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     const struct wined3d_d3d_info *d3d_info = context->d3d_info;
 661     struct shader_arb_priv *priv = device->shader_priv;
 662
 663     if (!from_shader_select)
 664     {
 665         const struct wined3d_shader *vshader = state->vertex_shader, *pshader = state->pixel_shader;
 666         if (vshader
 667                 && (stateblock->changed.vertexShaderConstantsB & vshader->reg_maps.boolean_constants
 668                 || (!gl_info->supported[NV_VERTEX_PROGRAM2_OPTION]
 669                 && (stateblock->changed.vertexShaderConstantsI
 670                 & vshader->reg_maps.integer_constants & ~vshader->reg_maps.local_int_consts))))
 671         {
 672             TRACE("bool/integer vertex shader constants potentially modified, forcing shader reselection.\n");
 673             shader_arb_select(context,
 674                     useVertexShader ? WINED3D_SHADER_MODE_SHADER : WINED3D_SHADER_MODE_FFP,
 675                     usePixelShader ? WINED3D_SHADER_MODE_SHADER : WINED3D_SHADER_MODE_FFP);
 676         }
 677         else if (pshader
 678                 && (stateblock->changed.pixelShaderConstantsB & pshader->reg_maps.boolean_constants
 679                 || (!gl_info->supported[NV_FRAGMENT_PROGRAM_OPTION]
 680                 && (stateblock->changed.pixelShaderConstantsI
 681                 & pshader->reg_maps.integer_constants & ~pshader->reg_maps.local_int_consts))))
 682         {
 683             TRACE("bool/integer pixel shader constants potentially modified, forcing shader reselection.\n");
 684             shader_arb_select(context,
 685                     useVertexShader ? WINED3D_SHADER_MODE_SHADER : WINED3D_SHADER_MODE_FFP,
 686                     usePixelShader ? WINED3D_SHADER_MODE_SHADER : WINED3D_SHADER_MODE_FFP);
 687         }
 688     }
 689
 690     if (context != priv->last_context)
 691     {
 692         memset(priv->vshader_const_dirty, 1,
 693                 sizeof(*priv->vshader_const_dirty) * d3d_info->limits.vs_uniform_count);
 694         priv->highest_dirty_vs_const = d3d_info->limits.vs_uniform_count;
 695
 696         memset(priv->pshader_const_dirty, 1,
 697                 sizeof(*priv->pshader_const_dirty) * d3d_info->limits.ps_uniform_count);
 698         priv->highest_dirty_ps_const = d3d_info->limits.ps_uniform_count;
 699
 700         priv->last_context = context;
 701     }
 702
 703     if (useVertexShader)
 704     {
 705         struct wined3d_shader *vshader = state->vertex_shader;
 706         const struct arb_vs_compiled_shader *gl_shader = priv->compiled_vprog;
 707
 708         /* Load DirectX 9 float constants for vertex shader */
 709         priv->highest_dirty_vs_const = shader_arb_load_constantsF(vshader, gl_info, GL_VERTEX_PROGRAM_ARB,
 710                 priv->highest_dirty_vs_const, state->vs_consts_f, priv->vshader_const_dirty);
 711         shader_arb_vs_local_constants(gl_shader, context, state);
 712     }
 713
 714     if (usePixelShader)
 715     {
 716         struct wined3d_shader *pshader = state->pixel_shader;
 717         const struct arb_ps_compiled_shader *gl_shader = priv->compiled_fprog;
 718         UINT rt_height = state->fb->render_targets[0]->resource.height;
 719
 720         /* Load DirectX 9 float constants for pixel shader */
 721         priv->highest_dirty_ps_const = shader_arb_load_constantsF(pshader, gl_info, GL_FRAGMENT_PROGRAM_ARB,
 722                 priv->highest_dirty_ps_const, state->ps_consts_f, priv->pshader_const_dirty);
 723         shader_arb_ps_local_constants(gl_shader, context, state, rt_height);
 724     }
 725 }
 726
 727 static void shader_arb_load_constants(const struct wined3d_context *context, BOOL ps, BOOL vs)
 728 {
 729     shader_arb_load_constants_internal(context, ps, vs, FALSE);
 730 }
 731
 732 static void shader_arb_update_float_vertex_constants(struct wined3d_device *device, UINT start, UINT count)
 733 {
 734     struct wined3d_context *context = context_get_current();
 735     struct shader_arb_priv *priv = device->shader_priv;
 736
 737     /* We don't want shader constant dirtification to be an O(contexts), so just dirtify the active
 738      * context. On a context switch the old context will be fully dirtified */
 739     if (!context || context->swapchain->device != device) return;
 740
 741     memset(priv->vshader_const_dirty + start, 1, sizeof(*priv->vshader_const_dirty) * count);
 742     priv->highest_dirty_vs_const = max(priv->highest_dirty_vs_const, start + count);
 743 }
 744
 745 static void shader_arb_update_float_pixel_constants(struct wined3d_device *device, UINT start, UINT count)
 746 {
 747     struct wined3d_context *context = context_get_current();
 748     struct shader_arb_priv *priv = device->shader_priv;
 749
 750     /* We don't want shader constant dirtification to be an O(contexts), so just dirtify the active
 751      * context. On a context switch the old context will be fully dirtified */
 752     if (!context || context->swapchain->device != device) return;
 753
 754     memset(priv->pshader_const_dirty + start, 1, sizeof(*priv->pshader_const_dirty) * count);
 755     priv->highest_dirty_ps_const = max(priv->highest_dirty_ps_const, start + count);
 756 }
 757
 758 /* Generate the variable & register declarations for the ARB_vertex_program output target */
 759 static void shader_generate_arb_declarations(const struct wined3d_shader *shader,
 760         const struct wined3d_shader_reg_maps *reg_maps, struct wined3d_shader_buffer *buffer,
 761         const struct wined3d_gl_info *gl_info, DWORD *num_clipplanes,
 762         const struct shader_arb_ctx_priv *ctx)
 763 {
 764     DWORD i;
 765     char pshader = shader_is_pshader_version(reg_maps->shader_version.type);
 766     const struct wined3d_shader_lconst *lconst;
 767     unsigned max_constantsF;
 768     DWORD map;
 769
 770     /* In pixel shaders, all private constants are program local, we don't need anything
 771      * from program.env. Thus we can advertise the full set of constants in pixel shaders.
 772      * If we need a private constant the GL implementation will squeeze it in somewhere
 773      *
 774      * With vertex shaders we need the posFixup and on some GL implementations 4 helper
 775      * immediate values. The posFixup is loaded using program.env for now, so always
 776      * subtract one from the number of constants. If the shader uses indirect addressing,
 777      * account for the helper const too because we have to declare all available d3d constants
 778      * and don't know which are actually used.
 779      */
 780     if (pshader)
 781     {
 782         max_constantsF = gl_info->limits.arb_ps_native_constants;
 783         /* 24 is the minimum MAX_PROGRAM_ENV_PARAMETERS_ARB value. */
 784         if (max_constantsF < 24)
 785             max_constantsF = gl_info->limits.arb_ps_float_constants;
 786     }
 787     else
 788     {
 789         const struct arb_vshader_private *shader_data = shader->backend_data;
 790         max_constantsF = gl_info->limits.arb_vs_native_constants;
 791         /* 96 is the minimum MAX_PROGRAM_ENV_PARAMETERS_ARB value.
 792          * Also prevents max_constantsF from becoming less than 0 and
 793          * wrapping . */
 794         if (max_constantsF < 96)
 795             max_constantsF = gl_info->limits.arb_vs_float_constants;
 796
 797         if (reg_maps->usesrelconstF)
 798         {
 799             DWORD highest_constf = 0, clip_limit;
 800
 801             max_constantsF -= reserved_vs_const(shader_data, reg_maps, gl_info);
 802             max_constantsF -= count_bits(reg_maps->integer_constants);
 803             max_constantsF -= gl_info->reserved_arb_constants;
 804
 805             for (i = 0; i < shader->limits.constant_float; ++i)
 806             {
 807                 DWORD idx = i >> 5;
 808                 DWORD shift = i & 0x1f;
 809                 if(reg_maps->constf[idx] & (1 << shift)) highest_constf = i;
 810             }
 811
 812             if(use_nv_clip(gl_info) && ctx->target_version >= NV2)
 813             {
 814                 if(ctx->cur_vs_args->super.clip_enabled)
 815                     clip_limit = gl_info->limits.clipplanes;
 816                 else
 817                     clip_limit = 0;
 818             }
 819             else
 820             {
 821                 unsigned int mask = ctx->cur_vs_args->clip.boolclip.clipplane_mask;
 822                 clip_limit = min(count_bits(mask), 4);
 823             }
 824             *num_clipplanes = min(clip_limit, max_constantsF - highest_constf - 1);
 825             max_constantsF -= *num_clipplanes;
 826             if(*num_clipplanes < clip_limit)
 827             {
 828                 WARN("Only %u clipplanes out of %u enabled\n", *num_clipplanes, gl_info->limits.clipplanes);
 829             }
 830         }
 831         else
 832         {
 833             if (ctx->target_version >= NV2) *num_clipplanes = gl_info->limits.clipplanes;
 834             else *num_clipplanes = min(gl_info->limits.clipplanes, 4);
 835         }
 836     }
 837
 838     for (i = 0, map = reg_maps->temporary; map; map >>= 1, ++i)
 839     {
 840         if (map & 1) shader_addline(buffer, "TEMP R%u;\n", i);
 841     }
 842
 843     for (i = 0, map = reg_maps->address; map; map >>= 1, ++i)
 844     {
 845         if (map & 1) shader_addline(buffer, "ADDRESS A%u;\n", i);
 846     }
 847
 848     if (pshader && reg_maps->shader_version.major == 1 && reg_maps->shader_version.minor <= 3)
 849     {
 850         for (i = 0, map = reg_maps->texcoord; map; map >>= 1, ++i)
 851         {
 852             if (map & 1) shader_addline(buffer, "TEMP T%u;\n", i);
 853         }
 854     }
 855
 856     if (!shader->load_local_constsF)
 857     {
 858         LIST_FOR_EACH_ENTRY(lconst, &shader->constantsF, struct wined3d_shader_lconst, entry)
 859         {
 860             const float *value;
 861             value = (const float *)lconst->value;
 862             shader_addline(buffer, "PARAM C%u = {%.8e, %.8e, %.8e, %.8e};\n", lconst->idx,
 863                            value[0], value[1], value[2], value[3]);
 864         }
 865     }
 866
 867     /* After subtracting privately used constants from the hardware limit(they are loaded as
 868      * local constants), make sure the shader doesn't violate the env constant limit
 869      */
 870     if(pshader)
 871     {
 872         max_constantsF = min(max_constantsF, gl_info->limits.arb_ps_float_constants);
 873     }
 874     else
 875     {
 876         max_constantsF = min(max_constantsF, gl_info->limits.arb_vs_float_constants);
 877     }
 878
 879     /* Avoid declaring more constants than needed */
 880     max_constantsF = min(max_constantsF, shader->limits.constant_float);
 881
 882     /* we use the array-based constants array if the local constants are marked for loading,
 883      * because then we use indirect addressing, or when the local constant list is empty,
 884      * because then we don't know if we're using indirect addressing or not. If we're hardcoding
 885      * local constants do not declare the loaded constants as an array because ARB compilers usually
 886      * do not optimize unused constants away
 887      */
 888     if (reg_maps->usesrelconstF)
 889     {
 890         /* Need to PARAM the environment parameters (constants) so we can use relative addressing */
 891         shader_addline(buffer, "PARAM C[%d] = { program.env[0..%d] };\n",
 892                     max_constantsF, max_constantsF - 1);
 893     } else {
 894         for(i = 0; i < max_constantsF; i++) {
 895             DWORD idx, mask;
 896             idx = i >> 5;
 897             mask = 1 << (i & 0x1f);
 898             if (!shader_constant_is_local(shader, i) && (reg_maps->constf[idx] & mask))
 899             {
 900                 shader_addline(buffer, "PARAM C%d = program.env[%d];\n",i, i);
 901             }
 902         }
 903     }
 904 }
 905
 906 static const char * const shift_tab[] = {
 907     "dummy",     /*  0 (none) */
 908     "coefmul.x", /*  1 (x2)   */
 909     "coefmul.y", /*  2 (x4)   */
 910     "coefmul.z", /*  3 (x8)   */
 911     "coefmul.w", /*  4 (x16)  */
 912     "dummy",     /*  5 (x32)  */
 913     "dummy",     /*  6 (x64)  */
 914     "dummy",     /*  7 (x128) */
 915     "dummy",     /*  8 (d256) */
 916     "dummy",     /*  9 (d128) */
 917     "dummy",     /* 10 (d64)  */
 918     "dummy",     /* 11 (d32)  */
 919     "coefdiv.w", /* 12 (d16)  */
 920     "coefdiv.z", /* 13 (d8)   */
 921     "coefdiv.y", /* 14 (d4)   */
 922     "coefdiv.x"  /* 15 (d2)   */
 923 };
 924
 925 static void shader_arb_get_write_mask(const struct wined3d_shader_instruction *ins,
 926         const struct wined3d_shader_dst_param *dst, char *write_mask)
 927 {
 928     char *ptr = write_mask;
 929
 930     if (dst->write_mask != WINED3DSP_WRITEMASK_ALL)
 931     {
 932         *ptr++ = '.';
 933         if (dst->write_mask & WINED3DSP_WRITEMASK_0) *ptr++ = 'x';
 934         if (dst->write_mask & WINED3DSP_WRITEMASK_1) *ptr++ = 'y';
 935         if (dst->write_mask & WINED3DSP_WRITEMASK_2) *ptr++ = 'z';
 936         if (dst->write_mask & WINED3DSP_WRITEMASK_3) *ptr++ = 'w';
 937     }
 938
 939     *ptr = '\0';
 940 }
 941
 942 static void shader_arb_get_swizzle(const struct wined3d_shader_src_param *param, BOOL fixup, char *swizzle_str)
 943 {
 944     /* For registers of type WINED3DDECLTYPE_D3DCOLOR, data is stored as "bgra",
 945      * but addressed as "rgba". To fix this we need to swap the register's x
 946      * and z components. */
 947     const char *swizzle_chars = fixup ? "zyxw" : "xyzw";
 948     char *ptr = swizzle_str;
 949
 950     /* swizzle bits fields: wwzzyyxx */
 951     DWORD swizzle = param->swizzle;
 952     DWORD swizzle_x = swizzle & 0x03;
 953     DWORD swizzle_y = (swizzle >> 2) & 0x03;
 954     DWORD swizzle_z = (swizzle >> 4) & 0x03;
 955     DWORD swizzle_w = (swizzle >> 6) & 0x03;
 956
 957     /* If the swizzle is the default swizzle (ie, "xyzw"), we don't need to
 958      * generate a swizzle string. Unless we need to our own swizzling. */
 959     if (swizzle != WINED3DSP_NOSWIZZLE || fixup)
 960     {
 961         *ptr++ = '.';
 962         if (swizzle_x == swizzle_y && swizzle_x == swizzle_z && swizzle_x == swizzle_w) {
 963             *ptr++ = swizzle_chars[swizzle_x];
 964         } else {
 965             *ptr++ = swizzle_chars[swizzle_x];
 966             *ptr++ = swizzle_chars[swizzle_y];
 967             *ptr++ = swizzle_chars[swizzle_z];
 968             *ptr++ = swizzle_chars[swizzle_w];
 969         }
 970     }
 971
 972     *ptr = '\0';
 973 }
 974
 975 static void shader_arb_request_a0(const struct wined3d_shader_instruction *ins, const char *src)
 976 {
 977     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
 978     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
 979
 980     if (!strcmp(priv->addr_reg, src)) return;
 981
 982     strcpy(priv->addr_reg, src);
 983     shader_addline(buffer, "ARL A0.x, %s;\n", src);
 984 }
 985
 986 static void shader_arb_get_src_param(const struct wined3d_shader_instruction *ins,
 987         const struct wined3d_shader_src_param *src, unsigned int tmpreg, char *outregstr);
 988
 989 static void shader_arb_get_register_name(const struct wined3d_shader_instruction *ins,
 990         const struct wined3d_shader_register *reg, char *register_name, BOOL *is_color)
 991 {
 992     /* oPos, oFog and oPts in D3D */
 993     static const char * const rastout_reg_names[] = {"TMP_OUT", "TMP_FOGCOORD", "result.pointsize"};
 994     const struct wined3d_shader *shader = ins->ctx->shader;
 995     const struct wined3d_shader_reg_maps *reg_maps = ins->ctx->reg_maps;
 996     BOOL pshader = shader_is_pshader_version(reg_maps->shader_version.type);
 997     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
 998
 999     *is_color = FALSE;
1000
1001     switch (reg->type)
1002     {
1003         case WINED3DSPR_TEMP:
1004             sprintf(register_name, "R%u", reg->idx[0].offset);
1005             break;
1006
1007         case WINED3DSPR_INPUT:
1008             if (pshader)
1009             {
1010                 if (reg_maps->shader_version.major < 3)
1011                 {
1012                     if (!reg->idx[0].offset)
1013                         strcpy(register_name, "fragment.color.primary");
1014                     else
1015                         strcpy(register_name, "fragment.color.secondary");
1016                 }
1017                 else
1018                 {
1019                     if (reg->idx[0].rel_addr)
1020                     {
1021                         char rel_reg[50];
1022                         shader_arb_get_src_param(ins, reg->idx[0].rel_addr, 0, rel_reg);
1023
1024                         if (!strcmp(rel_reg, "**aL_emul**"))
1025                         {
1026                             DWORD idx = ctx->aL + reg->idx[0].offset;
1027                             if(idx < MAX_REG_INPUT)
1028                             {
1029                                 strcpy(register_name, ctx->ps_input[idx]);
1030                             }
1031                             else
1032                             {
1033                                 ERR("Pixel shader input register out of bounds: %u\n", idx);
1034                                 sprintf(register_name, "out_of_bounds_%u", idx);
1035                             }
1036                         }
1037                         else if (reg_maps->input_registers & 0x0300)
1038                         {
1039                             /* There are two ways basically:
1040                              *
1041                              * 1) Use the unrolling code that is used for loop emulation and unroll the loop.
1042                              *    That means trouble if the loop also contains a breakc or if the control values
1043                              *    aren't local constants.
1044                              * 2) Generate an if block that checks if aL.y < 8, == 8 or == 9 and selects the
1045                              *    source dynamically. The trouble is that we cannot simply read aL.y because it
1046                              *    is an ADDRESS register. We could however push it, load .zw with a value and use
1047                              *    ADAC to load the condition code register and pop it again afterwards
1048                              */
1049                             FIXME("Relative input register addressing with more than 8 registers\n");
1050
1051                             /* This is better than nothing for now */
1052                             sprintf(register_name, "fragment.texcoord[%s + %u]", rel_reg, reg->idx[0].offset);
1053                         }
1054                         else if(ctx->cur_ps_args->super.vp_mode != vertexshader)
1055                         {
1056                             /* This is problematic because we'd have to consult the ctx->ps_input strings
1057                              * for where to find the varying. Some may be "0.0", others can be texcoords or
1058                              * colors. This needs either a pipeline replacement to make the vertex shader feed
1059                              * proper varyings, or loop unrolling
1060                              *
1061                              * For now use the texcoords and hope for the best
1062                              */
1063                             FIXME("Non-vertex shader varying input with indirect addressing\n");
1064                             sprintf(register_name, "fragment.texcoord[%s + %u]", rel_reg, reg->idx[0].offset);
1065                         }
1066                         else
1067                         {
1068                             /* D3D supports indirect addressing only with aL in loop registers. The loop instruction
1069                              * pulls GL_NV_fragment_program2 in
1070                              */
1071                             sprintf(register_name, "fragment.texcoord[%s + %u]", rel_reg, reg->idx[0].offset);
1072                         }
1073                     }
1074                     else
1075                     {
1076                         if (reg->idx[0].offset < MAX_REG_INPUT)
1077                         {
1078                             strcpy(register_name, ctx->ps_input[reg->idx[0].offset]);
1079                         }
1080                         else
1081                         {
1082                             ERR("Pixel shader input register out of bounds: %u\n", reg->idx[0].offset);
1083                             sprintf(register_name, "out_of_bounds_%u", reg->idx[0].offset);
1084                         }
1085                     }
1086                 }
1087             }
1088             else
1089             {
1090                 if (ctx->cur_vs_args->super.swizzle_map & (1 << reg->idx[0].offset))
1091                     *is_color = TRUE;
1092                 sprintf(register_name, "vertex.attrib[%u]", reg->idx[0].offset);
1093             }
1094             break;
1095
1096         case WINED3DSPR_CONST:
1097             if (!pshader && reg->idx[0].rel_addr)
1098             {
1099                 const struct arb_vshader_private *shader_data = shader->backend_data;
1100                 UINT rel_offset = shader_data->rel_offset;
1101                 BOOL aL = FALSE;
1102                 char rel_reg[50];
1103                 if (reg_maps->shader_version.major < 2)
1104                 {
1105                     sprintf(rel_reg, "A0.x");
1106                 }
1107                 else
1108                 {
1109                     shader_arb_get_src_param(ins, reg->idx[0].rel_addr, 0, rel_reg);
1110                     if (ctx->target_version == ARB)
1111                     {
1112                         if (!strcmp(rel_reg, "**aL_emul**"))
1113                         {
1114                             aL = TRUE;
1115                         } else {
1116                             shader_arb_request_a0(ins, rel_reg);
1117                             sprintf(rel_reg, "A0.x");
1118                         }
1119                     }
1120                 }
1121                 if (aL)
1122                     sprintf(register_name, "C[%u]", ctx->aL + reg->idx[0].offset);
1123                 else if (reg->idx[0].offset >= rel_offset)
1124                     sprintf(register_name, "C[%s + %u]", rel_reg, reg->idx[0].offset - rel_offset);
1125                 else
1126                     sprintf(register_name, "C[%s - %u]", rel_reg, rel_offset - reg->idx[0].offset);
1127             }
1128             else
1129             {
1130                 if (reg_maps->usesrelconstF)
1131                     sprintf(register_name, "C[%u]", reg->idx[0].offset);
1132                 else
1133                     sprintf(register_name, "C%u", reg->idx[0].offset);
1134             }
1135             break;
1136
1137         case WINED3DSPR_TEXTURE: /* case WINED3DSPR_ADDR: */
1138             if (pshader)
1139             {
1140                 if (reg_maps->shader_version.major == 1
1141                         && reg_maps->shader_version.minor <= 3)
1142                     /* In ps <= 1.3, Tx is a temporary register as destination
1143                      * to all instructions, and as source to most instructions.
1144                      * For some instructions it is the texcoord input. Those
1145                      * instructions know about the special use. */
1146                     sprintf(register_name, "T%u", reg->idx[0].offset);
1147                 else
1148                     /* In ps 1.4 and 2.x Tx is always a (read-only) varying. */
1149                     sprintf(register_name, "fragment.texcoord[%u]", reg->idx[0].offset);
1150             }
1151             else
1152             {
1153                 if (reg_maps->shader_version.major == 1 || ctx->target_version >= NV2)
1154                     sprintf(register_name, "A%u", reg->idx[0].offset);
1155                 else
1156                     sprintf(register_name, "A%u_SHADOW", reg->idx[0].offset);
1157             }
1158             break;
1159
1160         case WINED3DSPR_COLOROUT:
1161             if (ctx->cur_ps_args->super.srgb_correction && !reg->idx[0].offset)
1162             {
1163                 strcpy(register_name, "TMP_COLOR");
1164             }
1165             else
1166             {
1167                 if (ctx->cur_ps_args->super.srgb_correction)
1168                     FIXME("sRGB correction on higher render targets.\n");
1169                 if (reg_maps->rt_mask > 1)
1170                     sprintf(register_name, "result.color[%u]", reg->idx[0].offset);
1171                 else
1172                     strcpy(register_name, "result.color");
1173             }
1174             break;
1175
1176         case WINED3DSPR_RASTOUT:
1177             if (reg->idx[0].offset == 1)
1178                 sprintf(register_name, "%s", ctx->fog_output);
1179             else
1180                 sprintf(register_name, "%s", rastout_reg_names[reg->idx[0].offset]);
1181             break;
1182
1183         case WINED3DSPR_DEPTHOUT:
1184             strcpy(register_name, "result.depth");
1185             break;
1186
1187         case WINED3DSPR_ATTROUT:
1188         /* case WINED3DSPR_OUTPUT: */
1189             if (pshader)
1190                 sprintf(register_name, "oD[%u]", reg->idx[0].offset);
1191             else
1192                 strcpy(register_name, ctx->color_output[reg->idx[0].offset]);
1193             break;
1194
1195         case WINED3DSPR_TEXCRDOUT:
1196             if (pshader)
1197                 sprintf(register_name, "oT[%u]", reg->idx[0].offset);
1198             else if (reg_maps->shader_version.major < 3)
1199                 strcpy(register_name, ctx->texcrd_output[reg->idx[0].offset]);
1200             else
1201                 strcpy(register_name, ctx->vs_output[reg->idx[0].offset]);
1202             break;
1203
1204         case WINED3DSPR_LOOP:
1205             if(ctx->target_version >= NV2)
1206             {
1207                 /* Pshader has an implicitly declared loop index counter A0.x that cannot be renamed */
1208                 if(pshader) sprintf(register_name, "A0.x");
1209                 else sprintf(register_name, "aL.y");
1210             }
1211             else
1212             {
1213                 /* Unfortunately this code cannot return the value of ctx->aL here. An immediate value
1214                  * would be valid, but if aL is used for indexing(its only use), there's likely an offset,
1215                  * thus the result would be something like C[15 + 30], which is not valid in the ARB program
1216                  * grammar. So return a marker for the emulated aL and intercept it in constant and varying
1217                  * indexing
1218                  */
1219                 sprintf(register_name, "**aL_emul**");
1220             }
1221
1222             break;
1223
1224         case WINED3DSPR_CONSTINT:
1225             sprintf(register_name, "I%u", reg->idx[0].offset);
1226             break;
1227
1228         case WINED3DSPR_MISCTYPE:
1229             if (!reg->idx[0].offset)
1230                 sprintf(register_name, "vpos");
1231             else if (reg->idx[0].offset == 1)
1232                 sprintf(register_name, "fragment.facing.x");
1233             else
1234                 FIXME("Unknown MISCTYPE register index %u.\n", reg->idx[0].offset);
1235             break;
1236
1237         default:
1238             FIXME("Unhandled register type %#x[%u].\n", reg->type, reg->idx[0].offset);
1239             sprintf(register_name, "unrecognized_register[%u]", reg->idx[0].offset);
1240             break;
1241     }
1242 }
1243
1244 static void shader_arb_get_dst_param(const struct wined3d_shader_instruction *ins,
1245         const struct wined3d_shader_dst_param *wined3d_dst, char *str)
1246 {
1247     char register_name[255];
1248     char write_mask[6];
1249     BOOL is_color;
1250
1251     shader_arb_get_register_name(ins, &wined3d_dst->reg, register_name, &is_color);
1252     strcpy(str, register_name);
1253
1254     shader_arb_get_write_mask(ins, wined3d_dst, write_mask);
1255     strcat(str, write_mask);
1256 }
1257
1258 static const char *shader_arb_get_fixup_swizzle(enum fixup_channel_source channel_source)
1259 {
1260     switch(channel_source)
1261     {
1262         case CHANNEL_SOURCE_ZERO: return "0";
1263         case CHANNEL_SOURCE_ONE: return "1";
1264         case CHANNEL_SOURCE_X: return "x";
1265         case CHANNEL_SOURCE_Y: return "y";
1266         case CHANNEL_SOURCE_Z: return "z";
1267         case CHANNEL_SOURCE_W: return "w";
1268         default:
1269             FIXME("Unhandled channel source %#x\n", channel_source);
1270             return "undefined";
1271     }
1272 }
1273
1274 static void gen_color_correction(struct wined3d_shader_buffer *buffer, const char *reg,
1275         DWORD dst_mask, const char *one, const char *two, struct color_fixup_desc fixup)
1276 {
1277     DWORD mask;
1278
1279     if (is_complex_fixup(fixup))
1280     {
1281         enum complex_fixup complex_fixup = get_complex_fixup(fixup);
1282         FIXME("Complex fixup (%#x) not supported\n", complex_fixup);
1283         return;
1284     }
1285
1286     mask = 0;
1287     if (fixup.x_source != CHANNEL_SOURCE_X) mask |= WINED3DSP_WRITEMASK_0;
1288     if (fixup.y_source != CHANNEL_SOURCE_Y) mask |= WINED3DSP_WRITEMASK_1;
1289     if (fixup.z_source != CHANNEL_SOURCE_Z) mask |= WINED3DSP_WRITEMASK_2;
1290     if (fixup.w_source != CHANNEL_SOURCE_W) mask |= WINED3DSP_WRITEMASK_3;
1291     mask &= dst_mask;
1292
1293     if (mask)
1294     {
1295         shader_addline(buffer, "SWZ %s, %s, %s, %s, %s, %s;\n", reg, reg,
1296                 shader_arb_get_fixup_swizzle(fixup.x_source), shader_arb_get_fixup_swizzle(fixup.y_source),
1297                 shader_arb_get_fixup_swizzle(fixup.z_source), shader_arb_get_fixup_swizzle(fixup.w_source));
1298     }
1299
1300     mask = 0;
1301     if (fixup.x_sign_fixup) mask |= WINED3DSP_WRITEMASK_0;
1302     if (fixup.y_sign_fixup) mask |= WINED3DSP_WRITEMASK_1;
1303     if (fixup.z_sign_fixup) mask |= WINED3DSP_WRITEMASK_2;
1304     if (fixup.w_sign_fixup) mask |= WINED3DSP_WRITEMASK_3;
1305     mask &= dst_mask;
1306
1307     if (mask)
1308     {
1309         char reg_mask[6];
1310         char *ptr = reg_mask;
1311
1312         if (mask != WINED3DSP_WRITEMASK_ALL)
1313         {
1314             *ptr++ = '.';
1315             if (mask & WINED3DSP_WRITEMASK_0) *ptr++ = 'x';
1316             if (mask & WINED3DSP_WRITEMASK_1) *ptr++ = 'y';
1317             if (mask & WINED3DSP_WRITEMASK_2) *ptr++ = 'z';
1318             if (mask & WINED3DSP_WRITEMASK_3) *ptr++ = 'w';
1319         }
1320         *ptr = '\0';
1321
1322         shader_addline(buffer, "MAD %s%s, %s, %s, -%s;\n", reg, reg_mask, reg, two, one);
1323     }
1324 }
1325
1326 static const char *shader_arb_get_modifier(const struct wined3d_shader_instruction *ins)
1327 {
1328     DWORD mod;
1329     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
1330     if (!ins->dst_count) return "";
1331
1332     mod = ins->dst[0].modifiers;
1333
1334     /* Silently ignore PARTIALPRECISION if its not supported */
1335     if(priv->target_version == ARB) mod &= ~WINED3DSPDM_PARTIALPRECISION;
1336
1337     if(mod & WINED3DSPDM_MSAMPCENTROID)
1338     {
1339         FIXME("Unhandled modifier WINED3DSPDM_MSAMPCENTROID\n");
1340         mod &= ~WINED3DSPDM_MSAMPCENTROID;
1341     }
1342
1343     switch(mod)
1344     {
1345         case WINED3DSPDM_SATURATE | WINED3DSPDM_PARTIALPRECISION:
1346             return "H_SAT";
1347
1348         case WINED3DSPDM_SATURATE:
1349             return "_SAT";
1350
1351         case WINED3DSPDM_PARTIALPRECISION:
1352             return "H";
1353
1354         case 0:
1355             return "";
1356
1357         default:
1358             FIXME("Unknown modifiers 0x%08x\n", mod);
1359             return "";
1360     }
1361 }
1362
1363 #define TEX_PROJ        0x1
1364 #define TEX_BIAS        0x2
1365 #define TEX_LOD         0x4
1366 #define TEX_DERIV       0x10
1367
1368 static void shader_hw_sample(const struct wined3d_shader_instruction *ins, DWORD sampler_idx,
1369         const char *dst_str, const char *coord_reg, WORD flags, const char *dsx, const char *dsy)
1370 {
1371     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1372     DWORD sampler_type = ins->ctx->reg_maps->sampler_type[sampler_idx];
1373     const char *tex_type;
1374     BOOL np2_fixup = FALSE;
1375     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
1376     const char *mod;
1377     BOOL pshader = shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type);
1378     const struct wined3d_shader *shader;
1379     const struct wined3d_device *device;
1380     const struct wined3d_gl_info *gl_info;
1381
1382     /* D3D vertex shader sampler IDs are vertex samplers(0-3), not global d3d samplers */
1383     if(!pshader) sampler_idx += MAX_FRAGMENT_SAMPLERS;
1384
1385     switch(sampler_type) {
1386         case WINED3DSTT_1D:
1387             tex_type = "1D";
1388             break;
1389
1390         case WINED3DSTT_2D:
1391             shader = ins->ctx->shader;
1392             device = shader->device;
1393             gl_info = &device->adapter->gl_info;
1394
1395             if (pshader && priv->cur_ps_args->super.np2_fixup & (1 << sampler_idx)
1396                     && gl_info->supported[ARB_TEXTURE_RECTANGLE])
1397                 tex_type = "RECT";
1398             else
1399                 tex_type = "2D";
1400             if (shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type))
1401             {
1402                 if (priv->cur_np2fixup_info->super.active & (1 << sampler_idx))
1403                 {
1404                     if (flags) FIXME("Only ordinary sampling from NP2 textures is supported.\n");
1405                     else np2_fixup = TRUE;
1406                 }
1407             }
1408             break;
1409
1410         case WINED3DSTT_VOLUME:
1411             tex_type = "3D";
1412             break;
1413
1414         case WINED3DSTT_CUBE:
1415             tex_type = "CUBE";
1416             break;
1417
1418         default:
1419             ERR("Unexpected texture type %d\n", sampler_type);
1420             tex_type = "";
1421     }
1422
1423     /* TEX, TXL, TXD and TXP do not support the "H" modifier,
1424      * so don't use shader_arb_get_modifier
1425      */
1426     if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE) mod = "_SAT";
1427     else mod = "";
1428
1429     /* Fragment samplers always have indentity mapping */
1430     if(sampler_idx >= MAX_FRAGMENT_SAMPLERS)
1431     {
1432         sampler_idx = priv->cur_vs_args->vertex.samplers[sampler_idx - MAX_FRAGMENT_SAMPLERS];
1433     }
1434
1435     if (flags & TEX_DERIV)
1436     {
1437         if(flags & TEX_PROJ) FIXME("Projected texture sampling with custom derivatives\n");
1438         if(flags & TEX_BIAS) FIXME("Biased texture sampling with custom derivatives\n");
1439         shader_addline(buffer, "TXD%s %s, %s, %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg,
1440                        dsx, dsy,sampler_idx, tex_type);
1441     }
1442     else if(flags & TEX_LOD)
1443     {
1444         if(flags & TEX_PROJ) FIXME("Projected texture sampling with explicit lod\n");
1445         if(flags & TEX_BIAS) FIXME("Biased texture sampling with explicit lod\n");
1446         shader_addline(buffer, "TXL%s %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg,
1447                        sampler_idx, tex_type);
1448     }
1449     else if (flags & TEX_BIAS)
1450     {
1451         /* Shouldn't be possible, but let's check for it */
1452         if(flags & TEX_PROJ) FIXME("Biased and Projected texture sampling\n");
1453         /* TXB takes the 4th component of the source vector automatically, as d3d. Nothing more to do */
1454         shader_addline(buffer, "TXB%s %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg, sampler_idx, tex_type);
1455     }
1456     else if (flags & TEX_PROJ)
1457     {
1458         shader_addline(buffer, "TXP%s %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg, sampler_idx, tex_type);
1459     }
1460     else
1461     {
1462         if (np2_fixup)
1463         {
1464             const unsigned char idx = priv->cur_np2fixup_info->super.idx[sampler_idx];
1465             shader_addline(buffer, "MUL TA, np2fixup[%u].%s, %s;\n", idx >> 1,
1466                            (idx % 2) ? "zwxy" : "xyzw", coord_reg);
1467
1468             shader_addline(buffer, "TEX%s %s, TA, texture[%u], %s;\n", mod, dst_str, sampler_idx, tex_type);
1469         }
1470         else
1471             shader_addline(buffer, "TEX%s %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg, sampler_idx, tex_type);
1472     }
1473
1474     if (pshader)
1475     {
1476         gen_color_correction(buffer, dst_str, ins->dst[0].write_mask,
1477                 arb_get_helper_value(WINED3D_SHADER_TYPE_PIXEL, ARB_ONE),
1478                 arb_get_helper_value(WINED3D_SHADER_TYPE_PIXEL, ARB_TWO),
1479                 priv->cur_ps_args->super.color_fixup[sampler_idx]);
1480     }
1481 }
1482
1483 static void shader_arb_get_src_param(const struct wined3d_shader_instruction *ins,
1484         const struct wined3d_shader_src_param *src, unsigned int tmpreg, char *outregstr)
1485 {
1486     /* Generate a line that does the input modifier computation and return the input register to use */
1487     BOOL is_color = FALSE;
1488     char regstr[256];
1489     char swzstr[20];
1490     int insert_line;
1491     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1492     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
1493     const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
1494     const char *two = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_TWO);
1495
1496     /* Assume a new line will be added */
1497     insert_line = 1;
1498
1499     /* Get register name */
1500     shader_arb_get_register_name(ins, &src->reg, regstr, &is_color);
1501     shader_arb_get_swizzle(src, is_color, swzstr);
1502
1503     switch (src->modifiers)
1504     {
1505     case WINED3DSPSM_NONE:
1506         sprintf(outregstr, "%s%s", regstr, swzstr);
1507         insert_line = 0;
1508         break;
1509     case WINED3DSPSM_NEG:
1510         sprintf(outregstr, "-%s%s", regstr, swzstr);
1511         insert_line = 0;
1512         break;
1513     case WINED3DSPSM_BIAS:
1514         shader_addline(buffer, "ADD T%c, %s, -coefdiv.x;\n", 'A' + tmpreg, regstr);
1515         break;
1516     case WINED3DSPSM_BIASNEG:
1517         shader_addline(buffer, "ADD T%c, -%s, coefdiv.x;\n", 'A' + tmpreg, regstr);
1518         break;
1519     case WINED3DSPSM_SIGN:
1520         shader_addline(buffer, "MAD T%c, %s, %s, -%s;\n", 'A' + tmpreg, regstr, two, one);
1521         break;
1522     case WINED3DSPSM_SIGNNEG:
1523         shader_addline(buffer, "MAD T%c, %s, -%s, %s;\n", 'A' + tmpreg, regstr, two, one);
1524         break;
1525     case WINED3DSPSM_COMP:
1526         shader_addline(buffer, "SUB T%c, %s, %s;\n", 'A' + tmpreg, one, regstr);
1527         break;
1528     case WINED3DSPSM_X2:
1529         shader_addline(buffer, "ADD T%c, %s, %s;\n", 'A' + tmpreg, regstr, regstr);
1530         break;
1531     case WINED3DSPSM_X2NEG:
1532         shader_addline(buffer, "ADD T%c, -%s, -%s;\n", 'A' + tmpreg, regstr, regstr);
1533         break;
1534     case WINED3DSPSM_DZ:
1535         shader_addline(buffer, "RCP T%c, %s.z;\n", 'A' + tmpreg, regstr);
1536         shader_addline(buffer, "MUL T%c, %s, T%c;\n", 'A' + tmpreg, regstr, 'A' + tmpreg);
1537         break;
1538     case WINED3DSPSM_DW:
1539         shader_addline(buffer, "RCP T%c, %s.w;\n", 'A' + tmpreg, regstr);
1540         shader_addline(buffer, "MUL T%c, %s, T%c;\n", 'A' + tmpreg, regstr, 'A' + tmpreg);
1541         break;
1542     case WINED3DSPSM_ABS:
1543         if(ctx->target_version >= NV2) {
1544             sprintf(outregstr, "|%s%s|", regstr, swzstr);
1545             insert_line = 0;
1546         } else {
1547             shader_addline(buffer, "ABS T%c, %s;\n", 'A' + tmpreg, regstr);
1548         }
1549         break;
1550     case WINED3DSPSM_ABSNEG:
1551         if(ctx->target_version >= NV2) {
1552             sprintf(outregstr, "-|%s%s|", regstr, swzstr);
1553         } else {
1554             shader_addline(buffer, "ABS T%c, %s;\n", 'A' + tmpreg, regstr);
1555             sprintf(outregstr, "-T%c%s", 'A' + tmpreg, swzstr);
1556         }
1557         insert_line = 0;
1558         break;
1559     default:
1560         sprintf(outregstr, "%s%s", regstr, swzstr);
1561         insert_line = 0;
1562     }
1563
1564     /* Return modified or original register, with swizzle */
1565     if (insert_line)
1566         sprintf(outregstr, "T%c%s", 'A' + tmpreg, swzstr);
1567 }
1568
1569 static void pshader_hw_bem(const struct wined3d_shader_instruction *ins)
1570 {
1571     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1572     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1573     DWORD sampler_code = dst->reg.idx[0].offset;
1574     char dst_name[50];
1575     char src_name[2][50];
1576
1577     shader_arb_get_dst_param(ins, dst, dst_name);
1578
1579     /* Sampling the perturbation map in Tsrc was done already, including the signedness correction if needed
1580      *
1581      * Keep in mind that src_name[1] can be "TB" and src_name[0] can be "TA" because modifiers like _x2 are valid
1582      * with bem. So delay loading the first parameter until after the perturbation calculation which needs two
1583      * temps is done.
