature_idx;
4544
4545         args->clip.boolclip.clip_texcoord = shader_priv->clipplane_emulation + 1;
4546     }
4547     else
4548     {
4549         args->ps_signature = ~0;
4550         if (!device->vs_clipping && device->adapter->fragment_pipe == &arbfp_fragment_pipeline)
4551         {
4552             args->clip.boolclip.clip_texcoord = ffp_clip_emul(state) ? gl_info->limits.texture_stages : 0;
4553         }
4554         /* Otherwise: Setting boolclip_compare set clip_texcoord to 0 */
4555     }
4556
4557     if (args->clip.boolclip.clip_texcoord)
4558     {
4559         if (state->render_states[WINED3D_RS_CLIPPING])
4560             args->clip.boolclip.clipplane_mask = (unsigned char)state->render_states[WINED3D_RS_CLIPPLANEENABLE];
4561         /* clipplane_mask was set to 0 by setting boolclip_compare to 0 */
4562     }
4563
4564     /* This forces all local boolean constants to 1 to make them stateblock independent */
4565     args->clip.boolclip.bools = shader->reg_maps.local_bool_consts;
4566     /* TODO: Figure out if it would be better to store bool constants as bitmasks in the stateblock */
4567     for(i = 0; i < MAX_CONST_B; i++)
4568     {
4569         if (state->vs_consts_b[i])
4570             args->clip.boolclip.bools |= ( 1 << i);
4571     }
4572
4573     args->vertex.samplers[0] = device->texUnitMap[MAX_FRAGMENT_SAMPLERS + 0];
4574     args->vertex.samplers[1] = device->texUnitMap[MAX_FRAGMENT_SAMPLERS + 1];
4575     args->vertex.samplers[2] = device->texUnitMap[MAX_FRAGMENT_SAMPLERS + 2];
4576     args->vertex.samplers[3] = 0;
4577
4578     /* Skip if unused or local */
4579     int_skip = ~shader->reg_maps.integer_constants | shader->reg_maps.local_int_consts;
4580     /* This is about flow control, not clipping. */
4581     if (int_skip == 0xffff || gl_info->supported[NV_VERTEX_PROGRAM2_OPTION])
4582     {
4583         memset(args->loop_ctrl, 0, sizeof(args->loop_ctrl));
4584         return;
4585     }
4586
4587     for(i = 0; i < MAX_CONST_I; i++)
4588     {
4589         if(int_skip & (1 << i))
4590         {
4591             args->loop_ctrl[i][0] = 0;
4592             args->loop_ctrl[i][1] = 0;
4593             args->loop_ctrl[i][2] = 0;
4594         }
4595         else
4596         {
4597             args->loop_ctrl[i][0] = state->vs_consts_i[i * 4];
4598             args->loop_ctrl[i][1] = state->vs_consts_i[i * 4 + 1];
4599             args->loop_ctrl[i][2] = state->vs_consts_i[i * 4 + 2];
4600         }
4601     }
4602 }
4603
4604 /* Context activation is done by the caller. */
4605 static void shader_arb_select(const struct wined3d_context *context, enum wined3d_shader_mode vertex_mode,
4606         enum wined3d_shader_mode fragment_mode)
4607 {
4608     struct wined3d_device *device = context->swapchain->device;
4609     struct shader_arb_priv *priv = device->shader_priv;
4610     const struct wined3d_gl_info *gl_info = context->gl_info;
4611     const struct wined3d_state *state = &device->stateBlock->state;
4612     int i;
4613
4614     /* Deal with pixel shaders first so the vertex shader arg function has the input signature ready */
4615     if (fragment_mode == WINED3D_SHADER_MODE_SHADER)
4616     {
4617         struct wined3d_shader *ps = state->pixel_shader;
4618         struct arb_ps_compile_args compile_args;
4619         struct arb_ps_compiled_shader *compiled;
4620
4621         TRACE("Using pixel shader %p.\n", ps);
4622         find_arb_ps_compile_args(state, ps, &compile_args);
4623         compiled = find_arb_pshader(ps, &compile_args);
4624         priv->current_fprogram_id = compiled->prgId;
4625         priv->compiled_fprog = compiled;
4626
4627         /* Bind the fragment program */
4628         GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, priv->current_fprogram_id));
4629         checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, priv->current_fprogram_id);");
4630
4631         if (!priv->use_arbfp_fixed_func)
4632             priv->fragment_pipe->enable_extension(gl_info, FALSE);
4633
4634         /* Enable OpenGL fragment programs. */
4635         gl_info->gl_ops.gl.p_glEnable(GL_FRAGMENT_PROGRAM_ARB);
4636         checkGLcall("glEnable(GL_FRAGMENT_PROGRAM_ARB);");
4637
4638         TRACE("(%p) : Bound fragment program %u and enabled GL_FRAGMENT_PROGRAM_ARB\n",
4639                 device, priv->current_fprogram_id);
4640
4641         /* Pixel Shader 1.x constants are clamped to [-1;1], Pixel Shader 2.0 constants are not. If switching between
4642          * a 1.x and newer shader, reload the first 8 constants
4643          */
4644         if (priv->last_ps_const_clamped != ((struct arb_pshader_private *)ps->backend_data)->clamp_consts)
4645         {
4646             priv->last_ps_const_clamped = ((struct arb_pshader_private *)ps->backend_data)->clamp_consts;
4647             priv->highest_dirty_ps_const = max(priv->highest_dirty_ps_const, 8);
4648             for(i = 0; i < 8; i++)
4649             {
4650                 priv->pshader_const_dirty[i] = 1;
4651             }
4652             /* Also takes care of loading local constants */
4653             shader_arb_load_constants_internal(context, TRUE, FALSE, TRUE);
4654         }
4655         else
4656         {
4657             UINT rt_height = state->fb->render_targets[0]->resource.height;
4658             shader_arb_ps_local_constants(compiled, context, state, rt_height);
4659         }
4660
4661         /* Force constant reloading for the NP2 fixup (see comment in shader_glsl_select for more info) */
4662         if (compiled->np2fixup_info.super.active)
4663             shader_arb_load_np2fixup_constants(priv, gl_info, state);
4664     }
4665     else
4666     {
4667         if (gl_info->supported[ARB_FRAGMENT_PROGRAM] && !priv->use_arbfp_fixed_func)
4668         {
4669             /* Disable only if we're not using arbfp fixed function fragment
4670              * processing. If this is used, keep GL_FRAGMENT_PROGRAM_ARB
4671              * enabled, and the fixed function pipeline will bind the fixed
4672              * function replacement shader. */
4673             gl_info->gl_ops.gl.p_glDisable(GL_FRAGMENT_PROGRAM_ARB);
4674             checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
4675             priv->current_fprogram_id = 0;
4676         }
4677         priv->fragment_pipe->enable_extension(gl_info, fragment_mode == WINED3D_SHADER_MODE_FFP);
4678     }
4679
4680     if (vertex_mode == WINED3D_SHADER_MODE_SHADER)
4681     {
4682         struct wined3d_shader *vs = state->vertex_shader;
4683         struct arb_vs_compile_args compile_args;
4684         struct arb_vs_compiled_shader *compiled;
4685
4686         TRACE("Using vertex shader %p\n", vs);
4687         find_arb_vs_compile_args(state, vs, &compile_args);
4688         compiled = find_arb_vshader(vs, &compile_args);
4689         priv->current_vprogram_id = compiled->prgId;
4690         priv->compiled_vprog = compiled;
4691
4692         /* Bind the vertex program */
4693         GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->current_vprogram_id));
4694         checkGLcall("glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->current_vprogram_id);");
4695
4696         /* Enable OpenGL vertex programs */
4697         gl_info->gl_ops.gl.p_glEnable(GL_VERTEX_PROGRAM_ARB);
4698         checkGLcall("glEnable(GL_VERTEX_PROGRAM_ARB);");
4699         TRACE("(%p) : Bound vertex program %u and enabled GL_VERTEX_PROGRAM_ARB\n", device, priv->current_vprogram_id);
4700         shader_arb_vs_local_constants(compiled, context, state);
4701
4702         if(priv->last_vs_color_unclamp != compiled->need_color_unclamp) {
4703             priv->last_vs_color_unclamp = compiled->need_color_unclamp;
4704
4705             if (gl_info->supported[ARB_COLOR_BUFFER_FLOAT])
4706             {
4707                 GL_EXTCALL(glClampColorARB(GL_CLAMP_VERTEX_COLOR_ARB, !compiled->need_color_unclamp));
4708                 checkGLcall("glClampColorARB");
4709             } else {
4710                 FIXME("vertex color clamp needs to be changed, but extension not supported.\n");
4711             }
4712         }
4713     }
4714     else if (gl_info->supported[ARB_VERTEX_PROGRAM])
4715     {
4716         priv->current_vprogram_id = 0;
4717         gl_info->gl_ops.gl.p_glDisable(GL_VERTEX_PROGRAM_ARB);
4718         checkGLcall("glDisable(GL_VERTEX_PROGRAM_ARB)");
4719     }
4720     priv->vertex_pipe->vp_enable(gl_info, vertex_mode == WINED3D_SHADER_MODE_FFP);
4721 }
4722
4723 /* Context activation is done by the caller. */
4724 static void shader_arb_select_depth_blt(void *shader_priv, const struct wined3d_gl_info *gl_info,
4725         enum tex_types tex_type, const SIZE *ds_mask_size)
4726 {
4727     const float mask[] = {0.0f, 0.0f, (float)ds_mask_size->cx, (float)ds_mask_size->cy};
4728     BOOL masked = ds_mask_size->cx && ds_mask_size->cy;
4729     struct shader_arb_priv *priv = shader_priv;
4730     GLuint *blt_fprogram;
4731
4732     if (!priv->depth_blt_vprogram_id) priv->depth_blt_vprogram_id = create_arb_blt_vertex_program(gl_info);
4733     GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->depth_blt_vprogram_id));
4734     gl_info->gl_ops.gl.p_glEnable(GL_VERTEX_PROGRAM_ARB);
4735
4736     blt_fprogram = masked ? &priv->depth_blt_fprogram_id_masked[tex_type] : &priv->depth_blt_fprogram_id_full[tex_type];
4737     if (!*blt_fprogram) *blt_fprogram = create_arb_blt_fragment_program(gl_info, tex_type, masked);
4738     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, *blt_fprogram));
4739     if (masked) GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, 0, mask));
4740     gl_info->gl_ops.gl.p_glEnable(GL_FRAGMENT_PROGRAM_ARB);
4741 }
4742
4743 /* Context activation is done by the caller. */
4744 static void shader_arb_deselect_depth_blt(void *shader_priv, const struct wined3d_gl_info *gl_info)
4745 {
4746     struct shader_arb_priv *priv = shader_priv;
4747
4748     if (priv->current_vprogram_id) {
4749         GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->current_vprogram_id));
4750         checkGLcall("glBindProgramARB(GL_VERTEX_PROGRAM_ARB, vertexShader->prgId);");
4751
4752         TRACE("Bound vertex program %u and enabled GL_VERTEX_PROGRAM_ARB.\n", priv->current_vprogram_id);
4753     }
4754     else
4755     {
4756         gl_info->gl_ops.gl.p_glDisable(GL_VERTEX_PROGRAM_ARB);
4757         checkGLcall("glDisable(GL_VERTEX_PROGRAM_ARB)");
4758     }
4759
4760     if (priv->current_fprogram_id) {
4761         GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, priv->current_fprogram_id));
4762         checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, pixelShader->prgId);");
4763
4764         TRACE("Bound fragment program %u and enabled GL_FRAGMENT_PROGRAM_ARB.\n", priv->current_fprogram_id);
4765     }
4766     else if(!priv->use_arbfp_fixed_func)
4767     {
4768         gl_info->gl_ops.gl.p_glDisable(GL_FRAGMENT_PROGRAM_ARB);
4769         checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
4770     }
4771 }
4772
4773 static void shader_arb_destroy(struct wined3d_shader *shader)
4774 {
4775     struct wined3d_device *device = shader->device;
4776     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4777
4778     if (shader_is_pshader_version(shader->reg_maps.shader_version.type))
4779     {
4780         struct arb_pshader_private *shader_data = shader->backend_data;
4781         UINT i;
4782
4783         if(!shader_data) return; /* This can happen if a shader was never compiled */
4784
4785         if (shader_data->num_gl_shaders)
4786         {
4787             struct wined3d_context *context = context_acquire(device, NULL);
4788
4789             for (i = 0; i < shader_data->num_gl_shaders; ++i)
4790             {
4791                 GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId));
4792                 checkGLcall("GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId))");
4793             }
4794
4795             context_release(context);
4796         }
4797
4798         HeapFree(GetProcessHeap(), 0, shader_data->gl_shaders);
4799         HeapFree(GetProcessHeap(), 0, shader_data);
4800         shader->backend_data = NULL;
4801     }
4802     else
4803     {
4804         struct arb_vshader_private *shader_data = shader->backend_data;
4805         UINT i;
4806
4807         if(!shader_data) return; /* This can happen if a shader was never compiled */
4808
4809         if (shader_data->num_gl_shaders)
4810         {
4811             struct wined3d_context *context = context_acquire(device, NULL);
4812
4813             for (i = 0; i < shader_data->num_gl_shaders; ++i)
4814             {
4815                 GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId));
4816                 checkGLcall("GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId))");
4817             }
4818
4819             context_release(context);
4820         }
4821
4822         HeapFree(GetProcessHeap(), 0, shader_data->gl_shaders);
4823         HeapFree(GetProcessHeap(), 0, shader_data);
4824         shader->backend_data = NULL;
4825     }
4826 }
4827
4828 static int sig_tree_compare(const void *key, const struct wine_rb_entry *entry)
4829 {
4830     struct ps_signature *e = WINE_RB_ENTRY_VALUE(entry, struct ps_signature, entry);
4831     return compare_sig(key, e->sig);
4832 }
4833
4834 static const struct wine_rb_functions sig_tree_functions =
4835 {
4836     wined3d_rb_alloc,
4837     wined3d_rb_realloc,
4838     wined3d_rb_free,
4839     sig_tree_compare
4840 };
4841
4842 static HRESULT shader_arb_alloc(struct wined3d_device *device, const struct wined3d_vertex_pipe_ops *vertex_pipe,
4843         const struct fragment_pipeline *fragment_pipe)
4844 {
4845     struct shader_arb_priv *priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*priv));
4846     struct fragment_caps fragment_caps;
4847     void *vertex_priv, *fragment_priv;
4848
4849     if (!(vertex_priv = vertex_pipe->vp_alloc(&arb_program_shader_backend, priv)))
4850     {
4851         ERR("Failed to initialize vertex pipe.\n");
4852         HeapFree(GetProcessHeap(), 0, priv);
4853         return E_FAIL;
4854     }
4855
4856     if (!(fragment_priv = fragment_pipe->alloc_private(&arb_program_shader_backend, priv)))
4857     {
4858         ERR("Failed to initialize fragment pipe.\n");
4859         vertex_pipe->vp_free(device);
4860         HeapFree(GetProcessHeap(), 0, priv);
4861         return E_FAIL;
4862     }
4863
4864     priv->vshader_const_dirty = HeapAlloc(GetProcessHeap(), 0,
4865             sizeof(*priv->vshader_const_dirty) * device->d3d_vshader_constantF);
4866     if (!priv->vshader_const_dirty)
4867         goto fail;
4868     memset(priv->vshader_const_dirty, 1,
4869            sizeof(*priv->vshader_const_dirty) * device->d3d_vshader_constantF);
4870
4871     priv->pshader_const_dirty = HeapAlloc(GetProcessHeap(), 0,
4872             sizeof(*priv->pshader_const_dirty) * device->d3d_pshader_constantF);
4873     if (!priv->pshader_const_dirty)
4874         goto fail;
4875     memset(priv->pshader_const_dirty, 1,
4876             sizeof(*priv->pshader_const_dirty) * device->d3d_pshader_constantF);
4877
4878     if(wine_rb_init(&priv->signature_tree, &sig_tree_functions) == -1)
4879     {
4880         ERR("RB tree init failed\n");
4881         goto fail;
4882     }
4883
4884     priv->vertex_pipe = vertex_pipe;
4885     priv->fragment_pipe = fragment_pipe;
4886     fragment_pipe->get_caps(&device->adapter->gl_info, &fragment_caps);
4887     priv->ffp_proj_control = fragment_caps.wined3d_caps & WINED3D_FRAGMENT_CAP_PROJ_CONTROL;
4888
4889     device->vertex_priv = vertex_priv;
4890     device->fragment_priv = fragment_priv;
4891     device->shader_priv = priv;
4892
4893     return WINED3D_OK;
4894
4895 fail:
4896     HeapFree(GetProcessHeap(), 0, priv->pshader_const_dirty);
4897     HeapFree(GetProcessHeap(), 0, priv->vshader_const_dirty);
4898     fragment_pipe->free_private(device);
4899     vertex_pipe->vp_free(device);
4900     HeapFree(GetProcessHeap(), 0, priv);
4901     return E_OUTOFMEMORY;
4902 }
4903
4904 static void release_signature(struct wine_rb_entry *entry, void *context)
4905 {
4906     struct ps_signature *sig = WINE_RB_ENTRY_VALUE(entry, struct ps_signature, entry);
4907     int i;
4908     for(i = 0; i < MAX_REG_INPUT; i++)
4909     {
4910         HeapFree(GetProcessHeap(), 0, (char *) sig->sig[i].semantic_name);
4911     }
4912     HeapFree(GetProcessHeap(), 0, sig->sig);
4913     HeapFree(GetProcessHeap(), 0, sig);
4914 }
4915
4916 /* Context activation is done by the caller. */
4917 static void shader_arb_free(struct wined3d_device *device)
4918 {
4919     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4920     struct shader_arb_priv *priv = device->shader_priv;
4921     int i;
4922
4923     if (priv->depth_blt_vprogram_id)
4924         GL_EXTCALL(glDeleteProgramsARB(1, &priv->depth_blt_vprogram_id));
4925
4926     for (i = 0; i < tex_type_count; ++i)
4927     {
