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

   1 /*
   2  * Pixel and vertex shaders implementation using ARB_vertex_program
   3  * and ARB_fragment_program GL extensions.
   4  *
   5  * Copyright 2002-2003 Jason Edmeades
   6  * Copyright 2002-2003 Raphael Junqueira
   7  * Copyright 2004 Christian Costa
   8  * Copyright 2005 Oliver Stieber
   9  * Copyright 2006 Ivan Gyurdiev
  10  * Copyright 2006 Jason Green
  11  * Copyright 2006 Henri Verbeet
  12  * Copyright 2007-2008 Stefan Dösinger for CodeWeavers
  13  * Copyright 2009 Henri Verbeet for CodeWeavers
  14  *
  15  * This library is free software; you can redistribute it and/or
  16  * modify it under the terms of the GNU Lesser General Public
  17  * License as published by the Free Software Foundation; either
  18  * version 2.1 of the License, or (at your option) any later version.
  19  *
  20  * This library is distributed in the hope that it will be useful,
  21  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  22  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  23  * Lesser General Public License for more details.
  24  *
  25  * You should have received a copy of the GNU Lesser General Public
  26  * License along with this library; if not, write to the Free Software
  27  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301, USA
  28  */
  29
  30 #include "config.h"
  31 #include "wine/port.h"
  32
  33 #include <math.h>
  34 #include <stdio.h>
  35
  36 #include "wined3d_private.h"
  37
  38 WINE_DEFAULT_DEBUG_CHANNEL(d3d_shader);
  39 WINE_DECLARE_DEBUG_CHANNEL(d3d_constants);
  40 WINE_DECLARE_DEBUG_CHANNEL(d3d);
  41
  42 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
 331 /* GL locking for state handlers is done by the caller. */
 332 static BOOL need_rel_addr_const(const struct arb_vshader_private *shader_data,
 333         const struct wined3d_shader_reg_maps *reg_maps, const struct wined3d_gl_info *gl_info)
 334 {
 335     if (shader_data->rel_offset) return TRUE;
 336     if (!reg_maps->usesmova) return FALSE;
 337     return !gl_info->supported[NV_VERTEX_PROGRAM2_OPTION];
 338 }
 339
 340 /* Returns TRUE if result.clip from GL_NV_vertex_program2 should be used and FALSE otherwise */
 341 static inline BOOL use_nv_clip(const struct wined3d_gl_info *gl_info)
 342 {
 343     return gl_info->supported[NV_VERTEX_PROGRAM2_OPTION]
 344             && !(gl_info->quirks & WINED3D_QUIRK_NV_CLIP_BROKEN);
 345 }
 346
 347 static BOOL need_helper_const(const struct arb_vshader_private *shader_data,
 348         const struct wined3d_shader_reg_maps *reg_maps, const struct wined3d_gl_info *gl_info)
 349 {
 350     if (need_rel_addr_const(shader_data, reg_maps, gl_info)) return TRUE;
 351     if (!gl_info->supported[NV_VERTEX_PROGRAM]) return TRUE; /* Need to init colors. */
 352     if (gl_info->quirks & WINED3D_QUIRK_ARB_VS_OFFSET_LIMIT) return TRUE; /* Load the immval offset. */
 353     if (gl_info->quirks & WINED3D_QUIRK_SET_TEXCOORD_W) return TRUE; /* Have to init texcoords. */
 354     if (!use_nv_clip(gl_info)) return TRUE; /* Init the clip texcoord */
 355     if (reg_maps->usesnrm) return TRUE; /* 0.0 */
 356     if (reg_maps->usespow) return TRUE; /* EPS, 0.0 and 1.0 */
 357     if (reg_maps->fog) return TRUE; /* Clamping fog coord, 0.0 and 1.0 */
 358     return FALSE;
 359 }
 360
 361 static unsigned int reserved_vs_const(const struct arb_vshader_private *shader_data,
 362         const struct wined3d_shader_reg_maps *reg_maps, const struct wined3d_gl_info *gl_info)
 363 {
 364     unsigned int ret = 1;
 365     /* We use one PARAM for the pos fixup, and in some cases one to load
 366      * some immediate values into the shader. */
 367     if (need_helper_const(shader_data, reg_maps, gl_info)) ++ret;
 368     if (need_rel_addr_const(shader_data, reg_maps, gl_info)) ++ret;
 369     return ret;
 370 }
 371
 372 /* Loads floating point constants into the currently set ARB_vertex/fragment_program.
 373  * When constant_list == NULL, it will load all the constants.
 374  *
 375  * @target_type should be either GL_VERTEX_PROGRAM_ARB (for vertex shaders)
 376  *  or GL_FRAGMENT_PROGRAM_ARB (for pixel shaders)
 377  */
 378 /* GL locking is done by the caller */
 379 static unsigned int shader_arb_load_constantsF(const struct wined3d_shader *shader,
 380         const struct wined3d_gl_info *gl_info, GLuint target_type, unsigned int max_constants,
 381         const float *constants, char *dirty_consts)
 382 {
 383     struct wined3d_shader_lconst *lconst;
 384     DWORD i, j;
 385     unsigned int ret;
 386
 387     if (TRACE_ON(d3d_constants))
 388     {
 389         for(i = 0; i < max_constants; i++) {
 390             if(!dirty_consts[i]) continue;
 391             TRACE_(d3d_constants)("Loading constants %i: %f, %f, %f, %f\n", i,
 392                         constants[i * 4 + 0], constants[i * 4 + 1],
 393                         constants[i * 4 + 2], constants[i * 4 + 3]);
 394         }
 395     }
 396
 397     i = 0;
 398
 399     /* In 1.X pixel shaders constants are implicitly clamped in the range [-1;1] */
 400     if (target_type == GL_FRAGMENT_PROGRAM_ARB && shader->reg_maps.shader_version.major == 1)
 401     {
 402         float lcl_const[4];
 403         /* ps 1.x supports only 8 constants, clamp only those. When switching between 1.x and higher
 404          * shaders, the first 8 constants are marked dirty for reload
 405          */
 406         for(; i < min(8, max_constants); i++) {
 407             if(!dirty_consts[i]) continue;
 408             dirty_consts[i] = 0;
 409
 410             j = 4 * i;
 411             if (constants[j + 0] > 1.0f) lcl_const[0] = 1.0f;
 412             else if (constants[j + 0] < -1.0f) lcl_const[0] = -1.0f;
 413             else lcl_const[0] = constants[j + 0];
 414
 415             if (constants[j + 1] > 1.0f) lcl_const[1] = 1.0f;
 416             else if (constants[j + 1] < -1.0f) lcl_const[1] = -1.0f;
 417             else lcl_const[1] = constants[j + 1];
 418
 419             if (constants[j + 2] > 1.0f) lcl_const[2] = 1.0f;
 420             else if (constants[j + 2] < -1.0f) lcl_const[2] = -1.0f;
 421             else lcl_const[2] = constants[j + 2];
 422
 423             if (constants[j + 3] > 1.0f) lcl_const[3] = 1.0f;
 424             else if (constants[j + 3] < -1.0f) lcl_const[3] = -1.0f;
 425             else lcl_const[3] = constants[j + 3];
 426
 427             GL_EXTCALL(glProgramEnvParameter4fvARB(target_type, i, lcl_const));
 428         }
 429
 430         /* If further constants are dirty, reload them without clamping.
 431          *
 432          * The alternative is not to touch them, but then we cannot reset the dirty constant count
 433          * to zero. That's bad for apps that only use PS 1.x shaders, because in that case the code
 434          * above would always re-check the first 8 constants since max_constant remains at the init
 435          * value
 436          */
 437     }
 438
 439     if (gl_info->supported[EXT_GPU_PROGRAM_PARAMETERS])
 440     {
 441         /* TODO: Benchmark if we're better of with finding the dirty constants ourselves,
 442          * or just reloading *all* constants at once
 443          *
 444         GL_EXTCALL(glProgramEnvParameters4fvEXT(target_type, i, max_constants, constants + (i * 4)));
 445          */
 446         for(; i < max_constants; i++) {
 447             if(!dirty_consts[i]) continue;
 448
 449             /* Find the next block of dirty constants */
 450             dirty_consts[i] = 0;
 451             j = i;
 452             for(i++; (i < max_constants) && dirty_consts[i]; i++) {
 453                 dirty_consts[i] = 0;
 454             }
 455
 456             GL_EXTCALL(glProgramEnvParameters4fvEXT(target_type, j, i - j, constants + (j * 4)));
 457         }
 458     } else {
 459         for(; i < max_constants; i++) {
 460             if(dirty_consts[i]) {
 461                 dirty_consts[i] = 0;
 462                 GL_EXTCALL(glProgramEnvParameter4fvARB(target_type, i, constants + (i * 4)));
 463             }
 464         }
 465     }
 466     checkGLcall("glProgramEnvParameter4fvARB()");
 467
 468     /* Load immediate constants */
 469     if (shader->load_local_constsF)
 470     {
 471         if (TRACE_ON(d3d_shader))
 472         {
 473             LIST_FOR_EACH_ENTRY(lconst, &shader->constantsF, struct wined3d_shader_lconst, entry)
 474             {
 475                 GLfloat* values = (GLfloat*)lconst->value;
 476                 TRACE_(d3d_constants)("Loading local constants %i: %f, %f, %f, %f\n", lconst->idx,
 477                         values[0], values[1], values[2], values[3]);
 478             }
 479         }
 480         /* Immediate constants are clamped for 1.X shaders at loading times */
 481         ret = 0;
 482         LIST_FOR_EACH_ENTRY(lconst, &shader->constantsF, struct wined3d_shader_lconst, entry)
 483         {
 484             dirty_consts[lconst->idx] = 1; /* Dirtify so the non-immediate constant overwrites it next time */
 485             ret = max(ret, lconst->idx + 1);
 486             GL_EXTCALL(glProgramEnvParameter4fvARB(target_type, lconst->idx, (GLfloat*)lconst->value));
 487         }
 488         checkGLcall("glProgramEnvParameter4fvARB()");
 489         return ret; /* The loaded immediate constants need reloading for the next shader */
 490     } else {
 491         return 0; /* No constants are dirty now */
 492     }
 493 }
 494
 495 /**
 496  * Loads the texture dimensions for NP2 fixup into the currently set ARB_[vertex/fragment]_programs.
 497  */
 498 static void shader_arb_load_np2fixup_constants(void *shader_priv,
 499         const struct wined3d_gl_info *gl_info, const struct wined3d_state *state)
 500 {
 501     const struct shader_arb_priv * priv = shader_priv;
 502
 503     /* NP2 texcoord fixup is (currently) only done for pixelshaders. */
 504     if (!use_ps(state)) return;
 505
 506     if (priv->compiled_fprog && priv->compiled_fprog->np2fixup_info.super.active) {
 507         const struct arb_ps_np2fixup_info* const fixup = &priv->compiled_fprog->np2fixup_info;
 508         UINT i;
 509         WORD active = fixup->super.active;
 510         GLfloat np2fixup_constants[4 * MAX_FRAGMENT_SAMPLERS];
 511
 512         for (i = 0; active; active >>= 1, ++i)
 513         {
 514             const struct wined3d_texture *tex = state->textures[i];
 515             const unsigned char idx = fixup->super.idx[i];
 516             GLfloat *tex_dim = &np2fixup_constants[(idx >> 1) * 4];
 517
 518             if (!(active & 1)) continue;
 519
 520             if (!tex) {
 521                 FIXME("Nonexistent texture is flagged for NP2 texcoord fixup\n");
 522                 continue;
 523             }
 524
 525             if (idx % 2)
 526             {
 527                 tex_dim[2] = tex->pow2_matrix[0];
 528                 tex_dim[3] = tex->pow2_matrix[5];
 529             }
 530             else
 531             {
 532                 tex_dim[0] = tex->pow2_matrix[0];
 533                 tex_dim[1] = tex->pow2_matrix[5];
 534             }
 535         }
 536
 537         for (i = 0; i < fixup->super.num_consts; ++i) {
 538             GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB,
 539                                                    fixup->offset + i, &np2fixup_constants[i * 4]));
 540         }
 541     }
 542 }
 543
 544 /* GL locking is done by the caller. */
 545 static void shader_arb_ps_local_constants(const struct arb_ps_compiled_shader *gl_shader,
 546         const struct wined3d_context *context, const struct wined3d_state *state, UINT rt_height)
 547 {
 548     const struct wined3d_gl_info *gl_info = context->gl_info;
 549     unsigned char i;
 550
 551     for(i = 0; i < gl_shader->numbumpenvmatconsts; i++)
 552     {
 553         int texunit = gl_shader->bumpenvmatconst[i].texunit;
 554
 555         /* The state manager takes care that this function is always called if the bump env matrix changes */
 556         const float *data = (const float *)&state->texture_states[texunit][WINED3D_TSS_BUMPENV_MAT00];
 557         GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB,
 558                 gl_shader->bumpenvmatconst[i].const_num, data));
 559
 560         if (gl_shader->luminanceconst[i].const_num != WINED3D_CONST_NUM_UNUSED)
 561         {
 562             /* WINED3D_TSS_BUMPENVLSCALE and WINED3D_TSS_BUMPENVLOFFSET are next to each other.
 563              * point gl to the scale, and load 4 floats. x = scale, y = offset, z and w are junk, we
 564              * don't care about them. The pointers are valid for sure because the stateblock is bigger.
 565              * (they're WINED3D_TSS_TEXTURETRANSFORMFLAGS and WINED3D_TSS_ADDRESSW, so most likely 0 or NaN
 566             */
 567             const float *scale = (const float *)&state->texture_states[texunit][WINED3D_TSS_BUMPENV_LSCALE];
 568             GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB,
 569                     gl_shader->luminanceconst[i].const_num, scale));
 570         }
 571     }
 572     checkGLcall("Load bumpmap consts");
 573
 574     if(gl_shader->ycorrection != WINED3D_CONST_NUM_UNUSED)
 575     {
 576         /* ycorrection.x: Backbuffer height(onscreen) or 0(offscreen).
 577         * ycorrection.y: -1.0(onscreen), 1.0(offscreen)
 578         * ycorrection.z: 1.0
 579         * ycorrection.w: 0.0
 580         */
 581         float val[4];
 582         val[0] = context->render_offscreen ? 0.0f : (float) rt_height;
 583         val[1] = context->render_offscreen ? 1.0f : -1.0f;
 584         val[2] = 1.0f;
 585         val[3] = 0.0f;
 586         GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, gl_shader->ycorrection, val));
 587         checkGLcall("y correction loading");
 588     }
 589
 590     if (!gl_shader->num_int_consts) return;
 591
 592     for(i = 0; i < MAX_CONST_I; i++)
 593     {
 594         if(gl_shader->int_consts[i] != WINED3D_CONST_NUM_UNUSED)
 595         {
 596             float val[4];
 597             val[0] = (float)state->ps_consts_i[4 * i];
 598             val[1] = (float)state->ps_consts_i[4 * i + 1];
 599             val[2] = (float)state->ps_consts_i[4 * i + 2];
 600             val[3] = -1.0f;
 601
 602             GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, gl_shader->int_consts[i], val));
 603         }
 604     }
 605     checkGLcall("Load ps int consts");
 606 }
 607
 608 /* GL locking is done by the caller. */
 609 static void shader_arb_vs_local_constants(const struct arb_vs_compiled_shader *gl_shader,
 610         const struct wined3d_context *context, const struct wined3d_state *state)
 611 {
 612     const struct wined3d_gl_info *gl_info = context->gl_info;
 613     float position_fixup[4];
 614     unsigned char i;
 615
 616     /* Upload the position fixup */
 617     shader_get_position_fixup(context, state, position_fixup);
 618     GL_EXTCALL(glProgramLocalParameter4fvARB(GL_VERTEX_PROGRAM_ARB, gl_shader->pos_fixup, position_fixup));
 619
 620     if (!gl_shader->num_int_consts) return;
 621
 622     for(i = 0; i < MAX_CONST_I; i++)
 623     {
 624         if(gl_shader->int_consts[i] != WINED3D_CONST_NUM_UNUSED)
 625         {
 626             float val[4];
 627             val[0] = (float)state->vs_consts_i[4 * i];
 628             val[1] = (float)state->vs_consts_i[4 * i + 1];
 629             val[2] = (float)state->vs_consts_i[4 * i + 2];
 630             val[3] = -1.0f;
 631
 632             GL_EXTCALL(glProgramLocalParameter4fvARB(GL_VERTEX_PROGRAM_ARB, gl_shader->int_consts[i], val));
 633         }
 634     }
 635     checkGLcall("Load vs int consts");
 636 }
 637
 638 static void shader_arb_select(const struct wined3d_context *context, BOOL usePS, BOOL useVS);
 639
 640 /**
 641  * Loads the app-supplied constants into the currently set ARB_[vertex/fragment]_programs.
 642  *
 643  * We only support float constants in ARB at the moment, so don't
 644  * worry about the Integers or Booleans
 645  */
 646 /* GL locking is done by the caller (state handler) */
 647 static void shader_arb_load_constants_internal(const struct wined3d_context *context,
 648         BOOL usePixelShader, BOOL useVertexShader, BOOL from_shader_select)
 649 {
 650     struct wined3d_device *device = context->swapchain->device;
 651     const struct wined3d_stateblock *stateblock = device->stateBlock;
 652     const struct wined3d_state *state = &stateblock->state;
 653     const struct wined3d_gl_info *gl_info = context->gl_info;
 654     struct shader_arb_priv *priv = device->shader_priv;
 655
 656     if (!from_shader_select)
 657     {
 658         const struct wined3d_shader *vshader = state->vertex_shader, *pshader = state->pixel_shader;
 659         if (vshader
 660                 && (stateblock->changed.vertexShaderConstantsB & vshader->reg_maps.boolean_constants
 661                 || (!gl_info->supported[NV_VERTEX_PROGRAM2_OPTION]
 662                 && (stateblock->changed.vertexShaderConstantsI
 663                 & vshader->reg_maps.integer_constants & ~vshader->reg_maps.local_int_consts))))
 664         {
 665             TRACE("bool/integer vertex shader constants potentially modified, forcing shader reselection.\n");
 666             shader_arb_select(context, usePixelShader, useVertexShader);
 667         }
 668         else if (pshader
 669                 && (stateblock->changed.pixelShaderConstantsB & pshader->reg_maps.boolean_constants
 670                 || (!gl_info->supported[NV_FRAGMENT_PROGRAM_OPTION]
 671                 && (stateblock->changed.pixelShaderConstantsI
 672                 & pshader->reg_maps.integer_constants & ~pshader->reg_maps.local_int_consts))))
 673         {
 674             TRACE("bool/integer pixel shader constants potentially modified, forcing shader reselection.\n");
 675             shader_arb_select(context, usePixelShader, useVertexShader);
 676         }
 677     }
 678
 679     if (context != priv->last_context)
 680     {
 681         memset(priv->vshader_const_dirty, 1,
 682                 sizeof(*priv->vshader_const_dirty) * device->d3d_vshader_constantF);
 683         priv->highest_dirty_vs_const = device->d3d_vshader_constantF;
 684
 685         memset(priv->pshader_const_dirty, 1,
 686                 sizeof(*priv->pshader_const_dirty) * device->d3d_pshader_constantF);
 687         priv->highest_dirty_ps_const = device->d3d_pshader_constantF;
 688
 689         priv->last_context = context;
 690     }
 691
 692     if (useVertexShader)
 693     {
 694         struct wined3d_shader *vshader = state->vertex_shader;
 695         const struct arb_vs_compiled_shader *gl_shader = priv->compiled_vprog;
 696
 697         /* Load DirectX 9 float constants for vertex shader */
 698         priv->highest_dirty_vs_const = shader_arb_load_constantsF(vshader, gl_info, GL_VERTEX_PROGRAM_ARB,
 699                 priv->highest_dirty_vs_const, state->vs_consts_f, priv->vshader_const_dirty);
 700         shader_arb_vs_local_constants(gl_shader, context, state);
 701     }
 702
 703     if (usePixelShader)
 704     {
 705         struct wined3d_shader *pshader = state->pixel_shader;
 706         const struct arb_ps_compiled_shader *gl_shader = priv->compiled_fprog;
 707         UINT rt_height = state->fb->render_targets[0]->resource.height;
 708
 709         /* Load DirectX 9 float constants for pixel shader */
 710         priv->highest_dirty_ps_const = shader_arb_load_constantsF(pshader, gl_info, GL_FRAGMENT_PROGRAM_ARB,
 711                 priv->highest_dirty_ps_const, state->ps_consts_f, priv->pshader_const_dirty);
 712         shader_arb_ps_local_constants(gl_shader, context, state, rt_height);
 713     }
 714 }
 715
 716 static void shader_arb_load_constants(const struct wined3d_context *context, BOOL ps, BOOL vs)
 717 {
 718     shader_arb_load_constants_internal(context, ps, vs, FALSE);
 719 }
 720
 721 static void shader_arb_update_float_vertex_constants(struct wined3d_device *device, UINT start, UINT count)
 722 {
 723     struct wined3d_context *context = context_get_current();
 724     struct shader_arb_priv *priv = device->shader_priv;
 725
 726     /* We don't want shader constant dirtification to be an O(contexts), so just dirtify the active
 727      * context. On a context switch the old context will be fully dirtified */
 728     if (!context || context->swapchain->device != device) return;
 729
 730     memset(priv->vshader_const_dirty + start, 1, sizeof(*priv->vshader_const_dirty) * count);
 731     priv->highest_dirty_vs_const = max(priv->highest_dirty_vs_const, start + count);
 732 }
 733
 734 static void shader_arb_update_float_pixel_constants(struct wined3d_device *device, UINT start, UINT count)
 735 {
 736     struct wined3d_context *context = context_get_current();
 737     struct shader_arb_priv *priv = device->shader_priv;
 738
 739     /* We don't want shader constant dirtification to be an O(contexts), so just dirtify the active
 740      * context. On a context switch the old context will be fully dirtified */
 741     if (!context || context->swapchain->device != device) return;
 742
 743     memset(priv->pshader_const_dirty + start, 1, sizeof(*priv->pshader_const_dirty) * count);
 744     priv->highest_dirty_ps_const = max(priv->highest_dirty_ps_const, start + count);
 745 }
 746
 747 static DWORD *local_const_mapping(const struct wined3d_shader *shader)
 748 {
 749     const struct wined3d_shader_lconst *lconst;
 750     DWORD *ret;
 751     DWORD idx = 0;
 752
 753     if (shader->load_local_constsF || list_empty(&shader->constantsF))
 754         return NULL;
 755
 756     ret = HeapAlloc(GetProcessHeap(), 0, sizeof(DWORD) * shader->limits.constant_float);
 757     if (!ret)
 758     {
 759         ERR("Out of memory\n");
 760         return NULL;
 761     }
 762
 763     LIST_FOR_EACH_ENTRY(lconst, &shader->constantsF, struct wined3d_shader_lconst, entry)
 764     {
 765         ret[lconst->idx] = idx++;
 766     }
 767     return ret;
 768 }
 769
 770 /* Generate the variable & register declarations for the ARB_vertex_program output target */
 771 static DWORD shader_generate_arb_declarations(const struct wined3d_shader *shader,
 772         const struct wined3d_shader_reg_maps *reg_maps, struct wined3d_shader_buffer *buffer,
 773         const struct wined3d_gl_info *gl_info, const DWORD *lconst_map,
 774         DWORD *num_clipplanes, const struct shader_arb_ctx_priv *ctx)
 775 {
 776     DWORD i, next_local = 0;
 777     char pshader = shader_is_pshader_version(reg_maps->shader_version.type);
 778     const struct wined3d_shader_lconst *lconst;
 779     unsigned max_constantsF;
 780     DWORD map;
 781
 782     /* In pixel shaders, all private constants are program local, we don't need anything
 783      * from program.env. Thus we can advertise the full set of constants in pixel shaders.
 784      * If we need a private constant the GL implementation will squeeze it in somewhere
 785      *
 786      * With vertex shaders we need the posFixup and on some GL implementations 4 helper
 787      * immediate values. The posFixup is loaded using program.env for now, so always
 788      * subtract one from the number of constants. If the shader uses indirect addressing,
 789      * account for the helper const too because we have to declare all available d3d constants
 790      * and don't know which are actually used.
 791      */
 792     if (pshader)
 793     {
 794         max_constantsF = gl_info->limits.arb_ps_native_constants;
 795         /* 24 is the minimum MAX_PROGRAM_ENV_PARAMETERS_ARB value. */
 796         if (max_constantsF < 24)
 797             max_constantsF = gl_info->limits.arb_ps_float_constants;
 798     }
 799     else
 800     {
 801         const struct arb_vshader_private *shader_data = shader->backend_data;
 802         max_constantsF = gl_info->limits.arb_vs_native_constants;
 803         /* 96 is the minimum MAX_PROGRAM_ENV_PARAMETERS_ARB value.
 804          * Also prevents max_constantsF from becoming less than 0 and
 805          * wrapping . */
 806         if (max_constantsF < 96)
 807             max_constantsF = gl_info->limits.arb_vs_float_constants;
 808
 809         if (reg_maps->usesrelconstF)
 810         {
 811             DWORD highest_constf = 0, clip_limit;
 812
 813             max_constantsF -= reserved_vs_const(shader_data, reg_maps, gl_info);
 814             max_constantsF -= count_bits(reg_maps->integer_constants);
 815
 816             for (i = 0; i < shader->limits.constant_float; ++i)
 817             {
 818                 DWORD idx = i >> 5;
 819                 DWORD shift = i & 0x1f;
 820                 if(reg_maps->constf[idx] & (1 << shift)) highest_constf = i;
 821             }
 822
 823             if(use_nv_clip(gl_info) && ctx->target_version >= NV2)
 824             {
 825                 if(ctx->cur_vs_args->super.clip_enabled)
 826                     clip_limit = gl_info->limits.clipplanes;
 827                 else
 828                     clip_limit = 0;
 829             }
 830             else
 831             {
 832                 unsigned int mask = ctx->cur_vs_args->clip.boolclip.clipplane_mask;
 833                 clip_limit = min(count_bits(mask), 4);
 834             }
 835             *num_clipplanes = min(clip_limit, max_constantsF - highest_constf - 1);
 836             max_constantsF -= *num_clipplanes;
 837             if(*num_clipplanes < clip_limit)
 838             {
 839                 WARN("Only %u clipplanes out of %u enabled\n", *num_clipplanes, gl_info->limits.clipplanes);
 840             }
 841         }
 842         else
 843         {
 844             if (ctx->target_version >= NV2) *num_clipplanes = gl_info->limits.clipplanes;
 845             else *num_clipplanes = min(gl_info->limits.clipplanes, 4);
 846         }
 847     }
 848
 849     for (i = 0, map = reg_maps->temporary; map; map >>= 1, ++i)
 850     {
 851         if (map & 1) shader_addline(buffer, "TEMP R%u;\n", i);
 852     }
 853
 854     for (i = 0, map = reg_maps->address; map; map >>= 1, ++i)
 855     {
 856         if (map & 1) shader_addline(buffer, "ADDRESS A%u;\n", i);
 857     }
 858
 859     if (pshader && reg_maps->shader_version.major == 1 && reg_maps->shader_version.minor <= 3)
 860     {
 861         for (i = 0, map = reg_maps->texcoord; map; map >>= 1, ++i)
 862         {
 863             if (map & 1) shader_addline(buffer, "TEMP T%u;\n", i);
 864         }
 865     }
 866
 867     /* Load local constants using the program-local space,
 868      * this avoids reloading them each time the shader is used
 869      */
 870     if (lconst_map)
 871     {
 872         LIST_FOR_EACH_ENTRY(lconst, &shader->constantsF, struct wined3d_shader_lconst, entry)
 873         {
 874             shader_addline(buffer, "PARAM C%u = program.local[%u];\n", lconst->idx,
 875                            lconst_map[lconst->idx]);
 876             next_local = max(next_local, lconst_map[lconst->idx] + 1);
 877         }
 878     }
 879
 880     /* After subtracting privately used constants from the hardware limit(they are loaded as
 881      * local constants), make sure the shader doesn't violate the env constant limit
 882      */
 883     if(pshader)
 884     {
 885         max_constantsF = min(max_constantsF, gl_info->limits.arb_ps_float_constants);
 886     }
 887     else
 888     {
 889         max_constantsF = min(max_constantsF, gl_info->limits.arb_vs_float_constants);
 890     }
 891
 892     /* Avoid declaring more constants than needed */
 893     max_constantsF = min(max_constantsF, shader->limits.constant_float);
 894
 895     /* we use the array-based constants array if the local constants are marked for loading,
 896      * because then we use indirect addressing, or when the local constant list is empty,
 897      * because then we don't know if we're using indirect addressing or not. If we're hardcoding
 898      * local constants do not declare the loaded constants as an array because ARB compilers usually
 899      * do not optimize unused constants away
 900      */
 901     if (reg_maps->usesrelconstF)
 902     {
 903         /* Need to PARAM the environment parameters (constants) so we can use relative addressing */
 904         shader_addline(buffer, "PARAM C[%d] = { program.env[0..%d] };\n",
 905                     max_constantsF, max_constantsF - 1);
 906     } else {
 907         for(i = 0; i < max_constantsF; i++) {
 908             DWORD idx, mask;
 909             idx = i >> 5;
 910             mask = 1 << (i & 0x1f);
 911             if (!shader_constant_is_local(shader, i) && (reg_maps->constf[idx] & mask))
 912             {
 913                 shader_addline(buffer, "PARAM C%d = program.env[%d];\n",i, i);
 914             }
 915         }
 916     }
 917
 918     return next_local;
 919 }
 920
 921 static const char * const shift_tab[] = {
 922     "dummy",     /*  0 (none) */
 923     "coefmul.x", /*  1 (x2)   */
 924     "coefmul.y", /*  2 (x4)   */
 925     "coefmul.z", /*  3 (x8)   */
 926     "coefmul.w", /*  4 (x16)  */
 927     "dummy",     /*  5 (x32)  */
 928     "dummy",     /*  6 (x64)  */
 929     "dummy",     /*  7 (x128) */
 930     "dummy",     /*  8 (d256) */
 931     "dummy",     /*  9 (d128) */
 932     "dummy",     /* 10 (d64)  */
 933     "dummy",     /* 11 (d32)  */
 934     "coefdiv.w", /* 12 (d16)  */
 935     "coefdiv.z", /* 13 (d8)   */
 936     "coefdiv.y", /* 14 (d4)   */
 937     "coefdiv.x"  /* 15 (d2)   */
 938 };
 939
 940 static void shader_arb_get_write_mask(const struct wined3d_shader_instruction *ins,
 941         const struct wined3d_shader_dst_param *dst, char *write_mask)
 942 {
 943     char *ptr = write_mask;
 944
 945     if (dst->write_mask != WINED3DSP_WRITEMASK_ALL)
 946     {
 947         *ptr++ = '.';
 948         if (dst->write_mask & WINED3DSP_WRITEMASK_0) *ptr++ = 'x';
 949         if (dst->write_mask & WINED3DSP_WRITEMASK_1) *ptr++ = 'y';
 950         if (dst->write_mask & WINED3DSP_WRITEMASK_2) *ptr++ = 'z';
 951         if (dst->write_mask & WINED3DSP_WRITEMASK_3) *ptr++ = 'w';
 952     }
 953
 954     *ptr = '\0';
 955 }
 956
 957 static void shader_arb_get_swizzle(const struct wined3d_shader_src_param *param, BOOL fixup, char *swizzle_str)
 958 {
 959     /* For registers of type WINED3DDECLTYPE_D3DCOLOR, data is stored as "bgra",
 960      * but addressed as "rgba". To fix this we need to swap the register's x
 961      * and z components. */
 962     const char *swizzle_chars = fixup ? "zyxw" : "xyzw";
 963     char *ptr = swizzle_str;
 964
 965     /* swizzle bits fields: wwzzyyxx */
 966     DWORD swizzle = param->swizzle;
 967     DWORD swizzle_x = swizzle & 0x03;
 968     DWORD swizzle_y = (swizzle >> 2) & 0x03;
 969     DWORD swizzle_z = (swizzle >> 4) & 0x03;
 970     DWORD swizzle_w = (swizzle >> 6) & 0x03;
 971
 972     /* If the swizzle is the default swizzle (ie, "xyzw"), we don't need to
 973      * generate a swizzle string. Unless we need to our own swizzling. */
 974     if (swizzle != WINED3DSP_NOSWIZZLE || fixup)
 975     {
 976         *ptr++ = '.';
 977         if (swizzle_x == swizzle_y && swizzle_x == swizzle_z && swizzle_x == swizzle_w) {
 978             *ptr++ = swizzle_chars[swizzle_x];
 979         } else {
 980             *ptr++ = swizzle_chars[swizzle_x];
 981             *ptr++ = swizzle_chars[swizzle_y];
 982             *ptr++ = swizzle_chars[swizzle_z];
 983             *ptr++ = swizzle_chars[swizzle_w];
 984         }
 985     }
 986
 987     *ptr = '\0';
 988 }
 989
 990 static void shader_arb_request_a0(const struct wined3d_shader_instruction *ins, const char *src)
 991 {
 992     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
 993     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
 994
 995     if (!strcmp(priv->addr_reg, src)) return;
 996
 997     strcpy(priv->addr_reg, src);
 998     shader_addline(buffer, "ARL A0.x, %s;\n", src);
 999 }
1000
1001 static void shader_arb_get_src_param(const struct wined3d_shader_instruction *ins,
1002         const struct wined3d_shader_src_param *src, unsigned int tmpreg, char *outregstr);
1003
1004 static void shader_arb_get_register_name(const struct wined3d_shader_instruction *ins,
1005         const struct wined3d_shader_register *reg, char *register_name, BOOL *is_color)
1006 {
1007     /* oPos, oFog and oPts in D3D */
1008     static const char * const rastout_reg_names[] = {"TMP_OUT", "TMP_FOGCOORD", "result.pointsize"};
1009     const struct wined3d_shader *shader = ins->ctx->shader;
1010     const struct wined3d_shader_reg_maps *reg_maps = ins->ctx->reg_maps;
1011     BOOL pshader = shader_is_pshader_version(reg_maps->shader_version.type);
1012     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
1013
1014     *is_color = FALSE;
1015
1016     switch (reg->type)
1017     {
1018         case WINED3DSPR_TEMP:
1019             sprintf(register_name, "R%u", reg->idx[0].offset);
1020             break;
1021
1022         case WINED3DSPR_INPUT:
1023             if (pshader)
1024             {
1025                 if (reg_maps->shader_version.major < 3)
1026                 {
1027                     if (!reg->idx[0].offset)
1028                         strcpy(register_name, "fragment.color.primary");
1029                     else
1030                         strcpy(register_name, "fragment.color.secondary");
1031                 }
1032                 else
1033                 {
1034                     if (reg->idx[0].rel_addr)
1035                     {
1036                         char rel_reg[50];
1037                         shader_arb_get_src_param(ins, reg->idx[0].rel_addr, 0, rel_reg);
1038
1039                         if (!strcmp(rel_reg, "**aL_emul**"))
1040                         {
1041                             DWORD idx = ctx->aL + reg->idx[0].offset;
1042                             if(idx < MAX_REG_INPUT)
1043                             {
1044                                 strcpy(register_name, ctx->ps_input[idx]);
1045                             }
1046                             else
1047                             {
1048                                 ERR("Pixel shader input register out of bounds: %u\n", idx);
1049                                 sprintf(register_name, "out_of_bounds_%u", idx);
1050                             }
1051                         }
1052                         else if (reg_maps->input_registers & 0x0300)
1053                         {
1054                             /* There are two ways basically:
1055                              *
1056                              * 1) Use the unrolling code that is used for loop emulation and unroll the loop.
1057                              *    That means trouble if the loop also contains a breakc or if the control values
1058                              *    aren't local constants.
1059                              * 2) Generate an if block that checks if aL.y < 8, == 8 or == 9 and selects the
1060                              *    source dynamically. The trouble is that we cannot simply read aL.y because it
1061                              *    is an ADDRESS register. We could however push it, load .zw with a value and use
1062                              *    ADAC to load the condition code register and pop it again afterwards
1063                              */
1064                             FIXME("Relative input register addressing with more than 8 registers\n");
1065
1066                             /* This is better than nothing for now */
1067                             sprintf(register_name, "fragment.texcoord[%s + %u]", rel_reg, reg->idx[0].offset);
1068                         }
1069                         else if(ctx->cur_ps_args->super.vp_mode != vertexshader)
1070                         {
1071                             /* This is problematic because we'd have to consult the ctx->ps_input strings
1072                              * for where to find the varying. Some may be "0.0", others can be texcoords or
1073                              * colors. This needs either a pipeline replacement to make the vertex shader feed
1074                              * proper varyings, or loop unrolling
1075                              *
1076                              * For now use the texcoords and hope for the best
1077                              */
1078                             FIXME("Non-vertex shader varying input with indirect addressing\n");
1079                             sprintf(register_name, "fragment.texcoord[%s + %u]", rel_reg, reg->idx[0].offset);
1080                         }
1081                         else
1082                         {
1083                             /* D3D supports indirect addressing only with aL in loop registers. The loop instruction
1084                              * pulls GL_NV_fragment_program2 in
1085                              */
1086                             sprintf(register_name, "fragment.texcoord[%s + %u]", rel_reg, reg->idx[0].offset);
1087                         }
1088                     }
1089                     else
1090                     {
1091                         if (reg->idx[0].offset < MAX_REG_INPUT)
1092                         {
1093                             strcpy(register_name, ctx->ps_input[reg->idx[0].offset]);
1094                         }
1095                         else
1096                         {
1097                             ERR("Pixel shader input register out of bounds: %u\n", reg->idx[0].offset);
1098                             sprintf(register_name, "out_of_bounds_%u", reg->idx[0].offset);
1099                         }
1100                     }
1101                 }
1102             }
1103             else
1104             {
1105                 if (ctx->cur_vs_args->super.swizzle_map & (1 << reg->idx[0].offset))
1106                     *is_color = TRUE;
1107                 sprintf(register_name, "vertex.attrib[%u]", reg->idx[0].offset);
1108             }
1109             break;
1110
1111         case WINED3DSPR_CONST:
1112             if (!pshader && reg->idx[0].rel_addr)
1113             {
1114                 const struct arb_vshader_private *shader_data = shader->backend_data;
1115                 UINT rel_offset = shader_data->rel_offset;
1116                 BOOL aL = FALSE;
1117                 char rel_reg[50];
1118                 if (reg_maps->shader_version.major < 2)
1119                 {
1120                     sprintf(rel_reg, "A0.x");
1121                 }
1122                 else
1123                 {
1124                     shader_arb_get_src_param(ins, reg->idx[0].rel_addr, 0, rel_reg);
1125                     if (ctx->target_version == ARB)
1126                     {
1127                         if (!strcmp(rel_reg, "**aL_emul**"))
1128                         {
1129                             aL = TRUE;
1130                         } else {
1131                             shader_arb_request_a0(ins, rel_reg);
1132                             sprintf(rel_reg, "A0.x");
1133                         }
1134                     }
1135                 }
1136                 if (aL)
1137                     sprintf(register_name, "C[%u]", ctx->aL + reg->idx[0].offset);
1138                 else if (reg->idx[0].offset >= rel_offset)
1139                     sprintf(register_name, "C[%s + %u]", rel_reg, reg->idx[0].offset - rel_offset);
1140                 else
1141                     sprintf(register_name, "C[%s - %u]", rel_reg, rel_offset - reg->idx[0].offset);
1142             }
1143             else
1144             {
1145                 if (reg_maps->usesrelconstF)
1146                     sprintf(register_name, "C[%u]", reg->idx[0].offset);
1147                 else
1148                     sprintf(register_name, "C%u", reg->idx[0].offset);
1149             }
1150             break;
1151
1152         case WINED3DSPR_TEXTURE: /* case WINED3DSPR_ADDR: */
1153             if (pshader)
1154             {
1155                 if (reg_maps->shader_version.major == 1
1156                         && reg_maps->shader_version.minor <= 3)
1157                     /* In ps <= 1.3, Tx is a temporary register as destination
1158                      * to all instructions, and as source to most instructions.