1584      */
1585     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1586     shader_addline(buffer, "SWZ TA, bumpenvmat%d, x, z, 0, 0;\n", sampler_code);
1587     shader_addline(buffer, "DP3 TC.r, TA, %s;\n", src_name[1]);
1588     shader_addline(buffer, "SWZ TA, bumpenvmat%d, y, w, 0, 0;\n", sampler_code);
1589     shader_addline(buffer, "DP3 TC.g, TA, %s;\n", src_name[1]);
1590
1591     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
1592     shader_addline(buffer, "ADD %s, %s, TC;\n", dst_name, src_name[0]);
1593 }
1594
1595 static DWORD negate_modifiers(DWORD mod, char *extra_char)
1596 {
1597     *extra_char = ' ';
1598     switch(mod)
1599     {
1600         case WINED3DSPSM_NONE:      return WINED3DSPSM_NEG;
1601         case WINED3DSPSM_NEG:       return WINED3DSPSM_NONE;
1602         case WINED3DSPSM_BIAS:      return WINED3DSPSM_BIASNEG;
1603         case WINED3DSPSM_BIASNEG:   return WINED3DSPSM_BIAS;
1604         case WINED3DSPSM_SIGN:      return WINED3DSPSM_SIGNNEG;
1605         case WINED3DSPSM_SIGNNEG:   return WINED3DSPSM_SIGN;
1606         case WINED3DSPSM_COMP:      *extra_char = '-'; return WINED3DSPSM_COMP;
1607         case WINED3DSPSM_X2:        return WINED3DSPSM_X2NEG;
1608         case WINED3DSPSM_X2NEG:     return WINED3DSPSM_X2;
1609         case WINED3DSPSM_DZ:        *extra_char = '-'; return WINED3DSPSM_DZ;
1610         case WINED3DSPSM_DW:        *extra_char = '-'; return WINED3DSPSM_DW;
1611         case WINED3DSPSM_ABS:       return WINED3DSPSM_ABSNEG;
1612         case WINED3DSPSM_ABSNEG:    return WINED3DSPSM_ABS;
1613     }
1614     FIXME("Unknown modifier %u\n", mod);
1615     return mod;
1616 }
1617
1618 static void pshader_hw_cnd(const struct wined3d_shader_instruction *ins)
1619 {
1620     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1621     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1622     char dst_name[50];
1623     char src_name[3][50];
1624     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
1625             ins->ctx->reg_maps->shader_version.minor);
1626
1627     shader_arb_get_dst_param(ins, dst, dst_name);
1628     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1629
1630     /* The coissue flag changes the semantic of the cnd instruction in <= 1.3 shaders */
1631     if (shader_version <= WINED3D_SHADER_VERSION(1, 3) && ins->coissue)
1632     {
1633         shader_addline(buffer, "MOV%s %s, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name[1]);
1634     }
1635     else
1636     {
1637         struct wined3d_shader_src_param src0_copy = ins->src[0];
1638         char extra_neg;
1639
1640         /* src0 may have a negate srcmod set, so we can't blindly add "-" to the name */
1641         src0_copy.modifiers = negate_modifiers(src0_copy.modifiers, &extra_neg);
1642
1643         shader_arb_get_src_param(ins, &src0_copy, 0, src_name[0]);
1644         shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
1645         shader_addline(buffer, "ADD TA, %c%s, coefdiv.x;\n", extra_neg, src_name[0]);
1646         shader_addline(buffer, "CMP%s %s, TA, %s, %s;\n", shader_arb_get_modifier(ins),
1647                 dst_name, src_name[1], src_name[2]);
1648     }
1649 }
1650
1651 static void pshader_hw_cmp(const struct wined3d_shader_instruction *ins)
1652 {
1653     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1654     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1655     char dst_name[50];
1656     char src_name[3][50];
1657
1658     shader_arb_get_dst_param(ins, dst, dst_name);
1659
1660     /* Generate input register names (with modifiers) */
1661     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
1662     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1663     shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
1664
1665     shader_addline(buffer, "CMP%s %s, %s, %s, %s;\n", shader_arb_get_modifier(ins),
1666             dst_name, src_name[0], src_name[2], src_name[1]);
1667 }
1668
1669 /** Process the WINED3DSIO_DP2ADD instruction in ARB.
1670  * dst = dot2(src0, src1) + src2 */
1671 static void pshader_hw_dp2add(const struct wined3d_shader_instruction *ins)
1672 {
1673     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1674     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1675     char dst_name[50];
1676     char src_name[3][50];
1677     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
1678
1679     shader_arb_get_dst_param(ins, dst, dst_name);
1680     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
1681     shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
1682
1683     if(ctx->target_version >= NV3)
1684     {
1685         /* GL_NV_fragment_program2 has a 1:1 matching instruction */
1686         shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1687         shader_addline(buffer, "DP2A%s %s, %s, %s, %s;\n", shader_arb_get_modifier(ins),
1688                        dst_name, src_name[0], src_name[1], src_name[2]);
1689     }
1690     else if(ctx->target_version >= NV2)
1691     {
1692         /* dst.x = src2.?, src0.x, src1.x + src0.y * src1.y
1693          * dst.y = src2.?, src0.x, src1.z + src0.y * src1.w
1694          * dst.z = src2.?, src0.x, src1.x + src0.y * src1.y
1695          * dst.z = src2.?, src0.x, src1.z + src0.y * src1.w
1696          *
1697          * Make sure that src1.zw = src1.xy, then we get a classic dp2add
1698          *
1699          * .xyxy and other swizzles that we could get with this are not valid in
1700          * plain ARBfp, but luckily the NV extension grammar lifts this limitation.
1701          */
1702         struct wined3d_shader_src_param tmp_param = ins->src[1];
1703         DWORD swizzle = tmp_param.swizzle & 0xf; /* Selects .xy */
1704         tmp_param.swizzle = swizzle | (swizzle << 4); /* Creates .xyxy */
1705
1706         shader_arb_get_src_param(ins, &tmp_param, 1, src_name[1]);
1707
1708         shader_addline(buffer, "X2D%s %s, %s, %s, %s;\n", shader_arb_get_modifier(ins),
1709                        dst_name, src_name[2], src_name[0], src_name[1]);
1710     }
1711     else
1712     {
1713         shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1714         /* 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]
1715         * src_name[0] can be TA, but TA is a private temp for modifiers, so it is save to overwrite
1716         */
1717         shader_addline(buffer, "MOV TA, %s;\n", src_name[0]);
1718         shader_addline(buffer, "MOV TA.z, 0.0;\n");
1719         shader_addline(buffer, "DP3 TA, TA, %s;\n", src_name[1]);
1720         shader_addline(buffer, "ADD%s %s, TA, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name[2]);
1721     }
1722 }
1723
1724 /* Map the opcode 1-to-1 to the GL code */
1725 static void shader_hw_map2gl(const struct wined3d_shader_instruction *ins)
1726 {
1727     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1728     const char *instruction;
1729     char arguments[256], dst_str[50];
1730     unsigned int i;
1731     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1732
1733     switch (ins->handler_idx)
1734     {
1735         case WINED3DSIH_ABS: instruction = "ABS"; break;
1736         case WINED3DSIH_ADD: instruction = "ADD"; break;
1737         case WINED3DSIH_CRS: instruction = "XPD"; break;
1738         case WINED3DSIH_DP3: instruction = "DP3"; break;
1739         case WINED3DSIH_DP4: instruction = "DP4"; break;
1740         case WINED3DSIH_DST: instruction = "DST"; break;
1741         case WINED3DSIH_FRC: instruction = "FRC"; break;
1742         case WINED3DSIH_LIT: instruction = "LIT"; break;
1743         case WINED3DSIH_LRP: instruction = "LRP"; break;
1744         case WINED3DSIH_MAD: instruction = "MAD"; break;
1745         case WINED3DSIH_MAX: instruction = "MAX"; break;
1746         case WINED3DSIH_MIN: instruction = "MIN"; break;
1747         case WINED3DSIH_MOV: instruction = "MOV"; break;
1748         case WINED3DSIH_MUL: instruction = "MUL"; break;
1749         case WINED3DSIH_SGE: instruction = "SGE"; break;
1750         case WINED3DSIH_SLT: instruction = "SLT"; break;
1751         case WINED3DSIH_SUB: instruction = "SUB"; break;
1752         case WINED3DSIH_MOVA:instruction = "ARR"; break;
1753         case WINED3DSIH_DSX: instruction = "DDX"; break;
1754         default: instruction = "";
1755             FIXME("Unhandled opcode %#x\n", ins->handler_idx);
1756             break;
1757     }
1758
1759     /* Note that shader_arb_add_dst_param() adds spaces. */
1760     arguments[0] = '\0';
1761     shader_arb_get_dst_param(ins, dst, dst_str);
1762     for (i = 0; i < ins->src_count; ++i)
1763     {
1764         char operand[100];
1765         strcat(arguments, ", ");
1766         shader_arb_get_src_param(ins, &ins->src[i], i, operand);
1767         strcat(arguments, operand);
1768     }
1769     shader_addline(buffer, "%s%s %s%s;\n", instruction, shader_arb_get_modifier(ins), dst_str, arguments);
1770 }
1771
1772 static void shader_hw_nop(const struct wined3d_shader_instruction *ins) {}
1773
1774 static void shader_hw_mov(const struct wined3d_shader_instruction *ins)
1775 {
1776     const struct wined3d_shader *shader = ins->ctx->shader;
1777     const struct wined3d_shader_reg_maps *reg_maps = ins->ctx->reg_maps;
1778     BOOL pshader = shader_is_pshader_version(reg_maps->shader_version.type);
1779     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
1780     const char *zero = arb_get_helper_value(reg_maps->shader_version.type, ARB_ZERO);
1781     const char *one = arb_get_helper_value(reg_maps->shader_version.type, ARB_ONE);
1782     const char *two = arb_get_helper_value(reg_maps->shader_version.type, ARB_TWO);
1783
1784     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1785     char src0_param[256];
1786
1787     if (ins->handler_idx == WINED3DSIH_MOVA)
1788     {
1789         const struct arb_vshader_private *shader_data = shader->backend_data;
1790         char write_mask[6];
1791         const char *offset = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_VS_REL_OFFSET);
1792
1793         if(ctx->target_version >= NV2) {
1794             shader_hw_map2gl(ins);
1795             return;
1796         }
1797         shader_arb_get_src_param(ins, &ins->src[0], 0, src0_param);
1798         shader_arb_get_write_mask(ins, &ins->dst[0], write_mask);
1799
1800         /* This implements the mova formula used in GLSL. The first two instructions
1801          * prepare the sign() part. Note that it is fine to have my_sign(0.0) = 1.0
1802          * in this case:
1803          * mova A0.x, 0.0
1804          *
1805          * A0.x = arl(floor(abs(0.0) + 0.5) * 1.0) = floor(0.5) = 0.0 since arl does a floor
1806          *
1807          * The ARL is performed when A0 is used - the requested component is read from A0_SHADOW into
1808          * A0.x. We can use the overwritten component of A0_shadow as temporary storage for the sign.
1809          */
1810         shader_addline(buffer, "SGE A0_SHADOW%s, %s, %s;\n", write_mask, src0_param, zero);
1811         shader_addline(buffer, "MAD A0_SHADOW%s, A0_SHADOW, %s, -%s;\n", write_mask, two, one);
1812
1813         shader_addline(buffer, "ABS TA%s, %s;\n", write_mask, src0_param);
1814         shader_addline(buffer, "ADD TA%s, TA, rel_addr_const.x;\n", write_mask);
1815         shader_addline(buffer, "FLR TA%s, TA;\n", write_mask);
1816         if (shader_data->rel_offset)
1817         {
1818             shader_addline(buffer, "ADD TA%s, TA, %s;\n", write_mask, offset);
1819         }
1820         shader_addline(buffer, "MUL A0_SHADOW%s, TA, A0_SHADOW;\n", write_mask);
1821
1822         ((struct shader_arb_ctx_priv *)ins->ctx->backend_data)->addr_reg[0] = '\0';
1823     }
1824     else if (reg_maps->shader_version.major == 1
1825           && !shader_is_pshader_version(reg_maps->shader_version.type)
1826           && ins->dst[0].reg.type == WINED3DSPR_ADDR)
1827     {
1828         const struct arb_vshader_private *shader_data = shader->backend_data;
1829         src0_param[0] = '\0';
1830
1831         if (shader_data->rel_offset)
1832         {
1833             const char *offset = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_VS_REL_OFFSET);
1834             shader_arb_get_src_param(ins, &ins->src[0], 0, src0_param);
1835             shader_addline(buffer, "ADD TA.x, %s, %s;\n", src0_param, offset);
1836             shader_addline(buffer, "ARL A0.x, TA.x;\n");
1837         }
1838         else
1839         {
1840             /* Apple's ARB_vertex_program implementation does not accept an ARL source argument
1841              * with more than one component. Thus replicate the first source argument over all
1842              * 4 components. For example, .xyzw -> .x (or better: .xxxx), .zwxy -> .z, etc) */
1843             struct wined3d_shader_src_param tmp_src = ins->src[0];
1844             tmp_src.swizzle = (tmp_src.swizzle & 0x3) * 0x55;
1845             shader_arb_get_src_param(ins, &tmp_src, 0, src0_param);
1846             shader_addline(buffer, "ARL A0.x, %s;\n", src0_param);
1847         }
1848     }
1849     else if (ins->dst[0].reg.type == WINED3DSPR_COLOROUT && !ins->dst[0].reg.idx[0].offset && pshader)
1850     {
1851         if (ctx->cur_ps_args->super.srgb_correction && shader->u.ps.color0_mov)
1852         {
1853             shader_addline(buffer, "#mov handled in srgb write code\n");
1854             return;
1855         }
1856         shader_hw_map2gl(ins);
1857     }
1858     else
1859     {
1860         shader_hw_map2gl(ins);
1861     }
1862 }
1863
1864 static void pshader_hw_texkill(const struct wined3d_shader_instruction *ins)
1865 {
1866     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1867     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1868     char reg_dest[40];
1869
1870     /* No swizzles are allowed in d3d's texkill. PS 1.x ignores the 4th component as documented,
1871      * but >= 2.0 honors it (undocumented, but tested by the d3d9 testsuite)
1872      */
1873     shader_arb_get_dst_param(ins, dst, reg_dest);
1874
1875     if (ins->ctx->reg_maps->shader_version.major >= 2)
1876     {
1877         const char *kilsrc = "TA";
1878         BOOL is_color;
1879
1880         shader_arb_get_register_name(ins, &dst->reg, reg_dest, &is_color);
1881         if(dst->write_mask == WINED3DSP_WRITEMASK_ALL)
1882         {
1883             kilsrc = reg_dest;
1884         }
1885         else
1886         {
1887             /* Sigh. KIL doesn't support swizzles/writemasks. KIL passes a writemask, but ".xy" for example
1888              * is not valid as a swizzle in ARB (needs ".xyyy"). Use SWZ to load the register properly, and set
1889              * masked out components to 0(won't kill)
1890              */
1891             char x = '0', y = '0', z = '0', w = '0';
1892             if(dst->write_mask & WINED3DSP_WRITEMASK_0) x = 'x';
1893             if(dst->write_mask & WINED3DSP_WRITEMASK_1) y = 'y';
1894             if(dst->write_mask & WINED3DSP_WRITEMASK_2) z = 'z';
1895             if(dst->write_mask & WINED3DSP_WRITEMASK_3) w = 'w';
1896             shader_addline(buffer, "SWZ TA, %s, %c, %c, %c, %c;\n", reg_dest, x, y, z, w);
1897         }
1898         shader_addline(buffer, "KIL %s;\n", kilsrc);
1899     }
1900     else
1901     {
1902         /* ARB fp doesn't like swizzles on the parameter of the KIL instruction. To mask the 4th component,
1903          * copy the register into our general purpose TMP variable, overwrite .w and pass TMP to KIL
1904          *
1905          * ps_1_3 shaders use the texcoord incarnation of the Tx register. ps_1_4 shaders can use the same,
1906          * or pass in any temporary register(in shader phase 2)
1907          */
1908         if (ins->ctx->reg_maps->shader_version.minor <= 3)
1909             sprintf(reg_dest, "fragment.texcoord[%u]", dst->reg.idx[0].offset);
1910         else
1911             shader_arb_get_dst_param(ins, dst, reg_dest);
1912         shader_addline(buffer, "SWZ TA, %s, x, y, z, 1;\n", reg_dest);
1913         shader_addline(buffer, "KIL TA;\n");
1914     }
1915 }
1916
1917 static void pshader_hw_tex(const struct wined3d_shader_instruction *ins)
1918 {
1919     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
1920     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1921     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
1922             ins->ctx->reg_maps->shader_version.minor);
1923     struct wined3d_shader_src_param src;
1924
1925     char reg_dest[40];
1926     char reg_coord[40];
1927     DWORD reg_sampler_code;
1928     WORD myflags = 0;
1929     BOOL swizzle_coord = FALSE;
1930
1931     /* All versions have a destination register */
1932     shader_arb_get_dst_param(ins, dst, reg_dest);
1933
1934     /* 1.0-1.4: Use destination register number as texture code.
1935        2.0+: Use provided sampler number as texure code. */
1936     if (shader_version < WINED3D_SHADER_VERSION(2,0))
1937         reg_sampler_code = dst->reg.idx[0].offset;
1938     else
1939         reg_sampler_code = ins->src[1].reg.idx[0].offset;
1940
1941     /* 1.0-1.3: Use the texcoord varying.
1942        1.4+: Use provided coordinate source register. */
1943     if (shader_version < WINED3D_SHADER_VERSION(1,4))
1944         sprintf(reg_coord, "fragment.texcoord[%u]", reg_sampler_code);
1945     else {
1946         /* TEX is the only instruction that can handle DW and DZ natively */
1947         src = ins->src[0];
1948         if(src.modifiers == WINED3DSPSM_DW) src.modifiers = WINED3DSPSM_NONE;
1949         if(src.modifiers == WINED3DSPSM_DZ) src.modifiers = WINED3DSPSM_NONE;
1950         shader_arb_get_src_param(ins, &src, 0, reg_coord);
1951     }
1952
1953     /* projection flag:
1954      * 1.1, 1.2, 1.3: Use WINED3D_TSS_TEXTURETRANSFORMFLAGS
1955      * 1.4: Use WINED3DSPSM_DZ or WINED3DSPSM_DW on src[0]
1956      * 2.0+: Use WINED3DSI_TEXLD_PROJECT on the opcode
1957      */
1958     if (shader_version < WINED3D_SHADER_VERSION(1,4))
1959     {
1960         DWORD flags = 0;
1961         if (reg_sampler_code < MAX_TEXTURES)
1962             flags = priv->cur_ps_args->super.tex_transform >> reg_sampler_code * WINED3D_PSARGS_TEXTRANSFORM_SHIFT;
1963         if (flags & WINED3D_PSARGS_PROJECTED)
1964         {
1965             myflags |= TEX_PROJ;
1966             if ((flags & ~WINED3D_PSARGS_PROJECTED) == WINED3D_TTFF_COUNT3)
1967                 swizzle_coord = TRUE;
1968         }
1969     }
1970     else if (shader_version < WINED3D_SHADER_VERSION(2,0))
1971     {
1972         enum wined3d_shader_src_modifier src_mod = ins->src[0].modifiers;
1973         if (src_mod == WINED3DSPSM_DZ)
1974         {
1975             swizzle_coord = TRUE;
1976             myflags |= TEX_PROJ;
1977         } else if(src_mod == WINED3DSPSM_DW) {
1978             myflags |= TEX_PROJ;
1979         }
1980     } else {
1981         if (ins->flags & WINED3DSI_TEXLD_PROJECT) myflags |= TEX_PROJ;
1982         if (ins->flags & WINED3DSI_TEXLD_BIAS) myflags |= TEX_BIAS;
1983     }
1984
1985     if (swizzle_coord)
1986     {
1987         /* TXP cannot handle DZ natively, so move the z coordinate to .w.
1988          * reg_coord is a read-only varying register, so we need a temp reg */
1989         shader_addline(ins->ctx->buffer, "SWZ TA, %s, x, y, z, z;\n", reg_coord);
1990         strcpy(reg_coord, "TA");
1991     }
1992
1993     shader_hw_sample(ins, reg_sampler_code, reg_dest, reg_coord, myflags, NULL, NULL);
1994 }
1995
1996 static void pshader_hw_texcoord(const struct wined3d_shader_instruction *ins)
1997 {
1998     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1999     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2000     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
2001             ins->ctx->reg_maps->shader_version.minor);
2002     char dst_str[50];
2003
2004     if (shader_version < WINED3D_SHADER_VERSION(1,4))
2005     {
2006         DWORD reg = dst->reg.idx[0].offset;
2007
2008         shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2009         shader_addline(buffer, "MOV_SAT %s, fragment.texcoord[%u];\n", dst_str, reg);
2010     } else {
2011         char reg_src[40];
2012
2013         shader_arb_get_src_param(ins, &ins->src[0], 0, reg_src);
2014         shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2015         shader_addline(buffer, "MOV %s, %s;\n", dst_str, reg_src);
2016     }
2017 }
2018
2019 static void pshader_hw_texreg2ar(const struct wined3d_shader_instruction *ins)
2020 {
2021      struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2022      DWORD flags = 0;
2023
2024      DWORD reg1 = ins->dst[0].reg.idx[0].offset;
2025      char dst_str[50];
2026      char src_str[50];
2027
2028      /* Note that texreg2ar treats Tx as a temporary register, not as a varying */
2029      shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2030      shader_arb_get_src_param(ins, &ins->src[0], 0, src_str);
2031      /* Move .x first in case src_str is "TA" */
2032      shader_addline(buffer, "MOV TA.y, %s.x;\n", src_str);
2033      shader_addline(buffer, "MOV TA.x, %s.w;\n", src_str);
2034      if (reg1 < MAX_TEXTURES)
2035      {
2036          struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2037          flags = priv->cur_ps_args->super.tex_transform >> reg1 * WINED3D_PSARGS_TEXTRANSFORM_SHIFT;
2038      }
2039      shader_hw_sample(ins, reg1, dst_str, "TA", flags & WINED3D_PSARGS_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2040 }
2041
2042 static void pshader_hw_texreg2gb(const struct wined3d_shader_instruction *ins)
2043 {
2044      struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2045
2046      DWORD reg1 = ins->dst[0].reg.idx[0].offset;
2047      char dst_str[50];
2048      char src_str[50];
2049
2050      /* Note that texreg2gb treats Tx as a temporary register, not as a varying */
2051      shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2052      shader_arb_get_src_param(ins, &ins->src[0], 0, src_str);
2053      shader_addline(buffer, "MOV TA.x, %s.y;\n", src_str);
2054      shader_addline(buffer, "MOV TA.y, %s.z;\n", src_str);
2055      shader_hw_sample(ins, reg1, dst_str, "TA", 0, NULL, NULL);
2056 }
2057
2058 static void pshader_hw_texreg2rgb(const struct wined3d_shader_instruction *ins)
2059 {
2060     DWORD reg1 = ins->dst[0].reg.idx[0].offset;
2061     char dst_str[50];
2062     char src_str[50];
2063
2064     /* Note that texreg2rg treats Tx as a temporary register, not as a varying */
2065     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2066     shader_arb_get_src_param(ins, &ins->src[0], 0, src_str);
2067     shader_hw_sample(ins, reg1, dst_str, src_str, 0, NULL, NULL);
2068 }
2069
2070 static void pshader_hw_texbem(const struct wined3d_shader_instruction *ins)
2071 {
2072     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2073     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2074     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2075     char reg_coord[40], dst_reg[50], src_reg[50];
2076     DWORD reg_dest_code;
2077
2078     /* All versions have a destination register. The Tx where the texture coordinates come
2079      * from is the varying incarnation of the texture register
2080      */
2081     reg_dest_code = dst->reg.idx[0].offset;
2082     shader_arb_get_dst_param(ins, &ins->dst[0], dst_reg);
2083     shader_arb_get_src_param(ins, &ins->src[0], 0, src_reg);
2084     sprintf(reg_coord, "fragment.texcoord[%u]", reg_dest_code);
2085
2086     /* Sampling the perturbation map in Tsrc was done already, including the signedness correction if needed
2087      * The Tx in which the perturbation map is stored is the tempreg incarnation of the texture register
2088      *
2089      * GL_NV_fragment_program_option could handle this in one instruction via X2D:
2090      * X2D TA.xy, fragment.texcoord, T%u, bumpenvmat%u.xzyw
2091      *
2092      * However, the NV extensions are never enabled for <= 2.0 shaders because of the performance penalty that
2093      * comes with it, and texbem is an 1.x only instruction. No 1.x instruction forces us to enable the NV
2094      * extension.
2095      */
2096     shader_addline(buffer, "SWZ TB, bumpenvmat%d, x, z, 0, 0;\n", reg_dest_code);
2097     shader_addline(buffer, "DP3 TA.x, TB, %s;\n", src_reg);
2098     shader_addline(buffer, "SWZ TB, bumpenvmat%d, y, w, 0, 0;\n", reg_dest_code);
2099     shader_addline(buffer, "DP3 TA.y, TB, %s;\n", src_reg);
2100
2101     /* with projective textures, texbem only divides the static texture coord, not the displacement,
2102      * so we can't let the GL handle this.
2103      */
2104     if ((priv->cur_ps_args->super.tex_transform >> reg_dest_code * WINED3D_PSARGS_TEXTRANSFORM_SHIFT)
2105             & WINED3D_PSARGS_PROJECTED)
2106     {
2107         shader_addline(buffer, "RCP TB.w, %s.w;\n", reg_coord);
2108         shader_addline(buffer, "MUL TB.xy, %s, TB.w;\n", reg_coord);
2109         shader_addline(buffer, "ADD TA.xy, TA, TB;\n");
2110     } else {
2111         shader_addline(buffer, "ADD TA.xy, TA, %s;\n", reg_coord);
2112     }
2113
2114     shader_hw_sample(ins, reg_dest_code, dst_reg, "TA", 0, NULL, NULL);
2115
2116     if (ins->handler_idx == WINED3DSIH_TEXBEML)
2117     {
2118         /* No src swizzles are allowed, so this is ok */
2119         shader_addline(buffer, "MAD TA, %s.z, luminance%d.x, luminance%d.y;\n",
2120                        src_reg, reg_dest_code, reg_dest_code);
2121         shader_addline(buffer, "MUL %s, %s, TA;\n", dst_reg, dst_reg);
2122     }
2123 }
2124
2125 static void pshader_hw_texm3x2pad(const struct wined3d_shader_instruction *ins)
2126 {
2127     DWORD reg = ins->dst[0].reg.idx[0].offset;
2128     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2129     char src0_name[50], dst_name[50];
2130     BOOL is_color;
2131     struct wined3d_shader_register tmp_reg = ins->dst[0].reg;
2132
2133     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2134     /* The next instruction will be a texm3x2tex or texm3x2depth that writes to the uninitialized
2135      * T<reg+1> register. Use this register to store the calculated vector
2136      */
2137     tmp_reg.idx[0].offset = reg + 1;
2138     shader_arb_get_register_name(ins, &tmp_reg, dst_name, &is_color);
2139     shader_addline(buffer, "DP3 %s.x, fragment.texcoord[%u], %s;\n", dst_name, reg, src0_name);
2140 }
2141
2142 static void pshader_hw_texm3x2tex(const struct wined3d_shader_instruction *ins)
2143 {
2144     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2145     DWORD flags;
2146     DWORD reg = ins->dst[0].reg.idx[0].offset;
2147     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2148     char dst_str[50];
2149     char src0_name[50];
2150     char dst_reg[50];
2151     BOOL is_color;
2152
2153     /* We know that we're writing to the uninitialized T<reg> register, so use it for temporary storage */
2154     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_reg, &is_color);
2155
2156     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2157     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2158     shader_addline(buffer, "DP3 %s.y, fragment.texcoord[%u], %s;\n", dst_reg, reg, src0_name);
2159     flags = reg < MAX_TEXTURES ? priv->cur_ps_args->super.tex_transform >> reg * WINED3D_PSARGS_TEXTRANSFORM_SHIFT : 0;
2160     shader_hw_sample(ins, reg, dst_str, dst_reg, flags & WINED3D_PSARGS_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2161 }
2162
2163 static void pshader_hw_texm3x3pad(const struct wined3d_shader_instruction *ins)
2164 {
2165     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
2166     DWORD reg = ins->dst[0].reg.idx[0].offset;
2167     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2168     char src0_name[50], dst_name[50];
2169     struct wined3d_shader_register tmp_reg = ins->dst[0].reg;
2170     BOOL is_color;
2171
2172     /* There are always 2 texm3x3pad instructions followed by one texm3x3[tex,vspec, ...] instruction, with
2173      * incrementing ins->dst[0].register_idx numbers. So the pad instruction already knows the final destination
2174      * register, and this register is uninitialized(otherwise the assembler complains that it is 'redeclared')
2175      */
2176     tmp_reg.idx[0].offset = reg + 2 - tex_mx->current_row;
2177     shader_arb_get_register_name(ins, &tmp_reg, dst_name, &is_color);
2178
2179     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2180     shader_addline(buffer, "DP3 %s.%c, fragment.texcoord[%u], %s;\n",
2181                    dst_name, 'x' + tex_mx->current_row, reg, src0_name);
2182     tex_mx->texcoord_w[tex_mx->current_row++] = reg;
2183 }
2184
2185 static void pshader_hw_texm3x3tex(const struct wined3d_shader_instruction *ins)
2186 {
2187     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2188     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
2189     DWORD flags;
2190     DWORD reg = ins->dst[0].reg.idx[0].offset;
2191     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2192     char dst_str[50];
2193     char src0_name[50], dst_name[50];
2194     BOOL is_color;
2195
2196     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2197     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2198     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_name, reg, src0_name);
2199
2200     /* Sample the texture using the calculated coordinates */
2201     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2202     flags = reg < MAX_TEXTURES ? priv->cur_ps_args->super.tex_transform >> reg * WINED3D_PSARGS_TEXTRANSFORM_SHIFT : 0;
2203     shader_hw_sample(ins, reg, dst_str, dst_name, flags & WINED3D_PSARGS_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2204     tex_mx->current_row = 0;
2205 }
2206
2207 static void pshader_hw_texm3x3vspec(const struct wined3d_shader_instruction *ins)
2208 {
2209     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2210     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
2211     DWORD flags;
2212     DWORD reg = ins->dst[0].reg.idx[0].offset;
2213     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2214     char dst_str[50];
2215     char src0_name[50];
2216     char dst_reg[50];
2217     BOOL is_color;
2218
2219     /* Get the dst reg without writemask strings. We know this register is uninitialized, so we can use all
2220      * components for temporary data storage
2221      */
2222     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_reg, &is_color);
2223     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2224     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_reg, reg, src0_name);
2225
2226     /* Construct the eye-ray vector from w coordinates */
2227     shader_addline(buffer, "MOV TB.x, fragment.texcoord[%u].w;\n", tex_mx->texcoord_w[0]);
2228     shader_addline(buffer, "MOV TB.y, fragment.texcoord[%u].w;\n", tex_mx->texcoord_w[1]);
2229     shader_addline(buffer, "MOV TB.z, fragment.texcoord[%u].w;\n", reg);
2230
2231     /* Calculate reflection vector
2232      */
2233     shader_addline(buffer, "DP3 %s.w, %s, TB;\n", dst_reg, dst_reg);
2234     /* The .w is ignored when sampling, so I can use TB.w to calculate dot(N, N) */
2235     shader_addline(buffer, "DP3 TB.w, %s, %s;\n", dst_reg, dst_reg);
2236     shader_addline(buffer, "RCP TB.w, TB.w;\n");
2237     shader_addline(buffer, "MUL %s.w, %s.w, TB.w;\n", dst_reg, dst_reg);
2238     shader_addline(buffer, "MUL %s, %s.w, %s;\n", dst_reg, dst_reg, dst_reg);
2239     shader_addline(buffer, "MAD %s, coefmul.x, %s, -TB;\n", dst_reg, dst_reg);
2240
2241     /* Sample the texture using the calculated coordinates */
2242     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2243     flags = reg < MAX_TEXTURES ? priv->cur_ps_args->super.tex_transform >> reg * WINED3D_PSARGS_TEXTRANSFORM_SHIFT : 0;
2244     shader_hw_sample(ins, reg, dst_str, dst_reg, flags & WINED3D_PSARGS_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2245     tex_mx->current_row = 0;
2246 }
2247
2248 static void pshader_hw_texm3x3spec(const struct wined3d_shader_instruction *ins)
2249 {
2250     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2251     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
2252     DWORD flags;
2253     DWORD reg = ins->dst[0].reg.idx[0].offset;
2254     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2255     char dst_str[50];
2256     char src0_name[50];
2257     char src1_name[50];
2258     char dst_reg[50];
2259     BOOL is_color;
2260
2261     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2262     shader_arb_get_src_param(ins, &ins->src[0], 1, src1_name);
2263     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_reg, &is_color);
2264     /* Note: dst_reg.xy is input here, generated by two texm3x3pad instructions */
2265     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_reg, reg, src0_name);
2266
2267     /* Calculate reflection vector.
2268      *
2269      *                   dot(N, E)
2270      * dst_reg.xyz = 2 * --------- * N - E
2271      *                   dot(N, N)
2272      *
2273      * Which normalizes the normal vector
2274      */
2275     shader_addline(buffer, "DP3 %s.w, %s, %s;\n", dst_reg, dst_reg, src1_name);
2276     shader_addline(buffer, "DP3 TC.w, %s, %s;\n", dst_reg, dst_reg);
2277     shader_addline(buffer, "RCP TC.w, TC.w;\n");
2278     shader_addline(buffer, "MUL %s.w, %s.w, TC.w;\n", dst_reg, dst_reg);
2279     shader_addline(buffer, "MUL %s, %s.w, %s;\n", dst_reg, dst_reg, dst_reg);
2280     shader_addline(buffer, "MAD %s, coefmul.x, %s, -%s;\n", dst_reg, dst_reg, src1_name);
2281
2282     /* Sample the texture using the calculated coordinates */
2283     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2284     flags = reg < MAX_TEXTURES ? priv->cur_ps_args->super.tex_transform >> reg * WINED3D_PSARGS_TEXTRANSFORM_SHIFT : 0;
2285     shader_hw_sample(ins, reg, dst_str, dst_reg, flags & WINED3D_PSARGS_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2286     tex_mx->current_row = 0;
2287 }
2288
2289 static void pshader_hw_texdepth(const struct wined3d_shader_instruction *ins)
2290 {
2291     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2292     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2293     char dst_name[50];
2294     const char *zero = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ZERO);
2295     const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
2296
2297     /* texdepth has an implicit destination, the fragment depth value. It's only parameter,
2298      * which is essentially an input, is the destination register because it is the first
2299      * parameter. According to the msdn, this must be register r5, but let's keep it more flexible
2300      * here(writemasks/swizzles are not valid on texdepth)
2301      */
2302     shader_arb_get_dst_param(ins, dst, dst_name);
2303
2304     /* According to the msdn, the source register(must be r5) is unusable after
2305      * the texdepth instruction, so we're free to modify it
2306      */
2307     shader_addline(buffer, "MIN %s.y, %s.y, %s;\n", dst_name, dst_name, one);
2308
2309     /* How to deal with the special case dst_name.g == 0? if r != 0, then
2310      * the r * (1 / 0) will give infinity, which is clamped to 1.0, the correct
2311      * result. But if r = 0.0, then 0 * inf = 0, which is incorrect.
2312      */
2313     shader_addline(buffer, "RCP %s.y, %s.y;\n", dst_name, dst_name);
2314     shader_addline(buffer, "MUL TA.x, %s.x, %s.y;\n", dst_name, dst_name);
2315     shader_addline(buffer, "MIN TA.x, TA.x, %s;\n", one);
2316     shader_addline(buffer, "MAX result.depth, TA.x, %s;\n", zero);
2317 }
2318
2319 /** Process the WINED3DSIO_TEXDP3TEX instruction in ARB:
2320  * Take a 3-component dot product of the TexCoord[dstreg] and src,
2321  * then perform a 1D texture lookup from stage dstregnum, place into dst. */
2322 static void pshader_hw_texdp3tex(const struct wined3d_shader_instruction *ins)
2323 {
2324     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2325     DWORD sampler_idx = ins->dst[0].reg.idx[0].offset;
2326     char src0[50];
2327     char dst_str[50];
2328
2329     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2330     shader_addline(buffer, "MOV TB, 0.0;\n");
2331     shader_addline(buffer, "DP3 TB.x, fragment.texcoord[%u], %s;\n", sampler_idx, src0);
2332
2333     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2334     shader_hw_sample(ins, sampler_idx, dst_str, "TB", 0 /* Only one coord, can't be projected */, NULL, NULL);
2335 }
2336
2337 /** Process the WINED3DSIO_TEXDP3 instruction in ARB:
2338  * Take a 3-component dot product of the TexCoord[dstreg] and src. */
2339 static void pshader_hw_texdp3(const struct wined3d_shader_instruction *ins)
2340 {
2341     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2342     char src0[50];
2343     char dst_str[50];
2344     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2345
2346     /* Handle output register */
2347     shader_arb_get_dst_param(ins, dst, dst_str);
2348     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2349     shader_addline(buffer, "DP3 %s, fragment.texcoord[%u], %s;\n", dst_str, dst->reg.idx[0].offset, src0);
2350 }
2351
2352 /** Process the WINED3DSIO_TEXM3X3 instruction in ARB
2353  * Perform the 3rd row of a 3x3 matrix multiply */
2354 static void pshader_hw_texm3x3(const struct wined3d_shader_instruction *ins)
2355 {
2356     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2357     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2358     char dst_str[50], dst_name[50];
2359     char src0[50];
2360     BOOL is_color;
2361
2362     shader_arb_get_dst_param(ins, dst, dst_str);
2363     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2364     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2365     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_name, dst->reg.idx[0].offset, src0);
2366     shader_addline(buffer, "MOV %s, %s;\n", dst_str, dst_name);
2367 }
2368
2369 /** Process the WINED3DSIO_TEXM3X2DEPTH instruction in ARB:
2370  * Last row of a 3x2 matrix multiply, use the result to calculate the depth:
2371  * Calculate tmp0.y = TexCoord[dstreg] . src.xyz;  (tmp0.x has already been calculated)
2372  * depth = (tmp0.y == 0.0) ? 1.0 : tmp0.x / tmp0.y
2373  */
2374 static void pshader_hw_texm3x2depth(const struct wined3d_shader_instruction *ins)
2375 {
2376     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2377     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2378     char src0[50], dst_name[50];
2379     BOOL is_color;
2380     const char *zero = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ZERO);
2381     const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
2382
2383     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2384     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2385     shader_addline(buffer, "DP3 %s.y, fragment.texcoord[%u], %s;\n", dst_name, dst->reg.idx[0].offset, src0);
2386
2387     /* How to deal with the special case dst_name.g == 0? if r != 0, then
2388      * the r * (1 / 0) will give infinity, which is clamped to 1.0, the correct
2389      * result. But if r = 0.0, then 0 * inf = 0, which is incorrect.