4928         if (priv->depth_blt_fprogram_id_full[i])
4929         {
4930             GL_EXTCALL(glDeleteProgramsARB(1, &priv->depth_blt_fprogram_id_full[i]));
4931         }
4932         if (priv->depth_blt_fprogram_id_masked[i])
4933         {
4934             GL_EXTCALL(glDeleteProgramsARB(1, &priv->depth_blt_fprogram_id_masked[i]));
4935         }
4936     }
4937
4938     wine_rb_destroy(&priv->signature_tree, release_signature, NULL);
4939     HeapFree(GetProcessHeap(), 0, priv->pshader_const_dirty);
4940     HeapFree(GetProcessHeap(), 0, priv->vshader_const_dirty);
4941     priv->fragment_pipe->free_private(device);
4942     priv->vertex_pipe->vp_free(device);
4943     HeapFree(GetProcessHeap(), 0, device->shader_priv);
4944 }
4945
4946 static void shader_arb_context_destroyed(void *shader_priv, const struct wined3d_context *context)
4947 {
4948     struct shader_arb_priv *priv = shader_priv;
4949
4950     if (priv->last_context == context)
4951         priv->last_context = NULL;
4952 }
4953
4954 static void shader_arb_get_caps(const struct wined3d_gl_info *gl_info, struct shader_caps *caps)
4955 {
4956     if (gl_info->supported[ARB_VERTEX_PROGRAM])
4957     {
4958         DWORD vs_consts;
4959         UINT vs_version;
4960
4961         /* 96 is the minimum allowed value of MAX_PROGRAM_ENV_PARAMETERS_ARB
4962          * for vertex programs. If the native limit is less than that it's
4963          * not very useful, and e.g. Mesa swrast returns 0, probably to
4964          * indicate it's a software implementation. */
4965         if (gl_info->limits.arb_vs_native_constants < 96)
4966             vs_consts = gl_info->limits.arb_vs_float_constants;
4967         else
4968             vs_consts = min(gl_info->limits.arb_vs_float_constants, gl_info->limits.arb_vs_native_constants);
4969
4970         if (gl_info->supported[NV_VERTEX_PROGRAM3])
4971         {
4972             vs_version = 3;
4973             TRACE("Hardware vertex shader version 3.0 enabled (NV_VERTEX_PROGRAM3)\n");
4974         }
4975         else if (vs_consts >= 256)
4976         {
4977             /* Shader Model 2.0 requires at least 256 vertex shader constants */
4978             vs_version = 2;
4979             TRACE("Hardware vertex shader version 2.0 enabled (ARB_PROGRAM)\n");
4980         }
4981         else
4982         {
4983             vs_version = 1;
4984             TRACE("Hardware vertex shader version 1.1 enabled (ARB_PROGRAM)\n");
4985         }
4986         caps->vs_version = min(wined3d_settings.max_sm_vs, vs_version);
4987         caps->vs_uniform_count = vs_consts;
4988     }
4989     else
4990     {
4991         caps->vs_version = 0;
4992         caps->vs_uniform_count = 0;
4993     }
4994
4995     caps->gs_version = 0;
4996
4997     if (gl_info->supported[ARB_FRAGMENT_PROGRAM])
4998     {
4999         DWORD ps_consts;
5000         UINT ps_version;
5001
5002         /* Similar as above for vertex programs, but the minimum for fragment
5003          * programs is 24. */
5004         if (gl_info->limits.arb_ps_native_constants < 24)
5005             ps_consts = gl_info->limits.arb_ps_float_constants;
5006         else
5007             ps_consts = min(gl_info->limits.arb_ps_float_constants, gl_info->limits.arb_ps_native_constants);
5008
5009         if (gl_info->supported[NV_FRAGMENT_PROGRAM2])
5010         {
5011             ps_version = 3;
5012             TRACE("Hardware pixel shader version 3.0 enabled (NV_FRAGMENT_PROGRAM2)\n");
5013         }
5014         else if (ps_consts >= 32)
5015         {
5016             /* Shader Model 2.0 requires at least 32 pixel shader constants */
5017             ps_version = 2;
5018             TRACE("Hardware pixel shader version 2.0 enabled (ARB_PROGRAM)\n");
5019         }
5020         else
5021         {
5022             ps_version = 1;
5023             TRACE("Hardware pixel shader version 1.4 enabled (ARB_PROGRAM)\n");
5024         }
5025         caps->ps_version = min(wined3d_settings.max_sm_ps, ps_version);
5026         caps->ps_uniform_count = ps_consts;
5027         caps->ps_1x_max_value = 8.0f;
5028     }
5029     else
5030     {
5031         caps->ps_version = 0;
5032         caps->ps_uniform_count = 0;
5033         caps->ps_1x_max_value = 0.0f;
5034     }
5035
5036     caps->wined3d_caps = WINED3D_SHADER_CAP_SRGB_WRITE;
5037     if (use_nv_clip(gl_info))
5038         caps->wined3d_caps |= WINED3D_SHADER_CAP_VS_CLIPPING;
5039 }
5040
5041 static BOOL shader_arb_color_fixup_supported(struct color_fixup_desc fixup)
5042 {
5043     if (TRACE_ON(d3d_shader) && TRACE_ON(d3d))
5044     {
5045         TRACE("Checking support for color_fixup:\n");
5046         dump_color_fixup_desc(fixup);
5047     }
5048
5049     /* We support everything except complex conversions. */
5050     if (!is_complex_fixup(fixup))
5051     {
5052         TRACE("[OK]\n");
5053         return TRUE;
5054     }
5055
5056     TRACE("[FAILED]\n");
5057     return FALSE;
5058 }
5059
5060 static void shader_arb_add_instruction_modifiers(const struct wined3d_shader_instruction *ins) {
5061     DWORD shift;
5062     char write_mask[20], regstr[50];
5063     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
5064     BOOL is_color = FALSE;
5065     const struct wined3d_shader_dst_param *dst;
5066
5067     if (!ins->dst_count) return;
5068
5069     dst = &ins->dst[0];
5070     shift = dst->shift;
5071     if (!shift) return; /* Saturate alone is handled by the instructions */
5072
5073     shader_arb_get_write_mask(ins, dst, write_mask);
5074     shader_arb_get_register_name(ins, &dst->reg, regstr, &is_color);
5075
5076     /* Generate a line that does the output modifier computation
5077      * FIXME: _SAT vs shift? _SAT alone is already handled in the instructions, if this
5078      * maps problems in e.g. _d4_sat modify shader_arb_get_modifier
5079      */
5080     shader_addline(buffer, "MUL%s %s%s, %s, %s;\n", shader_arb_get_modifier(ins),
5081                    regstr, write_mask, regstr, shift_tab[shift]);
5082 }
5083
5084 static const SHADER_HANDLER shader_arb_instruction_handler_table[WINED3DSIH_TABLE_SIZE] =
5085 {
5086     /* WINED3DSIH_ABS                   */ shader_hw_map2gl,
5087     /* WINED3DSIH_ADD                   */ shader_hw_map2gl,
5088     /* WINED3DSIH_AND                   */ NULL,
5089     /* WINED3DSIH_BEM                   */ pshader_hw_bem,
5090     /* WINED3DSIH_BREAK                 */ shader_hw_break,
5091     /* WINED3DSIH_BREAKC                */ shader_hw_breakc,
5092     /* WINED3DSIH_BREAKP                */ NULL,
5093     /* WINED3DSIH_CALL                  */ shader_hw_call,
5094     /* WINED3DSIH_CALLNZ                */ NULL,
5095     /* WINED3DSIH_CMP                   */ pshader_hw_cmp,
5096     /* WINED3DSIH_CND                   */ pshader_hw_cnd,
5097     /* WINED3DSIH_CRS                   */ shader_hw_map2gl,
5098     /* WINED3DSIH_CUT                   */ NULL,
5099     /* WINED3DSIH_DCL                   */ shader_hw_nop,
5100     /* WINED3DSIH_DCL_CONSTANT_BUFFER   */ shader_hw_nop,
5101     /* WINED3DSIH_DCL_INPUT_PRIMITIVE   */ shader_hw_nop,
5102     /* WINED3DSIH_DCL_OUTPUT_TOPOLOGY   */ shader_hw_nop,
5103     /* WINED3DSIH_DCL_VERTICES_OUT      */ shader_hw_nop,
5104     /* WINED3DSIH_DEF                   */ shader_hw_nop,
5105     /* WINED3DSIH_DEFB                  */ shader_hw_nop,
5106     /* WINED3DSIH_DEFI                  */ shader_hw_nop,
5107     /* WINED3DSIH_DIV                   */ NULL,
5108     /* WINED3DSIH_DP2ADD                */ pshader_hw_dp2add,
5109     /* WINED3DSIH_DP3                   */ shader_hw_map2gl,
5110     /* WINED3DSIH_DP4                   */ shader_hw_map2gl,
5111     /* WINED3DSIH_DST                   */ shader_hw_map2gl,
5112     /* WINED3DSIH_DSX                   */ shader_hw_map2gl,
5113     /* WINED3DSIH_DSY                   */ shader_hw_dsy,
5114     /* WINED3DSIH_ELSE                  */ shader_hw_else,
5115     /* WINED3DSIH_EMIT                  */ NULL,
5116     /* WINED3DSIH_ENDIF                 */ shader_hw_endif,
5117     /* WINED3DSIH_ENDLOOP               */ shader_hw_endloop,
5118     /* WINED3DSIH_ENDREP                */ shader_hw_endrep,
5119     /* WINED3DSIH_EQ                    */ NULL,
5120     /* WINED3DSIH_EXP                   */ shader_hw_scalar_op,
5121     /* WINED3DSIH_EXPP                  */ shader_hw_scalar_op,
5122     /* WINED3DSIH_FRC                   */ shader_hw_map2gl,
5123     /* WINED3DSIH_FTOI                  */ NULL,
5124     /* WINED3DSIH_GE                    */ NULL,
5125     /* WINED3DSIH_IADD                  */ NULL,
5126     /* WINED3DSIH_IEQ                   */ NULL,
5127     /* WINED3DSIH_IF                    */ NULL /* Hardcoded into the shader */,
5128     /* WINED3DSIH_IFC                   */ shader_hw_ifc,
5129     /* WINED3DSIH_IGE                   */ NULL,
5130     /* WINED3DSIH_IMUL                  */ NULL,
5131     /* WINED3DSIH_ITOF                  */ NULL,
5132     /* WINED3DSIH_LABEL                 */ shader_hw_label,
5133     /* WINED3DSIH_LD                    */ NULL,
5134     /* WINED3DSIH_LIT                   */ shader_hw_map2gl,
5135     /* WINED3DSIH_LOG                   */ shader_hw_log,
5136     /* WINED3DSIH_LOGP                  */ shader_hw_log,
5137     /* WINED3DSIH_LOOP                  */ shader_hw_loop,
5138     /* WINED3DSIH_LRP                   */ shader_hw_lrp,
5139     /* WINED3DSIH_LT                    */ NULL,
5140     /* WINED3DSIH_M3x2                  */ shader_hw_mnxn,
5141     /* WINED3DSIH_M3x3                  */ shader_hw_mnxn,
5142     /* WINED3DSIH_M3x4                  */ shader_hw_mnxn,
5143     /* WINED3DSIH_M4x3                  */ shader_hw_mnxn,
5144     /* WINED3DSIH_M4x4                  */ shader_hw_mnxn,
5145     /* WINED3DSIH_MAD                   */ shader_hw_map2gl,
5146     /* WINED3DSIH_MAX                   */ shader_hw_map2gl,
5147     /* WINED3DSIH_MIN                   */ shader_hw_map2gl,
5148     /* WINED3DSIH_MOV                   */ shader_hw_mov,
5149     /* WINED3DSIH_MOVA                  */ shader_hw_mov,
5150     /* WINED3DSIH_MOVC                  */ NULL,
5151     /* WINED3DSIH_MUL                   */ shader_hw_map2gl,
5152     /* WINED3DSIH_NOP                   */ shader_hw_nop,
5153     /* WINED3DSIH_NRM                   */ shader_hw_nrm,
5154     /* WINED3DSIH_PHASE                 */ shader_hw_nop,
5155     /* WINED3DSIH_POW                   */ shader_hw_pow,
5156     /* WINED3DSIH_RCP                   */ shader_hw_rcp,
5157     /* WINED3DSIH_REP                   */ shader_hw_rep,
5158     /* WINED3DSIH_RET                   */ shader_hw_ret,
5159     /* WINED3DSIH_ROUND_NI              */ NULL,
5160     /* WINED3DSIH_RSQ                   */ shader_hw_scalar_op,
5161     /* WINED3DSIH_SAMPLE                */ NULL,
5162     /* WINED3DSIH_SAMPLE_GRAD           */ NULL,
5163     /* WINED3DSIH_SAMPLE_LOD            */ NULL,
5164     /* WINED3DSIH_SETP                  */ NULL,
5165     /* WINED3DSIH_SGE                   */ shader_hw_map2gl,
5166     /* WINED3DSIH_SGN                   */ shader_hw_sgn,
5167     /* WINED3DSIH_SINCOS                */ shader_hw_sincos,
5168     /* WINED3DSIH_SLT                   */ shader_hw_map2gl,
5169     /* WINED3DSIH_SQRT                  */ NULL,
5170     /* WINED3DSIH_SUB                   */ shader_hw_map2gl,
5171     /* WINED3DSIH_TEX                   */ pshader_hw_tex,
5172     /* WINED3DSIH_TEXBEM                */ pshader_hw_texbem,
5173     /* WINED3DSIH_TEXBEML               */ pshader_hw_texbem,
5174     /* WINED3DSIH_TEXCOORD              */ pshader_hw_texcoord,
5175     /* WINED3DSIH_TEXDEPTH              */ pshader_hw_texdepth,
5176     /* WINED3DSIH_TEXDP3                */ pshader_hw_texdp3,
5177     /* WINED3DSIH_TEXDP3TEX             */ pshader_hw_texdp3tex,
5178     /* WINED3DSIH_TEXKILL               */ pshader_hw_texkill,
5179     /* WINED3DSIH_TEXLDD                */ shader_hw_texldd,
5180     /* WINED3DSIH_TEXLDL                */ shader_hw_texldl,
5181     /* WINED3DSIH_TEXM3x2DEPTH          */ pshader_hw_texm3x2depth,
5182     /* WINED3DSIH_TEXM3x2PAD            */ pshader_hw_texm3x2pad,
5183     /* WINED3DSIH_TEXM3x2TEX            */ pshader_hw_texm3x2tex,
5184     /* WINED3DSIH_TEXM3x3               */ pshader_hw_texm3x3,
5185     /* WINED3DSIH_TEXM3x3DIFF           */ NULL,
5186     /* WINED3DSIH_TEXM3x3PAD            */ pshader_hw_texm3x3pad,
5187     /* WINED3DSIH_TEXM3x3SPEC           */ pshader_hw_texm3x3spec,
5188     /* WINED3DSIH_TEXM3x3TEX            */ pshader_hw_texm3x3tex,
5189     /* WINED3DSIH_TEXM3x3VSPEC          */ pshader_hw_texm3x3vspec,
5190     /* WINED3DSIH_TEXREG2AR             */ pshader_hw_texreg2ar,
5191     /* WINED3DSIH_TEXREG2GB             */ pshader_hw_texreg2gb,
5192     /* WINED3DSIH_TEXREG2RGB            */ pshader_hw_texreg2rgb,
5193     /* WINED3DSIH_UDIV                  */ NULL,
5194     /* WINED3DSIH_USHR                  */ NULL,
5195     /* WINED3DSIH_UTOF                  */ NULL,
5196     /* WINED3DSIH_XOR                   */ NULL,
5197 };
5198
5199 static BOOL get_bool_const(const struct wined3d_shader_instruction *ins,
5200         const struct wined3d_shader *shader, DWORD idx)
5201 {
5202     const struct wined3d_shader_reg_maps *reg_maps = ins->ctx->reg_maps;
5203     BOOL vshader = shader_is_vshader_version(reg_maps->shader_version.type);
5204     const struct wined3d_shader_lconst *constant;
5205     WORD bools = 0;
5206     WORD flag = (1 << idx);
5207     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
5208
5209     if (reg_maps->local_bool_consts & flag)
5210     {
5211         /* What good is a if(bool) with a hardcoded local constant? I don't know, but handle it */
5212         LIST_FOR_EACH_ENTRY(constant, &shader->constantsB, struct wined3d_shader_lconst, entry)
5213         {
5214             if (constant->idx == idx)
5215             {
5216                 return constant->value[0];
5217             }
5218         }
5219         ERR("Local constant not found\n");
5220         return FALSE;
5221     }
5222     else
5223     {
5224         if(vshader) bools = priv->cur_vs_args->clip.boolclip.bools;
5225         else bools = priv->cur_ps_args->bools;
5226         return bools & flag;
5227     }
5228 }
5229
5230 static void get_loop_control_const(const struct wined3d_shader_instruction *ins,
5231         const struct wined3d_shader *shader, UINT idx, struct wined3d_shader_loop_control *loop_control)
5232 {
5233     const struct wined3d_shader_reg_maps *reg_maps = ins->ctx->reg_maps;
5234     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
5235
5236     /* Integer constants can either be a local constant, or they can be stored in the shader
5237      * type specific compile args. */
5238     if (reg_maps->local_int_consts & (1 << idx))
5239     {
5240         const struct wined3d_shader_lconst *constant;
5241
5242         LIST_FOR_EACH_ENTRY(constant, &shader->constantsI, struct wined3d_shader_lconst, entry)
5243         {
5244             if (constant->idx == idx)
5245             {
5246                 loop_control->count = constant->value[0];
5247                 loop_control->start = constant->value[1];
5248                 /* Step is signed. */
5249                 loop_control->step = (int)constant->value[2];
5250                 return;
5251             }
5252         }
5253         /* If this happens the flag was set incorrectly */
5254         ERR("Local constant not found\n");
5255         loop_control->count = 0;
5256         loop_control->start = 0;
5257         loop_control->step = 0;
5258         return;
5259     }
5260
5261     switch (reg_maps->shader_version.type)
5262     {
5263         case WINED3D_SHADER_TYPE_VERTEX:
5264             /* Count and aL start value are unsigned */
5265             loop_control->count = priv->cur_vs_args->loop_ctrl[idx][0];
5266             loop_control->start = priv->cur_vs_args->loop_ctrl[idx][1];
5267             /* Step is signed. */
5268             loop_control->step = ((char)priv->cur_vs_args->loop_ctrl[idx][2]);
5269             break;
5270
5271         case WINED3D_SHADER_TYPE_PIXEL:
5272             loop_control->count = priv->cur_ps_args->loop_ctrl[idx][0];
5273             loop_control->start = priv->cur_ps_args->loop_ctrl[idx][1];
5274             loop_control->step = ((char)priv->cur_ps_args->loop_ctrl[idx][2]);
5275             break;
5276
5277         default:
5278             FIXME("Unhandled shader type %#x.\n", reg_maps->shader_version.type);
5279             break;
5280     }
5281 }
5282
5283 static void record_instruction(struct list *list, const struct wined3d_shader_instruction *ins)
5284 {
5285     unsigned int i;
5286     struct wined3d_shader_dst_param *dst_param = NULL;
5287     struct wined3d_shader_src_param *src_param = NULL, *rel_addr = NULL;
5288     struct recorded_instruction *rec = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*rec));
5289     if(!rec)
5290     {