1159                      * For some instructions it is the texcoord input. Those
1160                      * instructions know about the special use. */
1161                     sprintf(register_name, "T%u", reg->idx[0].offset);
1162                 else
1163                     /* In ps 1.4 and 2.x Tx is always a (read-only) varying. */
1164                     sprintf(register_name, "fragment.texcoord[%u]", reg->idx[0].offset);
1165             }
1166             else
1167             {
1168                 if (reg_maps->shader_version.major == 1 || ctx->target_version >= NV2)
1169                     sprintf(register_name, "A%u", reg->idx[0].offset);
1170                 else
1171                     sprintf(register_name, "A%u_SHADOW", reg->idx[0].offset);
1172             }
1173             break;
1174
1175         case WINED3DSPR_COLOROUT:
1176             if (ctx->cur_ps_args->super.srgb_correction && !reg->idx[0].offset)
1177             {
1178                 strcpy(register_name, "TMP_COLOR");
1179             }
1180             else
1181             {
1182                 if (ctx->cur_ps_args->super.srgb_correction)
1183                     FIXME("sRGB correction on higher render targets.\n");
1184                 if (reg_maps->rt_mask > 1)
1185                     sprintf(register_name, "result.color[%u]", reg->idx[0].offset);
1186                 else
1187                     strcpy(register_name, "result.color");
1188             }
1189             break;
1190
1191         case WINED3DSPR_RASTOUT:
1192             if (reg->idx[0].offset == 1)
1193                 sprintf(register_name, "%s", ctx->fog_output);
1194             else
1195                 sprintf(register_name, "%s", rastout_reg_names[reg->idx[0].offset]);
1196             break;
1197
1198         case WINED3DSPR_DEPTHOUT:
1199             strcpy(register_name, "result.depth");
1200             break;
1201
1202         case WINED3DSPR_ATTROUT:
1203         /* case WINED3DSPR_OUTPUT: */
1204             if (pshader)
1205                 sprintf(register_name, "oD[%u]", reg->idx[0].offset);
1206             else
1207                 strcpy(register_name, ctx->color_output[reg->idx[0].offset]);
1208             break;
1209
1210         case WINED3DSPR_TEXCRDOUT:
1211             if (pshader)
1212                 sprintf(register_name, "oT[%u]", reg->idx[0].offset);
1213             else if (reg_maps->shader_version.major < 3)
1214                 strcpy(register_name, ctx->texcrd_output[reg->idx[0].offset]);
1215             else
1216                 strcpy(register_name, ctx->vs_output[reg->idx[0].offset]);
1217             break;
1218
1219         case WINED3DSPR_LOOP:
1220             if(ctx->target_version >= NV2)
1221             {
1222                 /* Pshader has an implicitly declared loop index counter A0.x that cannot be renamed */
1223                 if(pshader) sprintf(register_name, "A0.x");
1224                 else sprintf(register_name, "aL.y");
1225             }
1226             else
1227             {
1228                 /* Unfortunately this code cannot return the value of ctx->aL here. An immediate value
1229                  * would be valid, but if aL is used for indexing(its only use), there's likely an offset,
1230                  * thus the result would be something like C[15 + 30], which is not valid in the ARB program
1231                  * grammar. So return a marker for the emulated aL and intercept it in constant and varying
1232                  * indexing
1233                  */
1234                 sprintf(register_name, "**aL_emul**");
1235             }
1236
1237             break;
1238
1239         case WINED3DSPR_CONSTINT:
1240             sprintf(register_name, "I%u", reg->idx[0].offset);
1241             break;
1242
1243         case WINED3DSPR_MISCTYPE:
1244             if (!reg->idx[0].offset)
1245                 sprintf(register_name, "vpos");
1246             else if (reg->idx[0].offset == 1)
1247                 sprintf(register_name, "fragment.facing.x");
1248             else
1249                 FIXME("Unknown MISCTYPE register index %u.\n", reg->idx[0].offset);
1250             break;
1251
1252         default:
1253             FIXME("Unhandled register type %#x[%u].\n", reg->type, reg->idx[0].offset);
1254             sprintf(register_name, "unrecognized_register[%u]", reg->idx[0].offset);
1255             break;
1256     }
1257 }
1258
1259 static void shader_arb_get_dst_param(const struct wined3d_shader_instruction *ins,
1260         const struct wined3d_shader_dst_param *wined3d_dst, char *str)
1261 {
1262     char register_name[255];
1263     char write_mask[6];
1264     BOOL is_color;
1265
1266     shader_arb_get_register_name(ins, &wined3d_dst->reg, register_name, &is_color);
1267     strcpy(str, register_name);
1268
1269     shader_arb_get_write_mask(ins, wined3d_dst, write_mask);
1270     strcat(str, write_mask);
1271 }
1272
1273 static const char *shader_arb_get_fixup_swizzle(enum fixup_channel_source channel_source)
1274 {
1275     switch(channel_source)
1276     {
1277         case CHANNEL_SOURCE_ZERO: return "0";
1278         case CHANNEL_SOURCE_ONE: return "1";
1279         case CHANNEL_SOURCE_X: return "x";
1280         case CHANNEL_SOURCE_Y: return "y";
1281         case CHANNEL_SOURCE_Z: return "z";
1282         case CHANNEL_SOURCE_W: return "w";
1283         default:
1284             FIXME("Unhandled channel source %#x\n", channel_source);
1285             return "undefined";
1286     }
1287 }
1288
1289 static void gen_color_correction(struct wined3d_shader_buffer *buffer, const char *reg,
1290         DWORD dst_mask, const char *one, const char *two, struct color_fixup_desc fixup)
1291 {
1292     DWORD mask;
1293
1294     if (is_complex_fixup(fixup))
1295     {
1296         enum complex_fixup complex_fixup = get_complex_fixup(fixup);
1297         FIXME("Complex fixup (%#x) not supported\n", complex_fixup);
1298         return;
1299     }
1300
1301     mask = 0;
1302     if (fixup.x_source != CHANNEL_SOURCE_X) mask |= WINED3DSP_WRITEMASK_0;
1303     if (fixup.y_source != CHANNEL_SOURCE_Y) mask |= WINED3DSP_WRITEMASK_1;
1304     if (fixup.z_source != CHANNEL_SOURCE_Z) mask |= WINED3DSP_WRITEMASK_2;
1305     if (fixup.w_source != CHANNEL_SOURCE_W) mask |= WINED3DSP_WRITEMASK_3;
1306     mask &= dst_mask;
1307
1308     if (mask)
1309     {
1310         shader_addline(buffer, "SWZ %s, %s, %s, %s, %s, %s;\n", reg, reg,
1311                 shader_arb_get_fixup_swizzle(fixup.x_source), shader_arb_get_fixup_swizzle(fixup.y_source),
1312                 shader_arb_get_fixup_swizzle(fixup.z_source), shader_arb_get_fixup_swizzle(fixup.w_source));
1313     }
1314
1315     mask = 0;
1316     if (fixup.x_sign_fixup) mask |= WINED3DSP_WRITEMASK_0;
1317     if (fixup.y_sign_fixup) mask |= WINED3DSP_WRITEMASK_1;
1318     if (fixup.z_sign_fixup) mask |= WINED3DSP_WRITEMASK_2;
1319     if (fixup.w_sign_fixup) mask |= WINED3DSP_WRITEMASK_3;
1320     mask &= dst_mask;
1321
1322     if (mask)
1323     {
1324         char reg_mask[6];
1325         char *ptr = reg_mask;
1326
1327         if (mask != WINED3DSP_WRITEMASK_ALL)
1328         {
1329             *ptr++ = '.';
1330             if (mask & WINED3DSP_WRITEMASK_0) *ptr++ = 'x';
1331             if (mask & WINED3DSP_WRITEMASK_1) *ptr++ = 'y';
1332             if (mask & WINED3DSP_WRITEMASK_2) *ptr++ = 'z';
1333             if (mask & WINED3DSP_WRITEMASK_3) *ptr++ = 'w';
1334         }
1335         *ptr = '\0';
1336
1337         shader_addline(buffer, "MAD %s%s, %s, %s, -%s;\n", reg, reg_mask, reg, two, one);
1338     }
1339 }
1340
1341 static const char *shader_arb_get_modifier(const struct wined3d_shader_instruction *ins)
1342 {
1343     DWORD mod;
1344     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
1345     if (!ins->dst_count) return "";
1346
1347     mod = ins->dst[0].modifiers;
1348
1349     /* Silently ignore PARTIALPRECISION if its not supported */
1350     if(priv->target_version == ARB) mod &= ~WINED3DSPDM_PARTIALPRECISION;
1351
1352     if(mod & WINED3DSPDM_MSAMPCENTROID)
1353     {
1354         FIXME("Unhandled modifier WINED3DSPDM_MSAMPCENTROID\n");
1355         mod &= ~WINED3DSPDM_MSAMPCENTROID;
1356     }
1357
1358     switch(mod)
1359     {
1360         case WINED3DSPDM_SATURATE | WINED3DSPDM_PARTIALPRECISION:
1361             return "H_SAT";
1362
1363         case WINED3DSPDM_SATURATE:
1364             return "_SAT";
1365
1366         case WINED3DSPDM_PARTIALPRECISION:
1367             return "H";
1368
1369         case 0:
1370             return "";
1371
1372         default:
1373             FIXME("Unknown modifiers 0x%08x\n", mod);
1374             return "";
1375     }
1376 }
1377
1378 #define TEX_PROJ        0x1
1379 #define TEX_BIAS        0x2
1380 #define TEX_LOD         0x4
1381 #define TEX_DERIV       0x10
1382
1383 static void shader_hw_sample(const struct wined3d_shader_instruction *ins, DWORD sampler_idx,
1384         const char *dst_str, const char *coord_reg, WORD flags, const char *dsx, const char *dsy)
1385 {
1386     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1387     DWORD sampler_type = ins->ctx->reg_maps->sampler_type[sampler_idx];
1388     const struct wined3d_shader *shader = ins->ctx->shader;
1389     const struct wined3d_texture *texture;
1390     const char *tex_type;
1391     BOOL np2_fixup = FALSE;
1392     struct wined3d_device *device = shader->device;
1393     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
1394     const char *mod;
1395     BOOL pshader = shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type);
1396
1397     /* D3D vertex shader sampler IDs are vertex samplers(0-3), not global d3d samplers */
1398     if(!pshader) sampler_idx += MAX_FRAGMENT_SAMPLERS;
1399
1400     switch(sampler_type) {
1401         case WINED3DSTT_1D:
1402             tex_type = "1D";
1403             break;
1404
1405         case WINED3DSTT_2D:
1406             texture = device->stateBlock->state.textures[sampler_idx];
1407             if (texture && texture->target == GL_TEXTURE_RECTANGLE_ARB)
1408             {
1409                 tex_type = "RECT";
1410             } else {
1411                 tex_type = "2D";
1412             }
1413             if (shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type))
1414             {
1415                 if (priv->cur_np2fixup_info->super.active & (1 << sampler_idx))
1416                 {
1417                     if (flags) FIXME("Only ordinary sampling from NP2 textures is supported.\n");
1418                     else np2_fixup = TRUE;
1419                 }
1420             }
1421             break;
1422
1423         case WINED3DSTT_VOLUME:
1424             tex_type = "3D";
1425             break;
1426
1427         case WINED3DSTT_CUBE:
1428             tex_type = "CUBE";
1429             break;
1430
1431         default:
1432             ERR("Unexpected texture type %d\n", sampler_type);
1433             tex_type = "";
1434     }
1435
1436     /* TEX, TXL, TXD and TXP do not support the "H" modifier,
1437      * so don't use shader_arb_get_modifier
1438      */
1439     if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE) mod = "_SAT";
1440     else mod = "";
1441
1442     /* Fragment samplers always have indentity mapping */
1443     if(sampler_idx >= MAX_FRAGMENT_SAMPLERS)
1444     {
1445         sampler_idx = priv->cur_vs_args->vertex.samplers[sampler_idx - MAX_FRAGMENT_SAMPLERS];
1446     }
1447
1448     if (flags & TEX_DERIV)
1449     {
1450         if(flags & TEX_PROJ) FIXME("Projected texture sampling with custom derivatives\n");
1451         if(flags & TEX_BIAS) FIXME("Biased texture sampling with custom derivatives\n");
1452         shader_addline(buffer, "TXD%s %s, %s, %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg,
1453                        dsx, dsy,sampler_idx, tex_type);
1454     }
1455     else if(flags & TEX_LOD)
1456     {
1457         if(flags & TEX_PROJ) FIXME("Projected texture sampling with explicit lod\n");
1458         if(flags & TEX_BIAS) FIXME("Biased texture sampling with explicit lod\n");
1459         shader_addline(buffer, "TXL%s %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg,
1460                        sampler_idx, tex_type);
1461     }
1462     else if (flags & TEX_BIAS)
1463     {
1464         /* Shouldn't be possible, but let's check for it */
1465         if(flags & TEX_PROJ) FIXME("Biased and Projected texture sampling\n");
1466         /* TXB takes the 4th component of the source vector automatically, as d3d. Nothing more to do */
1467         shader_addline(buffer, "TXB%s %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg, sampler_idx, tex_type);
1468     }
1469     else if (flags & TEX_PROJ)
1470     {
1471         shader_addline(buffer, "TXP%s %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg, sampler_idx, tex_type);
1472     }
1473     else
1474     {
1475         if (np2_fixup)
1476         {
1477             const unsigned char idx = priv->cur_np2fixup_info->super.idx[sampler_idx];
1478             shader_addline(buffer, "MUL TA, np2fixup[%u].%s, %s;\n", idx >> 1,
1479                            (idx % 2) ? "zwxy" : "xyzw", coord_reg);
1480
1481             shader_addline(buffer, "TEX%s %s, TA, texture[%u], %s;\n", mod, dst_str, sampler_idx, tex_type);
1482         }
1483         else
1484             shader_addline(buffer, "TEX%s %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg, sampler_idx, tex_type);
1485     }
1486
1487     if (pshader)
1488     {
1489         gen_color_correction(buffer, dst_str, ins->dst[0].write_mask,
1490                 arb_get_helper_value(WINED3D_SHADER_TYPE_PIXEL, ARB_ONE),
1491                 arb_get_helper_value(WINED3D_SHADER_TYPE_PIXEL, ARB_TWO),
1492                 priv->cur_ps_args->super.color_fixup[sampler_idx]);
1493     }
1494 }
1495
1496 static void shader_arb_get_src_param(const struct wined3d_shader_instruction *ins,
1497         const struct wined3d_shader_src_param *src, unsigned int tmpreg, char *outregstr)
1498 {
1499     /* Generate a line that does the input modifier computation and return the input register to use */
1500     BOOL is_color = FALSE;
1501     char regstr[256];
1502     char swzstr[20];
1503     int insert_line;
1504     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1505     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
1506     const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
1507     const char *two = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_TWO);
1508
1509     /* Assume a new line will be added */
1510     insert_line = 1;
1511
1512     /* Get register name */
1513     shader_arb_get_register_name(ins, &src->reg, regstr, &is_color);
1514     shader_arb_get_swizzle(src, is_color, swzstr);
1515
1516     switch (src->modifiers)
1517     {
1518     case WINED3DSPSM_NONE:
1519         sprintf(outregstr, "%s%s", regstr, swzstr);
1520         insert_line = 0;
1521         break;
1522     case WINED3DSPSM_NEG:
1523         sprintf(outregstr, "-%s%s", regstr, swzstr);
1524         insert_line = 0;
1525         break;
1526     case WINED3DSPSM_BIAS:
1527         shader_addline(buffer, "ADD T%c, %s, -coefdiv.x;\n", 'A' + tmpreg, regstr);
1528         break;
1529     case WINED3DSPSM_BIASNEG:
1530         shader_addline(buffer, "ADD T%c, -%s, coefdiv.x;\n", 'A' + tmpreg, regstr);
1531         break;
1532     case WINED3DSPSM_SIGN:
1533         shader_addline(buffer, "MAD T%c, %s, %s, -%s;\n", 'A' + tmpreg, regstr, two, one);
1534         break;
1535     case WINED3DSPSM_SIGNNEG:
1536         shader_addline(buffer, "MAD T%c, %s, -%s, %s;\n", 'A' + tmpreg, regstr, two, one);
1537         break;
1538     case WINED3DSPSM_COMP:
1539         shader_addline(buffer, "SUB T%c, %s, %s;\n", 'A' + tmpreg, one, regstr);
1540         break;
1541     case WINED3DSPSM_X2:
1542         shader_addline(buffer, "ADD T%c, %s, %s;\n", 'A' + tmpreg, regstr, regstr);
1543         break;
1544     case WINED3DSPSM_X2NEG:
1545         shader_addline(buffer, "ADD T%c, -%s, -%s;\n", 'A' + tmpreg, regstr, regstr);
1546         break;
1547     case WINED3DSPSM_DZ:
1548         shader_addline(buffer, "RCP T%c, %s.z;\n", 'A' + tmpreg, regstr);
1549         shader_addline(buffer, "MUL T%c, %s, T%c;\n", 'A' + tmpreg, regstr, 'A' + tmpreg);
1550         break;
1551     case WINED3DSPSM_DW:
1552         shader_addline(buffer, "RCP T%c, %s.w;\n", 'A' + tmpreg, regstr);
1553         shader_addline(buffer, "MUL T%c, %s, T%c;\n", 'A' + tmpreg, regstr, 'A' + tmpreg);
1554         break;
1555     case WINED3DSPSM_ABS:
1556         if(ctx->target_version >= NV2) {
1557             sprintf(outregstr, "|%s%s|", regstr, swzstr);
1558             insert_line = 0;
1559         } else {
1560             shader_addline(buffer, "ABS T%c, %s;\n", 'A' + tmpreg, regstr);
1561         }
1562         break;
1563     case WINED3DSPSM_ABSNEG:
1564         if(ctx->target_version >= NV2) {
1565             sprintf(outregstr, "-|%s%s|", regstr, swzstr);
1566         } else {
1567             shader_addline(buffer, "ABS T%c, %s;\n", 'A' + tmpreg, regstr);
1568             sprintf(outregstr, "-T%c%s", 'A' + tmpreg, swzstr);
1569         }
1570         insert_line = 0;
1571         break;
1572     default:
1573         sprintf(outregstr, "%s%s", regstr, swzstr);
1574         insert_line = 0;
1575     }
1576
1577     /* Return modified or original register, with swizzle */
1578     if (insert_line)
1579         sprintf(outregstr, "T%c%s", 'A' + tmpreg, swzstr);
1580 }
1581
1582 static void pshader_hw_bem(const struct wined3d_shader_instruction *ins)
1583 {
1584     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1585     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1586     DWORD sampler_code = dst->reg.idx[0].offset;
1587     char dst_name[50];
1588     char src_name[2][50];
1589
1590     shader_arb_get_dst_param(ins, dst, dst_name);
1591
1592     /* Sampling the perturbation map in Tsrc was done already, including the signedness correction if needed
1593      *
1594      * Keep in mind that src_name[1] can be "TB" and src_name[0] can be "TA" because modifiers like _x2 are valid
1595      * with bem. So delay loading the first parameter until after the perturbation calculation which needs two
1596      * temps is done.
1597      */
1598     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1599     shader_addline(buffer, "SWZ TA, bumpenvmat%d, x, z, 0, 0;\n", sampler_code);
1600     shader_addline(buffer, "DP3 TC.r, TA, %s;\n", src_name[1]);
1601     shader_addline(buffer, "SWZ TA, bumpenvmat%d, y, w, 0, 0;\n", sampler_code);
1602     shader_addline(buffer, "DP3 TC.g, TA, %s;\n", src_name[1]);
1603
1604     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
1605     shader_addline(buffer, "ADD %s, %s, TC;\n", dst_name, src_name[0]);
1606 }
1607
1608 static DWORD negate_modifiers(DWORD mod, char *extra_char)
1609 {
1610     *extra_char = ' ';
1611     switch(mod)
1612     {
1613         case WINED3DSPSM_NONE:      return WINED3DSPSM_NEG;
1614         case WINED3DSPSM_NEG:       return WINED3DSPSM_NONE;
1615         case WINED3DSPSM_BIAS:      return WINED3DSPSM_BIASNEG;
1616         case WINED3DSPSM_BIASNEG:   return WINED3DSPSM_BIAS;
1617         case WINED3DSPSM_SIGN:      return WINED3DSPSM_SIGNNEG;
1618         case WINED3DSPSM_SIGNNEG:   return WINED3DSPSM_SIGN;
1619         case WINED3DSPSM_COMP:      *extra_char = '-'; return WINED3DSPSM_COMP;
1620         case WINED3DSPSM_X2:        return WINED3DSPSM_X2NEG;
1621         case WINED3DSPSM_X2NEG:     return WINED3DSPSM_X2;
1622         case WINED3DSPSM_DZ:        *extra_char = '-'; return WINED3DSPSM_DZ;
1623         case WINED3DSPSM_DW:        *extra_char = '-'; return WINED3DSPSM_DW;
1624         case WINED3DSPSM_ABS:       return WINED3DSPSM_ABSNEG;
1625         case WINED3DSPSM_ABSNEG:    return WINED3DSPSM_ABS;
1626     }
1627     FIXME("Unknown modifier %u\n", mod);
1628     return mod;
1629 }
1630
1631 static void pshader_hw_cnd(const struct wined3d_shader_instruction *ins)
1632 {
1633     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1634     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1635     char dst_name[50];
1636     char src_name[3][50];
1637     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
1638             ins->ctx->reg_maps->shader_version.minor);
1639
1640     shader_arb_get_dst_param(ins, dst, dst_name);
1641     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1642
1643     /* The coissue flag changes the semantic of the cnd instruction in <= 1.3 shaders */
1644     if (shader_version <= WINED3D_SHADER_VERSION(1, 3) && ins->coissue)
1645     {
1646         shader_addline(buffer, "MOV%s %s, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name[1]);
1647     }
1648     else
1649     {
1650         struct wined3d_shader_src_param src0_copy = ins->src[0];
1651         char extra_neg;
1652
1653         /* src0 may have a negate srcmod set, so we can't blindly add "-" to the name */
1654         src0_copy.modifiers = negate_modifiers(src0_copy.modifiers, &extra_neg);
1655
1656         shader_arb_get_src_param(ins, &src0_copy, 0, src_name[0]);
1657         shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
1658         shader_addline(buffer, "ADD TA, %c%s, coefdiv.x;\n", extra_neg, src_name[0]);
1659         shader_addline(buffer, "CMP%s %s, TA, %s, %s;\n", shader_arb_get_modifier(ins),
1660                 dst_name, src_name[1], src_name[2]);
1661     }
1662 }
1663
1664 static void pshader_hw_cmp(const struct wined3d_shader_instruction *ins)
1665 {
1666     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1667     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1668     char dst_name[50];
1669     char src_name[3][50];
1670
1671     shader_arb_get_dst_param(ins, dst, dst_name);
1672
1673     /* Generate input register names (with modifiers) */
1674     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
1675     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1676     shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
1677
1678     shader_addline(buffer, "CMP%s %s, %s, %s, %s;\n", shader_arb_get_modifier(ins),
1679             dst_name, src_name[0], src_name[2], src_name[1]);
1680 }
1681
1682 /** Process the WINED3DSIO_DP2ADD instruction in ARB.
1683  * dst = dot2(src0, src1) + src2 */
1684 static void pshader_hw_dp2add(const struct wined3d_shader_instruction *ins)
1685 {
1686     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1687     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1688     char dst_name[50];
1689     char src_name[3][50];
1690     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
1691
1692     shader_arb_get_dst_param(ins, dst, dst_name);
1693     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
1694     shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
1695
1696     if(ctx->target_version >= NV3)
1697     {
1698         /* GL_NV_fragment_program2 has a 1:1 matching instruction */
1699         shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1700         shader_addline(buffer, "DP2A%s %s, %s, %s, %s;\n", shader_arb_get_modifier(ins),
1701                        dst_name, src_name[0], src_name[1], src_name[2]);
1702     }
1703     else if(ctx->target_version >= NV2)
1704     {
1705         /* dst.x = src2.?, src0.x, src1.x + src0.y * src1.y
1706          * dst.y = src2.?, src0.x, src1.z + src0.y * src1.w
1707          * dst.z = src2.?, src0.x, src1.x + src0.y * src1.y
1708          * dst.z = src2.?, src0.x, src1.z + src0.y * src1.w
1709          *
1710          * Make sure that src1.zw = src1.xy, then we get a classic dp2add
1711          *
1712          * .xyxy and other swizzles that we could get with this are not valid in
1713          * plain ARBfp, but luckily the NV extension grammar lifts this limitation.
1714          */
1715         struct wined3d_shader_src_param tmp_param = ins->src[1];
1716         DWORD swizzle = tmp_param.swizzle & 0xf; /* Selects .xy */
1717         tmp_param.swizzle = swizzle | (swizzle << 4); /* Creates .xyxy */
1718
1719         shader_arb_get_src_param(ins, &tmp_param, 1, src_name[1]);
1720
1721         shader_addline(buffer, "X2D%s %s, %s, %s, %s;\n", shader_arb_get_modifier(ins),
1722                        dst_name, src_name[2], src_name[0], src_name[1]);
1723     }
1724     else
1725     {
1726         shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1727         /* 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]
1728         * src_name[0] can be TA, but TA is a private temp for modifiers, so it is save to overwrite
1729         */
1730         shader_addline(buffer, "MOV TA, %s;\n", src_name[0]);
1731         shader_addline(buffer, "MOV TA.z, 0.0;\n");
1732         shader_addline(buffer, "DP3 TA, TA, %s;\n", src_name[1]);
1733         shader_addline(buffer, "ADD%s %s, TA, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name[2]);
1734     }
1735 }
1736
1737 /* Map the opcode 1-to-1 to the GL code */
1738 static void shader_hw_map2gl(const struct wined3d_shader_instruction *ins)
1739 {
1740     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1741     const char *instruction;
1742     char arguments[256], dst_str[50];
1743     unsigned int i;
1744     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1745
1746     switch (ins->handler_idx)
1747     {
1748         case WINED3DSIH_ABS: instruction = "ABS"; break;
1749         case WINED3DSIH_ADD: instruction = "ADD"; break;
1750         case WINED3DSIH_CRS: instruction = "XPD"; break;
1751         case WINED3DSIH_DP3: instruction = "DP3"; break;
1752         case WINED3DSIH_DP4: instruction = "DP4"; break;
1753         case WINED3DSIH_DST: instruction = "DST"; break;
1754         case WINED3DSIH_FRC: instruction = "FRC"; break;
1755         case WINED3DSIH_LIT: instruction = "LIT"; break;
1756         case WINED3DSIH_LRP: instruction = "LRP"; break;
1757         case WINED3DSIH_MAD: instruction = "MAD"; break;
1758         case WINED3DSIH_MAX: instruction = "MAX"; break;
1759         case WINED3DSIH_MIN: instruction = "MIN"; break;
1760         case WINED3DSIH_MOV: instruction = "MOV"; break;
1761         case WINED3DSIH_MUL: instruction = "MUL"; break;
1762         case WINED3DSIH_SGE: instruction = "SGE"; break;
1763         case WINED3DSIH_SLT: instruction = "SLT"; break;
1764         case WINED3DSIH_SUB: instruction = "SUB"; break;
1765         case WINED3DSIH_MOVA:instruction = "ARR"; break;
1766         case WINED3DSIH_DSX: instruction = "DDX"; break;
1767         default: instruction = "";
1768             FIXME("Unhandled opcode %#x\n", ins->handler_idx);
1769             break;
1770     }
1771
1772     /* Note that shader_arb_add_dst_param() adds spaces. */
1773     arguments[0] = '\0';
1774     shader_arb_get_dst_param(ins, dst, dst_str);
1775     for (i = 0; i < ins->src_count; ++i)
1776     {
1777         char operand[100];
1778         strcat(arguments, ", ");
1779         shader_arb_get_src_param(ins, &ins->src[i], i, operand);
1780         strcat(arguments, operand);
1781     }
1782     shader_addline(buffer, "%s%s %s%s;\n", instruction, shader_arb_get_modifier(ins), dst_str, arguments);
1783 }
1784
1785 static void shader_hw_nop(const struct wined3d_shader_instruction *ins) {}
1786
1787 static void shader_hw_mov(const struct wined3d_shader_instruction *ins)
1788 {
1789     const struct wined3d_shader *shader = ins->ctx->shader;
1790     const struct wined3d_shader_reg_maps *reg_maps = ins->ctx->reg_maps;
1791     BOOL pshader = shader_is_pshader_version(reg_maps->shader_version.type);
1792     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
1793     const char *zero = arb_get_helper_value(reg_maps->shader_version.type, ARB_ZERO);
1794     const char *one = arb_get_helper_value(reg_maps->shader_version.type, ARB_ONE);
1795     const char *two = arb_get_helper_value(reg_maps->shader_version.type, ARB_TWO);
1796
1797     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1798     char src0_param[256];
1799
1800     if (ins->handler_idx == WINED3DSIH_MOVA)
1801     {
1802         const struct arb_vshader_private *shader_data = shader->backend_data;
1803         char write_mask[6];
1804         const char *offset = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_VS_REL_OFFSET);
1805
1806         if(ctx->target_version >= NV2) {
1807             shader_hw_map2gl(ins);
1808             return;
1809         }
1810         shader_arb_get_src_param(ins, &ins->src[0], 0, src0_param);
1811         shader_arb_get_write_mask(ins, &ins->dst[0], write_mask);
1812
1813         /* This implements the mova formula used in GLSL. The first two instructions
1814          * prepare the sign() part. Note that it is fine to have my_sign(0.0) = 1.0
1815          * in this case:
1816          * mova A0.x, 0.0
1817          *
1818          * A0.x = arl(floor(abs(0.0) + 0.5) * 1.0) = floor(0.5) = 0.0 since arl does a floor
1819          *
1820          * The ARL is performed when A0 is used - the requested component is read from A0_SHADOW into
1821          * A0.x. We can use the overwritten component of A0_shadow as temporary storage for the sign.
1822          */
1823         shader_addline(buffer, "SGE A0_SHADOW%s, %s, %s;\n", write_mask, src0_param, zero);
1824         shader_addline(buffer, "MAD A0_SHADOW%s, A0_SHADOW, %s, -%s;\n", write_mask, two, one);
1825
1826         shader_addline(buffer, "ABS TA%s, %s;\n", write_mask, src0_param);
1827         shader_addline(buffer, "ADD TA%s, TA, rel_addr_const.x;\n", write_mask);
1828         shader_addline(buffer, "FLR TA%s, TA;\n", write_mask);
1829         if (shader_data->rel_offset)
1830         {
1831             shader_addline(buffer, "ADD TA%s, TA, %s;\n", write_mask, offset);
1832         }
1833         shader_addline(buffer, "MUL A0_SHADOW%s, TA, A0_SHADOW;\n", write_mask);
1834
1835         ((struct shader_arb_ctx_priv *)ins->ctx->backend_data)->addr_reg[0] = '\0';
1836     }
1837     else if (reg_maps->shader_version.major == 1
1838           && !shader_is_pshader_version(reg_maps->shader_version.type)
1839           && ins->dst[0].reg.type == WINED3DSPR_ADDR)
1840     {
1841         const struct arb_vshader_private *shader_data = shader->backend_data;
1842         src0_param[0] = '\0';
1843
1844         if (shader_data->rel_offset)
1845         {
1846             const char *offset = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_VS_REL_OFFSET);
1847             shader_arb_get_src_param(ins, &ins->src[0], 0, src0_param);
1848             shader_addline(buffer, "ADD TA.x, %s, %s;\n", src0_param, offset);
1849             shader_addline(buffer, "ARL A0.x, TA.x;\n");
1850         }
1851         else
1852         {
1853             /* Apple's ARB_vertex_program implementation does not accept an ARL source argument
1854              * with more than one component. Thus replicate the first source argument over all
1855              * 4 components. For example, .xyzw -> .x (or better: .xxxx), .zwxy -> .z, etc) */
1856             struct wined3d_shader_src_param tmp_src = ins->src[0];
1857             tmp_src.swizzle = (tmp_src.swizzle & 0x3) * 0x55;
1858             shader_arb_get_src_param(ins, &tmp_src, 0, src0_param);
1859             shader_addline(buffer, "ARL A0.x, %s;\n", src0_param);
1860         }
1861     }
1862     else if (ins->dst[0].reg.type == WINED3DSPR_COLOROUT && !ins->dst[0].reg.idx[0].offset && pshader)
1863     {
1864         if (ctx->cur_ps_args->super.srgb_correction && shader->u.ps.color0_mov)
1865         {
1866             shader_addline(buffer, "#mov handled in srgb write code\n");
1867             return;
1868         }
1869         shader_hw_map2gl(ins);
1870     }
1871     else
1872     {
1873         shader_hw_map2gl(ins);
1874     }
1875 }
1876
1877 static void pshader_hw_texkill(const struct wined3d_shader_instruction *ins)
1878 {
1879     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1880     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1881     char reg_dest[40];
1882
1883     /* No swizzles are allowed in d3d's texkill. PS 1.x ignores the 4th component as documented,
1884      * but >= 2.0 honors it (undocumented, but tested by the d3d9 testsuite)
1885      */
1886     shader_arb_get_dst_param(ins, dst, reg_dest);
1887
1888     if (ins->ctx->reg_maps->shader_version.major >= 2)
1889     {
1890         const char *kilsrc = "TA";
1891         BOOL is_color;
1892
1893         shader_arb_get_register_name(ins, &dst->reg, reg_dest, &is_color);
1894         if(dst->write_mask == WINED3DSP_WRITEMASK_ALL)
1895         {
1896             kilsrc = reg_dest;
1897         }
1898         else
1899         {
1900             /* Sigh. KIL doesn't support swizzles/writemasks. KIL passes a writemask, but ".xy" for example
1901              * is not valid as a swizzle in ARB (needs ".xyyy"). Use SWZ to load the register properly, and set
1902              * masked out components to 0(won't kill)
1903              */
1904             char x = '0', y = '0', z = '0', w = '0';
1905             if(dst->write_mask & WINED3DSP_WRITEMASK_0) x = 'x';
1906             if(dst->write_mask & WINED3DSP_WRITEMASK_1) y = 'y';
1907             if(dst->write_mask & WINED3DSP_WRITEMASK_2) z = 'z';
1908             if(dst->write_mask & WINED3DSP_WRITEMASK_3) w = 'w';
1909             shader_addline(buffer, "SWZ TA, %s, %c, %c, %c, %c;\n", reg_dest, x, y, z, w);
1910         }
1911         shader_addline(buffer, "KIL %s;\n", kilsrc);
1912     }
1913     else
1914     {
1915         /* ARB fp doesn't like swizzles on the parameter of the KIL instruction. To mask the 4th component,
1916          * copy the register into our general purpose TMP variable, overwrite .w and pass TMP to KIL
1917          *
1918          * ps_1_3 shaders use the texcoord incarnation of the Tx register. ps_1_4 shaders can use the same,
1919          * or pass in any temporary register(in shader phase 2)
1920          */
1921         if (ins->ctx->reg_maps->shader_version.minor <= 3)
1922             sprintf(reg_dest, "fragment.texcoord[%u]", dst->reg.idx[0].offset);
1923         else
1924             shader_arb_get_dst_param(ins, dst, reg_dest);
1925         shader_addline(buffer, "SWZ TA, %s, x, y, z, 1;\n", reg_dest);
1926         shader_addline(buffer, "KIL TA;\n");
1927     }
1928 }
1929
1930 static void pshader_hw_tex(const struct wined3d_shader_instruction *ins)
1931 {
1932     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
1933     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1934     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
1935             ins->ctx->reg_maps->shader_version.minor);
1936     struct wined3d_shader_src_param src;
1937
1938     char reg_dest[40];
1939     char reg_coord[40];
1940     DWORD reg_sampler_code;
1941     WORD myflags = 0;
1942     BOOL swizzle_coord = FALSE;
1943
1944     /* All versions have a destination register */
1945     shader_arb_get_dst_param(ins, dst, reg_dest);
1946
1947     /* 1.0-1.4: Use destination register number as texture code.
1948        2.0+: Use provided sampler number as texure code. */
1949     if (shader_version < WINED3D_SHADER_VERSION(2,0))
1950         reg_sampler_code = dst->reg.idx[0].offset;
1951     else
1952         reg_sampler_code = ins->src[1].reg.idx[0].offset;
1953
1954     /* 1.0-1.3: Use the texcoord varying.
1955        1.4+: Use provided coordinate source register. */
1956     if (shader_version < WINED3D_SHADER_VERSION(1,4))
1957         sprintf(reg_coord, "fragment.texcoord[%u]", reg_sampler_code);
1958     else {
1959         /* TEX is the only instruction that can handle DW and DZ natively */
1960         src = ins->src[0];
1961         if(src.modifiers == WINED3DSPSM_DW) src.modifiers = WINED3DSPSM_NONE;
1962         if(src.modifiers == WINED3DSPSM_DZ) src.modifiers = WINED3DSPSM_NONE;
1963         shader_arb_get_src_param(ins, &src, 0, reg_coord);
1964     }
1965
1966     /* projection flag:
1967      * 1.1, 1.2, 1.3: Use WINED3D_TSS_TEXTURETRANSFORMFLAGS
1968      * 1.4: Use WINED3DSPSM_DZ or WINED3DSPSM_DW on src[0]
1969      * 2.0+: Use WINED3DSI_TEXLD_PROJECT on the opcode
1970      */
1971     if (shader_version < WINED3D_SHADER_VERSION(1,4))
1972     {
1973         DWORD flags = 0;
1974         if (reg_sampler_code < MAX_TEXTURES)
1975             flags = priv->cur_ps_args->super.tex_transform >> reg_sampler_code * WINED3D_PSARGS_TEXTRANSFORM_SHIFT;
1976         if (flags & WINED3D_PSARGS_PROJECTED)
1977         {
1978             myflags |= TEX_PROJ;
1979             if ((flags & ~WINED3D_PSARGS_PROJECTED) == WINED3D_TTFF_COUNT3)
1980                 swizzle_coord = TRUE;
1981         }
1982     }
1983     else if (shader_version < WINED3D_SHADER_VERSION(2,0))
1984     {
1985         enum wined3d_shader_src_modifier src_mod = ins->src[0].modifiers;
1986         if (src_mod == WINED3DSPSM_DZ)
1987         {
1988             swizzle_coord = TRUE;
1989             myflags |= TEX_PROJ;
1990         } else if(src_mod == WINED3DSPSM_DW) {
1991             myflags |= TEX_PROJ;
1992         }
1993     } else {
1994         if (ins->flags & WINED3DSI_TEXLD_PROJECT) myflags |= TEX_PROJ;
1995         if (ins->flags & WINED3DSI_TEXLD_BIAS) myflags |= TEX_BIAS;
1996     }
1997
1998     if (swizzle_coord)
1999     {
2000         /* TXP cannot handle DZ natively, so move the z coordinate to .w.