2390      */
2391     shader_addline(buffer, "RCP %s.y, %s.y;\n", dst_name, dst_name);
2392     shader_addline(buffer, "MUL %s.x, %s.x, %s.y;\n", dst_name, dst_name, dst_name);
2393     shader_addline(buffer, "MIN %s.x, %s.x, %s;\n", dst_name, dst_name, one);
2394     shader_addline(buffer, "MAX result.depth, %s.x, %s;\n", dst_name, zero);
2395 }
2396
2397 /** Handles transforming all WINED3DSIO_M?x? opcodes for
2398     Vertex/Pixel shaders to ARB_vertex_program codes */
2399 static void shader_hw_mnxn(const struct wined3d_shader_instruction *ins)
2400 {
2401     int i;
2402     int nComponents = 0;
2403     struct wined3d_shader_dst_param tmp_dst = {{0}};
2404     struct wined3d_shader_src_param tmp_src[2] = {{{0}}};
2405     struct wined3d_shader_instruction tmp_ins;
2406
2407     memset(&tmp_ins, 0, sizeof(tmp_ins));
2408
2409     /* Set constants for the temporary argument */
2410     tmp_ins.ctx = ins->ctx;
2411     tmp_ins.dst_count = 1;
2412     tmp_ins.dst = &tmp_dst;
2413     tmp_ins.src_count = 2;
2414     tmp_ins.src = tmp_src;
2415
2416     switch(ins->handler_idx)
2417     {
2418         case WINED3DSIH_M4x4:
2419             nComponents = 4;
2420             tmp_ins.handler_idx = WINED3DSIH_DP4;
2421             break;
2422         case WINED3DSIH_M4x3:
2423             nComponents = 3;
2424             tmp_ins.handler_idx = WINED3DSIH_DP4;
2425             break;
2426         case WINED3DSIH_M3x4:
2427             nComponents = 4;
2428             tmp_ins.handler_idx = WINED3DSIH_DP3;
2429             break;
2430         case WINED3DSIH_M3x3:
2431             nComponents = 3;
2432             tmp_ins.handler_idx = WINED3DSIH_DP3;
2433             break;
2434         case WINED3DSIH_M3x2:
2435             nComponents = 2;
2436             tmp_ins.handler_idx = WINED3DSIH_DP3;
2437             break;
2438         default:
2439             FIXME("Unhandled opcode %#x\n", ins->handler_idx);
2440             break;
2441     }
2442
2443     tmp_dst = ins->dst[0];
2444     tmp_src[0] = ins->src[0];
2445     tmp_src[1] = ins->src[1];
2446     for (i = 0; i < nComponents; ++i)
2447     {
2448         tmp_dst.write_mask = WINED3DSP_WRITEMASK_0 << i;
2449         shader_hw_map2gl(&tmp_ins);
2450         ++tmp_src[1].reg.idx[0].offset;
2451     }
2452 }
2453
2454 static void shader_hw_rcp(const struct wined3d_shader_instruction *ins)
2455 {
2456     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2457
2458     char dst[50];
2459     char src[50];
2460
2461     shader_arb_get_dst_param(ins, &ins->dst[0], dst); /* Destination */
2462     shader_arb_get_src_param(ins, &ins->src[0], 0, src);
2463     if (ins->src[0].swizzle == WINED3DSP_NOSWIZZLE)
2464     {
2465         /* Dx sdk says .x is used if no swizzle is given, but our test shows that
2466          * .w is used
2467          */
2468         strcat(src, ".w");
2469     }
2470
2471     shader_addline(buffer, "RCP%s %s, %s;\n", shader_arb_get_modifier(ins), dst, src);
2472 }
2473
2474 static void shader_hw_scalar_op(const struct wined3d_shader_instruction *ins)
2475 {
2476     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2477     const char *instruction;
2478
2479     char dst[50];
2480     char src[50];
2481
2482     switch(ins->handler_idx)
2483     {
2484         case WINED3DSIH_RSQ:  instruction = "RSQ"; break;
2485         case WINED3DSIH_RCP:  instruction = "RCP"; break;
2486         case WINED3DSIH_EXP:  instruction = "EX2"; break;
2487         case WINED3DSIH_EXPP: instruction = "EXP"; break;
2488         default: instruction = "";
2489             FIXME("Unhandled opcode %#x\n", ins->handler_idx);
2490             break;
2491     }
2492
2493     shader_arb_get_dst_param(ins, &ins->dst[0], dst); /* Destination */
2494     shader_arb_get_src_param(ins, &ins->src[0], 0, src);
2495     if (ins->src[0].swizzle == WINED3DSP_NOSWIZZLE)
2496     {
2497         /* Dx sdk says .x is used if no swizzle is given, but our test shows that
2498          * .w is used
2499          */
2500         strcat(src, ".w");
2501     }
2502
2503     shader_addline(buffer, "%s%s %s, %s;\n", instruction, shader_arb_get_modifier(ins), dst, src);
2504 }
2505
2506 static void shader_hw_nrm(const struct wined3d_shader_instruction *ins)
2507 {
2508     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2509     char dst_name[50];
2510     char src_name[50];
2511     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2512     BOOL pshader = shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type);
2513     const char *zero = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ZERO);
2514
2515     shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2516     shader_arb_get_src_param(ins, &ins->src[0], 1 /* Use TB */, src_name);
2517
2518     /* In D3D, NRM of a vector with length zero returns zero. Catch this situation, as
2519      * otherwise NRM or RSQ would return NaN */
2520     if(pshader && priv->target_version >= NV3)
2521     {
2522         /* GL_NV_fragment_program2's NRM needs protection against length zero vectors too
2523          *
2524          * TODO: Find out if DP3+NRM+MOV is really faster than DP3+RSQ+MUL
2525          */
2526         shader_addline(buffer, "DP3C TA, %s, %s;\n", src_name, src_name);
2527         shader_addline(buffer, "NRM%s %s, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name);
2528         shader_addline(buffer, "MOV %s (EQ), %s;\n", dst_name, zero);
2529     }
2530     else if(priv->target_version >= NV2)
2531     {
2532         shader_addline(buffer, "DP3C TA.x, %s, %s;\n", src_name, src_name);
2533         shader_addline(buffer, "RSQ TA.x (NE), TA.x;\n");
2534         shader_addline(buffer, "MUL%s %s, %s, TA.x;\n", shader_arb_get_modifier(ins), dst_name,
2535                        src_name);
2536     }
2537     else
2538     {
2539         const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
2540
2541         shader_addline(buffer, "DP3 TA.x, %s, %s;\n", src_name, src_name);
2542         /* Pass any non-zero value to RSQ if the input vector has a length of zero. The
2543          * RSQ result doesn't matter, as long as multiplying it by 0 returns 0.
2544          */
2545         shader_addline(buffer, "SGE TA.y, -TA.x, %s;\n", zero);
2546         shader_addline(buffer, "MAD TA.x, %s, TA.y, TA.x;\n", one);
2547
2548         shader_addline(buffer, "RSQ TA.x, TA.x;\n");
2549         /* dst.w = src[0].w * 1 / (src.x^2 + src.y^2 + src.z^2)^(1/2) according to msdn*/
2550         shader_addline(buffer, "MUL%s %s, %s, TA.x;\n", shader_arb_get_modifier(ins), dst_name,
2551                     src_name);
2552     }
2553 }
2554
2555 static void shader_hw_lrp(const struct wined3d_shader_instruction *ins)
2556 {
2557     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2558     char dst_name[50];
2559     char src_name[3][50];
2560
2561     /* ARB_fragment_program has a convenient LRP instruction */
2562     if(shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type)) {
2563         shader_hw_map2gl(ins);
2564         return;
2565     }
2566
2567     shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2568     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
2569     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
2570     shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
2571
2572     shader_addline(buffer, "SUB TA, %s, %s;\n", src_name[1], src_name[2]);
2573     shader_addline(buffer, "MAD%s %s, %s, TA, %s;\n", shader_arb_get_modifier(ins),
2574                    dst_name, src_name[0], src_name[2]);
2575 }
2576
2577 static void shader_hw_sincos(const struct wined3d_shader_instruction *ins)
2578 {
2579     /* This instruction exists in ARB, but the d3d instruction takes two extra parameters which
2580      * must contain fixed constants. So we need a separate function to filter those constants and
2581      * can't use map2gl
2582      */
2583     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2584     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2585     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2586     char dst_name[50];
2587     char src_name0[50], src_name1[50], src_name2[50];
2588     BOOL is_color;
2589
2590     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name0);
2591     if(shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type)) {
2592         shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2593         /* No modifiers are supported on SCS */
2594         shader_addline(buffer, "SCS %s, %s;\n", dst_name, src_name0);
2595
2596         if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE)
2597         {
2598             shader_arb_get_register_name(ins, &dst->reg, src_name0, &is_color);
2599             shader_addline(buffer, "MOV_SAT %s, %s;\n", dst_name, src_name0);
2600         }
2601     } else if(priv->target_version >= NV2) {
2602         shader_arb_get_register_name(ins, &dst->reg, dst_name, &is_color);
2603
2604         /* Sincos writemask must be .x, .y or .xy */
2605         if(dst->write_mask & WINED3DSP_WRITEMASK_0)
2606             shader_addline(buffer, "COS%s %s.x, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name0);
2607         if(dst->write_mask & WINED3DSP_WRITEMASK_1)
2608             shader_addline(buffer, "SIN%s %s.y, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name0);
2609     } else {
2610         /* Approximate sine and cosine with a taylor series, as per math textbook. The application passes 8
2611          * helper constants(D3DSINCOSCONST1 and D3DSINCOSCONST2) in src1 and src2.
2612          *
2613          * sin(x) = x - x^3/3! + x^5/5! - x^7/7! + ...
2614          * cos(x) = 1 - x^2/2! + x^4/4! - x^6/6! + ...
2615          *
2616          * The constants we get are:
2617          *
2618          *  +1   +1,     -1     -1     +1      +1      -1       -1
2619          *      ---- ,  ---- , ---- , ----- , ----- , ----- , ------
2620          *      1!*2    2!*4   3!*8   4!*16   5!*32   6!*64   7!*128
2621          *
2622          * If used with x^2, x^3, x^4 etc they calculate sin(x/2) and cos(x/2):
2623          *
2624          * (x/2)^2 = x^2 / 4
2625          * (x/2)^3 = x^3 / 8
2626          * (x/2)^4 = x^4 / 16
2627          * (x/2)^5 = x^5 / 32
2628          * etc
2629          *
2630          * To get the final result:
2631          * sin(x) = 2 * sin(x/2) * cos(x/2)
2632          * cos(x) = cos(x/2)^2 - sin(x/2)^2
2633          * (from sin(x+y) and cos(x+y) rules)
2634          *
2635          * As per MSDN, dst.z is undefined after the operation, and so is
2636          * dst.x and dst.y if they're masked out by the writemask. Ie
2637          * sincos dst.y, src1, c0, c1
2638          * returns the sine in dst.y. dst.x and dst.z are undefined, dst.w is not touched. The assembler
2639          * vsa.exe also stops with an error if the dest register is the same register as the source
2640          * register. This means we can use dest.xyz as temporary storage. The assembler vsa.exe output also
2641          * indicates that sincos consumes 8 instruction slots in vs_2_0(and, strangely, in vs_3_0).
2642          */
2643         shader_arb_get_src_param(ins, &ins->src[1], 1, src_name1);
2644         shader_arb_get_src_param(ins, &ins->src[2], 2, src_name2);
2645         shader_arb_get_register_name(ins, &dst->reg, dst_name, &is_color);
2646
2647         shader_addline(buffer, "MUL %s.x, %s, %s;\n", dst_name, src_name0, src_name0);  /* x ^ 2 */
2648         shader_addline(buffer, "MUL TA.y, %s.x, %s;\n", dst_name, src_name0);           /* x ^ 3 */
2649         shader_addline(buffer, "MUL %s.y, TA.y, %s;\n", dst_name, src_name0);           /* x ^ 4 */
2650         shader_addline(buffer, "MUL TA.z, %s.y, %s;\n", dst_name, src_name0);           /* x ^ 5 */
2651         shader_addline(buffer, "MUL %s.z, TA.z, %s;\n", dst_name, src_name0);           /* x ^ 6 */
2652         shader_addline(buffer, "MUL TA.w, %s.z, %s;\n", dst_name, src_name0);           /* x ^ 7 */
2653
2654         /* sin(x/2)
2655          *
2656          * Unfortunately we don't get the constants in a DP4-capable form. Is there a way to
2657          * properly merge that with MULs in the code above?
2658          * The swizzles .yz and xw however fit into the .yzxw swizzle added to ps_2_0. Maybe
2659          * we can merge the sine and cosine MAD rows to calculate them together.
2660          */
2661         shader_addline(buffer, "MUL TA.x, %s, %s.w;\n", src_name0, src_name2); /* x^1, +1/(1!*2) */
2662         shader_addline(buffer, "MAD TA.x, TA.y, %s.x, TA.x;\n", src_name2); /* -1/(3!*8) */
2663         shader_addline(buffer, "MAD TA.x, TA.z, %s.w, TA.x;\n", src_name1); /* +1/(5!*32) */
2664         shader_addline(buffer, "MAD TA.x, TA.w, %s.x, TA.x;\n", src_name1); /* -1/(7!*128) */
2665
2666         /* cos(x/2) */
2667         shader_addline(buffer, "MAD TA.y, %s.x, %s.y, %s.z;\n", dst_name, src_name2, src_name2); /* -1/(2!*4), +1.0 */
2668         shader_addline(buffer, "MAD TA.y, %s.y, %s.z, TA.y;\n", dst_name, src_name1); /* +1/(4!*16) */
2669         shader_addline(buffer, "MAD TA.y, %s.z, %s.y, TA.y;\n", dst_name, src_name1); /* -1/(6!*64) */
2670
2671         if(dst->write_mask & WINED3DSP_WRITEMASK_0) {
2672             /* cos x */
2673             shader_addline(buffer, "MUL TA.z, TA.y, TA.y;\n");
2674             shader_addline(buffer, "MAD %s.x, -TA.x, TA.x, TA.z;\n", dst_name);
2675         }
2676         if(dst->write_mask & WINED3DSP_WRITEMASK_1) {
2677             /* sin x */
2678             shader_addline(buffer, "MUL %s.y, TA.x, TA.y;\n", dst_name);
2679             shader_addline(buffer, "ADD %s.y, %s.y, %s.y;\n", dst_name, dst_name, dst_name);
2680         }
2681     }
2682 }
2683
2684 static void shader_hw_sgn(const struct wined3d_shader_instruction *ins)
2685 {
2686     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2687     char dst_name[50];
2688     char src_name[50];
2689     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
2690
2691     shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2692     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name);
2693
2694     /* SGN is only valid in vertex shaders */
2695     if(ctx->target_version >= NV2) {
2696         shader_addline(buffer, "SSG%s %s, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name);
2697         return;
2698     }
2699
2700     /* If SRC > 0.0, -SRC < SRC = TRUE, otherwise false.
2701      * if SRC < 0.0,  SRC < -SRC = TRUE. If neither is true, src = 0.0
2702      */
2703     if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE) {
2704         shader_addline(buffer, "SLT %s, -%s, %s;\n", dst_name, src_name, src_name);
2705     } else {
2706         /* src contains TA? Write to the dest first. This won't overwrite our destination.
2707          * Then use TA, and calculate the final result
2708          *
2709          * Not reading from TA? Store the first result in TA to avoid overwriting the
2710          * destination if src reg = dst reg
2711          */
2712         if(strstr(src_name, "TA"))
2713         {
2714             shader_addline(buffer, "SLT %s,  %s, -%s;\n", dst_name, src_name, src_name);
2715             shader_addline(buffer, "SLT TA, -%s, %s;\n", src_name, src_name);
2716             shader_addline(buffer, "ADD %s, %s, -TA;\n", dst_name, dst_name);
2717         }
2718         else
2719         {
2720             shader_addline(buffer, "SLT TA, -%s, %s;\n", src_name, src_name);
2721             shader_addline(buffer, "SLT %s,  %s, -%s;\n", dst_name, src_name, src_name);
2722             shader_addline(buffer, "ADD %s, TA, -%s;\n", dst_name, dst_name);
2723         }
2724     }
2725 }
2726
2727 static void shader_hw_dsy(const struct wined3d_shader_instruction *ins)
2728 {
2729     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2730     char src[50];
2731     char dst[50];
2732     char dst_name[50];
2733     BOOL is_color;
2734
2735     shader_arb_get_dst_param(ins, &ins->dst[0], dst);
2736     shader_arb_get_src_param(ins, &ins->src[0], 0, src);
2737     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2738
2739     shader_addline(buffer, "DDY %s, %s;\n", dst, src);
2740     shader_addline(buffer, "MUL%s %s, %s, ycorrection.y;\n", shader_arb_get_modifier(ins), dst, dst_name);
2741 }
2742
2743 static DWORD abs_modifier(DWORD mod, BOOL *need_abs)
2744 {
2745     *need_abs = FALSE;
2746
2747     switch(mod)
2748     {
2749         case WINED3DSPSM_NONE:      return WINED3DSPSM_ABS;
2750         case WINED3DSPSM_NEG:       return WINED3DSPSM_ABS;
2751         case WINED3DSPSM_BIAS:      *need_abs = TRUE; return WINED3DSPSM_BIAS;
2752         case WINED3DSPSM_BIASNEG:   *need_abs = TRUE; return WINED3DSPSM_BIASNEG;
2753         case WINED3DSPSM_SIGN:      *need_abs = TRUE; return WINED3DSPSM_SIGN;
2754         case WINED3DSPSM_SIGNNEG:   *need_abs = TRUE; return WINED3DSPSM_SIGNNEG;
2755         case WINED3DSPSM_COMP:      *need_abs = TRUE; return WINED3DSPSM_COMP;
2756         case WINED3DSPSM_X2:        *need_abs = TRUE; return WINED3DSPSM_X2;
2757         case WINED3DSPSM_X2NEG:     *need_abs = TRUE; return WINED3DSPSM_X2NEG;
2758         case WINED3DSPSM_DZ:        *need_abs = TRUE; return WINED3DSPSM_DZ;
2759         case WINED3DSPSM_DW:        *need_abs = TRUE; return WINED3DSPSM_DW;
2760         case WINED3DSPSM_ABS:       return WINED3DSPSM_ABS;
2761         case WINED3DSPSM_ABSNEG:    return WINED3DSPSM_ABS;
2762     }
2763     FIXME("Unknown modifier %u\n", mod);
2764     return mod;
2765 }
2766
2767 static void shader_hw_log(const struct wined3d_shader_instruction *ins)
2768 {
2769     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2770     char src0[50], dst[50];
2771     struct wined3d_shader_src_param src0_copy = ins->src[0];
2772     BOOL need_abs = FALSE;
2773     const char *instr;
2774
2775     switch(ins->handler_idx)
2776     {
2777         case WINED3DSIH_LOG:  instr = "LG2"; break;
2778         case WINED3DSIH_LOGP: instr = "LOG"; break;
2779         default:
2780             ERR("Unexpected instruction %d\n", ins->handler_idx);
2781             return;
2782     }
2783
2784     /* LOG and LOGP operate on the absolute value of the input */
2785     src0_copy.modifiers = abs_modifier(src0_copy.modifiers, &need_abs);
2786
2787     shader_arb_get_dst_param(ins, &ins->dst[0], dst);
2788     shader_arb_get_src_param(ins, &src0_copy, 0, src0);
2789
2790     if(need_abs)
2791     {
2792         shader_addline(buffer, "ABS TA, %s;\n", src0);
2793         shader_addline(buffer, "%s%s %s, TA;\n", instr, shader_arb_get_modifier(ins), dst);
2794     }
2795     else
2796     {
2797         shader_addline(buffer, "%s%s %s, %s;\n", instr, shader_arb_get_modifier(ins), dst, src0);
2798     }
2799 }
2800
2801 static void shader_hw_pow(const struct wined3d_shader_instruction *ins)
2802 {
2803     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2804     char src0[50], src1[50], dst[50];
2805     struct wined3d_shader_src_param src0_copy = ins->src[0];
2806     BOOL need_abs = FALSE;
2807     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2808     const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
2809
2810     /* POW operates on the absolute value of the input */
2811     src0_copy.modifiers = abs_modifier(src0_copy.modifiers, &need_abs);
2812
2813     shader_arb_get_dst_param(ins, &ins->dst[0], dst);
2814     shader_arb_get_src_param(ins, &src0_copy, 0, src0);
2815     shader_arb_get_src_param(ins, &ins->src[1], 1, src1);
2816
2817     if (need_abs)
2818         shader_addline(buffer, "ABS TA.x, %s;\n", src0);
2819     else
2820         shader_addline(buffer, "MOV TA.x, %s;\n", src0);
2821
2822     if (priv->target_version >= NV2)
2823     {
2824         shader_addline(buffer, "MOVC TA.y, %s;\n", src1);
2825         shader_addline(buffer, "POW%s %s, TA.x, TA.y;\n", shader_arb_get_modifier(ins), dst);
2826         shader_addline(buffer, "MOV %s (EQ.y), %s;\n", dst, one);
2827     }
2828     else
2829     {
2830         const char *zero = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ZERO);
2831         const char *flt_eps = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_EPS);
2832
2833         shader_addline(buffer, "ABS TA.y, %s;\n", src1);
2834         shader_addline(buffer, "SGE TA.y, -TA.y, %s;\n", zero);
2835         /* Possibly add flt_eps to avoid getting float special values */
2836         shader_addline(buffer, "MAD TA.z, TA.y, %s, %s;\n", flt_eps, src1);
2837         shader_addline(buffer, "POW%s TA.x, TA.x, TA.z;\n", shader_arb_get_modifier(ins));
2838         shader_addline(buffer, "MAD TA.x, -TA.x, TA.y, TA.x;\n");
2839         shader_addline(buffer, "MAD %s, TA.y, %s, TA.x;\n", dst, one);
2840     }
2841 }
2842
2843 static void shader_hw_loop(const struct wined3d_shader_instruction *ins)
2844 {
2845     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2846     char src_name[50];
2847     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2848
2849     /* src0 is aL */
2850     shader_arb_get_src_param(ins, &ins->src[1], 0, src_name);
2851
2852     if(vshader)
2853     {
2854         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2855         struct list *e = list_head(&priv->control_frames);
2856         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2857
2858         if(priv->loop_depth > 1) shader_addline(buffer, "PUSHA aL;\n");
2859         /* The constant loader makes sure to load -1 into iX.w */
2860         shader_addline(buffer, "ARLC aL, %s.xywz;\n", src_name);
2861         shader_addline(buffer, "BRA loop_%u_end (LE.x);\n", control_frame->no.loop);
2862         shader_addline(buffer, "loop_%u_start:\n", control_frame->no.loop);
2863     }
2864     else
2865     {
2866         shader_addline(buffer, "LOOP %s;\n", src_name);
2867     }
2868 }
2869
2870 static void shader_hw_rep(const struct wined3d_shader_instruction *ins)
2871 {
2872     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2873     char src_name[50];
2874     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2875
2876     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name);
2877
2878     /* The constant loader makes sure to load -1 into iX.w */
2879     if(vshader)
2880     {
2881         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2882         struct list *e = list_head(&priv->control_frames);
2883         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2884
2885         if(priv->loop_depth > 1) shader_addline(buffer, "PUSHA aL;\n");
2886
2887         shader_addline(buffer, "ARLC aL, %s.xywz;\n", src_name);
2888         shader_addline(buffer, "BRA loop_%u_end (LE.x);\n", control_frame->no.loop);
2889         shader_addline(buffer, "loop_%u_start:\n", control_frame->no.loop);
2890     }
2891     else
2892     {
2893         shader_addline(buffer, "REP %s;\n", src_name);
2894     }
2895 }
2896
2897 static void shader_hw_endloop(const struct wined3d_shader_instruction *ins)
2898 {
2899     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2900     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2901
2902     if(vshader)
2903     {
2904         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2905         struct list *e = list_head(&priv->control_frames);
2906         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2907
2908         shader_addline(buffer, "ARAC aL.xy, aL;\n");
2909         shader_addline(buffer, "BRA loop_%u_start (GT.x);\n", control_frame->no.loop);
2910         shader_addline(buffer, "loop_%u_end:\n", control_frame->no.loop);
2911
2912         if(priv->loop_depth > 1) shader_addline(buffer, "POPA aL;\n");
2913     }
2914     else
2915     {
2916         shader_addline(buffer, "ENDLOOP;\n");
2917     }
2918 }
2919
2920 static void shader_hw_endrep(const struct wined3d_shader_instruction *ins)
2921 {
2922     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2923     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2924
2925     if(vshader)
2926     {
2927         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2928         struct list *e = list_head(&priv->control_frames);
2929         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2930
2931         shader_addline(buffer, "ARAC aL.xy, aL;\n");
2932         shader_addline(buffer, "BRA loop_%u_start (GT.x);\n", control_frame->no.loop);
2933         shader_addline(buffer, "loop_%u_end:\n", control_frame->no.loop);
2934
2935         if(priv->loop_depth > 1) shader_addline(buffer, "POPA aL;\n");
2936     }
2937     else
2938     {
2939         shader_addline(buffer, "ENDREP;\n");
2940     }
2941 }
2942
2943 static const struct control_frame *find_last_loop(const struct shader_arb_ctx_priv *priv)
2944 {
2945     struct control_frame *control_frame;
2946
2947     LIST_FOR_EACH_ENTRY(control_frame, &priv->control_frames, struct control_frame, entry)
2948     {
2949         if(control_frame->type == LOOP || control_frame->type == REP) return control_frame;
2950     }
2951     ERR("Could not find loop for break\n");
2952     return NULL;
2953 }
2954
2955 static void shader_hw_break(const struct wined3d_shader_instruction *ins)
2956 {
2957     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2958     const struct control_frame *control_frame = find_last_loop(ins->ctx->backend_data);
2959     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2960
2961     if(vshader)
2962     {
2963         shader_addline(buffer, "BRA loop_%u_end;\n", control_frame->no.loop);
2964     }
2965     else
2966     {
2967         shader_addline(buffer, "BRK;\n");
2968     }
2969 }
2970
2971 static const char *get_compare(enum wined3d_shader_rel_op op)
2972 {
2973     switch (op)
2974     {
2975         case WINED3D_SHADER_REL_OP_GT: return "GT";
2976         case WINED3D_SHADER_REL_OP_EQ: return "EQ";
2977         case WINED3D_SHADER_REL_OP_GE: return "GE";
2978         case WINED3D_SHADER_REL_OP_LT: return "LT";
2979         case WINED3D_SHADER_REL_OP_NE: return "NE";
2980         case WINED3D_SHADER_REL_OP_LE: return "LE";
2981         default:
2982             FIXME("Unrecognized operator %#x.\n", op);
2983             return "(\?\?)";
2984     }
2985 }
2986
2987 static enum wined3d_shader_rel_op invert_compare(enum wined3d_shader_rel_op op)
2988 {
2989     switch (op)
2990     {
2991         case WINED3D_SHADER_REL_OP_GT: return WINED3D_SHADER_REL_OP_LE;
2992         case WINED3D_SHADER_REL_OP_EQ: return WINED3D_SHADER_REL_OP_NE;
2993         case WINED3D_SHADER_REL_OP_GE: return WINED3D_SHADER_REL_OP_LT;
2994         case WINED3D_SHADER_REL_OP_LT: return WINED3D_SHADER_REL_OP_GE;
2995         case WINED3D_SHADER_REL_OP_NE: return WINED3D_SHADER_REL_OP_EQ;
2996         case WINED3D_SHADER_REL_OP_LE: return WINED3D_SHADER_REL_OP_GT;
2997         default:
2998             FIXME("Unrecognized operator %#x.\n", op);
2999             return -1;
3000     }
3001 }
3002
3003 static void shader_hw_breakc(const struct wined3d_shader_instruction *ins)
3004 {
3005     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3006     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
3007     const struct control_frame *control_frame = find_last_loop(ins->ctx->backend_data);
3008     char src_name0[50];
3009     char src_name1[50];
3010     const char *comp = get_compare(ins->flags);
3011
3012     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name0);
3013     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name1);
3014
3015     if(vshader)
3016     {
3017         /* SUBC CC, src0, src1" works only in pixel shaders, so use TA to throw
3018          * away the subtraction result
3019          */
3020         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
3021         shader_addline(buffer, "BRA loop_%u_end (%s.x);\n", control_frame->no.loop, comp);
3022     }
3023     else
3024     {
3025         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
3026         shader_addline(buffer, "BRK (%s.x);\n", comp);
3027     }
3028 }
3029
3030 static void shader_hw_ifc(const struct wined3d_shader_instruction *ins)
3031 {
3032     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3033     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3034     struct list *e = list_head(&priv->control_frames);
3035     struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
3036     const char *comp;
3037     char src_name0[50];
3038     char src_name1[50];
3039     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
3040
3041     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name0);
3042     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name1);
3043
3044     if(vshader)
3045     {
3046         /* Invert the flag. We jump to the else label if the condition is NOT true */
3047         comp = get_compare(invert_compare(ins->flags));
3048         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
3049         shader_addline(buffer, "BRA ifc_%u_else (%s.x);\n", control_frame->no.ifc, comp);
3050     }
3051     else
3052     {
3053         comp = get_compare(ins->flags);
3054         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
3055         shader_addline(buffer, "IF %s.x;\n", comp);
3056     }
3057 }
3058
3059 static void shader_hw_else(const struct wined3d_shader_instruction *ins)
3060 {
3061     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3062     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3063     struct list *e = list_head(&priv->control_frames);
3064     struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
3065     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
3066
3067     if(vshader)
3068     {
3069         shader_addline(buffer, "BRA ifc_%u_endif;\n", control_frame->no.ifc);
3070         shader_addline(buffer, "ifc_%u_else:\n", control_frame->no.ifc);
3071         control_frame->had_else = TRUE;
3072     }
3073     else
3074     {
3075         shader_addline(buffer, "ELSE;\n");
3076     }
3077 }
3078
3079 static void shader_hw_endif(const struct wined3d_shader_instruction *ins)
3080 {
3081     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3082     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3083     struct list *e = list_head(&priv->control_frames);
3084     struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
3085     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
3086
3087     if(vshader)
3088     {
3089         if(control_frame->had_else)
3090         {
3091             shader_addline(buffer, "ifc_%u_endif:\n", control_frame->no.ifc);
3092         }
3093         else
3094         {
3095             shader_addline(buffer, "#No else branch. else is endif\n");
3096             shader_addline(buffer, "ifc_%u_else:\n", control_frame->no.ifc);
3097         }
3098     }
3099     else
3100     {
3101         shader_addline(buffer, "ENDIF;\n");
3102     }
3103 }
3104
3105 static void shader_hw_texldd(const struct wined3d_shader_instruction *ins)
3106 {
3107     DWORD sampler_idx = ins->src[1].reg.idx[0].offset;
3108     char reg_dest[40];
3109     char reg_src[3][40];
3110     WORD flags = TEX_DERIV;
3111
3112     shader_arb_get_dst_param(ins, &ins->dst[0], reg_dest);
3113     shader_arb_get_src_param(ins, &ins->src[0], 0, reg_src[0]);
3114     shader_arb_get_src_param(ins, &ins->src[2], 1, reg_src[1]);
3115     shader_arb_get_src_param(ins, &ins->src[3], 2, reg_src[2]);
3116
3117     if (ins->flags & WINED3DSI_TEXLD_PROJECT) flags |= TEX_PROJ;
3118     if (ins->flags & WINED3DSI_TEXLD_BIAS) flags |= TEX_BIAS;
3119
3120     shader_hw_sample(ins, sampler_idx, reg_dest, reg_src[0], flags, reg_src[1], reg_src[2]);
3121 }
3122
3123 static void shader_hw_texldl(const struct wined3d_shader_instruction *ins)
3124 {
3125     DWORD sampler_idx = ins->src[1].reg.idx[0].offset;
3126     char reg_dest[40];
3127     char reg_coord[40];
3128     WORD flags = TEX_LOD;
3129
3130     shader_arb_get_dst_param(ins, &ins->dst[0], reg_dest);
3131     shader_arb_get_src_param(ins, &ins->src[0], 0, reg_coord);
3132
3133     if (ins->flags & WINED3DSI_TEXLD_PROJECT) flags |= TEX_PROJ;
3134     if (ins->flags & WINED3DSI_TEXLD_BIAS) flags |= TEX_BIAS;
3135
3136     shader_hw_sample(ins, sampler_idx, reg_dest, reg_coord, flags, NULL, NULL);
3137 }
3138
3139 static void shader_hw_label(const struct wined3d_shader_instruction *ins)
3140 {
3141     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3142     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3143
3144     priv->in_main_func = FALSE;
3145     /* Call instructions activate the NV extensions, not labels and rets. If there is an uncalled
3146      * subroutine, don't generate a label that will make GL complain
3147      */
3148     if(priv->target_version == ARB) return;
3149
3150     shader_addline(buffer, "l%u:\n", ins->src[0].reg.idx[0].offset);
3151 }
3152
3153 static void vshader_add_footer(struct shader_arb_ctx_priv *priv_ctx,
3154         const struct arb_vshader_private *shader_data, const struct arb_vs_compile_args *args,
3155         const struct wined3d_shader_reg_maps *reg_maps, const struct wined3d_gl_info *gl_info,
3156         struct wined3d_shader_buffer *buffer)
3157 {
3158     unsigned int i;
3159
3160     /* The D3DRS_FOGTABLEMODE render state defines if the shader-generated fog coord is used
3161      * or if the fragment depth is used. If the fragment depth is used(FOGTABLEMODE != NONE),
3162      * the fog frag coord is thrown away. If the fog frag coord is used, but not written by
3163      * the shader, it is set to 0.0(fully fogged, since start = 1.0, end = 0.0)
3164      */
3165     if (args->super.fog_src == VS_FOG_Z)
3166     {
3167         shader_addline(buffer, "MOV result.fogcoord, TMP_OUT.z;\n");
3168     }
3169     else
3170     {
3171         if (!reg_maps->fog)
3172         {
3173             /* posFixup.x is always 1.0, so we can safely use it */
3174             shader_addline(buffer, "ADD result.fogcoord, posFixup.x, -posFixup.x;\n");
3175         }
3176         else
3177         {
3178             /* Clamp fogcoord */
3179             const char *zero = arb_get_helper_value(reg_maps->shader_version.type, ARB_ZERO);
3180             const char *one = arb_get_helper_value(reg_maps->shader_version.type, ARB_ONE);
3181
3182             shader_addline(buffer, "MIN TMP_FOGCOORD.x, TMP_FOGCOORD.x, %s;\n", one);
3183             shader_addline(buffer, "MAX result.fogcoord.x, TMP_FOGCOORD.x, %s;\n", zero);
3184         }
3185     }
3186
3187     /* Clipplanes are always stored without y inversion */
3188     if (use_nv_clip(gl_info) && priv_ctx->target_version >= NV2)
3189     {
3190         if (args->super.clip_enabled)
3191         {
3192             for (i = 0; i < priv_ctx->vs_clipplanes; i++)
3193             {
3194                 shader_addline(buffer, "DP4 result.clip[%u].x, TMP_OUT, state.clip[%u].plane;\n", i, i);
3195             }
3196         }
3197     }
3198     else if (args->clip.boolclip.clip_texcoord)
3199     {
3200         unsigned int cur_clip = 0;
3201         char component[4] = {'x', 'y', 'z', 'w'};
3202         const char *zero = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_ZERO);
3203
3204         for (i = 0; i < gl_info->limits.clipplanes; ++i)
3205         {
3206             if (args->clip.boolclip.clipplane_mask & (1 << i))
3207             {
3208                 shader_addline(buffer, "DP4 TA.%c, TMP_OUT, state.clip[%u].plane;\n",
3209                                component[cur_clip++], i);
3210             }
3211         }
3212         switch (cur_clip)
3213         {
3214             case 0:
3215                 shader_addline(buffer, "MOV TA, %s;\n", zero);
3216                 break;
3217             case 1:
3218                 shader_addline(buffer, "MOV TA.yzw, %s;\n", zero);
3219                 break;
3220             case 2:
3221                 shader_addline(buffer, "MOV TA.zw, %s;\n", zero);
3222                 break;
3223             case 3:
3224                 shader_addline(buffer, "MOV TA.w, %s;\n", zero);
3225                 break;
3226         }
3227         shader_addline(buffer, "MOV result.texcoord[%u], TA;\n",
3228                        args->clip.boolclip.clip_texcoord - 1);
3229     }
3230
3231     /* Write the final position.
3232      *
3233      * OpenGL coordinates specify the center of the pixel while d3d coords specify
3234      * the corner. The offsets are stored in z and w in posFixup. posFixup.y contains
3235      * 1.0 or -1.0 to turn the rendering upside down for offscreen rendering. PosFixup.x
3236      * contains 1.0 to allow a mad, but arb vs swizzles are too restricted for that.