5291         ERR("Out of memory\n");
5292         return;
5293     }
5294
5295     rec->ins = *ins;
5296     dst_param = HeapAlloc(GetProcessHeap(), 0, sizeof(*dst_param));
5297     if(!dst_param) goto free;
5298     *dst_param = *ins->dst;
5299     if (ins->dst->reg.idx[0].rel_addr)
5300     {
5301         rel_addr = HeapAlloc(GetProcessHeap(), 0, sizeof(*rel_addr));
5302         if (!rel_addr)
5303             goto free;
5304         *rel_addr = *ins->dst->reg.idx[0].rel_addr;
5305         dst_param->reg.idx[0].rel_addr = rel_addr;
5306     }
5307     rec->ins.dst = dst_param;
5308
5309     src_param = HeapAlloc(GetProcessHeap(), 0, sizeof(*src_param) * ins->src_count);
5310     if (!src_param)
5311         goto free;
5312     for (i = 0; i < ins->src_count; ++i)
5313     {
5314         src_param[i] = ins->src[i];
5315         if (ins->src[i].reg.idx[0].rel_addr)
5316         {
5317             rel_addr = HeapAlloc(GetProcessHeap(), 0, sizeof(*rel_addr));
5318             if (!rel_addr)
5319                 goto free;
5320             *rel_addr = *ins->src[i].reg.idx[0].rel_addr;
5321             src_param[i].reg.idx[0].rel_addr = rel_addr;
5322         }
5323     }
5324     rec->ins.src = src_param;
5325     list_add_tail(list, &rec->entry);
5326     return;
5327
5328 free:
5329     ERR("Out of memory\n");
5330     if(dst_param)
5331     {
5332         HeapFree(GetProcessHeap(), 0, (void *)dst_param->reg.idx[0].rel_addr);
5333         HeapFree(GetProcessHeap(), 0, dst_param);
5334     }
5335     if(src_param)
5336     {
5337         for(i = 0; i < ins->src_count; i++)
5338         {
5339             HeapFree(GetProcessHeap(), 0, (void *)src_param[i].reg.idx[0].rel_addr);
5340         }
5341         HeapFree(GetProcessHeap(), 0, src_param);
5342     }
5343     HeapFree(GetProcessHeap(), 0, rec);
5344 }
5345
5346 static void free_recorded_instruction(struct list *list)
5347 {
5348     struct recorded_instruction *rec_ins, *entry2;
5349     unsigned int i;
5350
5351     LIST_FOR_EACH_ENTRY_SAFE(rec_ins, entry2, list, struct recorded_instruction, entry)
5352     {
5353         list_remove(&rec_ins->entry);
5354         if (rec_ins->ins.dst)
5355         {
5356             HeapFree(GetProcessHeap(), 0, (void *)rec_ins->ins.dst->reg.idx[0].rel_addr);
5357             HeapFree(GetProcessHeap(), 0, (void *)rec_ins->ins.dst);
5358         }
5359         if (rec_ins->ins.src)
5360         {
5361             for (i = 0; i < rec_ins->ins.src_count; ++i)
5362             {
5363                 HeapFree(GetProcessHeap(), 0, (void *)rec_ins->ins.src[i].reg.idx[0].rel_addr);
5364             }
5365             HeapFree(GetProcessHeap(), 0, (void *)rec_ins->ins.src);
5366         }
5367         HeapFree(GetProcessHeap(), 0, rec_ins);
5368     }
5369 }
5370
5371 static void pop_control_frame(const struct wined3d_shader_instruction *ins)
5372 {
5373     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
5374     struct control_frame *control_frame;
5375
5376     if (ins->handler_idx == WINED3DSIH_ENDLOOP || ins->handler_idx == WINED3DSIH_ENDREP)
5377     {
5378         struct list *e = list_head(&priv->control_frames);
5379         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5380         list_remove(&control_frame->entry);
5381         HeapFree(GetProcessHeap(), 0, control_frame);
5382         priv->loop_depth--;
5383     }
5384     else if (ins->handler_idx == WINED3DSIH_ENDIF)
5385     {
5386         /* Non-ifc ENDIFs were already handled previously. */
5387         struct list *e = list_head(&priv->control_frames);
5388         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5389         list_remove(&control_frame->entry);
5390         HeapFree(GetProcessHeap(), 0, control_frame);
5391     }
5392 }
5393
5394 static void shader_arb_handle_instruction(const struct wined3d_shader_instruction *ins) {
5395     SHADER_HANDLER hw_fct;
5396     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
5397     const struct wined3d_shader *shader = ins->ctx->shader;
5398     struct control_frame *control_frame;
5399     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
5400     BOOL bool_const;
5401
5402     if(ins->handler_idx == WINED3DSIH_LOOP || ins->handler_idx == WINED3DSIH_REP)
5403     {
5404         control_frame = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*control_frame));
5405         list_add_head(&priv->control_frames, &control_frame->entry);
5406
5407         if(ins->handler_idx == WINED3DSIH_LOOP) control_frame->type = LOOP;
5408         if(ins->handler_idx == WINED3DSIH_REP) control_frame->type = REP;
5409
5410         if(priv->target_version >= NV2)
5411         {
5412             control_frame->no.loop = priv->num_loops++;
5413             priv->loop_depth++;
5414         }
5415         else
5416         {
5417             /* Don't bother recording when we're in a not used if branch */
5418             if(priv->muted)
5419             {
5420                 return;
5421             }
5422
5423             if(!priv->recording)
5424             {
5425                 list_init(&priv->record);
5426                 priv->recording = TRUE;
5427                 control_frame->outer_loop = TRUE;
5428                 get_loop_control_const(ins, shader, ins->src[0].reg.idx[0].offset, &control_frame->loop_control);
5429                 return; /* Instruction is handled */
5430             }
5431             /* Record this loop in the outer loop's recording */
5432         }
5433     }
5434     else if(ins->handler_idx == WINED3DSIH_ENDLOOP || ins->handler_idx == WINED3DSIH_ENDREP)
5435     {
5436         if(priv->target_version >= NV2)
5437         {
5438             /* Nothing to do. The control frame is popped after the HW instr handler */
5439         }
5440         else
5441         {
5442             struct list *e = list_head(&priv->control_frames);
5443             control_frame = LIST_ENTRY(e, struct control_frame, entry);
5444             list_remove(&control_frame->entry);
5445
5446             if(control_frame->outer_loop)
5447             {
5448                 unsigned int iteration;
5449                 int aL = 0;
5450                 struct list copy;
5451
5452                 /* Turn off recording before playback */
5453                 priv->recording = FALSE;
5454
5455                 /* Move the recorded instructions to a separate list and get them out of the private data
5456                  * structure. If there are nested loops, the shader_arb_handle_instruction below will
5457                  * be recorded again, thus priv->record might be overwritten
5458                  */
5459                 list_init(&copy);
5460                 list_move_tail(&copy, &priv->record);
5461                 list_init(&priv->record);
5462
5463                 if(ins->handler_idx == WINED3DSIH_ENDLOOP)
5464                 {
5465                     shader_addline(buffer, "#unrolling loop: %u iterations, aL=%u, inc %d\n",
5466                                    control_frame->loop_control.count, control_frame->loop_control.start,
5467                                    control_frame->loop_control.step);
5468                     aL = control_frame->loop_control.start;
5469                 }
5470                 else
5471                 {
5472                     shader_addline(buffer, "#unrolling rep: %u iterations\n", control_frame->loop_control.count);
5473                 }
5474
5475                 for (iteration = 0; iteration < control_frame->loop_control.count; ++iteration)
5476                 {
5477                     struct recorded_instruction *rec_ins;
5478                     if(ins->handler_idx == WINED3DSIH_ENDLOOP)
5479                     {
5480                         priv->aL = aL;
5481                         shader_addline(buffer, "#Iteration %u, aL=%d\n", iteration, aL);
5482                     }
5483                     else
5484                     {
5485                         shader_addline(buffer, "#Iteration %u\n", iteration);
5486                     }
5487
5488                     LIST_FOR_EACH_ENTRY(rec_ins, &copy, struct recorded_instruction, entry)
5489                     {
5490                         shader_arb_handle_instruction(&rec_ins->ins);
5491                     }
5492
5493                     if(ins->handler_idx == WINED3DSIH_ENDLOOP)
5494                     {
5495                         aL += control_frame->loop_control.step;
5496                     }
5497                 }
5498                 shader_addline(buffer, "#end loop/rep\n");
5499
5500                 free_recorded_instruction(&copy);
5501                 HeapFree(GetProcessHeap(), 0, control_frame);
5502                 return; /* Instruction is handled */
5503             }
5504             else
5505             {
5506                 /* This is a nested loop. Proceed to the normal recording function */
5507                 HeapFree(GetProcessHeap(), 0, control_frame);
5508             }
5509         }
5510     }
5511
5512     if(priv->recording)
5513     {
5514         record_instruction(&priv->record, ins);
5515         return;
5516     }
5517
5518     /* boolean if */
5519     if(ins->handler_idx == WINED3DSIH_IF)
5520     {
5521         control_frame = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*control_frame));
5522         list_add_head(&priv->control_frames, &control_frame->entry);
5523         control_frame->type = IF;
5524
5525         bool_const = get_bool_const(ins, shader, ins->src[0].reg.idx[0].offset);
5526         if (ins->src[0].modifiers == WINED3DSPSM_NOT)
5527             bool_const = !bool_const;
5528         if (!priv->muted && !bool_const)
5529         {
5530             shader_addline(buffer, "#if(FALSE){\n");
5531             priv->muted = TRUE;
5532             control_frame->muting = TRUE;
5533         }
5534         else shader_addline(buffer, "#if(TRUE) {\n");
5535
5536         return; /* Instruction is handled */
5537     }
5538     else if(ins->handler_idx == WINED3DSIH_IFC)
5539     {
5540         /* IF(bool) and if_cond(a, b) use the same ELSE and ENDIF tokens */
5541         control_frame = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*control_frame));
5542         control_frame->type = IFC;
5543         control_frame->no.ifc = priv->num_ifcs++;
5544         list_add_head(&priv->control_frames, &control_frame->entry);
5545     }
5546     else if(ins->handler_idx == WINED3DSIH_ELSE)
5547     {
5548         struct list *e = list_head(&priv->control_frames);
5549         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5550
5551         if(control_frame->type == IF)
5552         {
5553             shader_addline(buffer, "#} else {\n");
5554             if(!priv->muted && !control_frame->muting)
5555             {
5556                 priv->muted = TRUE;
5557                 control_frame->muting = TRUE;
5558             }
5559             else if(control_frame->muting) priv->muted = FALSE;
5560             return; /* Instruction is handled. */
5561         }
5562         /* In case of an ifc, generate a HW shader instruction */
5563         if (control_frame->type != IFC)
5564             ERR("Control frame does not match.\n");
5565     }
5566     else if(ins->handler_idx == WINED3DSIH_ENDIF)
5567     {
5568         struct list *e = list_head(&priv->control_frames);
5569         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5570
5571         if(control_frame->type == IF)
5572         {
5573             shader_addline(buffer, "#} endif\n");
5574             if(control_frame->muting) priv->muted = FALSE;
5575             list_remove(&control_frame->entry);
5576             HeapFree(GetProcessHeap(), 0, control_frame);
5577             return; /* Instruction is handled */
5578         }
5579         /* In case of an ifc, generate a HW shader instruction */
5580         if (control_frame->type != IFC)
5581             ERR("Control frame does not match.\n");
5582     }
5583
5584     if(priv->muted)
5585     {
5586         pop_control_frame(ins);
5587         return;
5588     }
5589
5590     /* Select handler */
5591     hw_fct = shader_arb_instruction_handler_table[ins->handler_idx];
5592
5593     /* Unhandled opcode */
5594     if (!hw_fct)
5595     {
5596         FIXME("Backend can't handle opcode %#x\n", ins->handler_idx);
5597         return;
5598     }
5599     hw_fct(ins);
5600
5601     pop_control_frame(ins);
5602
5603     shader_arb_add_instruction_modifiers(ins);
5604 }
5605
5606 static BOOL shader_arb_has_ffp_proj_control(void *shader_priv)
5607 {
5608     struct shader_arb_priv *priv = shader_priv;
5609
5610     return priv->ffp_proj_control;
5611 }
5612
5613 const struct wined3d_shader_backend_ops arb_program_shader_backend =
5614 {
5615     shader_arb_handle_instruction,
5616     shader_arb_select,
5617     shader_arb_select_depth_blt,
5618     shader_arb_deselect_depth_blt,
5619     shader_arb_update_float_vertex_constants,
5620     shader_arb_update_float_pixel_constants,
5621     shader_arb_load_constants,
5622     shader_arb_load_np2fixup_constants,
5623     shader_arb_destroy,
5624     shader_arb_alloc,
5625     shader_arb_free,
5626     shader_arb_context_destroyed,
5627     shader_arb_get_caps,
5628     shader_arb_color_fixup_supported,
5629     shader_arb_has_ffp_proj_control,
5630 };
5631
5632 /* ARB_fragment_program fixed function pipeline replacement definitions */
5633 #define ARB_FFP_CONST_TFACTOR           0
5634 #define ARB_FFP_CONST_SPECULAR_ENABLE   ((ARB_FFP_CONST_TFACTOR) + 1)
5635 #define ARB_FFP_CONST_CONSTANT(i)       ((ARB_FFP_CONST_SPECULAR_ENABLE) + 1 + i)
5636 #define ARB_FFP_CONST_BUMPMAT(i)        ((ARB_FFP_CONST_CONSTANT(7)) + 1 + i)
5637 #define ARB_FFP_CONST_LUMINANCE(i)      ((ARB_FFP_CONST_BUMPMAT(7)) + 1 + i)
5638
5639 struct arbfp_ffp_desc
5640 {
5641     struct ffp_frag_desc parent;
5642     GLuint shader;
5643 };
5644
5645 /* Context activation is done by the caller. */
5646 static void arbfp_enable(const struct wined3d_gl_info *gl_info, BOOL enable)
5647 {
5648     if (enable)
5649     {
5650         gl_info->gl_ops.gl.p_glEnable(GL_FRAGMENT_PROGRAM_ARB);
5651         checkGLcall("glEnable(GL_FRAGMENT_PROGRAM_ARB)");
5652     }
5653     else
5654     {
5655         gl_info->gl_ops.gl.p_glDisable(GL_FRAGMENT_PROGRAM_ARB);
5656         checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
5657     }
5658 }
5659
5660 static void *arbfp_alloc(const struct wined3d_shader_backend_ops *shader_backend, void *shader_priv)
5661 {
5662     struct shader_arb_priv *priv;
5663
5664     /* Share private data between the shader backend and the pipeline
5665      * replacement, if both are the arb implementation. This is needed to
5666      * figure out whether ARBfp should be disabled if no pixel shader is bound
5667      * or not. */
5668     if (shader_backend == &arb_program_shader_backend)
5669         priv = shader_priv;
5670     else if (!(priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*priv))))
5671         return NULL;
5672
5673     if (wine_rb_init(&priv->fragment_shaders, &wined3d_ffp_frag_program_rb_functions) == -1)
5674     {
5675         ERR("Failed to initialize rbtree.\n");
5676         if (priv != shader_priv)
5677             HeapFree(GetProcessHeap(), 0, priv);
5678         return NULL;
5679     }
5680     priv->use_arbfp_fixed_func = TRUE;
5681
5682     return priv;
5683 }
5684
5685 /* Context activation is done by the caller. */
5686 static void arbfp_free_ffpshader(struct wine_rb_entry *entry, void *context)
5687 {
5688     const struct wined3d_gl_info *gl_info = context;
5689     struct arbfp_ffp_desc *entry_arb = WINE_RB_ENTRY_VALUE(entry, struct arbfp_ffp_desc, parent.entry);
5690
5691     GL_EXTCALL(glDeleteProgramsARB(1, &entry_arb->shader));
5692     checkGLcall("glDeleteProgramsARB(1, &entry_arb->shader)");
5693     HeapFree(GetProcessHeap(), 0, entry_arb);
5694 }
5695
5696 /* Context activation is done by the caller. */
5697 static void arbfp_free(struct wined3d_device *device)
5698 {
5699     struct shader_arb_priv *priv = device->fragment_priv;
5700
5701     wine_rb_destroy(&priv->fragment_shaders, arbfp_free_ffpshader, &device->adapter->gl_info);
5702     priv->use_arbfp_fixed_func = FALSE;
5703
5704     if (device->shader_backend != &arb_program_shader_backend)
5705     {
5706         HeapFree(GetProcessHeap(), 0, device->fragment_priv);
5707     }
5708 }
5709
5710 static void arbfp_get_caps(const struct wined3d_gl_info *gl_info, struct fragment_caps *caps)
5711 {
5712     caps->wined3d_caps = WINED3D_FRAGMENT_CAP_PROJ_CONTROL
5713             | WINED3D_FRAGMENT_CAP_SRGB_WRITE;
5714     caps->PrimitiveMiscCaps = WINED3DPMISCCAPS_TSSARGTEMP;
5715     caps->TextureOpCaps =  WINED3DTEXOPCAPS_DISABLE                     |
5716                            WINED3DTEXOPCAPS_SELECTARG1                  |
5717                            WINED3DTEXOPCAPS_SELECTARG2                  |
5718                            WINED3DTEXOPCAPS_MODULATE4X                  |