2001          * reg_coord is a read-only varying register, so we need a temp reg */
2002         shader_addline(ins->ctx->buffer, "SWZ TA, %s, x, y, z, z;\n", reg_coord);
2003         strcpy(reg_coord, "TA");
2004     }
2005
2006     shader_hw_sample(ins, reg_sampler_code, reg_dest, reg_coord, myflags, NULL, NULL);
2007 }
2008
2009 static void pshader_hw_texcoord(const struct wined3d_shader_instruction *ins)
2010 {
2011     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2012     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2013     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
2014             ins->ctx->reg_maps->shader_version.minor);
2015     char dst_str[50];
2016
2017     if (shader_version < WINED3D_SHADER_VERSION(1,4))
2018     {
2019         DWORD reg = dst->reg.idx[0].offset;
2020
2021         shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2022         shader_addline(buffer, "MOV_SAT %s, fragment.texcoord[%u];\n", dst_str, reg);
2023     } else {
2024         char reg_src[40];
2025
2026         shader_arb_get_src_param(ins, &ins->src[0], 0, reg_src);
2027         shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2028         shader_addline(buffer, "MOV %s, %s;\n", dst_str, reg_src);
2029    }
2030 }
2031
2032 static void pshader_hw_texreg2ar(const struct wined3d_shader_instruction *ins)
2033 {
2034      struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2035      DWORD flags = 0;
2036
2037      DWORD reg1 = ins->dst[0].reg.idx[0].offset;
2038      char dst_str[50];
2039      char src_str[50];
2040
2041      /* Note that texreg2ar treats Tx as a temporary register, not as a varying */
2042      shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2043      shader_arb_get_src_param(ins, &ins->src[0], 0, src_str);
2044      /* Move .x first in case src_str is "TA" */
2045      shader_addline(buffer, "MOV TA.y, %s.x;\n", src_str);
2046      shader_addline(buffer, "MOV TA.x, %s.w;\n", src_str);
2047      if (reg1 < MAX_TEXTURES)
2048      {
2049          struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2050          flags = priv->cur_ps_args->super.tex_transform >> reg1 * WINED3D_PSARGS_TEXTRANSFORM_SHIFT;
2051      }
2052      shader_hw_sample(ins, reg1, dst_str, "TA", flags & WINED3D_PSARGS_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2053 }
2054
2055 static void pshader_hw_texreg2gb(const struct wined3d_shader_instruction *ins)
2056 {
2057      struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2058
2059      DWORD reg1 = ins->dst[0].reg.idx[0].offset;
2060      char dst_str[50];
2061      char src_str[50];
2062
2063      /* Note that texreg2gb treats Tx as a temporary register, not as a varying */
2064      shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2065      shader_arb_get_src_param(ins, &ins->src[0], 0, src_str);
2066      shader_addline(buffer, "MOV TA.x, %s.y;\n", src_str);
2067      shader_addline(buffer, "MOV TA.y, %s.z;\n", src_str);
2068      shader_hw_sample(ins, reg1, dst_str, "TA", 0, NULL, NULL);
2069 }
2070
2071 static void pshader_hw_texreg2rgb(const struct wined3d_shader_instruction *ins)
2072 {
2073     DWORD reg1 = ins->dst[0].reg.idx[0].offset;
2074     char dst_str[50];
2075     char src_str[50];
2076
2077     /* Note that texreg2rg treats Tx as a temporary register, not as a varying */
2078     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2079     shader_arb_get_src_param(ins, &ins->src[0], 0, src_str);
2080     shader_hw_sample(ins, reg1, dst_str, src_str, 0, NULL, NULL);
2081 }
2082
2083 static void pshader_hw_texbem(const struct wined3d_shader_instruction *ins)
2084 {
2085     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2086     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2087     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2088     char reg_coord[40], dst_reg[50], src_reg[50];
2089     DWORD reg_dest_code;
2090
2091     /* All versions have a destination register. The Tx where the texture coordinates come
2092      * from is the varying incarnation of the texture register
2093      */
2094     reg_dest_code = dst->reg.idx[0].offset;
2095     shader_arb_get_dst_param(ins, &ins->dst[0], dst_reg);
2096     shader_arb_get_src_param(ins, &ins->src[0], 0, src_reg);
2097     sprintf(reg_coord, "fragment.texcoord[%u]", reg_dest_code);
2098
2099     /* Sampling the perturbation map in Tsrc was done already, including the signedness correction if needed
2100      * The Tx in which the perturbation map is stored is the tempreg incarnation of the texture register
2101      *
2102      * GL_NV_fragment_program_option could handle this in one instruction via X2D:
2103      * X2D TA.xy, fragment.texcoord, T%u, bumpenvmat%u.xzyw
2104      *
2105      * However, the NV extensions are never enabled for <= 2.0 shaders because of the performance penalty that
2106      * comes with it, and texbem is an 1.x only instruction. No 1.x instruction forces us to enable the NV
2107      * extension.
2108      */
2109     shader_addline(buffer, "SWZ TB, bumpenvmat%d, x, z, 0, 0;\n", reg_dest_code);
2110     shader_addline(buffer, "DP3 TA.x, TB, %s;\n", src_reg);
2111     shader_addline(buffer, "SWZ TB, bumpenvmat%d, y, w, 0, 0;\n", reg_dest_code);
2112     shader_addline(buffer, "DP3 TA.y, TB, %s;\n", src_reg);
2113
2114     /* with projective textures, texbem only divides the static texture coord, not the displacement,
2115      * so we can't let the GL handle this.
2116      */
2117     if ((priv->cur_ps_args->super.tex_transform >> reg_dest_code * WINED3D_PSARGS_TEXTRANSFORM_SHIFT)
2118             & WINED3D_PSARGS_PROJECTED)
2119     {
2120         shader_addline(buffer, "RCP TB.w, %s.w;\n", reg_coord);
2121         shader_addline(buffer, "MUL TB.xy, %s, TB.w;\n", reg_coord);
2122         shader_addline(buffer, "ADD TA.xy, TA, TB;\n");
2123     } else {
2124         shader_addline(buffer, "ADD TA.xy, TA, %s;\n", reg_coord);
2125     }
2126
2127     shader_hw_sample(ins, reg_dest_code, dst_reg, "TA", 0, NULL, NULL);
2128
2129     if (ins->handler_idx == WINED3DSIH_TEXBEML)
2130     {
2131         /* No src swizzles are allowed, so this is ok */
2132         shader_addline(buffer, "MAD TA, %s.z, luminance%d.x, luminance%d.y;\n",
2133                        src_reg, reg_dest_code, reg_dest_code);
2134         shader_addline(buffer, "MUL %s, %s, TA;\n", dst_reg, dst_reg);
2135     }
2136 }
2137
2138 static void pshader_hw_texm3x2pad(const struct wined3d_shader_instruction *ins)
2139 {
2140     DWORD reg = ins->dst[0].reg.idx[0].offset;
2141     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2142     char src0_name[50], dst_name[50];
2143     BOOL is_color;
2144     struct wined3d_shader_register tmp_reg = ins->dst[0].reg;
2145
2146     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2147     /* The next instruction will be a texm3x2tex or texm3x2depth that writes to the uninitialized
2148      * T<reg+1> register. Use this register to store the calculated vector
2149      */
2150     tmp_reg.idx[0].offset = reg + 1;
2151     shader_arb_get_register_name(ins, &tmp_reg, dst_name, &is_color);
2152     shader_addline(buffer, "DP3 %s.x, fragment.texcoord[%u], %s;\n", dst_name, reg, src0_name);
2153 }
2154
2155 static void pshader_hw_texm3x2tex(const struct wined3d_shader_instruction *ins)
2156 {
2157     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2158     DWORD flags;
2159     DWORD reg = ins->dst[0].reg.idx[0].offset;
2160     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2161     char dst_str[50];
2162     char src0_name[50];
2163     char dst_reg[50];
2164     BOOL is_color;
2165
2166     /* We know that we're writing to the uninitialized T<reg> register, so use it for temporary storage */
2167     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_reg, &is_color);
2168
2169     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2170     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2171     shader_addline(buffer, "DP3 %s.y, fragment.texcoord[%u], %s;\n", dst_reg, reg, src0_name);
2172     flags = reg < MAX_TEXTURES ? priv->cur_ps_args->super.tex_transform >> reg * WINED3D_PSARGS_TEXTRANSFORM_SHIFT : 0;
2173     shader_hw_sample(ins, reg, dst_str, dst_reg, flags & WINED3D_PSARGS_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2174 }
2175
2176 static void pshader_hw_texm3x3pad(const struct wined3d_shader_instruction *ins)
2177 {
2178     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
2179     DWORD reg = ins->dst[0].reg.idx[0].offset;
2180     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2181     char src0_name[50], dst_name[50];
2182     struct wined3d_shader_register tmp_reg = ins->dst[0].reg;
2183     BOOL is_color;
2184
2185     /* There are always 2 texm3x3pad instructions followed by one texm3x3[tex,vspec, ...] instruction, with
2186      * incrementing ins->dst[0].register_idx numbers. So the pad instruction already knows the final destination
2187      * register, and this register is uninitialized(otherwise the assembler complains that it is 'redeclared')
2188      */
2189     tmp_reg.idx[0].offset = reg + 2 - tex_mx->current_row;
2190     shader_arb_get_register_name(ins, &tmp_reg, dst_name, &is_color);
2191
2192     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2193     shader_addline(buffer, "DP3 %s.%c, fragment.texcoord[%u], %s;\n",
2194                    dst_name, 'x' + tex_mx->current_row, reg, src0_name);
2195     tex_mx->texcoord_w[tex_mx->current_row++] = reg;
2196 }
2197
2198 static void pshader_hw_texm3x3tex(const struct wined3d_shader_instruction *ins)
2199 {
2200     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2201     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
2202     DWORD flags;
2203     DWORD reg = ins->dst[0].reg.idx[0].offset;
2204     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2205     char dst_str[50];
2206     char src0_name[50], dst_name[50];
2207     BOOL is_color;
2208
2209     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2210     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2211     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_name, reg, src0_name);
2212
2213     /* Sample the texture using the calculated coordinates */
2214     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2215     flags = reg < MAX_TEXTURES ? priv->cur_ps_args->super.tex_transform >> reg * WINED3D_PSARGS_TEXTRANSFORM_SHIFT : 0;
2216     shader_hw_sample(ins, reg, dst_str, dst_name, flags & WINED3D_PSARGS_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2217     tex_mx->current_row = 0;
2218 }
2219
2220 static void pshader_hw_texm3x3vspec(const struct wined3d_shader_instruction *ins)
2221 {
2222     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2223     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
2224     DWORD flags;
2225     DWORD reg = ins->dst[0].reg.idx[0].offset;
2226     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2227     char dst_str[50];
2228     char src0_name[50];
2229     char dst_reg[50];
2230     BOOL is_color;
2231
2232     /* Get the dst reg without writemask strings. We know this register is uninitialized, so we can use all
2233      * components for temporary data storage
2234      */
2235     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_reg, &is_color);
2236     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2237     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_reg, reg, src0_name);
2238
2239     /* Construct the eye-ray vector from w coordinates */
2240     shader_addline(buffer, "MOV TB.x, fragment.texcoord[%u].w;\n", tex_mx->texcoord_w[0]);
2241     shader_addline(buffer, "MOV TB.y, fragment.texcoord[%u].w;\n", tex_mx->texcoord_w[1]);
2242     shader_addline(buffer, "MOV TB.z, fragment.texcoord[%u].w;\n", reg);
2243
2244     /* Calculate reflection vector
2245      */
2246     shader_addline(buffer, "DP3 %s.w, %s, TB;\n", dst_reg, dst_reg);
2247     /* The .w is ignored when sampling, so I can use TB.w to calculate dot(N, N) */
2248     shader_addline(buffer, "DP3 TB.w, %s, %s;\n", dst_reg, dst_reg);
2249     shader_addline(buffer, "RCP TB.w, TB.w;\n");
2250     shader_addline(buffer, "MUL %s.w, %s.w, TB.w;\n", dst_reg, dst_reg);
2251     shader_addline(buffer, "MUL %s, %s.w, %s;\n", dst_reg, dst_reg, dst_reg);
2252     shader_addline(buffer, "MAD %s, coefmul.x, %s, -TB;\n", dst_reg, dst_reg);
2253
2254     /* Sample the texture using the calculated coordinates */
2255     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2256     flags = reg < MAX_TEXTURES ? priv->cur_ps_args->super.tex_transform >> reg * WINED3D_PSARGS_TEXTRANSFORM_SHIFT : 0;
2257     shader_hw_sample(ins, reg, dst_str, dst_reg, flags & WINED3D_PSARGS_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2258     tex_mx->current_row = 0;
2259 }
2260
2261 static void pshader_hw_texm3x3spec(const struct wined3d_shader_instruction *ins)
2262 {
2263     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2264     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
2265     DWORD flags;
2266     DWORD reg = ins->dst[0].reg.idx[0].offset;
2267     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2268     char dst_str[50];
2269     char src0_name[50];
2270     char src1_name[50];
2271     char dst_reg[50];
2272     BOOL is_color;
2273
2274     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2275     shader_arb_get_src_param(ins, &ins->src[0], 1, src1_name);
2276     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_reg, &is_color);
2277     /* Note: dst_reg.xy is input here, generated by two texm3x3pad instructions */
2278     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_reg, reg, src0_name);
2279
2280     /* Calculate reflection vector.
2281      *
2282      *                   dot(N, E)
2283      * dst_reg.xyz = 2 * --------- * N - E
2284      *                   dot(N, N)
2285      *
2286      * Which normalizes the normal vector
2287      */
2288     shader_addline(buffer, "DP3 %s.w, %s, %s;\n", dst_reg, dst_reg, src1_name);
2289     shader_addline(buffer, "DP3 TC.w, %s, %s;\n", dst_reg, dst_reg);
2290     shader_addline(buffer, "RCP TC.w, TC.w;\n");
2291     shader_addline(buffer, "MUL %s.w, %s.w, TC.w;\n", dst_reg, dst_reg);
2292     shader_addline(buffer, "MUL %s, %s.w, %s;\n", dst_reg, dst_reg, dst_reg);
2293     shader_addline(buffer, "MAD %s, coefmul.x, %s, -%s;\n", dst_reg, dst_reg, src1_name);
2294
2295     /* Sample the texture using the calculated coordinates */
2296     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2297     flags = reg < MAX_TEXTURES ? priv->cur_ps_args->super.tex_transform >> reg * WINED3D_PSARGS_TEXTRANSFORM_SHIFT : 0;
2298     shader_hw_sample(ins, reg, dst_str, dst_reg, flags & WINED3D_PSARGS_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2299     tex_mx->current_row = 0;
2300 }
2301
2302 static void pshader_hw_texdepth(const struct wined3d_shader_instruction *ins)
2303 {
2304     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2305     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2306     char dst_name[50];
2307     const char *zero = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ZERO);
2308     const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
2309
2310     /* texdepth has an implicit destination, the fragment depth value. It's only parameter,
2311      * which is essentially an input, is the destination register because it is the first
2312      * parameter. According to the msdn, this must be register r5, but let's keep it more flexible
2313      * here(writemasks/swizzles are not valid on texdepth)
2314      */
2315     shader_arb_get_dst_param(ins, dst, dst_name);
2316
2317     /* According to the msdn, the source register(must be r5) is unusable after
2318      * the texdepth instruction, so we're free to modify it
2319      */
2320     shader_addline(buffer, "MIN %s.y, %s.y, %s;\n", dst_name, dst_name, one);
2321
2322     /* How to deal with the special case dst_name.g == 0? if r != 0, then
2323      * the r * (1 / 0) will give infinity, which is clamped to 1.0, the correct
2324      * result. But if r = 0.0, then 0 * inf = 0, which is incorrect.
2325      */
2326     shader_addline(buffer, "RCP %s.y, %s.y;\n", dst_name, dst_name);
2327     shader_addline(buffer, "MUL TA.x, %s.x, %s.y;\n", dst_name, dst_name);
2328     shader_addline(buffer, "MIN TA.x, TA.x, %s;\n", one);
2329     shader_addline(buffer, "MAX result.depth, TA.x, %s;\n", zero);
2330 }
2331
2332 /** Process the WINED3DSIO_TEXDP3TEX instruction in ARB:
2333  * Take a 3-component dot product of the TexCoord[dstreg] and src,
2334  * then perform a 1D texture lookup from stage dstregnum, place into dst. */
2335 static void pshader_hw_texdp3tex(const struct wined3d_shader_instruction *ins)
2336 {
2337     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2338     DWORD sampler_idx = ins->dst[0].reg.idx[0].offset;
2339     char src0[50];
2340     char dst_str[50];
2341
2342     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2343     shader_addline(buffer, "MOV TB, 0.0;\n");
2344     shader_addline(buffer, "DP3 TB.x, fragment.texcoord[%u], %s;\n", sampler_idx, src0);
2345
2346     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2347     shader_hw_sample(ins, sampler_idx, dst_str, "TB", 0 /* Only one coord, can't be projected */, NULL, NULL);
2348 }
2349
2350 /** Process the WINED3DSIO_TEXDP3 instruction in ARB:
2351  * Take a 3-component dot product of the TexCoord[dstreg] and src. */
2352 static void pshader_hw_texdp3(const struct wined3d_shader_instruction *ins)
2353 {
2354     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2355     char src0[50];
2356     char dst_str[50];
2357     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2358
2359     /* Handle output register */
2360     shader_arb_get_dst_param(ins, dst, dst_str);
2361     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2362     shader_addline(buffer, "DP3 %s, fragment.texcoord[%u], %s;\n", dst_str, dst->reg.idx[0].offset, src0);
2363 }
2364
2365 /** Process the WINED3DSIO_TEXM3X3 instruction in ARB
2366  * Perform the 3rd row of a 3x3 matrix multiply */
2367 static void pshader_hw_texm3x3(const struct wined3d_shader_instruction *ins)
2368 {
2369     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2370     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2371     char dst_str[50], dst_name[50];
2372     char src0[50];
2373     BOOL is_color;
2374
2375     shader_arb_get_dst_param(ins, dst, dst_str);
2376     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2377     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2378     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_name, dst->reg.idx[0].offset, src0);
2379     shader_addline(buffer, "MOV %s, %s;\n", dst_str, dst_name);
2380 }
2381
2382 /** Process the WINED3DSIO_TEXM3X2DEPTH instruction in ARB:
2383  * Last row of a 3x2 matrix multiply, use the result to calculate the depth:
2384  * Calculate tmp0.y = TexCoord[dstreg] . src.xyz;  (tmp0.x has already been calculated)
2385  * depth = (tmp0.y == 0.0) ? 1.0 : tmp0.x / tmp0.y
2386  */
2387 static void pshader_hw_texm3x2depth(const struct wined3d_shader_instruction *ins)
2388 {
2389     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2390     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2391     char src0[50], dst_name[50];
2392     BOOL is_color;
2393     const char *zero = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ZERO);
2394     const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
2395
2396     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2397     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2398     shader_addline(buffer, "DP3 %s.y, fragment.texcoord[%u], %s;\n", dst_name, dst->reg.idx[0].offset, src0);
2399
2400     /* How to deal with the special case dst_name.g == 0? if r != 0, then
2401      * the r * (1 / 0) will give infinity, which is clamped to 1.0, the correct
2402      * result. But if r = 0.0, then 0 * inf = 0, which is incorrect.
2403      */
2404     shader_addline(buffer, "RCP %s.y, %s.y;\n", dst_name, dst_name);
2405     shader_addline(buffer, "MUL %s.x, %s.x, %s.y;\n", dst_name, dst_name, dst_name);
2406     shader_addline(buffer, "MIN %s.x, %s.x, %s;\n", dst_name, dst_name, one);
2407     shader_addline(buffer, "MAX result.depth, %s.x, %s;\n", dst_name, zero);
2408 }
2409
2410 /** Handles transforming all WINED3DSIO_M?x? opcodes for
2411     Vertex/Pixel shaders to ARB_vertex_program codes */
2412 static void shader_hw_mnxn(const struct wined3d_shader_instruction *ins)
2413 {
2414     int i;
2415     int nComponents = 0;
2416     struct wined3d_shader_dst_param tmp_dst = {{0}};
2417     struct wined3d_shader_src_param tmp_src[2] = {{{0}}};
2418     struct wined3d_shader_instruction tmp_ins;
2419
2420     memset(&tmp_ins, 0, sizeof(tmp_ins));
2421
2422     /* Set constants for the temporary argument */
2423     tmp_ins.ctx = ins->ctx;
2424     tmp_ins.dst_count = 1;
2425     tmp_ins.dst = &tmp_dst;
2426     tmp_ins.src_count = 2;
2427     tmp_ins.src = tmp_src;
2428
2429     switch(ins->handler_idx)
2430     {
2431         case WINED3DSIH_M4x4:
2432             nComponents = 4;
2433             tmp_ins.handler_idx = WINED3DSIH_DP4;
2434             break;
2435         case WINED3DSIH_M4x3:
2436             nComponents = 3;
2437             tmp_ins.handler_idx = WINED3DSIH_DP4;
2438             break;
2439         case WINED3DSIH_M3x4:
2440             nComponents = 4;
2441             tmp_ins.handler_idx = WINED3DSIH_DP3;
2442             break;
2443         case WINED3DSIH_M3x3:
2444             nComponents = 3;
2445             tmp_ins.handler_idx = WINED3DSIH_DP3;
2446             break;
2447         case WINED3DSIH_M3x2:
2448             nComponents = 2;
2449             tmp_ins.handler_idx = WINED3DSIH_DP3;
2450             break;
2451         default:
2452             FIXME("Unhandled opcode %#x\n", ins->handler_idx);
2453             break;
2454     }
2455
2456     tmp_dst = ins->dst[0];
2457     tmp_src[0] = ins->src[0];
2458     tmp_src[1] = ins->src[1];
2459     for (i = 0; i < nComponents; ++i)
2460     {
2461         tmp_dst.write_mask = WINED3DSP_WRITEMASK_0 << i;
2462         shader_hw_map2gl(&tmp_ins);
2463         ++tmp_src[1].reg.idx[0].offset;
2464     }
2465 }
2466
2467 static void shader_hw_rcp(const struct wined3d_shader_instruction *ins)
2468 {
2469     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2470
2471     char dst[50];
2472     char src[50];
2473
2474     shader_arb_get_dst_param(ins, &ins->dst[0], dst); /* Destination */
2475     shader_arb_get_src_param(ins, &ins->src[0], 0, src);
2476     if (ins->src[0].swizzle == WINED3DSP_NOSWIZZLE)
2477     {
2478         /* Dx sdk says .x is used if no swizzle is given, but our test shows that
2479          * .w is used
2480          */
2481         strcat(src, ".w");
2482     }
2483
2484     shader_addline(buffer, "RCP%s %s, %s;\n", shader_arb_get_modifier(ins), dst, src);
2485 }
2486
2487 static void shader_hw_scalar_op(const struct wined3d_shader_instruction *ins)
2488 {
2489     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2490     const char *instruction;
2491
2492     char dst[50];
2493     char src[50];
2494
2495     switch(ins->handler_idx)
2496     {
2497         case WINED3DSIH_RSQ:  instruction = "RSQ"; break;
2498         case WINED3DSIH_RCP:  instruction = "RCP"; break;
2499         case WINED3DSIH_EXP:  instruction = "EX2"; break;
2500         case WINED3DSIH_EXPP: instruction = "EXP"; break;
2501         default: instruction = "";
2502             FIXME("Unhandled opcode %#x\n", ins->handler_idx);
2503             break;
2504     }
2505
2506     shader_arb_get_dst_param(ins, &ins->dst[0], dst); /* Destination */
2507     shader_arb_get_src_param(ins, &ins->src[0], 0, src);
2508     if (ins->src[0].swizzle == WINED3DSP_NOSWIZZLE)
2509     {
2510         /* Dx sdk says .x is used if no swizzle is given, but our test shows that
2511          * .w is used
2512          */
2513         strcat(src, ".w");
2514     }
2515
2516     shader_addline(buffer, "%s%s %s, %s;\n", instruction, shader_arb_get_modifier(ins), dst, src);
2517 }
2518
2519 static void shader_hw_nrm(const struct wined3d_shader_instruction *ins)
2520 {
2521     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2522     char dst_name[50];
2523     char src_name[50];
2524     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2525     BOOL pshader = shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type);
2526     const char *zero = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ZERO);
2527
2528     shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2529     shader_arb_get_src_param(ins, &ins->src[0], 1 /* Use TB */, src_name);
2530
2531     /* In D3D, NRM of a vector with length zero returns zero. Catch this situation, as
2532      * otherwise NRM or RSQ would return NaN */
2533     if(pshader && priv->target_version >= NV3)
2534     {
2535         /* GL_NV_fragment_program2's NRM needs protection against length zero vectors too
2536          *
2537          * TODO: Find out if DP3+NRM+MOV is really faster than DP3+RSQ+MUL
2538          */
2539         shader_addline(buffer, "DP3C TA, %s, %s;\n", src_name, src_name);
2540         shader_addline(buffer, "NRM%s %s, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name);
2541         shader_addline(buffer, "MOV %s (EQ), %s;\n", dst_name, zero);
2542     }
2543     else if(priv->target_version >= NV2)
2544     {
2545         shader_addline(buffer, "DP3C TA.x, %s, %s;\n", src_name, src_name);
2546         shader_addline(buffer, "RSQ TA.x (NE), TA.x;\n");
2547         shader_addline(buffer, "MUL%s %s, %s, TA.x;\n", shader_arb_get_modifier(ins), dst_name,
2548                        src_name);
2549     }
2550     else
2551     {
2552         const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
2553
2554         shader_addline(buffer, "DP3 TA.x, %s, %s;\n", src_name, src_name);
2555         /* Pass any non-zero value to RSQ if the input vector has a length of zero. The
2556          * RSQ result doesn't matter, as long as multiplying it by 0 returns 0.
2557          */
2558         shader_addline(buffer, "SGE TA.y, -TA.x, %s;\n", zero);
2559         shader_addline(buffer, "MAD TA.x, %s, TA.y, TA.x;\n", one);
2560
2561         shader_addline(buffer, "RSQ TA.x, TA.x;\n");
2562         /* dst.w = src[0].w * 1 / (src.x^2 + src.y^2 + src.z^2)^(1/2) according to msdn*/
2563         shader_addline(buffer, "MUL%s %s, %s, TA.x;\n", shader_arb_get_modifier(ins), dst_name,
2564                     src_name);
2565     }
2566 }
2567
2568 static void shader_hw_lrp(const struct wined3d_shader_instruction *ins)
2569 {
2570     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2571     char dst_name[50];
2572     char src_name[3][50];
2573
2574     /* ARB_fragment_program has a convenient LRP instruction */
2575     if(shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type)) {
2576         shader_hw_map2gl(ins);
2577         return;
2578     }
2579
2580     shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2581     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
2582     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
2583     shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
2584
2585     shader_addline(buffer, "SUB TA, %s, %s;\n", src_name[1], src_name[2]);
2586     shader_addline(buffer, "MAD%s %s, %s, TA, %s;\n", shader_arb_get_modifier(ins),
2587                    dst_name, src_name[0], src_name[2]);
2588 }
2589
2590 static void shader_hw_sincos(const struct wined3d_shader_instruction *ins)
2591 {
2592     /* This instruction exists in ARB, but the d3d instruction takes two extra parameters which
2593      * must contain fixed constants. So we need a separate function to filter those constants and
2594      * can't use map2gl
2595      */
2596     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2597     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2598     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2599     char dst_name[50];
2600     char src_name0[50], src_name1[50], src_name2[50];
2601     BOOL is_color;
2602
2603     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name0);
2604     if(shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type)) {
2605         shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2606         /* No modifiers are supported on SCS */
2607         shader_addline(buffer, "SCS %s, %s;\n", dst_name, src_name0);
2608
2609         if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE)
2610         {
2611             shader_arb_get_register_name(ins, &dst->reg, src_name0, &is_color);
2612             shader_addline(buffer, "MOV_SAT %s, %s;\n", dst_name, src_name0);
2613         }
2614     } else if(priv->target_version >= NV2) {
2615         shader_arb_get_register_name(ins, &dst->reg, dst_name, &is_color);
2616
2617         /* Sincos writemask must be .x, .y or .xy */
2618         if(dst->write_mask & WINED3DSP_WRITEMASK_0)
2619             shader_addline(buffer, "COS%s %s.x, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name0);
2620         if(dst->write_mask & WINED3DSP_WRITEMASK_1)
2621             shader_addline(buffer, "SIN%s %s.y, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name0);
2622     } else {
2623         /* Approximate sine and cosine with a taylor series, as per math textbook. The application passes 8
2624          * helper constants(D3DSINCOSCONST1 and D3DSINCOSCONST2) in src1 and src2.
2625          *
2626          * sin(x) = x - x^3/3! + x^5/5! - x^7/7! + ...
2627          * cos(x) = 1 - x^2/2! + x^4/4! - x^6/6! + ...
2628          *
2629          * The constants we get are:
2630          *
2631          *  +1   +1,     -1     -1     +1      +1      -1       -1
2632          *      ---- ,  ---- , ---- , ----- , ----- , ----- , ------
2633          *      1!*2    2!*4   3!*8   4!*16   5!*32   6!*64   7!*128
2634          *
2635          * If used with x^2, x^3, x^4 etc they calculate sin(x/2) and cos(x/2):
2636          *
2637          * (x/2)^2 = x^2 / 4
2638          * (x/2)^3 = x^3 / 8
2639          * (x/2)^4 = x^4 / 16
2640          * (x/2)^5 = x^5 / 32
2641          * etc
2642          *
2643          * To get the final result:
2644          * sin(x) = 2 * sin(x/2) * cos(x/2)
2645          * cos(x) = cos(x/2)^2 - sin(x/2)^2
2646          * (from sin(x+y) and cos(x+y) rules)
2647          *
2648          * As per MSDN, dst.z is undefined after the operation, and so is
2649          * dst.x and dst.y if they're masked out by the writemask. Ie
2650          * sincos dst.y, src1, c0, c1
2651          * returns the sine in dst.y. dst.x and dst.z are undefined, dst.w is not touched. The assembler
2652          * vsa.exe also stops with an error if the dest register is the same register as the source
2653          * register. This means we can use dest.xyz as temporary storage. The assembler vsa.exe output also
2654          * indicates that sincos consumes 8 instruction slots in vs_2_0(and, strangely, in vs_3_0).
2655          */
2656         shader_arb_get_src_param(ins, &ins->src[1], 1, src_name1);
2657         shader_arb_get_src_param(ins, &ins->src[2], 2, src_name2);
2658         shader_arb_get_register_name(ins, &dst->reg, dst_name, &is_color);
2659
2660         shader_addline(buffer, "MUL %s.x, %s, %s;\n", dst_name, src_name0, src_name0);  /* x ^ 2 */
2661         shader_addline(buffer, "MUL TA.y, %s.x, %s;\n", dst_name, src_name0);           /* x ^ 3 */
2662         shader_addline(buffer, "MUL %s.y, TA.y, %s;\n", dst_name, src_name0);           /* x ^ 4 */
2663         shader_addline(buffer, "MUL TA.z, %s.y, %s;\n", dst_name, src_name0);           /* x ^ 5 */
2664         shader_addline(buffer, "MUL %s.z, TA.z, %s;\n", dst_name, src_name0);           /* x ^ 6 */
2665         shader_addline(buffer, "MUL TA.w, %s.z, %s;\n", dst_name, src_name0);           /* x ^ 7 */
2666
2667         /* sin(x/2)
2668          *
2669          * Unfortunately we don't get the constants in a DP4-capable form. Is there a way to
2670          * properly merge that with MULs in the code above?
2671          * The swizzles .yz and xw however fit into the .yzxw swizzle added to ps_2_0. Maybe
2672          * we can merge the sine and cosine MAD rows to calculate them together.
2673          */
2674         shader_addline(buffer, "MUL TA.x, %s, %s.w;\n", src_name0, src_name2); /* x^1, +1/(1!*2) */
2675         shader_addline(buffer, "MAD TA.x, TA.y, %s.x, TA.x;\n", src_name2); /* -1/(3!*8) */
2676         shader_addline(buffer, "MAD TA.x, TA.z, %s.w, TA.x;\n", src_name1); /* +1/(5!*32) */
2677         shader_addline(buffer, "MAD TA.x, TA.w, %s.x, TA.x;\n", src_name1); /* -1/(7!*128) */
2678
2679         /* cos(x/2) */
2680         shader_addline(buffer, "MAD TA.y, %s.x, %s.y, %s.z;\n", dst_name, src_name2, src_name2); /* -1/(2!*4), +1.0 */
2681         shader_addline(buffer, "MAD TA.y, %s.y, %s.z, TA.y;\n", dst_name, src_name1); /* +1/(4!*16) */
2682         shader_addline(buffer, "MAD TA.y, %s.z, %s.y, TA.y;\n", dst_name, src_name1); /* -1/(6!*64) */
2683
2684         if(dst->write_mask & WINED3DSP_WRITEMASK_0) {
2685             /* cos x */
2686             shader_addline(buffer, "MUL TA.z, TA.y, TA.y;\n");
2687             shader_addline(buffer, "MAD %s.x, -TA.x, TA.x, TA.z;\n", dst_name);
2688         }
2689         if(dst->write_mask & WINED3DSP_WRITEMASK_1) {
2690             /* sin x */
2691             shader_addline(buffer, "MUL %s.y, TA.x, TA.y;\n", dst_name);
2692             shader_addline(buffer, "ADD %s.y, %s.y, %s.y;\n", dst_name, dst_name, dst_name);
2693         }
2694     }
2695 }
2696
2697 static void shader_hw_sgn(const struct wined3d_shader_instruction *ins)
2698 {
2699     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2700     char dst_name[50];
2701     char src_name[50];
2702     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
2703
2704     shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2705     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name);
2706
2707     /* SGN is only valid in vertex shaders */
2708     if(ctx->target_version >= NV2) {
2709         shader_addline(buffer, "SSG%s %s, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name);
2710         return;
2711     }
2712
2713     /* If SRC > 0.0, -SRC < SRC = TRUE, otherwise false.
2714      * if SRC < 0.0,  SRC < -SRC = TRUE. If neither is true, src = 0.0
2715      */
2716     if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE) {
2717         shader_addline(buffer, "SLT %s, -%s, %s;\n", dst_name, src_name, src_name);
2718     } else {
2719         /* src contains TA? Write to the dest first. This won't overwrite our destination.