3237      */
3238     shader_addline(buffer, "MUL TA, posFixup, TMP_OUT.w;\n");
3239     shader_addline(buffer, "ADD TMP_OUT.x, TMP_OUT.x, TA.z;\n");
3240     shader_addline(buffer, "MAD TMP_OUT.y, TMP_OUT.y, posFixup.y, TA.w;\n");
3241
3242     /* Z coord [0;1]->[-1;1] mapping, see comment in transform_projection in state.c
3243      * and the glsl equivalent
3244      */
3245     if (need_helper_const(shader_data, reg_maps, gl_info))
3246     {
3247         const char *two = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_TWO);
3248         shader_addline(buffer, "MAD TMP_OUT.z, TMP_OUT.z, %s, -TMP_OUT.w;\n", two);
3249     }
3250     else
3251     {
3252         shader_addline(buffer, "ADD TMP_OUT.z, TMP_OUT.z, TMP_OUT.z;\n");
3253         shader_addline(buffer, "ADD TMP_OUT.z, TMP_OUT.z, -TMP_OUT.w;\n");
3254     }
3255
3256     shader_addline(buffer, "MOV result.position, TMP_OUT;\n");
3257
3258     priv_ctx->footer_written = TRUE;
3259 }
3260
3261 static void shader_hw_ret(const struct wined3d_shader_instruction *ins)
3262 {
3263     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3264     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3265     const struct wined3d_shader *shader = ins->ctx->shader;
3266     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
3267
3268     if(priv->target_version == ARB) return;
3269
3270     if(vshader)
3271     {
3272         if (priv->in_main_func) vshader_add_footer(priv, shader->backend_data,
3273                 priv->cur_vs_args, ins->ctx->reg_maps, ins->ctx->gl_info, buffer);
3274     }
3275
3276     shader_addline(buffer, "RET;\n");
3277 }
3278
3279 static void shader_hw_call(const struct wined3d_shader_instruction *ins)
3280 {
3281     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3282     shader_addline(buffer, "CAL l%u;\n", ins->src[0].reg.idx[0].offset);
3283 }
3284
3285 /* Context activation is done by the caller. */
3286 static GLuint create_arb_blt_vertex_program(const struct wined3d_gl_info *gl_info)
3287 {
3288     GLuint program_id = 0;
3289     GLint pos;
3290
3291     const char *blt_vprogram =
3292         "!!ARBvp1.0\n"
3293         "PARAM c[1] = { { 1, 0.5 } };\n"
3294         "MOV result.position, vertex.position;\n"
3295         "MOV result.color, c[0].x;\n"
3296         "MOV result.texcoord[0], vertex.texcoord[0];\n"
3297         "END\n";
3298
3299     GL_EXTCALL(glGenProgramsARB(1, &program_id));
3300     GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, program_id));
3301     GL_EXTCALL(glProgramStringARB(GL_VERTEX_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
3302             strlen(blt_vprogram), blt_vprogram));
3303     checkGLcall("glProgramStringARB()");
3304
3305     gl_info->gl_ops.gl.p_glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
3306     if (pos != -1)
3307     {
3308         FIXME("Vertex program error at position %d: %s\n\n", pos,
3309             debugstr_a((const char *)gl_info->gl_ops.gl.p_glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
3310         shader_arb_dump_program_source(blt_vprogram);
3311     }
3312     else
3313     {
3314         GLint native;
3315
3316         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
3317         checkGLcall("glGetProgramivARB()");
3318         if (!native) WARN("Program exceeds native resource limits.\n");
3319     }
3320
3321     return program_id;
3322 }
3323
3324 /* Context activation is done by the caller. */
3325 static GLuint create_arb_blt_fragment_program(const struct wined3d_gl_info *gl_info,
3326         enum tex_types tex_type, BOOL masked)
3327 {
3328     GLuint program_id = 0;
3329     const char *fprogram;
3330     GLint pos;
3331
3332     static const char * const blt_fprograms_full[tex_type_count] =
3333     {
3334         /* tex_1d */
3335         NULL,
3336         /* tex_2d */
3337         "!!ARBfp1.0\n"
3338         "TEMP R0;\n"
3339         "TEX R0.x, fragment.texcoord[0], texture[0], 2D;\n"
3340         "MOV result.depth.z, R0.x;\n"
3341         "END\n",
3342         /* tex_3d */
3343         NULL,
3344         /* tex_cube */
3345         "!!ARBfp1.0\n"
3346         "TEMP R0;\n"
3347         "TEX R0.x, fragment.texcoord[0], texture[0], CUBE;\n"
3348         "MOV result.depth.z, R0.x;\n"
3349         "END\n",
3350         /* tex_rect */
3351         "!!ARBfp1.0\n"
3352         "TEMP R0;\n"
3353         "TEX R0.x, fragment.texcoord[0], texture[0], RECT;\n"
3354         "MOV result.depth.z, R0.x;\n"
3355         "END\n",
3356     };
3357
3358     static const char * const blt_fprograms_masked[tex_type_count] =
3359     {
3360         /* tex_1d */
3361         NULL,
3362         /* tex_2d */
3363         "!!ARBfp1.0\n"
3364         "PARAM mask = program.local[0];\n"
3365         "TEMP R0;\n"
3366         "SLT R0.xy, fragment.position, mask.zwzw;\n"
3367         "MUL R0.x, R0.x, R0.y;\n"
3368         "KIL -R0.x;\n"
3369         "TEX R0.x, fragment.texcoord[0], texture[0], 2D;\n"
3370         "MOV result.depth.z, R0.x;\n"
3371         "END\n",
3372         /* tex_3d */
3373         NULL,
3374         /* tex_cube */
3375         "!!ARBfp1.0\n"
3376         "PARAM mask = program.local[0];\n"
3377         "TEMP R0;\n"
3378         "SLT R0.xy, fragment.position, mask.zwzw;\n"
3379         "MUL R0.x, R0.x, R0.y;\n"
3380         "KIL -R0.x;\n"
3381         "TEX R0.x, fragment.texcoord[0], texture[0], CUBE;\n"
3382         "MOV result.depth.z, R0.x;\n"
3383         "END\n",
3384         /* tex_rect */
3385         "!!ARBfp1.0\n"
3386         "PARAM mask = program.local[0];\n"
3387         "TEMP R0;\n"
3388         "SLT R0.xy, fragment.position, mask.zwzw;\n"
3389         "MUL R0.x, R0.x, R0.y;\n"
3390         "KIL -R0.x;\n"
3391         "TEX R0.x, fragment.texcoord[0], texture[0], RECT;\n"
3392         "MOV result.depth.z, R0.x;\n"
3393         "END\n",
3394     };
3395
3396     fprogram = masked ? blt_fprograms_masked[tex_type] : blt_fprograms_full[tex_type];
3397     if (!fprogram)
3398     {
3399         FIXME("tex_type %#x not supported, falling back to tex_2d\n", tex_type);
3400         tex_type = tex_2d;
3401         fprogram = masked ? blt_fprograms_masked[tex_type] : blt_fprograms_full[tex_type];
3402     }
3403
3404     GL_EXTCALL(glGenProgramsARB(1, &program_id));
3405     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, program_id));
3406     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB, strlen(fprogram), fprogram));
3407     checkGLcall("glProgramStringARB()");
3408
3409     gl_info->gl_ops.gl.p_glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
3410     if (pos != -1)
3411     {
3412         FIXME("Fragment program error at position %d: %s\n\n", pos,
3413             debugstr_a((const char *)gl_info->gl_ops.gl.p_glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
3414         shader_arb_dump_program_source(fprogram);
3415     }
3416     else
3417     {
3418         GLint native;
3419
3420         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
3421         checkGLcall("glGetProgramivARB()");
3422         if (!native) WARN("Program exceeds native resource limits.\n");
3423     }
3424
3425     return program_id;
3426 }
3427
3428 static void arbfp_add_sRGB_correction(struct wined3d_shader_buffer *buffer, const char *fragcolor,
3429         const char *tmp1, const char *tmp2, const char *tmp3, const char *tmp4, BOOL condcode)
3430 {
3431     /* Perform sRGB write correction. See GLX_EXT_framebuffer_sRGB */
3432
3433     if(condcode)
3434     {
3435         /* Sigh. MOVC CC doesn't work, so use one of the temps as dummy dest */
3436         shader_addline(buffer, "SUBC %s, %s.x, srgb_consts1.y;\n", tmp1, fragcolor);
3437         /* Calculate the > 0.0031308 case */
3438         shader_addline(buffer, "POW %s.x (GE), %s.x, srgb_consts1.z;\n", fragcolor, fragcolor);
3439         shader_addline(buffer, "POW %s.y (GE), %s.y, srgb_consts1.z;\n", fragcolor, fragcolor);
3440         shader_addline(buffer, "POW %s.z (GE), %s.z, srgb_consts1.z;\n", fragcolor, fragcolor);
3441         shader_addline(buffer, "MUL %s.xyz (GE), %s, srgb_consts1.w;\n", fragcolor, fragcolor);
3442         shader_addline(buffer, "SUB %s.xyz (GE), %s, srgb_consts2.x;\n", fragcolor, fragcolor);
3443         /* Calculate the < case */
3444         shader_addline(buffer, "MUL %s.xyz (LT), srgb_consts1.x, %s;\n", fragcolor, fragcolor);
3445     }
3446     else
3447     {
3448         /* Calculate the > 0.0031308 case */
3449         shader_addline(buffer, "POW %s.x, %s.x, srgb_consts1.z;\n", tmp1, fragcolor);
3450         shader_addline(buffer, "POW %s.y, %s.y, srgb_consts1.z;\n", tmp1, fragcolor);
3451         shader_addline(buffer, "POW %s.z, %s.z, srgb_consts1.z;\n", tmp1, fragcolor);
3452         shader_addline(buffer, "MUL %s, %s, srgb_consts1.w;\n", tmp1, tmp1);
3453         shader_addline(buffer, "SUB %s, %s, srgb_consts2.x;\n", tmp1, tmp1);
3454         /* Calculate the < case */
3455         shader_addline(buffer, "MUL %s, srgb_consts1.x, %s;\n", tmp2, fragcolor);
3456         /* Get 1.0 / 0.0 masks for > 0.0031308 and < 0.0031308 */
3457         shader_addline(buffer, "SLT %s, srgb_consts1.y, %s;\n", tmp3, fragcolor);
3458         shader_addline(buffer, "SGE %s, srgb_consts1.y, %s;\n", tmp4, fragcolor);
3459         /* Store the components > 0.0031308 in the destination */
3460         shader_addline(buffer, "MUL %s.xyz, %s, %s;\n", fragcolor, tmp1, tmp3);
3461         /* Add the components that are < 0.0031308 */
3462         shader_addline(buffer, "MAD %s.xyz, %s, %s, %s;\n", fragcolor, tmp2, tmp4, fragcolor);
3463         /* Move everything into result.color at once. Nvidia hardware cannot handle partial
3464         * result.color writes(.rgb first, then .a), or handle overwriting already written
3465         * components. The assembler uses a temporary register in this case, which is usually
3466         * not allocated from one of our registers that were used earlier.
3467         */
3468     }
3469     /* [0.0;1.0] clamping. Not needed, this is done implicitly */
3470 }
3471
3472 static const DWORD *find_loop_control_values(const struct wined3d_shader *shader, DWORD idx)
3473 {
3474     const struct wined3d_shader_lconst *constant;
3475
3476     LIST_FOR_EACH_ENTRY(constant, &shader->constantsI, struct wined3d_shader_lconst, entry)
3477     {
3478         if (constant->idx == idx)
3479         {
3480             return constant->value;
3481         }
3482     }
3483     return NULL;
3484 }
3485
3486 static void init_ps_input(const struct wined3d_shader *shader,
3487         const struct arb_ps_compile_args *args, struct shader_arb_ctx_priv *priv)
3488 {
3489     static const char * const texcoords[8] =
3490     {
3491         "fragment.texcoord[0]", "fragment.texcoord[1]", "fragment.texcoord[2]", "fragment.texcoord[3]",
3492         "fragment.texcoord[4]", "fragment.texcoord[5]", "fragment.texcoord[6]", "fragment.texcoord[7]"
3493     };
3494     unsigned int i;
3495     const struct wined3d_shader_signature_element *sig = shader->input_signature;
3496     const char *semantic_name;
3497     DWORD semantic_idx;
3498
3499     switch(args->super.vp_mode)
3500     {
3501         case pretransformed:
3502         case fixedfunction:
3503             /* The pixelshader has to collect the varyings on its own. In any case properly load
3504              * color0 and color1. In the case of pretransformed vertices also load texcoords. Set
3505              * other attribs to 0.0.
3506              *
3507              * For fixedfunction this behavior is correct, according to the tests. For pretransformed
3508              * we'd either need a replacement shader that can load other attribs like BINORMAL, or
3509              * load the texcoord attrib pointers to match the pixel shader signature
3510              */
3511             for(i = 0; i < MAX_REG_INPUT; i++)
3512             {
3513                 semantic_name = sig[i].semantic_name;
3514                 semantic_idx = sig[i].semantic_idx;
3515                 if (!semantic_name) continue;
3516
3517                 if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_COLOR))
3518                 {
3519                     if (!semantic_idx) priv->ps_input[i] = "fragment.color.primary";
3520                     else if(semantic_idx == 1) priv->ps_input[i] = "fragment.color.secondary";
3521                     else priv->ps_input[i] = "0.0";
3522                 }
3523                 else if(args->super.vp_mode == fixedfunction)
3524                 {
3525                     priv->ps_input[i] = "0.0";
3526                 }
3527                 else if(shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_TEXCOORD))
3528                 {
3529                     if(semantic_idx < 8) priv->ps_input[i] = texcoords[semantic_idx];
3530                     else priv->ps_input[i] = "0.0";
3531                 }
3532                 else if(shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_FOG))
3533                 {
3534                     if (!semantic_idx) priv->ps_input[i] = "fragment.fogcoord";
3535                     else priv->ps_input[i] = "0.0";
3536                 }
3537                 else
3538                 {
3539                     priv->ps_input[i] = "0.0";
3540                 }
3541
3542                 TRACE("v%u, semantic %s%u is %s\n", i, semantic_name, semantic_idx, priv->ps_input[i]);
3543             }
3544             break;
3545
3546         case vertexshader:
3547             /* That one is easy. The vertex shaders provide v0-v7 in fragment.texcoord and v8 and v9 in
3548              * fragment.color
3549              */
3550             for(i = 0; i < 8; i++)
3551             {
3552                 priv->ps_input[i] = texcoords[i];
3553             }
3554             priv->ps_input[8] = "fragment.color.primary";
3555             priv->ps_input[9] = "fragment.color.secondary";
3556             break;
3557     }
3558 }
3559
3560 /* Context activation is done by the caller. */
3561 static GLuint shader_arb_generate_pshader(const struct wined3d_shader *shader,
3562         const struct wined3d_gl_info *gl_info, struct wined3d_shader_buffer *buffer,
3563         const struct arb_ps_compile_args *args, struct arb_ps_compiled_shader *compiled)
3564 {
3565     const struct wined3d_shader_reg_maps *reg_maps = &shader->reg_maps;
3566     const DWORD *function = shader->function;
3567     GLuint retval;
3568     char fragcolor[16];
3569     DWORD next_local = 0;
3570     struct shader_arb_ctx_priv priv_ctx;
3571     BOOL dcl_td = FALSE;
3572     BOOL want_nv_prog = FALSE;
3573     struct arb_pshader_private *shader_priv = shader->backend_data;
3574     GLint errPos;
3575     DWORD map;
3576
3577     char srgbtmp[4][4];
3578     unsigned int i, found = 0;
3579
3580     for (i = 0, map = reg_maps->temporary; map; map >>= 1, ++i)
3581     {
3582         if (!(map & 1)
3583                 || (shader->u.ps.color0_mov && i == shader->u.ps.color0_reg)
3584                 || (reg_maps->shader_version.major < 2 && !i))
3585             continue;
3586
3587         sprintf(srgbtmp[found], "R%u", i);
3588         ++found;
3589         if (found == 4) break;
3590     }
3591
3592     switch(found) {
3593         case 0:
3594             sprintf(srgbtmp[0], "TA");
3595             sprintf(srgbtmp[1], "TB");
3596             sprintf(srgbtmp[2], "TC");
3597             sprintf(srgbtmp[3], "TD");
3598             dcl_td = TRUE;
3599             break;
3600         case 1:
3601             sprintf(srgbtmp[1], "TA");
3602             sprintf(srgbtmp[2], "TB");
3603             sprintf(srgbtmp[3], "TC");
3604             break;
3605         case 2:
3606             sprintf(srgbtmp[2], "TA");
3607             sprintf(srgbtmp[3], "TB");
3608             break;
3609         case 3:
3610             sprintf(srgbtmp[3], "TA");
3611             break;
3612         case 4:
3613             break;
3614     }
3615
3616     /*  Create the hw ARB shader */
3617     memset(&priv_ctx, 0, sizeof(priv_ctx));
3618     priv_ctx.cur_ps_args = args;
3619     priv_ctx.compiled_fprog = compiled;
3620     priv_ctx.cur_np2fixup_info = &compiled->np2fixup_info;
3621     init_ps_input(shader, args, &priv_ctx);
3622     list_init(&priv_ctx.control_frames);
3623
3624     /* Avoid enabling NV_fragment_program* if we do not need it.
3625      *
3626      * Enabling GL_NV_fragment_program_option causes the driver to occupy a temporary register,
3627      * and it slows down the shader execution noticeably(about 5%). Usually our instruction emulation
3628      * is faster than what we gain from using higher native instructions. There are some things though
3629      * that cannot be emulated. In that case enable the extensions.
3630      * If the extension is enabled, instruction handlers that support both ways will use it.
3631      *
3632      * Testing shows no performance difference between OPTION NV_fragment_program2 and NV_fragment_program.
3633      * So enable the best we can get.
3634      */
3635     if(reg_maps->usesdsx || reg_maps->usesdsy || reg_maps->loop_depth > 0 || reg_maps->usestexldd ||
3636        reg_maps->usestexldl || reg_maps->usesfacing || reg_maps->usesifc || reg_maps->usescall)
3637     {
3638         want_nv_prog = TRUE;
3639     }
3640
3641     shader_addline(buffer, "!!ARBfp1.0\n");
3642     if (want_nv_prog && gl_info->supported[NV_FRAGMENT_PROGRAM2])
3643     {
3644         shader_addline(buffer, "OPTION NV_fragment_program2;\n");
3645         priv_ctx.target_version = NV3;
3646     }
3647     else if (want_nv_prog && gl_info->supported[NV_FRAGMENT_PROGRAM_OPTION])
3648     {
3649         shader_addline(buffer, "OPTION NV_fragment_program;\n");
3650         priv_ctx.target_version = NV2;
3651     } else {
3652         if(want_nv_prog)
3653         {
3654             /* This is an error - either we're advertising the wrong shader version, or aren't enforcing some
3655              * limits properly
3656              */
3657             ERR("The shader requires instructions that are not available in plain GL_ARB_fragment_program\n");
3658             ERR("Try GLSL\n");
3659         }
3660         priv_ctx.target_version = ARB;
3661     }
3662
3663     if (reg_maps->rt_mask > 1)
3664     {
3665         shader_addline(buffer, "OPTION ARB_draw_buffers;\n");
3666     }
3667
3668     if (reg_maps->shader_version.major < 3)
3669     {
3670         switch(args->super.fog) {
3671             case FOG_OFF:
3672                 break;
3673             case FOG_LINEAR:
3674                 shader_addline(buffer, "OPTION ARB_fog_linear;\n");
3675                 break;
3676             case FOG_EXP:
3677                 shader_addline(buffer, "OPTION ARB_fog_exp;\n");
3678                 break;
3679             case FOG_EXP2:
3680                 shader_addline(buffer, "OPTION ARB_fog_exp2;\n");
3681                 break;
3682         }
3683     }
3684
3685     /* For now always declare the temps. At least the Nvidia assembler optimizes completely
3686      * unused temps away(but occupies them for the whole shader if they're used once). Always
3687      * declaring them avoids tricky bookkeeping work
3688      */
3689     shader_addline(buffer, "TEMP TA;\n");      /* Used for modifiers */
3690     shader_addline(buffer, "TEMP TB;\n");      /* Used for modifiers */
3691     shader_addline(buffer, "TEMP TC;\n");      /* Used for modifiers */
3692     if(dcl_td) shader_addline(buffer, "TEMP TD;\n"); /* Used for sRGB writing */
3693     shader_addline(buffer, "PARAM coefdiv = { 0.5, 0.25, 0.125, 0.0625 };\n");
3694     shader_addline(buffer, "PARAM coefmul = { 2, 4, 8, 16 };\n");
3695     shader_addline(buffer, "PARAM ps_helper_const = { 0.0, 1.0, %1.10f, 0.0 };\n", eps);
3696
3697     if (reg_maps->shader_version.major < 2)
3698     {
3699         strcpy(fragcolor, "R0");
3700     }
3701     else
3702     {
3703         if (args->super.srgb_correction)
3704         {
3705             if (shader->u.ps.color0_mov)
3706             {
3707                 sprintf(fragcolor, "R%u", shader->u.ps.color0_reg);
3708             }
3709             else
3710             {
3711                 shader_addline(buffer, "TEMP TMP_COLOR;\n");
3712                 strcpy(fragcolor, "TMP_COLOR");
3713             }
3714         } else {
3715             strcpy(fragcolor, "result.color");
3716         }
3717     }
3718
3719     if(args->super.srgb_correction) {
3720         shader_addline(buffer, "PARAM srgb_consts1 = {%f, %f, %f, %f};\n",
3721                        srgb_mul_low, srgb_cmp, srgb_pow, srgb_mul_high);
3722         shader_addline(buffer, "PARAM srgb_consts2 = {%f, %f, %f, %f};\n",
3723                        srgb_sub_high, 0.0, 0.0, 0.0);
3724     }
3725
3726     /* Base Declarations */
3727     shader_generate_arb_declarations(shader, reg_maps, buffer, gl_info, NULL, &priv_ctx);
3728
3729     for (i = 0, map = reg_maps->bumpmat; map; map >>= 1, ++i)
3730     {
3731         unsigned char bump_const;
3732
3733         if (!(map & 1)) continue;
3734
3735         bump_const = compiled->numbumpenvmatconsts;
3736         compiled->bumpenvmatconst[bump_const].const_num = WINED3D_CONST_NUM_UNUSED;
3737         compiled->bumpenvmatconst[bump_const].texunit = i;
3738         compiled->luminanceconst[bump_const].const_num = WINED3D_CONST_NUM_UNUSED;
3739         compiled->luminanceconst[bump_const].texunit = i;
3740
3741         /* We can fit the constants into the constant limit for sure because texbem, texbeml, bem and beml are only supported
3742          * in 1.x shaders, and GL_ARB_fragment_program has a constant limit of 24 constants. So in the worst case we're loading
3743          * 8 shader constants, 8 bump matrices and 8 luminance parameters and are perfectly fine. (No NP2 fixup on bumpmapped
3744          * textures due to conditional NP2 restrictions)
3745          *
3746          * Use local constants to load the bump env parameters, not program.env. This avoids collisions with d3d constants of
3747          * shaders in newer shader models. Since the bump env parameters have to share their space with NP2 fixup constants,
3748          * their location is shader dependent anyway and they cannot be loaded globally.
3749          */
3750         compiled->bumpenvmatconst[bump_const].const_num = next_local++;
3751         shader_addline(buffer, "PARAM bumpenvmat%d = program.local[%d];\n",
3752                        i, compiled->bumpenvmatconst[bump_const].const_num);
3753         compiled->numbumpenvmatconsts = bump_const + 1;
3754
3755         if (!(reg_maps->luminanceparams & (1 << i))) continue;
3756
3757         compiled->luminanceconst[bump_const].const_num = next_local++;
3758         shader_addline(buffer, "PARAM luminance%d = program.local[%d];\n",
3759                        i, compiled->luminanceconst[bump_const].const_num);
3760     }
3761
3762     for(i = 0; i < MAX_CONST_I; i++)
3763     {
3764         compiled->int_consts[i] = WINED3D_CONST_NUM_UNUSED;
3765         if (reg_maps->integer_constants & (1 << i) && priv_ctx.target_version >= NV2)
3766         {
3767             const DWORD *control_values = find_loop_control_values(shader, i);
3768
3769             if(control_values)
3770             {
3771                 shader_addline(buffer, "PARAM I%u = {%u, %u, %u, -1};\n", i,
3772                                 control_values[0], control_values[1], control_values[2]);
3773             }
3774             else
3775             {
3776                 compiled->int_consts[i] = next_local;
3777                 compiled->num_int_consts++;
3778                 shader_addline(buffer, "PARAM I%u = program.local[%u];\n", i, next_local++);
3779             }
3780         }
3781     }
3782
3783     if(reg_maps->vpos || reg_maps->usesdsy)
3784     {
3785         compiled->ycorrection = next_local;
3786         shader_addline(buffer, "PARAM ycorrection = program.local[%u];\n", next_local++);
3787
3788         if(reg_maps->vpos)
3789         {
3790             shader_addline(buffer, "TEMP vpos;\n");
3791             /* ycorrection.x: Backbuffer height(onscreen) or 0(offscreen).
3792              * ycorrection.y: -1.0(onscreen), 1.0(offscreen)
3793              * ycorrection.z: 1.0
3794              * ycorrection.w: 0.0
3795              */
3796             shader_addline(buffer, "MAD vpos, fragment.position, ycorrection.zyww, ycorrection.wxww;\n");
3797             shader_addline(buffer, "FLR vpos.xy, vpos;\n");
3798         }
3799     }
3800     else
3801     {
3802         compiled->ycorrection = WINED3D_CONST_NUM_UNUSED;
3803     }
3804
3805     /* Load constants to fixup NP2 texcoords if there are still free constants left:
3806      * Constants (texture dimensions) for the NP2 fixup are loaded as local program parameters. This will consume
3807      * at most 8 (MAX_FRAGMENT_SAMPLERS / 2) parameters, which is highly unlikely, since the application had to
3808      * use 16 NP2 textures at the same time. In case that we run out of constants the fixup is simply not
3809      * applied / activated. This will probably result in wrong rendering of the texture, but will save us from
3810      * shader compilation errors and the subsequent errors when drawing with this shader. */
3811     if (priv_ctx.cur_ps_args->super.np2_fixup) {
3812         unsigned char cur_fixup_sampler = 0;
3813
3814         struct arb_ps_np2fixup_info* const fixup = priv_ctx.cur_np2fixup_info;
3815         const WORD map = priv_ctx.cur_ps_args->super.np2_fixup;
3816         const UINT max_lconsts = gl_info->limits.arb_ps_local_constants;
3817
3818         fixup->offset = next_local;
3819         fixup->super.active = 0;
3820
3821         for (i = 0; i < MAX_FRAGMENT_SAMPLERS; ++i) {
3822             if (!(map & (1 << i))) continue;
3823
3824             if (fixup->offset + (cur_fixup_sampler >> 1) < max_lconsts) {
3825                 fixup->super.active |= (1 << i);
3826                 fixup->super.idx[i] = cur_fixup_sampler++;
3827             } else {
3828                 FIXME("No free constant found to load NP2 fixup data into shader. "
3829                       "Sampling from this texture will probably look wrong.\n");
3830                 break;
3831             }
3832         }
3833
3834         fixup->super.num_consts = (cur_fixup_sampler + 1) >> 1;
3835         if (fixup->super.num_consts) {
3836             shader_addline(buffer, "PARAM np2fixup[%u] = { program.env[%u..%u] };\n",
3837                            fixup->super.num_consts, fixup->offset, fixup->super.num_consts + fixup->offset - 1);
3838         }
3839     }
3840
3841     if (shader_priv->clipplane_emulation != ~0U && args->clip)
3842     {
3843         shader_addline(buffer, "KIL fragment.texcoord[%u];\n", shader_priv->clipplane_emulation);
3844     }
3845
3846     /* Base Shader Body */
3847     shader_generate_main(shader, buffer, reg_maps, function, &priv_ctx);
3848
3849     if(args->super.srgb_correction) {
3850         arbfp_add_sRGB_correction(buffer, fragcolor, srgbtmp[0], srgbtmp[1], srgbtmp[2], srgbtmp[3],
3851                                   priv_ctx.target_version >= NV2);
3852     }
3853
3854     if(strcmp(fragcolor, "result.color")) {
3855         shader_addline(buffer, "MOV result.color, %s;\n", fragcolor);
3856     }
3857     shader_addline(buffer, "END\n");
3858
3859     /* TODO: change to resource.glObjectHandle or something like that */
3860     GL_EXTCALL(glGenProgramsARB(1, &retval));
3861
3862     TRACE("Creating a hw pixel shader, prg=%d\n", retval);
3863     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, retval));
3864
3865     TRACE("Created hw pixel shader, prg=%d\n", retval);
3866     /* Create the program and check for errors */
3867     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
3868                buffer->content_size, buffer->buffer));
3869     checkGLcall("glProgramStringARB()");
3870
3871     gl_info->gl_ops.gl.p_glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &errPos);
3872     if (errPos != -1)
3873     {
3874         FIXME("HW PixelShader Error at position %d: %s\n\n",
3875               errPos, debugstr_a((const char *)gl_info->gl_ops.gl.p_glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
3876         shader_arb_dump_program_source(buffer->buffer);
3877         retval = 0;
3878     }
3879     else
3880     {
3881         GLint native;
3882
3883         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
3884         checkGLcall("glGetProgramivARB()");
3885         if (!native) WARN("Program exceeds native resource limits.\n");
3886     }
3887
3888     return retval;
3889 }
3890
3891 static int compare_sig(const struct wined3d_shader_signature_element *sig1, const struct wined3d_shader_signature_element *sig2)
3892 {
3893     unsigned int i;
3894     int ret;
3895
3896     for(i = 0; i < MAX_REG_INPUT; i++)
3897     {
3898         if (!sig1[i].semantic_name || !sig2[i].semantic_name)
3899         {
3900             /* Compare pointers, not contents. One string is NULL(element does not exist), the other one is not NULL */
3901             if(sig1[i].semantic_name != sig2[i].semantic_name) return sig1[i].semantic_name < sig2[i].semantic_name ? -1 : 1;
3902             continue;
3903         }
3904
3905         if ((ret = strcmp(sig1[i].semantic_name, sig2[i].semantic_name))) return ret;
3906         if(sig1[i].semantic_idx    != sig2[i].semantic_idx)    return sig1[i].semantic_idx    < sig2[i].semantic_idx    ? -1 : 1;
3907         if(sig1[i].sysval_semantic != sig2[i].sysval_semantic) return sig1[i].sysval_semantic < sig2[i].sysval_semantic ? -1 : 1;
3908         if(sig1[i].component_type  != sig2[i].component_type)  return sig1[i].component_type  < sig2[i].component_type  ? -1 : 1;
3909         if(sig1[i].register_idx    != sig2[i].register_idx)    return sig1[i].register_idx    < sig2[i].register_idx    ? -1 : 1;
3910         if(sig1[i].mask            != sig2[i].mask)            return sig1[i].mask            < sig2[i].mask            ? -1 : 1;
3911     }
3912     return 0;
3913 }
3914
3915 static struct wined3d_shader_signature_element *clone_sig(const struct wined3d_shader_signature_element *sig)
3916 {
3917     struct wined3d_shader_signature_element *new;
3918     int i;
3919     char *name;
3920
3921     new = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*new) * MAX_REG_INPUT);
3922     for(i = 0; i < MAX_REG_INPUT; i++)
3923     {
3924         if (!sig[i].semantic_name) continue;
3925
3926         new[i] = sig[i];
3927         /* Clone the semantic string */
3928         name = HeapAlloc(GetProcessHeap(), 0, strlen(sig[i].semantic_name) + 1);
3929         strcpy(name, sig[i].semantic_name);
3930         new[i].semantic_name = name;
3931     }
3932     return new;
3933 }
3934
3935 static DWORD find_input_signature(struct shader_arb_priv *priv, const struct wined3d_shader_signature_element *sig)
3936 {
3937     struct wine_rb_entry *entry = wine_rb_get(&priv->signature_tree, sig);
3938     struct ps_signature *found_sig;
3939
3940     if (entry)
3941     {
3942         found_sig = WINE_RB_ENTRY_VALUE(entry, struct ps_signature, entry);
3943         TRACE("Found existing signature %u\n", found_sig->idx);
3944         return found_sig->idx;
3945     }
3946     found_sig = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*sig));
3947     found_sig->sig = clone_sig(sig);
3948     found_sig->idx = priv->ps_sig_number++;
3949     TRACE("New signature stored and assigned number %u\n", found_sig->idx);
3950     if(wine_rb_put(&priv->signature_tree, sig, &found_sig->entry) == -1)
3951     {
3952         ERR("Failed to insert program entry.\n");
3953     }
3954     return found_sig->idx;
3955 }
3956
3957 static void init_output_registers(const struct wined3d_shader *shader, DWORD sig_num,
3958         struct shader_arb_ctx_priv *priv_ctx, struct arb_vs_compiled_shader *compiled)
3959 {
3960     unsigned int i, j;
3961     static const char * const texcoords[8] =
3962     {
3963         "result.texcoord[0]", "result.texcoord[1]", "result.texcoord[2]", "result.texcoord[3]",
3964         "result.texcoord[4]", "result.texcoord[5]", "result.texcoord[6]", "result.texcoord[7]"
3965     };
3966     struct wined3d_device *device = shader->device;
3967     const struct wined3d_shader_signature_element *sig;
3968     const char *semantic_name;
3969     DWORD semantic_idx, reg_idx;
3970
3971     /* Write generic input varyings 0 to 7 to result.texcoord[], varying 8 to result.color.primary
3972      * and varying 9 to result.color.secondary
3973      */
3974     static const char * const decl_idx_to_string[MAX_REG_INPUT] =
3975     {
3976         "result.texcoord[0]", "result.texcoord[1]", "result.texcoord[2]", "result.texcoord[3]",
3977         "result.texcoord[4]", "result.texcoord[5]", "result.texcoord[6]", "result.texcoord[7]",
3978         "result.color.primary", "result.color.secondary"
3979     };
3980
3981     if(sig_num == ~0)
3982     {
3983         TRACE("Pixel shader uses builtin varyings\n");
3984         /* Map builtins to builtins */
3985         for(i = 0; i < 8; i++)
3986         {
3987             priv_ctx->texcrd_output[i] = texcoords[i];
3988         }
3989         priv_ctx->color_output[0] = "result.color.primary";
3990         priv_ctx->color_output[1] = "result.color.secondary";
3991         priv_ctx->fog_output = "TMP_FOGCOORD";
3992
3993         /* Map declared regs to builtins. Use "TA" to /dev/null unread output */
3994         for (i = 0; i < (sizeof(shader->output_signature) / sizeof(*shader->output_signature)); ++i)
3995         {
3996             semantic_name = shader->output_signature[i].semantic_name;
3997             if (!semantic_name) continue;
3998
3999             if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_POSITION))
4000             {
4001                 TRACE("o%u is TMP_OUT\n", i);
4002                 if (!shader->output_signature[i].semantic_idx) priv_ctx->vs_output[i] = "TMP_OUT";
4003                 else priv_ctx->vs_output[i] = "TA";
4004             }
4005             else if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_PSIZE))
4006             {
4007                 TRACE("o%u is result.pointsize\n", i);
4008                 if (!shader->output_signature[i].semantic_idx) priv_ctx->vs_output[i] = "result.pointsize";
4009                 else priv_ctx->vs_output[i] = "TA";
4010             }
4011             else if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_COLOR))
4012             {
4013                 TRACE("o%u is result.color.?, idx %u\n", i, shader->output_signature[i].semantic_idx);
4014                 if (!shader->output_signature[i].semantic_idx)
4015                     priv_ctx->vs_output[i] = "result.color.primary";
4016                 else if (shader->output_signature[i].semantic_idx == 1)
4017                     priv_ctx->vs_output[i] = "result.color.secondary";
4018                 else priv_ctx->vs_output[i] = "TA";
4019             }
4020             else if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_TEXCOORD))
4021             {
4022                 TRACE("o%u is %s\n", i, texcoords[shader->output_signature[i].semantic_idx]);
4023                 if (shader->output_signature[i].semantic_idx >= 8) priv_ctx->vs_output[i] = "TA";
4024                 else priv_ctx->vs_output[i] = texcoords[shader->output_signature[i].semantic_idx];
4025             }
4026             else if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_FOG))
4027             {
4028                 TRACE("o%u is result.fogcoord\n", i);
4029                 if (shader->output_signature[i].semantic_idx > 0) priv_ctx->vs_output[i] = "TA";
4030                 else priv_ctx->vs_output[i] = "result.fogcoord";
4031             }
4032             else
4033             {
4034                 priv_ctx->vs_output[i] = "TA";
4035             }
4036         }
4037         return;
4038     }
4039
4040     /* Instead of searching for the signature in the signature list, read the one from the current pixel shader.
4041      * Its maybe not the shader where the signature came from, but it is the same signature and faster to find
4042      */
4043     sig = device->stateBlock->state.pixel_shader->input_signature;
4044     TRACE("Pixel shader uses declared varyings\n");
4045
4046     /* Map builtin to declared. /dev/null the results by default to the TA temp reg */
4047     for(i = 0; i < 8; i++)
4048     {
4049         priv_ctx->texcrd_output[i] = "TA";
4050     }
4051     priv_ctx->color_output[0] = "TA";
4052     priv_ctx->color_output[1] = "TA";
4053     priv_ctx->fog_output = "TA";
4054
4055     for(i = 0; i < MAX_REG_INPUT; i++)
4056     {
4057         semantic_name = sig[i].semantic_name;
4058         semantic_idx = sig[i].semantic_idx;
4059         reg_idx = sig[i].register_idx;
4060         if (!semantic_name) continue;
4061
4062         /* If a declared input register is not written by builtin arguments, don't write to it.
4063          * GL_NV_vertex_program makes sure the input defaults to 0.0, which is correct with D3D
4064          *
4065          * Don't care about POSITION and PSIZE here - this is a builtin vertex shader, position goes
4066          * to TMP_OUT in any case
4067          */
4068         if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_TEXCOORD))
4069         {
4070             if (semantic_idx < 8)
4071                 priv_ctx->texcrd_output[semantic_idx] = decl_idx_to_string[reg_idx];
4072         }
4073         else if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_COLOR))
4074         {
4075             if (semantic_idx < 2)
4076                 priv_ctx->color_output[semantic_idx] = decl_idx_to_string[reg_idx];
4077         }
4078         else if(shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_FOG))
4079         {
4080             if (!semantic_idx)
4081                 priv_ctx->fog_output = decl_idx_to_string[reg_idx];
4082         }
4083         else
4084         {
4085             continue;
4086         }
4087
4088         if (!strcmp(decl_idx_to_string[reg_idx], "result.color.primary")
4089                 || !strcmp(decl_idx_to_string[reg_idx], "result.color.secondary"))
4090         {
4091             compiled->need_color_unclamp = TRUE;
4092         }
4093     }
4094
4095     /* Map declared to declared */
4096     for (i = 0; i < (sizeof(shader->output_signature) / sizeof(*shader->output_signature)); ++i)
4097     {
4098         /* Write unread output to TA to throw them away */
4099         priv_ctx->vs_output[i] = "TA";
4100         semantic_name = shader->output_signature[i].semantic_name;
4101         if (!semantic_name) continue;
4102
4103         if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_POSITION)
4104                 && !shader->output_signature[i].semantic_idx)
4105         {
4106             priv_ctx->vs_output[i] = "TMP_OUT";
4107             continue;
4108         }
4109         else if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_PSIZE)
4110                 && !shader->output_signature[i].semantic_idx)
4111         {
4112             priv_ctx->vs_output[i] = "result.pointsize";
4113             continue;
4114         }
4115
4116         for(j = 0; j < MAX_REG_INPUT; j++)
4117         {
4118             if (!sig[j].semantic_name) continue;
4119
4120             if (!strcmp(sig[j].semantic_name, semantic_name)
4121                     && sig[j].semantic_idx == shader->output_signature[i].semantic_idx)
4122             {
4123                 priv_ctx->vs_output[i] = decl_idx_to_string[sig[j].register_idx];
4124
4125                 if (!strcmp(priv_ctx->vs_output[i], "result.color.primary")
4126                         || !strcmp(priv_ctx->vs_output[i], "result.color.secondary"))
4127                 {
4128                     compiled->need_color_unclamp = TRUE;
4129                 }
4130             }
4131         }
4132     }
4133 }
4134
4135 /* Context activation is done by the caller. */
4136 static GLuint shader_arb_generate_vshader(const struct wined3d_shader *shader,
4137         const struct wined3d_gl_info *gl_info, struct wined3d_shader_buffer *buffer,
4138         const struct arb_vs_compile_args *args, struct arb_vs_compiled_shader *compiled)
4139 {
4140     const struct arb_vshader_private *shader_data = shader->backend_data;
4141     const struct wined3d_shader_reg_maps *reg_maps = &shader->reg_maps;
4142     struct shader_arb_priv *priv = shader->device->shader_priv;
4143     const DWORD *function = shader->function;
4144     GLuint ret;
4145     DWORD next_local = 0;
4146     struct shader_arb_ctx_priv priv_ctx;
4147     unsigned int i;
4148     GLint errPos;
4149
4150     memset(&priv_ctx, 0, sizeof(priv_ctx));
4151     priv_ctx.cur_vs_args = args;
4152     list_init(&priv_ctx.control_frames);
4153     init_output_registers(shader, args->ps_signature, &priv_ctx, compiled);
4154
4155     /*  Create the hw ARB shader */
4156     shader_addline(buffer, "!!ARBvp1.0\n");
4157
4158     /* Always enable the NV extension if available. Unlike fragment shaders, there is no
4159      * mesurable performance penalty, and we can always make use of it for clipplanes.