5719                            WINED3DTEXOPCAPS_MODULATE2X                  |
5720                            WINED3DTEXOPCAPS_MODULATE                    |
5721                            WINED3DTEXOPCAPS_ADDSIGNED2X                 |
5722                            WINED3DTEXOPCAPS_ADDSIGNED                   |
5723                            WINED3DTEXOPCAPS_ADD                         |
5724                            WINED3DTEXOPCAPS_SUBTRACT                    |
5725                            WINED3DTEXOPCAPS_ADDSMOOTH                   |
5726                            WINED3DTEXOPCAPS_BLENDCURRENTALPHA           |
5727                            WINED3DTEXOPCAPS_BLENDFACTORALPHA            |
5728                            WINED3DTEXOPCAPS_BLENDTEXTUREALPHA           |
5729                            WINED3DTEXOPCAPS_BLENDDIFFUSEALPHA           |
5730                            WINED3DTEXOPCAPS_BLENDTEXTUREALPHAPM         |
5731                            WINED3DTEXOPCAPS_MODULATEALPHA_ADDCOLOR      |
5732                            WINED3DTEXOPCAPS_MODULATECOLOR_ADDALPHA      |
5733                            WINED3DTEXOPCAPS_MODULATEINVCOLOR_ADDALPHA   |
5734                            WINED3DTEXOPCAPS_MODULATEINVALPHA_ADDCOLOR   |
5735                            WINED3DTEXOPCAPS_DOTPRODUCT3                 |
5736                            WINED3DTEXOPCAPS_MULTIPLYADD                 |
5737                            WINED3DTEXOPCAPS_LERP                        |
5738                            WINED3DTEXOPCAPS_BUMPENVMAP                  |
5739                            WINED3DTEXOPCAPS_BUMPENVMAPLUMINANCE;
5740
5741     /* TODO: Implement WINED3DTEXOPCAPS_PREMODULATE */
5742
5743     caps->MaxTextureBlendStages   = 8;
5744     caps->MaxSimultaneousTextures = min(gl_info->limits.fragment_samplers, 8);
5745 }
5746
5747 static void state_texfactor_arbfp(struct wined3d_context *context,
5748         const struct wined3d_state *state, DWORD state_id)
5749 {
5750     struct wined3d_device *device = context->swapchain->device;
5751     const struct wined3d_gl_info *gl_info = context->gl_info;
5752     float col[4];
5753
5754     /* Don't load the parameter if we're using an arbfp pixel shader,
5755      * otherwise we'll overwrite application provided constants. */
5756     if (device->shader_backend == &arb_program_shader_backend)
5757     {
5758         struct shader_arb_priv *priv;
5759
5760         if (use_ps(state)) return;
5761
5762         priv = device->shader_priv;
5763         priv->pshader_const_dirty[ARB_FFP_CONST_TFACTOR] = 1;
5764         priv->highest_dirty_ps_const = max(priv->highest_dirty_ps_const, ARB_FFP_CONST_TFACTOR + 1);
5765     }
5766
5767     D3DCOLORTOGLFLOAT4(state->render_states[WINED3D_RS_TEXTUREFACTOR], col);
5768     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_TFACTOR, col));
5769     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_TFACTOR, col)");
5770 }
5771
5772 static void state_arb_specularenable(struct wined3d_context *context,
5773         const struct wined3d_state *state, DWORD state_id)
5774 {
5775     struct wined3d_device *device = context->swapchain->device;
5776     const struct wined3d_gl_info *gl_info = context->gl_info;
5777     float col[4];
5778
5779     /* Don't load the parameter if we're using an arbfp pixel shader, otherwise we'll overwrite
5780      * application provided constants
5781      */
5782     if (device->shader_backend == &arb_program_shader_backend)
5783     {
5784         struct shader_arb_priv *priv;
5785
5786         if (use_ps(state)) return;
5787
5788         priv = device->shader_priv;
5789         priv->pshader_const_dirty[ARB_FFP_CONST_SPECULAR_ENABLE] = 1;
5790         priv->highest_dirty_ps_const = max(priv->highest_dirty_ps_const, ARB_FFP_CONST_SPECULAR_ENABLE + 1);
5791     }
5792
5793     if (state->render_states[WINED3D_RS_SPECULARENABLE])
5794     {
5795         /* The specular color has no alpha */
5796         col[0] = 1.0f; col[1] = 1.0f;
5797         col[2] = 1.0f; col[3] = 0.0f;
5798     } else {
5799         col[0] = 0.0f; col[1] = 0.0f;
5800         col[2] = 0.0f; col[3] = 0.0f;
5801     }
5802     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_SPECULAR_ENABLE, col));
5803     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_SPECULAR_ENABLE, col)");
5804 }
5805
5806 static void set_bumpmat_arbfp(struct wined3d_context *context, const struct wined3d_state *state, DWORD state_id)
5807 {
5808     DWORD stage = (state_id - STATE_TEXTURESTAGE(0, 0)) / (WINED3D_HIGHEST_TEXTURE_STATE + 1);
5809     struct wined3d_device *device = context->swapchain->device;
5810     const struct wined3d_gl_info *gl_info = context->gl_info;
5811     float mat[2][2];
5812
5813     if (use_ps(state))
5814     {
5815         /* The pixel shader has to know the bump env matrix. Do a constants
5816          * update. */
5817         if (stage && (state->pixel_shader->reg_maps.bumpmat & (1 << stage)))
5818             context->load_constants = 1;
5819
5820         if(device->shader_backend == &arb_program_shader_backend) {
5821             /* Exit now, don't set the bumpmat below, otherwise we may overwrite pixel shader constants */
5822             return;
5823         }
5824     }
5825     else if (device->shader_backend == &arb_program_shader_backend)
5826     {
5827         struct shader_arb_priv *priv = device->shader_priv;
5828         priv->pshader_const_dirty[ARB_FFP_CONST_BUMPMAT(stage)] = 1;
5829         priv->highest_dirty_ps_const = max(priv->highest_dirty_ps_const, ARB_FFP_CONST_BUMPMAT(stage) + 1);
5830     }
5831
5832     mat[0][0] = *((float *)&state->texture_states[stage][WINED3D_TSS_BUMPENV_MAT00]);
5833     mat[0][1] = *((float *)&state->texture_states[stage][WINED3D_TSS_BUMPENV_MAT01]);
5834     mat[1][0] = *((float *)&state->texture_states[stage][WINED3D_TSS_BUMPENV_MAT10]);
5835     mat[1][1] = *((float *)&state->texture_states[stage][WINED3D_TSS_BUMPENV_MAT11]);
5836
5837     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_BUMPMAT(stage), &mat[0][0]));
5838     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_BUMPMAT(stage), &mat[0][0])");
5839 }
5840
5841 static void tex_bumpenvlum_arbfp(struct wined3d_context *context,
5842         const struct wined3d_state *state, DWORD state_id)
5843 {
5844     DWORD stage = (state_id - STATE_TEXTURESTAGE(0, 0)) / (WINED3D_HIGHEST_TEXTURE_STATE + 1);
5845     struct wined3d_device *device = context->swapchain->device;
5846     const struct wined3d_gl_info *gl_info = context->gl_info;
5847     float param[4];
5848
5849     if (use_ps(state))
5850     {
5851         /* The pixel shader has to know the luminance offset. Do a constants
5852          * update. */
5853         if (stage && (state->pixel_shader->reg_maps.luminanceparams & (1 << stage)))
5854             context->load_constants = 1;
5855
5856         if(device->shader_backend == &arb_program_shader_backend) {
5857             /* Exit now, don't set the bumpmat below, otherwise we may overwrite pixel shader constants */
5858             return;
5859         }
5860     }
5861     else if (device->shader_backend == &arb_program_shader_backend)
5862     {
5863         struct shader_arb_priv *priv = device->shader_priv;
5864         priv->pshader_const_dirty[ARB_FFP_CONST_LUMINANCE(stage)] = 1;
5865         priv->highest_dirty_ps_const = max(priv->highest_dirty_ps_const, ARB_FFP_CONST_LUMINANCE(stage) + 1);
5866     }
5867
5868     param[0] = *((float *)&state->texture_states[stage][WINED3D_TSS_BUMPENV_LSCALE]);
5869     param[1] = *((float *)&state->texture_states[stage][WINED3D_TSS_BUMPENV_LOFFSET]);
5870     param[2] = 0.0f;
5871     param[3] = 0.0f;
5872
5873     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_LUMINANCE(stage), param));
5874     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_LUMINANCE(stage), param)");
5875 }
5876
5877 static const char *get_argreg(struct wined3d_shader_buffer *buffer, DWORD argnum, unsigned int stage, DWORD arg)
5878 {
5879     const char *ret;
5880
5881     if(arg == ARG_UNUSED) return "unused"; /* This is the marker for unused registers */
5882
5883     switch(arg & WINED3DTA_SELECTMASK) {
5884         case WINED3DTA_DIFFUSE:
5885             ret = "fragment.color.primary"; break;
5886
5887         case WINED3DTA_CURRENT:
5888             if (!stage) ret = "fragment.color.primary";
5889             else ret = "ret";
5890             break;
5891
5892         case WINED3DTA_TEXTURE:
5893             switch(stage) {
5894                 case 0: ret = "tex0"; break;
5895                 case 1: ret = "tex1"; break;
5896                 case 2: ret = "tex2"; break;
5897                 case 3: ret = "tex3"; break;
5898                 case 4: ret = "tex4"; break;
5899                 case 5: ret = "tex5"; break;
5900                 case 6: ret = "tex6"; break;
5901                 case 7: ret = "tex7"; break;
5902                 default: ret = "unknown texture";
5903             }
5904             break;
5905
5906         case WINED3DTA_TFACTOR:
5907             ret = "tfactor"; break;
5908
5909         case WINED3DTA_SPECULAR:
5910             ret = "fragment.color.secondary"; break;
5911
5912         case WINED3DTA_TEMP:
5913             ret = "tempreg"; break;
5914
5915         case WINED3DTA_CONSTANT:
5916             FIXME("Implement perstage constants\n");
5917             switch(stage) {
5918                 case 0: ret = "const0"; break;
5919                 case 1: ret = "const1"; break;
5920                 case 2: ret = "const2"; break;
5921                 case 3: ret = "const3"; break;
5922                 case 4: ret = "const4"; break;
5923                 case 5: ret = "const5"; break;
5924                 case 6: ret = "const6"; break;
5925                 case 7: ret = "const7"; break;
5926                 default: ret = "unknown constant";
5927             }
5928             break;
5929
5930         default:
5931             return "unknown";
5932     }
5933
5934     if(arg & WINED3DTA_COMPLEMENT) {
5935         shader_addline(buffer, "SUB arg%u, const.x, %s;\n", argnum, ret);
5936         if(argnum == 0) ret = "arg0";
5937         if(argnum == 1) ret = "arg1";
5938         if(argnum == 2) ret = "arg2";
5939     }
5940     if(arg & WINED3DTA_ALPHAREPLICATE) {
5941         shader_addline(buffer, "MOV arg%u, %s.w;\n", argnum, ret);
5942         if(argnum == 0) ret = "arg0";
5943         if(argnum == 1) ret = "arg1";
5944         if(argnum == 2) ret = "arg2";
5945     }
5946     return ret;
5947 }
5948
5949 static void gen_ffp_instr(struct wined3d_shader_buffer *buffer, unsigned int stage, BOOL color,
5950         BOOL alpha, DWORD dst, DWORD op, DWORD dw_arg0, DWORD dw_arg1, DWORD dw_arg2)
5951 {
5952     const char *dstmask, *dstreg, *arg0, *arg1, *arg2;
5953     unsigned int mul = 1;
5954
5955     if(color && alpha) dstmask = "";
5956     else if(color) dstmask = ".xyz";
5957     else dstmask = ".w";
5958
5959     if(dst == tempreg) dstreg = "tempreg";
5960     else dstreg = "ret";
5961
5962     arg0 = get_argreg(buffer, 0, stage, dw_arg0);
5963     arg1 = get_argreg(buffer, 1, stage, dw_arg1);
5964     arg2 = get_argreg(buffer, 2, stage, dw_arg2);
5965
5966     switch (op)
5967     {
5968         case WINED3D_TOP_DISABLE:
5969             if (!stage)
5970                 shader_addline(buffer, "MOV %s%s, fragment.color.primary;\n", dstreg, dstmask);
5971             break;
5972
5973         case WINED3D_TOP_SELECT_ARG2:
5974             arg1 = arg2;
5975             /* FALLTHROUGH */
5976         case WINED3D_TOP_SELECT_ARG1:
5977             shader_addline(buffer, "MOV %s%s, %s;\n", dstreg, dstmask, arg1);
5978             break;
5979
5980         case WINED3D_TOP_MODULATE_4X:
5981             mul = 2;
5982             /* FALLTHROUGH */
5983         case WINED3D_TOP_MODULATE_2X:
5984             mul *= 2;
5985             /* FALLTHROUGH */
5986         case WINED3D_TOP_MODULATE:
5987             shader_addline(buffer, "MUL %s%s, %s, %s;\n", dstreg, dstmask, arg1, arg2);
5988             break;
5989
5990         case WINED3D_TOP_ADD_SIGNED_2X:
5991             mul = 2;
5992             /* FALLTHROUGH */
5993         case WINED3D_TOP_ADD_SIGNED:
5994             shader_addline(buffer, "SUB arg2, %s, const.w;\n", arg2);
5995             arg2 = "arg2";
5996             /* FALLTHROUGH */
5997         case WINED3D_TOP_ADD:
5998             shader_addline(buffer, "ADD_SAT %s%s, %s, %s;\n", dstreg, dstmask, arg1, arg2);
5999             break;
6000
6001         case WINED3D_TOP_SUBTRACT:
6002             shader_addline(buffer, "SUB_SAT %s%s, %s, %s;\n", dstreg, dstmask, arg1, arg2);
6003             break;
6004
6005         case WINED3D_TOP_ADD_SMOOTH:
6006             shader_addline(buffer, "SUB arg1, const.x, %s;\n", arg1);
6007             shader_addline(buffer, "MAD_SAT %s%s, arg1, %s, %s;\n", dstreg, dstmask, arg2, arg1);
6008             break;
6009
6010         case WINED3D_TOP_BLEND_CURRENT_ALPHA:
6011             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_CURRENT);
6012             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
6013             break;
6014         case WINED3D_TOP_BLEND_FACTOR_ALPHA:
6015             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_TFACTOR);
6016             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
6017             break;
6018         case WINED3D_TOP_BLEND_TEXTURE_ALPHA:
6019             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_TEXTURE);
6020             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
6021             break;
6022         case WINED3D_TOP_BLEND_DIFFUSE_ALPHA:
6023             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_DIFFUSE);
6024             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
6025             break;
6026
6027         case WINED3D_TOP_BLEND_TEXTURE_ALPHA_PM:
6028             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_TEXTURE);
6029             shader_addline(buffer, "SUB arg0.w, const.x, %s.w;\n", arg0);
6030             shader_addline(buffer, "MAD_SAT %s%s, %s, arg0.w, %s;\n", dstreg, dstmask, arg2, arg1);
6031             break;
6032
6033         /* D3DTOP_PREMODULATE ???? */
6034
6035         case WINED3D_TOP_MODULATE_INVALPHA_ADD_COLOR:
6036             shader_addline(buffer, "SUB arg0.w, const.x, %s;\n", arg1);
6037             shader_addline(buffer, "MAD_SAT %s%s, arg0.w, %s, %s;\n", dstreg, dstmask, arg2, arg1);
6038             break;
6039         case WINED3D_TOP_MODULATE_ALPHA_ADD_COLOR:
6040             shader_addline(buffer, "MAD_SAT %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg1, arg2, arg1);
6041             break;
6042         case WINED3D_TOP_MODULATE_INVCOLOR_ADD_ALPHA:
6043             shader_addline(buffer, "SUB arg0, const.x, %s;\n", arg1);
6044             shader_addline(buffer, "MAD_SAT %s%s, arg0, %s, %s.w;\n", dstreg, dstmask, arg2, arg1);
6045             break;
6046         case WINED3D_TOP_MODULATE_COLOR_ADD_ALPHA:
6047             shader_addline(buffer, "MAD_SAT %s%s, %s, %s, %s.w;\n", dstreg, dstmask, arg1, arg2, arg1);
6048             break;
6049
6050         case WINED3D_TOP_DOTPRODUCT3:
6051             mul = 4;
6052             shader_addline(buffer, "SUB arg1, %s, const.w;\n", arg1);
6053             shader_addline(buffer, "SUB arg2, %s, const.w;\n", arg2);
6054             shader_addline(buffer, "DP3_SAT %s%s, arg1, arg2;\n", dstreg, dstmask);
6055             break;
6056
6057         case WINED3D_TOP_MULTIPLY_ADD:
6058             shader_addline(buffer, "MAD_SAT %s%s, %s, %s, %s;\n", dstreg, dstmask, arg1, arg2, arg0);
6059             break;
6060
6061         case WINED3D_TOP_LERP:
6062             /* The msdn is not quite right here */
6063             shader_addline(buffer, "LRP %s%s, %s, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
6064             break;
6065
6066         case WINED3D_TOP_BUMPENVMAP:
6067         case WINED3D_TOP_BUMPENVMAP_LUMINANCE:
6068             /* Those are handled in the first pass of the shader(generation pass 1 and 2) already */
6069             break;
6070
6071         default:
6072             FIXME("Unhandled texture op %08x\n", op);
6073     }
6074
6075     if (mul == 2)
6076         shader_addline(buffer, "MUL_SAT %s%s, %s, const.y;\n", dstreg, dstmask, dstreg);
6077     else if (mul == 4)
6078         shader_addline(buffer, "MUL_SAT %s%s, %s, const.z;\n", dstreg, dstmask, dstreg);
6079 }
6080
6081 static GLuint gen_arbfp_ffp_shader(const struct ffp_frag_settings *settings, const struct wined3d_gl_info *gl_info)
6082 {
6083     unsigned int stage;
6084     struct wined3d_shader_buffer buffer;
6085     BOOL tex_read[MAX_TEXTURES] = {FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE};
6086     BOOL bump_used[MAX_TEXTURES] = {FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE};
6087     BOOL luminance_used[MAX_TEXTURES] = {FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE};
6088     UINT lowest_disabled_stage;