2720          * Then use TA, and calculate the final result
2721          *
2722          * Not reading from TA? Store the first result in TA to avoid overwriting the
2723          * destination if src reg = dst reg
2724          */
2725         if(strstr(src_name, "TA"))
2726         {
2727             shader_addline(buffer, "SLT %s,  %s, -%s;\n", dst_name, src_name, src_name);
2728             shader_addline(buffer, "SLT TA, -%s, %s;\n", src_name, src_name);
2729             shader_addline(buffer, "ADD %s, %s, -TA;\n", dst_name, dst_name);
2730         }
2731         else
2732         {
2733             shader_addline(buffer, "SLT TA, -%s, %s;\n", src_name, src_name);
2734             shader_addline(buffer, "SLT %s,  %s, -%s;\n", dst_name, src_name, src_name);
2735             shader_addline(buffer, "ADD %s, TA, -%s;\n", dst_name, dst_name);
2736         }
2737     }
2738 }
2739
2740 static void shader_hw_dsy(const struct wined3d_shader_instruction *ins)
2741 {
2742     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2743     char src[50];
2744     char dst[50];
2745     char dst_name[50];
2746     BOOL is_color;
2747
2748     shader_arb_get_dst_param(ins, &ins->dst[0], dst);
2749     shader_arb_get_src_param(ins, &ins->src[0], 0, src);
2750     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2751
2752     shader_addline(buffer, "DDY %s, %s;\n", dst, src);
2753     shader_addline(buffer, "MUL%s %s, %s, ycorrection.y;\n", shader_arb_get_modifier(ins), dst, dst_name);
2754 }
2755
2756 static DWORD abs_modifier(DWORD mod, BOOL *need_abs)
2757 {
2758     *need_abs = FALSE;
2759
2760     switch(mod)
2761     {
2762         case WINED3DSPSM_NONE:      return WINED3DSPSM_ABS;
2763         case WINED3DSPSM_NEG:       return WINED3DSPSM_ABS;
2764         case WINED3DSPSM_BIAS:      *need_abs = TRUE; return WINED3DSPSM_BIAS;
2765         case WINED3DSPSM_BIASNEG:   *need_abs = TRUE; return WINED3DSPSM_BIASNEG;
2766         case WINED3DSPSM_SIGN:      *need_abs = TRUE; return WINED3DSPSM_SIGN;
2767         case WINED3DSPSM_SIGNNEG:   *need_abs = TRUE; return WINED3DSPSM_SIGNNEG;
2768         case WINED3DSPSM_COMP:      *need_abs = TRUE; return WINED3DSPSM_COMP;
2769         case WINED3DSPSM_X2:        *need_abs = TRUE; return WINED3DSPSM_X2;
2770         case WINED3DSPSM_X2NEG:     *need_abs = TRUE; return WINED3DSPSM_X2NEG;
2771         case WINED3DSPSM_DZ:        *need_abs = TRUE; return WINED3DSPSM_DZ;
2772         case WINED3DSPSM_DW:        *need_abs = TRUE; return WINED3DSPSM_DW;
2773         case WINED3DSPSM_ABS:       return WINED3DSPSM_ABS;
2774         case WINED3DSPSM_ABSNEG:    return WINED3DSPSM_ABS;
2775     }
2776     FIXME("Unknown modifier %u\n", mod);
2777     return mod;
2778 }
2779
2780 static void shader_hw_log(const struct wined3d_shader_instruction *ins)
2781 {
2782     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2783     char src0[50], dst[50];
2784     struct wined3d_shader_src_param src0_copy = ins->src[0];
2785     BOOL need_abs = FALSE;
2786     const char *instr;
2787
2788     switch(ins->handler_idx)
2789     {
2790         case WINED3DSIH_LOG:  instr = "LG2"; break;
2791         case WINED3DSIH_LOGP: instr = "LOG"; break;
2792         default:
2793             ERR("Unexpected instruction %d\n", ins->handler_idx);
2794             return;
2795     }
2796
2797     /* LOG and LOGP operate on the absolute value of the input */
2798     src0_copy.modifiers = abs_modifier(src0_copy.modifiers, &need_abs);
2799
2800     shader_arb_get_dst_param(ins, &ins->dst[0], dst);
2801     shader_arb_get_src_param(ins, &src0_copy, 0, src0);
2802
2803     if(need_abs)
2804     {
2805         shader_addline(buffer, "ABS TA, %s;\n", src0);
2806         shader_addline(buffer, "%s%s %s, TA;\n", instr, shader_arb_get_modifier(ins), dst);
2807     }
2808     else
2809     {
2810         shader_addline(buffer, "%s%s %s, %s;\n", instr, shader_arb_get_modifier(ins), dst, src0);
2811     }
2812 }
2813
2814 static void shader_hw_pow(const struct wined3d_shader_instruction *ins)
2815 {
2816     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2817     char src0[50], src1[50], dst[50];
2818     struct wined3d_shader_src_param src0_copy = ins->src[0];
2819     BOOL need_abs = FALSE;
2820     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2821     const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
2822
2823     /* POW operates on the absolute value of the input */
2824     src0_copy.modifiers = abs_modifier(src0_copy.modifiers, &need_abs);
2825
2826     shader_arb_get_dst_param(ins, &ins->dst[0], dst);
2827     shader_arb_get_src_param(ins, &src0_copy, 0, src0);
2828     shader_arb_get_src_param(ins, &ins->src[1], 1, src1);
2829
2830     if (need_abs)
2831         shader_addline(buffer, "ABS TA.x, %s;\n", src0);
2832     else
2833         shader_addline(buffer, "MOV TA.x, %s;\n", src0);
2834
2835     if (priv->target_version >= NV2)
2836     {
2837         shader_addline(buffer, "MOVC TA.y, %s;\n", src1);
2838         shader_addline(buffer, "POW%s %s, TA.x, TA.y;\n", shader_arb_get_modifier(ins), dst);
2839         shader_addline(buffer, "MOV %s (EQ.y), %s;\n", dst, one);
2840     }
2841     else
2842     {
2843         const char *zero = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ZERO);
2844         const char *flt_eps = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_EPS);
2845
2846         shader_addline(buffer, "ABS TA.y, %s;\n", src1);
2847         shader_addline(buffer, "SGE TA.y, -TA.y, %s;\n", zero);
2848         /* Possibly add flt_eps to avoid getting float special values */
2849         shader_addline(buffer, "MAD TA.z, TA.y, %s, %s;\n", flt_eps, src1);
2850         shader_addline(buffer, "POW%s TA.x, TA.x, TA.z;\n", shader_arb_get_modifier(ins));
2851         shader_addline(buffer, "MAD TA.x, -TA.x, TA.y, TA.x;\n");
2852         shader_addline(buffer, "MAD %s, TA.y, %s, TA.x;\n", dst, one);
2853     }
2854 }
2855
2856 static void shader_hw_loop(const struct wined3d_shader_instruction *ins)
2857 {
2858     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2859     char src_name[50];
2860     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2861
2862     /* src0 is aL */
2863     shader_arb_get_src_param(ins, &ins->src[1], 0, src_name);
2864
2865     if(vshader)
2866     {
2867         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2868         struct list *e = list_head(&priv->control_frames);
2869         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2870
2871         if(priv->loop_depth > 1) shader_addline(buffer, "PUSHA aL;\n");
2872         /* The constant loader makes sure to load -1 into iX.w */
2873         shader_addline(buffer, "ARLC aL, %s.xywz;\n", src_name);
2874         shader_addline(buffer, "BRA loop_%u_end (LE.x);\n", control_frame->no.loop);
2875         shader_addline(buffer, "loop_%u_start:\n", control_frame->no.loop);
2876     }
2877     else
2878     {
2879         shader_addline(buffer, "LOOP %s;\n", src_name);
2880     }
2881 }
2882
2883 static void shader_hw_rep(const struct wined3d_shader_instruction *ins)
2884 {
2885     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2886     char src_name[50];
2887     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2888
2889     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name);
2890
2891     /* The constant loader makes sure to load -1 into iX.w */
2892     if(vshader)
2893     {
2894         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2895         struct list *e = list_head(&priv->control_frames);
2896         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2897
2898         if(priv->loop_depth > 1) shader_addline(buffer, "PUSHA aL;\n");
2899
2900         shader_addline(buffer, "ARLC aL, %s.xywz;\n", src_name);
2901         shader_addline(buffer, "BRA loop_%u_end (LE.x);\n", control_frame->no.loop);
2902         shader_addline(buffer, "loop_%u_start:\n", control_frame->no.loop);
2903     }
2904     else
2905     {
2906         shader_addline(buffer, "REP %s;\n", src_name);
2907     }
2908 }
2909
2910 static void shader_hw_endloop(const struct wined3d_shader_instruction *ins)
2911 {
2912     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2913     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2914
2915     if(vshader)
2916     {
2917         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2918         struct list *e = list_head(&priv->control_frames);
2919         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2920
2921         shader_addline(buffer, "ARAC aL.xy, aL;\n");
2922         shader_addline(buffer, "BRA loop_%u_start (GT.x);\n", control_frame->no.loop);
2923         shader_addline(buffer, "loop_%u_end:\n", control_frame->no.loop);
2924
2925         if(priv->loop_depth > 1) shader_addline(buffer, "POPA aL;\n");
2926     }
2927     else
2928     {
2929         shader_addline(buffer, "ENDLOOP;\n");
2930     }
2931 }
2932
2933 static void shader_hw_endrep(const struct wined3d_shader_instruction *ins)
2934 {
2935     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2936     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2937
2938     if(vshader)
2939     {
2940         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2941         struct list *e = list_head(&priv->control_frames);
2942         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2943
2944         shader_addline(buffer, "ARAC aL.xy, aL;\n");
2945         shader_addline(buffer, "BRA loop_%u_start (GT.x);\n", control_frame->no.loop);
2946         shader_addline(buffer, "loop_%u_end:\n", control_frame->no.loop);
2947
2948         if(priv->loop_depth > 1) shader_addline(buffer, "POPA aL;\n");
2949     }
2950     else
2951     {
2952         shader_addline(buffer, "ENDREP;\n");
2953     }
2954 }
2955
2956 static const struct control_frame *find_last_loop(const struct shader_arb_ctx_priv *priv)
2957 {
2958     struct control_frame *control_frame;
2959
2960     LIST_FOR_EACH_ENTRY(control_frame, &priv->control_frames, struct control_frame, entry)
2961     {
2962         if(control_frame->type == LOOP || control_frame->type == REP) return control_frame;
2963     }
2964     ERR("Could not find loop for break\n");
2965     return NULL;
2966 }
2967
2968 static void shader_hw_break(const struct wined3d_shader_instruction *ins)
2969 {
2970     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2971     const struct control_frame *control_frame = find_last_loop(ins->ctx->backend_data);
2972     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2973
2974     if(vshader)
2975     {
2976         shader_addline(buffer, "BRA loop_%u_end;\n", control_frame->no.loop);
2977     }
2978     else
2979     {
2980         shader_addline(buffer, "BRK;\n");
2981     }
2982 }
2983
2984 static const char *get_compare(enum wined3d_shader_rel_op op)
2985 {
2986     switch (op)
2987     {
2988         case WINED3D_SHADER_REL_OP_GT: return "GT";
2989         case WINED3D_SHADER_REL_OP_EQ: return "EQ";
2990         case WINED3D_SHADER_REL_OP_GE: return "GE";
2991         case WINED3D_SHADER_REL_OP_LT: return "LT";
2992         case WINED3D_SHADER_REL_OP_NE: return "NE";
2993         case WINED3D_SHADER_REL_OP_LE: return "LE";
2994         default:
2995             FIXME("Unrecognized operator %#x.\n", op);
2996             return "(\?\?)";
2997     }
2998 }
2999
3000 static enum wined3d_shader_rel_op invert_compare(enum wined3d_shader_rel_op op)
3001 {
3002     switch (op)
3003     {
3004         case WINED3D_SHADER_REL_OP_GT: return WINED3D_SHADER_REL_OP_LE;
3005         case WINED3D_SHADER_REL_OP_EQ: return WINED3D_SHADER_REL_OP_NE;
3006         case WINED3D_SHADER_REL_OP_GE: return WINED3D_SHADER_REL_OP_LT;
3007         case WINED3D_SHADER_REL_OP_LT: return WINED3D_SHADER_REL_OP_GE;
3008         case WINED3D_SHADER_REL_OP_NE: return WINED3D_SHADER_REL_OP_EQ;
3009         case WINED3D_SHADER_REL_OP_LE: return WINED3D_SHADER_REL_OP_GT;
3010         default:
3011             FIXME("Unrecognized operator %#x.\n", op);
3012             return -1;
3013     }
3014 }
3015
3016 static void shader_hw_breakc(const struct wined3d_shader_instruction *ins)
3017 {
3018     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3019     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
3020     const struct control_frame *control_frame = find_last_loop(ins->ctx->backend_data);
3021     char src_name0[50];
3022     char src_name1[50];
3023     const char *comp = get_compare(ins->flags);
3024
3025     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name0);
3026     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name1);
3027
3028     if(vshader)
3029     {
3030         /* SUBC CC, src0, src1" works only in pixel shaders, so use TA to throw
3031          * away the subtraction result
3032          */
3033         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
3034         shader_addline(buffer, "BRA loop_%u_end (%s.x);\n", control_frame->no.loop, comp);
3035     }
3036     else
3037     {
3038         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
3039         shader_addline(buffer, "BRK (%s.x);\n", comp);
3040     }
3041 }
3042
3043 static void shader_hw_ifc(const struct wined3d_shader_instruction *ins)
3044 {
3045     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3046     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3047     struct list *e = list_head(&priv->control_frames);
3048     struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
3049     const char *comp;
3050     char src_name0[50];
3051     char src_name1[50];
3052     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
3053
3054     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name0);
3055     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name1);
3056
3057     if(vshader)
3058     {
3059         /* Invert the flag. We jump to the else label if the condition is NOT true */
3060         comp = get_compare(invert_compare(ins->flags));
3061         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
3062         shader_addline(buffer, "BRA ifc_%u_else (%s.x);\n", control_frame->no.ifc, comp);
3063     }
3064     else
3065     {
3066         comp = get_compare(ins->flags);
3067         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
3068         shader_addline(buffer, "IF %s.x;\n", comp);
3069     }
3070 }
3071
3072 static void shader_hw_else(const struct wined3d_shader_instruction *ins)
3073 {
3074     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3075     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3076     struct list *e = list_head(&priv->control_frames);
3077     struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
3078     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
3079
3080     if(vshader)
3081     {
3082         shader_addline(buffer, "BRA ifc_%u_endif;\n", control_frame->no.ifc);
3083         shader_addline(buffer, "ifc_%u_else:\n", control_frame->no.ifc);
3084         control_frame->had_else = TRUE;
3085     }
3086     else
3087     {
3088         shader_addline(buffer, "ELSE;\n");
3089     }
3090 }
3091
3092 static void shader_hw_endif(const struct wined3d_shader_instruction *ins)
3093 {
3094     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3095     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3096     struct list *e = list_head(&priv->control_frames);
3097     struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
3098     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
3099
3100     if(vshader)
3101     {
3102         if(control_frame->had_else)
3103         {
3104             shader_addline(buffer, "ifc_%u_endif:\n", control_frame->no.ifc);
3105         }
3106         else
3107         {
3108             shader_addline(buffer, "#No else branch. else is endif\n");
3109             shader_addline(buffer, "ifc_%u_else:\n", control_frame->no.ifc);
3110         }
3111     }
3112     else
3113     {
3114         shader_addline(buffer, "ENDIF;\n");
3115     }
3116 }
3117
3118 static void shader_hw_texldd(const struct wined3d_shader_instruction *ins)
3119 {
3120     DWORD sampler_idx = ins->src[1].reg.idx[0].offset;
3121     char reg_dest[40];
3122     char reg_src[3][40];
3123     WORD flags = TEX_DERIV;
3124
3125     shader_arb_get_dst_param(ins, &ins->dst[0], reg_dest);
3126     shader_arb_get_src_param(ins, &ins->src[0], 0, reg_src[0]);
3127     shader_arb_get_src_param(ins, &ins->src[2], 1, reg_src[1]);
3128     shader_arb_get_src_param(ins, &ins->src[3], 2, reg_src[2]);
3129
3130     if (ins->flags & WINED3DSI_TEXLD_PROJECT) flags |= TEX_PROJ;
3131     if (ins->flags & WINED3DSI_TEXLD_BIAS) flags |= TEX_BIAS;
3132
3133     shader_hw_sample(ins, sampler_idx, reg_dest, reg_src[0], flags, reg_src[1], reg_src[2]);
3134 }
3135
3136 static void shader_hw_texldl(const struct wined3d_shader_instruction *ins)
3137 {
3138     DWORD sampler_idx = ins->src[1].reg.idx[0].offset;
3139     char reg_dest[40];
3140     char reg_coord[40];
3141     WORD flags = TEX_LOD;
3142
3143     shader_arb_get_dst_param(ins, &ins->dst[0], reg_dest);
3144     shader_arb_get_src_param(ins, &ins->src[0], 0, reg_coord);
3145
3146     if (ins->flags & WINED3DSI_TEXLD_PROJECT) flags |= TEX_PROJ;
3147     if (ins->flags & WINED3DSI_TEXLD_BIAS) flags |= TEX_BIAS;
3148
3149     shader_hw_sample(ins, sampler_idx, reg_dest, reg_coord, flags, NULL, NULL);
3150 }
3151
3152 static void shader_hw_label(const struct wined3d_shader_instruction *ins)
3153 {
3154     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3155     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3156
3157     priv->in_main_func = FALSE;
3158     /* Call instructions activate the NV extensions, not labels and rets. If there is an uncalled
3159      * subroutine, don't generate a label that will make GL complain
3160      */
3161     if(priv->target_version == ARB) return;
3162
3163     shader_addline(buffer, "l%u:\n", ins->src[0].reg.idx[0].offset);
3164 }
3165
3166 static void vshader_add_footer(struct shader_arb_ctx_priv *priv_ctx,
3167         const struct arb_vshader_private *shader_data, const struct arb_vs_compile_args *args,
3168         const struct wined3d_shader_reg_maps *reg_maps, const struct wined3d_gl_info *gl_info,
3169         struct wined3d_shader_buffer *buffer)
3170 {
3171     unsigned int i;
3172
3173     /* The D3DRS_FOGTABLEMODE render state defines if the shader-generated fog coord is used
3174      * or if the fragment depth is used. If the fragment depth is used(FOGTABLEMODE != NONE),
3175      * the fog frag coord is thrown away. If the fog frag coord is used, but not written by
3176      * the shader, it is set to 0.0(fully fogged, since start = 1.0, end = 0.0)
3177      */
3178     if (args->super.fog_src == VS_FOG_Z)
3179     {
3180         shader_addline(buffer, "MOV result.fogcoord, TMP_OUT.z;\n");
3181     }
3182     else
3183     {
3184         if (!reg_maps->fog)
3185         {
3186             /* posFixup.x is always 1.0, so we can safely use it */
3187             shader_addline(buffer, "ADD result.fogcoord, posFixup.x, -posFixup.x;\n");
3188         }
3189         else
3190         {
3191             /* Clamp fogcoord */
3192             const char *zero = arb_get_helper_value(reg_maps->shader_version.type, ARB_ZERO);
3193             const char *one = arb_get_helper_value(reg_maps->shader_version.type, ARB_ONE);
3194
3195             shader_addline(buffer, "MIN TMP_FOGCOORD.x, TMP_FOGCOORD.x, %s;\n", one);
3196             shader_addline(buffer, "MAX result.fogcoord.x, TMP_FOGCOORD.x, %s;\n", zero);
3197         }
3198     }
3199
3200     /* Clipplanes are always stored without y inversion */
3201     if (use_nv_clip(gl_info) && priv_ctx->target_version >= NV2)
3202     {
3203         if (args->super.clip_enabled)
3204         {
3205             for (i = 0; i < priv_ctx->vs_clipplanes; i++)
3206             {
3207                 shader_addline(buffer, "DP4 result.clip[%u].x, TMP_OUT, state.clip[%u].plane;\n", i, i);
3208             }
3209         }
3210     }
3211     else if (args->clip.boolclip.clip_texcoord)
3212     {
3213         unsigned int cur_clip = 0;
3214         char component[4] = {'x', 'y', 'z', 'w'};
3215         const char *zero = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_ZERO);
3216
3217         for (i = 0; i < gl_info->limits.clipplanes; ++i)
3218         {
3219             if (args->clip.boolclip.clipplane_mask & (1 << i))
3220             {
3221                 shader_addline(buffer, "DP4 TA.%c, TMP_OUT, state.clip[%u].plane;\n",
3222                                component[cur_clip++], i);
3223             }
3224         }
3225         switch (cur_clip)
3226         {
3227             case 0:
3228                 shader_addline(buffer, "MOV TA, %s;\n", zero);
3229                 break;
3230             case 1:
3231                 shader_addline(buffer, "MOV TA.yzw, %s;\n", zero);
3232                 break;
3233             case 2:
3234                 shader_addline(buffer, "MOV TA.zw, %s;\n", zero);
3235                 break;
3236             case 3:
3237                 shader_addline(buffer, "MOV TA.w, %s;\n", zero);
3238                 break;
3239         }
3240         shader_addline(buffer, "MOV result.texcoord[%u], TA;\n",
3241                        args->clip.boolclip.clip_texcoord - 1);
3242     }
3243
3244     /* Write the final position.
3245      *
3246      * OpenGL coordinates specify the center of the pixel while d3d coords specify
3247      * the corner. The offsets are stored in z and w in posFixup. posFixup.y contains
3248      * 1.0 or -1.0 to turn the rendering upside down for offscreen rendering. PosFixup.x
3249      * contains 1.0 to allow a mad, but arb vs swizzles are too restricted for that.
3250      */
3251     shader_addline(buffer, "MUL TA, posFixup, TMP_OUT.w;\n");
3252     shader_addline(buffer, "ADD TMP_OUT.x, TMP_OUT.x, TA.z;\n");
3253     shader_addline(buffer, "MAD TMP_OUT.y, TMP_OUT.y, posFixup.y, TA.w;\n");
3254
3255     /* Z coord [0;1]->[-1;1] mapping, see comment in transform_projection in state.c
3256      * and the glsl equivalent
3257      */
3258     if (need_helper_const(shader_data, reg_maps, gl_info))
3259     {
3260         const char *two = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_TWO);
3261         shader_addline(buffer, "MAD TMP_OUT.z, TMP_OUT.z, %s, -TMP_OUT.w;\n", two);
3262     }
3263     else
3264     {
3265         shader_addline(buffer, "ADD TMP_OUT.z, TMP_OUT.z, TMP_OUT.z;\n");
3266         shader_addline(buffer, "ADD TMP_OUT.z, TMP_OUT.z, -TMP_OUT.w;\n");
3267     }
3268
3269     shader_addline(buffer, "MOV result.position, TMP_OUT;\n");
3270
3271     priv_ctx->footer_written = TRUE;
3272 }
3273
3274 static void shader_hw_ret(const struct wined3d_shader_instruction *ins)
3275 {
3276     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3277     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3278     const struct wined3d_shader *shader = ins->ctx->shader;
3279     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
3280
3281     if(priv->target_version == ARB) return;
3282
3283     if(vshader)
3284     {
3285         if (priv->in_main_func) vshader_add_footer(priv, shader->backend_data,
3286                 priv->cur_vs_args, ins->ctx->reg_maps, ins->ctx->gl_info, buffer);
3287     }
3288
3289     shader_addline(buffer, "RET;\n");
3290 }
3291
3292 static void shader_hw_call(const struct wined3d_shader_instruction *ins)
3293 {
3294     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3295     shader_addline(buffer, "CAL l%u;\n", ins->src[0].reg.idx[0].offset);
3296 }
3297
3298 /* GL locking is done by the caller */
3299 static GLuint create_arb_blt_vertex_program(const struct wined3d_gl_info *gl_info)
3300 {
3301     GLuint program_id = 0;
3302     GLint pos;
3303
3304     const char *blt_vprogram =
3305         "!!ARBvp1.0\n"
3306         "PARAM c[1] = { { 1, 0.5 } };\n"
3307         "MOV result.position, vertex.position;\n"
3308         "MOV result.color, c[0].x;\n"
3309         "MOV result.texcoord[0], vertex.texcoord[0];\n"
3310         "END\n";
3311
3312     GL_EXTCALL(glGenProgramsARB(1, &program_id));
3313     GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, program_id));
3314     GL_EXTCALL(glProgramStringARB(GL_VERTEX_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
3315             strlen(blt_vprogram), blt_vprogram));
3316     checkGLcall("glProgramStringARB()");
3317
3318     gl_info->gl_ops.gl.p_glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
3319     if (pos != -1)
3320     {
3321         FIXME("Vertex program error at position %d: %s\n\n", pos,
3322             debugstr_a((const char *)gl_info->gl_ops.gl.p_glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
3323         shader_arb_dump_program_source(blt_vprogram);
3324     }
3325     else
3326     {
3327         GLint native;
3328
3329         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
3330         checkGLcall("glGetProgramivARB()");
3331         if (!native) WARN("Program exceeds native resource limits.\n");
3332     }
3333
3334     return program_id;
3335 }
3336
3337 /* GL locking is done by the caller */
3338 static GLuint create_arb_blt_fragment_program(const struct wined3d_gl_info *gl_info,
3339         enum tex_types tex_type, BOOL masked)
3340 {
3341     GLuint program_id = 0;
3342     const char *fprogram;
3343     GLint pos;
3344
3345     static const char * const blt_fprograms_full[tex_type_count] =
3346     {
3347         /* tex_1d */
3348         NULL,
3349         /* tex_2d */
3350         "!!ARBfp1.0\n"
3351         "TEMP R0;\n"
3352         "TEX R0.x, fragment.texcoord[0], texture[0], 2D;\n"
3353         "MOV result.depth.z, R0.x;\n"
3354         "END\n",
3355         /* tex_3d */
3356         NULL,
3357         /* tex_cube */
3358         "!!ARBfp1.0\n"
3359         "TEMP R0;\n"
3360         "TEX R0.x, fragment.texcoord[0], texture[0], CUBE;\n"
3361         "MOV result.depth.z, R0.x;\n"
3362         "END\n",
3363         /* tex_rect */
3364         "!!ARBfp1.0\n"
3365         "TEMP R0;\n"
3366         "TEX R0.x, fragment.texcoord[0], texture[0], RECT;\n"
3367         "MOV result.depth.z, R0.x;\n"
3368         "END\n",
3369     };
3370
3371     static const char * const blt_fprograms_masked[tex_type_count] =
3372     {
3373         /* tex_1d */
3374         NULL,
3375         /* tex_2d */
3376         "!!ARBfp1.0\n"
3377         "PARAM mask = program.local[0];\n"
3378         "TEMP R0;\n"
3379         "SLT R0.xy, fragment.position, mask.zwzw;\n"
3380         "MUL R0.x, R0.x, R0.y;\n"
3381         "KIL -R0.x;\n"
3382         "TEX R0.x, fragment.texcoord[0], texture[0], 2D;\n"
3383         "MOV result.depth.z, R0.x;\n"
3384         "END\n",
3385         /* tex_3d */
3386         NULL,
3387         /* tex_cube */
3388         "!!ARBfp1.0\n"
3389         "PARAM mask = program.local[0];\n"
3390         "TEMP R0;\n"
3391         "SLT R0.xy, fragment.position, mask.zwzw;\n"
3392         "MUL R0.x, R0.x, R0.y;\n"
3393         "KIL -R0.x;\n"
3394         "TEX R0.x, fragment.texcoord[0], texture[0], CUBE;\n"
3395         "MOV result.depth.z, R0.x;\n"
3396         "END\n",
3397         /* tex_rect */
3398         "!!ARBfp1.0\n"
3399         "PARAM mask = program.local[0];\n"
3400         "TEMP R0;\n"
3401         "SLT R0.xy, fragment.position, mask.zwzw;\n"
3402         "MUL R0.x, R0.x, R0.y;\n"
3403         "KIL -R0.x;\n"
3404         "TEX R0.x, fragment.texcoord[0], texture[0], RECT;\n"
3405         "MOV result.depth.z, R0.x;\n"
3406         "END\n",
3407     };
3408
3409     fprogram = masked ? blt_fprograms_masked[tex_type] : blt_fprograms_full[tex_type];
3410     if (!fprogram)
3411     {
3412         FIXME("tex_type %#x not supported, falling back to tex_2d\n", tex_type);
3413         tex_type = tex_2d;
3414         fprogram = masked ? blt_fprograms_masked[tex_type] : blt_fprograms_full[tex_type];
3415     }
3416
3417     GL_EXTCALL(glGenProgramsARB(1, &program_id));
3418     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, program_id));
3419     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB, strlen(fprogram), fprogram));
3420     checkGLcall("glProgramStringARB()");
3421
3422     gl_info->gl_ops.gl.p_glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
3423     if (pos != -1)
3424     {
3425         FIXME("Fragment program error at position %d: %s\n\n", pos,
3426             debugstr_a((const char *)gl_info->gl_ops.gl.p_glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
3427         shader_arb_dump_program_source(fprogram);
3428     }
3429     else
3430     {
3431         GLint native;
3432
3433         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
3434         checkGLcall("glGetProgramivARB()");
3435         if (!native) WARN("Program exceeds native resource limits.\n");
3436     }
3437
3438     return program_id;
3439 }
3440
3441 static void arbfp_add_sRGB_correction(struct wined3d_shader_buffer *buffer, const char *fragcolor,
3442         const char *tmp1, const char *tmp2, const char *tmp3, const char *tmp4, BOOL condcode)
3443 {
3444     /* Perform sRGB write correction. See GLX_EXT_framebuffer_sRGB */
3445
3446     if(condcode)
3447     {
3448         /* Sigh. MOVC CC doesn't work, so use one of the temps as dummy dest */
3449         shader_addline(buffer, "SUBC %s, %s.x, srgb_consts1.y;\n", tmp1, fragcolor);
3450         /* Calculate the > 0.0031308 case */
3451         shader_addline(buffer, "POW %s.x (GE), %s.x, srgb_consts1.z;\n", fragcolor, fragcolor);
3452         shader_addline(buffer, "POW %s.y (GE), %s.y, srgb_consts1.z;\n", fragcolor, fragcolor);
3453         shader_addline(buffer, "POW %s.z (GE), %s.z, srgb_consts1.z;\n", fragcolor, fragcolor);
3454         shader_addline(buffer, "MUL %s.xyz (GE), %s, srgb_consts1.w;\n", fragcolor, fragcolor);
3455         shader_addline(buffer, "SUB %s.xyz (GE), %s, srgb_consts2.x;\n", fragcolor, fragcolor);
3456         /* Calculate the < case */
3457         shader_addline(buffer, "MUL %s.xyz (LT), srgb_consts1.x, %s;\n", fragcolor, fragcolor);
3458     }
3459     else
3460     {
3461         /* Calculate the > 0.0031308 case */
3462         shader_addline(buffer, "POW %s.x, %s.x, srgb_consts1.z;\n", tmp1, fragcolor);
3463         shader_addline(buffer, "POW %s.y, %s.y, srgb_consts1.z;\n", tmp1, fragcolor);
3464         shader_addline(buffer, "POW %s.z, %s.z, srgb_consts1.z;\n", tmp1, fragcolor);
3465         shader_addline(buffer, "MUL %s, %s, srgb_consts1.w;\n", tmp1, tmp1);
3466         shader_addline(buffer, "SUB %s, %s, srgb_consts2.x;\n", tmp1, tmp1);
3467         /* Calculate the < case */
3468         shader_addline(buffer, "MUL %s, srgb_consts1.x, %s;\n", tmp2, fragcolor);
3469         /* Get 1.0 / 0.0 masks for > 0.0031308 and < 0.0031308 */
3470         shader_addline(buffer, "SLT %s, srgb_consts1.y, %s;\n", tmp3, fragcolor);
3471         shader_addline(buffer, "SGE %s, srgb_consts1.y, %s;\n", tmp4, fragcolor);
3472         /* Store the components > 0.0031308 in the destination */
3473         shader_addline(buffer, "MUL %s.xyz, %s, %s;\n", fragcolor, tmp1, tmp3);
3474         /* Add the components that are < 0.0031308 */
3475         shader_addline(buffer, "MAD %s.xyz, %s, %s, %s;\n", fragcolor, tmp2, tmp4, fragcolor);
3476         /* Move everything into result.color at once. Nvidia hardware cannot handle partial
3477         * result.color writes(.rgb first, then .a), or handle overwriting already written
3478         * components. The assembler uses a temporary register in this case, which is usually
3479         * not allocated from one of our registers that were used earlier.
3480         */
3481     }
3482     /* [0.0;1.0] clamping. Not needed, this is done implicitly */
3483 }
3484
3485 static const DWORD *find_loop_control_values(const struct wined3d_shader *shader, DWORD idx)
3486 {
3487     const struct wined3d_shader_lconst *constant;
3488
3489     LIST_FOR_EACH_ENTRY(constant, &shader->constantsI, struct wined3d_shader_lconst, entry)
3490     {
3491         if (constant->idx == idx)
3492         {
3493             return constant->value;
3494         }
3495     }
3496     return NULL;
3497 }
3498
3499 static void init_ps_input(const struct wined3d_shader *shader,
3500         const struct arb_ps_compile_args *args, struct shader_arb_ctx_priv *priv)
3501 {
3502     static const char * const texcoords[8] =
3503     {
3504         "fragment.texcoord[0]", "fragment.texcoord[1]", "fragment.texcoord[2]", "fragment.texcoord[3]",
3505         "fragment.texcoord[4]", "fragment.texcoord[5]", "fragment.texcoord[6]", "fragment.texcoord[7]"
3506     };
3507     unsigned int i;
3508     const struct wined3d_shader_signature_element *sig = shader->input_signature;
3509     const char *semantic_name;
3510     DWORD semantic_idx;
3511
3512     switch(args->super.vp_mode)
3513     {
3514         case pretransformed:
3515         case fixedfunction:
3516             /* The pixelshader has to collect the varyings on its own. In any case properly load
3517              * color0 and color1. In the case of pretransformed vertices also load texcoords. Set
3518              * other attribs to 0.0.
3519              *
3520              * For fixedfunction this behavior is correct, according to the tests. For pretransformed
3521              * we'd either need a replacement shader that can load other attribs like BINORMAL, or
3522              * load the texcoord attrib pointers to match the pixel shader signature
3523              */
3524             for(i = 0; i < MAX_REG_INPUT; i++)
3525             {
3526                 semantic_name = sig[i].semantic_name;
3527                 semantic_idx = sig[i].semantic_idx;
3528                 if (!semantic_name) continue;
3529
3530                 if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_COLOR))
3531                 {
3532                     if (!semantic_idx) priv->ps_input[i] = "fragment.color.primary";
3533                     else if(semantic_idx == 1) priv->ps_input[i] = "fragment.color.secondary";
3534                     else priv->ps_input[i] = "0.0";
3535                 }
3536                 else if(args->super.vp_mode == fixedfunction)
3537                 {
3538                     priv->ps_input[i] = "0.0";
3539                 }
3540                 else if(shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_TEXCOORD))
3541                 {
3542                     if(semantic_idx < 8) priv->ps_input[i] = texcoords[semantic_idx];
3543                     else priv->ps_input[i] = "0.0";
3544                 }
3545                 else if(shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_FOG))
3546                 {
3547                     if (!semantic_idx) priv->ps_input[i] = "fragment.fogcoord";
3548                     else priv->ps_input[i] = "0.0";
3549                 }
3550                 else
3551                 {
3552                     priv->ps_input[i] = "0.0";
3553                 }
3554
3555                 TRACE("v%u, semantic %s%u is %s\n", i, semantic_name, semantic_idx, priv->ps_input[i]);
3556             }
3557             break;
3558
3559         case vertexshader:
3560             /* That one is easy. The vertex shaders provide v0-v7 in fragment.texcoord and v8 and v9 in
3561              * fragment.color
3562              */
3563             for(i = 0; i < 8; i++)
3564             {
3565                 priv->ps_input[i] = texcoords[i];
3566             }
3567             priv->ps_input[8] = "fragment.color.primary";
3568             priv->ps_input[9] = "fragment.color.secondary";
3569             break;
3570     }
3571 }
3572
3573 /* GL locking is done by the caller */
3574 static GLuint shader_arb_generate_pshader(const struct wined3d_shader *shader,
3575         const struct wined3d_gl_info *gl_info, struct wined3d_shader_buffer *buffer,
3576         const struct arb_ps_compile_args *args, struct arb_ps_compiled_shader *compiled)
3577 {
3578     const struct wined3d_shader_reg_maps *reg_maps = &shader->reg_maps;
3579     const struct wined3d_shader_lconst *lconst;
3580     const DWORD *function = shader->function;
3581     GLuint retval;
3582     char fragcolor[16];
3583     DWORD *lconst_map = local_const_mapping(shader), next_local;
3584     struct shader_arb_ctx_priv priv_ctx;
3585     BOOL dcl_td = FALSE;
3586     BOOL want_nv_prog = FALSE;
3587     struct arb_pshader_private *shader_priv = shader->backend_data;
3588     GLint errPos;
3589     DWORD map;
3590
3591     char srgbtmp[4][4];
3592     unsigned int i, found = 0;
3593
3594     for (i = 0, map = reg_maps->temporary; map; map >>= 1, ++i)
3595     {
3596         if (!(map & 1)
3597                 || (shader->u.ps.color0_mov && i == shader->u.ps.color0_reg)
3598                 || (reg_maps->shader_version.major < 2 && !i))
3599             continue;
3600
3601         sprintf(srgbtmp[found], "R%u", i);
3602         ++found;
3603         if (found == 4) break;
3604     }
3605
3606     switch(found) {
3607         case 0:
3608             sprintf(srgbtmp[0], "TA");
3609             sprintf(srgbtmp[1], "TB");
3610             sprintf(srgbtmp[2], "TC");
3611             sprintf(srgbtmp[3], "TD");
3612             dcl_td = TRUE;
3613             break;
3614         case 1:
3615             sprintf(srgbtmp[1], "TA");
3616             sprintf(srgbtmp[2], "TB");
3617             sprintf(srgbtmp[3], "TC");
3618             break;
3619         case 2:
3620             sprintf(srgbtmp[2], "TA");
3621             sprintf(srgbtmp[3], "TB");
3622             break;
3623         case 3:
3624             sprintf(srgbtmp[3], "TA");
3625             break;
3626         case 4:
3627             break;
3628     }
3629
3630     /*  Create the hw ARB shader */
3631     memset(&priv_ctx, 0, sizeof(priv_ctx));
3632     priv_ctx.cur_ps_args = args;
3633     priv_ctx.compiled_fprog = compiled;
3634     priv_ctx.cur_np2fixup_info = &compiled->np2fixup_info;
3635     init_ps_input(shader, args, &priv_ctx);
3636     list_init(&priv_ctx.control_frames);
3637
3638     /* Avoid enabling NV_fragment_program* if we do not need it.
3639      *
3640      * Enabling GL_NV_fragment_program_option causes the driver to occupy a temporary register,
3641      * and it slows down the shader execution noticeably(about 5%). Usually our instruction emulation
3642      * is faster than what we gain from using higher native instructions. There are some things though
3643      * that cannot be emulated. In that case enable the extensions.
3644      * If the extension is enabled, instruction handlers that support both ways will use it.
3645      *
3646      * Testing shows no performance difference between OPTION NV_fragment_program2 and NV_fragment_program.
3647      * So enable the best we can get.
3648      */
3649     if(reg_maps->usesdsx || reg_maps->usesdsy || reg_maps->loop_depth > 0 || reg_maps->usestexldd ||
3650        reg_maps->usestexldl || reg_maps->usesfacing || reg_maps->usesifc || reg_maps->usescall)
3651     {
3652         want_nv_prog = TRUE;
3653     }
3654
3655     shader_addline(buffer, "!!ARBfp1.0\n");
3656     if (want_nv_prog && gl_info->supported[NV_FRAGMENT_PROGRAM2])
3657     {
3658         shader_addline(buffer, "OPTION NV_fragment_program2;\n");
3659         priv_ctx.target_version = NV3;
3660     }
3661     else if (want_nv_prog && gl_info->supported[NV_FRAGMENT_PROGRAM_OPTION])
3662     {
3663         shader_addline(buffer, "OPTION NV_fragment_program;\n");
3664         priv_ctx.target_version = NV2;
3665     } else {
3666         if(want_nv_prog)
3667         {
3668             /* This is an error - either we're advertising the wrong shader version, or aren't enforcing some
3669              * limits properly
3670              */
3671             ERR("The shader requires instructions that are not available in plain GL_ARB_fragment_program\n");
3672             ERR("Try GLSL\n");
3673         }
3674         priv_ctx.target_version = ARB;
3675     }
3676
3677     if (reg_maps->rt_mask > 1)
3678     {
3679         shader_addline(buffer, "OPTION ARB_draw_buffers;\n");
3680     }
3681
3682     if (reg_maps->shader_version.major < 3)
3683     {
3684         switch(args->super.fog) {
3685             case FOG_OFF:
3686                 break;
3687             case FOG_LINEAR:
3688                 shader_addline(buffer, "OPTION ARB_fog_linear;\n");
3689                 break;
3690             case FOG_EXP:
3691                 shader_addline(buffer, "OPTION ARB_fog_exp;\n");
3692                 break;
3693             case FOG_EXP2:
3694                 shader_addline(buffer, "OPTION ARB_fog_exp2;\n");
3695                 break;
3696         }
3697     }
3698
3699     /* For now always declare the temps. At least the Nvidia assembler optimizes completely
3700      * unused temps away(but occupies them for the whole shader if they're used once). Always
3701      * declaring them avoids tricky bookkeeping work
3702      */
3703     shader_addline(buffer, "TEMP TA;\n");      /* Used for modifiers */
3704     shader_addline(buffer, "TEMP TB;\n");      /* Used for modifiers */
3705     shader_addline(buffer, "TEMP TC;\n");      /* Used for modifiers */
3706     if(dcl_td) shader_addline(buffer, "TEMP TD;\n"); /* Used for sRGB writing */
3707     shader_addline(buffer, "PARAM coefdiv = { 0.5, 0.25, 0.125, 0.0625 };\n");
3708     shader_addline(buffer, "PARAM coefmul = { 2, 4, 8, 16 };\n");
3709     shader_addline(buffer, "PARAM ps_helper_const = { 0.0, 1.0, %1.10f, 0.0 };\n", eps);
3710
3711     if (reg_maps->shader_version.major < 2)
3712     {
3713         strcpy(fragcolor, "R0");
3714     }
3715     else
3716     {
3717         if (args->super.srgb_correction)
3718         {
3719             if (shader->u.ps.color0_mov)
3720             {
3721                 sprintf(fragcolor, "R%u", shader->u.ps.color0_reg);
3722             }
3723             else
3724             {
3725                 shader_addline(buffer, "TEMP TMP_COLOR;\n");
3726                 strcpy(fragcolor, "TMP_COLOR");
3727             }
3728         } else {
3729             strcpy(fragcolor, "result.color");
3730         }
3731     }
3732
3733     if(args->super.srgb_correction) {
3734         shader_addline(buffer, "PARAM srgb_consts1 = {%f, %f, %f, %f};\n",
3735                        srgb_mul_low, srgb_cmp, srgb_pow, srgb_mul_high);
3736         shader_addline(buffer, "PARAM srgb_consts2 = {%f, %f, %f, %f};\n",
3737                        srgb_sub_high, 0.0, 0.0, 0.0);
3738     }
3739
3740     /* Base Declarations */
3741     next_local = shader_generate_arb_declarations(shader, reg_maps,
3742             buffer, gl_info, lconst_map, NULL, &priv_ctx);
3743
3744     for (i = 0, map = reg_maps->bumpmat; map; map >>= 1, ++i)
3745     {
3746         unsigned char bump_const;
3747
3748         if (!(map & 1)) continue;
3749
3750         bump_const = compiled->numbumpenvmatconsts;
3751         compiled->bumpenvmatconst[bump_const].const_num = WINED3D_CONST_NUM_UNUSED;
3752         compiled->bumpenvmatconst[bump_const].texunit = i;
3753         compiled->luminanceconst[bump_const].const_num = WINED3D_CONST_NUM_UNUSED;
3754         compiled->luminanceconst[bump_const].texunit = i;
3755
3756         /* We can fit the constants into the constant limit for sure because texbem, texbeml, bem and beml are only supported
3757          * in 1.x shaders, and GL_ARB_fragment_program has a constant limit of 24 constants. So in the worst case we're loading
3758          * 8 shader constants, 8 bump matrices and 8 luminance parameters and are perfectly fine. (No NP2 fixup on bumpmapped
3759          * textures due to conditional NP2 restrictions)
3760          *
3761          * Use local constants to load the bump env parameters, not program.env. This avoids collisions with d3d constants of
3762          * shaders in newer shader models. Since the bump env parameters have to share their space with NP2 fixup constants,
3763          * their location is shader dependent anyway and they cannot be loaded globally.