4160      */
4161     if (gl_info->supported[NV_VERTEX_PROGRAM3])
4162     {
4163         shader_addline(buffer, "OPTION NV_vertex_program3;\n");
4164         priv_ctx.target_version = NV3;
4165         shader_addline(buffer, "ADDRESS aL;\n");
4166     }
4167     else if (gl_info->supported[NV_VERTEX_PROGRAM2_OPTION])
4168     {
4169         shader_addline(buffer, "OPTION NV_vertex_program2;\n");
4170         priv_ctx.target_version = NV2;
4171         shader_addline(buffer, "ADDRESS aL;\n");
4172     } else {
4173         priv_ctx.target_version = ARB;
4174     }
4175
4176     shader_addline(buffer, "TEMP TMP_OUT;\n");
4177     if (reg_maps->fog)
4178         shader_addline(buffer, "TEMP TMP_FOGCOORD;\n");
4179     if (need_helper_const(shader_data, reg_maps, gl_info))
4180     {
4181         shader_addline(buffer, "PARAM helper_const = { 0.0, 1.0, 2.0, %1.10f};\n", eps);
4182     }
4183     if (need_rel_addr_const(shader_data, reg_maps, gl_info))
4184     {
4185         shader_addline(buffer, "PARAM rel_addr_const = { 0.5, %d.0, 0.0, 0.0 };\n", shader_data->rel_offset);
4186         shader_addline(buffer, "TEMP A0_SHADOW;\n");
4187     }
4188
4189     shader_addline(buffer, "TEMP TA;\n");
4190     shader_addline(buffer, "TEMP TB;\n");
4191
4192     /* Base Declarations */
4193     shader_generate_arb_declarations(shader, reg_maps, buffer, gl_info,
4194             &priv_ctx.vs_clipplanes, &priv_ctx);
4195
4196     for(i = 0; i < MAX_CONST_I; i++)
4197     {
4198         compiled->int_consts[i] = WINED3D_CONST_NUM_UNUSED;
4199         if(reg_maps->integer_constants & (1 << i) && priv_ctx.target_version >= NV2)
4200         {
4201             const DWORD *control_values = find_loop_control_values(shader, i);
4202
4203             if(control_values)
4204             {
4205                 shader_addline(buffer, "PARAM I%u = {%u, %u, %u, -1};\n", i,
4206                                 control_values[0], control_values[1], control_values[2]);
4207             }
4208             else
4209             {
4210                 compiled->int_consts[i] = next_local;
4211                 compiled->num_int_consts++;
4212                 shader_addline(buffer, "PARAM I%u = program.local[%u];\n", i, next_local++);
4213             }
4214         }
4215     }
4216
4217     /* We need a constant to fixup the final position */
4218     shader_addline(buffer, "PARAM posFixup = program.local[%u];\n", next_local);
4219     compiled->pos_fixup = next_local++;
4220
4221     /* Initialize output parameters. GL_ARB_vertex_program does not require special initialization values
4222      * for output parameters. D3D in theory does not do that either, but some applications depend on a
4223      * proper initialization of the secondary color, and programs using the fixed function pipeline without
4224      * a replacement shader depend on the texcoord.w being set properly.
4225      *
4226      * GL_NV_vertex_program defines that all output values are initialized to {0.0, 0.0, 0.0, 1.0}. This
4227      * assertion is in effect even when using GL_ARB_vertex_program without any NV specific additions. So
4228      * skip this if NV_vertex_program is supported. Otherwise, initialize the secondary color. For the tex-
4229      * coords, we have a flag in the opengl caps. Many cards do not require the texcoord being set, and
4230      * this can eat a number of instructions, so skip it unless this cap is set as well
4231      */
4232     if (!gl_info->supported[NV_VERTEX_PROGRAM])
4233     {
4234         const char *color_init = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_0001);
4235         shader_addline(buffer, "MOV result.color.secondary, %s;\n", color_init);
4236
4237         if (gl_info->quirks & WINED3D_QUIRK_SET_TEXCOORD_W && !priv->ffp_proj_control)
4238         {
4239             int i;
4240             const char *one = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_ONE);
4241             for(i = 0; i < min(8, MAX_REG_TEXCRD); i++)
4242             {
4243                 if (reg_maps->texcoord_mask[i] && reg_maps->texcoord_mask[i] != WINED3DSP_WRITEMASK_ALL)
4244                     shader_addline(buffer, "MOV result.texcoord[%u].w, %s\n", i, one);
4245             }
4246         }
4247     }
4248
4249     /* The shader starts with the main function */
4250     priv_ctx.in_main_func = TRUE;
4251     /* Base Shader Body */
4252     shader_generate_main(shader, buffer, reg_maps, function, &priv_ctx);
4253
4254     if (!priv_ctx.footer_written) vshader_add_footer(&priv_ctx,
4255             shader_data, args, reg_maps, gl_info, buffer);
4256
4257     shader_addline(buffer, "END\n");
4258
4259     /* TODO: change to resource.glObjectHandle or something like that */
4260     GL_EXTCALL(glGenProgramsARB(1, &ret));
4261
4262     TRACE("Creating a hw vertex shader, prg=%d\n", ret);
4263     GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, ret));
4264
4265     TRACE("Created hw vertex shader, prg=%d\n", ret);
4266     /* Create the program and check for errors */
4267     GL_EXTCALL(glProgramStringARB(GL_VERTEX_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
4268                buffer->content_size, buffer->buffer));
4269     checkGLcall("glProgramStringARB()");
4270
4271     gl_info->gl_ops.gl.p_glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &errPos);
4272     if (errPos != -1)
4273     {
4274         FIXME("HW VertexShader Error at position %d: %s\n\n",
4275               errPos, debugstr_a((const char *)gl_info->gl_ops.gl.p_glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
4276         shader_arb_dump_program_source(buffer->buffer);
4277         ret = -1;
4278     }
4279     else
4280     {
4281         GLint native;
4282
4283         GL_EXTCALL(glGetProgramivARB(GL_VERTEX_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
4284         checkGLcall("glGetProgramivARB()");
4285         if (!native) WARN("Program exceeds native resource limits.\n");
4286     }
4287
4288     return ret;
4289 }
4290
4291 /* Context activation is done by the caller. */
4292 static struct arb_ps_compiled_shader *find_arb_pshader(struct wined3d_shader *shader,
4293         const struct arb_ps_compile_args *args)
4294 {
4295     struct wined3d_device *device = shader->device;
4296     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4297     const struct wined3d_d3d_info *d3d_info = &device->adapter->d3d_info;
4298     UINT i;
4299     DWORD new_size;
4300     struct arb_ps_compiled_shader *new_array;
4301     struct wined3d_shader_buffer buffer;
4302     struct arb_pshader_private *shader_data;
4303     GLuint ret;
4304
4305     if (!shader->backend_data)
4306     {
4307         struct shader_arb_priv *priv = device->shader_priv;
4308
4309         shader->backend_data = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data));
4310         shader_data = shader->backend_data;
4311         shader_data->clamp_consts = shader->reg_maps.shader_version.major == 1;
4312
4313         if (shader->reg_maps.shader_version.major < 3)
4314             shader_data->input_signature_idx = ~0;
4315         else
4316             shader_data->input_signature_idx = find_input_signature(priv, shader->input_signature);
4317
4318         TRACE("Shader got assigned input signature index %u\n", shader_data->input_signature_idx);
4319
4320         if (!d3d_info->vs_clipping)
4321             shader_data->clipplane_emulation = shader_find_free_input_register(&shader->reg_maps,
4322                     d3d_info->limits.ffp_blend_stages - 1);
4323         else
4324             shader_data->clipplane_emulation = ~0U;
4325     }
4326     shader_data = shader->backend_data;
4327
4328     /* Usually we have very few GL shaders for each d3d shader(just 1 or maybe 2),
4329      * so a linear search is more performant than a hashmap or a binary search
4330      * (cache coherency etc)
4331      */
4332     for (i = 0; i < shader_data->num_gl_shaders; ++i)
4333     {
4334         if (!memcmp(&shader_data->gl_shaders[i].args, args, sizeof(*args)))
4335             return &shader_data->gl_shaders[i];
4336     }
4337
4338     TRACE("No matching GL shader found, compiling a new shader\n");
4339     if(shader_data->shader_array_size == shader_data->num_gl_shaders) {
4340         if (shader_data->num_gl_shaders)
4341         {
4342             new_size = shader_data->shader_array_size + max(1, shader_data->shader_array_size / 2);
4343             new_array = HeapReAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, shader_data->gl_shaders,
4344                                     new_size * sizeof(*shader_data->gl_shaders));
4345         } else {
4346             new_array = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data->gl_shaders));
4347             new_size = 1;
4348         }
4349
4350         if(!new_array) {
4351             ERR("Out of memory\n");
4352             return 0;
4353         }
4354         shader_data->gl_shaders = new_array;
4355         shader_data->shader_array_size = new_size;
4356     }
4357
4358     shader_data->gl_shaders[shader_data->num_gl_shaders].args = *args;
4359
4360     pixelshader_update_samplers(shader, args->super.tex_types);
4361
4362     if (!shader_buffer_init(&buffer))
4363     {
4364         ERR("Failed to initialize shader buffer.\n");
4365         return 0;
4366     }
4367
4368     ret = shader_arb_generate_pshader(shader, gl_info, &buffer, args,
4369             &shader_data->gl_shaders[shader_data->num_gl_shaders]);
4370     shader_buffer_free(&buffer);
4371     shader_data->gl_shaders[shader_data->num_gl_shaders].prgId = ret;
4372
4373     return &shader_data->gl_shaders[shader_data->num_gl_shaders++];
4374 }
4375
4376 static inline BOOL vs_args_equal(const struct arb_vs_compile_args *stored, const struct arb_vs_compile_args *new,
4377                                  const DWORD use_map, BOOL skip_int) {
4378     if((stored->super.swizzle_map & use_map) != new->super.swizzle_map) return FALSE;
4379     if(stored->super.clip_enabled != new->super.clip_enabled) return FALSE;
4380     if(stored->super.fog_src != new->super.fog_src) return FALSE;
4381     if(stored->clip.boolclip_compare != new->clip.boolclip_compare) return FALSE;
4382     if(stored->ps_signature != new->ps_signature) return FALSE;
4383     if(stored->vertex.samplers_compare != new->vertex.samplers_compare) return FALSE;
4384     if(skip_int) return TRUE;
4385
4386     return !memcmp(stored->loop_ctrl, new->loop_ctrl, sizeof(stored->loop_ctrl));
4387 }
4388
4389 static struct arb_vs_compiled_shader *find_arb_vshader(struct wined3d_shader *shader,
4390         const struct arb_vs_compile_args *args)
4391 {
4392     struct wined3d_device *device = shader->device;
4393     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4394     DWORD use_map = device->stream_info.use_map;
4395     UINT i;
4396     DWORD new_size;
4397     struct arb_vs_compiled_shader *new_array;
4398     struct wined3d_shader_buffer buffer;
4399     struct arb_vshader_private *shader_data;
4400     GLuint ret;
4401
4402     if (!shader->backend_data)
4403     {
4404         const struct wined3d_shader_reg_maps *reg_maps = &shader->reg_maps;
4405
4406         shader->backend_data = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data));
4407         shader_data = shader->backend_data;
4408
4409         if ((gl_info->quirks & WINED3D_QUIRK_ARB_VS_OFFSET_LIMIT)
4410                 && reg_maps->min_rel_offset <= reg_maps->max_rel_offset)
4411         {
4412             if (reg_maps->max_rel_offset - reg_maps->min_rel_offset > 127)
4413             {
4414                 FIXME("The difference between the minimum and maximum relative offset is > 127.\n");
4415                 FIXME("Which this OpenGL implementation does not support. Try using GLSL.\n");
4416                 FIXME("Min: %u, Max: %u.\n", reg_maps->min_rel_offset, reg_maps->max_rel_offset);
4417             }
4418             else if (reg_maps->max_rel_offset - reg_maps->min_rel_offset > 63)
4419                 shader_data->rel_offset = reg_maps->min_rel_offset + 63;
4420             else if (reg_maps->max_rel_offset > 63)
4421                 shader_data->rel_offset = reg_maps->min_rel_offset;
4422         }
4423     }
4424     shader_data = shader->backend_data;
4425
4426     /* Usually we have very few GL shaders for each d3d shader(just 1 or maybe 2),
4427      * so a linear search is more performant than a hashmap or a binary search
4428      * (cache coherency etc)
4429      */
4430     for(i = 0; i < shader_data->num_gl_shaders; i++) {
4431         if (vs_args_equal(&shader_data->gl_shaders[i].args, args,
4432                 use_map, gl_info->supported[NV_VERTEX_PROGRAM2_OPTION]))
4433         {
4434             return &shader_data->gl_shaders[i];
4435         }
4436     }
4437
4438     TRACE("No matching GL shader found, compiling a new shader\n");
4439
4440     if(shader_data->shader_array_size == shader_data->num_gl_shaders) {
4441         if (shader_data->num_gl_shaders)
4442         {
4443             new_size = shader_data->shader_array_size + max(1, shader_data->shader_array_size / 2);
4444             new_array = HeapReAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, shader_data->gl_shaders,
4445                                     new_size * sizeof(*shader_data->gl_shaders));
4446         } else {
4447             new_array = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data->gl_shaders));
4448             new_size = 1;
4449         }
4450
4451         if(!new_array) {
4452             ERR("Out of memory\n");
4453             return 0;
4454         }
4455         shader_data->gl_shaders = new_array;
4456         shader_data->shader_array_size = new_size;
4457     }
4458
4459     shader_data->gl_shaders[shader_data->num_gl_shaders].args = *args;
4460
4461     if (!shader_buffer_init(&buffer))
4462     {
4463         ERR("Failed to initialize shader buffer.\n");
4464         return 0;
4465     }
4466
4467     ret = shader_arb_generate_vshader(shader, gl_info, &buffer, args,
4468             &shader_data->gl_shaders[shader_data->num_gl_shaders]);
4469     shader_buffer_free(&buffer);
4470     shader_data->gl_shaders[shader_data->num_gl_shaders].prgId = ret;
4471
4472     return &shader_data->gl_shaders[shader_data->num_gl_shaders++];
4473 }
4474
4475 static void find_arb_ps_compile_args(const struct wined3d_state *state,
4476         const struct wined3d_context *context, const struct wined3d_shader *shader,
4477         struct arb_ps_compile_args *args)
4478 {
4479     const struct wined3d_gl_info *gl_info = context->gl_info;
4480     const struct wined3d_d3d_info *d3d_info = context->d3d_info;
4481     int i;
4482     WORD int_skip;
4483
4484     find_ps_compile_args(state, shader, &args->super);
4485
4486     /* This forces all local boolean constants to 1 to make them stateblock independent */
4487     args->bools = shader->reg_maps.local_bool_consts;
4488
4489     for(i = 0; i < MAX_CONST_B; i++)
4490     {
4491         if (state->ps_consts_b[i])
4492             args->bools |= ( 1 << i);
4493     }
4494
4495     /* Only enable the clip plane emulation KIL if at least one clipplane is enabled. The KIL instruction
4496      * is quite expensive because it forces the driver to disable early Z discards. It is cheaper to
4497      * duplicate the shader than have a no-op KIL instruction in every shader
4498      */
4499     if (!d3d_info->vs_clipping && use_vs(state)
4500             && state->render_states[WINED3D_RS_CLIPPING]
4501             && state->render_states[WINED3D_RS_CLIPPLANEENABLE])
4502         args->clip = 1;
4503     else
4504         args->clip = 0;
4505
4506     /* Skip if unused or local, or supported natively */
4507     int_skip = ~shader->reg_maps.integer_constants | shader->reg_maps.local_int_consts;
4508     if (int_skip == 0xffff || gl_info->supported[NV_FRAGMENT_PROGRAM_OPTION])
4509     {
4510         memset(args->loop_ctrl, 0, sizeof(args->loop_ctrl));
4511         return;
4512     }
4513
4514     for(i = 0; i < MAX_CONST_I; i++)
4515     {
4516         if(int_skip & (1 << i))
4517         {
4518             args->loop_ctrl[i][0] = 0;
4519             args->loop_ctrl[i][1] = 0;
4520             args->loop_ctrl[i][2] = 0;
4521         }
4522         else
4523         {
4524             args->loop_ctrl[i][0] = state->ps_consts_i[i * 4];
4525             args->loop_ctrl[i][1] = state->ps_consts_i[i * 4 + 1];
4526             args->loop_ctrl[i][2] = state->ps_consts_i[i * 4 + 2];
4527         }
4528     }
4529 }
4530
4531 static void find_arb_vs_compile_args(const struct wined3d_state *state,
4532         const struct wined3d_context *context, const struct wined3d_shader *shader,
4533         struct arb_vs_compile_args *args)
4534 {
4535     struct wined3d_device *device = shader->device;
4536     const struct wined3d_adapter *adapter = device->adapter;
4537     const struct wined3d_gl_info *gl_info = context->gl_info;
4538     const struct wined3d_d3d_info *d3d_info = context->d3d_info;
4539     int i;
4540     WORD int_skip;
4541
4542     find_vs_compile_args(state, shader, &args->super);
4543
4544     args->clip.boolclip_compare = 0;
4545     if (use_ps(state))
4546     {
4547         const struct wined3d_shader *ps = state->pixel_shader;
4548         const struct arb_pshader_private *shader_priv = ps->backend_data;
4549         args->ps_signature = shader_priv->input_signature_idx;
4550
4551         args->clip.boolclip.clip_texcoord = shader_priv->clipplane_emulation + 1;
4552     }
4553     else
4554     {
4555         args->ps_signature = ~0;
4556         if (!d3d_info->vs_clipping && adapter->fragment_pipe == &arbfp_fragment_pipeline)
4557         {
4558             args->clip.boolclip.clip_texcoord = ffp_clip_emul(state) ? d3d_info->limits.ffp_blend_stages : 0;
4559         }
4560         /* Otherwise: Setting boolclip_compare set clip_texcoord to 0 */
4561     }
4562
4563     if (args->clip.boolclip.clip_texcoord)
4564     {
4565         if (state->render_states[WINED3D_RS_CLIPPING])
4566             args->clip.boolclip.clipplane_mask = (unsigned char)state->render_states[WINED3D_RS_CLIPPLANEENABLE];
4567         /* clipplane_mask was set to 0 by setting boolclip_compare to 0 */
4568     }
4569
4570     /* This forces all local boolean constants to 1 to make them stateblock independent */
4571     args->clip.boolclip.bools = shader->reg_maps.local_bool_consts;
4572     /* TODO: Figure out if it would be better to store bool constants as bitmasks in the stateblock */
4573     for(i = 0; i < MAX_CONST_B; i++)
4574     {
4575         if (state->vs_consts_b[i])
4576             args->clip.boolclip.bools |= ( 1 << i);
4577     }
4578
4579     args->vertex.samplers[0] = device->texUnitMap[MAX_FRAGMENT_SAMPLERS + 0];
4580     args->vertex.samplers[1] = device->texUnitMap[MAX_FRAGMENT_SAMPLERS + 1];
4581     args->vertex.samplers[2] = device->texUnitMap[MAX_FRAGMENT_SAMPLERS + 2];
4582     args->vertex.samplers[3] = 0;
4583
4584     /* Skip if unused or local */
4585     int_skip = ~shader->reg_maps.integer_constants | shader->reg_maps.local_int_consts;
4586     /* This is about flow control, not clipping. */
4587     if (int_skip == 0xffff || gl_info->supported[NV_VERTEX_PROGRAM2_OPTION])
4588     {
4589         memset(args->loop_ctrl, 0, sizeof(args->loop_ctrl));
4590         return;
4591     }
4592
4593     for(i = 0; i < MAX_CONST_I; i++)
4594     {
4595         if(int_skip & (1 << i))
4596         {
4597             args->loop_ctrl[i][0] = 0;
4598             args->loop_ctrl[i][1] = 0;
4599             args->loop_ctrl[i][2] = 0;
4600         }
4601         else
4602         {
4603             args->loop_ctrl[i][0] = state->vs_consts_i[i * 4];
4604             args->loop_ctrl[i][1] = state->vs_consts_i[i * 4 + 1];
4605             args->loop_ctrl[i][2] = state->vs_consts_i[i * 4 + 2];
4606         }
4607     }
4608 }
4609
4610 /* Context activation is done by the caller. */
4611 static void shader_arb_select(const struct wined3d_context *context, enum wined3d_shader_mode vertex_mode,
4612         enum wined3d_shader_mode fragment_mode)
4613 {
4614     struct wined3d_device *device = context->swapchain->device;
4615     struct shader_arb_priv *priv = device->shader_priv;
4616     const struct wined3d_gl_info *gl_info = context->gl_info;
4617     const struct wined3d_state *state = &device->stateBlock->state;
4618     int i;
4619
4620     /* Deal with pixel shaders first so the vertex shader arg function has the input signature ready */
4621     if (fragment_mode == WINED3D_SHADER_MODE_SHADER)
4622     {
4623         struct wined3d_shader *ps = state->pixel_shader;
4624         struct arb_ps_compile_args compile_args;
4625         struct arb_ps_compiled_shader *compiled;
4626
4627         TRACE("Using pixel shader %p.\n", ps);
4628         find_arb_ps_compile_args(state, context, ps, &compile_args);
4629         compiled = find_arb_pshader(ps, &compile_args);
4630         priv->current_fprogram_id = compiled->prgId;
4631         priv->compiled_fprog = compiled;
4632
4633         /* Bind the fragment program */
4634         GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, priv->current_fprogram_id));
4635         checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, priv->current_fprogram_id);");
4636
4637         if (!priv->use_arbfp_fixed_func)
4638             priv->fragment_pipe->enable_extension(gl_info, FALSE);
4639
4640         /* Enable OpenGL fragment programs. */
4641         gl_info->gl_ops.gl.p_glEnable(GL_FRAGMENT_PROGRAM_ARB);
4642         checkGLcall("glEnable(GL_FRAGMENT_PROGRAM_ARB);");
4643
4644         TRACE("(%p) : Bound fragment program %u and enabled GL_FRAGMENT_PROGRAM_ARB\n",
4645                 device, priv->current_fprogram_id);
4646
4647         /* Pixel Shader 1.x constants are clamped to [-1;1], Pixel Shader 2.0 constants are not. If switching between
4648          * a 1.x and newer shader, reload the first 8 constants
4649          */
4650         if (priv->last_ps_const_clamped != ((struct arb_pshader_private *)ps->backend_data)->clamp_consts)
4651         {
4652             priv->last_ps_const_clamped = ((struct arb_pshader_private *)ps->backend_data)->clamp_consts;
4653             priv->highest_dirty_ps_const = max(priv->highest_dirty_ps_const, 8);
4654             for(i = 0; i < 8; i++)
4655             {
4656                 priv->pshader_const_dirty[i] = 1;
4657             }
4658             /* Also takes care of loading local constants */
4659             shader_arb_load_constants_internal(context, TRUE, FALSE, TRUE);
4660         }
4661         else
4662         {
4663             UINT rt_height = state->fb->render_targets[0]->resource.height;
4664             shader_arb_ps_local_constants(compiled, context, state, rt_height);
4665         }
4666
4667         /* Force constant reloading for the NP2 fixup (see comment in shader_glsl_select for more info) */
4668         if (compiled->np2fixup_info.super.active)
4669             shader_arb_load_np2fixup_constants(priv, gl_info, state);
4670     }
4671     else
4672     {
4673         if (gl_info->supported[ARB_FRAGMENT_PROGRAM] && !priv->use_arbfp_fixed_func)
4674         {
4675             /* Disable only if we're not using arbfp fixed function fragment
4676              * processing. If this is used, keep GL_FRAGMENT_PROGRAM_ARB
4677              * enabled, and the fixed function pipeline will bind the fixed
4678              * function replacement shader. */
4679             gl_info->gl_ops.gl.p_glDisable(GL_FRAGMENT_PROGRAM_ARB);
4680             checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
4681             priv->current_fprogram_id = 0;
4682         }
4683         priv->fragment_pipe->enable_extension(gl_info, fragment_mode == WINED3D_SHADER_MODE_FFP);
4684     }
4685
4686     if (vertex_mode == WINED3D_SHADER_MODE_SHADER)
4687     {
4688         struct wined3d_shader *vs = state->vertex_shader;
4689         struct arb_vs_compile_args compile_args;
4690         struct arb_vs_compiled_shader *compiled;
4691
4692         TRACE("Using vertex shader %p\n", vs);
4693         find_arb_vs_compile_args(state, context, vs, &compile_args);
4694         compiled = find_arb_vshader(vs, &compile_args);
4695         priv->current_vprogram_id = compiled->prgId;
4696         priv->compiled_vprog = compiled;
4697
4698         /* Bind the vertex program */
4699         GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->current_vprogram_id));
4700         checkGLcall("glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->current_vprogram_id);");
4701
4702         /* Enable OpenGL vertex programs */
4703         gl_info->gl_ops.gl.p_glEnable(GL_VERTEX_PROGRAM_ARB);
4704         checkGLcall("glEnable(GL_VERTEX_PROGRAM_ARB);");
4705         TRACE("(%p) : Bound vertex program %u and enabled GL_VERTEX_PROGRAM_ARB\n", device, priv->current_vprogram_id);
4706         shader_arb_vs_local_constants(compiled, context, state);
4707
4708         if(priv->last_vs_color_unclamp != compiled->need_color_unclamp) {
4709             priv->last_vs_color_unclamp = compiled->need_color_unclamp;
4710
4711             if (gl_info->supported[ARB_COLOR_BUFFER_FLOAT])
4712             {
4713                 GL_EXTCALL(glClampColorARB(GL_CLAMP_VERTEX_COLOR_ARB, !compiled->need_color_unclamp));
4714                 checkGLcall("glClampColorARB");
4715             } else {
4716                 FIXME("vertex color clamp needs to be changed, but extension not supported.\n");
4717             }
4718         }
4719     }
4720     else if (gl_info->supported[ARB_VERTEX_PROGRAM])
4721     {
4722         priv->current_vprogram_id = 0;
4723         gl_info->gl_ops.gl.p_glDisable(GL_VERTEX_PROGRAM_ARB);
4724         checkGLcall("glDisable(GL_VERTEX_PROGRAM_ARB)");
4725     }
4726     priv->vertex_pipe->vp_enable(gl_info, vertex_mode == WINED3D_SHADER_MODE_FFP);
4727 }
4728
4729 /* Context activation is done by the caller. */
4730 static void shader_arb_select_depth_blt(void *shader_priv, const struct wined3d_gl_info *gl_info,
4731         enum tex_types tex_type, const SIZE *ds_mask_size)
4732 {
4733     const float mask[] = {0.0f, 0.0f, (float)ds_mask_size->cx, (float)ds_mask_size->cy};
4734     BOOL masked = ds_mask_size->cx && ds_mask_size->cy;
4735     struct shader_arb_priv *priv = shader_priv;
4736     GLuint *blt_fprogram;
4737
4738     if (!priv->depth_blt_vprogram_id) priv->depth_blt_vprogram_id = create_arb_blt_vertex_program(gl_info);
4739     GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->depth_blt_vprogram_id));
4740     gl_info->gl_ops.gl.p_glEnable(GL_VERTEX_PROGRAM_ARB);
4741
4742     blt_fprogram = masked ? &priv->depth_blt_fprogram_id_masked[tex_type] : &priv->depth_blt_fprogram_id_full[tex_type];
4743     if (!*blt_fprogram) *blt_fprogram = create_arb_blt_fragment_program(gl_info, tex_type, masked);
4744     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, *blt_fprogram));
4745     if (masked) GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, 0, mask));
4746     gl_info->gl_ops.gl.p_glEnable(GL_FRAGMENT_PROGRAM_ARB);
4747 }
4748
4749 /* Context activation is done by the caller. */
4750 static void shader_arb_deselect_depth_blt(void *shader_priv, const struct wined3d_gl_info *gl_info)
4751 {
4752     struct shader_arb_priv *priv = shader_priv;
4753
4754     if (priv->current_vprogram_id) {
4755         GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->current_vprogram_id));
4756         checkGLcall("glBindProgramARB(GL_VERTEX_PROGRAM_ARB, vertexShader->prgId);");
4757
4758         TRACE("Bound vertex program %u and enabled GL_VERTEX_PROGRAM_ARB.\n", priv->current_vprogram_id);
4759     }
4760     else
4761     {
4762         gl_info->gl_ops.gl.p_glDisable(GL_VERTEX_PROGRAM_ARB);
4763         checkGLcall("glDisable(GL_VERTEX_PROGRAM_ARB)");
4764     }
4765
4766     if (priv->current_fprogram_id) {
4767         GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, priv->current_fprogram_id));
4768         checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, pixelShader->prgId);");
4769
4770         TRACE("Bound fragment program %u and enabled GL_FRAGMENT_PROGRAM_ARB.\n", priv->current_fprogram_id);
4771     }
4772     else if(!priv->use_arbfp_fixed_func)
4773     {
4774         gl_info->gl_ops.gl.p_glDisable(GL_FRAGMENT_PROGRAM_ARB);
4775         checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
4776     }
4777 }
4778
4779 static void shader_arb_destroy(struct wined3d_shader *shader)
4780 {
4781     struct wined3d_device *device = shader->device;
4782     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4783
4784     if (shader_is_pshader_version(shader->reg_maps.shader_version.type))
4785     {
4786         struct arb_pshader_private *shader_data = shader->backend_data;
4787         UINT i;
4788
4789         if(!shader_data) return; /* This can happen if a shader was never compiled */
4790
4791         if (shader_data->num_gl_shaders)
4792         {
4793             struct wined3d_context *context = context_acquire(device, NULL);
4794
4795             for (i = 0; i < shader_data->num_gl_shaders; ++i)
4796             {
4797                 GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId));
4798                 checkGLcall("GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId))");
4799             }
4800
4801             context_release(context);
4802         }
4803
4804         HeapFree(GetProcessHeap(), 0, shader_data->gl_shaders);
4805         HeapFree(GetProcessHeap(), 0, shader_data);
4806         shader->backend_data = NULL;
4807     }
4808     else
4809     {
4810         struct arb_vshader_private *shader_data = shader->backend_data;
4811         UINT i;
4812
4813         if(!shader_data) return; /* This can happen if a shader was never compiled */
4814
4815         if (shader_data->num_gl_shaders)
4816         {
4817             struct wined3d_context *context = context_acquire(device, NULL);
4818
4819             for (i = 0; i < shader_data->num_gl_shaders; ++i)
4820             {
4821                 GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId));
4822                 checkGLcall("GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId))");
4823             }
4824
4825             context_release(context);
4826         }
4827
4828         HeapFree(GetProcessHeap(), 0, shader_data->gl_shaders);
4829         HeapFree(GetProcessHeap(), 0, shader_data);
4830         shader->backend_data = NULL;
4831     }
4832 }
4833
4834 static int sig_tree_compare(const void *key, const struct wine_rb_entry *entry)
4835 {
4836     struct ps_signature *e = WINE_RB_ENTRY_VALUE(entry, struct ps_signature, entry);
4837     return compare_sig(key, e->sig);
4838 }
4839
4840 static const struct wine_rb_functions sig_tree_functions =
4841 {
4842     wined3d_rb_alloc,
4843     wined3d_rb_realloc,
4844     wined3d_rb_free,
4845     sig_tree_compare
4846 };
4847
4848 static HRESULT shader_arb_alloc(struct wined3d_device *device, const struct wined3d_vertex_pipe_ops *vertex_pipe,
4849         const struct fragment_pipeline *fragment_pipe)
4850 {
4851     struct shader_arb_priv *priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*priv));
4852     struct fragment_caps fragment_caps;
4853     void *vertex_priv, *fragment_priv;
4854     const struct wined3d_d3d_info *d3d_info = &device->adapter->d3d_info;
4855
4856     if (!(vertex_priv = vertex_pipe->vp_alloc(&arb_program_shader_backend, priv)))
4857     {
4858         ERR("Failed to initialize vertex pipe.\n");
4859         HeapFree(GetProcessHeap(), 0, priv);
4860         return E_FAIL;
4861     }
4862
4863     if (!(fragment_priv = fragment_pipe->alloc_private(&arb_program_shader_backend, priv)))
4864     {
4865         ERR("Failed to initialize fragment pipe.\n");
4866         vertex_pipe->vp_free(device);
4867         HeapFree(GetProcessHeap(), 0, priv);
4868         return E_FAIL;
4869     }
4870
4871     priv->vshader_const_dirty = HeapAlloc(GetProcessHeap(), 0,
4872             sizeof(*priv->vshader_const_dirty) * d3d_info->limits.vs_uniform_count);
4873     if (!priv->vshader_const_dirty)
4874         goto fail;
4875     memset(priv->vshader_const_dirty, 1,
4876            sizeof(*priv->vshader_const_dirty) * d3d_info->limits.vs_uniform_count);
4877
4878     priv->pshader_const_dirty = HeapAlloc(GetProcessHeap(), 0,
4879             sizeof(*priv->pshader_const_dirty) * d3d_info->limits.ps_uniform_count);
4880     if (!priv->pshader_const_dirty)
4881         goto fail;
4882     memset(priv->pshader_const_dirty, 1,
4883             sizeof(*priv->pshader_const_dirty) * d3d_info->limits.ps_uniform_count);
4884
4885     if(wine_rb_init(&priv->signature_tree, &sig_tree_functions) == -1)
4886     {
4887         ERR("RB tree init failed\n");
4888         goto fail;
4889     }
4890
4891     priv->vertex_pipe = vertex_pipe;
4892     priv->fragment_pipe = fragment_pipe;
4893     fragment_pipe->get_caps(&device->adapter->gl_info, &fragment_caps);
4894     priv->ffp_proj_control = fragment_caps.wined3d_caps & WINED3D_FRAGMENT_CAP_PROJ_CONTROL;
4895
4896     device->vertex_priv = vertex_priv;
4897     device->fragment_priv = fragment_priv;
4898     device->shader_priv = priv;
4899
4900     return WINED3D_OK;
4901
4902 fail:
4903     HeapFree(GetProcessHeap(), 0, priv->pshader_const_dirty);
4904     HeapFree(GetProcessHeap(), 0, priv->vshader_const_dirty);
4905     fragment_pipe->free_private(device);
4906     vertex_pipe->vp_free(device);
4907     HeapFree(GetProcessHeap(), 0, priv);
4908     return E_OUTOFMEMORY;
4909 }
4910
4911 static void release_signature(struct wine_rb_entry *entry, void *context)
4912 {
4913     struct ps_signature *sig = WINE_RB_ENTRY_VALUE(entry, struct ps_signature, entry);
4914     int i;
4915     for(i = 0; i < MAX_REG_INPUT; i++)
4916     {
4917         HeapFree(GetProcessHeap(), 0, (char *) sig->sig[i].semantic_name);
4918     }
4919     HeapFree(GetProcessHeap(), 0, sig->sig);
4920     HeapFree(GetProcessHeap(), 0, sig);
4921 }
4922
4923 /* Context activation is done by the caller. */
4924 static void shader_arb_free(struct wined3d_device *device)
4925 {
4926     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4927     struct shader_arb_priv *priv = device->shader_priv;
4928     int i;
4929
4930     if (priv->depth_blt_vprogram_id)
4931         GL_EXTCALL(glDeleteProgramsARB(1, &priv->depth_blt_vprogram_id));
4932
4933     for (i = 0; i < tex_type_count; ++i)
4934     {
4935         if (priv->depth_blt_fprogram_id_full[i])
4936         {
4937             GL_EXTCALL(glDeleteProgramsARB(1, &priv->depth_blt_fprogram_id_full[i]));
4938         }
4939         if (priv->depth_blt_fprogram_id_masked[i])
4940         {
4941             GL_EXTCALL(glDeleteProgramsARB(1, &priv->depth_blt_fprogram_id_masked[i]));
4942         }
4943     }
4944
4945     wine_rb_destroy(&priv->signature_tree, release_signature, NULL);
4946     HeapFree(GetProcessHeap(), 0, priv->pshader_const_dirty);
4947     HeapFree(GetProcessHeap(), 0, priv->vshader_const_dirty);
4948     priv->fragment_pipe->free_private(device);
4949     priv->vertex_pipe->vp_free(device);
4950     HeapFree(GetProcessHeap(), 0, device->shader_priv);
4951 }
4952
4953 static void shader_arb_context_destroyed(void *shader_priv, const struct wined3d_context *context)
4954 {
4955     struct shader_arb_priv *priv = shader_priv;
4956
4957     if (priv->last_context == context)
4958         priv->last_context = NULL;
4959 }
4960
4961 static void shader_arb_get_caps(const struct wined3d_gl_info *gl_info, struct shader_caps *caps)
4962 {
4963     if (gl_info->supported[ARB_VERTEX_PROGRAM])
4964     {
4965         DWORD vs_consts;
4966         UINT vs_version;
4967
4968         /* 96 is the minimum allowed value of MAX_PROGRAM_ENV_PARAMETERS_ARB
4969          * for vertex programs. If the native limit is less than that it's
4970          * not very useful, and e.g. Mesa swrast returns 0, probably to
4971          * indicate it's a software implementation. */
4972         if (gl_info->limits.arb_vs_native_constants < 96)
4973             vs_consts = gl_info->limits.arb_vs_float_constants;
4974         else
4975             vs_consts = min(gl_info->limits.arb_vs_float_constants, gl_info->limits.arb_vs_native_constants);
4976
4977         if (gl_info->supported[NV_VERTEX_PROGRAM3])
4978         {
4979             vs_version = 3;
4980             TRACE("Hardware vertex shader version 3.0 enabled (NV_VERTEX_PROGRAM3)\n");
4981         }
4982         else if (vs_consts >= 256)
4983         {
4984             /* Shader Model 2.0 requires at least 256 vertex shader constants */
4985             vs_version = 2;
4986             TRACE("Hardware vertex shader version 2.0 enabled (ARB_PROGRAM)\n");
4987         }
4988         else
4989         {
4990             vs_version = 1;
4991             TRACE("Hardware vertex shader version 1.1 enabled (ARB_PROGRAM)\n");
4992         }
4993         caps->vs_version = min(wined3d_settings.max_sm_vs, vs_version);
4994         caps->vs_uniform_count = vs_consts;
4995     }
4996     else
4997     {
4998         caps->vs_version = 0;
4999         caps->vs_uniform_count = 0;
5000     }
5001
5002     caps->gs_version = 0;
5003
5004     if (gl_info->supported[ARB_FRAGMENT_PROGRAM])
5005     {
5006         DWORD ps_consts;
5007         UINT ps_version;
5008
5009         /* Similar as above for vertex programs, but the minimum for fragment
5010          * programs is 24. */
5011         if (gl_info->limits.arb_ps_native_constants < 24)
5012             ps_consts = gl_info->limits.arb_ps_float_constants;
5013         else
5014             ps_consts = min(gl_info->limits.arb_ps_float_constants, gl_info->limits.arb_ps_native_constants);
5015
5016         if (gl_info->supported[NV_FRAGMENT_PROGRAM2])
5017         {
5018             ps_version = 3;
5019             TRACE("Hardware pixel shader version 3.0 enabled (NV_FRAGMENT_PROGRAM2)\n");
5020         }
5021         else if (ps_consts >= 32)
5022         {
5023             /* Shader Model 2.0 requires at least 32 pixel shader constants */
5024             ps_version = 2;
5025             TRACE("Hardware pixel shader version 2.0 enabled (ARB_PROGRAM)\n");
5026         }
5027         else
5028         {
5029             ps_version = 1;
5030             TRACE("Hardware pixel shader version 1.4 enabled (ARB_PROGRAM)\n");
5031         }
5032         caps->ps_version = min(wined3d_settings.max_sm_ps, ps_version);
5033         caps->ps_uniform_count = ps_consts;
5034         caps->ps_1x_max_value = 8.0f;
5035     }
5036     else
5037     {
5038         caps->ps_version = 0;
5039         caps->ps_uniform_count = 0;
5040         caps->ps_1x_max_value = 0.0f;
5041     }
5042
5043     caps->wined3d_caps = WINED3D_SHADER_CAP_SRGB_WRITE;
5044     if (use_nv_clip(gl_info))
5045         caps->wined3d_caps |= WINED3D_SHADER_CAP_VS_CLIPPING;
5046 }
5047
5048 static BOOL shader_arb_color_fixup_supported(struct color_fixup_desc fixup)
5049 {
5050     if (TRACE_ON(d3d_shader) && TRACE_ON(d3d))
5051     {
5052         TRACE("Checking support for color_fixup:\n");
5053         dump_color_fixup_desc(fixup);
5054     }
5055
5056     /* We support everything except complex conversions. */
5057     if (!is_complex_fixup(fixup))
5058     {
5059         TRACE("[OK]\n");
5060         return TRUE;
5061     }
5062
5063     TRACE("[FAILED]\n");
5064     return FALSE;
5065 }
5066
5067 static void shader_arb_add_instruction_modifiers(const struct wined3d_shader_instruction *ins) {
5068     DWORD shift;
5069     char write_mask[20], regstr[50];
5070     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
5071     BOOL is_color = FALSE;
5072     const struct wined3d_shader_dst_param *dst;
5073
5074     if (!ins->dst_count) return;
5075
5076     dst = &ins->dst[0];
5077     shift = dst->shift;
5078     if (!shift) return; /* Saturate alone is handled by the instructions */
5079
5080     shader_arb_get_write_mask(ins, dst, write_mask);
5081     shader_arb_get_register_name(ins, &dst->reg, regstr, &is_color);
5082
5083     /* Generate a line that does the output modifier computation
5084      * FIXME: _SAT vs shift? _SAT alone is already handled in the instructions, if this
5085      * maps problems in e.g. _d4_sat modify shader_arb_get_modifier
5086      */
5087     shader_addline(buffer, "MUL%s %s%s, %s, %s;\n", shader_arb_get_modifier(ins),
5088                    regstr, write_mask, regstr, shift_tab[shift]);
5089 }
5090
5091 static const SHADER_HANDLER shader_arb_instruction_handler_table[WINED3DSIH_TABLE_SIZE] =
5092 {
5093     /* WINED3DSIH_ABS                   */ shader_hw_map2gl,
5094     /* WINED3DSIH_ADD                   */ shader_hw_map2gl,
5095     /* WINED3DSIH_AND                   */ NULL,
5096     /* WINED3DSIH_BEM                   */ pshader_hw_bem,
5097     /* WINED3DSIH_BREAK                 */ shader_hw_break,