6089     const char *textype;
6090     const char *instr, *sat;
6091     char colorcor_dst[8];
6092     GLuint ret;
6093     DWORD arg0, arg1, arg2;
6094     BOOL tempreg_used = FALSE, tfactor_used = FALSE;
6095     BOOL op_equal;
6096     const char *final_combiner_src = "ret";
6097     GLint pos;
6098
6099     /* Find out which textures are read */
6100     for (stage = 0; stage < MAX_TEXTURES; ++stage)
6101     {
6102         if (settings->op[stage].cop == WINED3D_TOP_DISABLE)
6103             break;
6104         arg0 = settings->op[stage].carg0 & WINED3DTA_SELECTMASK;
6105         arg1 = settings->op[stage].carg1 & WINED3DTA_SELECTMASK;
6106         arg2 = settings->op[stage].carg2 & WINED3DTA_SELECTMASK;
6107         if(arg0 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
6108         if(arg1 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
6109         if(arg2 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
6110
6111         if (settings->op[stage].cop == WINED3D_TOP_BLEND_TEXTURE_ALPHA)
6112             tex_read[stage] = TRUE;
6113         if (settings->op[stage].cop == WINED3D_TOP_BLEND_TEXTURE_ALPHA_PM)
6114             tex_read[stage] = TRUE;
6115         if (settings->op[stage].cop == WINED3D_TOP_BUMPENVMAP)
6116         {
6117             bump_used[stage] = TRUE;
6118             tex_read[stage] = TRUE;
6119         }
6120         if (settings->op[stage].cop == WINED3D_TOP_BUMPENVMAP_LUMINANCE)
6121         {
6122             bump_used[stage] = TRUE;
6123             tex_read[stage] = TRUE;
6124             luminance_used[stage] = TRUE;
6125         }
6126         else if (settings->op[stage].cop == WINED3D_TOP_BLEND_FACTOR_ALPHA)
6127         {
6128             tfactor_used = TRUE;
6129         }
6130
6131         if(arg0 == WINED3DTA_TFACTOR || arg1 == WINED3DTA_TFACTOR || arg2 == WINED3DTA_TFACTOR) {
6132             tfactor_used = TRUE;
6133         }
6134
6135         if(settings->op[stage].dst == tempreg) tempreg_used = TRUE;
6136         if(arg0 == WINED3DTA_TEMP || arg1 == WINED3DTA_TEMP || arg2 == WINED3DTA_TEMP) {
6137             tempreg_used = TRUE;
6138         }
6139
6140         if (settings->op[stage].aop == WINED3D_TOP_DISABLE)
6141             continue;
6142         arg0 = settings->op[stage].aarg0 & WINED3DTA_SELECTMASK;
6143         arg1 = settings->op[stage].aarg1 & WINED3DTA_SELECTMASK;
6144         arg2 = settings->op[stage].aarg2 & WINED3DTA_SELECTMASK;
6145         if(arg0 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
6146         if(arg1 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
6147         if(arg2 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
6148
6149         if(arg0 == WINED3DTA_TEMP || arg1 == WINED3DTA_TEMP || arg2 == WINED3DTA_TEMP) {
6150             tempreg_used = TRUE;
6151         }
6152         if(arg0 == WINED3DTA_TFACTOR || arg1 == WINED3DTA_TFACTOR || arg2 == WINED3DTA_TFACTOR) {
6153             tfactor_used = TRUE;
6154         }
6155     }
6156     lowest_disabled_stage = stage;
6157
6158     /* Shader header */
6159     if (!shader_buffer_init(&buffer))
6160     {
6161         ERR("Failed to initialize shader buffer.\n");
6162         return 0;
6163     }
6164
6165     shader_addline(&buffer, "!!ARBfp1.0\n");
6166
6167     switch(settings->fog) {
6168         case FOG_OFF:                                                         break;
6169         case FOG_LINEAR: shader_addline(&buffer, "OPTION ARB_fog_linear;\n"); break;
6170         case FOG_EXP:    shader_addline(&buffer, "OPTION ARB_fog_exp;\n");    break;
6171         case FOG_EXP2:   shader_addline(&buffer, "OPTION ARB_fog_exp2;\n");   break;
6172         default: FIXME("Unexpected fog setting %d\n", settings->fog);
6173     }
6174
6175     shader_addline(&buffer, "PARAM const = {1, 2, 4, 0.5};\n");
6176     shader_addline(&buffer, "TEMP TMP;\n");
6177     shader_addline(&buffer, "TEMP ret;\n");
6178     if(tempreg_used || settings->sRGB_write) shader_addline(&buffer, "TEMP tempreg;\n");
6179     shader_addline(&buffer, "TEMP arg0;\n");
6180     shader_addline(&buffer, "TEMP arg1;\n");
6181     shader_addline(&buffer, "TEMP arg2;\n");
6182     for(stage = 0; stage < MAX_TEXTURES; stage++) {
6183         if(!tex_read[stage]) continue;
6184         shader_addline(&buffer, "TEMP tex%u;\n", stage);
6185         if(!bump_used[stage]) continue;
6186         shader_addline(&buffer, "PARAM bumpmat%u = program.env[%u];\n", stage, ARB_FFP_CONST_BUMPMAT(stage));
6187         if(!luminance_used[stage]) continue;
6188         shader_addline(&buffer, "PARAM luminance%u = program.env[%u];\n", stage, ARB_FFP_CONST_LUMINANCE(stage));
6189     }
6190     if(tfactor_used) {
6191         shader_addline(&buffer, "PARAM tfactor = program.env[%u];\n", ARB_FFP_CONST_TFACTOR);
6192     }
6193     shader_addline(&buffer, "PARAM specular_enable = program.env[%u];\n", ARB_FFP_CONST_SPECULAR_ENABLE);
6194
6195     if(settings->sRGB_write) {
6196         shader_addline(&buffer, "PARAM srgb_consts1 = {%f, %f, %f, %f};\n",
6197                        srgb_mul_low, srgb_cmp, srgb_pow, srgb_mul_high);
6198         shader_addline(&buffer, "PARAM srgb_consts2 = {%f, %f, %f, %f};\n",
6199                        srgb_sub_high, 0.0, 0.0, 0.0);
6200     }
6201
6202     if (lowest_disabled_stage < 7 && settings->emul_clipplanes)
6203         shader_addline(&buffer, "KIL fragment.texcoord[7];\n");
6204
6205     /* Generate texture sampling instructions) */
6206     for (stage = 0; stage < MAX_TEXTURES && settings->op[stage].cop != WINED3D_TOP_DISABLE; ++stage)
6207     {
6208         if (!tex_read[stage])
6209             continue;
6210
6211         switch(settings->op[stage].tex_type) {
6212             case tex_1d:                    textype = "1D";     break;
6213             case tex_2d:                    textype = "2D";     break;
6214             case tex_3d:                    textype = "3D";     break;
6215             case tex_cube:                  textype = "CUBE";   break;
6216             case tex_rect:                  textype = "RECT";   break;
6217             default: textype = "unexpected_textype";   break;
6218         }
6219
6220         if (settings->op[stage].cop == WINED3D_TOP_BUMPENVMAP
6221                 || settings->op[stage].cop == WINED3D_TOP_BUMPENVMAP_LUMINANCE)
6222             sat = "";
6223         else
6224             sat = "_SAT";
6225
6226         if(settings->op[stage].projected == proj_none) {
6227             instr = "TEX";
6228         } else if(settings->op[stage].projected == proj_count4 ||
6229                   settings->op[stage].projected == proj_count3) {
6230             instr = "TXP";
6231         } else {
6232             FIXME("Unexpected projection mode %d\n", settings->op[stage].projected);
6233             instr = "TXP";
6234         }
6235
6236         if (stage > 0
6237                 && (settings->op[stage - 1].cop == WINED3D_TOP_BUMPENVMAP
6238                 || settings->op[stage - 1].cop == WINED3D_TOP_BUMPENVMAP_LUMINANCE))
6239         {
6240             shader_addline(&buffer, "SWZ arg1, bumpmat%u, x, z, 0, 0;\n", stage - 1);
6241             shader_addline(&buffer, "DP3 ret.x, arg1, tex%u;\n", stage - 1);
6242             shader_addline(&buffer, "SWZ arg1, bumpmat%u, y, w, 0, 0;\n", stage - 1);
6243             shader_addline(&buffer, "DP3 ret.y, arg1, tex%u;\n", stage - 1);
6244
6245             /* with projective textures, texbem only divides the static texture coord, not the displacement,
6246              * so multiply the displacement with the dividing parameter before passing it to TXP
6247              */
6248             if (settings->op[stage].projected != proj_none) {
6249                 if(settings->op[stage].projected == proj_count4) {
6250                     shader_addline(&buffer, "MOV ret.w, fragment.texcoord[%u].w;\n", stage);
6251                     shader_addline(&buffer, "MUL ret.xyz, ret, fragment.texcoord[%u].w, fragment.texcoord[%u];\n", stage, stage);
6252                 } else {
6253                     shader_addline(&buffer, "MOV ret.w, fragment.texcoord[%u].z;\n", stage);
6254                     shader_addline(&buffer, "MAD ret.xyz, ret, fragment.texcoord[%u].z, fragment.texcoord[%u];\n", stage, stage);
6255                 }
6256             } else {
6257                 shader_addline(&buffer, "ADD ret, ret, fragment.texcoord[%u];\n", stage);
6258             }
6259
6260             shader_addline(&buffer, "%s%s tex%u, ret, texture[%u], %s;\n",
6261                     instr, sat, stage, stage, textype);
6262             if (settings->op[stage - 1].cop == WINED3D_TOP_BUMPENVMAP_LUMINANCE)
6263             {
6264                 shader_addline(&buffer, "MAD_SAT ret.x, tex%u.z, luminance%u.x, luminance%u.y;\n",
6265                                stage - 1, stage - 1, stage - 1);
6266                 shader_addline(&buffer, "MUL tex%u, tex%u, ret.x;\n", stage, stage);
6267             }
6268         } else if(settings->op[stage].projected == proj_count3) {
6269             shader_addline(&buffer, "MOV ret, fragment.texcoord[%u];\n", stage);
6270             shader_addline(&buffer, "MOV ret.w, ret.z;\n");
6271             shader_addline(&buffer, "%s%s tex%u, ret, texture[%u], %s;\n",
6272                             instr, sat, stage, stage, textype);
6273         } else {
6274             shader_addline(&buffer, "%s%s tex%u, fragment.texcoord[%u], texture[%u], %s;\n",
6275                             instr, sat, stage, stage, stage, textype);
6276         }
6277
6278         sprintf(colorcor_dst, "tex%u", stage);
6279         gen_color_correction(&buffer, colorcor_dst, WINED3DSP_WRITEMASK_ALL, "const.x", "const.y",
6280                 settings->op[stage].color_fixup);
6281     }
6282
6283     /* Generate the main shader */
6284     for (stage = 0; stage < MAX_TEXTURES; ++stage)
6285     {
6286         if (settings->op[stage].cop == WINED3D_TOP_DISABLE)
6287         {
6288             if (!stage)
6289                 final_combiner_src = "fragment.color.primary";
6290             break;
6291         }
6292
6293         if (settings->op[stage].cop == WINED3D_TOP_SELECT_ARG1
6294                 && settings->op[stage].aop == WINED3D_TOP_SELECT_ARG1)
6295             op_equal = settings->op[stage].carg1 == settings->op[stage].aarg1;
6296         else if (settings->op[stage].cop == WINED3D_TOP_SELECT_ARG1
6297                 && settings->op[stage].aop == WINED3D_TOP_SELECT_ARG2)
6298             op_equal = settings->op[stage].carg1 == settings->op[stage].aarg2;
6299         else if (settings->op[stage].cop == WINED3D_TOP_SELECT_ARG2
6300                 && settings->op[stage].aop == WINED3D_TOP_SELECT_ARG1)
6301             op_equal = settings->op[stage].carg2 == settings->op[stage].aarg1;
6302         else if (settings->op[stage].cop == WINED3D_TOP_SELECT_ARG2
6303                 && settings->op[stage].aop == WINED3D_TOP_SELECT_ARG2)
6304             op_equal = settings->op[stage].carg2 == settings->op[stage].aarg2;
6305         else
6306             op_equal = settings->op[stage].aop   == settings->op[stage].cop
6307                     && settings->op[stage].carg0 == settings->op[stage].aarg0
6308                     && settings->op[stage].carg1 == settings->op[stage].aarg1
6309                     && settings->op[stage].carg2 == settings->op[stage].aarg2;
6310
6311         if (settings->op[stage].aop == WINED3D_TOP_DISABLE)
6312         {
6313             gen_ffp_instr(&buffer, stage, TRUE, FALSE, settings->op[stage].dst,
6314                           settings->op[stage].cop, settings->op[stage].carg0,
6315                           settings->op[stage].carg1, settings->op[stage].carg2);
6316             if (!stage)
6317                 shader_addline(&buffer, "MOV ret.w, fragment.color.primary.w;\n");
6318         }
6319         else if (op_equal)
6320         {
6321             gen_ffp_instr(&buffer, stage, TRUE, TRUE, settings->op[stage].dst,
6322                           settings->op[stage].cop, settings->op[stage].carg0,
6323                           settings->op[stage].carg1, settings->op[stage].carg2);
6324         } else {
6325             gen_ffp_instr(&buffer, stage, TRUE, FALSE, settings->op[stage].dst,
6326                           settings->op[stage].cop, settings->op[stage].carg0,
6327                           settings->op[stage].carg1, settings->op[stage].carg2);
6328             gen_ffp_instr(&buffer, stage, FALSE, TRUE, settings->op[stage].dst,
6329                           settings->op[stage].aop, settings->op[stage].aarg0,
6330                           settings->op[stage].aarg1, settings->op[stage].aarg2);
6331         }
6332     }
6333
6334     if(settings->sRGB_write) {
6335         shader_addline(&buffer, "MAD ret, fragment.color.secondary, specular_enable, %s;\n", final_combiner_src);
6336         arbfp_add_sRGB_correction(&buffer, "ret", "arg0", "arg1", "arg2", "tempreg", FALSE);
6337         shader_addline(&buffer, "MOV result.color, ret;\n");
6338     } else {
6339         shader_addline(&buffer, "MAD result.color, fragment.color.secondary, specular_enable, %s;\n", final_combiner_src);
6340     }
6341
6342     /* Footer */
6343     shader_addline(&buffer, "END\n");
6344
6345     /* Generate the shader */
6346     GL_EXTCALL(glGenProgramsARB(1, &ret));
6347     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, ret));
6348     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
6349             strlen(buffer.buffer), buffer.buffer));
6350     checkGLcall("glProgramStringARB()");
6351
6352     gl_info->gl_ops.gl.p_glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
6353     if (pos != -1)
6354     {
6355         FIXME("Fragment program error at position %d: %s\n\n", pos,
6356               debugstr_a((const char *)gl_info->gl_ops.gl.p_glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
6357         shader_arb_dump_program_source(buffer.buffer);
6358     }
6359     else
6360     {
6361         GLint native;
6362
6363         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
6364         checkGLcall("glGetProgramivARB()");
6365         if (!native) WARN("Program exceeds native resource limits.\n");
6366     }
6367
6368     shader_buffer_free(&buffer);
6369     return ret;
6370 }
6371
6372 static void fragment_prog_arbfp(struct wined3d_context *context, const struct wined3d_state *state, DWORD state_id)
6373 {
6374     const struct wined3d_device *device = context->swapchain->device;
6375     const struct wined3d_gl_info *gl_info = context->gl_info;
6376     struct shader_arb_priv *priv = device->fragment_priv;
6377     BOOL use_pshader = use_ps(state);
6378     struct ffp_frag_settings settings;
6379     const struct arbfp_ffp_desc *desc;
6380     unsigned int i;
6381
6382     TRACE("context %p, state %p, state_id %#x.\n", context, state, state_id);
6383
6384     if (isStateDirty(context, STATE_RENDER(WINED3D_RS_FOGENABLE)))
6385     {
6386         if (!use_pshader && device->shader_backend == &arb_program_shader_backend && context->last_was_pshader)
6387         {
6388             /* Reload fixed function constants since they collide with the
6389              * pixel shader constants. */
6390             for (i = 0; i < MAX_TEXTURES; ++i)
6391             {
6392                 set_bumpmat_arbfp(context, state, STATE_TEXTURESTAGE(i, WINED3D_TSS_BUMPENV_MAT00));
6393             }
6394             state_texfactor_arbfp(context, state, STATE_RENDER(WINED3D_RS_TEXTUREFACTOR));
6395             state_arb_specularenable(context, state, STATE_RENDER(WINED3D_RS_SPECULARENABLE));
6396         }
6397         else if (use_pshader)
6398         {
6399             context->select_shader = 1;
6400         }
6401         return;
6402     }
6403
6404     if (!use_pshader)
6405     {
6406         /* Find or create a shader implementing the fixed function pipeline
6407          * settings, then activate it. */
6408         gen_ffp_frag_op(device, state, &settings, FALSE);
6409         desc = (const struct arbfp_ffp_desc *)find_ffp_frag_shader(&priv->fragment_shaders, &settings);
6410         if(!desc) {
6411             struct arbfp_ffp_desc *new_desc = HeapAlloc(GetProcessHeap(), 0, sizeof(*new_desc));
6412             if (!new_desc)
6413             {
6414                 ERR("Out of memory\n");
6415                 return;
6416             }
6417
6418             new_desc->parent.settings = settings;
6419             new_desc->shader = gen_arbfp_ffp_shader(&settings, gl_info);
6420             add_ffp_frag_shader(&priv->fragment_shaders, &new_desc->parent);
6421             TRACE("Allocated fixed function replacement shader descriptor %p\n", new_desc);
6422             desc = new_desc;
6423         }
6424
6425         /* Now activate the replacement program. GL_FRAGMENT_PROGRAM_ARB is already active (however, note the
6426          * comment above the shader_select call below). If e.g. GLSL is active, the shader_select call will
6427          * deactivate it.