3764          */
3765         compiled->bumpenvmatconst[bump_const].const_num = next_local++;
3766         shader_addline(buffer, "PARAM bumpenvmat%d = program.local[%d];\n",
3767                        i, compiled->bumpenvmatconst[bump_const].const_num);
3768         compiled->numbumpenvmatconsts = bump_const + 1;
3769
3770         if (!(reg_maps->luminanceparams & (1 << i))) continue;
3771
3772         compiled->luminanceconst[bump_const].const_num = next_local++;
3773         shader_addline(buffer, "PARAM luminance%d = program.local[%d];\n",
3774                        i, compiled->luminanceconst[bump_const].const_num);
3775     }
3776
3777     for(i = 0; i < MAX_CONST_I; i++)
3778     {
3779         compiled->int_consts[i] = WINED3D_CONST_NUM_UNUSED;
3780         if (reg_maps->integer_constants & (1 << i) && priv_ctx.target_version >= NV2)
3781         {
3782             const DWORD *control_values = find_loop_control_values(shader, i);
3783
3784             if(control_values)
3785             {
3786                 shader_addline(buffer, "PARAM I%u = {%u, %u, %u, -1};\n", i,
3787                                 control_values[0], control_values[1], control_values[2]);
3788             }
3789             else
3790             {
3791                 compiled->int_consts[i] = next_local;
3792                 compiled->num_int_consts++;
3793                 shader_addline(buffer, "PARAM I%u = program.local[%u];\n", i, next_local++);
3794             }
3795         }
3796     }
3797
3798     if(reg_maps->vpos || reg_maps->usesdsy)
3799     {
3800         compiled->ycorrection = next_local;
3801         shader_addline(buffer, "PARAM ycorrection = program.local[%u];\n", next_local++);
3802
3803         if(reg_maps->vpos)
3804         {
3805             shader_addline(buffer, "TEMP vpos;\n");
3806             /* ycorrection.x: Backbuffer height(onscreen) or 0(offscreen).
3807              * ycorrection.y: -1.0(onscreen), 1.0(offscreen)
3808              * ycorrection.z: 1.0
3809              * ycorrection.w: 0.0
3810              */
3811             shader_addline(buffer, "MAD vpos, fragment.position, ycorrection.zyww, ycorrection.wxww;\n");
3812             shader_addline(buffer, "FLR vpos.xy, vpos;\n");
3813         }
3814     }
3815     else
3816     {
3817         compiled->ycorrection = WINED3D_CONST_NUM_UNUSED;
3818     }
3819
3820     /* Load constants to fixup NP2 texcoords if there are still free constants left:
3821      * Constants (texture dimensions) for the NP2 fixup are loaded as local program parameters. This will consume
3822      * at most 8 (MAX_FRAGMENT_SAMPLERS / 2) parameters, which is highly unlikely, since the application had to
3823      * use 16 NP2 textures at the same time. In case that we run out of constants the fixup is simply not
3824      * applied / activated. This will probably result in wrong rendering of the texture, but will save us from
3825      * shader compilation errors and the subsequent errors when drawing with this shader. */
3826     if (priv_ctx.cur_ps_args->super.np2_fixup) {
3827         unsigned char cur_fixup_sampler = 0;
3828
3829         struct arb_ps_np2fixup_info* const fixup = priv_ctx.cur_np2fixup_info;
3830         const WORD map = priv_ctx.cur_ps_args->super.np2_fixup;
3831         const UINT max_lconsts = gl_info->limits.arb_ps_local_constants;
3832
3833         fixup->offset = next_local;
3834         fixup->super.active = 0;
3835
3836         for (i = 0; i < MAX_FRAGMENT_SAMPLERS; ++i) {
3837             if (!(map & (1 << i))) continue;
3838
3839             if (fixup->offset + (cur_fixup_sampler >> 1) < max_lconsts) {
3840                 fixup->super.active |= (1 << i);
3841                 fixup->super.idx[i] = cur_fixup_sampler++;
3842             } else {
3843                 FIXME("No free constant found to load NP2 fixup data into shader. "
3844                       "Sampling from this texture will probably look wrong.\n");
3845                 break;
3846             }
3847         }
3848
3849         fixup->super.num_consts = (cur_fixup_sampler + 1) >> 1;
3850         if (fixup->super.num_consts) {
3851             shader_addline(buffer, "PARAM np2fixup[%u] = { program.env[%u..%u] };\n",
3852                            fixup->super.num_consts, fixup->offset, fixup->super.num_consts + fixup->offset - 1);
3853         }
3854     }
3855
3856     if (shader_priv->clipplane_emulation != ~0U && args->clip)
3857     {
3858         shader_addline(buffer, "KIL fragment.texcoord[%u];\n", shader_priv->clipplane_emulation);
3859     }
3860
3861     /* Base Shader Body */
3862     shader_generate_main(shader, buffer, reg_maps, function, &priv_ctx);
3863
3864     if(args->super.srgb_correction) {
3865         arbfp_add_sRGB_correction(buffer, fragcolor, srgbtmp[0], srgbtmp[1], srgbtmp[2], srgbtmp[3],
3866                                   priv_ctx.target_version >= NV2);
3867     }
3868
3869     if(strcmp(fragcolor, "result.color")) {
3870         shader_addline(buffer, "MOV result.color, %s;\n", fragcolor);
3871     }
3872     shader_addline(buffer, "END\n");
3873
3874     /* TODO: change to resource.glObjectHandle or something like that */
3875     GL_EXTCALL(glGenProgramsARB(1, &retval));
3876
3877     TRACE("Creating a hw pixel shader, prg=%d\n", retval);
3878     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, retval));
3879
3880     TRACE("Created hw pixel shader, prg=%d\n", retval);
3881     /* Create the program and check for errors */
3882     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
3883                buffer->bsize, buffer->buffer));
3884     checkGLcall("glProgramStringARB()");
3885
3886     gl_info->gl_ops.gl.p_glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &errPos);
3887     if (errPos != -1)
3888     {
3889         FIXME("HW PixelShader Error at position %d: %s\n\n",
3890               errPos, debugstr_a((const char *)gl_info->gl_ops.gl.p_glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
3891         shader_arb_dump_program_source(buffer->buffer);
3892         retval = 0;
3893     }
3894     else
3895     {
3896         GLint native;
3897
3898         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
3899         checkGLcall("glGetProgramivARB()");
3900         if (!native) WARN("Program exceeds native resource limits.\n");
3901     }
3902
3903     /* Load immediate constants */
3904     if (lconst_map)
3905     {
3906         LIST_FOR_EACH_ENTRY(lconst, &shader->constantsF, struct wined3d_shader_lconst, entry)
3907         {
3908             const float *value = (const float *)lconst->value;
3909             GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, lconst_map[lconst->idx], value));
3910             checkGLcall("glProgramLocalParameter4fvARB");
3911         }
3912         HeapFree(GetProcessHeap(), 0, lconst_map);
3913     }
3914
3915     return retval;
3916 }
3917
3918 static int compare_sig(const struct wined3d_shader_signature_element *sig1, const struct wined3d_shader_signature_element *sig2)
3919 {
3920     unsigned int i;
3921     int ret;
3922
3923     for(i = 0; i < MAX_REG_INPUT; i++)
3924     {
3925         if (!sig1[i].semantic_name || !sig2[i].semantic_name)
3926         {
3927             /* Compare pointers, not contents. One string is NULL(element does not exist), the other one is not NULL */
3928             if(sig1[i].semantic_name != sig2[i].semantic_name) return sig1[i].semantic_name < sig2[i].semantic_name ? -1 : 1;
3929             continue;
3930         }
3931
3932         if ((ret = strcmp(sig1[i].semantic_name, sig2[i].semantic_name))) return ret;
3933         if(sig1[i].semantic_idx    != sig2[i].semantic_idx)    return sig1[i].semantic_idx    < sig2[i].semantic_idx    ? -1 : 1;
3934         if(sig1[i].sysval_semantic != sig2[i].sysval_semantic) return sig1[i].sysval_semantic < sig2[i].sysval_semantic ? -1 : 1;
3935         if(sig1[i].component_type  != sig2[i].component_type)  return sig1[i].component_type  < sig2[i].component_type  ? -1 : 1;
3936         if(sig1[i].register_idx    != sig2[i].register_idx)    return sig1[i].register_idx    < sig2[i].register_idx    ? -1 : 1;
3937         if(sig1[i].mask            != sig2[i].mask)            return sig1[i].mask            < sig2[i].mask            ? -1 : 1;
3938     }
3939     return 0;
3940 }
3941
3942 static struct wined3d_shader_signature_element *clone_sig(const struct wined3d_shader_signature_element *sig)
3943 {
3944     struct wined3d_shader_signature_element *new;
3945     int i;
3946     char *name;
3947
3948     new = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*new) * MAX_REG_INPUT);
3949     for(i = 0; i < MAX_REG_INPUT; i++)
3950     {
3951         if (!sig[i].semantic_name) continue;
3952
3953         new[i] = sig[i];
3954         /* Clone the semantic string */
3955         name = HeapAlloc(GetProcessHeap(), 0, strlen(sig[i].semantic_name) + 1);
3956         strcpy(name, sig[i].semantic_name);
3957         new[i].semantic_name = name;
3958     }
3959     return new;
3960 }
3961
3962 static DWORD find_input_signature(struct shader_arb_priv *priv, const struct wined3d_shader_signature_element *sig)
3963 {
3964     struct wine_rb_entry *entry = wine_rb_get(&priv->signature_tree, sig);
3965     struct ps_signature *found_sig;
3966
3967     if (entry)
3968     {
3969         found_sig = WINE_RB_ENTRY_VALUE(entry, struct ps_signature, entry);
3970         TRACE("Found existing signature %u\n", found_sig->idx);
3971         return found_sig->idx;
3972     }
3973     found_sig = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*sig));
3974     found_sig->sig = clone_sig(sig);
3975     found_sig->idx = priv->ps_sig_number++;
3976     TRACE("New signature stored and assigned number %u\n", found_sig->idx);
3977     if(wine_rb_put(&priv->signature_tree, sig, &found_sig->entry) == -1)
3978     {
3979         ERR("Failed to insert program entry.\n");
3980     }
3981     return found_sig->idx;
3982 }
3983
3984 static void init_output_registers(const struct wined3d_shader *shader, DWORD sig_num,
3985         struct shader_arb_ctx_priv *priv_ctx, struct arb_vs_compiled_shader *compiled)
3986 {
3987     unsigned int i, j;
3988     static const char * const texcoords[8] =
3989     {
3990         "result.texcoord[0]", "result.texcoord[1]", "result.texcoord[2]", "result.texcoord[3]",
3991         "result.texcoord[4]", "result.texcoord[5]", "result.texcoord[6]", "result.texcoord[7]"
3992     };
3993     struct wined3d_device *device = shader->device;
3994     const struct wined3d_shader_signature_element *sig;
3995     const char *semantic_name;
3996     DWORD semantic_idx, reg_idx;
3997
3998     /* Write generic input varyings 0 to 7 to result.texcoord[], varying 8 to result.color.primary
3999      * and varying 9 to result.color.secondary
4000      */
4001     static const char * const decl_idx_to_string[MAX_REG_INPUT] =
4002     {
4003         "result.texcoord[0]", "result.texcoord[1]", "result.texcoord[2]", "result.texcoord[3]",
4004         "result.texcoord[4]", "result.texcoord[5]", "result.texcoord[6]", "result.texcoord[7]",
4005         "result.color.primary", "result.color.secondary"
4006     };
4007
4008     if(sig_num == ~0)
4009     {
4010         TRACE("Pixel shader uses builtin varyings\n");
4011         /* Map builtins to builtins */
4012         for(i = 0; i < 8; i++)
4013         {
4014             priv_ctx->texcrd_output[i] = texcoords[i];
4015         }
4016         priv_ctx->color_output[0] = "result.color.primary";
4017         priv_ctx->color_output[1] = "result.color.secondary";
4018         priv_ctx->fog_output = "TMP_FOGCOORD";
4019
4020         /* Map declared regs to builtins. Use "TA" to /dev/null unread output */
4021         for (i = 0; i < (sizeof(shader->output_signature) / sizeof(*shader->output_signature)); ++i)
4022         {
4023             semantic_name = shader->output_signature[i].semantic_name;
4024             if (!semantic_name) continue;
4025
4026             if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_POSITION))
4027             {
4028                 TRACE("o%u is TMP_OUT\n", i);
4029                 if (!shader->output_signature[i].semantic_idx) priv_ctx->vs_output[i] = "TMP_OUT";
4030                 else priv_ctx->vs_output[i] = "TA";
4031             }
4032             else if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_PSIZE))
4033             {
4034                 TRACE("o%u is result.pointsize\n", i);
4035                 if (!shader->output_signature[i].semantic_idx) priv_ctx->vs_output[i] = "result.pointsize";
4036                 else priv_ctx->vs_output[i] = "TA";
4037             }
4038             else if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_COLOR))
4039             {
4040                 TRACE("o%u is result.color.?, idx %u\n", i, shader->output_signature[i].semantic_idx);
4041                 if (!shader->output_signature[i].semantic_idx)
4042                     priv_ctx->vs_output[i] = "result.color.primary";
4043                 else if (shader->output_signature[i].semantic_idx == 1)
4044                     priv_ctx->vs_output[i] = "result.color.secondary";
4045                 else priv_ctx->vs_output[i] = "TA";
4046             }
4047             else if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_TEXCOORD))
4048             {
4049                 TRACE("o%u is %s\n", i, texcoords[shader->output_signature[i].semantic_idx]);
4050                 if (shader->output_signature[i].semantic_idx >= 8) priv_ctx->vs_output[i] = "TA";
4051                 else priv_ctx->vs_output[i] = texcoords[shader->output_signature[i].semantic_idx];
4052             }
4053             else if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_FOG))
4054             {
4055                 TRACE("o%u is result.fogcoord\n", i);
4056                 if (shader->output_signature[i].semantic_idx > 0) priv_ctx->vs_output[i] = "TA";
4057                 else priv_ctx->vs_output[i] = "result.fogcoord";
4058             }
4059             else
4060             {
4061                 priv_ctx->vs_output[i] = "TA";
4062             }
4063         }
4064         return;
4065     }
4066
4067     /* Instead of searching for the signature in the signature list, read the one from the current pixel shader.
4068      * Its maybe not the shader where the signature came from, but it is the same signature and faster to find
4069      */
4070     sig = device->stateBlock->state.pixel_shader->input_signature;
4071     TRACE("Pixel shader uses declared varyings\n");
4072
4073     /* Map builtin to declared. /dev/null the results by default to the TA temp reg */
4074     for(i = 0; i < 8; i++)
4075     {
4076         priv_ctx->texcrd_output[i] = "TA";
4077     }
4078     priv_ctx->color_output[0] = "TA";
4079     priv_ctx->color_output[1] = "TA";
4080     priv_ctx->fog_output = "TA";
4081
4082     for(i = 0; i < MAX_REG_INPUT; i++)
4083     {
4084         semantic_name = sig[i].semantic_name;
4085         semantic_idx = sig[i].semantic_idx;
4086         reg_idx = sig[i].register_idx;
4087         if (!semantic_name) continue;
4088
4089         /* If a declared input register is not written by builtin arguments, don't write to it.
4090          * GL_NV_vertex_program makes sure the input defaults to 0.0, which is correct with D3D
4091          *
4092          * Don't care about POSITION and PSIZE here - this is a builtin vertex shader, position goes
4093          * to TMP_OUT in any case
4094          */
4095         if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_TEXCOORD))
4096         {
4097             if (semantic_idx < 8)
4098                 priv_ctx->texcrd_output[semantic_idx] = decl_idx_to_string[reg_idx];
4099         }
4100         else if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_COLOR))
4101         {
4102             if (semantic_idx < 2)
4103                 priv_ctx->color_output[semantic_idx] = decl_idx_to_string[reg_idx];
4104         }
4105         else if(shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_FOG))
4106         {
4107             if (!semantic_idx)
4108                 priv_ctx->fog_output = decl_idx_to_string[reg_idx];
4109         }
4110         else
4111         {
4112             continue;
4113         }
4114
4115         if (!strcmp(decl_idx_to_string[reg_idx], "result.color.primary")
4116                 || !strcmp(decl_idx_to_string[reg_idx], "result.color.secondary"))
4117         {
4118             compiled->need_color_unclamp = TRUE;
4119         }
4120     }
4121
4122     /* Map declared to declared */
4123     for (i = 0; i < (sizeof(shader->output_signature) / sizeof(*shader->output_signature)); ++i)
4124     {
4125         /* Write unread output to TA to throw them away */
4126         priv_ctx->vs_output[i] = "TA";
4127         semantic_name = shader->output_signature[i].semantic_name;
4128         if (!semantic_name) continue;
4129
4130         if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_POSITION)
4131                 && !shader->output_signature[i].semantic_idx)
4132         {
4133             priv_ctx->vs_output[i] = "TMP_OUT";
4134             continue;
4135         }
4136         else if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_PSIZE)
4137                 && !shader->output_signature[i].semantic_idx)
4138         {
4139             priv_ctx->vs_output[i] = "result.pointsize";
4140             continue;
4141         }
4142
4143         for(j = 0; j < MAX_REG_INPUT; j++)
4144         {
4145             if (!sig[j].semantic_name) continue;
4146
4147             if (!strcmp(sig[j].semantic_name, semantic_name)
4148                     && sig[j].semantic_idx == shader->output_signature[i].semantic_idx)
4149             {
4150                 priv_ctx->vs_output[i] = decl_idx_to_string[sig[j].register_idx];
4151
4152                 if (!strcmp(priv_ctx->vs_output[i], "result.color.primary")
4153                         || !strcmp(priv_ctx->vs_output[i], "result.color.secondary"))
4154                 {
4155                     compiled->need_color_unclamp = TRUE;
4156                 }
4157             }
4158         }
4159     }
4160 }
4161
4162 /* GL locking is done by the caller */
4163 static GLuint shader_arb_generate_vshader(const struct wined3d_shader *shader,
4164         const struct wined3d_gl_info *gl_info, struct wined3d_shader_buffer *buffer,
4165         const struct arb_vs_compile_args *args, struct arb_vs_compiled_shader *compiled)
4166 {
4167     const struct arb_vshader_private *shader_data = shader->backend_data;
4168     const struct wined3d_shader_reg_maps *reg_maps = &shader->reg_maps;
4169     const struct wined3d_shader_lconst *lconst;
4170     const DWORD *function = shader->function;
4171     GLuint ret;
4172     DWORD next_local, *lconst_map = local_const_mapping(shader);
4173     struct shader_arb_ctx_priv priv_ctx;
4174     unsigned int i;
4175     GLint errPos;
4176
4177     memset(&priv_ctx, 0, sizeof(priv_ctx));
4178     priv_ctx.cur_vs_args = args;
4179     list_init(&priv_ctx.control_frames);
4180     init_output_registers(shader, args->ps_signature, &priv_ctx, compiled);
4181
4182     /*  Create the hw ARB shader */
4183     shader_addline(buffer, "!!ARBvp1.0\n");
4184
4185     /* Always enable the NV extension if available. Unlike fragment shaders, there is no
4186      * mesurable performance penalty, and we can always make use of it for clipplanes.
4187      */
4188     if (gl_info->supported[NV_VERTEX_PROGRAM3])
4189     {
4190         shader_addline(buffer, "OPTION NV_vertex_program3;\n");
4191         priv_ctx.target_version = NV3;
4192         shader_addline(buffer, "ADDRESS aL;\n");
4193     }
4194     else if (gl_info->supported[NV_VERTEX_PROGRAM2_OPTION])
4195     {
4196         shader_addline(buffer, "OPTION NV_vertex_program2;\n");
4197         priv_ctx.target_version = NV2;
4198         shader_addline(buffer, "ADDRESS aL;\n");
4199     } else {
4200         priv_ctx.target_version = ARB;
4201     }
4202
4203     shader_addline(buffer, "TEMP TMP_OUT;\n");
4204     if (reg_maps->fog)
4205         shader_addline(buffer, "TEMP TMP_FOGCOORD;\n");
4206     if (need_helper_const(shader_data, reg_maps, gl_info))
4207     {
4208         shader_addline(buffer, "PARAM helper_const = { 0.0, 1.0, 2.0, %1.10f};\n", eps);
4209     }
4210     if (need_rel_addr_const(shader_data, reg_maps, gl_info))
4211     {
4212         shader_addline(buffer, "PARAM rel_addr_const = { 0.5, %d.0, 0.0, 0.0 };\n", shader_data->rel_offset);
4213         shader_addline(buffer, "TEMP A0_SHADOW;\n");
4214     }
4215
4216     shader_addline(buffer, "TEMP TA;\n");
4217     shader_addline(buffer, "TEMP TB;\n");
4218
4219     /* Base Declarations */
4220     next_local = shader_generate_arb_declarations(shader, reg_maps, buffer,
4221             gl_info, lconst_map, &priv_ctx.vs_clipplanes, &priv_ctx);
4222
4223     for(i = 0; i < MAX_CONST_I; i++)
4224     {
4225         compiled->int_consts[i] = WINED3D_CONST_NUM_UNUSED;
4226         if(reg_maps->integer_constants & (1 << i) && priv_ctx.target_version >= NV2)
4227         {
4228             const DWORD *control_values = find_loop_control_values(shader, i);
4229
4230             if(control_values)
4231             {
4232                 shader_addline(buffer, "PARAM I%u = {%u, %u, %u, -1};\n", i,
4233                                 control_values[0], control_values[1], control_values[2]);
4234             }
4235             else
4236             {
4237                 compiled->int_consts[i] = next_local;
4238                 compiled->num_int_consts++;
4239                 shader_addline(buffer, "PARAM I%u = program.local[%u];\n", i, next_local++);
4240             }
4241         }
4242     }
4243
4244     /* We need a constant to fixup the final position */
4245     shader_addline(buffer, "PARAM posFixup = program.local[%u];\n", next_local);
4246     compiled->pos_fixup = next_local++;
4247
4248     /* Initialize output parameters. GL_ARB_vertex_program does not require special initialization values
4249      * for output parameters. D3D in theory does not do that either, but some applications depend on a
4250      * proper initialization of the secondary color, and programs using the fixed function pipeline without
4251      * a replacement shader depend on the texcoord.w being set properly.
4252      *
4253      * GL_NV_vertex_program defines that all output values are initialized to {0.0, 0.0, 0.0, 1.0}. This
4254      * assertion is in effect even when using GL_ARB_vertex_program without any NV specific additions. So
4255      * skip this if NV_vertex_program is supported. Otherwise, initialize the secondary color. For the tex-
4256      * coords, we have a flag in the opengl caps. Many cards do not require the texcoord being set, and
4257      * this can eat a number of instructions, so skip it unless this cap is set as well
4258      */
4259     if (!gl_info->supported[NV_VERTEX_PROGRAM])
4260     {
4261         struct wined3d_device *device = shader->device;
4262         const char *color_init = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_0001);
4263         shader_addline(buffer, "MOV result.color.secondary, %s;\n", color_init);
4264
4265         if (gl_info->quirks & WINED3D_QUIRK_SET_TEXCOORD_W && !device->frag_pipe->ffp_proj_control)
4266         {
4267             int i;
4268             const char *one = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_ONE);
4269             for(i = 0; i < min(8, MAX_REG_TEXCRD); i++)
4270             {
4271                 if (reg_maps->texcoord_mask[i] && reg_maps->texcoord_mask[i] != WINED3DSP_WRITEMASK_ALL)
4272                     shader_addline(buffer, "MOV result.texcoord[%u].w, %s\n", i, one);
4273             }
4274         }
4275     }
4276
4277     /* The shader starts with the main function */
4278     priv_ctx.in_main_func = TRUE;
4279     /* Base Shader Body */
4280     shader_generate_main(shader, buffer, reg_maps, function, &priv_ctx);
4281
4282     if (!priv_ctx.footer_written) vshader_add_footer(&priv_ctx,
4283             shader_data, args, reg_maps, gl_info, buffer);
4284
4285     shader_addline(buffer, "END\n");
4286
4287     /* TODO: change to resource.glObjectHandle or something like that */
4288     GL_EXTCALL(glGenProgramsARB(1, &ret));
4289
4290     TRACE("Creating a hw vertex shader, prg=%d\n", ret);
4291     GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, ret));
4292
4293     TRACE("Created hw vertex shader, prg=%d\n", ret);
4294     /* Create the program and check for errors */
4295     GL_EXTCALL(glProgramStringARB(GL_VERTEX_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
4296                buffer->bsize, buffer->buffer));
4297     checkGLcall("glProgramStringARB()");
4298
4299     gl_info->gl_ops.gl.p_glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &errPos);
4300     if (errPos != -1)
4301     {
4302         FIXME("HW VertexShader Error at position %d: %s\n\n",
4303               errPos, debugstr_a((const char *)gl_info->gl_ops.gl.p_glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
4304         shader_arb_dump_program_source(buffer->buffer);
4305         ret = -1;
4306     }
4307     else
4308     {
4309         GLint native;
4310
4311         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
4312         checkGLcall("glGetProgramivARB()");
4313         if (!native) WARN("Program exceeds native resource limits.\n");
4314
4315         /* Load immediate constants */
4316         if (lconst_map)
4317         {
4318             LIST_FOR_EACH_ENTRY(lconst, &shader->constantsF, struct wined3d_shader_lconst, entry)
4319             {
4320                 const float *value = (const float *)lconst->value;
4321                 GL_EXTCALL(glProgramLocalParameter4fvARB(GL_VERTEX_PROGRAM_ARB, lconst_map[lconst->idx], value));
4322             }
4323         }
4324     }
4325     HeapFree(GetProcessHeap(), 0, lconst_map);
4326
4327     return ret;
4328 }
4329
4330 /* GL locking is done by the caller */
4331 static struct arb_ps_compiled_shader *find_arb_pshader(struct wined3d_shader *shader,
4332         const struct arb_ps_compile_args *args)
4333 {
4334     struct wined3d_device *device = shader->device;
4335     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4336     UINT i;
4337     DWORD new_size;
4338     struct arb_ps_compiled_shader *new_array;
4339     struct wined3d_shader_buffer buffer;
4340     struct arb_pshader_private *shader_data;
4341     GLuint ret;
4342
4343     if (!shader->backend_data)
4344     {
4345         struct shader_arb_priv *priv = device->shader_priv;
4346
4347         shader->backend_data = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data));
4348         shader_data = shader->backend_data;
4349         shader_data->clamp_consts = shader->reg_maps.shader_version.major == 1;
4350
4351         if (shader->reg_maps.shader_version.major < 3)
4352             shader_data->input_signature_idx = ~0;
4353         else
4354             shader_data->input_signature_idx = find_input_signature(priv, shader->input_signature);
4355
4356         TRACE("Shader got assigned input signature index %u\n", shader_data->input_signature_idx);
4357
4358         if (!device->vs_clipping)
4359             shader_data->clipplane_emulation = shader_find_free_input_register(&shader->reg_maps,
4360                     gl_info->limits.texture_stages - 1);
4361         else
4362             shader_data->clipplane_emulation = ~0U;
4363     }
4364     shader_data = shader->backend_data;
4365
4366     /* Usually we have very few GL shaders for each d3d shader(just 1 or maybe 2),
4367      * so a linear search is more performant than a hashmap or a binary search
4368      * (cache coherency etc)
4369      */
4370     for (i = 0; i < shader_data->num_gl_shaders; ++i)
4371     {
4372         if (!memcmp(&shader_data->gl_shaders[i].args, args, sizeof(*args)))
4373             return &shader_data->gl_shaders[i];
4374     }
4375
4376     TRACE("No matching GL shader found, compiling a new shader\n");
4377     if(shader_data->shader_array_size == shader_data->num_gl_shaders) {
4378         if (shader_data->num_gl_shaders)
4379         {
4380             new_size = shader_data->shader_array_size + max(1, shader_data->shader_array_size / 2);
4381             new_array = HeapReAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, shader_data->gl_shaders,
4382                                     new_size * sizeof(*shader_data->gl_shaders));
4383         } else {
4384             new_array = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data->gl_shaders));
4385             new_size = 1;
4386         }
4387
4388         if(!new_array) {
4389             ERR("Out of memory\n");
4390             return 0;
4391         }
4392         shader_data->gl_shaders = new_array;
4393         shader_data->shader_array_size = new_size;
4394     }
4395
4396     shader_data->gl_shaders[shader_data->num_gl_shaders].args = *args;
4397
4398     pixelshader_update_samplers(&shader->reg_maps, device->stateBlock->state.textures);
4399
4400     if (!shader_buffer_init(&buffer))
4401     {
4402         ERR("Failed to initialize shader buffer.\n");
4403         return 0;
4404     }
4405
4406     ret = shader_arb_generate_pshader(shader, gl_info, &buffer, args,
4407             &shader_data->gl_shaders[shader_data->num_gl_shaders]);
4408     shader_buffer_free(&buffer);
4409     shader_data->gl_shaders[shader_data->num_gl_shaders].prgId = ret;
4410
4411     return &shader_data->gl_shaders[shader_data->num_gl_shaders++];
4412 }
4413
4414 static inline BOOL vs_args_equal(const struct arb_vs_compile_args *stored, const struct arb_vs_compile_args *new,
4415                                  const DWORD use_map, BOOL skip_int) {
4416     if((stored->super.swizzle_map & use_map) != new->super.swizzle_map) return FALSE;
4417     if(stored->super.clip_enabled != new->super.clip_enabled) return FALSE;
4418     if(stored->super.fog_src != new->super.fog_src) return FALSE;
4419     if(stored->clip.boolclip_compare != new->clip.boolclip_compare) return FALSE;
4420     if(stored->ps_signature != new->ps_signature) return FALSE;
4421     if(stored->vertex.samplers_compare != new->vertex.samplers_compare) return FALSE;
4422     if(skip_int) return TRUE;
4423
4424     return !memcmp(stored->loop_ctrl, new->loop_ctrl, sizeof(stored->loop_ctrl));
4425 }
4426
4427 static struct arb_vs_compiled_shader *find_arb_vshader(struct wined3d_shader *shader,
4428         const struct arb_vs_compile_args *args)
4429 {
4430     struct wined3d_device *device = shader->device;
4431     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4432     DWORD use_map = device->strided_streams.use_map;
4433     UINT i;
4434     DWORD new_size;
4435     struct arb_vs_compiled_shader *new_array;
4436     struct wined3d_shader_buffer buffer;
4437     struct arb_vshader_private *shader_data;
4438     GLuint ret;
4439
4440     if (!shader->backend_data)
4441     {
4442         const struct wined3d_shader_reg_maps *reg_maps = &shader->reg_maps;
4443
4444         shader->backend_data = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data));
4445         shader_data = shader->backend_data;
4446
4447         if ((gl_info->quirks & WINED3D_QUIRK_ARB_VS_OFFSET_LIMIT)
4448                 && reg_maps->min_rel_offset <= reg_maps->max_rel_offset)
4449         {
4450             if (reg_maps->max_rel_offset - reg_maps->min_rel_offset > 127)
4451             {
4452                 FIXME("The difference between the minimum and maximum relative offset is > 127.\n");
4453                 FIXME("Which this OpenGL implementation does not support. Try using GLSL.\n");
4454                 FIXME("Min: %u, Max: %u.\n", reg_maps->min_rel_offset, reg_maps->max_rel_offset);
4455             }
4456             else if (reg_maps->max_rel_offset - reg_maps->min_rel_offset > 63)
4457                 shader_data->rel_offset = reg_maps->min_rel_offset + 63;
4458             else if (reg_maps->max_rel_offset > 63)
4459                 shader_data->rel_offset = reg_maps->min_rel_offset;
4460         }
4461     }
4462     shader_data = shader->backend_data;
4463
4464     /* Usually we have very few GL shaders for each d3d shader(just 1 or maybe 2),
4465      * so a linear search is more performant than a hashmap or a binary search
4466      * (cache coherency etc)
4467      */
4468     for(i = 0; i < shader_data->num_gl_shaders; i++) {
4469         if (vs_args_equal(&shader_data->gl_shaders[i].args, args,
4470                 use_map, gl_info->supported[NV_VERTEX_PROGRAM2_OPTION]))
4471         {
4472             return &shader_data->gl_shaders[i];
4473         }
4474     }
4475
4476     TRACE("No matching GL shader found, compiling a new shader\n");
4477
4478     if(shader_data->shader_array_size == shader_data->num_gl_shaders) {
4479         if (shader_data->num_gl_shaders)
4480         {
4481             new_size = shader_data->shader_array_size + max(1, shader_data->shader_array_size / 2);
4482             new_array = HeapReAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, shader_data->gl_shaders,
4483                                     new_size * sizeof(*shader_data->gl_shaders));
4484         } else {
4485             new_array = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data->gl_shaders));
4486             new_size = 1;
4487         }
4488
4489         if(!new_array) {
4490             ERR("Out of memory\n");
4491             return 0;
4492         }
4493         shader_data->gl_shaders = new_array;
4494         shader_data->shader_array_size = new_size;
4495     }
4496
4497     shader_data->gl_shaders[shader_data->num_gl_shaders].args = *args;
4498
4499     if (!shader_buffer_init(&buffer))
4500     {
4501         ERR("Failed to initialize shader buffer.\n");
4502         return 0;
4503     }
4504
4505     ret = shader_arb_generate_vshader(shader, gl_info, &buffer, args,
4506             &shader_data->gl_shaders[shader_data->num_gl_shaders]);
4507     shader_buffer_free(&buffer);
4508     shader_data->gl_shaders[shader_data->num_gl_shaders].prgId = ret;
4509
4510     return &shader_data->gl_shaders[shader_data->num_gl_shaders++];
4511 }
4512
4513 static void find_arb_ps_compile_args(const struct wined3d_state *state,
4514         const struct wined3d_shader *shader, struct arb_ps_compile_args *args)
4515 {
4516     struct wined3d_device *device = shader->device;
4517     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4518     int i;
4519     WORD int_skip;
4520
4521     find_ps_compile_args(state, shader, &args->super);
4522
4523     /* This forces all local boolean constants to 1 to make them stateblock independent */
4524     args->bools = shader->reg_maps.local_bool_consts;
4525
4526     for(i = 0; i < MAX_CONST_B; i++)
4527     {
4528         if (state->ps_consts_b[i])
4529             args->bools |= ( 1 << i);
4530     }
4531
4532     /* Only enable the clip plane emulation KIL if at least one clipplane is enabled. The KIL instruction
4533      * is quite expensive because it forces the driver to disable early Z discards. It is cheaper to
4534      * duplicate the shader than have a no-op KIL instruction in every shader
4535      */
4536     if (!device->vs_clipping && use_vs(state)
4537             && state->render_states[WINED3D_RS_CLIPPING]
4538             && state->render_states[WINED3D_RS_CLIPPLANEENABLE])
4539         args->clip = 1;
4540     else
4541         args->clip = 0;
4542
4543     /* Skip if unused or local, or supported natively */
4544     int_skip = ~shader->reg_maps.integer_constants | shader->reg_maps.local_int_consts;
4545     if (int_skip == 0xffff || gl_info->supported[NV_FRAGMENT_PROGRAM_OPTION])
4546     {
4547         memset(args->loop_ctrl, 0, sizeof(args->loop_ctrl));
4548         return;
4549     }
4550
4551     for(i = 0; i < MAX_CONST_I; i++)
4552     {
4553         if(int_skip & (1 << i))
4554         {
4555             args->loop_ctrl[i][0] = 0;
4556             args->loop_ctrl[i][1] = 0;
4557             args->loop_ctrl[i][2] = 0;
4558         }
4559         else
4560         {
4561             args->loop_ctrl[i][0] = state->ps_consts_i[i * 4];
4562             args->loop_ctrl[i][1] = state->ps_consts_i[i * 4 + 1];
4563             args->loop_ctrl[i][2] = state->ps_consts_i[i * 4 + 2];
4564         }
4565     }
4566 }
4567
4568 static void find_arb_vs_compile_args(const struct wined3d_state *state,
4569         const struct wined3d_shader *shader, struct arb_vs_compile_args *args)
4570 {
4571     struct wined3d_device *device = shader->device;
4572     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4573     int i;
4574     WORD int_skip;
4575
4576     find_vs_compile_args(state, shader, &args->super);
4577
4578     args->clip.boolclip_compare = 0;
4579     if (use_ps(state))
4580     {
4581         const struct wined3d_shader *ps = state->pixel_shader;
4582         const struct arb_pshader_private *shader_priv = ps->backend_data;
4583         args->ps_signature = shader_priv->input_signature_idx;
4584
4585         args->clip.boolclip.clip_texcoord = shader_priv->clipplane_emulation + 1;
4586     }
4587     else
4588     {
4589         args->ps_signature = ~0;
4590         if (!device->vs_clipping && device->adapter->fragment_pipe == &arbfp_fragment_pipeline)
4591         {
4592             args->clip.boolclip.clip_texcoord = ffp_clip_emul(state) ? gl_info->limits.texture_stages : 0;
4593         }
4594         /* Otherwise: Setting boolclip_compare set clip_texcoord to 0 */
4595     }
4596
4597     if (args->clip.boolclip.clip_texcoord)
4598     {
4599         if (state->render_states[WINED3D_RS_CLIPPING])
4600             args->clip.boolclip.clipplane_mask = (unsigned char)state->render_states[WINED3D_RS_CLIPPLANEENABLE];
4601         /* clipplane_mask was set to 0 by setting boolclip_compare to 0 */
4602     }
4603
4604     /* This forces all local boolean constants to 1 to make them stateblock independent */
4605     args->clip.boolclip.bools = shader->reg_maps.local_bool_consts;
4606     /* TODO: Figure out if it would be better to store bool constants as bitmasks in the stateblock */
4607     for(i = 0; i < MAX_CONST_B; i++)
4608     {
4609         if (state->vs_consts_b[i])
4610             args->clip.boolclip.bools |= ( 1 << i);
4611     }
4612
4613     args->vertex.samplers[0] = device->texUnitMap[MAX_FRAGMENT_SAMPLERS + 0];
4614     args->vertex.samplers[1] = device->texUnitMap[MAX_FRAGMENT_SAMPLERS + 1];
4615     args->vertex.samplers[2] = device->texUnitMap[MAX_FRAGMENT_SAMPLERS + 2];
4616     args->vertex.samplers[3] = 0;
4617
4618     /* Skip if unused or local */
4619     int_skip = ~shader->reg_maps.integer_constants | shader->reg_maps.local_int_consts;
4620     /* This is about flow control, not clipping. */
4621     if (int_skip == 0xffff || gl_info->supported[NV_VERTEX_PROGRAM2_OPTION])
4622     {
4623         memset(args->loop_ctrl, 0, sizeof(args->loop_ctrl));
4624         return;
4625     }
4626
4627     for(i = 0; i < MAX_CONST_I; i++)
4628     {
4629         if(int_skip & (1 << i))
4630         {
4631             args->loop_ctrl[i][0] = 0;
4632             args->loop_ctrl[i][1] = 0;
4633             args->loop_ctrl[i][2] = 0;
4634         }
4635         else
4636         {
4637             args->loop_ctrl[i][0] = state->vs_consts_i[i * 4];
4638             args->loop_ctrl[i][1] = state->vs_consts_i[i * 4 + 1];