5098     /* WINED3DSIH_BREAKC                */ shader_hw_breakc,
5099     /* WINED3DSIH_BREAKP                */ NULL,
5100     /* WINED3DSIH_CALL                  */ shader_hw_call,
5101     /* WINED3DSIH_CALLNZ                */ NULL,
5102     /* WINED3DSIH_CMP                   */ pshader_hw_cmp,
5103     /* WINED3DSIH_CND                   */ pshader_hw_cnd,
5104     /* WINED3DSIH_CRS                   */ shader_hw_map2gl,
5105     /* WINED3DSIH_CUT                   */ NULL,
5106     /* WINED3DSIH_DCL                   */ shader_hw_nop,
5107     /* WINED3DSIH_DCL_CONSTANT_BUFFER   */ shader_hw_nop,
5108     /* WINED3DSIH_DCL_INPUT_PRIMITIVE   */ shader_hw_nop,
5109     /* WINED3DSIH_DCL_OUTPUT_TOPOLOGY   */ shader_hw_nop,
5110     /* WINED3DSIH_DCL_VERTICES_OUT      */ shader_hw_nop,
5111     /* WINED3DSIH_DEF                   */ shader_hw_nop,
5112     /* WINED3DSIH_DEFB                  */ shader_hw_nop,
5113     /* WINED3DSIH_DEFI                  */ shader_hw_nop,
5114     /* WINED3DSIH_DIV                   */ NULL,
5115     /* WINED3DSIH_DP2ADD                */ pshader_hw_dp2add,
5116     /* WINED3DSIH_DP3                   */ shader_hw_map2gl,
5117     /* WINED3DSIH_DP4                   */ shader_hw_map2gl,
5118     /* WINED3DSIH_DST                   */ shader_hw_map2gl,
5119     /* WINED3DSIH_DSX                   */ shader_hw_map2gl,
5120     /* WINED3DSIH_DSY                   */ shader_hw_dsy,
5121     /* WINED3DSIH_ELSE                  */ shader_hw_else,
5122     /* WINED3DSIH_EMIT                  */ NULL,
5123     /* WINED3DSIH_ENDIF                 */ shader_hw_endif,
5124     /* WINED3DSIH_ENDLOOP               */ shader_hw_endloop,
5125     /* WINED3DSIH_ENDREP                */ shader_hw_endrep,
5126     /* WINED3DSIH_EQ                    */ NULL,
5127     /* WINED3DSIH_EXP                   */ shader_hw_scalar_op,
5128     /* WINED3DSIH_EXPP                  */ shader_hw_scalar_op,
5129     /* WINED3DSIH_FRC                   */ shader_hw_map2gl,
5130     /* WINED3DSIH_FTOI                  */ NULL,
5131     /* WINED3DSIH_GE                    */ NULL,
5132     /* WINED3DSIH_IADD                  */ NULL,
5133     /* WINED3DSIH_IEQ                   */ NULL,
5134     /* WINED3DSIH_IF                    */ NULL /* Hardcoded into the shader */,
5135     /* WINED3DSIH_IFC                   */ shader_hw_ifc,
5136     /* WINED3DSIH_IGE                   */ NULL,
5137     /* WINED3DSIH_IMUL                  */ NULL,
5138     /* WINED3DSIH_ITOF                  */ NULL,
5139     /* WINED3DSIH_LABEL                 */ shader_hw_label,
5140     /* WINED3DSIH_LD                    */ NULL,
5141     /* WINED3DSIH_LIT                   */ shader_hw_map2gl,
5142     /* WINED3DSIH_LOG                   */ shader_hw_log,
5143     /* WINED3DSIH_LOGP                  */ shader_hw_log,
5144     /* WINED3DSIH_LOOP                  */ shader_hw_loop,
5145     /* WINED3DSIH_LRP                   */ shader_hw_lrp,
5146     /* WINED3DSIH_LT                    */ NULL,
5147     /* WINED3DSIH_M3x2                  */ shader_hw_mnxn,
5148     /* WINED3DSIH_M3x3                  */ shader_hw_mnxn,
5149     /* WINED3DSIH_M3x4                  */ shader_hw_mnxn,
5150     /* WINED3DSIH_M4x3                  */ shader_hw_mnxn,
5151     /* WINED3DSIH_M4x4                  */ shader_hw_mnxn,
5152     /* WINED3DSIH_MAD                   */ shader_hw_map2gl,
5153     /* WINED3DSIH_MAX                   */ shader_hw_map2gl,
5154     /* WINED3DSIH_MIN                   */ shader_hw_map2gl,
5155     /* WINED3DSIH_MOV                   */ shader_hw_mov,
5156     /* WINED3DSIH_MOVA                  */ shader_hw_mov,
5157     /* WINED3DSIH_MOVC                  */ NULL,
5158     /* WINED3DSIH_MUL                   */ shader_hw_map2gl,
5159     /* WINED3DSIH_NOP                   */ shader_hw_nop,
5160     /* WINED3DSIH_NRM                   */ shader_hw_nrm,
5161     /* WINED3DSIH_PHASE                 */ shader_hw_nop,
5162     /* WINED3DSIH_POW                   */ shader_hw_pow,
5163     /* WINED3DSIH_RCP                   */ shader_hw_rcp,
5164     /* WINED3DSIH_REP                   */ shader_hw_rep,
5165     /* WINED3DSIH_RET                   */ shader_hw_ret,
5166     /* WINED3DSIH_ROUND_NI              */ NULL,
5167     /* WINED3DSIH_RSQ                   */ shader_hw_scalar_op,
5168     /* WINED3DSIH_SAMPLE                */ NULL,
5169     /* WINED3DSIH_SAMPLE_GRAD           */ NULL,
5170     /* WINED3DSIH_SAMPLE_LOD            */ NULL,
5171     /* WINED3DSIH_SETP                  */ NULL,
5172     /* WINED3DSIH_SGE                   */ shader_hw_map2gl,
5173     /* WINED3DSIH_SGN                   */ shader_hw_sgn,
5174     /* WINED3DSIH_SINCOS                */ shader_hw_sincos,
5175     /* WINED3DSIH_SLT                   */ shader_hw_map2gl,
5176     /* WINED3DSIH_SQRT                  */ NULL,
5177     /* WINED3DSIH_SUB                   */ shader_hw_map2gl,
5178     /* WINED3DSIH_TEX                   */ pshader_hw_tex,
5179     /* WINED3DSIH_TEXBEM                */ pshader_hw_texbem,
5180     /* WINED3DSIH_TEXBEML               */ pshader_hw_texbem,
5181     /* WINED3DSIH_TEXCOORD              */ pshader_hw_texcoord,
5182     /* WINED3DSIH_TEXDEPTH              */ pshader_hw_texdepth,
5183     /* WINED3DSIH_TEXDP3                */ pshader_hw_texdp3,
5184     /* WINED3DSIH_TEXDP3TEX             */ pshader_hw_texdp3tex,
5185     /* WINED3DSIH_TEXKILL               */ pshader_hw_texkill,
5186     /* WINED3DSIH_TEXLDD                */ shader_hw_texldd,
5187     /* WINED3DSIH_TEXLDL                */ shader_hw_texldl,
5188     /* WINED3DSIH_TEXM3x2DEPTH          */ pshader_hw_texm3x2depth,
5189     /* WINED3DSIH_TEXM3x2PAD            */ pshader_hw_texm3x2pad,
5190     /* WINED3DSIH_TEXM3x2TEX            */ pshader_hw_texm3x2tex,
5191     /* WINED3DSIH_TEXM3x3               */ pshader_hw_texm3x3,
5192     /* WINED3DSIH_TEXM3x3DIFF           */ NULL,
5193     /* WINED3DSIH_TEXM3x3PAD            */ pshader_hw_texm3x3pad,
5194     /* WINED3DSIH_TEXM3x3SPEC           */ pshader_hw_texm3x3spec,
5195     /* WINED3DSIH_TEXM3x3TEX            */ pshader_hw_texm3x3tex,
5196     /* WINED3DSIH_TEXM3x3VSPEC          */ pshader_hw_texm3x3vspec,
5197     /* WINED3DSIH_TEXREG2AR             */ pshader_hw_texreg2ar,
5198     /* WINED3DSIH_TEXREG2GB             */ pshader_hw_texreg2gb,
5199     /* WINED3DSIH_TEXREG2RGB            */ pshader_hw_texreg2rgb,
5200     /* WINED3DSIH_UDIV                  */ NULL,
5201     /* WINED3DSIH_USHR                  */ NULL,
5202     /* WINED3DSIH_UTOF                  */ NULL,
5203     /* WINED3DSIH_XOR                   */ NULL,
5204 };
5205
5206 static BOOL get_bool_const(const struct wined3d_shader_instruction *ins,
5207         const struct wined3d_shader *shader, DWORD idx)
5208 {
5209     const struct wined3d_shader_reg_maps *reg_maps = ins->ctx->reg_maps;
5210     BOOL vshader = shader_is_vshader_version(reg_maps->shader_version.type);
5211     const struct wined3d_shader_lconst *constant;
5212     WORD bools = 0;
5213     WORD flag = (1 << idx);
5214     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
5215
5216     if (reg_maps->local_bool_consts & flag)
5217     {
5218         /* What good is a if(bool) with a hardcoded local constant? I don't know, but handle it */
5219         LIST_FOR_EACH_ENTRY(constant, &shader->constantsB, struct wined3d_shader_lconst, entry)
5220         {
5221             if (constant->idx == idx)
5222             {
5223                 return constant->value[0];
5224             }
5225         }
5226         ERR("Local constant not found\n");
5227         return FALSE;
5228     }
5229     else
5230     {
5231         if(vshader) bools = priv->cur_vs_args->clip.boolclip.bools;
5232         else bools = priv->cur_ps_args->bools;
5233         return bools & flag;
5234     }
5235 }
5236
5237 static void get_loop_control_const(const struct wined3d_shader_instruction *ins,
5238         const struct wined3d_shader *shader, UINT idx, struct wined3d_shader_loop_control *loop_control)
5239 {
5240     const struct wined3d_shader_reg_maps *reg_maps = ins->ctx->reg_maps;
5241     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
5242
5243     /* Integer constants can either be a local constant, or they can be stored in the shader
5244      * type specific compile args. */
5245     if (reg_maps->local_int_consts & (1 << idx))
5246     {
5247         const struct wined3d_shader_lconst *constant;
5248
5249         LIST_FOR_EACH_ENTRY(constant, &shader->constantsI, struct wined3d_shader_lconst, entry)
5250         {
5251             if (constant->idx == idx)
5252             {
5253                 loop_control->count = constant->value[0];
5254                 loop_control->start = constant->value[1];
5255                 /* Step is signed. */
5256                 loop_control->step = (int)constant->value[2];
5257                 return;
5258             }
5259         }
5260         /* If this happens the flag was set incorrectly */
5261         ERR("Local constant not found\n");
5262         loop_control->count = 0;
5263         loop_control->start = 0;
5264         loop_control->step = 0;
5265         return;
5266     }
5267
5268     switch (reg_maps->shader_version.type)
5269     {
5270         case WINED3D_SHADER_TYPE_VERTEX:
5271             /* Count and aL start value are unsigned */
5272             loop_control->count = priv->cur_vs_args->loop_ctrl[idx][0];
5273             loop_control->start = priv->cur_vs_args->loop_ctrl[idx][1];
5274             /* Step is signed. */
5275             loop_control->step = ((char)priv->cur_vs_args->loop_ctrl[idx][2]);
5276             break;
5277
5278         case WINED3D_SHADER_TYPE_PIXEL:
5279             loop_control->count = priv->cur_ps_args->loop_ctrl[idx][0];
5280             loop_control->start = priv->cur_ps_args->loop_ctrl[idx][1];
5281             loop_control->step = ((char)priv->cur_ps_args->loop_ctrl[idx][2]);
5282             break;
5283
5284         default:
5285             FIXME("Unhandled shader type %#x.\n", reg_maps->shader_version.type);
5286             break;
5287     }
5288 }
5289
5290 static void record_instruction(struct list *list, const struct wined3d_shader_instruction *ins)
5291 {
5292     unsigned int i;
5293     struct wined3d_shader_dst_param *dst_param = NULL;
5294     struct wined3d_shader_src_param *src_param = NULL, *rel_addr = NULL;
5295     struct recorded_instruction *rec = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*rec));
5296     if(!rec)
5297     {
5298         ERR("Out of memory\n");
5299         return;
5300     }
5301
5302     rec->ins = *ins;
5303     dst_param = HeapAlloc(GetProcessHeap(), 0, sizeof(*dst_param));
5304     if(!dst_param) goto free;
5305     *dst_param = *ins->dst;
5306     if (ins->dst->reg.idx[0].rel_addr)
5307     {
5308         rel_addr = HeapAlloc(GetProcessHeap(), 0, sizeof(*rel_addr));
5309         if (!rel_addr)
5310             goto free;
5311         *rel_addr = *ins->dst->reg.idx[0].rel_addr;
5312         dst_param->reg.idx[0].rel_addr = rel_addr;
5313     }
5314     rec->ins.dst = dst_param;
5315
5316     src_param = HeapAlloc(GetProcessHeap(), 0, sizeof(*src_param) * ins->src_count);
5317     if (!src_param)
5318         goto free;
5319     for (i = 0; i < ins->src_count; ++i)
5320     {
5321         src_param[i] = ins->src[i];
5322         if (ins->src[i].reg.idx[0].rel_addr)
5323         {
5324             rel_addr = HeapAlloc(GetProcessHeap(), 0, sizeof(*rel_addr));
5325             if (!rel_addr)
5326                 goto free;
5327             *rel_addr = *ins->src[i].reg.idx[0].rel_addr;
5328             src_param[i].reg.idx[0].rel_addr = rel_addr;
5329         }
5330     }
5331     rec->ins.src = src_param;
5332     list_add_tail(list, &rec->entry);
5333     return;
5334
5335 free:
5336     ERR("Out of memory\n");
5337     if(dst_param)
5338     {
5339         HeapFree(GetProcessHeap(), 0, (void *)dst_param->reg.idx[0].rel_addr);
5340         HeapFree(GetProcessHeap(), 0, dst_param);
5341     }
5342     if(src_param)
5343     {
5344         for(i = 0; i < ins->src_count; i++)
5345         {
5346             HeapFree(GetProcessHeap(), 0, (void *)src_param[i].reg.idx[0].rel_addr);
5347         }
5348         HeapFree(GetProcessHeap(), 0, src_param);
5349     }
5350     HeapFree(GetProcessHeap(), 0, rec);
5351 }
5352
5353 static void free_recorded_instruction(struct list *list)
5354 {
5355     struct recorded_instruction *rec_ins, *entry2;
5356     unsigned int i;
5357
5358     LIST_FOR_EACH_ENTRY_SAFE(rec_ins, entry2, list, struct recorded_instruction, entry)
5359     {
5360         list_remove(&rec_ins->entry);
5361         if (rec_ins->ins.dst)
5362         {
5363             HeapFree(GetProcessHeap(), 0, (void *)rec_ins->ins.dst->reg.idx[0].rel_addr);
5364             HeapFree(GetProcessHeap(), 0, (void *)rec_ins->ins.dst);
5365         }
5366         if (rec_ins->ins.src)
5367         {
5368             for (i = 0; i < rec_ins->ins.src_count; ++i)
5369             {
5370                 HeapFree(GetProcessHeap(), 0, (void *)rec_ins->ins.src[i].reg.idx[0].rel_addr);
5371             }
5372             HeapFree(GetProcessHeap(), 0, (void *)rec_ins->ins.src);
5373         }
5374         HeapFree(GetProcessHeap(), 0, rec_ins);
5375     }
5376 }
5377
5378 static void pop_control_frame(const struct wined3d_shader_instruction *ins)
5379 {
5380     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
5381     struct control_frame *control_frame;
5382
5383     if (ins->handler_idx == WINED3DSIH_ENDLOOP || ins->handler_idx == WINED3DSIH_ENDREP)
5384     {
5385         struct list *e = list_head(&priv->control_frames);
5386         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5387         list_remove(&control_frame->entry);
5388         HeapFree(GetProcessHeap(), 0, control_frame);
5389         priv->loop_depth--;
5390     }
5391     else if (ins->handler_idx == WINED3DSIH_ENDIF)
5392     {
5393         /* Non-ifc ENDIFs were already handled previously. */
5394         struct list *e = list_head(&priv->control_frames);
5395         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5396         list_remove(&control_frame->entry);
5397         HeapFree(GetProcessHeap(), 0, control_frame);
5398     }
5399 }
5400
5401 static void shader_arb_handle_instruction(const struct wined3d_shader_instruction *ins) {
5402     SHADER_HANDLER hw_fct;
5403     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
5404     const struct wined3d_shader *shader = ins->ctx->shader;
5405     struct control_frame *control_frame;
5406     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
5407     BOOL bool_const;
5408
5409     if(ins->handler_idx == WINED3DSIH_LOOP || ins->handler_idx == WINED3DSIH_REP)
5410     {
5411         control_frame = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*control_frame));
5412         list_add_head(&priv->control_frames, &control_frame->entry);
5413
5414         if(ins->handler_idx == WINED3DSIH_LOOP) control_frame->type = LOOP;
5415         if(ins->handler_idx == WINED3DSIH_REP) control_frame->type = REP;
5416
5417         if(priv->target_version >= NV2)
5418         {
5419             control_frame->no.loop = priv->num_loops++;
5420             priv->loop_depth++;
5421         }
5422         else
5423         {
5424             /* Don't bother recording when we're in a not used if branch */
5425             if(priv->muted)
5426             {
5427                 return;
5428             }
5429
5430             if(!priv->recording)
5431             {
5432                 list_init(&priv->record);
5433                 priv->recording = TRUE;
5434                 control_frame->outer_loop = TRUE;
5435                 get_loop_control_const(ins, shader, ins->src[0].reg.idx[0].offset, &control_frame->loop_control);
5436                 return; /* Instruction is handled */
5437             }
5438             /* Record this loop in the outer loop's recording */
5439         }
5440     }
5441     else if(ins->handler_idx == WINED3DSIH_ENDLOOP || ins->handler_idx == WINED3DSIH_ENDREP)
5442     {
5443         if(priv->target_version >= NV2)
5444         {
5445             /* Nothing to do. The control frame is popped after the HW instr handler */
5446         }
5447         else
5448         {
5449             struct list *e = list_head(&priv->control_frames);
5450             control_frame = LIST_ENTRY(e, struct control_frame, entry);
5451             list_remove(&control_frame->entry);
5452
5453             if(control_frame->outer_loop)
5454             {
5455                 unsigned int iteration;
5456                 int aL = 0;
5457                 struct list copy;
5458
5459                 /* Turn off recording before playback */
5460                 priv->recording = FALSE;
5461
5462                 /* Move the recorded instructions to a separate list and get them out of the private data
5463                  * structure. If there are nested loops, the shader_arb_handle_instruction below will
5464                  * be recorded again, thus priv->record might be overwritten
5465                  */
5466                 list_init(&copy);
5467                 list_move_tail(&copy, &priv->record);
5468                 list_init(&priv->record);
5469
5470                 if(ins->handler_idx == WINED3DSIH_ENDLOOP)
5471                 {
5472                     shader_addline(buffer, "#unrolling loop: %u iterations, aL=%u, inc %d\n",
5473                                    control_frame->loop_control.count, control_frame->loop_control.start,
5474                                    control_frame->loop_control.step);
5475                     aL = control_frame->loop_control.start;
5476                 }
5477                 else
5478                 {
5479                     shader_addline(buffer, "#unrolling rep: %u iterations\n", control_frame->loop_control.count);
5480                 }
5481
5482                 for (iteration = 0; iteration < control_frame->loop_control.count; ++iteration)
5483                 {
5484                     struct recorded_instruction *rec_ins;
5485                     if(ins->handler_idx == WINED3DSIH_ENDLOOP)
5486                     {
5487                         priv->aL = aL;
5488                         shader_addline(buffer, "#Iteration %u, aL=%d\n", iteration, aL);
5489                     }
5490                     else
5491                     {
5492                         shader_addline(buffer, "#Iteration %u\n", iteration);
5493                     }
5494
5495                     LIST_FOR_EACH_ENTRY(rec_ins, &copy, struct recorded_instruction, entry)
5496                     {
5497                         shader_arb_handle_instruction(&rec_ins->ins);
5498                     }
5499
5500                     if(ins->handler_idx == WINED3DSIH_ENDLOOP)
5501                     {
5502                         aL += control_frame->loop_control.step;
5503                     }
5504                 }
5505                 shader_addline(buffer, "#end loop/rep\n");
5506
5507                 free_recorded_instruction(&copy);
5508                 HeapFree(GetProcessHeap(), 0, control_frame);
5509                 return; /* Instruction is handled */
5510             }
5511             else
5512             {
5513                 /* This is a nested loop. Proceed to the normal recording function */
5514                 HeapFree(GetProcessHeap(), 0, control_frame);
5515             }
5516         }
5517     }
5518
5519     if(priv->recording)
5520     {
5521         record_instruction(&priv->record, ins);
5522         return;
5523     }
5524
5525     /* boolean if */
5526     if(ins->handler_idx == WINED3DSIH_IF)
5527     {
5528         control_frame = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*control_frame));
5529         list_add_head(&priv->control_frames, &control_frame->entry);
5530         control_frame->type = IF;
5531
5532         bool_const = get_bool_const(ins, shader, ins->src[0].reg.idx[0].offset);
5533         if (ins->src[0].modifiers == WINED3DSPSM_NOT)
5534             bool_const = !bool_const;
5535         if (!priv->muted && !bool_const)
5536         {
5537             shader_addline(buffer, "#if(FALSE){\n");
5538             priv->muted = TRUE;
5539             control_frame->muting = TRUE;
5540         }
5541         else shader_addline(buffer, "#if(TRUE) {\n");
5542
5543         return; /* Instruction is handled */
5544     }
5545     else if(ins->handler_idx == WINED3DSIH_IFC)
5546     {
5547         /* IF(bool) and if_cond(a, b) use the same ELSE and ENDIF tokens */
5548         control_frame = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*control_frame));
5549         control_frame->type = IFC;
5550         control_frame->no.ifc = priv->num_ifcs++;
5551         list_add_head(&priv->control_frames, &control_frame->entry);
5552     }
5553     else if(ins->handler_idx == WINED3DSIH_ELSE)
5554     {
5555         struct list *e = list_head(&priv->control_frames);
5556         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5557
5558         if(control_frame->type == IF)
5559         {
5560             shader_addline(buffer, "#} else {\n");
5561             if(!priv->muted && !control_frame->muting)
5562             {
5563                 priv->muted = TRUE;
5564                 control_frame->muting = TRUE;
5565             }
5566             else if(control_frame->muting) priv->muted = FALSE;
5567             return; /* Instruction is handled. */
5568         }
5569         /* In case of an ifc, generate a HW shader instruction */
5570         if (control_frame->type != IFC)
5571             ERR("Control frame does not match.\n");
5572     }
5573     else if(ins->handler_idx == WINED3DSIH_ENDIF)
5574     {
5575         struct list *e = list_head(&priv->control_frames);
5576         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5577
5578         if(control_frame->type == IF)
5579         {
5580             shader_addline(buffer, "#} endif\n");
5581             if(control_frame->muting) priv->muted = FALSE;
5582             list_remove(&control_frame->entry);
5583             HeapFree(GetProcessHeap(), 0, control_frame);
5584             return; /* Instruction is handled */
5585         }
5586         /* In case of an ifc, generate a HW shader instruction */
5587         if (control_frame->type != IFC)
5588             ERR("Control frame does not match.\n");
5589     }
5590
5591     if(priv->muted)
5592     {
5593         pop_control_frame(ins);
5594         return;
5595     }
5596
5597     /* Select handler */
5598     hw_fct = shader_arb_instruction_handler_table[ins->handler_idx];
5599
5600     /* Unhandled opcode */
5601     if (!hw_fct)
5602     {
5603         FIXME("Backend can't handle opcode %#x\n", ins->handler_idx);
5604         return;
5605     }
5606     hw_fct(ins);
5607
5608     pop_control_frame(ins);
5609
5610     shader_arb_add_instruction_modifiers(ins);
5611 }
5612
5613 static BOOL shader_arb_has_ffp_proj_control(void *shader_priv)
5614 {
5615     struct shader_arb_priv *priv = shader_priv;
5616
5617     return priv->ffp_proj_control;
5618 }
5619
5620 const struct wined3d_shader_backend_ops arb_program_shader_backend =
5621 {
5622     shader_arb_handle_instruction,
5623     shader_arb_select,
5624     shader_arb_select_depth_blt,
5625     shader_arb_deselect_depth_blt,
5626     shader_arb_update_float_vertex_constants,
5627     shader_arb_update_float_pixel_constants,
5628     shader_arb_load_constants,
5629     shader_arb_load_np2fixup_constants,
5630     shader_arb_destroy,
5631     shader_arb_alloc,
5632     shader_arb_free,
5633     shader_arb_context_destroyed,
5634     shader_arb_get_caps,
5635     shader_arb_color_fixup_supported,
5636     shader_arb_has_ffp_proj_control,
5637 };
5638
5639 /* ARB_fragment_program fixed function pipeline replacement definitions */
5640 #define ARB_FFP_CONST_TFACTOR           0
5641 #define ARB_FFP_CONST_SPECULAR_ENABLE   ((ARB_FFP_CONST_TFACTOR) + 1)
5642 #define ARB_FFP_CONST_CONSTANT(i)       ((ARB_FFP_CONST_SPECULAR_ENABLE) + 1 + i)
5643 #define ARB_FFP_CONST_BUMPMAT(i)        ((ARB_FFP_CONST_CONSTANT(7)) + 1 + i)
5644 #define ARB_FFP_CONST_LUMINANCE(i)      ((ARB_FFP_CONST_BUMPMAT(7)) + 1 + i)
5645
5646 struct arbfp_ffp_desc
5647 {
5648     struct ffp_frag_desc parent;
5649     GLuint shader;
5650 };
5651
5652 /* Context activation is done by the caller. */
5653 static void arbfp_enable(const struct wined3d_gl_info *gl_info, BOOL enable)
5654 {
5655     if (enable)
5656     {
5657         gl_info->gl_ops.gl.p_glEnable(GL_FRAGMENT_PROGRAM_ARB);
5658         checkGLcall("glEnable(GL_FRAGMENT_PROGRAM_ARB)");
5659     }
5660     else
5661     {
5662         gl_info->gl_ops.gl.p_glDisable(GL_FRAGMENT_PROGRAM_ARB);
5663         checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
5664     }
5665 }
5666
5667 static void *arbfp_alloc(const struct wined3d_shader_backend_ops *shader_backend, void *shader_priv)
5668 {
5669     struct shader_arb_priv *priv;
5670
5671     /* Share private data between the shader backend and the pipeline
5672      * replacement, if both are the arb implementation. This is needed to
5673      * figure out whether ARBfp should be disabled if no pixel shader is bound
5674      * or not. */
5675     if (shader_backend == &arb_program_shader_backend)
5676         priv = shader_priv;
5677     else if (!(priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*priv))))
5678         return NULL;
5679
5680     if (wine_rb_init(&priv->fragment_shaders, &wined3d_ffp_frag_program_rb_functions) == -1)
5681     {
5682         ERR("Failed to initialize rbtree.\n");
5683         if (priv != shader_priv)
5684             HeapFree(GetProcessHeap(), 0, priv);
5685         return NULL;
5686     }
5687     priv->use_arbfp_fixed_func = TRUE;
5688
5689     return priv;
5690 }
5691
5692 /* Context activation is done by the caller. */
5693 static void arbfp_free_ffpshader(struct wine_rb_entry *entry, void *context)
5694 {
5695     const struct wined3d_gl_info *gl_info = context;
5696     struct arbfp_ffp_desc *entry_arb = WINE_RB_ENTRY_VALUE(entry, struct arbfp_ffp_desc, parent.entry);
5697
5698     GL_EXTCALL(glDeleteProgramsARB(1, &entry_arb->shader));
5699     checkGLcall("glDeleteProgramsARB(1, &entry_arb->shader)");
5700     HeapFree(GetProcessHeap(), 0, entry_arb);
5701 }
5702
5703 /* Context activation is done by the caller. */
5704 static void arbfp_free(struct wined3d_device *device)
5705 {
5706     struct shader_arb_priv *priv = device->fragment_priv;
5707
5708     wine_rb_destroy(&priv->fragment_shaders, arbfp_free_ffpshader, &device->adapter->gl_info);
5709     priv->use_arbfp_fixed_func = FALSE;
5710
5711     if (device->shader_backend != &arb_program_shader_backend)
5712     {
5713         HeapFree(GetProcessHeap(), 0, device->fragment_priv);
5714     }
5715 }
5716
5717 static void arbfp_get_caps(const struct wined3d_gl_info *gl_info, struct fragment_caps *caps)
5718 {
5719     caps->wined3d_caps = WINED3D_FRAGMENT_CAP_PROJ_CONTROL
5720             | WINED3D_FRAGMENT_CAP_SRGB_WRITE;
5721     caps->PrimitiveMiscCaps = WINED3DPMISCCAPS_TSSARGTEMP;
5722     caps->TextureOpCaps =  WINED3DTEXOPCAPS_DISABLE                     |
5723                            WINED3DTEXOPCAPS_SELECTARG1                  |
5724                            WINED3DTEXOPCAPS_SELECTARG2                  |
5725                            WINED3DTEXOPCAPS_MODULATE4X                  |
5726                            WINED3DTEXOPCAPS_MODULATE2X                  |
5727                            WINED3DTEXOPCAPS_MODULATE                    |
5728                            WINED3DTEXOPCAPS_ADDSIGNED2X                 |
5729                            WINED3DTEXOPCAPS_ADDSIGNED                   |
5730                            WINED3DTEXOPCAPS_ADD                         |
5731                            WINED3DTEXOPCAPS_SUBTRACT                    |
5732                            WINED3DTEXOPCAPS_ADDSMOOTH                   |
5733                            WINED3DTEXOPCAPS_BLENDCURRENTALPHA           |
5734                            WINED3DTEXOPCAPS_BLENDFACTORALPHA            |
5735                            WINED3DTEXOPCAPS_BLENDTEXTUREALPHA           |
5736                            WINED3DTEXOPCAPS_BLENDDIFFUSEALPHA           |
5737                            WINED3DTEXOPCAPS_BLENDTEXTUREALPHAPM         |
5738                            WINED3DTEXOPCAPS_MODULATEALPHA_ADDCOLOR      |
5739                            WINED3DTEXOPCAPS_MODULATECOLOR_ADDALPHA      |
5740                            WINED3DTEXOPCAPS_MODULATEINVCOLOR_ADDALPHA   |
5741                            WINED3DTEXOPCAPS_MODULATEINVALPHA_ADDCOLOR   |
5742                            WINED3DTEXOPCAPS_DOTPRODUCT3                 |
5743                            WINED3DTEXOPCAPS_MULTIPLYADD                 |
5744                            WINED3DTEXOPCAPS_LERP                        |
5745                            WINED3DTEXOPCAPS_BUMPENVMAP                  |
5746                            WINED3DTEXOPCAPS_BUMPENVMAPLUMINANCE;
5747
5748     /* TODO: Implement WINED3DTEXOPCAPS_PREMODULATE */
5749
5750     caps->MaxTextureBlendStages   = 8;
5751     caps->MaxSimultaneousTextures = min(gl_info->limits.fragment_samplers, 8);
5752 }
5753
5754 static void state_texfactor_arbfp(struct wined3d_context *context,
5755         const struct wined3d_state *state, DWORD state_id)
5756 {
5757     struct wined3d_device *device = context->swapchain->device;
5758     const struct wined3d_gl_info *gl_info = context->gl_info;
5759     float col[4];
5760
5761     /* Don't load the parameter if we're using an arbfp pixel shader,
5762      * otherwise we'll overwrite application provided constants. */
5763     if (device->shader_backend == &arb_program_shader_backend)
5764     {
5765         struct shader_arb_priv *priv;
5766
5767         if (use_ps(state)) return;
5768
5769         priv = device->shader_priv;
5770         priv->pshader_const_dirty[ARB_FFP_CONST_TFACTOR] = 1;
5771         priv->highest_dirty_ps_const = max(priv->highest_dirty_ps_const, ARB_FFP_CONST_TFACTOR + 1);
5772     }
5773
5774     D3DCOLORTOGLFLOAT4(state->render_states[WINED3D_RS_TEXTUREFACTOR], col);
5775     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_TFACTOR, col));
5776     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_TFACTOR, col)");
5777 }
5778
5779 static void state_arb_specularenable(struct wined3d_context *context,
5780         const struct wined3d_state *state, DWORD state_id)
5781 {
5782     struct wined3d_device *device = context->swapchain->device;
5783     const struct wined3d_gl_info *gl_info = context->gl_info;
5784     float col[4];
5785
5786     /* Don't load the parameter if we're using an arbfp pixel shader, otherwise we'll overwrite
5787      * application provided constants
5788      */
5789     if (device->shader_backend == &arb_program_shader_backend)
5790     {
5791         struct shader_arb_priv *priv;
5792
5793         if (use_ps(state)) return;
5794
5795         priv = device->shader_priv;
5796         priv->pshader_const_dirty[ARB_FFP_CONST_SPECULAR_ENABLE] = 1;
5797         priv->highest_dirty_ps_const = max(priv->highest_dirty_ps_const, ARB_FFP_CONST_SPECULAR_ENABLE + 1);
5798     }
5799
5800     if (state->render_states[WINED3D_RS_SPECULARENABLE])
5801     {
5802         /* The specular color has no alpha */
5803         col[0] = 1.0f; col[1] = 1.0f;
5804         col[2] = 1.0f; col[3] = 0.0f;
5805     } else {
5806         col[0] = 0.0f; col[1] = 0.0f;
5807         col[2] = 0.0f; col[3] = 0.0f;
5808     }
5809     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_SPECULAR_ENABLE, col));
5810     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_SPECULAR_ENABLE, col)");
5811 }
5812
5813 static void set_bumpmat_arbfp(struct wined3d_context *context, const struct wined3d_state *state, DWORD state_id)
5814 {
5815     DWORD stage = (state_id - STATE_TEXTURESTAGE(0, 0)) / (WINED3D_HIGHEST_TEXTURE_STATE + 1);
5816     struct wined3d_device *device = context->swapchain->device;
5817     const struct wined3d_gl_info *gl_info = context->gl_info;
5818     float mat[2][2];
5819
5820     if (use_ps(state))
5821     {
5822         /* The pixel shader has to know the bump env matrix. Do a constants
5823          * update. */
5824         if (stage && (state->pixel_shader->reg_maps.bumpmat & (1 << stage)))
5825             context->load_constants = 1;
5826
5827         if(device->shader_backend == &arb_program_shader_backend) {
5828             /* Exit now, don't set the bumpmat below, otherwise we may overwrite pixel shader constants */
5829             return;
5830         }
5831     }
5832     else if (device->shader_backend == &arb_program_shader_backend)
5833     {
5834         struct shader_arb_priv *priv = device->shader_priv;
5835         priv->pshader_const_dirty[ARB_FFP_CONST_BUMPMAT(stage)] = 1;
5836         priv->highest_dirty_ps_const = max(priv->highest_dirty_ps_const, ARB_FFP_CONST_BUMPMAT(stage) + 1);
5837     }
5838
5839     mat[0][0] = *((float *)&state->texture_states[stage][WINED3D_TSS_BUMPENV_MAT00]);
5840     mat[0][1] = *((float *)&state->texture_states[stage][WINED3D_TSS_BUMPENV_MAT01]);
5841     mat[1][0] = *((float *)&state->texture_states[stage][WINED3D_TSS_BUMPENV_MAT10]);
5842     mat[1][1] = *((float *)&state->texture_states[stage][WINED3D_TSS_BUMPENV_MAT11]);
5843
5844     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_BUMPMAT(stage), &mat[0][0]));
5845     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_BUMPMAT(stage), &mat[0][0])");
5846 }
5847
5848 static void tex_bumpenvlum_arbfp(struct wined3d_context *context,
5849         const struct wined3d_state *state, DWORD state_id)
5850 {
5851     DWORD stage = (state_id - STATE_TEXTURESTAGE(0, 0)) / (WINED3D_HIGHEST_TEXTURE_STATE + 1);
5852     struct wined3d_device *device = context->swapchain->device;
5853     const struct wined3d_gl_info *gl_info = context->gl_info;
5854     float param[4];
5855
5856     if (use_ps(state))
5857     {
5858         /* The pixel shader has to know the luminance offset. Do a constants
5859          * update. */
5860         if (stage && (state->pixel_shader->reg_maps.luminanceparams & (1 << stage)))
5861             context->load_constants = 1;
5862
5863         if(device->shader_backend == &arb_program_shader_backend) {
5864             /* Exit now, don't set the bumpmat below, otherwise we may overwrite pixel shader constants */
5865             return;
5866         }
5867     }
5868     else if (device->shader_backend == &arb_program_shader_backend)
5869     {
5870         struct shader_arb_priv *priv = device->shader_priv;
5871         priv->pshader_const_dirty[ARB_FFP_CONST_LUMINANCE(stage)] = 1;
5872         priv->highest_dirty_ps_const = max(priv->highest_dirty_ps_const, ARB_FFP_CONST_LUMINANCE(stage) + 1);
5873     }
5874
5875     param[0] = *((float *)&state->texture_states[stage][WINED3D_TSS_BUMPENV_LSCALE]);
5876     param[1] = *((float *)&state->texture_states[stage][WINED3D_TSS_BUMPENV_LOFFSET]);
5877     param[2] = 0.0f;
5878     param[3] = 0.0f;
5879
5880     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_LUMINANCE(stage), param));
5881     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_LUMINANCE(stage), param)");
5882 }
5883
5884 static const char *get_argreg(struct wined3d_shader_buffer *buffer, DWORD argnum, unsigned int stage, DWORD arg)
5885 {
5886     const char *ret;
5887
5888     if(arg == ARG_UNUSED) return "unused"; /* This is the marker for unused registers */
5889
5890     switch(arg & WINED3DTA_SELECTMASK) {
5891         case WINED3DTA_DIFFUSE:
5892             ret = "fragment.color.primary"; break;
5893
5894         case WINED3DTA_CURRENT:
5895             if (!stage) ret = "fragment.color.primary";
5896             else ret = "ret";
5897             break;
5898
5899         case WINED3DTA_TEXTURE:
5900             switch(stage) {
5901                 case 0: ret = "tex0"; break;
5902                 case 1: ret = "tex1"; break;
5903                 case 2: ret = "tex2"; break;
5904                 case 3: ret = "tex3"; break;
5905                 case 4: ret = "tex4"; break;
5906                 case 5: ret = "tex5"; break;
5907                 case 6: ret = "tex6"; break;
5908                 case 7: ret = "tex7"; break;
5909                 default: ret = "unknown texture";
5910             }
5911             break;
5912
5913         case WINED3DTA_TFACTOR:
5914             ret = "tfactor"; break;
5915
5916         case WINED3DTA_SPECULAR:
5917             ret = "fragment.color.secondary"; break;
5918
5919         case WINED3DTA_TEMP:
5920             ret = "tempreg"; break;
5921
5922         case WINED3DTA_CONSTANT:
5923             FIXME("Implement perstage constants\n");
5924             switch(stage) {
5925                 case 0: ret = "const0"; break;
5926                 case 1: ret = "const1"; break;
5927                 case 2: ret = "const2"; break;
5928                 case 3: ret = "const3"; break;
5929                 case 4: ret = "const4"; break;
5930                 case 5: ret = "const5"; break;
5931                 case 6: ret = "const6"; break;
5932                 case 7: ret = "const7"; break;
5933                 default: ret = "unknown constant";
5934             }
5935             break;
5936
5937         default:
5938             return "unknown";
5939     }
5940
5941     if(arg & WINED3DTA_COMPLEMENT) {
5942         shader_addline(buffer, "SUB arg%u, const.x, %s;\n", argnum, ret);
5943         if(argnum == 0) ret = "arg0";
5944         if(argnum == 1) ret = "arg1";
5945         if(argnum == 2) ret = "arg2";
5946     }
5947     if(arg & WINED3DTA_ALPHAREPLICATE) {
5948         shader_addline(buffer, "MOV arg%u, %s.w;\n", argnum, ret);
5949         if(argnum == 0) ret = "arg0";
5950         if(argnum == 1) ret = "arg1";
5951         if(argnum == 2) ret = "arg2";
5952     }
5953     return ret;
5954 }
5955
5956 static void gen_ffp_instr(struct wined3d_shader_buffer *buffer, unsigned int stage, BOOL color,
5957         BOOL alpha, DWORD dst, DWORD op, DWORD dw_arg0, DWORD dw_arg1, DWORD dw_arg2)
5958 {
5959     const char *dstmask, *dstreg, *arg0, *arg1, *arg2;
5960     unsigned int mul = 1;
5961
5962     if(color && alpha) dstmask = "";
5963     else if(color) dstmask = ".xyz";
5964     else dstmask = ".w";
5965
5966     if(dst == tempreg) dstreg = "tempreg";
5967     else dstreg = "ret";
5968
5969     arg0 = get_argreg(buffer, 0, stage, dw_arg0);
5970     arg1 = get_argreg(buffer, 1, stage, dw_arg1);
5971     arg2 = get_argreg(buffer, 2, stage, dw_arg2);
5972
5973     switch (op)
5974     {
5975         case WINED3D_TOP_DISABLE:
5976             if (!stage)
5977                 shader_addline(buffer, "MOV %s%s, fragment.color.primary;\n", dstreg, dstmask);
5978             break;
5979
5980         case WINED3D_TOP_SELECT_ARG2:
5981             arg1 = arg2;
5982             /* FALLTHROUGH */
5983         case WINED3D_TOP_SELECT_ARG1:
5984             shader_addline(buffer, "MOV %s%s, %s;\n", dstreg, dstmask, arg1);
5985             break;
5986
5987         case WINED3D_TOP_MODULATE_4X:
5988             mul = 2;
5989             /* FALLTHROUGH */
5990         case WINED3D_TOP_MODULATE_2X:
5991             mul *= 2;
5992             /* FALLTHROUGH */
5993         case WINED3D_TOP_MODULATE:
5994             shader_addline(buffer, "MUL %s%s, %s, %s;\n", dstreg, dstmask, arg1, arg2);
5995             break;
5996
5997         case WINED3D_TOP_ADD_SIGNED_2X:
5998             mul = 2;
5999             /* FALLTHROUGH */
6000         case WINED3D_TOP_ADD_SIGNED:
6001             shader_addline(buffer, "SUB arg2, %s, const.w;\n", arg2);
6002             arg2 = "arg2";
6003             /* FALLTHROUGH */
6004         case WINED3D_TOP_ADD:
6005             shader_addline(buffer, "ADD_SAT %s%s, %s, %s;\n", dstreg, dstmask, arg1, arg2);
6006             break;
6007
6008         case WINED3D_TOP_SUBTRACT:
6009             shader_addline(buffer, "SUB_SAT %s%s, %s, %s;\n", dstreg, dstmask, arg1, arg2);
6010             break;
6011
6012         case WINED3D_TOP_ADD_SMOOTH:
6013             shader_addline(buffer, "SUB arg1, const.x, %s;\n", arg1);
6014             shader_addline(buffer, "MAD_SAT %s%s, arg1, %s, %s;\n", dstreg, dstmask, arg2, arg1);
6015             break;
6016
6017         case WINED3D_TOP_BLEND_CURRENT_ALPHA:
6018             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_CURRENT);
6019             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
6020             break;
6021         case WINED3D_TOP_BLEND_FACTOR_ALPHA:
6022             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_TFACTOR);
6023             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