6428          */
6429         GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, desc->shader));
6430         checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, desc->shader)");
6431         priv->current_fprogram_id = desc->shader;
6432
6433         if (device->shader_backend == &arb_program_shader_backend && context->last_was_pshader)
6434         {
6435             /* Reload fixed function constants since they collide with the
6436              * pixel shader constants. */
6437             for (i = 0; i < MAX_TEXTURES; ++i)
6438             {
6439                 set_bumpmat_arbfp(context, state, STATE_TEXTURESTAGE(i, WINED3D_TSS_BUMPENV_MAT00));
6440             }
6441             state_texfactor_arbfp(context, state, STATE_RENDER(WINED3D_RS_TEXTUREFACTOR));
6442             state_arb_specularenable(context, state, STATE_RENDER(WINED3D_RS_SPECULARENABLE));
6443         }
6444         context->last_was_pshader = FALSE;
6445     } else {
6446         context->last_was_pshader = TRUE;
6447     }
6448
6449     context->select_shader = 1;
6450     context->load_constants = 1;
6451 }
6452
6453 /* We can't link the fog states to the fragment state directly since the
6454  * vertex pipeline links them to FOGENABLE. A different linking in different
6455  * pipeline parts can't be expressed in the combined state table, so we need
6456  * to handle that with a forwarding function. The other invisible side effect
6457  * is that changing the fog start and fog end (which links to FOGENABLE in
6458  * vertex) results in the fragment_prog_arbfp function being called because
6459  * FOGENABLE is dirty, which calls this function here. */
6460 static void state_arbfp_fog(struct wined3d_context *context, const struct wined3d_state *state, DWORD state_id)
6461 {
6462     enum fogsource new_source;
6463
6464     TRACE("context %p, state %p, state_id %#x.\n", context, state, state_id);
6465
6466     if (!isStateDirty(context, STATE_PIXELSHADER))
6467         fragment_prog_arbfp(context, state, state_id);
6468
6469     if (!state->render_states[WINED3D_RS_FOGENABLE])
6470         return;
6471
6472     if (state->render_states[WINED3D_RS_FOGTABLEMODE] == WINED3D_FOG_NONE)
6473     {
6474         if (use_vs(state))
6475         {
6476             new_source = FOGSOURCE_VS;
6477         }
6478         else
6479         {
6480             if (state->render_states[WINED3D_RS_FOGVERTEXMODE] == WINED3D_FOG_NONE || context->last_was_rhw)
6481                 new_source = FOGSOURCE_COORD;
6482             else
6483                 new_source = FOGSOURCE_FFP;
6484         }
6485     }
6486     else
6487     {
6488         new_source = FOGSOURCE_FFP;
6489     }
6490
6491     if (new_source != context->fog_source)
6492     {
6493         context->fog_source = new_source;
6494         state_fogstartend(context, state, STATE_RENDER(WINED3D_RS_FOGSTART));
6495     }
6496 }
6497
6498 static void textransform(struct wined3d_context *context, const struct wined3d_state *state, DWORD state_id)
6499 {
6500     if (!isStateDirty(context, STATE_PIXELSHADER))
6501         fragment_prog_arbfp(context, state, state_id);
6502 }
6503
6504 static const struct StateEntryTemplate arbfp_fragmentstate_template[] =
6505 {
6506     {STATE_RENDER(WINED3D_RS_TEXTUREFACTOR),              { STATE_RENDER(WINED3D_RS_TEXTUREFACTOR),             state_texfactor_arbfp   }, WINED3D_GL_EXT_NONE             },
6507     {STATE_TEXTURESTAGE(0, WINED3D_TSS_COLOR_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6508     {STATE_TEXTURESTAGE(0, WINED3D_TSS_COLOR_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6509     {STATE_TEXTURESTAGE(0, WINED3D_TSS_COLOR_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6510     {STATE_TEXTURESTAGE(0, WINED3D_TSS_COLOR_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6511     {STATE_TEXTURESTAGE(0, WINED3D_TSS_ALPHA_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6512     {STATE_TEXTURESTAGE(0, WINED3D_TSS_ALPHA_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6513     {STATE_TEXTURESTAGE(0, WINED3D_TSS_ALPHA_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6514     {STATE_TEXTURESTAGE(0, WINED3D_TSS_ALPHA_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6515     {STATE_TEXTURESTAGE(0, WINED3D_TSS_RESULT_ARG),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6516     {STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6517     {STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6518     {STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6519     {STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6520     {STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6521     {STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6522     {STATE_TEXTURESTAGE(1, WINED3D_TSS_COLOR_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6523     {STATE_TEXTURESTAGE(1, WINED3D_TSS_COLOR_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6524     {STATE_TEXTURESTAGE(1, WINED3D_TSS_COLOR_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6525     {STATE_TEXTURESTAGE(1, WINED3D_TSS_COLOR_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6526     {STATE_TEXTURESTAGE(1, WINED3D_TSS_ALPHA_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6527     {STATE_TEXTURESTAGE(1, WINED3D_TSS_ALPHA_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6528     {STATE_TEXTURESTAGE(1, WINED3D_TSS_ALPHA_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6529     {STATE_TEXTURESTAGE(1, WINED3D_TSS_ALPHA_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6530     {STATE_TEXTURESTAGE(1, WINED3D_TSS_RESULT_ARG),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6531     {STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6532     {STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6533     {STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6534     {STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6535     {STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6536     {STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6537     {STATE_TEXTURESTAGE(2, WINED3D_TSS_COLOR_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6538     {STATE_TEXTURESTAGE(2, WINED3D_TSS_COLOR_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6539     {STATE_TEXTURESTAGE(2, WINED3D_TSS_COLOR_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6540     {STATE_TEXTURESTAGE(2, WINED3D_TSS_COLOR_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6541     {STATE_TEXTURESTAGE(2, WINED3D_TSS_ALPHA_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6542     {STATE_TEXTURESTAGE(2, WINED3D_TSS_ALPHA_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6543     {STATE_TEXTURESTAGE(2, WINED3D_TSS_ALPHA_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6544     {STATE_TEXTURESTAGE(2, WINED3D_TSS_ALPHA_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6545     {STATE_TEXTURESTAGE(2, WINED3D_TSS_RESULT_ARG),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6546     {STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6547     {STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6548     {STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6549     {STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6550     {STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6551     {STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6552     {STATE_TEXTURESTAGE(3, WINED3D_TSS_COLOR_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6553     {STATE_TEXTURESTAGE(3, WINED3D_TSS_COLOR_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6554     {STATE_TEXTURESTAGE(3, WINED3D_TSS_COLOR_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6555     {STATE_TEXTURESTAGE(3, WINED3D_TSS_COLOR_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6556     {STATE_TEXTURESTAGE(3, WINED3D_TSS_ALPHA_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6557     {STATE_TEXTURESTAGE(3, WINED3D_TSS_ALPHA_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6558     {STATE_TEXTURESTAGE(3, WINED3D_TSS_ALPHA_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6559     {STATE_TEXTURESTAGE(3, WINED3D_TSS_ALPHA_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6560     {STATE_TEXTURESTAGE(3, WINED3D_TSS_RESULT_ARG),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6561     {STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6562     {STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6563     {STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6564     {STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6565     {STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6566     {STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6567     {STATE_TEXTURESTAGE(4, WINED3D_TSS_COLOR_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6568     {STATE_TEXTURESTAGE(4, WINED3D_TSS_COLOR_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6569     {STATE_TEXTURESTAGE(4, WINED3D_TSS_COLOR_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6570     {STATE_TEXTURESTAGE(4, WINED3D_TSS_COLOR_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6571     {STATE_TEXTURESTAGE(4, WINED3D_TSS_ALPHA_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6572     {STATE_TEXTURESTAGE(4, WINED3D_TSS_ALPHA_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6573     {STATE_TEXTURESTAGE(4, WINED3D_TSS_ALPHA_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6574     {STATE_TEXTURESTAGE(4, WINED3D_TSS_ALPHA_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6575     {STATE_TEXTURESTAGE(4, WINED3D_TSS_RESULT_ARG),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6576     {STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6577     {STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6578     {STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6579     {STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6580     {STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6581     {STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6582     {STATE_TEXTURESTAGE(5, WINED3D_TSS_COLOR_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6583     {STATE_TEXTURESTAGE(5, WINED3D_TSS_COLOR_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6584     {STATE_TEXTURESTAGE(5, WINED3D_TSS_COLOR_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6585     {STATE_TEXTURESTAGE(5, WINED3D_TSS_COLOR_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6586     {STATE_TEXTURESTAGE(5, WINED3D_TSS_ALPHA_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6587     {STATE_TEXTURESTAGE(5, WINED3D_TSS_ALPHA_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6588     {STATE_TEXTURESTAGE(5, WINED3D_TSS_ALPHA_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6589     {STATE_TEXTURESTAGE(5, WINED3D_TSS_ALPHA_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6590     {STATE_TEXTURESTAGE(5, WINED3D_TSS_RESULT_ARG),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6591     {STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6592     {STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6593     {STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6594     {STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6595     {STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6596     {STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6597     {STATE_TEXTURESTAGE(6, WINED3D_TSS_COLOR_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6598     {STATE_TEXTURESTAGE(6, WINED3D_TSS_COLOR_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6599     {STATE_TEXTURESTAGE(6, WINED3D_TSS_COLOR_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6600     {STATE_TEXTURESTAGE(6, WINED3D_TSS_COLOR_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6601     {STATE_TEXTURESTAGE(6, WINED3D_TSS_ALPHA_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6602     {STATE_TEXTURESTAGE(6, WINED3D_TSS_ALPHA_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6603     {STATE_TEXTURESTAGE(6, WINED3D_TSS_ALPHA_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6604     {STATE_TEXTURESTAGE(6, WINED3D_TSS_ALPHA_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6605     {STATE_TEXTURESTAGE(6, WINED3D_TSS_RESULT_ARG),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6606     {STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6607     {STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6608     {STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6609     {STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6610     {STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6611     {STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6612     {STATE_TEXTURESTAGE(7, WINED3D_TSS_COLOR_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6613     {STATE_TEXTURESTAGE(7, WINED3D_TSS_COLOR_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6614     {STATE_TEXTURESTAGE(7, WINED3D_TSS_COLOR_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6615     {STATE_TEXTURESTAGE(7, WINED3D_TSS_COLOR_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6616     {STATE_TEXTURESTAGE(7, WINED3D_TSS_ALPHA_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6617     {STATE_TEXTURESTAGE(7, WINED3D_TSS_ALPHA_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6618     {STATE_TEXTURESTAGE(7, WINED3D_TSS_ALPHA_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6619     {STATE_TEXTURESTAGE(7, WINED3D_TSS_ALPHA_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6620     {STATE_TEXTURESTAGE(7, WINED3D_TSS_RESULT_ARG),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6621     {STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6622     {STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6623     {STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6624     {STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6625     {STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6626     {STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6627     {STATE_PIXELSHADER,                                   { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6628     {STATE_RENDER(WINED3D_RS_FOGENABLE),                  { STATE_RENDER(WINED3D_RS_FOGENABLE),                 state_arbfp_fog         }, WINED3D_GL_EXT_NONE             },
6629     {STATE_RENDER(WINED3D_RS_FOGTABLEMODE),               { STATE_RENDER(WINED3D_RS_FOGENABLE),                 NULL                    }, WINED3D_GL_EXT_NONE             },
6630     {STATE_RENDER(WINED3D_RS_FOGVERTEXMODE),              { STATE_RENDER(WINED3D_RS_FOGENABLE),                 NULL                    }, WINED3D_GL_EXT_NONE             },
6631     {STATE_RENDER(WINED3D_RS_FOGSTART),                   { STATE_RENDER(WINED3D_RS_FOGSTART),                  state_fogstartend       }, WINED3D_GL_EXT_NONE             },
6632     {STATE_RENDER(WINED3D_RS_FOGEND),                     { STATE_RENDER(WINED3D_RS_FOGSTART),                  NULL                    }, WINED3D_GL_EXT_NONE             },
6633     {STATE_RENDER(WINED3D_RS_SRGBWRITEENABLE),            { STATE_RENDER(WINED3D_RS_SRGBWRITEENABLE),           state_srgbwrite         }, ARB_FRAMEBUFFER_SRGB            },
6634     {STATE_RENDER(WINED3D_RS_SRGBWRITEENABLE),            { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6635     {STATE_RENDER(WINED3D_RS_FOGCOLOR),                   { STATE_RENDER(WINED3D_RS_FOGCOLOR),                  state_fogcolor          }, WINED3D_GL_EXT_NONE             },
6636     {STATE_RENDER(WINED3D_RS_FOGDENSITY),                 { STATE_RENDER(WINED3D_RS_FOGDENSITY),                state_fogdensity        }, WINED3D_GL_EXT_NONE             },
6637     {STATE_TEXTURESTAGE(0,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(0, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6638     {STATE_TEXTURESTAGE(1,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(1, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6639     {STATE_TEXTURESTAGE(2,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(2, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6640     {STATE_TEXTURESTAGE(3,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(3, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6641     {STATE_TEXTURESTAGE(4,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(4, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6642     {STATE_TEXTURESTAGE(5,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(5, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6643     {STATE_TEXTURESTAGE(6,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(6, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6644     {STATE_TEXTURESTAGE(7,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(7, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6645     {STATE_RENDER(WINED3D_RS_SPECULARENABLE),             { STATE_RENDER(WINED3D_RS_SPECULARENABLE),            state_arb_specularenable}, WINED3D_GL_EXT_NONE             },
6646     {0 /* Terminate */,                                   { 0,                                                  0                       }, WINED3D_GL_EXT_NONE             },
6647 };
6648
6649 const struct fragment_pipeline arbfp_fragment_pipeline = {
6650     arbfp_enable,
6651     arbfp_get_caps,
6652     arbfp_alloc,
6653     arbfp_free,
6654     shader_arb_color_fixup_supported,
6655     arbfp_fragmentstate_template,
6656 };
6657
6658 struct arbfp_blit_priv {
6659     GLenum yuy2_rect_shader, yuy2_2d_shader;
6660     GLenum uyvy_rect_shader, uyvy_2d_shader;
6661     GLenum yv12_rect_shader, yv12_2d_shader;
6662     GLenum p8_rect_shader, p8_2d_shader;
6663     GLuint palette_texture;
6664 };
6665
6666 static HRESULT arbfp_blit_alloc(struct wined3d_device *device)
6667 {
6668     device->blit_priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(struct arbfp_blit_priv));
6669     if(!device->blit_priv) {
6670         ERR("Out of memory\n");
6671         return E_OUTOFMEMORY;
6672     }
6673     return WINED3D_OK;
6674 }
6675
6676 /* Context activation is done by the caller. */
6677 static void arbfp_blit_free(struct wined3d_device *device)
6678 {
6679     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
6680     struct arbfp_blit_priv *priv = device->blit_priv;
6681
6682     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yuy2_rect_shader));
6683     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yuy2_2d_shader));
6684     GL_EXTCALL(glDeleteProgramsARB(1, &priv->uyvy_rect_shader));
6685     GL_EXTCALL(glDeleteProgramsARB(1, &priv->uyvy_2d_shader));
6686     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yv12_rect_shader));
6687     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yv12_2d_shader));
6688     GL_EXTCALL(glDeleteProgramsARB(1, &priv->p8_rect_shader));
6689     GL_EXTCALL(glDeleteProgramsARB(1, &priv->p8_2d_shader));
6690     checkGLcall("Delete yuv and p8 programs");
6691
6692     if (priv->palette_texture)
6693         gl_info->gl_ops.gl.p_glDeleteTextures(1, &priv->palette_texture);
6694
6695     HeapFree(GetProcessHeap(), 0, device->blit_priv);
6696     device->blit_priv = NULL;
6697 }
6698
6699 static BOOL gen_planar_yuv_read(struct wined3d_shader_buffer *buffer, enum complex_fixup fixup,
6700         GLenum textype, char *luminance)
6701 {
6702     char chroma;
6703     const char *tex, *texinstr;
6704
6705     if (fixup == COMPLEX_FIXUP_UYVY) {
6706         chroma = 'x';
6707         *luminance = 'w';
6708     } else {
6709         chroma = 'w';
6710         *luminance = 'x';
6711     }
6712     switch(textype) {
6713         case GL_TEXTURE_2D:             tex = "2D";     texinstr = "TXP"; break;
6714         case GL_TEXTURE_RECTANGLE_ARB:  tex = "RECT";   texinstr = "TEX"; break;
6715         default:
6716             /* This is more tricky than just replacing the texture type - we have to navigate
6717              * properly in the texture to find the correct chroma values
6718              */
6719             FIXME("Implement yuv correction for non-2d, non-rect textures\n");
6720             return FALSE;
6721     }
6722
6723     /* First we have to read the chroma values. This means we need at least two pixels(no filtering),
6724      * or 4 pixels(with filtering). To get the unmodified chromas, we have to rid ourselves of the
6725      * filtering when we sample the texture.