4639             args->loop_ctrl[i][2] = state->vs_consts_i[i * 4 + 2];
4640         }
4641     }
4642 }
4643
4644 /* GL locking is done by the caller */
4645 static void shader_arb_select(const struct wined3d_context *context, BOOL usePS, BOOL useVS)
4646 {
4647     struct wined3d_device *device = context->swapchain->device;
4648     struct shader_arb_priv *priv = device->shader_priv;
4649     const struct wined3d_gl_info *gl_info = context->gl_info;
4650     const struct wined3d_state *state = &device->stateBlock->state;
4651     int i;
4652
4653     /* Deal with pixel shaders first so the vertex shader arg function has the input signature ready */
4654     if (usePS)
4655     {
4656         struct wined3d_shader *ps = state->pixel_shader;
4657         struct arb_ps_compile_args compile_args;
4658         struct arb_ps_compiled_shader *compiled;
4659
4660         TRACE("Using pixel shader %p.\n", ps);
4661         find_arb_ps_compile_args(state, ps, &compile_args);
4662         compiled = find_arb_pshader(ps, &compile_args);
4663         priv->current_fprogram_id = compiled->prgId;
4664         priv->compiled_fprog = compiled;
4665
4666         /* Bind the fragment program */
4667         GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, priv->current_fprogram_id));
4668         checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, priv->current_fprogram_id);");
4669
4670         if (!priv->use_arbfp_fixed_func)
4671         {
4672             /* Enable OpenGL fragment programs. */
4673             gl_info->gl_ops.gl.p_glEnable(GL_FRAGMENT_PROGRAM_ARB);
4674             checkGLcall("glEnable(GL_FRAGMENT_PROGRAM_ARB);");
4675         }
4676         TRACE("(%p) : Bound fragment program %u and enabled GL_FRAGMENT_PROGRAM_ARB\n",
4677                 device, priv->current_fprogram_id);
4678
4679         /* Pixel Shader 1.x constants are clamped to [-1;1], Pixel Shader 2.0 constants are not. If switching between
4680          * a 1.x and newer shader, reload the first 8 constants
4681          */
4682         if (priv->last_ps_const_clamped != ((struct arb_pshader_private *)ps->backend_data)->clamp_consts)
4683         {
4684             priv->last_ps_const_clamped = ((struct arb_pshader_private *)ps->backend_data)->clamp_consts;
4685             priv->highest_dirty_ps_const = max(priv->highest_dirty_ps_const, 8);
4686             for(i = 0; i < 8; i++)
4687             {
4688                 priv->pshader_const_dirty[i] = 1;
4689             }
4690             /* Also takes care of loading local constants */
4691             shader_arb_load_constants_internal(context, TRUE, FALSE, TRUE);
4692         }
4693         else
4694         {
4695             UINT rt_height = state->fb->render_targets[0]->resource.height;
4696             shader_arb_ps_local_constants(compiled, context, state, rt_height);
4697         }
4698
4699         /* Force constant reloading for the NP2 fixup (see comment in shader_glsl_select for more info) */
4700         if (compiled->np2fixup_info.super.active)
4701             shader_arb_load_np2fixup_constants(priv, gl_info, state);
4702     }
4703     else if (gl_info->supported[ARB_FRAGMENT_PROGRAM] && !priv->use_arbfp_fixed_func)
4704     {
4705         /* Disable only if we're not using arbfp fixed function fragment processing. If this is used,
4706         * keep GL_FRAGMENT_PROGRAM_ARB enabled, and the fixed function pipeline will bind the fixed function
4707         * replacement shader
4708         */
4709         gl_info->gl_ops.gl.p_glDisable(GL_FRAGMENT_PROGRAM_ARB);
4710         checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
4711         priv->current_fprogram_id = 0;
4712     }
4713
4714     if (useVS)
4715     {
4716         struct wined3d_shader *vs = state->vertex_shader;
4717         struct arb_vs_compile_args compile_args;
4718         struct arb_vs_compiled_shader *compiled;
4719
4720         TRACE("Using vertex shader %p\n", vs);
4721         find_arb_vs_compile_args(state, vs, &compile_args);
4722         compiled = find_arb_vshader(vs, &compile_args);
4723         priv->current_vprogram_id = compiled->prgId;
4724         priv->compiled_vprog = compiled;
4725
4726         /* Bind the vertex program */
4727         GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->current_vprogram_id));
4728         checkGLcall("glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->current_vprogram_id);");
4729
4730         /* Enable OpenGL vertex programs */
4731         gl_info->gl_ops.gl.p_glEnable(GL_VERTEX_PROGRAM_ARB);
4732         checkGLcall("glEnable(GL_VERTEX_PROGRAM_ARB);");
4733         TRACE("(%p) : Bound vertex program %u and enabled GL_VERTEX_PROGRAM_ARB\n", device, priv->current_vprogram_id);
4734         shader_arb_vs_local_constants(compiled, context, state);
4735
4736         if(priv->last_vs_color_unclamp != compiled->need_color_unclamp) {
4737             priv->last_vs_color_unclamp = compiled->need_color_unclamp;
4738
4739             if (gl_info->supported[ARB_COLOR_BUFFER_FLOAT])
4740             {
4741                 GL_EXTCALL(glClampColorARB(GL_CLAMP_VERTEX_COLOR_ARB, !compiled->need_color_unclamp));
4742                 checkGLcall("glClampColorARB");
4743             } else {
4744                 FIXME("vertex color clamp needs to be changed, but extension not supported.\n");
4745             }
4746         }
4747     }
4748     else if (gl_info->supported[ARB_VERTEX_PROGRAM])
4749     {
4750         priv->current_vprogram_id = 0;
4751         gl_info->gl_ops.gl.p_glDisable(GL_VERTEX_PROGRAM_ARB);
4752         checkGLcall("glDisable(GL_VERTEX_PROGRAM_ARB)");
4753     }
4754 }
4755
4756 /* GL locking is done by the caller */
4757 static void shader_arb_select_depth_blt(void *shader_priv, const struct wined3d_gl_info *gl_info,
4758         enum tex_types tex_type, const SIZE *ds_mask_size)
4759 {
4760     const float mask[] = {0.0f, 0.0f, (float)ds_mask_size->cx, (float)ds_mask_size->cy};
4761     BOOL masked = ds_mask_size->cx && ds_mask_size->cy;
4762     struct shader_arb_priv *priv = shader_priv;
4763     GLuint *blt_fprogram;
4764
4765     if (!priv->depth_blt_vprogram_id) priv->depth_blt_vprogram_id = create_arb_blt_vertex_program(gl_info);
4766     GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->depth_blt_vprogram_id));
4767     gl_info->gl_ops.gl.p_glEnable(GL_VERTEX_PROGRAM_ARB);
4768
4769     blt_fprogram = masked ? &priv->depth_blt_fprogram_id_masked[tex_type] : &priv->depth_blt_fprogram_id_full[tex_type];
4770     if (!*blt_fprogram) *blt_fprogram = create_arb_blt_fragment_program(gl_info, tex_type, masked);
4771     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, *blt_fprogram));
4772     if (masked) GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, 0, mask));
4773     gl_info->gl_ops.gl.p_glEnable(GL_FRAGMENT_PROGRAM_ARB);
4774 }
4775
4776 /* GL locking is done by the caller */
4777 static void shader_arb_deselect_depth_blt(void *shader_priv, const struct wined3d_gl_info *gl_info)
4778 {
4779     struct shader_arb_priv *priv = shader_priv;
4780
4781     if (priv->current_vprogram_id) {
4782         GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->current_vprogram_id));
4783         checkGLcall("glBindProgramARB(GL_VERTEX_PROGRAM_ARB, vertexShader->prgId);");
4784
4785         TRACE("Bound vertex program %u and enabled GL_VERTEX_PROGRAM_ARB.\n", priv->current_vprogram_id);
4786     }
4787     else
4788     {
4789         gl_info->gl_ops.gl.p_glDisable(GL_VERTEX_PROGRAM_ARB);
4790         checkGLcall("glDisable(GL_VERTEX_PROGRAM_ARB)");
4791     }
4792
4793     if (priv->current_fprogram_id) {
4794         GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, priv->current_fprogram_id));
4795         checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, pixelShader->prgId);");
4796
4797         TRACE("Bound fragment program %u and enabled GL_FRAGMENT_PROGRAM_ARB.\n", priv->current_fprogram_id);
4798     }
4799     else if(!priv->use_arbfp_fixed_func)
4800     {
4801         gl_info->gl_ops.gl.p_glDisable(GL_FRAGMENT_PROGRAM_ARB);
4802         checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
4803     }
4804 }
4805
4806 static void shader_arb_destroy(struct wined3d_shader *shader)
4807 {
4808     struct wined3d_device *device = shader->device;
4809     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4810
4811     if (shader_is_pshader_version(shader->reg_maps.shader_version.type))
4812     {
4813         struct arb_pshader_private *shader_data = shader->backend_data;
4814         UINT i;
4815
4816         if(!shader_data) return; /* This can happen if a shader was never compiled */
4817
4818         if (shader_data->num_gl_shaders)
4819         {
4820             struct wined3d_context *context = context_acquire(device, NULL);
4821
4822             ENTER_GL();
4823             for (i = 0; i < shader_data->num_gl_shaders; ++i)
4824             {
4825                 GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId));
4826                 checkGLcall("GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId))");
4827             }
4828             LEAVE_GL();
4829
4830             context_release(context);
4831         }
4832
4833         HeapFree(GetProcessHeap(), 0, shader_data->gl_shaders);
4834         HeapFree(GetProcessHeap(), 0, shader_data);
4835         shader->backend_data = NULL;
4836     }
4837     else
4838     {
4839         struct arb_vshader_private *shader_data = shader->backend_data;
4840         UINT i;
4841
4842         if(!shader_data) return; /* This can happen if a shader was never compiled */
4843
4844         if (shader_data->num_gl_shaders)
4845         {
4846             struct wined3d_context *context = context_acquire(device, NULL);
4847
4848             ENTER_GL();
4849             for (i = 0; i < shader_data->num_gl_shaders; ++i)
4850             {
4851                 GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId));
4852                 checkGLcall("GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId))");
4853             }
4854             LEAVE_GL();
4855
4856             context_release(context);
4857         }
4858
4859         HeapFree(GetProcessHeap(), 0, shader_data->gl_shaders);
4860         HeapFree(GetProcessHeap(), 0, shader_data);
4861         shader->backend_data = NULL;
4862     }
4863 }
4864
4865 static int sig_tree_compare(const void *key, const struct wine_rb_entry *entry)
4866 {
4867     struct ps_signature *e = WINE_RB_ENTRY_VALUE(entry, struct ps_signature, entry);
4868     return compare_sig(key, e->sig);
4869 }
4870
4871 static const struct wine_rb_functions sig_tree_functions =
4872 {
4873     wined3d_rb_alloc,
4874     wined3d_rb_realloc,
4875     wined3d_rb_free,
4876     sig_tree_compare
4877 };
4878
4879 static HRESULT shader_arb_alloc(struct wined3d_device *device)
4880 {
4881     struct shader_arb_priv *priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*priv));
4882
4883     priv->vshader_const_dirty = HeapAlloc(GetProcessHeap(), 0,
4884             sizeof(*priv->vshader_const_dirty) * device->d3d_vshader_constantF);
4885     if (!priv->vshader_const_dirty)
4886         goto fail;
4887     memset(priv->vshader_const_dirty, 1,
4888            sizeof(*priv->vshader_const_dirty) * device->d3d_vshader_constantF);
4889
4890     priv->pshader_const_dirty = HeapAlloc(GetProcessHeap(), 0,
4891             sizeof(*priv->pshader_const_dirty) * device->d3d_pshader_constantF);
4892     if (!priv->pshader_const_dirty)
4893         goto fail;
4894     memset(priv->pshader_const_dirty, 1,
4895             sizeof(*priv->pshader_const_dirty) * device->d3d_pshader_constantF);
4896
4897     if(wine_rb_init(&priv->signature_tree, &sig_tree_functions) == -1)
4898     {
4899         ERR("RB tree init failed\n");
4900         goto fail;
4901     }
4902     device->shader_priv = priv;
4903     return WINED3D_OK;
4904
4905 fail:
4906     HeapFree(GetProcessHeap(), 0, priv->pshader_const_dirty);
4907     HeapFree(GetProcessHeap(), 0, priv->vshader_const_dirty);
4908     HeapFree(GetProcessHeap(), 0, priv);
4909     return E_OUTOFMEMORY;
4910 }
4911
4912 static void release_signature(struct wine_rb_entry *entry, void *context)
4913 {
4914     struct ps_signature *sig = WINE_RB_ENTRY_VALUE(entry, struct ps_signature, entry);
4915     int i;
4916     for(i = 0; i < MAX_REG_INPUT; i++)
4917     {
4918         HeapFree(GetProcessHeap(), 0, (char *) sig->sig[i].semantic_name);
4919     }
4920     HeapFree(GetProcessHeap(), 0, sig->sig);
4921     HeapFree(GetProcessHeap(), 0, sig);
4922 }
4923
4924 /* Context activation is done by the caller. */
4925 static void shader_arb_free(struct wined3d_device *device)
4926 {
4927     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4928     struct shader_arb_priv *priv = device->shader_priv;
4929     int i;
4930
4931     ENTER_GL();
4932     if(priv->depth_blt_vprogram_id) {
4933         GL_EXTCALL(glDeleteProgramsARB(1, &priv->depth_blt_vprogram_id));
4934     }
4935     for (i = 0; i < tex_type_count; ++i)
4936     {
4937         if (priv->depth_blt_fprogram_id_full[i])
4938         {
4939             GL_EXTCALL(glDeleteProgramsARB(1, &priv->depth_blt_fprogram_id_full[i]));
4940         }
4941         if (priv->depth_blt_fprogram_id_masked[i])
4942         {
4943             GL_EXTCALL(glDeleteProgramsARB(1, &priv->depth_blt_fprogram_id_masked[i]));
4944         }
4945     }
4946     LEAVE_GL();
4947
4948     wine_rb_destroy(&priv->signature_tree, release_signature, NULL);
4949     HeapFree(GetProcessHeap(), 0, priv->pshader_const_dirty);
4950     HeapFree(GetProcessHeap(), 0, priv->vshader_const_dirty);
4951     HeapFree(GetProcessHeap(), 0, device->shader_priv);
4952 }
4953
4954 static void shader_arb_context_destroyed(void *shader_priv, const struct wined3d_context *context)
4955 {
4956     struct shader_arb_priv *priv = shader_priv;
4957
4958     if (priv->last_context == context)
4959         priv->last_context = NULL;
4960 }
4961
4962 static void shader_arb_get_caps(const struct wined3d_gl_info *gl_info, struct shader_caps *caps)
4963 {
4964     if (gl_info->supported[ARB_VERTEX_PROGRAM])
4965     {
4966         DWORD vs_consts;
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             caps->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             caps->vs_version = 2;
4986             TRACE("Hardware vertex shader version 2.0 enabled (ARB_PROGRAM)\n");
4987         }
4988         else
4989         {
4990             caps->vs_version = 1;
4991             TRACE("Hardware vertex shader version 1.1 enabled (ARB_PROGRAM)\n");
4992         }
4993         caps->vs_uniform_count = vs_consts;
4994     }
4995     else
4996     {
4997         caps->vs_version = 0;
4998         caps->vs_uniform_count = 0;
4999     }
5000
5001     caps->gs_version = 0;
5002
5003     if (gl_info->supported[ARB_FRAGMENT_PROGRAM])
5004     {
5005         DWORD ps_consts;
5006
5007         /* Similar as above for vertex programs, but the minimum for fragment
5008          * programs is 24. */
5009         if (gl_info->limits.arb_ps_native_constants < 24)
5010             ps_consts = gl_info->limits.arb_ps_float_constants;
5011         else
5012             ps_consts = min(gl_info->limits.arb_ps_float_constants, gl_info->limits.arb_ps_native_constants);
5013
5014         if (gl_info->supported[NV_FRAGMENT_PROGRAM2])
5015         {
5016             caps->ps_version = 3;
5017             TRACE("Hardware pixel shader version 3.0 enabled (NV_FRAGMENT_PROGRAM2)\n");
5018         }
5019         else if (ps_consts >= 32)
5020         {
5021             /* Shader Model 2.0 requires at least 32 pixel shader constants */
5022             caps->ps_version = 2;
5023             TRACE("Hardware pixel shader version 2.0 enabled (ARB_PROGRAM)\n");
5024         }
5025         else
5026         {
5027             caps->ps_version = 1;
5028             TRACE("Hardware pixel shader version 1.4 enabled (ARB_PROGRAM)\n");
5029         }
5030         caps->ps_uniform_count = ps_consts;
5031         caps->ps_1x_max_value = 8.0f;
5032     }
5033     else
5034     {
5035         caps->ps_version = 0;
5036         caps->ps_uniform_count = 0;
5037         caps->ps_1x_max_value = 0.0f;
5038     }
5039
5040     caps->vs_clipping = use_nv_clip(gl_info);
5041 }
5042
5043 static BOOL shader_arb_color_fixup_supported(struct color_fixup_desc fixup)
5044 {
5045     if (TRACE_ON(d3d_shader) && TRACE_ON(d3d))
5046     {
5047         TRACE("Checking support for color_fixup:\n");
5048         dump_color_fixup_desc(fixup);
5049     }
5050
5051     /* We support everything except complex conversions. */
5052     if (!is_complex_fixup(fixup))
5053     {
5054         TRACE("[OK]\n");
5055         return TRUE;
5056     }
5057
5058     TRACE("[FAILED]\n");
5059     return FALSE;
5060 }
5061
5062 static void shader_arb_add_instruction_modifiers(const struct wined3d_shader_instruction *ins) {
5063     DWORD shift;
5064     char write_mask[20], regstr[50];
5065     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
5066     BOOL is_color = FALSE;
5067     const struct wined3d_shader_dst_param *dst;
5068
5069     if (!ins->dst_count) return;
5070
5071     dst = &ins->dst[0];
5072     shift = dst->shift;
5073     if (!shift) return; /* Saturate alone is handled by the instructions */
5074
5075     shader_arb_get_write_mask(ins, dst, write_mask);
5076     shader_arb_get_register_name(ins, &dst->reg, regstr, &is_color);
5077
5078     /* Generate a line that does the output modifier computation
5079      * FIXME: _SAT vs shift? _SAT alone is already handled in the instructions, if this
5080      * maps problems in e.g. _d4_sat modify shader_arb_get_modifier
5081      */
5082     shader_addline(buffer, "MUL%s %s%s, %s, %s;\n", shader_arb_get_modifier(ins),
5083                    regstr, write_mask, regstr, shift_tab[shift]);
5084 }
5085
5086 static const SHADER_HANDLER shader_arb_instruction_handler_table[WINED3DSIH_TABLE_SIZE] =
5087 {
5088     /* WINED3DSIH_ABS                   */ shader_hw_map2gl,
5089     /* WINED3DSIH_ADD                   */ shader_hw_map2gl,
5090     /* WINED3DSIH_AND                   */ NULL,
5091     /* WINED3DSIH_BEM                   */ pshader_hw_bem,
5092     /* WINED3DSIH_BREAK                 */ shader_hw_break,
5093     /* WINED3DSIH_BREAKC                */ shader_hw_breakc,
5094     /* WINED3DSIH_BREAKP                */ NULL,
5095     /* WINED3DSIH_CALL                  */ shader_hw_call,
5096     /* WINED3DSIH_CALLNZ                */ NULL,
5097     /* WINED3DSIH_CMP                   */ pshader_hw_cmp,
5098     /* WINED3DSIH_CND                   */ pshader_hw_cnd,
5099     /* WINED3DSIH_CRS                   */ shader_hw_map2gl,
5100     /* WINED3DSIH_CUT                   */ NULL,
5101     /* WINED3DSIH_DCL                   */ shader_hw_nop,
5102     /* WINED3DSIH_DCL_CONSTANT_BUFFER   */ shader_hw_nop,
5103     /* WINED3DSIH_DCL_INPUT_PRIMITIVE   */ shader_hw_nop,
5104     /* WINED3DSIH_DCL_OUTPUT_TOPOLOGY   */ shader_hw_nop,
5105     /* WINED3DSIH_DCL_VERTICES_OUT      */ shader_hw_nop,
5106     /* WINED3DSIH_DEF                   */ shader_hw_nop,
5107     /* WINED3DSIH_DEFB                  */ shader_hw_nop,
5108     /* WINED3DSIH_DEFI                  */ shader_hw_nop,
5109     /* WINED3DSIH_DIV                   */ NULL,
5110     /* WINED3DSIH_DP2ADD                */ pshader_hw_dp2add,
5111     /* WINED3DSIH_DP3                   */ shader_hw_map2gl,
5112     /* WINED3DSIH_DP4                   */ shader_hw_map2gl,
5113     /* WINED3DSIH_DST                   */ shader_hw_map2gl,
5114     /* WINED3DSIH_DSX                   */ shader_hw_map2gl,
5115     /* WINED3DSIH_DSY                   */ shader_hw_dsy,
5116     /* WINED3DSIH_ELSE                  */ shader_hw_else,
5117     /* WINED3DSIH_EMIT                  */ NULL,
5118     /* WINED3DSIH_ENDIF                 */ shader_hw_endif,
5119     /* WINED3DSIH_ENDLOOP               */ shader_hw_endloop,
5120     /* WINED3DSIH_ENDREP                */ shader_hw_endrep,
5121     /* WINED3DSIH_EQ                    */ NULL,
5122     /* WINED3DSIH_EXP                   */ shader_hw_scalar_op,
5123     /* WINED3DSIH_EXPP                  */ shader_hw_scalar_op,
5124     /* WINED3DSIH_FRC                   */ shader_hw_map2gl,
5125     /* WINED3DSIH_FTOI                  */ NULL,
5126     /* WINED3DSIH_GE                    */ NULL,
5127     /* WINED3DSIH_IADD                  */ NULL,
5128     /* WINED3DSIH_IEQ                   */ NULL,
5129     /* WINED3DSIH_IF                    */ NULL /* Hardcoded into the shader */,
5130     /* WINED3DSIH_IFC                   */ shader_hw_ifc,
5131     /* WINED3DSIH_IGE                   */ NULL,
5132     /* WINED3DSIH_IMUL                  */ NULL,
5133     /* WINED3DSIH_ITOF                  */ NULL,
5134     /* WINED3DSIH_LABEL                 */ shader_hw_label,
5135     /* WINED3DSIH_LD                    */ NULL,
5136     /* WINED3DSIH_LIT                   */ shader_hw_map2gl,
5137     /* WINED3DSIH_LOG                   */ shader_hw_log,
5138     /* WINED3DSIH_LOGP                  */ shader_hw_log,
5139     /* WINED3DSIH_LOOP                  */ shader_hw_loop,
5140     /* WINED3DSIH_LRP                   */ shader_hw_lrp,
5141     /* WINED3DSIH_LT                    */ NULL,
5142     /* WINED3DSIH_M3x2                  */ shader_hw_mnxn,
5143     /* WINED3DSIH_M3x3                  */ shader_hw_mnxn,
5144     /* WINED3DSIH_M3x4                  */ shader_hw_mnxn,
5145     /* WINED3DSIH_M4x3                  */ shader_hw_mnxn,
5146     /* WINED3DSIH_M4x4                  */ shader_hw_mnxn,
5147     /* WINED3DSIH_MAD                   */ shader_hw_map2gl,
5148     /* WINED3DSIH_MAX                   */ shader_hw_map2gl,
5149     /* WINED3DSIH_MIN                   */ shader_hw_map2gl,
5150     /* WINED3DSIH_MOV                   */ shader_hw_mov,
5151     /* WINED3DSIH_MOVA                  */ shader_hw_mov,
5152     /* WINED3DSIH_MOVC                  */ NULL,
5153     /* WINED3DSIH_MUL                   */ shader_hw_map2gl,
5154     /* WINED3DSIH_NOP                   */ shader_hw_nop,
5155     /* WINED3DSIH_NRM                   */ shader_hw_nrm,
5156     /* WINED3DSIH_PHASE                 */ shader_hw_nop,
5157     /* WINED3DSIH_POW                   */ shader_hw_pow,
5158     /* WINED3DSIH_RCP                   */ shader_hw_rcp,
5159     /* WINED3DSIH_REP                   */ shader_hw_rep,
5160     /* WINED3DSIH_RET                   */ shader_hw_ret,
5161     /* WINED3DSIH_ROUND_NI              */ NULL,
5162     /* WINED3DSIH_RSQ                   */ shader_hw_scalar_op,
5163     /* WINED3DSIH_SAMPLE                */ NULL,
5164     /* WINED3DSIH_SAMPLE_GRAD           */ NULL,
5165     /* WINED3DSIH_SAMPLE_LOD            */ NULL,
5166     /* WINED3DSIH_SETP                  */ NULL,
5167     /* WINED3DSIH_SGE                   */ shader_hw_map2gl,
5168     /* WINED3DSIH_SGN                   */ shader_hw_sgn,
5169     /* WINED3DSIH_SINCOS                */ shader_hw_sincos,
5170     /* WINED3DSIH_SLT                   */ shader_hw_map2gl,
5171     /* WINED3DSIH_SQRT                  */ NULL,
5172     /* WINED3DSIH_SUB                   */ shader_hw_map2gl,
5173     /* WINED3DSIH_TEX                   */ pshader_hw_tex,
5174     /* WINED3DSIH_TEXBEM                */ pshader_hw_texbem,
5175     /* WINED3DSIH_TEXBEML               */ pshader_hw_texbem,
5176     /* WINED3DSIH_TEXCOORD              */ pshader_hw_texcoord,
5177     /* WINED3DSIH_TEXDEPTH              */ pshader_hw_texdepth,
5178     /* WINED3DSIH_TEXDP3                */ pshader_hw_texdp3,
5179     /* WINED3DSIH_TEXDP3TEX             */ pshader_hw_texdp3tex,
5180     /* WINED3DSIH_TEXKILL               */ pshader_hw_texkill,
5181     /* WINED3DSIH_TEXLDD                */ shader_hw_texldd,
5182     /* WINED3DSIH_TEXLDL                */ shader_hw_texldl,
5183     /* WINED3DSIH_TEXM3x2DEPTH          */ pshader_hw_texm3x2depth,
5184     /* WINED3DSIH_TEXM3x2PAD            */ pshader_hw_texm3x2pad,
5185     /* WINED3DSIH_TEXM3x2TEX            */ pshader_hw_texm3x2tex,
5186     /* WINED3DSIH_TEXM3x3               */ pshader_hw_texm3x3,
5187     /* WINED3DSIH_TEXM3x3DIFF           */ NULL,
5188     /* WINED3DSIH_TEXM3x3PAD            */ pshader_hw_texm3x3pad,
5189     /* WINED3DSIH_TEXM3x3SPEC           */ pshader_hw_texm3x3spec,
5190     /* WINED3DSIH_TEXM3x3TEX            */ pshader_hw_texm3x3tex,
5191     /* WINED3DSIH_TEXM3x3VSPEC          */ pshader_hw_texm3x3vspec,
5192     /* WINED3DSIH_TEXREG2AR             */ pshader_hw_texreg2ar,
5193     /* WINED3DSIH_TEXREG2GB             */ pshader_hw_texreg2gb,
5194     /* WINED3DSIH_TEXREG2RGB            */ pshader_hw_texreg2rgb,
5195     /* WINED3DSIH_UDIV                  */ NULL,
5196     /* WINED3DSIH_USHR                  */ NULL,
5197     /* WINED3DSIH_UTOF                  */ NULL,
5198     /* WINED3DSIH_XOR                   */ NULL,
5199 };
5200
5201 static BOOL get_bool_const(const struct wined3d_shader_instruction *ins,
5202         const struct wined3d_shader *shader, DWORD idx)
5203 {
5204     const struct wined3d_shader_reg_maps *reg_maps = ins->ctx->reg_maps;
5205     BOOL vshader = shader_is_vshader_version(reg_maps->shader_version.type);
5206     const struct wined3d_shader_lconst *constant;
5207     WORD bools = 0;
5208     WORD flag = (1 << idx);
5209     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
5210
5211     if (reg_maps->local_bool_consts & flag)
5212     {
5213         /* What good is a if(bool) with a hardcoded local constant? I don't know, but handle it */
5214         LIST_FOR_EACH_ENTRY(constant, &shader->constantsB, struct wined3d_shader_lconst, entry)
5215         {
5216             if (constant->idx == idx)
5217             {
5218                 return constant->value[0];
5219             }
5220         }
5221         ERR("Local constant not found\n");
5222         return FALSE;
5223     }
5224     else
5225     {
5226         if(vshader) bools = priv->cur_vs_args->clip.boolclip.bools;
5227         else bools = priv->cur_ps_args->bools;
5228         return bools & flag;
5229     }
5230 }
5231
5232 static void get_loop_control_const(const struct wined3d_shader_instruction *ins,
5233         const struct wined3d_shader *shader, UINT idx, struct wined3d_shader_loop_control *loop_control)
5234 {
5235     const struct wined3d_shader_reg_maps *reg_maps = ins->ctx->reg_maps;
5236     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
5237
5238     /* Integer constants can either be a local constant, or they can be stored in the shader
5239      * type specific compile args. */
5240     if (reg_maps->local_int_consts & (1 << idx))
5241     {
5242         const struct wined3d_shader_lconst *constant;
5243
5244         LIST_FOR_EACH_ENTRY(constant, &shader->constantsI, struct wined3d_shader_lconst, entry)
5245         {
5246             if (constant->idx == idx)
5247             {
5248                 loop_control->count = constant->value[0];
5249                 loop_control->start = constant->value[1];
5250                 /* Step is signed. */
5251                 loop_control->step = (int)constant->value[2];
5252                 return;
5253             }
5254         }
5255         /* If this happens the flag was set incorrectly */
5256         ERR("Local constant not found\n");
5257         loop_control->count = 0;
5258         loop_control->start = 0;
5259         loop_control->step = 0;
5260         return;
5261     }
5262
5263     switch (reg_maps->shader_version.type)
5264     {
5265         case WINED3D_SHADER_TYPE_VERTEX:
5266             /* Count and aL start value are unsigned */
5267             loop_control->count = priv->cur_vs_args->loop_ctrl[idx][0];
5268             loop_control->start = priv->cur_vs_args->loop_ctrl[idx][1];
5269             /* Step is signed. */
5270             loop_control->step = ((char)priv->cur_vs_args->loop_ctrl[idx][2]);
5271             break;
5272
5273         case WINED3D_SHADER_TYPE_PIXEL:
5274             loop_control->count = priv->cur_ps_args->loop_ctrl[idx][0];
5275             loop_control->start = priv->cur_ps_args->loop_ctrl[idx][1];
5276             loop_control->step = ((char)priv->cur_ps_args->loop_ctrl[idx][2]);
5277             break;
5278
5279         default:
5280             FIXME("Unhandled shader type %#x.\n", reg_maps->shader_version.type);
5281             break;
5282     }
5283 }
5284
5285 static void record_instruction(struct list *list, const struct wined3d_shader_instruction *ins)
5286 {
5287     unsigned int i;
5288     struct wined3d_shader_dst_param *dst_param = NULL;
5289     struct wined3d_shader_src_param *src_param = NULL, *rel_addr = NULL;
5290     struct recorded_instruction *rec = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*rec));
5291     if(!rec)
5292     {
5293         ERR("Out of memory\n");
5294         return;
5295     }
5296
5297     rec->ins = *ins;
5298     dst_param = HeapAlloc(GetProcessHeap(), 0, sizeof(*dst_param));
5299     if(!dst_param) goto free;
5300     *dst_param = *ins->dst;
5301     if (ins->dst->reg.idx[0].rel_addr)
5302     {
5303         rel_addr = HeapAlloc(GetProcessHeap(), 0, sizeof(*rel_addr));
5304         if (!rel_addr)
5305             goto free;
5306         *rel_addr = *ins->dst->reg.idx[0].rel_addr;
5307         dst_param->reg.idx[0].rel_addr = rel_addr;
5308     }
5309     rec->ins.dst = dst_param;
5310
5311     src_param = HeapAlloc(GetProcessHeap(), 0, sizeof(*src_param) * ins->src_count);
5312     if (!src_param)
5313         goto free;
5314     for (i = 0; i < ins->src_count; ++i)
5315     {
5316         src_param[i] = ins->src[i];
5317         if (ins->src[i].reg.idx[0].rel_addr)
5318         {
5319             rel_addr = HeapAlloc(GetProcessHeap(), 0, sizeof(*rel_addr));
5320             if (!rel_addr)
5321                 goto free;
5322             *rel_addr = *ins->src[i].reg.idx[0].rel_addr;
5323             src_param[i].reg.idx[0].rel_addr = rel_addr;
5324         }
5325     }
5326     rec->ins.src = src_param;
5327     list_add_tail(list, &rec->entry);
5328     return;
5329
5330 free:
5331     ERR("Out of memory\n");
5332     if(dst_param)
5333     {
5334         HeapFree(GetProcessHeap(), 0, (void *)dst_param->reg.idx[0].rel_addr);
5335         HeapFree(GetProcessHeap(), 0, dst_param);
5336     }
5337     if(src_param)
5338     {
5339         for(i = 0; i < ins->src_count; i++)
5340         {
5341             HeapFree(GetProcessHeap(), 0, (void *)src_param[i].reg.idx[0].rel_addr);
5342         }
5343         HeapFree(GetProcessHeap(), 0, src_param);
5344     }
5345     HeapFree(GetProcessHeap(), 0, rec);
5346 }
5347
5348 static void free_recorded_instruction(struct list *list)
5349 {
5350     struct recorded_instruction *rec_ins, *entry2;
5351     unsigned int i;
5352
5353     LIST_FOR_EACH_ENTRY_SAFE(rec_ins, entry2, list, struct recorded_instruction, entry)
5354     {
5355         list_remove(&rec_ins->entry);
5356         if (rec_ins->ins.dst)
5357         {
5358             HeapFree(GetProcessHeap(), 0, (void *)rec_ins->ins.dst->reg.idx[0].rel_addr);
5359             HeapFree(GetProcessHeap(), 0, (void *)rec_ins->ins.dst);
5360         }
5361         if (rec_ins->ins.src)
5362         {
5363             for (i = 0; i < rec_ins->ins.src_count; ++i)
5364             {
5365                 HeapFree(GetProcessHeap(), 0, (void *)rec_ins->ins.src[i].reg.idx[0].rel_addr);
5366             }
5367             HeapFree(GetProcessHeap(), 0, (void *)rec_ins->ins.src);
5368         }
5369         HeapFree(GetProcessHeap(), 0, rec_ins);
5370     }
5371 }
5372
5373 static void shader_arb_handle_instruction(const struct wined3d_shader_instruction *ins) {
5374     SHADER_HANDLER hw_fct;
5375     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
5376     const struct wined3d_shader *shader = ins->ctx->shader;
5377     struct control_frame *control_frame;
5378     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
5379     BOOL bool_const;
5380
5381     if(ins->handler_idx == WINED3DSIH_LOOP || ins->handler_idx == WINED3DSIH_REP)
5382     {
5383         control_frame = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*control_frame));
5384         list_add_head(&priv->control_frames, &control_frame->entry);
5385
5386         if(ins->handler_idx == WINED3DSIH_LOOP) control_frame->type = LOOP;
5387         if(ins->handler_idx == WINED3DSIH_REP) control_frame->type = REP;
5388
5389         if(priv->target_version >= NV2)
5390         {
5391             control_frame->no.loop = priv->num_loops++;
5392             priv->loop_depth++;
5393         }
5394         else
5395         {
5396             /* Don't bother recording when we're in a not used if branch */
5397             if(priv->muted)
5398             {
5399                 return;
5400             }
5401
5402             if(!priv->recording)
5403             {
5404                 list_init(&priv->record);
5405                 priv->recording = TRUE;
5406                 control_frame->outer_loop = TRUE;
5407                 get_loop_control_const(ins, shader, ins->src[0].reg.idx[0].offset, &control_frame->loop_control);
5408                 return; /* Instruction is handled */
5409             }
5410             /* Record this loop in the outer loop's recording */
5411         }
5412     }
5413     else if(ins->handler_idx == WINED3DSIH_ENDLOOP || ins->handler_idx == WINED3DSIH_ENDREP)
5414     {
5415         if(priv->target_version >= NV2)
5416         {
5417             /* Nothing to do. The control frame is popped after the HW instr handler */
5418         }
5419         else
5420         {
5421             struct list *e = list_head(&priv->control_frames);
5422             control_frame = LIST_ENTRY(e, struct control_frame, entry);
5423             list_remove(&control_frame->entry);
5424
5425             if(control_frame->outer_loop)
5426             {
5427                 unsigned int iteration;
5428                 int aL = 0;
5429                 struct list copy;
5430
5431                 /* Turn off recording before playback */
5432                 priv->recording = FALSE;
5433
5434                 /* Move the recorded instructions to a separate list and get them out of the private data
5435                  * structure. If there are nested loops, the shader_arb_handle_instruction below will
5436                  * be recorded again, thus priv->record might be overwritten
5437                  */
5438                 list_init(&copy);
5439                 list_move_tail(&copy, &priv->record);
5440                 list_init(&priv->record);
5441
5442                 if(ins->handler_idx == WINED3DSIH_ENDLOOP)
5443                 {
5444                     shader_addline(buffer, "#unrolling loop: %u iterations, aL=%u, inc %d\n",
5445                                    control_frame->loop_control.count, control_frame->loop_control.start,
5446                                    control_frame->loop_control.step);
5447                     aL = control_frame->loop_control.start;
5448                 }
5449                 else
5450                 {
5451                     shader_addline(buffer, "#unrolling rep: %u iterations\n", control_frame->loop_control.count);
5452                 }
5453
5454                 for (iteration = 0; iteration < control_frame->loop_control.count; ++iteration)
5455                 {
5456                     struct recorded_instruction *rec_ins;
5457                     if(ins->handler_idx == WINED3DSIH_ENDLOOP)
5458                     {
5459                         priv->aL = aL;
5460                         shader_addline(buffer, "#Iteration %u, aL=%d\n", iteration, aL);
5461                     }
5462                     else
5463                     {
5464                         shader_addline(buffer, "#Iteration %u\n", iteration);
5465                     }
5466
5467                     LIST_FOR_EACH_ENTRY(rec_ins, &copy, struct recorded_instruction, entry)
5468                     {
5469                         shader_arb_handle_instruction(&rec_ins->ins);
5470                     }
5471
5472                     if(ins->handler_idx == WINED3DSIH_ENDLOOP)
5473                     {
5474                         aL += control_frame->loop_control.step;
5475                     }
5476                 }
5477                 shader_addline(buffer, "#end loop/rep\n");
5478
5479                 free_recorded_instruction(&copy);
5480                 HeapFree(GetProcessHeap(), 0, control_frame);
5481                 return; /* Instruction is handled */
5482             }
5483             else
5484             {
5485                 /* This is a nested loop. Proceed to the normal recording function */
5486                 HeapFree(GetProcessHeap(), 0, control_frame);
5487             }
5488         }
5489     }
5490
5491     if(priv->recording)
5492     {
5493         record_instruction(&priv->record, ins);
5494         return;
5495     }
5496
5497     /* boolean if */
5498     if(ins->handler_idx == WINED3DSIH_IF)
5499     {
5500         control_frame = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*control_frame));
5501         list_add_head(&priv->control_frames, &control_frame->entry);
5502         control_frame->type = IF;
5503
5504         bool_const = get_bool_const(ins, shader, ins->src[0].reg.idx[0].offset);
5505         if (ins->src[0].modifiers == WINED3DSPSM_NOT)
5506             bool_const = !bool_const;
5507         if (!priv->muted && !bool_const)
5508         {
5509             shader_addline(buffer, "#if(FALSE){\n");
5510             priv->muted = TRUE;
5511             control_frame->muting = TRUE;
5512         }
5513         else shader_addline(buffer, "#if(TRUE) {\n");
5514
5515         return; /* Instruction is handled */
5516     }
5517     else if(ins->handler_idx == WINED3DSIH_IFC)
5518     {
5519         /* IF(bool) and if_cond(a, b) use the same ELSE and ENDIF tokens */
5520         control_frame = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*control_frame));
5521         control_frame->type = IFC;
5522         control_frame->no.ifc = priv->num_ifcs++;
5523         list_add_head(&priv->control_frames, &control_frame->entry);
5524     }
5525     else if(ins->handler_idx == WINED3DSIH_ELSE)
5526     {
5527         struct list *e = list_head(&priv->control_frames);
5528         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5529
5530         if(control_frame->type == IF)
5531         {
5532             shader_addline(buffer, "#} else {\n");
5533             if(!priv->muted && !control_frame->muting)
5534             {
5535                 priv->muted = TRUE;
5536                 control_frame->muting = TRUE;
5537             }
5538             else if(control_frame->muting) priv->muted = FALSE;
5539             return; /* Instruction is handled. */
5540         }
5541         /* In case of an ifc, generate a HW shader instruction */
5542     }
5543     else if(ins->handler_idx == WINED3DSIH_ENDIF)
5544     {
5545         struct list *e = list_head(&priv->control_frames);
5546         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5547
5548         if(control_frame->type == IF)
5549         {
5550             shader_addline(buffer, "#} endif\n");
5551             if(control_frame->muting) priv->muted = FALSE;
5552             list_remove(&control_frame->entry);
5553             HeapFree(GetProcessHeap(), 0, control_frame);