6024             break;
6025         case WINED3D_TOP_BLEND_TEXTURE_ALPHA:
6026             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_TEXTURE);
6027             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
6028             break;
6029         case WINED3D_TOP_BLEND_DIFFUSE_ALPHA:
6030             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_DIFFUSE);
6031             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
6032             break;
6033
6034         case WINED3D_TOP_BLEND_TEXTURE_ALPHA_PM:
6035             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_TEXTURE);
6036             shader_addline(buffer, "SUB arg0.w, const.x, %s.w;\n", arg0);
6037             shader_addline(buffer, "MAD_SAT %s%s, %s, arg0.w, %s;\n", dstreg, dstmask, arg2, arg1);
6038             break;
6039
6040         /* D3DTOP_PREMODULATE ???? */
6041
6042         case WINED3D_TOP_MODULATE_INVALPHA_ADD_COLOR:
6043             shader_addline(buffer, "SUB arg0.w, const.x, %s;\n", arg1);
6044             shader_addline(buffer, "MAD_SAT %s%s, arg0.w, %s, %s;\n", dstreg, dstmask, arg2, arg1);
6045             break;
6046         case WINED3D_TOP_MODULATE_ALPHA_ADD_COLOR:
6047             shader_addline(buffer, "MAD_SAT %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg1, arg2, arg1);
6048             break;
6049         case WINED3D_TOP_MODULATE_INVCOLOR_ADD_ALPHA:
6050             shader_addline(buffer, "SUB arg0, const.x, %s;\n", arg1);
6051             shader_addline(buffer, "MAD_SAT %s%s, arg0, %s, %s.w;\n", dstreg, dstmask, arg2, arg1);
6052             break;
6053         case WINED3D_TOP_MODULATE_COLOR_ADD_ALPHA:
6054             shader_addline(buffer, "MAD_SAT %s%s, %s, %s, %s.w;\n", dstreg, dstmask, arg1, arg2, arg1);
6055             break;
6056
6057         case WINED3D_TOP_DOTPRODUCT3:
6058             mul = 4;
6059             shader_addline(buffer, "SUB arg1, %s, const.w;\n", arg1);
6060             shader_addline(buffer, "SUB arg2, %s, const.w;\n", arg2);
6061             shader_addline(buffer, "DP3_SAT %s%s, arg1, arg2;\n", dstreg, dstmask);
6062             break;
6063
6064         case WINED3D_TOP_MULTIPLY_ADD:
6065             shader_addline(buffer, "MAD_SAT %s%s, %s, %s, %s;\n", dstreg, dstmask, arg1, arg2, arg0);
6066             break;
6067
6068         case WINED3D_TOP_LERP:
6069             /* The msdn is not quite right here */
6070             shader_addline(buffer, "LRP %s%s, %s, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
6071             break;
6072
6073         case WINED3D_TOP_BUMPENVMAP:
6074         case WINED3D_TOP_BUMPENVMAP_LUMINANCE:
6075             /* Those are handled in the first pass of the shader(generation pass 1 and 2) already */
6076             break;
6077
6078         default:
6079             FIXME("Unhandled texture op %08x\n", op);
6080     }
6081
6082     if (mul == 2)
6083         shader_addline(buffer, "MUL_SAT %s%s, %s, const.y;\n", dstreg, dstmask, dstreg);
6084     else if (mul == 4)
6085         shader_addline(buffer, "MUL_SAT %s%s, %s, const.z;\n", dstreg, dstmask, dstreg);
6086 }
6087
6088 static GLuint gen_arbfp_ffp_shader(const struct ffp_frag_settings *settings, const struct wined3d_gl_info *gl_info)
6089 {
6090     unsigned int stage;
6091     struct wined3d_shader_buffer buffer;
6092     BOOL tex_read[MAX_TEXTURES] = {FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE};
6093     BOOL bump_used[MAX_TEXTURES] = {FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE};
6094     BOOL luminance_used[MAX_TEXTURES] = {FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE};
6095     UINT lowest_disabled_stage;
6096     const char *textype;
6097     const char *instr, *sat;
6098     char colorcor_dst[8];
6099     GLuint ret;
6100     DWORD arg0, arg1, arg2;
6101     BOOL tempreg_used = FALSE, tfactor_used = FALSE;
6102     BOOL op_equal;
6103     const char *final_combiner_src = "ret";
6104     GLint pos;
6105
6106     /* Find out which textures are read */
6107     for (stage = 0; stage < MAX_TEXTURES; ++stage)
6108     {
6109         if (settings->op[stage].cop == WINED3D_TOP_DISABLE)
6110             break;
6111         arg0 = settings->op[stage].carg0 & WINED3DTA_SELECTMASK;
6112         arg1 = settings->op[stage].carg1 & WINED3DTA_SELECTMASK;
6113         arg2 = settings->op[stage].carg2 & WINED3DTA_SELECTMASK;
6114         if(arg0 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
6115         if(arg1 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
6116         if(arg2 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
6117
6118         if (settings->op[stage].cop == WINED3D_TOP_BLEND_TEXTURE_ALPHA)
6119             tex_read[stage] = TRUE;
6120         if (settings->op[stage].cop == WINED3D_TOP_BLEND_TEXTURE_ALPHA_PM)
6121             tex_read[stage] = TRUE;
6122         if (settings->op[stage].cop == WINED3D_TOP_BUMPENVMAP)
6123         {
6124             bump_used[stage] = TRUE;
6125             tex_read[stage] = TRUE;
6126         }
6127         if (settings->op[stage].cop == WINED3D_TOP_BUMPENVMAP_LUMINANCE)
6128         {
6129             bump_used[stage] = TRUE;
6130             tex_read[stage] = TRUE;
6131             luminance_used[stage] = TRUE;
6132         }
6133         else if (settings->op[stage].cop == WINED3D_TOP_BLEND_FACTOR_ALPHA)
6134         {
6135             tfactor_used = TRUE;
6136         }
6137
6138         if(arg0 == WINED3DTA_TFACTOR || arg1 == WINED3DTA_TFACTOR || arg2 == WINED3DTA_TFACTOR) {
6139             tfactor_used = TRUE;
6140         }
6141
6142         if(settings->op[stage].dst == tempreg) tempreg_used = TRUE;
6143         if(arg0 == WINED3DTA_TEMP || arg1 == WINED3DTA_TEMP || arg2 == WINED3DTA_TEMP) {
6144             tempreg_used = TRUE;
6145         }
6146
6147         if (settings->op[stage].aop == WINED3D_TOP_DISABLE)
6148             continue;
6149         arg0 = settings->op[stage].aarg0 & WINED3DTA_SELECTMASK;
6150         arg1 = settings->op[stage].aarg1 & WINED3DTA_SELECTMASK;
6151         arg2 = settings->op[stage].aarg2 & WINED3DTA_SELECTMASK;
6152         if(arg0 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
6153         if(arg1 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
6154         if(arg2 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
6155
6156         if(arg0 == WINED3DTA_TEMP || arg1 == WINED3DTA_TEMP || arg2 == WINED3DTA_TEMP) {
6157             tempreg_used = TRUE;
6158         }
6159         if(arg0 == WINED3DTA_TFACTOR || arg1 == WINED3DTA_TFACTOR || arg2 == WINED3DTA_TFACTOR) {
6160             tfactor_used = TRUE;
6161         }
6162     }
6163     lowest_disabled_stage = stage;
6164
6165     /* Shader header */
6166     if (!shader_buffer_init(&buffer))
6167     {
6168         ERR("Failed to initialize shader buffer.\n");
6169         return 0;
6170     }
6171
6172     shader_addline(&buffer, "!!ARBfp1.0\n");
6173
6174     switch(settings->fog) {
6175         case FOG_OFF:                                                         break;
6176         case FOG_LINEAR: shader_addline(&buffer, "OPTION ARB_fog_linear;\n"); break;
6177         case FOG_EXP:    shader_addline(&buffer, "OPTION ARB_fog_exp;\n");    break;
6178         case FOG_EXP2:   shader_addline(&buffer, "OPTION ARB_fog_exp2;\n");   break;
6179         default: FIXME("Unexpected fog setting %d\n", settings->fog);
6180     }
6181
6182     shader_addline(&buffer, "PARAM const = {1, 2, 4, 0.5};\n");
6183     shader_addline(&buffer, "TEMP TMP;\n");
6184     shader_addline(&buffer, "TEMP ret;\n");
6185     if(tempreg_used || settings->sRGB_write) shader_addline(&buffer, "TEMP tempreg;\n");
6186     shader_addline(&buffer, "TEMP arg0;\n");
6187     shader_addline(&buffer, "TEMP arg1;\n");
6188     shader_addline(&buffer, "TEMP arg2;\n");
6189     for(stage = 0; stage < MAX_TEXTURES; stage++) {
6190         if(!tex_read[stage]) continue;
6191         shader_addline(&buffer, "TEMP tex%u;\n", stage);
6192         if(!bump_used[stage]) continue;
6193         shader_addline(&buffer, "PARAM bumpmat%u = program.env[%u];\n", stage, ARB_FFP_CONST_BUMPMAT(stage));
6194         if(!luminance_used[stage]) continue;
6195         shader_addline(&buffer, "PARAM luminance%u = program.env[%u];\n", stage, ARB_FFP_CONST_LUMINANCE(stage));
6196     }
6197     if(tfactor_used) {
6198         shader_addline(&buffer, "PARAM tfactor = program.env[%u];\n", ARB_FFP_CONST_TFACTOR);
6199     }
6200     shader_addline(&buffer, "PARAM specular_enable = program.env[%u];\n", ARB_FFP_CONST_SPECULAR_ENABLE);
6201
6202     if(settings->sRGB_write) {
6203         shader_addline(&buffer, "PARAM srgb_consts1 = {%f, %f, %f, %f};\n",
6204                        srgb_mul_low, srgb_cmp, srgb_pow, srgb_mul_high);
6205         shader_addline(&buffer, "PARAM srgb_consts2 = {%f, %f, %f, %f};\n",
6206                        srgb_sub_high, 0.0, 0.0, 0.0);
6207     }
6208
6209     if (lowest_disabled_stage < 7 && settings->emul_clipplanes)
6210         shader_addline(&buffer, "KIL fragment.texcoord[7];\n");
6211
6212     /* Generate texture sampling instructions) */
6213     for (stage = 0; stage < MAX_TEXTURES && settings->op[stage].cop != WINED3D_TOP_DISABLE; ++stage)
6214     {
6215         if (!tex_read[stage])
6216             continue;
6217
6218         switch(settings->op[stage].tex_type) {
6219             case tex_1d:                    textype = "1D";     break;
6220             case tex_2d:                    textype = "2D";     break;
6221             case tex_3d:                    textype = "3D";     break;
6222             case tex_cube:                  textype = "CUBE";   break;
6223             case tex_rect:                  textype = "RECT";   break;
6224             default: textype = "unexpected_textype";   break;
6225         }
6226
6227         if (settings->op[stage].cop == WINED3D_TOP_BUMPENVMAP
6228                 || settings->op[stage].cop == WINED3D_TOP_BUMPENVMAP_LUMINANCE)
6229             sat = "";
6230         else
6231             sat = "_SAT";
6232
6233         if(settings->op[stage].projected == proj_none) {
6234             instr = "TEX";
6235         } else if(settings->op[stage].projected == proj_count4 ||
6236                   settings->op[stage].projected == proj_count3) {
6237             instr = "TXP";
6238         } else {
6239             FIXME("Unexpected projection mode %d\n", settings->op[stage].projected);
6240             instr = "TXP";
6241         }
6242
6243         if (stage > 0
6244                 && (settings->op[stage - 1].cop == WINED3D_TOP_BUMPENVMAP
6245                 || settings->op[stage - 1].cop == WINED3D_TOP_BUMPENVMAP_LUMINANCE))
6246         {
6247             shader_addline(&buffer, "SWZ arg1, bumpmat%u, x, z, 0, 0;\n", stage - 1);
6248             shader_addline(&buffer, "DP3 ret.x, arg1, tex%u;\n", stage - 1);
6249             shader_addline(&buffer, "SWZ arg1, bumpmat%u, y, w, 0, 0;\n", stage - 1);
6250             shader_addline(&buffer, "DP3 ret.y, arg1, tex%u;\n", stage - 1);
6251
6252             /* with projective textures, texbem only divides the static texture coord, not the displacement,
6253              * so multiply the displacement with the dividing parameter before passing it to TXP
6254              */
6255             if (settings->op[stage].projected != proj_none) {
6256                 if(settings->op[stage].projected == proj_count4) {
6257                     shader_addline(&buffer, "MOV ret.w, fragment.texcoord[%u].w;\n", stage);
6258                     shader_addline(&buffer, "MUL ret.xyz, ret, fragment.texcoord[%u].w, fragment.texcoord[%u];\n", stage, stage);
6259                 } else {
6260                     shader_addline(&buffer, "MOV ret.w, fragment.texcoord[%u].z;\n", stage);
6261                     shader_addline(&buffer, "MAD ret.xyz, ret, fragment.texcoord[%u].z, fragment.texcoord[%u];\n", stage, stage);
6262                 }
6263             } else {
6264                 shader_addline(&buffer, "ADD ret, ret, fragment.texcoord[%u];\n", stage);
6265             }
6266
6267             shader_addline(&buffer, "%s%s tex%u, ret, texture[%u], %s;\n",
6268                     instr, sat, stage, stage, textype);
6269             if (settings->op[stage - 1].cop == WINED3D_TOP_BUMPENVMAP_LUMINANCE)
6270             {
6271                 shader_addline(&buffer, "MAD_SAT ret.x, tex%u.z, luminance%u.x, luminance%u.y;\n",
6272                                stage - 1, stage - 1, stage - 1);
6273                 shader_addline(&buffer, "MUL tex%u, tex%u, ret.x;\n", stage, stage);
6274             }
6275         } else if(settings->op[stage].projected == proj_count3) {
6276             shader_addline(&buffer, "MOV ret, fragment.texcoord[%u];\n", stage);
6277             shader_addline(&buffer, "MOV ret.w, ret.z;\n");
6278             shader_addline(&buffer, "%s%s tex%u, ret, texture[%u], %s;\n",
6279                             instr, sat, stage, stage, textype);
6280         } else {
6281             shader_addline(&buffer, "%s%s tex%u, fragment.texcoord[%u], texture[%u], %s;\n",
6282                             instr, sat, stage, stage, stage, textype);
6283         }
6284
6285         sprintf(colorcor_dst, "tex%u", stage);
6286         gen_color_correction(&buffer, colorcor_dst, WINED3DSP_WRITEMASK_ALL, "const.x", "const.y",
6287                 settings->op[stage].color_fixup);
6288     }
6289
6290     /* Generate the main shader */
6291     for (stage = 0; stage < MAX_TEXTURES; ++stage)
6292     {
6293         if (settings->op[stage].cop == WINED3D_TOP_DISABLE)
6294         {
6295             if (!stage)
6296                 final_combiner_src = "fragment.color.primary";
6297             break;
6298         }
6299
6300         if (settings->op[stage].cop == WINED3D_TOP_SELECT_ARG1
6301                 && settings->op[stage].aop == WINED3D_TOP_SELECT_ARG1)
6302             op_equal = settings->op[stage].carg1 == settings->op[stage].aarg1;
6303         else if (settings->op[stage].cop == WINED3D_TOP_SELECT_ARG1
6304                 && settings->op[stage].aop == WINED3D_TOP_SELECT_ARG2)
6305             op_equal = settings->op[stage].carg1 == settings->op[stage].aarg2;
6306         else if (settings->op[stage].cop == WINED3D_TOP_SELECT_ARG2
6307                 && settings->op[stage].aop == WINED3D_TOP_SELECT_ARG1)
6308             op_equal = settings->op[stage].carg2 == settings->op[stage].aarg1;
6309         else if (settings->op[stage].cop == WINED3D_TOP_SELECT_ARG2
6310                 && settings->op[stage].aop == WINED3D_TOP_SELECT_ARG2)
6311             op_equal = settings->op[stage].carg2 == settings->op[stage].aarg2;
6312         else
6313             op_equal = settings->op[stage].aop   == settings->op[stage].cop
6314                     && settings->op[stage].carg0 == settings->op[stage].aarg0
6315                     && settings->op[stage].carg1 == settings->op[stage].aarg1
6316                     && settings->op[stage].carg2 == settings->op[stage].aarg2;
6317
6318         if (settings->op[stage].aop == WINED3D_TOP_DISABLE)
6319         {
6320             gen_ffp_instr(&buffer, stage, TRUE, FALSE, settings->op[stage].dst,
6321                           settings->op[stage].cop, settings->op[stage].carg0,
6322                           settings->op[stage].carg1, settings->op[stage].carg2);
6323             if (!stage)
6324                 shader_addline(&buffer, "MOV ret.w, fragment.color.primary.w;\n");
6325         }
6326         else if (op_equal)
6327         {
6328             gen_ffp_instr(&buffer, stage, TRUE, TRUE, settings->op[stage].dst,
6329                           settings->op[stage].cop, settings->op[stage].carg0,
6330                           settings->op[stage].carg1, settings->op[stage].carg2);
6331         } else {
6332             gen_ffp_instr(&buffer, stage, TRUE, FALSE, settings->op[stage].dst,
6333                           settings->op[stage].cop, settings->op[stage].carg0,
6334                           settings->op[stage].carg1, settings->op[stage].carg2);
6335             gen_ffp_instr(&buffer, stage, FALSE, TRUE, settings->op[stage].dst,
6336                           settings->op[stage].aop, settings->op[stage].aarg0,
6337                           settings->op[stage].aarg1, settings->op[stage].aarg2);
6338         }
6339     }
6340
6341     if(settings->sRGB_write) {
6342         shader_addline(&buffer, "MAD ret, fragment.color.secondary, specular_enable, %s;\n", final_combiner_src);
6343         arbfp_add_sRGB_correction(&buffer, "ret", "arg0", "arg1", "arg2", "tempreg", FALSE);
6344         shader_addline(&buffer, "MOV result.color, ret;\n");
6345     } else {
6346         shader_addline(&buffer, "MAD result.color, fragment.color.secondary, specular_enable, %s;\n", final_combiner_src);
6347     }
6348
6349     /* Footer */
6350     shader_addline(&buffer, "END\n");
6351
6352     /* Generate the shader */
6353     GL_EXTCALL(glGenProgramsARB(1, &ret));
6354     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, ret));
6355     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
6356             strlen(buffer.buffer), buffer.buffer));
6357     checkGLcall("glProgramStringARB()");
6358
6359     gl_info->gl_ops.gl.p_glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
6360     if (pos != -1)
6361     {
6362         FIXME("Fragment program error at position %d: %s\n\n", pos,
6363               debugstr_a((const char *)gl_info->gl_ops.gl.p_glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
6364         shader_arb_dump_program_source(buffer.buffer);
6365     }
6366     else
6367     {
6368         GLint native;
6369
6370         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
6371         checkGLcall("glGetProgramivARB()");
6372         if (!native) WARN("Program exceeds native resource limits.\n");
6373     }
6374
6375     shader_buffer_free(&buffer);
6376     return ret;
6377 }
6378
6379 static void fragment_prog_arbfp(struct wined3d_context *context, const struct wined3d_state *state, DWORD state_id)
6380 {
6381     const struct wined3d_device *device = context->swapchain->device;
6382     const struct wined3d_gl_info *gl_info = context->gl_info;
6383     struct shader_arb_priv *priv = device->fragment_priv;
6384     BOOL use_pshader = use_ps(state);
6385     struct ffp_frag_settings settings;
6386     const struct arbfp_ffp_desc *desc;
6387     unsigned int i;
6388
6389     TRACE("context %p, state %p, state_id %#x.\n", context, state, state_id);
6390
6391     if (isStateDirty(context, STATE_RENDER(WINED3D_RS_FOGENABLE)))
6392     {
6393         if (!use_pshader && device->shader_backend == &arb_program_shader_backend && context->last_was_pshader)
6394         {
6395             /* Reload fixed function constants since they collide with the
6396              * pixel shader constants. */
6397             for (i = 0; i < MAX_TEXTURES; ++i)
6398             {
6399                 set_bumpmat_arbfp(context, state, STATE_TEXTURESTAGE(i, WINED3D_TSS_BUMPENV_MAT00));
6400             }
6401             state_texfactor_arbfp(context, state, STATE_RENDER(WINED3D_RS_TEXTUREFACTOR));
6402             state_arb_specularenable(context, state, STATE_RENDER(WINED3D_RS_SPECULARENABLE));
6403         }
6404         else if (use_pshader)
6405         {
6406             context->select_shader = 1;
6407         }
6408         return;
6409     }
6410
6411     if (!use_pshader)
6412     {
6413         /* Find or create a shader implementing the fixed function pipeline
6414          * settings, then activate it. */
6415         gen_ffp_frag_op(context, state, &settings, FALSE);
6416         desc = (const struct arbfp_ffp_desc *)find_ffp_frag_shader(&priv->fragment_shaders, &settings);
6417         if(!desc) {
6418             struct arbfp_ffp_desc *new_desc = HeapAlloc(GetProcessHeap(), 0, sizeof(*new_desc));
6419             if (!new_desc)
6420             {
6421                 ERR("Out of memory\n");
6422                 return;
6423             }
6424
6425             new_desc->parent.settings = settings;
6426             new_desc->shader = gen_arbfp_ffp_shader(&settings, gl_info);
6427             add_ffp_frag_shader(&priv->fragment_shaders, &new_desc->parent);
6428             TRACE("Allocated fixed function replacement shader descriptor %p\n", new_desc);
6429             desc = new_desc;
6430         }
6431
6432         /* Now activate the replacement program. GL_FRAGMENT_PROGRAM_ARB is already active (however, note the
6433          * comment above the shader_select call below). If e.g. GLSL is active, the shader_select call will
6434          * deactivate it.
6435          */
6436         GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, desc->shader));
6437         checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, desc->shader)");
6438         priv->current_fprogram_id = desc->shader;
6439
6440         if (device->shader_backend == &arb_program_shader_backend && context->last_was_pshader)
6441         {
6442             /* Reload fixed function constants since they collide with the
6443              * pixel shader constants. */
6444             for (i = 0; i < MAX_TEXTURES; ++i)
6445             {
6446                 set_bumpmat_arbfp(context, state, STATE_TEXTURESTAGE(i, WINED3D_TSS_BUMPENV_MAT00));
6447             }
6448             state_texfactor_arbfp(context, state, STATE_RENDER(WINED3D_RS_TEXTUREFACTOR));
6449             state_arb_specularenable(context, state, STATE_RENDER(WINED3D_RS_SPECULARENABLE));
6450         }
6451         context->last_was_pshader = FALSE;
6452     } else {
6453         context->last_was_pshader = TRUE;
6454     }
6455
6456     context->select_shader = 1;
6457     context->load_constants = 1;
6458 }
6459
6460 /* We can't link the fog states to the fragment state directly since the
6461  * vertex pipeline links them to FOGENABLE. A different linking in different
6462  * pipeline parts can't be expressed in the combined state table, so we need
6463  * to handle that with a forwarding function. The other invisible side effect
6464  * is that changing the fog start and fog end (which links to FOGENABLE in
6465  * vertex) results in the fragment_prog_arbfp function being called because
6466  * FOGENABLE is dirty, which calls this function here. */
6467 static void state_arbfp_fog(struct wined3d_context *context, const struct wined3d_state *state, DWORD state_id)
6468 {
6469     enum fogsource new_source;
6470
6471     TRACE("context %p, state %p, state_id %#x.\n", context, state, state_id);
6472
6473     if (!isStateDirty(context, STATE_PIXELSHADER))
6474         fragment_prog_arbfp(context, state, state_id);
6475
6476     if (!state->render_states[WINED3D_RS_FOGENABLE])
6477         return;
6478
6479     if (state->render_states[WINED3D_RS_FOGTABLEMODE] == WINED3D_FOG_NONE)
6480     {
6481         if (use_vs(state))
6482         {
6483             new_source = FOGSOURCE_VS;
6484         }
6485         else
6486         {
6487             if (state->render_states[WINED3D_RS_FOGVERTEXMODE] == WINED3D_FOG_NONE || context->last_was_rhw)
6488                 new_source = FOGSOURCE_COORD;
6489             else
6490                 new_source = FOGSOURCE_FFP;
6491         }
6492     }
6493     else
6494     {
6495         new_source = FOGSOURCE_FFP;
6496     }
6497
6498     if (new_source != context->fog_source)
6499     {
6500         context->fog_source = new_source;
6501         state_fogstartend(context, state, STATE_RENDER(WINED3D_RS_FOGSTART));
6502     }
6503 }
6504
6505 static void textransform(struct wined3d_context *context, const struct wined3d_state *state, DWORD state_id)
6506 {
6507     if (!isStateDirty(context, STATE_PIXELSHADER))
6508         fragment_prog_arbfp(context, state, state_id);
6509 }
6510
6511 static const struct StateEntryTemplate arbfp_fragmentstate_template[] =
6512 {
6513     {STATE_RENDER(WINED3D_RS_TEXTUREFACTOR),              { STATE_RENDER(WINED3D_RS_TEXTUREFACTOR),             state_texfactor_arbfp   }, WINED3D_GL_EXT_NONE             },
6514     {STATE_TEXTURESTAGE(0, WINED3D_TSS_COLOR_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6515     {STATE_TEXTURESTAGE(0, WINED3D_TSS_COLOR_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6516     {STATE_TEXTURESTAGE(0, WINED3D_TSS_COLOR_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6517     {STATE_TEXTURESTAGE(0, WINED3D_TSS_COLOR_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6518     {STATE_TEXTURESTAGE(0, WINED3D_TSS_ALPHA_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6519     {STATE_TEXTURESTAGE(0, WINED3D_TSS_ALPHA_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6520     {STATE_TEXTURESTAGE(0, WINED3D_TSS_ALPHA_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6521     {STATE_TEXTURESTAGE(0, WINED3D_TSS_ALPHA_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6522     {STATE_TEXTURESTAGE(0, WINED3D_TSS_RESULT_ARG),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6523     {STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6524     {STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6525     {STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6526     {STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6527     {STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6528     {STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6529     {STATE_TEXTURESTAGE(1, WINED3D_TSS_COLOR_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6530     {STATE_TEXTURESTAGE(1, WINED3D_TSS_COLOR_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6531     {STATE_TEXTURESTAGE(1, WINED3D_TSS_COLOR_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6532     {STATE_TEXTURESTAGE(1, WINED3D_TSS_COLOR_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6533     {STATE_TEXTURESTAGE(1, WINED3D_TSS_ALPHA_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6534     {STATE_TEXTURESTAGE(1, WINED3D_TSS_ALPHA_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6535     {STATE_TEXTURESTAGE(1, WINED3D_TSS_ALPHA_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6536     {STATE_TEXTURESTAGE(1, WINED3D_TSS_ALPHA_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6537     {STATE_TEXTURESTAGE(1, WINED3D_TSS_RESULT_ARG),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6538     {STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6539     {STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6540     {STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6541     {STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6542     {STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6543     {STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6544     {STATE_TEXTURESTAGE(2, WINED3D_TSS_COLOR_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6545     {STATE_TEXTURESTAGE(2, WINED3D_TSS_COLOR_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6546     {STATE_TEXTURESTAGE(2, WINED3D_TSS_COLOR_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6547     {STATE_TEXTURESTAGE(2, WINED3D_TSS_COLOR_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6548     {STATE_TEXTURESTAGE(2, WINED3D_TSS_ALPHA_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6549     {STATE_TEXTURESTAGE(2, WINED3D_TSS_ALPHA_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6550     {STATE_TEXTURESTAGE(2, WINED3D_TSS_ALPHA_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6551     {STATE_TEXTURESTAGE(2, WINED3D_TSS_ALPHA_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6552     {STATE_TEXTURESTAGE(2, WINED3D_TSS_RESULT_ARG),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6553     {STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6554     {STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6555     {STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6556     {STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6557     {STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6558     {STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6559     {STATE_TEXTURESTAGE(3, WINED3D_TSS_COLOR_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6560     {STATE_TEXTURESTAGE(3, WINED3D_TSS_COLOR_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6561     {STATE_TEXTURESTAGE(3, WINED3D_TSS_COLOR_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6562     {STATE_TEXTURESTAGE(3, WINED3D_TSS_COLOR_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6563     {STATE_TEXTURESTAGE(3, WINED3D_TSS_ALPHA_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6564     {STATE_TEXTURESTAGE(3, WINED3D_TSS_ALPHA_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6565     {STATE_TEXTURESTAGE(3, WINED3D_TSS_ALPHA_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6566     {STATE_TEXTURESTAGE(3, WINED3D_TSS_ALPHA_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6567     {STATE_TEXTURESTAGE(3, WINED3D_TSS_RESULT_ARG),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6568     {STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6569     {STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6570     {STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6571     {STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6572     {STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6573     {STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6574     {STATE_TEXTURESTAGE(4, WINED3D_TSS_COLOR_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6575     {STATE_TEXTURESTAGE(4, WINED3D_TSS_COLOR_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6576     {STATE_TEXTURESTAGE(4, WINED3D_TSS_COLOR_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6577     {STATE_TEXTURESTAGE(4, WINED3D_TSS_COLOR_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6578     {STATE_TEXTURESTAGE(4, WINED3D_TSS_ALPHA_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6579     {STATE_TEXTURESTAGE(4, WINED3D_TSS_ALPHA_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6580     {STATE_TEXTURESTAGE(4, WINED3D_TSS_ALPHA_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6581     {STATE_TEXTURESTAGE(4, WINED3D_TSS_ALPHA_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6582     {STATE_TEXTURESTAGE(4, WINED3D_TSS_RESULT_ARG),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6583     {STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6584     {STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6585     {STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6586     {STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6587     {STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6588     {STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6589     {STATE_TEXTURESTAGE(5, WINED3D_TSS_COLOR_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6590     {STATE_TEXTURESTAGE(5, WINED3D_TSS_COLOR_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6591     {STATE_TEXTURESTAGE(5, WINED3D_TSS_COLOR_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6592     {STATE_TEXTURESTAGE(5, WINED3D_TSS_COLOR_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6593     {STATE_TEXTURESTAGE(5, WINED3D_TSS_ALPHA_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6594     {STATE_TEXTURESTAGE(5, WINED3D_TSS_ALPHA_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6595     {STATE_TEXTURESTAGE(5, WINED3D_TSS_ALPHA_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6596     {STATE_TEXTURESTAGE(5, WINED3D_TSS_ALPHA_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6597     {STATE_TEXTURESTAGE(5, WINED3D_TSS_RESULT_ARG),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6598     {STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6599     {STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6600     {STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6601     {STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6602     {STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6603     {STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6604     {STATE_TEXTURESTAGE(6, WINED3D_TSS_COLOR_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6605     {STATE_TEXTURESTAGE(6, WINED3D_TSS_COLOR_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6606     {STATE_TEXTURESTAGE(6, WINED3D_TSS_COLOR_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6607     {STATE_TEXTURESTAGE(6, WINED3D_TSS_COLOR_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6608     {STATE_TEXTURESTAGE(6, WINED3D_TSS_ALPHA_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6609     {STATE_TEXTURESTAGE(6, WINED3D_TSS_ALPHA_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6610     {STATE_TEXTURESTAGE(6, WINED3D_TSS_ALPHA_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6611     {STATE_TEXTURESTAGE(6, WINED3D_TSS_ALPHA_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6612     {STATE_TEXTURESTAGE(6, WINED3D_TSS_RESULT_ARG),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6613     {STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6614     {STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6615     {STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6616     {STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6617     {STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6618     {STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6619     {STATE_TEXTURESTAGE(7, WINED3D_TSS_COLOR_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6620     {STATE_TEXTURESTAGE(7, WINED3D_TSS_COLOR_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6621     {STATE_TEXTURESTAGE(7, WINED3D_TSS_COLOR_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6622     {STATE_TEXTURESTAGE(7, WINED3D_TSS_COLOR_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6623     {STATE_TEXTURESTAGE(7, WINED3D_TSS_ALPHA_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6624     {STATE_TEXTURESTAGE(7, WINED3D_TSS_ALPHA_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6625     {STATE_TEXTURESTAGE(7, WINED3D_TSS_ALPHA_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6626     {STATE_TEXTURESTAGE(7, WINED3D_TSS_ALPHA_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6627     {STATE_TEXTURESTAGE(7, WINED3D_TSS_RESULT_ARG),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6628     {STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6629     {STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6630     {STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6631     {STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6632     {STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6633     {STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6634     {STATE_PIXELSHADER,                                   { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6635     {STATE_RENDER(WINED3D_RS_FOGENABLE),                  { STATE_RENDER(WINED3D_RS_FOGENABLE),                 state_arbfp_fog         }, WINED3D_GL_EXT_NONE             },
6636     {STATE_RENDER(WINED3D_RS_FOGTABLEMODE),               { STATE_RENDER(WINED3D_RS_FOGENABLE),                 NULL                    }, WINED3D_GL_EXT_NONE             },
6637     {STATE_RENDER(WINED3D_RS_FOGVERTEXMODE),              { STATE_RENDER(WINED3D_RS_FOGENABLE),                 NULL                    }, WINED3D_GL_EXT_NONE             },
6638     {STATE_RENDER(WINED3D_RS_FOGSTART),                   { STATE_RENDER(WINED3D_RS_FOGSTART),                  state_fogstartend       }, WINED3D_GL_EXT_NONE             },
6639     {STATE_RENDER(WINED3D_RS_FOGEND),                     { STATE_RENDER(WINED3D_RS_FOGSTART),                  NULL                    }, WINED3D_GL_EXT_NONE             },
6640     {STATE_RENDER(WINED3D_RS_SRGBWRITEENABLE),            { STATE_RENDER(WINED3D_RS_SRGBWRITEENABLE),           state_srgbwrite         }, ARB_FRAMEBUFFER_SRGB            },
6641     {STATE_RENDER(WINED3D_RS_SRGBWRITEENABLE),            { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6642     {STATE_RENDER(WINED3D_RS_FOGCOLOR),                   { STATE_RENDER(WINED3D_RS_FOGCOLOR),                  state_fogcolor          }, WINED3D_GL_EXT_NONE             },
6643     {STATE_RENDER(WINED3D_RS_FOGDENSITY),                 { STATE_RENDER(WINED3D_RS_FOGDENSITY),                state_fogdensity        }, WINED3D_GL_EXT_NONE             },
6644     {STATE_TEXTURESTAGE(0,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(0, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6645     {STATE_TEXTURESTAGE(1,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(1, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6646     {STATE_TEXTURESTAGE(2,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(2, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6647     {STATE_TEXTURESTAGE(3,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(3, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6648     {STATE_TEXTURESTAGE(4,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(4, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6649     {STATE_TEXTURESTAGE(5,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(5, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6650     {STATE_TEXTURESTAGE(6,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(6, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6651     {STATE_TEXTURESTAGE(7,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(7, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6652     {STATE_RENDER(WINED3D_RS_SPECULARENABLE),             { STATE_RENDER(WINED3D_RS_SPECULARENABLE),            state_arb_specularenable}, WINED3D_GL_EXT_NONE             },
6653     {0 /* Terminate */,                                   { 0,                                                  0                       }, WINED3D_GL_EXT_NONE             },
6654 };
6655
6656 const struct fragment_pipeline arbfp_fragment_pipeline = {
6657     arbfp_enable,
6658     arbfp_get_caps,
6659     arbfp_alloc,
6660     arbfp_free,
6661     shader_arb_color_fixup_supported,
6662     arbfp_fragmentstate_template,
6663 };
6664
6665 struct arbfp_blit_priv {
6666     GLenum yuy2_rect_shader, yuy2_2d_shader;
6667     GLenum uyvy_rect_shader, uyvy_2d_shader;
6668     GLenum yv12_rect_shader, yv12_2d_shader;
6669     GLenum p8_rect_shader, p8_2d_shader;
6670     GLuint palette_texture;
6671 };
6672
6673 static HRESULT arbfp_blit_alloc(struct wined3d_device *device)
6674 {
6675     device->blit_priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(struct arbfp_blit_priv));
6676     if(!device->blit_priv) {
6677         ERR("Out of memory\n");
6678         return E_OUTOFMEMORY;
6679     }
6680     return WINED3D_OK;
6681 }
6682
6683 /* Context activation is done by the caller. */
6684 static void arbfp_blit_free(struct wined3d_device *device)
6685 {
6686     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
6687     struct arbfp_blit_priv *priv = device->blit_priv;
6688
6689     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yuy2_rect_shader));
6690     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yuy2_2d_shader));
6691     GL_EXTCALL(glDeleteProgramsARB(1, &priv->uyvy_rect_shader));
6692     GL_EXTCALL(glDeleteProgramsARB(1, &priv->uyvy_2d_shader));
6693     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yv12_rect_shader));
6694     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yv12_2d_shader));
6695     GL_EXTCALL(glDeleteProgramsARB(1, &priv->p8_rect_shader));
6696     GL_EXTCALL(glDeleteProgramsARB(1, &priv->p8_2d_shader));
6697     checkGLcall("Delete yuv and p8 programs");
6698
6699     if (priv->palette_texture)
6700         gl_info->gl_ops.gl.p_glDeleteTextures(1, &priv->palette_texture);
6701
6702     HeapFree(GetProcessHeap(), 0, device->blit_priv);
6703     device->blit_priv = NULL;
6704 }
6705
6706 static BOOL gen_planar_yuv_read(struct wined3d_shader_buffer *buffer, enum complex_fixup fixup,
6707         GLenum textype, char *luminance)
6708 {
6709     char chroma;
6710     const char *tex, *texinstr;
6711
6712     if (fixup == COMPLEX_FIXUP_UYVY) {
6713         chroma = 'x';
6714         *luminance = 'w';
6715     } else {
6716         chroma = 'w';
6717         *luminance = 'x';
6718     }
6719     switch(textype) {
6720         case GL_TEXTURE_2D:             tex = "2D";     texinstr = "TXP"; break;
6721         case GL_TEXTURE_RECTANGLE_ARB:  tex = "RECT";   texinstr = "TEX"; break;
6722         default:
6723             /* This is more tricky than just replacing the texture type - we have to navigate
6724              * properly in the texture to find the correct chroma values
6725              */
6726             FIXME("Implement yuv correction for non-2d, non-rect textures\n");
6727             return FALSE;
6728     }
6729
6730     /* First we have to read the chroma values. This means we need at least two pixels(no filtering),
6731      * or 4 pixels(with filtering). To get the unmodified chromas, we have to rid ourselves of the
6732      * filtering when we sample the texture.