6726      *
6727      * These are the rules for reading the chroma:
6728      *
6729      * Even pixel: Cr
6730      * Even pixel: U
6731      * Odd pixel: V
6732      *
6733      * So we have to get the sampling x position in non-normalized coordinates in integers
6734      */
6735     if(textype != GL_TEXTURE_RECTANGLE_ARB) {
6736         shader_addline(buffer, "MUL texcrd.xy, fragment.texcoord[0], size.x;\n");
6737         shader_addline(buffer, "MOV texcrd.w, size.x;\n");
6738     } else {
6739         shader_addline(buffer, "MOV texcrd, fragment.texcoord[0];\n");
6740     }
6741     /* We must not allow filtering between pixel x and x+1, this would mix U and V
6742      * Vertical filtering is ok. However, bear in mind that the pixel center is at
6743      * 0.5, so add 0.5.
6744      */
6745     shader_addline(buffer, "FLR texcrd.x, texcrd.x;\n");
6746     shader_addline(buffer, "ADD texcrd.x, texcrd.x, coef.y;\n");
6747
6748     /* Divide the x coordinate by 0.5 and get the fraction. This gives 0.25 and 0.75 for the
6749      * even and odd pixels respectively
6750      */
6751     shader_addline(buffer, "MUL texcrd2, texcrd, coef.y;\n");
6752     shader_addline(buffer, "FRC texcrd2, texcrd2;\n");
6753
6754     /* Sample Pixel 1 */
6755     shader_addline(buffer, "%s luminance, texcrd, texture[0], %s;\n", texinstr, tex);
6756
6757     /* Put the value into either of the chroma values */
6758     shader_addline(buffer, "SGE temp.x, texcrd2.x, coef.y;\n");
6759     shader_addline(buffer, "MUL chroma.x, luminance.%c, temp.x;\n", chroma);
6760     shader_addline(buffer, "SLT temp.x, texcrd2.x, coef.y;\n");
6761     shader_addline(buffer, "MUL chroma.y, luminance.%c, temp.x;\n", chroma);
6762
6763     /* Sample pixel 2. If we read an even pixel(SLT above returned 1), sample
6764      * the pixel right to the current one. Otherwise, sample the left pixel.
6765      * Bias and scale the SLT result to -1;1 and add it to the texcrd.x.
6766      */
6767     shader_addline(buffer, "MAD temp.x, temp.x, coef.z, -coef.x;\n");
6768     shader_addline(buffer, "ADD texcrd.x, texcrd, temp.x;\n");
6769     shader_addline(buffer, "%s luminance, texcrd, texture[0], %s;\n", texinstr, tex);
6770
6771     /* Put the value into the other chroma */
6772     shader_addline(buffer, "SGE temp.x, texcrd2.x, coef.y;\n");
6773     shader_addline(buffer, "MAD chroma.y, luminance.%c, temp.x, chroma.y;\n", chroma);
6774     shader_addline(buffer, "SLT temp.x, texcrd2.x, coef.y;\n");
6775     shader_addline(buffer, "MAD chroma.x, luminance.%c, temp.x, chroma.x;\n", chroma);
6776
6777     /* TODO: If filtering is enabled, sample a 2nd pair of pixels left or right of
6778      * the current one and lerp the two U and V values
6779      */
6780
6781     /* This gives the correctly filtered luminance value */
6782     shader_addline(buffer, "TEX luminance, fragment.texcoord[0], texture[0], %s;\n", tex);
6783
6784     return TRUE;
6785 }
6786
6787 static BOOL gen_yv12_read(struct wined3d_shader_buffer *buffer, GLenum textype, char *luminance)
6788 {
6789     const char *tex;
6790
6791     switch(textype) {
6792         case GL_TEXTURE_2D:             tex = "2D";     break;
6793         case GL_TEXTURE_RECTANGLE_ARB:  tex = "RECT";   break;
6794         default:
6795             FIXME("Implement yv12 correction for non-2d, non-rect textures\n");
6796             return FALSE;
6797     }
6798
6799     /* YV12 surfaces contain a WxH sized luminance plane, followed by a (W/2)x(H/2)
6800      * V and a (W/2)x(H/2) U plane, each with 8 bit per pixel. So the effective
6801      * bitdepth is 12 bits per pixel. Since the U and V planes have only half the
6802      * pitch of the luminance plane, the packing into the gl texture is a bit
6803      * unfortunate. If the whole texture is interpreted as luminance data it looks
6804      * approximately like this:
6805      *
6806      *        +----------------------------------+----
6807      *        |                                  |
6808      *        |                                  |
6809      *        |                                  |
6810      *        |                                  |
6811      *        |                                  |   2
6812      *        |            LUMINANCE             |   -
6813      *        |                                  |   3
6814      *        |                                  |
6815      *        |                                  |
6816      *        |                                  |
6817      *        |                                  |
6818      *        +----------------+-----------------+----
6819      *        |                |                 |
6820      *        |  U even rows   |  U odd rows     |
6821      *        |                |                 |   1
6822      *        +----------------+------------------   -
6823      *        |                |                 |   3
6824      *        |  V even rows   |  V odd rows     |
6825      *        |                |                 |
6826      *        +----------------+-----------------+----
6827      *        |                |                 |
6828      *        |     0.5        |       0.5       |
6829      *
6830      * So it appears as if there are 4 chroma images, but in fact the odd rows
6831      * in the chroma images are in the same row as the even ones. So its is
6832      * kinda tricky to read
6833      *
6834      * When reading from rectangle textures, keep in mind that the input y coordinates
6835      * go from 0 to d3d_height, whereas the opengl texture height is 1.5 * d3d_height
6836      */
6837     shader_addline(buffer, "PARAM yv12_coef = {%f, %f, %f, %f};\n",
6838             2.0f / 3.0f, 1.0f / 6.0f, (2.0f / 3.0f) + (1.0f / 6.0f), 1.0f / 3.0f);
6839
6840     shader_addline(buffer, "MOV texcrd, fragment.texcoord[0];\n");
6841     /* the chroma planes have only half the width */
6842     shader_addline(buffer, "MUL texcrd.x, texcrd.x, coef.y;\n");
6843
6844     /* The first value is between 2/3 and 5/6th of the texture's height, so scale+bias
6845      * the coordinate. Also read the right side of the image when reading odd lines
6846      *
6847      * Don't forget to clamp the y values in into the range, otherwise we'll get filtering
6848      * bleeding
6849      */
6850     if(textype == GL_TEXTURE_2D) {
6851
6852         shader_addline(buffer, "RCP chroma.w, size.y;\n");
6853
6854         shader_addline(buffer, "MUL texcrd2.y, texcrd.y, size.y;\n");
6855
6856         shader_addline(buffer, "FLR texcrd2.y, texcrd2.y;\n");
6857         shader_addline(buffer, "MAD texcrd.y, texcrd.y, yv12_coef.y, yv12_coef.x;\n");
6858
6859         /* Read odd lines from the right side(add size * 0.5 to the x coordinate */
6860         shader_addline(buffer, "ADD texcrd2.x, texcrd2.y, yv12_coef.y;\n"); /* To avoid 0.5 == 0.5 comparisons */
6861         shader_addline(buffer, "FRC texcrd2.x, texcrd2.x;\n");
6862         shader_addline(buffer, "SGE texcrd2.x, texcrd2.x, coef.y;\n");
6863         shader_addline(buffer, "MAD texcrd.x, texcrd2.x, coef.y, texcrd.x;\n");
6864
6865         /* clamp, keep the half pixel origin in mind */
6866         shader_addline(buffer, "MAD temp.y, coef.y, chroma.w, yv12_coef.x;\n");
6867         shader_addline(buffer, "MAX texcrd.y, temp.y, texcrd.y;\n");
6868         shader_addline(buffer, "MAD temp.y, -coef.y, chroma.w, yv12_coef.z;\n");
6869         shader_addline(buffer, "MIN texcrd.y, temp.y, texcrd.y;\n");
6870     } else {
6871         /* Read from [size - size+size/4] */
6872         shader_addline(buffer, "FLR texcrd.y, texcrd.y;\n");
6873         shader_addline(buffer, "MAD texcrd.y, texcrd.y, coef.w, size.y;\n");
6874
6875         /* Read odd lines from the right side(add size * 0.5 to the x coordinate */
6876         shader_addline(buffer, "ADD texcrd2.x, texcrd.y, yv12_coef.y;\n"); /* To avoid 0.5 == 0.5 comparisons */
6877         shader_addline(buffer, "FRC texcrd2.x, texcrd2.x;\n");
6878         shader_addline(buffer, "SGE texcrd2.x, texcrd2.x, coef.y;\n");
6879         shader_addline(buffer, "MUL texcrd2.x, texcrd2.x, size.x;\n");
6880         shader_addline(buffer, "MAD texcrd.x, texcrd2.x, coef.y, texcrd.x;\n");
6881
6882         /* Make sure to read exactly from the pixel center */
6883         shader_addline(buffer, "FLR texcrd.y, texcrd.y;\n");
6884         shader_addline(buffer, "ADD texcrd.y, texcrd.y, coef.y;\n");
6885
6886         /* Clamp */
6887         shader_addline(buffer, "MAD temp.y, size.y, coef.w, size.y;\n");
6888         shader_addline(buffer, "ADD temp.y, temp.y, -coef.y;\n");
6889         shader_addline(buffer, "MIN texcrd.y, temp.y, texcrd.y;\n");
6890         shader_addline(buffer, "ADD temp.y, size.y, -coef.y;\n");
6891         shader_addline(buffer, "MAX texcrd.y, temp.y, texcrd.y;\n");
6892     }
6893     /* Read the texture, put the result into the output register */
6894     shader_addline(buffer, "TEX temp, texcrd, texture[0], %s;\n", tex);
6895     shader_addline(buffer, "MOV chroma.x, temp.w;\n");
6896
6897     /* The other chroma value is 1/6th of the texture lower, from 5/6th to 6/6th
6898      * No need to clamp because we're just reusing the already clamped value from above
6899      */
6900     if(textype == GL_TEXTURE_2D) {
6901         shader_addline(buffer, "ADD texcrd.y, texcrd.y, yv12_coef.y;\n");
6902     } else {
6903         shader_addline(buffer, "MAD texcrd.y, size.y, coef.w, texcrd.y;\n");
6904     }
6905     shader_addline(buffer, "TEX temp, texcrd, texture[0], %s;\n", tex);
6906     shader_addline(buffer, "MOV chroma.y, temp.w;\n");
6907
6908     /* Sample the luminance value. It is in the top 2/3rd of the texture, so scale the y coordinate.
6909      * Clamp the y coordinate to prevent the chroma values from bleeding into the sampled luminance
6910      * values due to filtering
6911      */
6912     shader_addline(buffer, "MOV texcrd, fragment.texcoord[0];\n");
6913     if(textype == GL_TEXTURE_2D) {
6914         /* Multiply the y coordinate by 2/3 and clamp it */
6915         shader_addline(buffer, "MUL texcrd.y, texcrd.y, yv12_coef.x;\n");
6916         shader_addline(buffer, "MAD temp.y, -coef.y, chroma.w, yv12_coef.x;\n");
6917         shader_addline(buffer, "MIN texcrd.y, temp.y, texcrd.y;\n");
6918         shader_addline(buffer, "TEX luminance, texcrd, texture[0], %s;\n", tex);
6919     } else {
6920         /* Reading from texture_rectangles is pretty straightforward, just use the unmodified
6921          * texture coordinate. It is still a good idea to clamp it though, since the opengl texture
6922          * is bigger
6923          */
6924         shader_addline(buffer, "ADD temp.x, size.y, -coef.y;\n");
6925         shader_addline(buffer, "MIN texcrd.y, texcrd.y, size.x;\n");
6926         shader_addline(buffer, "TEX luminance, texcrd, texture[0], %s;\n", tex);
6927     }
6928     *luminance = 'a';
6929
6930     return TRUE;
6931 }
6932
6933 static GLuint gen_p8_shader(struct arbfp_blit_priv *priv,
6934         const struct wined3d_gl_info *gl_info, GLenum textype)
6935 {
6936     GLenum shader;
6937     struct wined3d_shader_buffer buffer;
6938     GLint pos;
6939
6940     /* Shader header */
6941     if (!shader_buffer_init(&buffer))
6942     {
6943         ERR("Failed to initialize shader buffer.\n");
6944         return 0;
6945     }
6946
6947     GL_EXTCALL(glGenProgramsARB(1, &shader));
6948     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader));
6949     if (!shader)
6950     {
6951         shader_buffer_free(&buffer);
6952         return 0;
6953     }
6954
6955     shader_addline(&buffer, "!!ARBfp1.0\n");
6956     shader_addline(&buffer, "TEMP index;\n");
6957
6958     /* { 255/256, 0.5/255*255/256, 0, 0 } */
6959     shader_addline(&buffer, "PARAM constants = { 0.996, 0.00195, 0, 0 };\n");
6960
6961     /* The alpha-component contains the palette index */
6962     if(textype == GL_TEXTURE_RECTANGLE_ARB)
6963         shader_addline(&buffer, "TXP index, fragment.texcoord[0], texture[0], RECT;\n");
6964     else
6965         shader_addline(&buffer, "TEX index, fragment.texcoord[0], texture[0], 2D;\n");
6966
6967     /* Scale the index by 255/256 and add a bias of '0.5' in order to sample in the middle */
6968     shader_addline(&buffer, "MAD index.a, index.a, constants.x, constants.y;\n");
6969
6970     /* Use the alpha-component as an index in the palette to get the final color */
6971     shader_addline(&buffer, "TEX result.color, index.a, texture[1], 1D;\n");
6972     shader_addline(&buffer, "END\n");
6973
6974     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
6975             strlen(buffer.buffer), buffer.buffer));
6976     checkGLcall("glProgramStringARB()");
6977
6978     gl_info->gl_ops.gl.p_glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
6979     if (pos != -1)
6980     {
6981         FIXME("Fragment program error at position %d: %s\n\n", pos,
6982               debugstr_a((const char *)gl_info->gl_ops.gl.p_glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
6983         shader_arb_dump_program_source(buffer.buffer);
6984     }
6985
6986     if (textype == GL_TEXTURE_RECTANGLE_ARB)
6987         priv->p8_rect_shader = shader;
6988     else
6989         priv->p8_2d_shader = shader;
6990
6991     shader_buffer_free(&buffer);
6992
6993     return shader;
6994 }
6995
6996 /* Context activation is done by the caller. */
6997 static void upload_palette(const struct wined3d_surface *surface, struct wined3d_context *context)
6998 {
6999     BYTE table[256][4];
7000     struct wined3d_device *device = surface->resource.device;
7001     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
7002     struct arbfp_blit_priv *priv = device->blit_priv;
7003     BOOL colorkey = (surface->CKeyFlags & WINEDDSD_CKSRCBLT) != 0;
7004
7005     d3dfmt_p8_init_palette(surface, table, colorkey);
7006
7007     if (gl_info->supported[APPLE_CLIENT_STORAGE])
7008     {
7009         gl_info->gl_ops.gl.p_glPixelStorei(GL_UNPACK_CLIENT_STORAGE_APPLE, GL_FALSE);
7010         checkGLcall("glPixelStorei(GL_UNPACK_CLIENT_STORAGE_APPLE, GL_FALSE)");
7011     }
7012
7013     if (!priv->palette_texture)
7014         gl_info->gl_ops.gl.p_glGenTextures(1, &priv->palette_texture);
7015
7016     GL_EXTCALL(glActiveTextureARB(GL_TEXTURE1));
7017     gl_info->gl_ops.gl.p_glBindTexture(GL_TEXTURE_1D, priv->palette_texture);
7018
7019     gl_info->gl_ops.gl.p_glTexEnvi(GL_TEXTURE_ENV, GL_TEXTURE_ENV_MODE, GL_REPLACE);
7020
7021     gl_info->gl_ops.gl.p_glTexParameteri(GL_TEXTURE_1D, GL_TEXTURE_MIN_FILTER, GL_NEAREST);
7022     /* Make sure we have discrete color levels. */
7023     gl_info->gl_ops.gl.p_glTexParameteri(GL_TEXTURE_1D, GL_TEXTURE_MAG_FILTER, GL_NEAREST);
7024     gl_info->gl_ops.gl.p_glTexParameteri(GL_TEXTURE_1D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
7025     /* Upload the palette */
7026     /* TODO: avoid unneeded uploads in the future by adding some SFLAG_PALETTE_DIRTY mechanism */
7027     gl_info->gl_ops.gl.p_glTexImage1D(GL_TEXTURE_1D, 0, GL_RGBA, 256, 0, GL_RGBA, GL_UNSIGNED_BYTE, table);
7028
7029     if (gl_info->supported[APPLE_CLIENT_STORAGE])
7030     {
7031         gl_info->gl_ops.gl.p_glPixelStorei(GL_UNPACK_CLIENT_STORAGE_APPLE, GL_TRUE);
7032         checkGLcall("glPixelStorei(GL_UNPACK_CLIENT_STORAGE_APPLE, GL_TRUE)");
7033     }
7034
7035     /* Switch back to unit 0 in which the 2D texture will be stored. */
7036     context_active_texture(context, gl_info, 0);
7037 }
7038
7039 /* Context activation is done by the caller. */
7040 static GLuint gen_yuv_shader(struct arbfp_blit_priv *priv, const struct wined3d_gl_info *gl_info,
7041         enum complex_fixup yuv_fixup, GLenum textype)
7042 {
7043     GLenum shader;
7044     struct wined3d_shader_buffer buffer;
7045     char luminance_component;
7046     GLint pos;
7047
7048     /* Shader header */
7049     if (!shader_buffer_init(&buffer))
7050     {
7051         ERR("Failed to initialize shader buffer.\n");
7052         return 0;
7053     }
7054
7055     GL_EXTCALL(glGenProgramsARB(1, &shader));
7056     checkGLcall("GL_EXTCALL(glGenProgramsARB(1, &shader))");
7057     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader));
7058     checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader)");
7059     if (!shader)
7060     {
7061         shader_buffer_free(&buffer);
7062         return 0;
7063     }
7064
7065     /* The YUY2 and UYVY formats contain two pixels packed into a 32 bit macropixel,
7066      * giving effectively 16 bit per pixel. The color consists of a luminance(Y) and
7067      * two chroma(U and V) values. Each macropixel has two luminance values, one for
7068      * each single pixel it contains, and one U and one V value shared between both
7069      * pixels.