5554             return; /* Instruction is handled */
5555         }
5556     }
5557
5558     if(priv->muted) return;
5559
5560     /* Select handler */
5561     hw_fct = shader_arb_instruction_handler_table[ins->handler_idx];
5562
5563     /* Unhandled opcode */
5564     if (!hw_fct)
5565     {
5566         FIXME("Backend can't handle opcode %#x\n", ins->handler_idx);
5567         return;
5568     }
5569     hw_fct(ins);
5570
5571     if(ins->handler_idx == WINED3DSIH_ENDLOOP || ins->handler_idx == WINED3DSIH_ENDREP)
5572     {
5573         struct list *e = list_head(&priv->control_frames);
5574         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5575         list_remove(&control_frame->entry);
5576         HeapFree(GetProcessHeap(), 0, control_frame);
5577         priv->loop_depth--;
5578     }
5579     else if(ins->handler_idx == WINED3DSIH_ENDIF)
5580     {
5581         /* Non-ifc ENDIFs don't reach that place because of the return in the if block above */
5582         struct list *e = list_head(&priv->control_frames);
5583         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5584         list_remove(&control_frame->entry);
5585         HeapFree(GetProcessHeap(), 0, control_frame);
5586     }
5587
5588
5589     shader_arb_add_instruction_modifiers(ins);
5590 }
5591
5592 const struct wined3d_shader_backend_ops arb_program_shader_backend =
5593 {
5594     shader_arb_handle_instruction,
5595     shader_arb_select,
5596     shader_arb_select_depth_blt,
5597     shader_arb_deselect_depth_blt,
5598     shader_arb_update_float_vertex_constants,
5599     shader_arb_update_float_pixel_constants,
5600     shader_arb_load_constants,
5601     shader_arb_load_np2fixup_constants,
5602     shader_arb_destroy,
5603     shader_arb_alloc,
5604     shader_arb_free,
5605     shader_arb_context_destroyed,
5606     shader_arb_get_caps,
5607     shader_arb_color_fixup_supported,
5608 };
5609
5610 /* ARB_fragment_program fixed function pipeline replacement definitions */
5611 #define ARB_FFP_CONST_TFACTOR           0
5612 #define ARB_FFP_CONST_SPECULAR_ENABLE   ((ARB_FFP_CONST_TFACTOR) + 1)
5613 #define ARB_FFP_CONST_CONSTANT(i)       ((ARB_FFP_CONST_SPECULAR_ENABLE) + 1 + i)
5614 #define ARB_FFP_CONST_BUMPMAT(i)        ((ARB_FFP_CONST_CONSTANT(7)) + 1 + i)
5615 #define ARB_FFP_CONST_LUMINANCE(i)      ((ARB_FFP_CONST_BUMPMAT(7)) + 1 + i)
5616
5617 struct arbfp_ffp_desc
5618 {
5619     struct ffp_frag_desc parent;
5620     GLuint shader;
5621     unsigned int num_textures_used;
5622 };
5623
5624 /* Context activation and GL locking are done by the caller. */
5625 static void arbfp_enable(const struct wined3d_gl_info *gl_info, BOOL enable)
5626 {
5627     if (enable)
5628     {
5629         gl_info->gl_ops.gl.p_glEnable(GL_FRAGMENT_PROGRAM_ARB);
5630         checkGLcall("glEnable(GL_FRAGMENT_PROGRAM_ARB)");
5631     }
5632     else
5633     {
5634         gl_info->gl_ops.gl.p_glDisable(GL_FRAGMENT_PROGRAM_ARB);
5635         checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
5636     }
5637 }
5638
5639 static HRESULT arbfp_alloc(struct wined3d_device *device)
5640 {
5641     struct shader_arb_priv *priv;
5642     /* Share private data between the shader backend and the pipeline replacement, if both
5643      * are the arb implementation. This is needed to figure out whether ARBfp should be disabled
5644      * if no pixel shader is bound or not
5645      */
5646     if (device->shader_backend == &arb_program_shader_backend)
5647     {
5648         device->fragment_priv = device->shader_priv;
5649     }
5650     else
5651     {
5652         device->fragment_priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(struct shader_arb_priv));
5653         if (!device->fragment_priv) return E_OUTOFMEMORY;
5654     }
5655     priv = device->fragment_priv;
5656     if (wine_rb_init(&priv->fragment_shaders, &wined3d_ffp_frag_program_rb_functions) == -1)
5657     {
5658         ERR("Failed to initialize rbtree.\n");
5659         HeapFree(GetProcessHeap(), 0, device->fragment_priv);
5660         return E_OUTOFMEMORY;
5661     }
5662     priv->use_arbfp_fixed_func = TRUE;
5663     return WINED3D_OK;
5664 }
5665
5666 /* Context activation is done by the caller. */
5667 static void arbfp_free_ffpshader(struct wine_rb_entry *entry, void *context)
5668 {
5669     const struct wined3d_gl_info *gl_info = context;
5670     struct arbfp_ffp_desc *entry_arb = WINE_RB_ENTRY_VALUE(entry, struct arbfp_ffp_desc, parent.entry);
5671
5672     ENTER_GL();
5673     GL_EXTCALL(glDeleteProgramsARB(1, &entry_arb->shader));
5674     checkGLcall("glDeleteProgramsARB(1, &entry_arb->shader)");
5675     HeapFree(GetProcessHeap(), 0, entry_arb);
5676     LEAVE_GL();
5677 }
5678
5679 /* Context activation is done by the caller. */
5680 static void arbfp_free(struct wined3d_device *device)
5681 {
5682     struct shader_arb_priv *priv = device->fragment_priv;
5683
5684     wine_rb_destroy(&priv->fragment_shaders, arbfp_free_ffpshader, &device->adapter->gl_info);
5685     priv->use_arbfp_fixed_func = FALSE;
5686
5687     if (device->shader_backend != &arb_program_shader_backend)
5688     {
5689         HeapFree(GetProcessHeap(), 0, device->fragment_priv);
5690     }
5691 }
5692
5693 static void arbfp_get_caps(const struct wined3d_gl_info *gl_info, struct fragment_caps *caps)
5694 {
5695     caps->PrimitiveMiscCaps = WINED3DPMISCCAPS_TSSARGTEMP;
5696     caps->TextureOpCaps =  WINED3DTEXOPCAPS_DISABLE                     |
5697                            WINED3DTEXOPCAPS_SELECTARG1                  |
5698                            WINED3DTEXOPCAPS_SELECTARG2                  |
5699                            WINED3DTEXOPCAPS_MODULATE4X                  |
5700                            WINED3DTEXOPCAPS_MODULATE2X                  |
5701                            WINED3DTEXOPCAPS_MODULATE                    |
5702                            WINED3DTEXOPCAPS_ADDSIGNED2X                 |
5703                            WINED3DTEXOPCAPS_ADDSIGNED                   |
5704                            WINED3DTEXOPCAPS_ADD                         |
5705                            WINED3DTEXOPCAPS_SUBTRACT                    |
5706                            WINED3DTEXOPCAPS_ADDSMOOTH                   |
5707                            WINED3DTEXOPCAPS_BLENDCURRENTALPHA           |
5708                            WINED3DTEXOPCAPS_BLENDFACTORALPHA            |
5709                            WINED3DTEXOPCAPS_BLENDTEXTUREALPHA           |
5710                            WINED3DTEXOPCAPS_BLENDDIFFUSEALPHA           |
5711                            WINED3DTEXOPCAPS_BLENDTEXTUREALPHAPM         |
5712                            WINED3DTEXOPCAPS_MODULATEALPHA_ADDCOLOR      |
5713                            WINED3DTEXOPCAPS_MODULATECOLOR_ADDALPHA      |
5714                            WINED3DTEXOPCAPS_MODULATEINVCOLOR_ADDALPHA   |
5715                            WINED3DTEXOPCAPS_MODULATEINVALPHA_ADDCOLOR   |
5716                            WINED3DTEXOPCAPS_DOTPRODUCT3                 |
5717                            WINED3DTEXOPCAPS_MULTIPLYADD                 |
5718                            WINED3DTEXOPCAPS_LERP                        |
5719                            WINED3DTEXOPCAPS_BUMPENVMAP                  |
5720                            WINED3DTEXOPCAPS_BUMPENVMAPLUMINANCE;
5721
5722     /* TODO: Implement WINED3DTEXOPCAPS_PREMODULATE */
5723
5724     caps->MaxTextureBlendStages   = 8;
5725     caps->MaxSimultaneousTextures = min(gl_info->limits.fragment_samplers, 8);
5726 }
5727
5728 static void state_texfactor_arbfp(struct wined3d_context *context,
5729         const struct wined3d_state *state, DWORD state_id)
5730 {
5731     struct wined3d_device *device = context->swapchain->device;
5732     const struct wined3d_gl_info *gl_info = context->gl_info;
5733     float col[4];
5734
5735     /* Don't load the parameter if we're using an arbfp pixel shader,
5736      * otherwise we'll overwrite application provided constants. */
5737     if (device->shader_backend == &arb_program_shader_backend)
5738     {
5739         struct shader_arb_priv *priv;
5740
5741         if (use_ps(state)) return;
5742
5743         priv = device->shader_priv;
5744         priv->pshader_const_dirty[ARB_FFP_CONST_TFACTOR] = 1;
5745         priv->highest_dirty_ps_const = max(priv->highest_dirty_ps_const, ARB_FFP_CONST_TFACTOR + 1);
5746     }
5747
5748     D3DCOLORTOGLFLOAT4(state->render_states[WINED3D_RS_TEXTUREFACTOR], col);
5749     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_TFACTOR, col));
5750     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_TFACTOR, col)");
5751 }
5752
5753 static void state_arb_specularenable(struct wined3d_context *context,
5754         const struct wined3d_state *state, DWORD state_id)
5755 {
5756     struct wined3d_device *device = context->swapchain->device;
5757     const struct wined3d_gl_info *gl_info = context->gl_info;
5758     float col[4];
5759
5760     /* Don't load the parameter if we're using an arbfp pixel shader, otherwise we'll overwrite
5761      * application provided constants
5762      */
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_SPECULAR_ENABLE] = 1;
5771         priv->highest_dirty_ps_const = max(priv->highest_dirty_ps_const, ARB_FFP_CONST_SPECULAR_ENABLE + 1);
5772     }
5773
5774     if (state->render_states[WINED3D_RS_SPECULARENABLE])
5775     {
5776         /* The specular color has no alpha */
5777         col[0] = 1.0f; col[1] = 1.0f;
5778         col[2] = 1.0f; col[3] = 0.0f;
5779     } else {
5780         col[0] = 0.0f; col[1] = 0.0f;
5781         col[2] = 0.0f; col[3] = 0.0f;
5782     }
5783     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_SPECULAR_ENABLE, col));
5784     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_SPECULAR_ENABLE, col)");
5785 }
5786
5787 static void set_bumpmat_arbfp(struct wined3d_context *context, const struct wined3d_state *state, DWORD state_id)
5788 {
5789     DWORD stage = (state_id - STATE_TEXTURESTAGE(0, 0)) / (WINED3D_HIGHEST_TEXTURE_STATE + 1);
5790     struct wined3d_device *device = context->swapchain->device;
5791     const struct wined3d_gl_info *gl_info = context->gl_info;
5792     float mat[2][2];
5793
5794     if (use_ps(state))
5795     {
5796         if (stage && (state->pixel_shader->reg_maps.bumpmat & (1 << stage)))
5797         {
5798             /* The pixel shader has to know the bump env matrix. Do a constants update if it isn't scheduled
5799              * anyway
5800              */
5801             if (!isStateDirty(context, STATE_PIXELSHADERCONSTANT))
5802                 context_apply_state(context, state, STATE_PIXELSHADERCONSTANT);
5803         }
5804
5805         if(device->shader_backend == &arb_program_shader_backend) {
5806             /* Exit now, don't set the bumpmat below, otherwise we may overwrite pixel shader constants */
5807             return;
5808         }
5809     }
5810     else if (device->shader_backend == &arb_program_shader_backend)
5811     {
5812         struct shader_arb_priv *priv = device->shader_priv;
5813         priv->pshader_const_dirty[ARB_FFP_CONST_BUMPMAT(stage)] = 1;
5814         priv->highest_dirty_ps_const = max(priv->highest_dirty_ps_const, ARB_FFP_CONST_BUMPMAT(stage) + 1);
5815     }
5816
5817     mat[0][0] = *((float *)&state->texture_states[stage][WINED3D_TSS_BUMPENV_MAT00]);
5818     mat[0][1] = *((float *)&state->texture_states[stage][WINED3D_TSS_BUMPENV_MAT01]);
5819     mat[1][0] = *((float *)&state->texture_states[stage][WINED3D_TSS_BUMPENV_MAT10]);
5820     mat[1][1] = *((float *)&state->texture_states[stage][WINED3D_TSS_BUMPENV_MAT11]);
5821
5822     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_BUMPMAT(stage), &mat[0][0]));
5823     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_BUMPMAT(stage), &mat[0][0])");
5824 }
5825
5826 static void tex_bumpenvlum_arbfp(struct wined3d_context *context,
5827         const struct wined3d_state *state, DWORD state_id)
5828 {
5829     DWORD stage = (state_id - STATE_TEXTURESTAGE(0, 0)) / (WINED3D_HIGHEST_TEXTURE_STATE + 1);
5830     struct wined3d_device *device = context->swapchain->device;
5831     const struct wined3d_gl_info *gl_info = context->gl_info;
5832     float param[4];
5833
5834     if (use_ps(state))
5835     {
5836         if (stage && (state->pixel_shader->reg_maps.luminanceparams & (1 << stage)))
5837         {
5838             /* The pixel shader has to know the luminance offset. Do a constants update if it
5839              * isn't scheduled anyway
5840              */
5841             if (!isStateDirty(context, STATE_PIXELSHADERCONSTANT))
5842                 context_apply_state(context, state, STATE_PIXELSHADERCONSTANT);
5843         }
5844
5845         if(device->shader_backend == &arb_program_shader_backend) {
5846             /* Exit now, don't set the bumpmat below, otherwise we may overwrite pixel shader constants */
5847             return;
5848         }
5849     }
5850     else if (device->shader_backend == &arb_program_shader_backend)
5851     {
5852         struct shader_arb_priv *priv = device->shader_priv;
5853         priv->pshader_const_dirty[ARB_FFP_CONST_LUMINANCE(stage)] = 1;
5854         priv->highest_dirty_ps_const = max(priv->highest_dirty_ps_const, ARB_FFP_CONST_LUMINANCE(stage) + 1);
5855     }
5856
5857     param[0] = *((float *)&state->texture_states[stage][WINED3D_TSS_BUMPENV_LSCALE]);
5858     param[1] = *((float *)&state->texture_states[stage][WINED3D_TSS_BUMPENV_LOFFSET]);
5859     param[2] = 0.0f;
5860     param[3] = 0.0f;
5861
5862     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_LUMINANCE(stage), param));
5863     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_LUMINANCE(stage), param)");
5864 }
5865
5866 static const char *get_argreg(struct wined3d_shader_buffer *buffer, DWORD argnum, unsigned int stage, DWORD arg)
5867 {
5868     const char *ret;
5869
5870     if(arg == ARG_UNUSED) return "unused"; /* This is the marker for unused registers */
5871
5872     switch(arg & WINED3DTA_SELECTMASK) {
5873         case WINED3DTA_DIFFUSE:
5874             ret = "fragment.color.primary"; break;
5875
5876         case WINED3DTA_CURRENT:
5877             if (!stage) ret = "fragment.color.primary";
5878             else ret = "ret";
5879             break;
5880
5881         case WINED3DTA_TEXTURE:
5882             switch(stage) {
5883                 case 0: ret = "tex0"; break;
5884                 case 1: ret = "tex1"; break;
5885                 case 2: ret = "tex2"; break;
5886                 case 3: ret = "tex3"; break;
5887                 case 4: ret = "tex4"; break;
5888                 case 5: ret = "tex5"; break;
5889                 case 6: ret = "tex6"; break;
5890                 case 7: ret = "tex7"; break;
5891                 default: ret = "unknown texture";
5892             }
5893             break;
5894
5895         case WINED3DTA_TFACTOR:
5896             ret = "tfactor"; break;
5897
5898         case WINED3DTA_SPECULAR:
5899             ret = "fragment.color.secondary"; break;
5900
5901         case WINED3DTA_TEMP:
5902             ret = "tempreg"; break;
5903
5904         case WINED3DTA_CONSTANT:
5905             FIXME("Implement perstage constants\n");
5906             switch(stage) {
5907                 case 0: ret = "const0"; break;
5908                 case 1: ret = "const1"; break;
5909                 case 2: ret = "const2"; break;
5910                 case 3: ret = "const3"; break;
5911                 case 4: ret = "const4"; break;
5912                 case 5: ret = "const5"; break;
5913                 case 6: ret = "const6"; break;
5914                 case 7: ret = "const7"; break;
5915                 default: ret = "unknown constant";
5916             }
5917             break;
5918
5919         default:
5920             return "unknown";
5921     }
5922
5923     if(arg & WINED3DTA_COMPLEMENT) {
5924         shader_addline(buffer, "SUB arg%u, const.x, %s;\n", argnum, ret);
5925         if(argnum == 0) ret = "arg0";
5926         if(argnum == 1) ret = "arg1";
5927         if(argnum == 2) ret = "arg2";
5928     }
5929     if(arg & WINED3DTA_ALPHAREPLICATE) {
5930         shader_addline(buffer, "MOV arg%u, %s.w;\n", argnum, ret);
5931         if(argnum == 0) ret = "arg0";
5932         if(argnum == 1) ret = "arg1";
5933         if(argnum == 2) ret = "arg2";
5934     }
5935     return ret;
5936 }
5937
5938 static void gen_ffp_instr(struct wined3d_shader_buffer *buffer, unsigned int stage, BOOL color,
5939         BOOL alpha, DWORD dst, DWORD op, DWORD dw_arg0, DWORD dw_arg1, DWORD dw_arg2)
5940 {
5941     const char *dstmask, *dstreg, *arg0, *arg1, *arg2;
5942     unsigned int mul = 1;
5943     BOOL mul_final_dest = FALSE;
5944
5945     if(color && alpha) dstmask = "";
5946     else if(color) dstmask = ".xyz";
5947     else dstmask = ".w";
5948
5949     if(dst == tempreg) dstreg = "tempreg";
5950     else dstreg = "ret";
5951
5952     arg0 = get_argreg(buffer, 0, stage, dw_arg0);
5953     arg1 = get_argreg(buffer, 1, stage, dw_arg1);
5954     arg2 = get_argreg(buffer, 2, stage, dw_arg2);
5955
5956     switch (op)
5957     {
5958         case WINED3D_TOP_DISABLE:
5959             if (!stage)
5960                 shader_addline(buffer, "MOV %s%s, fragment.color.primary;\n", dstreg, dstmask);
5961             break;
5962
5963         case WINED3D_TOP_SELECT_ARG2:
5964             arg1 = arg2;
5965             /* FALLTHROUGH */
5966         case WINED3D_TOP_SELECT_ARG1:
5967             shader_addline(buffer, "MOV %s%s, %s;\n", dstreg, dstmask, arg1);
5968             break;
5969
5970         case WINED3D_TOP_MODULATE_4X:
5971             mul = 2;
5972             /* FALLTHROUGH */
5973         case WINED3D_TOP_MODULATE_2X:
5974             mul *= 2;
5975             if (!strcmp(dstreg, "result.color"))
5976             {
5977                 dstreg = "ret";
5978                 mul_final_dest = TRUE;
5979             }
5980             /* FALLTHROUGH */
5981         case WINED3D_TOP_MODULATE:
5982             shader_addline(buffer, "MUL %s%s, %s, %s;\n", dstreg, dstmask, arg1, arg2);
5983             break;
5984
5985         case WINED3D_TOP_ADD_SIGNED_2X:
5986             mul = 2;
5987             if (!strcmp(dstreg, "result.color"))
5988             {
5989                 dstreg = "ret";
5990                 mul_final_dest = TRUE;
5991             }
5992             /* FALLTHROUGH */
5993         case WINED3D_TOP_ADD_SIGNED:
5994             shader_addline(buffer, "SUB arg2, %s, const.w;\n", arg2);
5995             arg2 = "arg2";
5996             /* FALLTHROUGH */
5997         case WINED3D_TOP_ADD:
5998             shader_addline(buffer, "ADD_SAT %s%s, %s, %s;\n", dstreg, dstmask, arg1, arg2);
5999             break;
6000
6001         case WINED3D_TOP_SUBTRACT:
6002             shader_addline(buffer, "SUB_SAT %s%s, %s, %s;\n", dstreg, dstmask, arg1, arg2);
6003             break;
6004
6005         case WINED3D_TOP_ADD_SMOOTH:
6006             shader_addline(buffer, "SUB arg1, const.x, %s;\n", arg1);
6007             shader_addline(buffer, "MAD_SAT %s%s, arg1, %s, %s;\n", dstreg, dstmask, arg2, arg1);
6008             break;
6009
6010         case WINED3D_TOP_BLEND_CURRENT_ALPHA:
6011             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_CURRENT);
6012             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
6013             break;
6014         case WINED3D_TOP_BLEND_FACTOR_ALPHA:
6015             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_TFACTOR);
6016             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
6017             break;
6018         case WINED3D_TOP_BLEND_TEXTURE_ALPHA:
6019             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_TEXTURE);
6020             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
6021             break;
6022         case WINED3D_TOP_BLEND_DIFFUSE_ALPHA:
6023             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_DIFFUSE);
6024             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
6025             break;
6026
6027         case WINED3D_TOP_BLEND_TEXTURE_ALPHA_PM:
6028             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_TEXTURE);
6029             shader_addline(buffer, "SUB arg0.w, const.x, %s.w;\n", arg0);
6030             shader_addline(buffer, "MAD_SAT %s%s, %s, arg0.w, %s;\n", dstreg, dstmask, arg2, arg1);
6031             break;
6032
6033         /* D3DTOP_PREMODULATE ???? */
6034
6035         case WINED3D_TOP_MODULATE_INVALPHA_ADD_COLOR:
6036             shader_addline(buffer, "SUB arg0.w, const.x, %s;\n", arg1);
6037             shader_addline(buffer, "MAD_SAT %s%s, arg0.w, %s, %s;\n", dstreg, dstmask, arg2, arg1);
6038             break;
6039         case WINED3D_TOP_MODULATE_ALPHA_ADD_COLOR:
6040             shader_addline(buffer, "MAD_SAT %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg1, arg2, arg1);
6041             break;
6042         case WINED3D_TOP_MODULATE_INVCOLOR_ADD_ALPHA:
6043             shader_addline(buffer, "SUB arg0, const.x, %s;\n", arg1);
6044             shader_addline(buffer, "MAD_SAT %s%s, arg0, %s, %s.w;\n", dstreg, dstmask, arg2, arg1);
6045             break;
6046         case WINED3D_TOP_MODULATE_COLOR_ADD_ALPHA:
6047             shader_addline(buffer, "MAD_SAT %s%s, %s, %s, %s.w;\n", dstreg, dstmask, arg1, arg2, arg1);
6048             break;
6049
6050         case WINED3D_TOP_DOTPRODUCT3:
6051             mul = 4;
6052             if (!strcmp(dstreg, "result.color"))
6053             {
6054                 dstreg = "ret";
6055                 mul_final_dest = TRUE;
6056             }
6057             shader_addline(buffer, "SUB arg1, %s, const.w;\n", arg1);
6058             shader_addline(buffer, "SUB arg2, %s, const.w;\n", arg2);
6059             shader_addline(buffer, "DP3_SAT %s%s, arg1, arg2;\n", dstreg, dstmask);
6060             break;
6061
6062         case WINED3D_TOP_MULTIPLY_ADD:
6063             shader_addline(buffer, "MAD_SAT %s%s, %s, %s, %s;\n", dstreg, dstmask, arg1, arg2, arg0);
6064             break;
6065
6066         case WINED3D_TOP_LERP:
6067             /* The msdn is not quite right here */
6068             shader_addline(buffer, "LRP %s%s, %s, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
6069             break;
6070
6071         case WINED3D_TOP_BUMPENVMAP:
6072         case WINED3D_TOP_BUMPENVMAP_LUMINANCE:
6073             /* Those are handled in the first pass of the shader(generation pass 1 and 2) already */
6074             break;
6075
6076         default:
6077             FIXME("Unhandled texture op %08x\n", op);
6078     }
6079
6080     if(mul == 2) {
6081         shader_addline(buffer, "MUL_SAT %s%s, %s, const.y;\n", mul_final_dest ? "result.color" : dstreg, dstmask, dstreg);
6082     } else if(mul == 4) {
6083         shader_addline(buffer, "MUL_SAT %s%s, %s, const.z;\n", mul_final_dest ? "result.color" : dstreg, dstmask, dstreg);
6084     }
6085 }
6086
6087 static GLuint gen_arbfp_ffp_shader(const struct ffp_frag_settings *settings, const struct wined3d_gl_info *gl_info)
6088 {
6089     unsigned int stage;
6090     struct wined3d_shader_buffer buffer;
6091     BOOL tex_read[MAX_TEXTURES] = {FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE};
6092     BOOL bump_used[MAX_TEXTURES] = {FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE};
6093     BOOL luminance_used[MAX_TEXTURES] = {FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE};
6094     UINT lowest_disabled_stage;
6095     const char *textype;
6096     const char *instr, *sat;
6097     char colorcor_dst[8];
6098     GLuint ret;
6099     DWORD arg0, arg1, arg2;
6100     BOOL tempreg_used = FALSE, tfactor_used = FALSE;
6101     BOOL op_equal;
6102     const char *final_combiner_src = "ret";
6103     GLint pos;
6104
6105     /* Find out which textures are read */
6106     for (stage = 0; stage < MAX_TEXTURES; ++stage)
6107     {
6108         if (settings->op[stage].cop == WINED3D_TOP_DISABLE)
6109             break;
6110         arg0 = settings->op[stage].carg0 & WINED3DTA_SELECTMASK;
6111         arg1 = settings->op[stage].carg1 & WINED3DTA_SELECTMASK;
6112         arg2 = settings->op[stage].carg2 & WINED3DTA_SELECTMASK;
6113         if(arg0 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
6114         if(arg1 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
6115         if(arg2 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
6116
6117         if (settings->op[stage].cop == WINED3D_TOP_BLEND_TEXTURE_ALPHA)
6118             tex_read[stage] = TRUE;
6119         if (settings->op[stage].cop == WINED3D_TOP_BLEND_TEXTURE_ALPHA_PM)
6120             tex_read[stage] = TRUE;
6121         if (settings->op[stage].cop == WINED3D_TOP_BUMPENVMAP)
6122         {
6123             bump_used[stage] = TRUE;
6124             tex_read[stage] = TRUE;
6125         }
6126         if (settings->op[stage].cop == WINED3D_TOP_BUMPENVMAP_LUMINANCE)
6127         {
6128             bump_used[stage] = TRUE;
6129             tex_read[stage] = TRUE;
6130             luminance_used[stage] = TRUE;
6131         }
6132         else if (settings->op[stage].cop == WINED3D_TOP_BLEND_FACTOR_ALPHA)
6133         {
6134             tfactor_used = TRUE;
6135         }
6136
6137         if(arg0 == WINED3DTA_TFACTOR || arg1 == WINED3DTA_TFACTOR || arg2 == WINED3DTA_TFACTOR) {
6138             tfactor_used = TRUE;
6139         }
6140
6141         if(settings->op[stage].dst == tempreg) tempreg_used = TRUE;
6142         if(arg0 == WINED3DTA_TEMP || arg1 == WINED3DTA_TEMP || arg2 == WINED3DTA_TEMP) {
6143             tempreg_used = TRUE;
6144         }
6145
6146         if (settings->op[stage].aop == WINED3D_TOP_DISABLE)
6147             continue;
6148         arg0 = settings->op[stage].aarg0 & WINED3DTA_SELECTMASK;
6149         arg1 = settings->op[stage].aarg1 & WINED3DTA_SELECTMASK;
6150         arg2 = settings->op[stage].aarg2 & WINED3DTA_SELECTMASK;
6151         if(arg0 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
6152         if(arg1 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
6153         if(arg2 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
6154
6155         if(arg0 == WINED3DTA_TEMP || arg1 == WINED3DTA_TEMP || arg2 == WINED3DTA_TEMP) {
6156             tempreg_used = TRUE;
6157         }
6158         if(arg0 == WINED3DTA_TFACTOR || arg1 == WINED3DTA_TFACTOR || arg2 == WINED3DTA_TFACTOR) {
6159             tfactor_used = TRUE;
6160         }
6161     }
6162     lowest_disabled_stage = stage;
6163
6164     /* Shader header */
6165     if (!shader_buffer_init(&buffer))
6166     {
6167         ERR("Failed to initialize shader buffer.\n");
6168         return 0;
6169     }
6170
6171     shader_addline(&buffer, "!!ARBfp1.0\n");
6172
6173     switch(settings->fog) {
6174         case FOG_OFF:                                                         break;
6175         case FOG_LINEAR: shader_addline(&buffer, "OPTION ARB_fog_linear;\n"); break;
6176         case FOG_EXP:    shader_addline(&buffer, "OPTION ARB_fog_exp;\n");    break;
6177         case FOG_EXP2:   shader_addline(&buffer, "OPTION ARB_fog_exp2;\n");   break;
6178         default: FIXME("Unexpected fog setting %d\n", settings->fog);
6179     }
6180
6181     shader_addline(&buffer, "PARAM const = {1, 2, 4, 0.5};\n");
6182     shader_addline(&buffer, "TEMP TMP;\n");
6183     shader_addline(&buffer, "TEMP ret;\n");
6184     if(tempreg_used || settings->sRGB_write) shader_addline(&buffer, "TEMP tempreg;\n");
6185     shader_addline(&buffer, "TEMP arg0;\n");
6186     shader_addline(&buffer, "TEMP arg1;\n");
6187     shader_addline(&buffer, "TEMP arg2;\n");
6188     for(stage = 0; stage < MAX_TEXTURES; stage++) {
6189         if(!tex_read[stage]) continue;
6190         shader_addline(&buffer, "TEMP tex%u;\n", stage);
6191         if(!bump_used[stage]) continue;
6192         shader_addline(&buffer, "PARAM bumpmat%u = program.env[%u];\n", stage, ARB_FFP_CONST_BUMPMAT(stage));
6193         if(!luminance_used[stage]) continue;
6194         shader_addline(&buffer, "PARAM luminance%u = program.env[%u];\n", stage, ARB_FFP_CONST_LUMINANCE(stage));
6195     }
6196     if(tfactor_used) {
6197         shader_addline(&buffer, "PARAM tfactor = program.env[%u];\n", ARB_FFP_CONST_TFACTOR);
6198     }
6199         shader_addline(&buffer, "PARAM specular_enable = program.env[%u];\n", ARB_FFP_CONST_SPECULAR_ENABLE);
6200
6201     if(settings->sRGB_write) {
6202         shader_addline(&buffer, "PARAM srgb_consts1 = {%f, %f, %f, %f};\n",
6203                        srgb_mul_low, srgb_cmp, srgb_pow, srgb_mul_high);
6204         shader_addline(&buffer, "PARAM srgb_consts2 = {%f, %f, %f, %f};\n",
6205                        srgb_sub_high, 0.0, 0.0, 0.0);
6206     }
6207
6208     if (lowest_disabled_stage < 7 && settings->emul_clipplanes)
6209         shader_addline(&buffer, "KIL fragment.texcoord[7];\n");
6210
6211     /* Generate texture sampling instructions) */
6212     for (stage = 0; stage < MAX_TEXTURES && settings->op[stage].cop != WINED3D_TOP_DISABLE; ++stage)
6213     {
6214         if (!tex_read[stage])
6215             continue;
6216
6217         switch(settings->op[stage].tex_type) {
6218             case tex_1d:                    textype = "1D";     break;
6219             case tex_2d:                    textype = "2D";     break;
6220             case tex_3d:                    textype = "3D";     break;
6221             case tex_cube:                  textype = "CUBE";   break;
6222             case tex_rect:                  textype = "RECT";   break;
6223             default: textype = "unexpected_textype";   break;
6224         }
6225
6226         if (settings->op[stage].cop == WINED3D_TOP_BUMPENVMAP
6227                 || settings->op[stage].cop == WINED3D_TOP_BUMPENVMAP_LUMINANCE)
6228             sat = "";
6229         else
6230             sat = "_SAT";
6231
6232         if(settings->op[stage].projected == proj_none) {
6233             instr = "TEX";
6234         } else if(settings->op[stage].projected == proj_count4 ||
6235                   settings->op[stage].projected == proj_count3) {
6236             instr = "TXP";
6237         } else {
6238             FIXME("Unexpected projection mode %d\n", settings->op[stage].projected);
6239             instr = "TXP";
6240         }
6241
6242         if (stage > 0
6243                 && (settings->op[stage - 1].cop == WINED3D_TOP_BUMPENVMAP
6244                 || settings->op[stage - 1].cop == WINED3D_TOP_BUMPENVMAP_LUMINANCE))
6245         {
6246             shader_addline(&buffer, "SWZ arg1, bumpmat%u, x, z, 0, 0;\n", stage - 1);
6247             shader_addline(&buffer, "DP3 ret.x, arg1, tex%u;\n", stage - 1);
6248             shader_addline(&buffer, "SWZ arg1, bumpmat%u, y, w, 0, 0;\n", stage - 1);
6249             shader_addline(&buffer, "DP3 ret.y, arg1, tex%u;\n", stage - 1);
6250
6251             /* with projective textures, texbem only divides the static texture coord, not the displacement,
6252              * so multiply the displacement with the dividing parameter before passing it to TXP
6253              */
6254             if (settings->op[stage].projected != proj_none) {
6255                 if(settings->op[stage].projected == proj_count4) {
6256                     shader_addline(&buffer, "MOV ret.w, fragment.texcoord[%u].w;\n", stage);
6257                     shader_addline(&buffer, "MUL ret.xyz, ret, fragment.texcoord[%u].w, fragment.texcoord[%u];\n", stage, stage);
6258                 } else {
6259                     shader_addline(&buffer, "MOV ret.w, fragment.texcoord[%u].z;\n", stage);
6260                     shader_addline(&buffer, "MAD ret.xyz, ret, fragment.texcoord[%u].z, fragment.texcoord[%u];\n", stage, stage);
6261                 }
6262             } else {
6263                 shader_addline(&buffer, "ADD ret, ret, fragment.texcoord[%u];\n", stage);
6264             }
6265
6266             shader_addline(&buffer, "%s%s tex%u, ret, texture[%u], %s;\n",
6267                     instr, sat, stage, stage, textype);
6268             if (settings->op[stage - 1].cop == WINED3D_TOP_BUMPENVMAP_LUMINANCE)
6269             {
6270                 shader_addline(&buffer, "MAD_SAT ret.x, tex%u.z, luminance%u.x, luminance%u.y;\n",
6271                                stage - 1, stage - 1, stage - 1);
6272                 shader_addline(&buffer, "MUL tex%u, tex%u, ret.x;\n", stage, stage);
6273             }
6274         } else if(settings->op[stage].projected == proj_count3) {
6275             shader_addline(&buffer, "MOV ret, fragment.texcoord[%u];\n", stage);
6276             shader_addline(&buffer, "MOV ret.w, ret.z;\n");
6277             shader_addline(&buffer, "%s%s tex%u, ret, texture[%u], %s;\n",
6278                             instr, sat, stage, stage, textype);
6279         } else {
6280             shader_addline(&buffer, "%s%s tex%u, fragment.texcoord[%u], texture[%u], %s;\n",
6281                             instr, sat, stage, stage, stage, textype);
6282         }
6283
6284         sprintf(colorcor_dst, "tex%u", stage);
6285         gen_color_correction(&buffer, colorcor_dst, WINED3DSP_WRITEMASK_ALL, "const.x", "const.y",
6286                 settings->op[stage].color_fixup);
6287     }
6288
6289     /* Generate the main shader */
6290     for (stage = 0; stage < MAX_TEXTURES; ++stage)
6291     {
6292         if (settings->op[stage].cop == WINED3D_TOP_DISABLE)
6293         {
6294             if (!stage)
6295                 final_combiner_src = "fragment.color.primary";
6296             break;
6297         }
6298
6299         if (settings->op[stage].cop == WINED3D_TOP_SELECT_ARG1
6300                 && settings->op[stage].aop == WINED3D_TOP_SELECT_ARG1)
6301             op_equal = settings->op[stage].carg1 == settings->op[stage].aarg1;
6302         else if (settings->op[stage].cop == WINED3D_TOP_SELECT_ARG1
6303                 && settings->op[stage].aop == WINED3D_TOP_SELECT_ARG2)
6304             op_equal = settings->op[stage].carg1 == settings->op[stage].aarg2;
6305         else if (settings->op[stage].cop == WINED3D_TOP_SELECT_ARG2
6306                 && settings->op[stage].aop == WINED3D_TOP_SELECT_ARG1)
6307             op_equal = settings->op[stage].carg2 == settings->op[stage].aarg1;
6308         else if (settings->op[stage].cop == WINED3D_TOP_SELECT_ARG2
6309                 && settings->op[stage].aop == WINED3D_TOP_SELECT_ARG2)
6310             op_equal = settings->op[stage].carg2 == settings->op[stage].aarg2;
6311         else
6312             op_equal = settings->op[stage].aop   == settings->op[stage].cop
6313                     && settings->op[stage].carg0 == settings->op[stage].aarg0
6314                     && settings->op[stage].carg1 == settings->op[stage].aarg1
6315                     && settings->op[stage].carg2 == settings->op[stage].aarg2;
6316
6317         if (settings->op[stage].aop == WINED3D_TOP_DISABLE)
6318         {
6319             gen_ffp_instr(&buffer, stage, TRUE, FALSE, settings->op[stage].dst,
6320                           settings->op[stage].cop, settings->op[stage].carg0,
6321                           settings->op[stage].carg1, settings->op[stage].carg2);
6322             if (!stage)
6323                 shader_addline(&buffer, "MOV ret.w, fragment.color.primary.w;\n");
6324         }
6325         else if (op_equal)
6326         {
6327             gen_ffp_instr(&buffer, stage, TRUE, TRUE, settings->op[stage].dst,
6328                           settings->op[stage].cop, settings->op[stage].carg0,
6329                           settings->op[stage].carg1, settings->op[stage].carg2);
6330         } else {
6331             gen_ffp_instr(&buffer, stage, TRUE, FALSE, settings->op[stage].dst,
6332                           settings->op[stage].cop, settings->op[stage].carg0,
6333                           settings->op[stage].carg1, settings->op[stage].carg2);
6334             gen_ffp_instr(&buffer, stage, FALSE, TRUE, settings->op[stage].dst,
6335                           settings->op[stage].aop, settings->op[stage].aarg0,
6336                           settings->op[stage].aarg1, settings->op[stage].aarg2);
6337         }
6338     }
6339
6340     if(settings->sRGB_write) {
6341         shader_addline(&buffer, "MAD ret, fragment.color.secondary, specular_enable, %s;\n", final_combiner_src);
6342         arbfp_add_sRGB_correction(&buffer, "ret", "arg0", "arg1", "arg2", "tempreg", FALSE);
6343         shader_addline(&buffer, "MOV result.color, ret;\n");
6344     } else {
6345         shader_addline(&buffer, "MAD result.color, fragment.color.secondary, specular_enable, %s;\n", final_combiner_src);
6346     }
6347
6348     /* Footer */
6349     shader_addline(&buffer, "END\n");
6350
6351     /* Generate the shader */
6352     GL_EXTCALL(glGenProgramsARB(1, &ret));
6353     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, ret));
6354     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
6355             strlen(buffer.buffer), buffer.buffer));
6356     checkGLcall("glProgramStringARB()");
6357
6358     gl_info->gl_ops.gl.p_glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
6359     if (pos != -1)
6360     {
6361         FIXME("Fragment program error at position %d: %s\n\n", pos,
6362               debugstr_a((const char *)gl_info->gl_ops.gl.p_glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
6363         shader_arb_dump_program_source(buffer.buffer);
6364     }
6365     else
6366     {
6367         GLint native;
6368
6369         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
6370         checkGLcall("glGetProgramivARB()");
6371         if (!native) WARN("Program exceeds native resource limits.\n");
6372     }
6373
6374     shader_buffer_free(&buffer);
6375     return ret;
6376 }
6377
6378 static void fragment_prog_arbfp(struct wined3d_context *context, const struct wined3d_state *state, DWORD state_id)
6379 {
6380     const struct wined3d_device *device = context->swapchain->device;
6381     const struct wined3d_gl_info *gl_info = context->gl_info;
6382     struct shader_arb_priv *priv = device->fragment_priv;
6383     BOOL use_vshader = use_vs(state);
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 && !isStateDirty(context, context->state_table[STATE_VSHADER].representative))
6405         {
6406             device->shader_backend->shader_select(context, use_pshader, use_vshader);
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(device, 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             new_desc->num_textures_used = 0;
6425             for (i = 0; i < gl_info->limits.texture_stages; ++i)
6426             {
6427                 if (settings.op[i].cop == WINED3D_TOP_DISABLE)
6428                     break;
6429                 new_desc->num_textures_used = i;
6430             }
6431
6432             memcpy(&new_desc->parent.settings, &settings, sizeof(settings));
6433             new_desc->shader = gen_arbfp_ffp_shader(&settings, gl_info);
6434             add_ffp_frag_shader(&priv->fragment_shaders, &new_desc->parent);
6435             TRACE("Allocated fixed function replacement shader descriptor %p\n", new_desc);
6436             desc = new_desc;
6437         }
6438
6439         /* Now activate the replacement program. GL_FRAGMENT_PROGRAM_ARB is already active (however, note the
6440          * comment above the shader_select call below). If e.g. GLSL is active, the shader_select call will
6441          * deactivate it.