6733      *
6734      * These are the rules for reading the chroma:
6735      *
6736      * Even pixel: Cr
6737      * Even pixel: U
6738      * Odd pixel: V
6739      *
6740      * So we have to get the sampling x position in non-normalized coordinates in integers
6741      */
6742     if(textype != GL_TEXTURE_RECTANGLE_ARB) {
6743         shader_addline(buffer, "MUL texcrd.xy, fragment.texcoord[0], size.x;\n");
6744         shader_addline(buffer, "MOV texcrd.w, size.x;\n");
6745     } else {
6746         shader_addline(buffer, "MOV texcrd, fragment.texcoord[0];\n");
6747     }
6748     /* We must not allow filtering between pixel x and x+1, this would mix U and V
6749      * Vertical filtering is ok. However, bear in mind that the pixel center is at
6750      * 0.5, so add 0.5.
6751      */
6752     shader_addline(buffer, "FLR texcrd.x, texcrd.x;\n");
6753     shader_addline(buffer, "ADD texcrd.x, texcrd.x, coef.y;\n");
6754
6755     /* Divide the x coordinate by 0.5 and get the fraction. This gives 0.25 and 0.75 for the
6756      * even and odd pixels respectively
6757      */
6758     shader_addline(buffer, "MUL texcrd2, texcrd, coef.y;\n");
6759     shader_addline(buffer, "FRC texcrd2, texcrd2;\n");
6760
6761     /* Sample Pixel 1 */
6762     shader_addline(buffer, "%s luminance, texcrd, texture[0], %s;\n", texinstr, tex);
6763
6764     /* Put the value into either of the chroma values */
6765     shader_addline(buffer, "SGE temp.x, texcrd2.x, coef.y;\n");
6766     shader_addline(buffer, "MUL chroma.x, luminance.%c, temp.x;\n", chroma);
6767     shader_addline(buffer, "SLT temp.x, texcrd2.x, coef.y;\n");
6768     shader_addline(buffer, "MUL chroma.y, luminance.%c, temp.x;\n", chroma);
6769
6770     /* Sample pixel 2. If we read an even pixel(SLT above returned 1), sample
6771      * the pixel right to the current one. Otherwise, sample the left pixel.
6772      * Bias and scale the SLT result to -1;1 and add it to the texcrd.x.
6773      */
6774     shader_addline(buffer, "MAD temp.x, temp.x, coef.z, -coef.x;\n");
6775     shader_addline(buffer, "ADD texcrd.x, texcrd, temp.x;\n");
6776     shader_addline(buffer, "%s luminance, texcrd, texture[0], %s;\n", texinstr, tex);
6777
6778     /* Put the value into the other chroma */
6779     shader_addline(buffer, "SGE temp.x, texcrd2.x, coef.y;\n");
6780     shader_addline(buffer, "MAD chroma.y, luminance.%c, temp.x, chroma.y;\n", chroma);
6781     shader_addline(buffer, "SLT temp.x, texcrd2.x, coef.y;\n");
6782     shader_addline(buffer, "MAD chroma.x, luminance.%c, temp.x, chroma.x;\n", chroma);
6783
6784     /* TODO: If filtering is enabled, sample a 2nd pair of pixels left or right of
6785      * the current one and lerp the two U and V values
6786      */
6787
6788     /* This gives the correctly filtered luminance value */
6789     shader_addline(buffer, "TEX luminance, fragment.texcoord[0], texture[0], %s;\n", tex);
6790
6791     return TRUE;
6792 }
6793
6794 static BOOL gen_yv12_read(struct wined3d_shader_buffer *buffer, GLenum textype, char *luminance)
6795 {
6796     const char *tex;
6797
6798     switch(textype) {
6799         case GL_TEXTURE_2D:             tex = "2D";     break;
6800         case GL_TEXTURE_RECTANGLE_ARB:  tex = "RECT";   break;
6801         default:
6802             FIXME("Implement yv12 correction for non-2d, non-rect textures\n");
6803             return FALSE;
6804     }
6805
6806     /* YV12 surfaces contain a WxH sized luminance plane, followed by a (W/2)x(H/2)
6807      * V and a (W/2)x(H/2) U plane, each with 8 bit per pixel. So the effective
6808      * bitdepth is 12 bits per pixel. Since the U and V planes have only half the
6809      * pitch of the luminance plane, the packing into the gl texture is a bit
6810      * unfortunate. If the whole texture is interpreted as luminance data it looks
6811      * approximately like this:
6812      *
6813      *        +----------------------------------+----
6814      *        |                                  |
6815      *        |                                  |
6816      *        |                                  |
6817      *        |                                  |
6818      *        |                                  |   2
6819      *        |            LUMINANCE             |   -
6820      *        |                                  |   3
6821      *        |                                  |
6822      *        |                                  |
6823      *        |                                  |
6824      *        |                                  |
6825      *        +----------------+-----------------+----
6826      *        |                |                 |
6827      *        |  U even rows   |  U odd rows     |
6828      *        |                |                 |   1
6829      *        +----------------+------------------   -
6830      *        |                |                 |   3
6831      *        |  V even rows   |  V odd rows     |
6832      *        |                |                 |
6833      *        +----------------+-----------------+----
6834      *        |                |                 |
6835      *        |     0.5        |       0.5       |
6836      *
6837      * So it appears as if there are 4 chroma images, but in fact the odd rows
6838      * in the chroma images are in the same row as the even ones. So its is
6839      * kinda tricky to read
6840      *
6841      * When reading from rectangle textures, keep in mind that the input y coordinates
6842      * go from 0 to d3d_height, whereas the opengl texture height is 1.5 * d3d_height
6843      */
6844     shader_addline(buffer, "PARAM yv12_coef = {%f, %f, %f, %f};\n",
6845             2.0f / 3.0f, 1.0f / 6.0f, (2.0f / 3.0f) + (1.0f / 6.0f), 1.0f / 3.0f);
6846
6847     shader_addline(buffer, "MOV texcrd, fragment.texcoord[0];\n");
6848     /* the chroma planes have only half the width */
6849     shader_addline(buffer, "MUL texcrd.x, texcrd.x, coef.y;\n");
6850
6851     /* The first value is between 2/3 and 5/6th of the texture's height, so scale+bias
6852      * the coordinate. Also read the right side of the image when reading odd lines
6853      *
6854      * Don't forget to clamp the y values in into the range, otherwise we'll get filtering
6855      * bleeding
6856      */
6857     if(textype == GL_TEXTURE_2D) {
6858
6859         shader_addline(buffer, "RCP chroma.w, size.y;\n");
6860
6861         shader_addline(buffer, "MUL texcrd2.y, texcrd.y, size.y;\n");
6862
6863         shader_addline(buffer, "FLR texcrd2.y, texcrd2.y;\n");
6864         shader_addline(buffer, "MAD texcrd.y, texcrd.y, yv12_coef.y, yv12_coef.x;\n");
6865
6866         /* Read odd lines from the right side(add size * 0.5 to the x coordinate */
6867         shader_addline(buffer, "ADD texcrd2.x, texcrd2.y, yv12_coef.y;\n"); /* To avoid 0.5 == 0.5 comparisons */
6868         shader_addline(buffer, "FRC texcrd2.x, texcrd2.x;\n");
6869         shader_addline(buffer, "SGE texcrd2.x, texcrd2.x, coef.y;\n");
6870         shader_addline(buffer, "MAD texcrd.x, texcrd2.x, coef.y, texcrd.x;\n");
6871
6872         /* clamp, keep the half pixel origin in mind */
6873         shader_addline(buffer, "MAD temp.y, coef.y, chroma.w, yv12_coef.x;\n");
6874         shader_addline(buffer, "MAX texcrd.y, temp.y, texcrd.y;\n");
6875         shader_addline(buffer, "MAD temp.y, -coef.y, chroma.w, yv12_coef.z;\n");
6876         shader_addline(buffer, "MIN texcrd.y, temp.y, texcrd.y;\n");
6877     } else {
6878         /* Read from [size - size+size/4] */
6879         shader_addline(buffer, "FLR texcrd.y, texcrd.y;\n");
6880         shader_addline(buffer, "MAD texcrd.y, texcrd.y, coef.w, size.y;\n");
6881
6882         /* Read odd lines from the right side(add size * 0.5 to the x coordinate */
6883         shader_addline(buffer, "ADD texcrd2.x, texcrd.y, yv12_coef.y;\n"); /* To avoid 0.5 == 0.5 comparisons */
6884         shader_addline(buffer, "FRC texcrd2.x, texcrd2.x;\n");
6885         shader_addline(buffer, "SGE texcrd2.x, texcrd2.x, coef.y;\n");
6886         shader_addline(buffer, "MUL texcrd2.x, texcrd2.x, size.x;\n");
6887         shader_addline(buffer, "MAD texcrd.x, texcrd2.x, coef.y, texcrd.x;\n");
6888
6889         /* Make sure to read exactly from the pixel center */
6890         shader_addline(buffer, "FLR texcrd.y, texcrd.y;\n");
6891         shader_addline(buffer, "ADD texcrd.y, texcrd.y, coef.y;\n");
6892
6893         /* Clamp */
6894         shader_addline(buffer, "MAD temp.y, size.y, coef.w, size.y;\n");
6895         shader_addline(buffer, "ADD temp.y, temp.y, -coef.y;\n");
6896         shader_addline(buffer, "MIN texcrd.y, temp.y, texcrd.y;\n");
6897         shader_addline(buffer, "ADD temp.y, size.y, -coef.y;\n");
6898         shader_addline(buffer, "MAX texcrd.y, temp.y, texcrd.y;\n");
6899     }
6900     /* Read the texture, put the result into the output register */
6901     shader_addline(buffer, "TEX temp, texcrd, texture[0], %s;\n", tex);
6902     shader_addline(buffer, "MOV chroma.x, temp.w;\n");
6903
6904     /* The other chroma value is 1/6th of the texture lower, from 5/6th to 6/6th
6905      * No need to clamp because we're just reusing the already clamped value from above
6906      */
6907     if(textype == GL_TEXTURE_2D) {
6908         shader_addline(buffer, "ADD texcrd.y, texcrd.y, yv12_coef.y;\n");
6909     } else {
6910         shader_addline(buffer, "MAD texcrd.y, size.y, coef.w, texcrd.y;\n");
6911     }
6912     shader_addline(buffer, "TEX temp, texcrd, texture[0], %s;\n", tex);
6913     shader_addline(buffer, "MOV chroma.y, temp.w;\n");
6914
6915     /* Sample the luminance value. It is in the top 2/3rd of the texture, so scale the y coordinate.
6916      * Clamp the y coordinate to prevent the chroma values from bleeding into the sampled luminance
6917      * values due to filtering
6918      */
6919     shader_addline(buffer, "MOV texcrd, fragment.texcoord[0];\n");
6920     if(textype == GL_TEXTURE_2D) {
6921         /* Multiply the y coordinate by 2/3 and clamp it */
6922         shader_addline(buffer, "MUL texcrd.y, texcrd.y, yv12_coef.x;\n");
6923         shader_addline(buffer, "MAD temp.y, -coef.y, chroma.w, yv12_coef.x;\n");
6924         shader_addline(buffer, "MIN texcrd.y, temp.y, texcrd.y;\n");
6925         shader_addline(buffer, "TEX luminance, texcrd, texture[0], %s;\n", tex);
6926     } else {
6927         /* Reading from texture_rectangles is pretty straightforward, just use the unmodified
6928          * texture coordinate. It is still a good idea to clamp it though, since the opengl texture
6929          * is bigger
6930          */
6931         shader_addline(buffer, "ADD temp.x, size.y, -coef.y;\n");
6932         shader_addline(buffer, "MIN texcrd.y, texcrd.y, size.x;\n");
6933         shader_addline(buffer, "TEX luminance, texcrd, texture[0], %s;\n", tex);
6934     }
6935     *luminance = 'a';
6936
6937     return TRUE;
6938 }
6939
6940 static GLuint gen_p8_shader(struct arbfp_blit_priv *priv,
6941         const struct wined3d_gl_info *gl_info, GLenum textype)
6942 {
6943     GLenum shader;
6944     struct wined3d_shader_buffer buffer;
6945     GLint pos;
6946
6947     /* Shader header */
6948     if (!shader_buffer_init(&buffer))
6949     {
6950         ERR("Failed to initialize shader buffer.\n");
6951         return 0;
6952     }
6953
6954     GL_EXTCALL(glGenProgramsARB(1, &shader));
6955     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader));
6956     if (!shader)
6957     {
6958         shader_buffer_free(&buffer);
6959         return 0;
6960     }
6961
6962     shader_addline(&buffer, "!!ARBfp1.0\n");
6963     shader_addline(&buffer, "TEMP index;\n");
6964
6965     /* { 255/256, 0.5/255*255/256, 0, 0 } */
6966     shader_addline(&buffer, "PARAM constants = { 0.996, 0.00195, 0, 0 };\n");
6967
6968     /* The alpha-component contains the palette index */
6969     if(textype == GL_TEXTURE_RECTANGLE_ARB)
6970         shader_addline(&buffer, "TXP index, fragment.texcoord[0], texture[0], RECT;\n");
6971     else
6972         shader_addline(&buffer, "TEX index, fragment.texcoord[0], texture[0], 2D;\n");
6973
6974     /* Scale the index by 255/256 and add a bias of '0.5' in order to sample in the middle */
6975     shader_addline(&buffer, "MAD index.a, index.a, constants.x, constants.y;\n");
6976
6977     /* Use the alpha-component as an index in the palette to get the final color */
6978     shader_addline(&buffer, "TEX result.color, index.a, texture[1], 1D;\n");
6979     shader_addline(&buffer, "END\n");
6980
6981     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
6982             strlen(buffer.buffer), buffer.buffer));
6983     checkGLcall("glProgramStringARB()");
6984
6985     gl_info->gl_ops.gl.p_glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
6986     if (pos != -1)
6987     {
6988         FIXME("Fragment program error at position %d: %s\n\n", pos,
6989               debugstr_a((const char *)gl_info->gl_ops.gl.p_glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
6990         shader_arb_dump_program_source(buffer.buffer);
6991     }
6992
6993     if (textype == GL_TEXTURE_RECTANGLE_ARB)
6994         priv->p8_rect_shader = shader;
6995     else
6996         priv->p8_2d_shader = shader;
6997
6998     shader_buffer_free(&buffer);
6999
7000     return shader;
7001 }
7002
7003 /* Context activation is done by the caller. */
7004 static void upload_palette(const struct wined3d_surface *surface, struct wined3d_context *context)
7005 {
7006     BYTE table[256][4];
7007     struct wined3d_device *device = surface->resource.device;
7008     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
7009     struct arbfp_blit_priv *priv = device->blit_priv;
7010     BOOL colorkey = (surface->CKeyFlags & WINEDDSD_CKSRCBLT) != 0;
7011
7012     d3dfmt_p8_init_palette(surface, table, colorkey);
7013
7014     if (gl_info->supported[APPLE_CLIENT_STORAGE])
7015     {
7016         gl_info->gl_ops.gl.p_glPixelStorei(GL_UNPACK_CLIENT_STORAGE_APPLE, GL_FALSE);
7017         checkGLcall("glPixelStorei(GL_UNPACK_CLIENT_STORAGE_APPLE, GL_FALSE)");
7018     }
7019
7020     if (!priv->palette_texture)
7021         gl_info->gl_ops.gl.p_glGenTextures(1, &priv->palette_texture);
7022
7023     GL_EXTCALL(glActiveTextureARB(GL_TEXTURE1));
7024     gl_info->gl_ops.gl.p_glBindTexture(GL_TEXTURE_1D, priv->palette_texture);
7025
7026     gl_info->gl_ops.gl.p_glTexEnvi(GL_TEXTURE_ENV, GL_TEXTURE_ENV_MODE, GL_REPLACE);
7027
7028     gl_info->gl_ops.gl.p_glTexParameteri(GL_TEXTURE_1D, GL_TEXTURE_MIN_FILTER, GL_NEAREST);
7029     /* Make sure we have discrete color levels. */
7030     gl_info->gl_ops.gl.p_glTexParameteri(GL_TEXTURE_1D, GL_TEXTURE_MAG_FILTER, GL_NEAREST);
7031     gl_info->gl_ops.gl.p_glTexParameteri(GL_TEXTURE_1D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
7032     /* Upload the palette */
7033     /* TODO: avoid unneeded uploads in the future by adding some SFLAG_PALETTE_DIRTY mechanism */
7034     gl_info->gl_ops.gl.p_glTexImage1D(GL_TEXTURE_1D, 0, GL_RGBA, 256, 0, GL_RGBA, GL_UNSIGNED_BYTE, table);
7035
7036     if (gl_info->supported[APPLE_CLIENT_STORAGE])
7037     {
7038         gl_info->gl_ops.gl.p_glPixelStorei(GL_UNPACK_CLIENT_STORAGE_APPLE, GL_TRUE);
7039         checkGLcall("glPixelStorei(GL_UNPACK_CLIENT_STORAGE_APPLE, GL_TRUE)");
7040     }
7041
7042     /* Switch back to unit 0 in which the 2D texture will be stored. */
7043     context_active_texture(context, gl_info, 0);
7044 }
7045
7046 /* Context activation is done by the caller. */
7047 static GLuint gen_yuv_shader(struct arbfp_blit_priv *priv, const struct wined3d_gl_info *gl_info,
7048         enum complex_fixup yuv_fixup, GLenum textype)
7049 {
7050     GLenum shader;
7051     struct wined3d_shader_buffer buffer;
7052     char luminance_component;
7053     GLint pos;
7054
7055     /* Shader header */
7056     if (!shader_buffer_init(&buffer))
7057     {
7058         ERR("Failed to initialize shader buffer.\n");
7059         return 0;
7060     }
7061
7062     GL_EXTCALL(glGenProgramsARB(1, &shader));
7063     checkGLcall("GL_EXTCALL(glGenProgramsARB(1, &shader))");
7064     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader));
7065     checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader)");
7066     if (!shader)
7067     {
7068         shader_buffer_free(&buffer);
7069         return 0;
7070     }
7071
7072     /* The YUY2 and UYVY formats contain two pixels packed into a 32 bit macropixel,
7073      * giving effectively 16 bit per pixel. The color consists of a luminance(Y) and
7074      * two chroma(U and V) values. Each macropixel has two luminance values, one for
7075      * each single pixel it contains, and one U and one V value shared between both
7076      * pixels.
7077      *
7078      * The data is loaded into an A8L8 texture. With YUY2, the luminance component
7079      * contains the luminance and alpha the chroma. With UYVY it is vice versa. Thus
7080      * take the format into account when generating the read swizzles
7081      *
7082      * Reading the Y value is straightforward - just sample the texture. The hardware
7083      * takes care of filtering in the horizontal and vertical direction.
7084      *
7085      * Reading the U and V values is harder. We have to avoid filtering horizontally,
7086      * because that would mix the U and V values of one pixel or two adjacent pixels.
7087      * Thus floor the texture coordinate and add 0.5 to get an unfiltered read,
7088      * regardless of the filtering setting. Vertical filtering works automatically
7089      * though - the U and V values of two rows are mixed nicely.
7090      *
7091      * Apart of avoiding filtering issues, the code has to know which value it just
7092      * read, and where it can find the other one. To determine this, it checks if
7093      * it sampled an even or odd pixel, and shifts the 2nd read accordingly.
7094      *
7095      * Handling horizontal filtering of U and V values requires reading a 2nd pair
7096      * of pixels, extracting U and V and mixing them. This is not implemented yet.
7097      *
7098      * An alternative implementation idea is to load the texture as A8R8G8B8 texture,
7099      * with width / 2. This way one read gives all 3 values, finding U and V is easy
7100      * in an unfiltered situation. Finding the luminance on the other hand requires
7101      * finding out if it is an odd or even pixel. The real drawback of this approach
7102      * is filtering. This would have to be emulated completely in the shader, reading
7103      * up two 2 packed pixels in up to 2 rows and interpolating both horizontally and
7104      * vertically. Beyond that it would require adjustments to the texture handling
7105      * code to deal with the width scaling
7106      */
7107     shader_addline(&buffer, "!!ARBfp1.0\n");
7108     shader_addline(&buffer, "TEMP luminance;\n");
7109     shader_addline(&buffer, "TEMP temp;\n");
7110     shader_addline(&buffer, "TEMP chroma;\n");
7111     shader_addline(&buffer, "TEMP texcrd;\n");
7112     shader_addline(&buffer, "TEMP texcrd2;\n");
7113     shader_addline(&buffer, "PARAM coef = {1.0, 0.5, 2.0, 0.25};\n");
7114     shader_addline(&buffer, "PARAM yuv_coef = {1.403, 0.344, 0.714, 1.770};\n");
7115     shader_addline(&buffer, "PARAM size = program.local[0];\n");
7116
7117     switch (yuv_fixup)
7118     {
7119         case COMPLEX_FIXUP_UYVY:
7120         case COMPLEX_FIXUP_YUY2:
7121             if (!gen_planar_yuv_read(&buffer, yuv_fixup, textype, &luminance_component))
7122             {
7123                 shader_buffer_free(&buffer);
7124                 return 0;
7125             }
7126             break;
7127
7128         case COMPLEX_FIXUP_YV12:
7129             if (!gen_yv12_read(&buffer, textype, &luminance_component))
7130             {
7131                 shader_buffer_free(&buffer);
7132                 return 0;
7133             }
7134             break;
7135
7136         default:
7137             FIXME("Unsupported YUV fixup %#x\n", yuv_fixup);
7138             shader_buffer_free(&buffer);
7139             return 0;
7140     }
7141
7142     /* Calculate the final result. Formula is taken from
7143      * http://www.fourcc.org/fccyvrgb.php. Note that the chroma
7144      * ranges from -0.5 to 0.5
7145      */
7146     shader_addline(&buffer, "SUB chroma.xy, chroma, coef.y;\n");
7147
7148     shader_addline(&buffer, "MAD result.color.x, chroma.x, yuv_coef.x, luminance.%c;\n", luminance_component);
7149     shader_addline(&buffer, "MAD temp.x, -chroma.y, yuv_coef.y, luminance.%c;\n", luminance_component);
7150     shader_addline(&buffer, "MAD result.color.y, -chroma.x, yuv_coef.z, temp.x;\n");
7151     shader_addline(&buffer, "MAD result.color.z, chroma.y, yuv_coef.w, luminance.%c;\n", luminance_component);
7152     shader_addline(&buffer, "END\n");
7153
7154     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
7155             strlen(buffer.buffer), buffer.buffer));
7156     checkGLcall("glProgramStringARB()");
7157
7158     gl_info->gl_ops.gl.p_glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
7159     if (pos != -1)
7160     {
7161         FIXME("Fragment program error at position %d: %s\n\n", pos,
7162               debugstr_a((const char *)gl_info->gl_ops.gl.p_glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
7163         shader_arb_dump_program_source(buffer.buffer);
7164     }
7165     else
7166     {
7167         GLint native;
7168
7169         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
7170         checkGLcall("glGetProgramivARB()");
7171         if (!native) WARN("Program exceeds native resource limits.\n");
7172     }
7173
7174     shader_buffer_free(&buffer);
7175
7176     switch (yuv_fixup)
7177     {
7178         case COMPLEX_FIXUP_YUY2:
7179             if (textype == GL_TEXTURE_RECTANGLE_ARB) priv->yuy2_rect_shader = shader;
7180             else priv->yuy2_2d_shader = shader;
7181             break;
7182
7183         case COMPLEX_FIXUP_UYVY:
7184             if (textype == GL_TEXTURE_RECTANGLE_ARB) priv->uyvy_rect_shader = shader;
7185             else priv->uyvy_2d_shader = shader;
7186             break;
7187
7188         case COMPLEX_FIXUP_YV12:
7189             if (textype == GL_TEXTURE_RECTANGLE_ARB) priv->yv12_rect_shader = shader;
7190             else priv->yv12_2d_shader = shader;
7191             break;
7192         default:
7193             ERR("Unsupported complex fixup: %d\n", yuv_fixup);
7194     }
7195
7196     return shader;
7197 }
7198
7199 /* Context activation is done by the caller. */
7200 static HRESULT arbfp_blit_set(void *blit_priv, struct wined3d_context *context, const struct wined3d_surface *surface)
7201 {
7202     GLenum shader;
7203     float size[4] = {(float) surface->pow2Width, (float) surface->pow2Height, 1.0f, 1.0f};
7204     struct arbfp_blit_priv *priv = blit_priv;
7205     enum complex_fixup fixup;
7206     const struct wined3d_gl_info *gl_info = context->gl_info;
7207     GLenum textype;
7208
7209     if (surface->container.type == WINED3D_CONTAINER_TEXTURE)
7210         textype = surface->container.u.texture->target;
7211     else
7212         textype = surface->texture_target;
7213
7214     if (surface->flags & SFLAG_CONVERTED)
7215     {
7216         gl_info->gl_ops.gl.p_glEnable(textype);
7217         checkGLcall("glEnable(textype)");
7218         return WINED3D_OK;
7219     }
7220
7221     if (!is_complex_fixup(surface->resource.format->color_fixup))
7222     {
7223         TRACE("Fixup:\n");
7224         dump_color_fixup_desc(surface->resource.format->color_fixup);
7225         /* Don't bother setting up a shader for unconverted formats */
7226         gl_info->gl_ops.gl.p_glEnable(textype);
7227         checkGLcall("glEnable(textype)");
7228         return WINED3D_OK;
7229     }
7230
7231     fixup = get_complex_fixup(surface->resource.format->color_fixup);
7232
7233     switch(fixup)
7234     {
7235         case COMPLEX_FIXUP_YUY2:
7236             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->yuy2_rect_shader : priv->yuy2_2d_shader;
7237             break;
7238
7239         case COMPLEX_FIXUP_UYVY:
7240             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->uyvy_rect_shader : priv->uyvy_2d_shader;
7241             break;
7242
7243         case COMPLEX_FIXUP_YV12:
7244             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->yv12_rect_shader : priv->yv12_2d_shader;
7245             break;
7246
7247         case COMPLEX_FIXUP_P8:
7248             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->p8_rect_shader : priv->p8_2d_shader;
7249             if (!shader) shader = gen_p8_shader(priv, gl_info, textype);
7250
7251             upload_palette(surface, context);
7252             break;
7253
7254         default:
7255             FIXME("Unsupported complex fixup %#x, not setting a shader\n", fixup);
7256             gl_info->gl_ops.gl.p_glEnable(textype);
7257             checkGLcall("glEnable(textype)");
7258             return E_NOTIMPL;
7259     }
7260
7261     if (!shader) shader = gen_yuv_shader(priv, gl_info, fixup, textype);
7262
7263     gl_info->gl_ops.gl.p_glEnable(GL_FRAGMENT_PROGRAM_ARB);
7264     checkGLcall("glEnable(GL_FRAGMENT_PROGRAM_ARB)");
7265     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader));
7266     checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader)");
7267     GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, 0, size));
7268     checkGLcall("glProgramLocalParameter4fvARB");
7269
7270     return WINED3D_OK;
7271 }
7272
7273 /* Context activation is done by the caller. */
7274 static void arbfp_blit_unset(const struct wined3d_gl_info *gl_info)
7275 {
7276     gl_info->gl_ops.gl.p_glDisable(GL_FRAGMENT_PROGRAM_ARB);
7277     checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
7278     gl_info->gl_ops.gl.p_glDisable(GL_TEXTURE_2D);
7279     checkGLcall("glDisable(GL_TEXTURE_2D)");
7280     if (gl_info->supported[ARB_TEXTURE_CUBE_MAP])
7281     {
7282         gl_info->gl_ops.gl.p_glDisable(GL_TEXTURE_CUBE_MAP_ARB);
7283         checkGLcall("glDisable(GL_TEXTURE_CUBE_MAP_ARB)");
7284     }
7285     if (gl_info->supported[ARB_TEXTURE_RECTANGLE])
7286     {
7287         gl_info->gl_ops.gl.p_glDisable(GL_TEXTURE_RECTANGLE_ARB);
7288         checkGLcall("glDisable(GL_TEXTURE_RECTANGLE_ARB)");
7289     }
7290 }
7291
7292 static BOOL arbfp_blit_supported(const struct wined3d_gl_info *gl_info, enum wined3d_blit_op blit_op,
7293         const RECT *src_rect, DWORD src_usage, enum wined3d_pool src_pool, const struct wined3d_format *src_format,
7294         const RECT *dst_rect, DWORD dst_usage, enum wined3d_pool dst_pool, const struct wined3d_format *dst_format)
7295 {
7296     enum complex_fixup src_fixup;
7297
7298     if (!gl_info->supported[ARB_FRAGMENT_PROGRAM])
7299         return FALSE;
7300
7301     if (blit_op != WINED3D_BLIT_OP_COLOR_BLIT)
7302     {
7303         TRACE("Unsupported blit_op=%d\n", blit_op);
7304         return FALSE;
7305     }
7306
7307     if (src_pool == WINED3D_POOL_SYSTEM_MEM || dst_pool == WINED3D_POOL_SYSTEM_MEM)
7308         return FALSE;
7309
7310     src_fixup = get_complex_fixup(src_format->color_fixup);
7311     if (TRACE_ON(d3d_shader) && TRACE_ON(d3d))
7312     {
7313         TRACE("Checking support for fixup:\n");
7314         dump_color_fixup_desc(src_format->color_fixup);
7315     }
7316
7317     if (!is_identity_fixup(dst_format->color_fixup))
7318     {
7319         TRACE("Destination fixups are not supported\n");
7320         return FALSE;
7321     }
7322
7323     if (is_identity_fixup(src_format->color_fixup))
7324     {
7325         TRACE("[OK]\n");
7326         return TRUE;
7327     }
7328
7329      /* We only support YUV conversions. */
7330     if (!is_complex_fixup(src_format->color_fixup))
7331     {
7332         TRACE("[FAILED]\n");
7333         return FALSE;
7334     }
7335
7336     switch(src_fixup)
7337     {
7338         case COMPLEX_FIXUP_YUY2:
7339         case COMPLEX_FIXUP_UYVY:
7340         case COMPLEX_FIXUP_YV12:
7341         case COMPLEX_FIXUP_P8:
7342             TRACE("[OK]\n");
7343             return TRUE;
7344
7345         default:
7346             FIXME("Unsupported YUV fixup %#x\n", src_fixup);
7347             TRACE("[FAILED]\n");
7348             return FALSE;
7349     }
7350 }
7351
7352 HRESULT arbfp_blit_surface(struct wined3d_device *device, DWORD filter,
7353         struct wined3d_surface *src_surface, const RECT *src_rect_in,
7354         struct wined3d_surface *dst_surface, const RECT *dst_rect_in)
7355 {
7356     struct wined3d_context *context;
7357     RECT src_rect = *src_rect_in;
7358     RECT dst_rect = *dst_rect_in;
7359
7360     /* Now load the surface */
7361     if (wined3d_settings.offscreen_rendering_mode != ORM_FBO
7362             && (src_surface->flags & (SFLAG_INTEXTURE | SFLAG_INDRAWABLE)) == SFLAG_INDRAWABLE)
7363     {
7364         /* Without FBO blits transferring from the drawable to the texture is
7365          * expensive, because we have to flip the data in sysmem. Since we can
7366          * flip in the blitter, we don't actually need that flip anyway. So we
7367          * use the surface's texture as scratch texture, and flip the source
7368          * rectangle instead. */
7369         surface_load_fb_texture(src_surface, FALSE);
7370
7371         src_rect.top = src_surface->resource.height - src_rect.top;
7372         src_rect.bottom = src_surface->resource.height - src_rect.bottom;
7373     }
7374     else
7375         surface_internal_preload(src_surface, SRGB_RGB);
7376
7377     /* Activate the destination context, set it up for blitting */
7378     context = context_acquire(device, dst_surface);
7379     context_apply_blit_state(context, device);
7380
7381     if (!surface_is_offscreen(dst_surface))
7382         surface_translate_drawable_coords(dst_surface, context->win_handle, &dst_rect);
7383
7384     arbfp_blit_set(device->blit_priv, context, src_surface);
7385
7386     /* Draw a textured quad */
7387     draw_textured_quad(src_surface, context, &src_rect, &dst_rect, filter);
7388
7389     /* Leave the opengl state valid for blitting */
7390     arbfp_blit_unset(context->gl_info);
7391
7392     if (wined3d_settings.strict_draw_ordering
7393             || (dst_surface->container.type == WINED3D_CONTAINER_SWAPCHAIN
7394             && (dst_surface->container.u.swapchain->front_buffer == dst_surface)))
7395         context->gl_info->gl_ops.gl.p_glFlush(); /* Flush to ensure ordering across contexts. */
7396
7397     context_release(context);
7398
7399     surface_modify_location(dst_surface, dst_surface->draw_binding, TRUE);
7400     return WINED3D_OK;
7401 }
7402
7403 /* Do not call while under the GL lock. */
7404 static HRESULT arbfp_blit_color_fill(struct wined3d_device *device, struct wined3d_surface *dst_surface,
7405         const RECT *dst_rect, const struct wined3d_color *color)
7406 {
7407     FIXME("Color filling not implemented by arbfp_blit\n");
7408     return WINED3DERR_INVALIDCALL;
7409 }
7410
7411 /* Do not call while under the GL lock. */
7412 static HRESULT arbfp_blit_depth_fill(struct wined3d_device *device,
7413         struct wined3d_surface *surface, const RECT *rect, float depth)
7414 {
7415     FIXME("Depth filling not implemented by arbfp_blit.\n");
7416     return WINED3DERR_INVALIDCALL;
7417 }
7418
7419 const struct blit_shader arbfp_blit = {
7420     arbfp_blit_alloc,
7421     arbfp_blit_free,
7422     arbfp_blit_set,
7423     arbfp_blit_unset,
7424     arbfp_blit_supported,
7425     arbfp_blit_color_fill,
7426     arbfp_blit_depth_fill,
7427 };