7070      *
7071      * The data is loaded into an A8L8 texture. With YUY2, the luminance component
7072      * contains the luminance and alpha the chroma. With UYVY it is vice versa. Thus
7073      * take the format into account when generating the read swizzles
7074      *
7075      * Reading the Y value is straightforward - just sample the texture. The hardware
7076      * takes care of filtering in the horizontal and vertical direction.
7077      *
7078      * Reading the U and V values is harder. We have to avoid filtering horizontally,
7079      * because that would mix the U and V values of one pixel or two adjacent pixels.
7080      * Thus floor the texture coordinate and add 0.5 to get an unfiltered read,
7081      * regardless of the filtering setting. Vertical filtering works automatically
7082      * though - the U and V values of two rows are mixed nicely.
7083      *
7084      * Apart of avoiding filtering issues, the code has to know which value it just
7085      * read, and where it can find the other one. To determine this, it checks if
7086      * it sampled an even or odd pixel, and shifts the 2nd read accordingly.
7087      *
7088      * Handling horizontal filtering of U and V values requires reading a 2nd pair
7089      * of pixels, extracting U and V and mixing them. This is not implemented yet.
7090      *
7091      * An alternative implementation idea is to load the texture as A8R8G8B8 texture,
7092      * with width / 2. This way one read gives all 3 values, finding U and V is easy
7093      * in an unfiltered situation. Finding the luminance on the other hand requires
7094      * finding out if it is an odd or even pixel. The real drawback of this approach
7095      * is filtering. This would have to be emulated completely in the shader, reading
7096      * up two 2 packed pixels in up to 2 rows and interpolating both horizontally and
7097      * vertically. Beyond that it would require adjustments to the texture handling
7098      * code to deal with the width scaling
7099      */
7100     shader_addline(&buffer, "!!ARBfp1.0\n");
7101     shader_addline(&buffer, "TEMP luminance;\n");
7102     shader_addline(&buffer, "TEMP temp;\n");
7103     shader_addline(&buffer, "TEMP chroma;\n");
7104     shader_addline(&buffer, "TEMP texcrd;\n");
7105     shader_addline(&buffer, "TEMP texcrd2;\n");
7106     shader_addline(&buffer, "PARAM coef = {1.0, 0.5, 2.0, 0.25};\n");
7107     shader_addline(&buffer, "PARAM yuv_coef = {1.403, 0.344, 0.714, 1.770};\n");
7108     shader_addline(&buffer, "PARAM size = program.local[0];\n");
7109
7110     switch (yuv_fixup)
7111     {
7112         case COMPLEX_FIXUP_UYVY:
7113         case COMPLEX_FIXUP_YUY2:
7114             if (!gen_planar_yuv_read(&buffer, yuv_fixup, textype, &luminance_component))
7115             {
7116                 shader_buffer_free(&buffer);
7117                 return 0;
7118             }
7119             break;
7120
7121         case COMPLEX_FIXUP_YV12:
7122             if (!gen_yv12_read(&buffer, textype, &luminance_component))
7123             {
7124                 shader_buffer_free(&buffer);
7125                 return 0;
7126             }
7127             break;
7128
7129         default:
7130             FIXME("Unsupported YUV fixup %#x\n", yuv_fixup);
7131             shader_buffer_free(&buffer);
7132             return 0;
7133     }
7134
7135     /* Calculate the final result. Formula is taken from
7136      * http://www.fourcc.org/fccyvrgb.php. Note that the chroma
7137      * ranges from -0.5 to 0.5
7138      */
7139     shader_addline(&buffer, "SUB chroma.xy, chroma, coef.y;\n");
7140
7141     shader_addline(&buffer, "MAD result.color.x, chroma.x, yuv_coef.x, luminance.%c;\n", luminance_component);
7142     shader_addline(&buffer, "MAD temp.x, -chroma.y, yuv_coef.y, luminance.%c;\n", luminance_component);
7143     shader_addline(&buffer, "MAD result.color.y, -chroma.x, yuv_coef.z, temp.x;\n");
7144     shader_addline(&buffer, "MAD result.color.z, chroma.y, yuv_coef.w, luminance.%c;\n", luminance_component);
7145     shader_addline(&buffer, "END\n");
7146
7147     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
7148             strlen(buffer.buffer), buffer.buffer));
7149     checkGLcall("glProgramStringARB()");
7150
7151     gl_info->gl_ops.gl.p_glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
7152     if (pos != -1)
7153     {
7154         FIXME("Fragment program error at position %d: %s\n\n", pos,
7155               debugstr_a((const char *)gl_info->gl_ops.gl.p_glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
7156         shader_arb_dump_program_source(buffer.buffer);
7157     }
7158     else
7159     {
7160         GLint native;
7161
7162         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
7163         checkGLcall("glGetProgramivARB()");
7164         if (!native) WARN("Program exceeds native resource limits.\n");
7165     }
7166
7167     shader_buffer_free(&buffer);
7168
7169     switch (yuv_fixup)
7170     {
7171         case COMPLEX_FIXUP_YUY2:
7172             if (textype == GL_TEXTURE_RECTANGLE_ARB) priv->yuy2_rect_shader = shader;
7173             else priv->yuy2_2d_shader = shader;
7174             break;
7175
7176         case COMPLEX_FIXUP_UYVY:
7177             if (textype == GL_TEXTURE_RECTANGLE_ARB) priv->uyvy_rect_shader = shader;
7178             else priv->uyvy_2d_shader = shader;
7179             break;
7180
7181         case COMPLEX_FIXUP_YV12:
7182             if (textype == GL_TEXTURE_RECTANGLE_ARB) priv->yv12_rect_shader = shader;
7183             else priv->yv12_2d_shader = shader;
7184             break;
7185         default:
7186             ERR("Unsupported complex fixup: %d\n", yuv_fixup);
7187     }
7188
7189     return shader;
7190 }
7191
7192 /* Context activation is done by the caller. */
7193 static HRESULT arbfp_blit_set(void *blit_priv, struct wined3d_context *context, const struct wined3d_surface *surface)
7194 {
7195     GLenum shader;
7196     float size[4] = {(float) surface->pow2Width, (float) surface->pow2Height, 1.0f, 1.0f};
7197     struct arbfp_blit_priv *priv = blit_priv;
7198     enum complex_fixup fixup;
7199     const struct wined3d_gl_info *gl_info = context->gl_info;
7200     GLenum textype;
7201
7202     if (surface->container.type == WINED3D_CONTAINER_TEXTURE)
7203         textype = surface->container.u.texture->target;
7204     else
7205         textype = surface->texture_target;
7206
7207     if (surface->flags & SFLAG_CONVERTED)
7208     {
7209         gl_info->gl_ops.gl.p_glEnable(textype);
7210         checkGLcall("glEnable(textype)");
7211         return WINED3D_OK;
7212     }
7213
7214     if (!is_complex_fixup(surface->resource.format->color_fixup))
7215     {
7216         TRACE("Fixup:\n");
7217         dump_color_fixup_desc(surface->resource.format->color_fixup);
7218         /* Don't bother setting up a shader for unconverted formats */
7219         gl_info->gl_ops.gl.p_glEnable(textype);
7220         checkGLcall("glEnable(textype)");
7221         return WINED3D_OK;
7222     }
7223
7224     fixup = get_complex_fixup(surface->resource.format->color_fixup);
7225
7226     switch(fixup)
7227     {
7228         case COMPLEX_FIXUP_YUY2:
7229             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->yuy2_rect_shader : priv->yuy2_2d_shader;
7230             break;
7231
7232         case COMPLEX_FIXUP_UYVY:
7233             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->uyvy_rect_shader : priv->uyvy_2d_shader;
7234             break;
7235
7236         case COMPLEX_FIXUP_YV12:
7237             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->yv12_rect_shader : priv->yv12_2d_shader;
7238             break;
7239
7240         case COMPLEX_FIXUP_P8:
7241             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->p8_rect_shader : priv->p8_2d_shader;
7242             if (!shader) shader = gen_p8_shader(priv, gl_info, textype);
7243
7244             upload_palette(surface, context);
7245             break;
7246
7247         default:
7248             FIXME("Unsupported complex fixup %#x, not setting a shader\n", fixup);
7249             gl_info->gl_ops.gl.p_glEnable(textype);
7250             checkGLcall("glEnable(textype)");
7251             return E_NOTIMPL;
7252     }
7253
7254     if (!shader) shader = gen_yuv_shader(priv, gl_info, fixup, textype);
7255
7256     gl_info->gl_ops.gl.p_glEnable(GL_FRAGMENT_PROGRAM_ARB);
7257     checkGLcall("glEnable(GL_FRAGMENT_PROGRAM_ARB)");
7258     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader));
7259     checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader)");
7260     GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, 0, size));
7261     checkGLcall("glProgramLocalParameter4fvARB");
7262
7263     return WINED3D_OK;
7264 }
7265
7266 /* Context activation is done by the caller. */
7267 static void arbfp_blit_unset(const struct wined3d_gl_info *gl_info)
7268 {
7269     gl_info->gl_ops.gl.p_glDisable(GL_FRAGMENT_PROGRAM_ARB);
7270     checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
7271     gl_info->gl_ops.gl.p_glDisable(GL_TEXTURE_2D);
7272     checkGLcall("glDisable(GL_TEXTURE_2D)");
7273     if (gl_info->supported[ARB_TEXTURE_CUBE_MAP])
7274     {
7275         gl_info->gl_ops.gl.p_glDisable(GL_TEXTURE_CUBE_MAP_ARB);
7276         checkGLcall("glDisable(GL_TEXTURE_CUBE_MAP_ARB)");
7277     }
7278     if (gl_info->supported[ARB_TEXTURE_RECTANGLE])
7279     {
7280         gl_info->gl_ops.gl.p_glDisable(GL_TEXTURE_RECTANGLE_ARB);
7281         checkGLcall("glDisable(GL_TEXTURE_RECTANGLE_ARB)");
7282     }
7283 }
7284
7285 static BOOL arbfp_blit_supported(const struct wined3d_gl_info *gl_info, enum wined3d_blit_op blit_op,
7286         const RECT *src_rect, DWORD src_usage, enum wined3d_pool src_pool, const struct wined3d_format *src_format,
7287         const RECT *dst_rect, DWORD dst_usage, enum wined3d_pool dst_pool, const struct wined3d_format *dst_format)
7288 {
7289     enum complex_fixup src_fixup;
7290
7291     if (!gl_info->supported[ARB_FRAGMENT_PROGRAM])
7292         return FALSE;
7293
7294     if (blit_op != WINED3D_BLIT_OP_COLOR_BLIT)
7295     {
7296         TRACE("Unsupported blit_op=%d\n", blit_op);
7297         return FALSE;
7298     }
7299
7300     if (src_pool == WINED3D_POOL_SYSTEM_MEM || dst_pool == WINED3D_POOL_SYSTEM_MEM)
7301         return FALSE;
7302
7303     src_fixup = get_complex_fixup(src_format->color_fixup);
7304     if (TRACE_ON(d3d_shader) && TRACE_ON(d3d))
7305     {
7306         TRACE("Checking support for fixup:\n");
7307         dump_color_fixup_desc(src_format->color_fixup);
7308     }
7309
7310     if (!is_identity_fixup(dst_format->color_fixup))
7311     {
7312         TRACE("Destination fixups are not supported\n");
7313         return FALSE;
7314     }
7315
7316     if (is_identity_fixup(src_format->color_fixup))
7317     {
7318         TRACE("[OK]\n");
7319         return TRUE;
7320     }
7321
7322      /* We only support YUV conversions. */
7323     if (!is_complex_fixup(src_format->color_fixup))
7324     {
7325         TRACE("[FAILED]\n");
7326         return FALSE;
7327     }
7328
7329     switch(src_fixup)
7330     {
7331         case COMPLEX_FIXUP_YUY2:
7332         case COMPLEX_FIXUP_UYVY:
7333         case COMPLEX_FIXUP_YV12:
7334         case COMPLEX_FIXUP_P8:
7335             TRACE("[OK]\n");
7336             return TRUE;
7337
7338         default:
7339             FIXME("Unsupported YUV fixup %#x\n", src_fixup);
7340             TRACE("[FAILED]\n");
7341             return FALSE;
7342     }
7343 }
7344
7345 HRESULT arbfp_blit_surface(struct wined3d_device *device, DWORD filter,
7346         struct wined3d_surface *src_surface, const RECT *src_rect_in,
7347         struct wined3d_surface *dst_surface, const RECT *dst_rect_in)
7348 {
7349     struct wined3d_context *context;
7350     RECT src_rect = *src_rect_in;
7351     RECT dst_rect = *dst_rect_in;
7352
7353     /* Now load the surface */
7354     if (wined3d_settings.offscreen_rendering_mode != ORM_FBO
7355             && (src_surface->flags & (SFLAG_INTEXTURE | SFLAG_INDRAWABLE)) == SFLAG_INDRAWABLE)
7356     {
7357         /* Without FBO blits transferring from the drawable to the texture is
7358          * expensive, because we have to flip the data in sysmem. Since we can
7359          * flip in the blitter, we don't actually need that flip anyway. So we
7360          * use the surface's texture as scratch texture, and flip the source
7361          * rectangle instead. */
7362         surface_load_fb_texture(src_surface, FALSE);
7363
7364         src_rect.top = src_surface->resource.height - src_rect.top;
7365         src_rect.bottom = src_surface->resource.height - src_rect.bottom;
7366     }
7367     else
7368         surface_internal_preload(src_surface, SRGB_RGB);
7369
7370     /* Activate the destination context, set it up for blitting */
7371     context = context_acquire(device, dst_surface);
7372     context_apply_blit_state(context, device);
7373
7374     if (!surface_is_offscreen(dst_surface))
7375         surface_translate_drawable_coords(dst_surface, context->win_handle, &dst_rect);
7376
7377     arbfp_blit_set(device->blit_priv, context, src_surface);
7378
7379     /* Draw a textured quad */
7380     draw_textured_quad(src_surface, context, &src_rect, &dst_rect, filter);
7381
7382     /* Leave the opengl state valid for blitting */
7383     arbfp_blit_unset(context->gl_info);
7384
7385     if (wined3d_settings.strict_draw_ordering
7386             || (dst_surface->container.type == WINED3D_CONTAINER_SWAPCHAIN
7387             && (dst_surface->container.u.swapchain->front_buffer == dst_surface)))
7388         context->gl_info->gl_ops.gl.p_glFlush(); /* Flush to ensure ordering across contexts. */
7389
7390     context_release(context);
7391
7392     surface_modify_location(dst_surface, dst_surface->draw_binding, TRUE);
7393     return WINED3D_OK;
7394 }
7395
7396 /* Do not call while under the GL lock. */
7397 static HRESULT arbfp_blit_color_fill(struct wined3d_device *device, struct wined3d_surface *dst_surface,
7398         const RECT *dst_rect, const struct wined3d_color *color)
7399 {
7400     FIXME("Color filling not implemented by arbfp_blit\n");
7401     return WINED3DERR_INVALIDCALL;
7402 }
7403
7404 /* Do not call while under the GL lock. */
7405 static HRESULT arbfp_blit_depth_fill(struct wined3d_device *device,
7406         struct wined3d_surface *surface, const RECT *rect, float depth)
7407 {
7408     FIXME("Depth filling not implemented by arbfp_blit.\n");
7409     return WINED3DERR_INVALIDCALL;
7410 }
7411
7412 const struct blit_shader arbfp_blit = {
7413     arbfp_blit_alloc,
7414     arbfp_blit_free,
7415     arbfp_blit_set,
7416     arbfp_blit_unset,
7417     arbfp_blit_supported,
7418     arbfp_blit_color_fill,
7419     arbfp_blit_depth_fill,
7420 };