6442          */
6443         GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, desc->shader));
6444         checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, desc->shader)");
6445         priv->current_fprogram_id = desc->shader;
6446
6447         if (device->shader_backend == &arb_program_shader_backend && context->last_was_pshader)
6448         {
6449             /* Reload fixed function constants since they collide with the
6450              * pixel shader constants. */
6451             for (i = 0; i < MAX_TEXTURES; ++i)
6452             {
6453                 set_bumpmat_arbfp(context, state, STATE_TEXTURESTAGE(i, WINED3D_TSS_BUMPENV_MAT00));
6454             }
6455             state_texfactor_arbfp(context, state, STATE_RENDER(WINED3D_RS_TEXTUREFACTOR));
6456             state_arb_specularenable(context, state, STATE_RENDER(WINED3D_RS_SPECULARENABLE));
6457         }
6458         context->last_was_pshader = FALSE;
6459     } else {
6460         context->last_was_pshader = TRUE;
6461     }
6462
6463     /* Finally, select the shader. If a pixel shader is used, it will be set and enabled by the shader backend.
6464      * If this shader backend is arbfp(most likely), then it will simply overwrite the last fixed function
6465      * replacement shader. If the shader backend is not ARB, it currently is important that the opengl implementation
6466      * type overwrites GL_ARB_fragment_program. This is currently the case with GLSL. If we really want to use
6467      * atifs or nvrc pixel shaders with arb fragment programs we'd have to disable GL_FRAGMENT_PROGRAM_ARB here
6468      *
6469      * Don't call shader_select if the vertex shader is dirty, because it will be called later on by the vertex
6470      * shader handler.
6471      */
6472     if (!isStateDirty(context, context->state_table[STATE_VSHADER].representative))
6473     {
6474         device->shader_backend->shader_select(context, use_pshader, use_vshader);
6475
6476         if (!isStateDirty(context, STATE_VERTEXSHADERCONSTANT) && (use_vshader || use_pshader))
6477             context_apply_state(context, state, STATE_VERTEXSHADERCONSTANT);
6478     }
6479     if (use_pshader)
6480         context_apply_state(context, state, STATE_PIXELSHADERCONSTANT);
6481 }
6482
6483 /* We can't link the fog states to the fragment state directly since the
6484  * vertex pipeline links them to FOGENABLE. A different linking in different
6485  * pipeline parts can't be expressed in the combined state table, so we need
6486  * to handle that with a forwarding function. The other invisible side effect
6487  * is that changing the fog start and fog end (which links to FOGENABLE in
6488  * vertex) results in the fragment_prog_arbfp function being called because
6489  * FOGENABLE is dirty, which calls this function here. */
6490 static void state_arbfp_fog(struct wined3d_context *context, const struct wined3d_state *state, DWORD state_id)
6491 {
6492     enum fogsource new_source;
6493
6494     TRACE("context %p, state %p, state_id %#x.\n", context, state, state_id);
6495
6496     if (!isStateDirty(context, STATE_PIXELSHADER))
6497         fragment_prog_arbfp(context, state, state_id);
6498
6499     if (!state->render_states[WINED3D_RS_FOGENABLE])
6500         return;
6501
6502     if (state->render_states[WINED3D_RS_FOGTABLEMODE] == WINED3D_FOG_NONE)
6503     {
6504         if (use_vs(state))
6505         {
6506             new_source = FOGSOURCE_VS;
6507         }
6508         else
6509         {
6510             if (state->render_states[WINED3D_RS_FOGVERTEXMODE] == WINED3D_FOG_NONE || context->last_was_rhw)
6511                 new_source = FOGSOURCE_COORD;
6512             else
6513                 new_source = FOGSOURCE_FFP;
6514         }
6515     }
6516     else
6517     {
6518         new_source = FOGSOURCE_FFP;
6519     }
6520
6521     if (new_source != context->fog_source)
6522     {
6523         context->fog_source = new_source;
6524         state_fogstartend(context, state, STATE_RENDER(WINED3D_RS_FOGSTART));
6525     }
6526 }
6527
6528 static void textransform(struct wined3d_context *context, const struct wined3d_state *state, DWORD state_id)
6529 {
6530     if (!isStateDirty(context, STATE_PIXELSHADER))
6531         fragment_prog_arbfp(context, state, state_id);
6532 }
6533
6534 static void state_srgbwrite(struct wined3d_context *context, const struct wined3d_state *state, DWORD state_id)
6535 {
6536     const struct wined3d_gl_info *gl_info = context->gl_info;
6537     const struct wined3d_surface *rt = state->fb->render_targets[0];
6538
6539     TRACE("context %p, state %p, state_id %#x.\n", context, state, state_id);
6540
6541     if (state->render_states[WINED3D_RS_SRGBWRITEENABLE]
6542             && rt->resource.format->flags & WINED3DFMT_FLAG_SRGB_WRITE)
6543         gl_info->gl_ops.gl.p_glEnable(GL_FRAMEBUFFER_SRGB);
6544     else
6545         gl_info->gl_ops.gl.p_glDisable(GL_FRAMEBUFFER_SRGB);
6546 }
6547
6548 static const struct StateEntryTemplate arbfp_fragmentstate_template[] =
6549 {
6550     {STATE_RENDER(WINED3D_RS_TEXTUREFACTOR),              { STATE_RENDER(WINED3D_RS_TEXTUREFACTOR),             state_texfactor_arbfp   }, WINED3D_GL_EXT_NONE             },
6551     {STATE_TEXTURESTAGE(0, WINED3D_TSS_COLOR_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6552     {STATE_TEXTURESTAGE(0, WINED3D_TSS_COLOR_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6553     {STATE_TEXTURESTAGE(0, WINED3D_TSS_COLOR_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6554     {STATE_TEXTURESTAGE(0, WINED3D_TSS_COLOR_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6555     {STATE_TEXTURESTAGE(0, WINED3D_TSS_ALPHA_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6556     {STATE_TEXTURESTAGE(0, WINED3D_TSS_ALPHA_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6557     {STATE_TEXTURESTAGE(0, WINED3D_TSS_ALPHA_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6558     {STATE_TEXTURESTAGE(0, WINED3D_TSS_ALPHA_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6559     {STATE_TEXTURESTAGE(0, WINED3D_TSS_RESULT_ARG),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6560     {STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6561     {STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6562     {STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6563     {STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6564     {STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6565     {STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6566     {STATE_TEXTURESTAGE(1, WINED3D_TSS_COLOR_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6567     {STATE_TEXTURESTAGE(1, WINED3D_TSS_COLOR_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6568     {STATE_TEXTURESTAGE(1, WINED3D_TSS_COLOR_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6569     {STATE_TEXTURESTAGE(1, WINED3D_TSS_COLOR_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6570     {STATE_TEXTURESTAGE(1, WINED3D_TSS_ALPHA_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6571     {STATE_TEXTURESTAGE(1, WINED3D_TSS_ALPHA_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6572     {STATE_TEXTURESTAGE(1, WINED3D_TSS_ALPHA_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6573     {STATE_TEXTURESTAGE(1, WINED3D_TSS_ALPHA_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6574     {STATE_TEXTURESTAGE(1, WINED3D_TSS_RESULT_ARG),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6575     {STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6576     {STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6577     {STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6578     {STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6579     {STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6580     {STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6581     {STATE_TEXTURESTAGE(2, WINED3D_TSS_COLOR_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6582     {STATE_TEXTURESTAGE(2, WINED3D_TSS_COLOR_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6583     {STATE_TEXTURESTAGE(2, WINED3D_TSS_COLOR_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6584     {STATE_TEXTURESTAGE(2, WINED3D_TSS_COLOR_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6585     {STATE_TEXTURESTAGE(2, WINED3D_TSS_ALPHA_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6586     {STATE_TEXTURESTAGE(2, WINED3D_TSS_ALPHA_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6587     {STATE_TEXTURESTAGE(2, WINED3D_TSS_ALPHA_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6588     {STATE_TEXTURESTAGE(2, WINED3D_TSS_ALPHA_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6589     {STATE_TEXTURESTAGE(2, WINED3D_TSS_RESULT_ARG),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6590     {STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6591     {STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6592     {STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6593     {STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6594     {STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6595     {STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6596     {STATE_TEXTURESTAGE(3, WINED3D_TSS_COLOR_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6597     {STATE_TEXTURESTAGE(3, WINED3D_TSS_COLOR_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6598     {STATE_TEXTURESTAGE(3, WINED3D_TSS_COLOR_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6599     {STATE_TEXTURESTAGE(3, WINED3D_TSS_COLOR_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6600     {STATE_TEXTURESTAGE(3, WINED3D_TSS_ALPHA_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6601     {STATE_TEXTURESTAGE(3, WINED3D_TSS_ALPHA_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6602     {STATE_TEXTURESTAGE(3, WINED3D_TSS_ALPHA_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6603     {STATE_TEXTURESTAGE(3, WINED3D_TSS_ALPHA_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6604     {STATE_TEXTURESTAGE(3, WINED3D_TSS_RESULT_ARG),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6605     {STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6606     {STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6607     {STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6608     {STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6609     {STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6610     {STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6611     {STATE_TEXTURESTAGE(4, WINED3D_TSS_COLOR_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6612     {STATE_TEXTURESTAGE(4, WINED3D_TSS_COLOR_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6613     {STATE_TEXTURESTAGE(4, WINED3D_TSS_COLOR_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6614     {STATE_TEXTURESTAGE(4, WINED3D_TSS_COLOR_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6615     {STATE_TEXTURESTAGE(4, WINED3D_TSS_ALPHA_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6616     {STATE_TEXTURESTAGE(4, WINED3D_TSS_ALPHA_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6617     {STATE_TEXTURESTAGE(4, WINED3D_TSS_ALPHA_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6618     {STATE_TEXTURESTAGE(4, WINED3D_TSS_ALPHA_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6619     {STATE_TEXTURESTAGE(4, WINED3D_TSS_RESULT_ARG),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6620     {STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6621     {STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6622     {STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6623     {STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6624     {STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6625     {STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6626     {STATE_TEXTURESTAGE(5, WINED3D_TSS_COLOR_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6627     {STATE_TEXTURESTAGE(5, WINED3D_TSS_COLOR_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6628     {STATE_TEXTURESTAGE(5, WINED3D_TSS_COLOR_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6629     {STATE_TEXTURESTAGE(5, WINED3D_TSS_COLOR_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6630     {STATE_TEXTURESTAGE(5, WINED3D_TSS_ALPHA_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6631     {STATE_TEXTURESTAGE(5, WINED3D_TSS_ALPHA_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6632     {STATE_TEXTURESTAGE(5, WINED3D_TSS_ALPHA_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6633     {STATE_TEXTURESTAGE(5, WINED3D_TSS_ALPHA_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6634     {STATE_TEXTURESTAGE(5, WINED3D_TSS_RESULT_ARG),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6635     {STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6636     {STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6637     {STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6638     {STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6639     {STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6640     {STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6641     {STATE_TEXTURESTAGE(6, WINED3D_TSS_COLOR_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6642     {STATE_TEXTURESTAGE(6, WINED3D_TSS_COLOR_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6643     {STATE_TEXTURESTAGE(6, WINED3D_TSS_COLOR_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6644     {STATE_TEXTURESTAGE(6, WINED3D_TSS_COLOR_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6645     {STATE_TEXTURESTAGE(6, WINED3D_TSS_ALPHA_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6646     {STATE_TEXTURESTAGE(6, WINED3D_TSS_ALPHA_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6647     {STATE_TEXTURESTAGE(6, WINED3D_TSS_ALPHA_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6648     {STATE_TEXTURESTAGE(6, WINED3D_TSS_ALPHA_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6649     {STATE_TEXTURESTAGE(6, WINED3D_TSS_RESULT_ARG),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6650     {STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6651     {STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6652     {STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6653     {STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6654     {STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6655     {STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6656     {STATE_TEXTURESTAGE(7, WINED3D_TSS_COLOR_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6657     {STATE_TEXTURESTAGE(7, WINED3D_TSS_COLOR_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6658     {STATE_TEXTURESTAGE(7, WINED3D_TSS_COLOR_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6659     {STATE_TEXTURESTAGE(7, WINED3D_TSS_COLOR_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6660     {STATE_TEXTURESTAGE(7, WINED3D_TSS_ALPHA_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6661     {STATE_TEXTURESTAGE(7, WINED3D_TSS_ALPHA_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6662     {STATE_TEXTURESTAGE(7, WINED3D_TSS_ALPHA_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6663     {STATE_TEXTURESTAGE(7, WINED3D_TSS_ALPHA_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6664     {STATE_TEXTURESTAGE(7, WINED3D_TSS_RESULT_ARG),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6665     {STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6666     {STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6667     {STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6668     {STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6669     {STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6670     {STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6671     {STATE_PIXELSHADER,                                   { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6672     {STATE_RENDER(WINED3D_RS_FOGENABLE),                  { STATE_RENDER(WINED3D_RS_FOGENABLE),                 state_arbfp_fog         }, WINED3D_GL_EXT_NONE             },
6673     {STATE_RENDER(WINED3D_RS_FOGTABLEMODE),               { STATE_RENDER(WINED3D_RS_FOGENABLE),                 NULL                    }, WINED3D_GL_EXT_NONE             },
6674     {STATE_RENDER(WINED3D_RS_FOGVERTEXMODE),              { STATE_RENDER(WINED3D_RS_FOGENABLE),                 NULL                    }, WINED3D_GL_EXT_NONE             },
6675     {STATE_RENDER(WINED3D_RS_FOGSTART),                   { STATE_RENDER(WINED3D_RS_FOGSTART),                  state_fogstartend       }, WINED3D_GL_EXT_NONE             },
6676     {STATE_RENDER(WINED3D_RS_FOGEND),                     { STATE_RENDER(WINED3D_RS_FOGSTART),                  NULL                    }, WINED3D_GL_EXT_NONE             },
6677     {STATE_RENDER(WINED3D_RS_SRGBWRITEENABLE),            { STATE_RENDER(WINED3D_RS_SRGBWRITEENABLE),           state_srgbwrite         }, ARB_FRAMEBUFFER_SRGB            },
6678     {STATE_RENDER(WINED3D_RS_SRGBWRITEENABLE),            { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6679     {STATE_RENDER(WINED3D_RS_FOGCOLOR),                   { STATE_RENDER(WINED3D_RS_FOGCOLOR),                  state_fogcolor          }, WINED3D_GL_EXT_NONE             },
6680     {STATE_RENDER(WINED3D_RS_FOGDENSITY),                 { STATE_RENDER(WINED3D_RS_FOGDENSITY),                state_fogdensity        }, WINED3D_GL_EXT_NONE             },
6681     {STATE_TEXTURESTAGE(0,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(0, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6682     {STATE_TEXTURESTAGE(1,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(1, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6683     {STATE_TEXTURESTAGE(2,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(2, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6684     {STATE_TEXTURESTAGE(3,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(3, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6685     {STATE_TEXTURESTAGE(4,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(4, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6686     {STATE_TEXTURESTAGE(5,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(5, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6687     {STATE_TEXTURESTAGE(6,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(6, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6688     {STATE_TEXTURESTAGE(7,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(7, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6689     {STATE_RENDER(WINED3D_RS_SPECULARENABLE),             { STATE_RENDER(WINED3D_RS_SPECULARENABLE),            state_arb_specularenable}, WINED3D_GL_EXT_NONE             },
6690     {0 /* Terminate */,                                   { 0,                                                  0                       }, WINED3D_GL_EXT_NONE             },
6691 };
6692
6693 const struct fragment_pipeline arbfp_fragment_pipeline = {
6694     arbfp_enable,
6695     arbfp_get_caps,
6696     arbfp_alloc,
6697     arbfp_free,
6698     shader_arb_color_fixup_supported,
6699     arbfp_fragmentstate_template,
6700     TRUE /* We can disable projected textures */
6701 };
6702
6703 struct arbfp_blit_priv {
6704     GLenum yuy2_rect_shader, yuy2_2d_shader;
6705     GLenum uyvy_rect_shader, uyvy_2d_shader;
6706     GLenum yv12_rect_shader, yv12_2d_shader;
6707     GLenum p8_rect_shader, p8_2d_shader;
6708     GLuint palette_texture;
6709 };
6710
6711 static HRESULT arbfp_blit_alloc(struct wined3d_device *device)
6712 {
6713     device->blit_priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(struct arbfp_blit_priv));
6714     if(!device->blit_priv) {
6715         ERR("Out of memory\n");
6716         return E_OUTOFMEMORY;
6717     }
6718     return WINED3D_OK;
6719 }
6720
6721 /* Context activation is done by the caller. */
6722 static void arbfp_blit_free(struct wined3d_device *device)
6723 {
6724     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
6725     struct arbfp_blit_priv *priv = device->blit_priv;
6726
6727     ENTER_GL();
6728     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yuy2_rect_shader));
6729     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yuy2_2d_shader));
6730     GL_EXTCALL(glDeleteProgramsARB(1, &priv->uyvy_rect_shader));
6731     GL_EXTCALL(glDeleteProgramsARB(1, &priv->uyvy_2d_shader));
6732     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yv12_rect_shader));
6733     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yv12_2d_shader));
6734     GL_EXTCALL(glDeleteProgramsARB(1, &priv->p8_rect_shader));
6735     GL_EXTCALL(glDeleteProgramsARB(1, &priv->p8_2d_shader));
6736     checkGLcall("Delete yuv and p8 programs");
6737
6738     if (priv->palette_texture)
6739         gl_info->gl_ops.gl.p_glDeleteTextures(1, &priv->palette_texture);
6740     LEAVE_GL();
6741
6742     HeapFree(GetProcessHeap(), 0, device->blit_priv);
6743     device->blit_priv = NULL;
6744 }
6745
6746 static BOOL gen_planar_yuv_read(struct wined3d_shader_buffer *buffer, enum complex_fixup fixup,
6747         GLenum textype, char *luminance)
6748 {
6749     char chroma;
6750     const char *tex, *texinstr;
6751
6752     if (fixup == COMPLEX_FIXUP_UYVY) {
6753         chroma = 'x';
6754         *luminance = 'w';
6755     } else {
6756         chroma = 'w';
6757         *luminance = 'x';
6758     }
6759     switch(textype) {
6760         case GL_TEXTURE_2D:             tex = "2D";     texinstr = "TXP"; break;
6761         case GL_TEXTURE_RECTANGLE_ARB:  tex = "RECT";   texinstr = "TEX"; break;
6762         default:
6763             /* This is more tricky than just replacing the texture type - we have to navigate
6764              * properly in the texture to find the correct chroma values
6765              */
6766             FIXME("Implement yuv correction for non-2d, non-rect textures\n");
6767             return FALSE;
6768     }
6769
6770     /* First we have to read the chroma values. This means we need at least two pixels(no filtering),
6771      * or 4 pixels(with filtering). To get the unmodified chromas, we have to rid ourselves of the
6772      * filtering when we sample the texture.
6773      *
6774      * These are the rules for reading the chroma:
6775      *
6776      * Even pixel: Cr
6777      * Even pixel: U
6778      * Odd pixel: V
6779      *
6780      * So we have to get the sampling x position in non-normalized coordinates in integers
6781      */
6782     if(textype != GL_TEXTURE_RECTANGLE_ARB) {
6783         shader_addline(buffer, "MUL texcrd.xy, fragment.texcoord[0], size.x;\n");
6784         shader_addline(buffer, "MOV texcrd.w, size.x;\n");
6785     } else {
6786         shader_addline(buffer, "MOV texcrd, fragment.texcoord[0];\n");
6787     }
6788     /* We must not allow filtering between pixel x and x+1, this would mix U and V
6789      * Vertical filtering is ok. However, bear in mind that the pixel center is at
6790      * 0.5, so add 0.5.
6791      */
6792     shader_addline(buffer, "FLR texcrd.x, texcrd.x;\n");
6793     shader_addline(buffer, "ADD texcrd.x, texcrd.x, coef.y;\n");
6794
6795     /* Divide the x coordinate by 0.5 and get the fraction. This gives 0.25 and 0.75 for the
6796      * even and odd pixels respectively
6797      */
6798     shader_addline(buffer, "MUL texcrd2, texcrd, coef.y;\n");
6799     shader_addline(buffer, "FRC texcrd2, texcrd2;\n");
6800
6801     /* Sample Pixel 1 */
6802     shader_addline(buffer, "%s luminance, texcrd, texture[0], %s;\n", texinstr, tex);
6803
6804     /* Put the value into either of the chroma values */
6805     shader_addline(buffer, "SGE temp.x, texcrd2.x, coef.y;\n");
6806     shader_addline(buffer, "MUL chroma.x, luminance.%c, temp.x;\n", chroma);
6807     shader_addline(buffer, "SLT temp.x, texcrd2.x, coef.y;\n");
6808     shader_addline(buffer, "MUL chroma.y, luminance.%c, temp.x;\n", chroma);
6809
6810     /* Sample pixel 2. If we read an even pixel(SLT above returned 1), sample
6811      * the pixel right to the current one. Otherwise, sample the left pixel.
6812      * Bias and scale the SLT result to -1;1 and add it to the texcrd.x.
6813      */
6814     shader_addline(buffer, "MAD temp.x, temp.x, coef.z, -coef.x;\n");
6815     shader_addline(buffer, "ADD texcrd.x, texcrd, temp.x;\n");
6816     shader_addline(buffer, "%s luminance, texcrd, texture[0], %s;\n", texinstr, tex);
6817
6818     /* Put the value into the other chroma */
6819     shader_addline(buffer, "SGE temp.x, texcrd2.x, coef.y;\n");
6820     shader_addline(buffer, "MAD chroma.y, luminance.%c, temp.x, chroma.y;\n", chroma);
6821     shader_addline(buffer, "SLT temp.x, texcrd2.x, coef.y;\n");
6822     shader_addline(buffer, "MAD chroma.x, luminance.%c, temp.x, chroma.x;\n", chroma);
6823
6824     /* TODO: If filtering is enabled, sample a 2nd pair of pixels left or right of
6825      * the current one and lerp the two U and V values
6826      */
6827
6828     /* This gives the correctly filtered luminance value */
6829     shader_addline(buffer, "TEX luminance, fragment.texcoord[0], texture[0], %s;\n", tex);
6830
6831     return TRUE;
6832 }
6833
6834 static BOOL gen_yv12_read(struct wined3d_shader_buffer *buffer, GLenum textype, char *luminance)
6835 {
6836     const char *tex;
6837
6838     switch(textype) {
6839         case GL_TEXTURE_2D:             tex = "2D";     break;
6840         case GL_TEXTURE_RECTANGLE_ARB:  tex = "RECT";   break;
6841         default:
6842             FIXME("Implement yv12 correction for non-2d, non-rect textures\n");
6843             return FALSE;
6844     }
6845
6846     /* YV12 surfaces contain a WxH sized luminance plane, followed by a (W/2)x(H/2)
6847      * V and a (W/2)x(H/2) U plane, each with 8 bit per pixel. So the effective
6848      * bitdepth is 12 bits per pixel. Since the U and V planes have only half the
6849      * pitch of the luminance plane, the packing into the gl texture is a bit
6850      * unfortunate. If the whole texture is interpreted as luminance data it looks
6851      * approximately like this:
6852      *
6853      *        +----------------------------------+----
6854      *        |                                  |
6855      *        |                                  |
6856      *        |                                  |
6857      *        |                                  |
6858      *        |                                  |   2
6859      *        |            LUMINANCE             |   -
6860      *        |                                  |   3
6861      *        |                                  |
6862      *        |                                  |
6863      *        |                                  |
6864      *        |                                  |
6865      *        +----------------+-----------------+----
6866      *        |                |                 |
6867      *        |  U even rows   |  U odd rows     |
6868      *        |                |                 |   1
6869      *        +----------------+------------------   -
6870      *        |                |                 |   3
6871      *        |  V even rows   |  V odd rows     |
6872      *        |                |                 |
6873      *        +----------------+-----------------+----
6874      *        |                |                 |
6875      *        |     0.5        |       0.5       |
6876      *
6877      * So it appears as if there are 4 chroma images, but in fact the odd rows
6878      * in the chroma images are in the same row as the even ones. So its is
6879      * kinda tricky to read
6880      *
6881      * When reading from rectangle textures, keep in mind that the input y coordinates
6882      * go from 0 to d3d_height, whereas the opengl texture height is 1.5 * d3d_height
6883      */
6884     shader_addline(buffer, "PARAM yv12_coef = {%f, %f, %f, %f};\n",
6885             2.0f / 3.0f, 1.0f / 6.0f, (2.0f / 3.0f) + (1.0f / 6.0f), 1.0f / 3.0f);
6886
6887     shader_addline(buffer, "MOV texcrd, fragment.texcoord[0];\n");
6888     /* the chroma planes have only half the width */
6889     shader_addline(buffer, "MUL texcrd.x, texcrd.x, coef.y;\n");
6890
6891     /* The first value is between 2/3 and 5/6th of the texture's height, so scale+bias
6892      * the coordinate. Also read the right side of the image when reading odd lines
6893      *
6894      * Don't forget to clamp the y values in into the range, otherwise we'll get filtering
6895      * bleeding
6896      */
6897     if(textype == GL_TEXTURE_2D) {
6898
6899         shader_addline(buffer, "RCP chroma.w, size.y;\n");
6900
6901         shader_addline(buffer, "MUL texcrd2.y, texcrd.y, size.y;\n");
6902
6903         shader_addline(buffer, "FLR texcrd2.y, texcrd2.y;\n");
6904         shader_addline(buffer, "MAD texcrd.y, texcrd.y, yv12_coef.y, yv12_coef.x;\n");
6905
6906         /* Read odd lines from the right side(add size * 0.5 to the x coordinate */
6907         shader_addline(buffer, "ADD texcrd2.x, texcrd2.y, yv12_coef.y;\n"); /* To avoid 0.5 == 0.5 comparisons */
6908         shader_addline(buffer, "FRC texcrd2.x, texcrd2.x;\n");
6909         shader_addline(buffer, "SGE texcrd2.x, texcrd2.x, coef.y;\n");
6910         shader_addline(buffer, "MAD texcrd.x, texcrd2.x, coef.y, texcrd.x;\n");
6911
6912         /* clamp, keep the half pixel origin in mind */
6913         shader_addline(buffer, "MAD temp.y, coef.y, chroma.w, yv12_coef.x;\n");
6914         shader_addline(buffer, "MAX texcrd.y, temp.y, texcrd.y;\n");
6915         shader_addline(buffer, "MAD temp.y, -coef.y, chroma.w, yv12_coef.z;\n");
6916         shader_addline(buffer, "MIN texcrd.y, temp.y, texcrd.y;\n");
6917     } else {
6918         /* Read from [size - size+size/4] */
6919         shader_addline(buffer, "FLR texcrd.y, texcrd.y;\n");
6920         shader_addline(buffer, "MAD texcrd.y, texcrd.y, coef.w, size.y;\n");
6921
6922         /* Read odd lines from the right side(add size * 0.5 to the x coordinate */
6923         shader_addline(buffer, "ADD texcrd2.x, texcrd.y, yv12_coef.y;\n"); /* To avoid 0.5 == 0.5 comparisons */
6924         shader_addline(buffer, "FRC texcrd2.x, texcrd2.x;\n");
6925         shader_addline(buffer, "SGE texcrd2.x, texcrd2.x, coef.y;\n");
6926         shader_addline(buffer, "MUL texcrd2.x, texcrd2.x, size.x;\n");
6927         shader_addline(buffer, "MAD texcrd.x, texcrd2.x, coef.y, texcrd.x;\n");
6928
6929         /* Make sure to read exactly from the pixel center */
6930         shader_addline(buffer, "FLR texcrd.y, texcrd.y;\n");
6931         shader_addline(buffer, "ADD texcrd.y, texcrd.y, coef.y;\n");
6932
6933         /* Clamp */
6934         shader_addline(buffer, "MAD temp.y, size.y, coef.w, size.y;\n");
6935         shader_addline(buffer, "ADD temp.y, temp.y, -coef.y;\n");
6936         shader_addline(buffer, "MIN texcrd.y, temp.y, texcrd.y;\n");
6937         shader_addline(buffer, "ADD temp.y, size.y, -coef.y;\n");
6938         shader_addline(buffer, "MAX texcrd.y, temp.y, texcrd.y;\n");
6939     }
6940     /* Read the texture, put the result into the output register */
6941     shader_addline(buffer, "TEX temp, texcrd, texture[0], %s;\n", tex);
6942     shader_addline(buffer, "MOV chroma.x, temp.w;\n");
6943
6944     /* The other chroma value is 1/6th of the texture lower, from 5/6th to 6/6th
6945      * No need to clamp because we're just reusing the already clamped value from above
6946      */
6947     if(textype == GL_TEXTURE_2D) {
6948         shader_addline(buffer, "ADD texcrd.y, texcrd.y, yv12_coef.y;\n");
6949     } else {
6950         shader_addline(buffer, "MAD texcrd.y, size.y, coef.w, texcrd.y;\n");
6951     }
6952     shader_addline(buffer, "TEX temp, texcrd, texture[0], %s;\n", tex);
6953     shader_addline(buffer, "MOV chroma.y, temp.w;\n");
6954
6955     /* Sample the luminance value. It is in the top 2/3rd of the texture, so scale the y coordinate.
6956      * Clamp the y coordinate to prevent the chroma values from bleeding into the sampled luminance
6957      * values due to filtering
6958      */
6959     shader_addline(buffer, "MOV texcrd, fragment.texcoord[0];\n");
6960     if(textype == GL_TEXTURE_2D) {
6961         /* Multiply the y coordinate by 2/3 and clamp it */
6962         shader_addline(buffer, "MUL texcrd.y, texcrd.y, yv12_coef.x;\n");
6963         shader_addline(buffer, "MAD temp.y, -coef.y, chroma.w, yv12_coef.x;\n");
6964         shader_addline(buffer, "MIN texcrd.y, temp.y, texcrd.y;\n");
6965         shader_addline(buffer, "TEX luminance, texcrd, texture[0], %s;\n", tex);
6966     } else {
6967         /* Reading from texture_rectangles is pretty straightforward, just use the unmodified
6968          * texture coordinate. It is still a good idea to clamp it though, since the opengl texture
6969          * is bigger
6970          */
6971         shader_addline(buffer, "ADD temp.x, size.y, -coef.y;\n");
6972         shader_addline(buffer, "MIN texcrd.y, texcrd.y, size.x;\n");
6973         shader_addline(buffer, "TEX luminance, texcrd, texture[0], %s;\n", tex);
6974     }
6975     *luminance = 'a';
6976
6977     return TRUE;
6978 }
6979
6980 static GLuint gen_p8_shader(struct arbfp_blit_priv *priv,
6981         const struct wined3d_gl_info *gl_info, GLenum textype)
6982 {
6983     GLenum shader;
6984     struct wined3d_shader_buffer buffer;
6985     GLint pos;
6986
6987     /* Shader header */
6988     if (!shader_buffer_init(&buffer))
6989     {
6990         ERR("Failed to initialize shader buffer.\n");
6991         return 0;
6992     }
6993
6994     ENTER_GL();
6995     GL_EXTCALL(glGenProgramsARB(1, &shader));
6996     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader));
6997     LEAVE_GL();
6998     if(!shader) {
6999         shader_buffer_free(&buffer);
7000         return 0;
7001     }
7002
7003     shader_addline(&buffer, "!!ARBfp1.0\n");
7004     shader_addline(&buffer, "TEMP index;\n");
7005
7006     /* { 255/256, 0.5/255*255/256, 0, 0 } */
7007     shader_addline(&buffer, "PARAM constants = { 0.996, 0.00195, 0, 0 };\n");
7008
7009     /* The alpha-component contains the palette index */
7010     if(textype == GL_TEXTURE_RECTANGLE_ARB)
7011         shader_addline(&buffer, "TXP index, fragment.texcoord[0], texture[0], RECT;\n");
7012     else
7013         shader_addline(&buffer, "TEX index, fragment.texcoord[0], texture[0], 2D;\n");
7014
7015     /* Scale the index by 255/256 and add a bias of '0.5' in order to sample in the middle */
7016     shader_addline(&buffer, "MAD index.a, index.a, constants.x, constants.y;\n");
7017
7018     /* Use the alpha-component as an index in the palette to get the final color */
7019     shader_addline(&buffer, "TEX result.color, index.a, texture[1], 1D;\n");
7020     shader_addline(&buffer, "END\n");
7021
7022     ENTER_GL();
7023     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
7024             strlen(buffer.buffer), buffer.buffer));
7025     checkGLcall("glProgramStringARB()");
7026
7027     gl_info->gl_ops.gl.p_glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
7028     if (pos != -1)
7029     {
7030         FIXME("Fragment program error at position %d: %s\n\n", pos,
7031               debugstr_a((const char *)gl_info->gl_ops.gl.p_glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
7032         shader_arb_dump_program_source(buffer.buffer);
7033     }
7034
7035     if (textype == GL_TEXTURE_RECTANGLE_ARB)
7036         priv->p8_rect_shader = shader;
7037     else
7038         priv->p8_2d_shader = shader;
7039
7040     shader_buffer_free(&buffer);
7041     LEAVE_GL();
7042
7043     return shader;
7044 }
7045
7046 /* Context activation is done by the caller. */
7047 static void upload_palette(const struct wined3d_surface *surface, struct wined3d_context *context)
7048 {
7049     BYTE table[256][4];
7050     struct wined3d_device *device = surface->resource.device;
7051     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
7052     struct arbfp_blit_priv *priv = device->blit_priv;
7053     BOOL colorkey = (surface->CKeyFlags & WINEDDSD_CKSRCBLT) != 0;
7054
7055     d3dfmt_p8_init_palette(surface, table, colorkey);
7056
7057     ENTER_GL();
7058
7059     if (gl_info->supported[APPLE_CLIENT_STORAGE])
7060     {
7061         gl_info->gl_ops.gl.p_glPixelStorei(GL_UNPACK_CLIENT_STORAGE_APPLE, GL_FALSE);
7062         checkGLcall("glPixelStorei(GL_UNPACK_CLIENT_STORAGE_APPLE, GL_FALSE)");
7063     }
7064
7065     if (!priv->palette_texture)
7066         gl_info->gl_ops.gl.p_glGenTextures(1, &priv->palette_texture);
7067
7068     GL_EXTCALL(glActiveTextureARB(GL_TEXTURE1));
7069     gl_info->gl_ops.gl.p_glBindTexture(GL_TEXTURE_1D, priv->palette_texture);
7070
7071     gl_info->gl_ops.gl.p_glTexEnvi(GL_TEXTURE_ENV, GL_TEXTURE_ENV_MODE, GL_REPLACE);
7072
7073     gl_info->gl_ops.gl.p_glTexParameteri(GL_TEXTURE_1D, GL_TEXTURE_MIN_FILTER, GL_NEAREST);
7074     /* Make sure we have discrete color levels. */
7075     gl_info->gl_ops.gl.p_glTexParameteri(GL_TEXTURE_1D, GL_TEXTURE_MAG_FILTER, GL_NEAREST);
7076     gl_info->gl_ops.gl.p_glTexParameteri(GL_TEXTURE_1D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
7077     /* Upload the palette */
7078     /* TODO: avoid unneeded uploads in the future by adding some SFLAG_PALETTE_DIRTY mechanism */
7079     gl_info->gl_ops.gl.p_glTexImage1D(GL_TEXTURE_1D, 0, GL_RGBA, 256, 0, GL_RGBA, GL_UNSIGNED_BYTE, table);
7080
7081     if (gl_info->supported[APPLE_CLIENT_STORAGE])
7082     {
7083         gl_info->gl_ops.gl.p_glPixelStorei(GL_UNPACK_CLIENT_STORAGE_APPLE, GL_TRUE);
7084         checkGLcall("glPixelStorei(GL_UNPACK_CLIENT_STORAGE_APPLE, GL_TRUE)");
7085     }
7086
7087     /* Switch back to unit 0 in which the 2D texture will be stored. */
7088     context_active_texture(context, gl_info, 0);
7089     LEAVE_GL();
7090 }
7091
7092 /* Context activation is done by the caller. */
7093 static GLuint gen_yuv_shader(struct arbfp_blit_priv *priv, const struct wined3d_gl_info *gl_info,
7094         enum complex_fixup yuv_fixup, GLenum textype)
7095 {
7096     GLenum shader;
7097     struct wined3d_shader_buffer buffer;
7098     char luminance_component;
7099     GLint pos;
7100
7101     /* Shader header */
7102     if (!shader_buffer_init(&buffer))
7103     {
7104         ERR("Failed to initialize shader buffer.\n");
7105         return 0;
7106     }
7107
7108     ENTER_GL();
7109     GL_EXTCALL(glGenProgramsARB(1, &shader));
7110     checkGLcall("GL_EXTCALL(glGenProgramsARB(1, &shader))");
7111     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader));
7112     checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader)");
7113     LEAVE_GL();
7114     if(!shader) {
7115         shader_buffer_free(&buffer);
7116         return 0;
7117     }
7118
7119     /* The YUY2 and UYVY formats contain two pixels packed into a 32 bit macropixel,
7120      * giving effectively 16 bit per pixel. The color consists of a luminance(Y) and
7121      * two chroma(U and V) values. Each macropixel has two luminance values, one for
7122      * each single pixel it contains, and one U and one V value shared between both
7123      * pixels.
7124      *
7125      * The data is loaded into an A8L8 texture. With YUY2, the luminance component
7126      * contains the luminance and alpha the chroma. With UYVY it is vice versa. Thus
7127      * take the format into account when generating the read swizzles
7128      *
7129      * Reading the Y value is straightforward - just sample the texture. The hardware
7130      * takes care of filtering in the horizontal and vertical direction.
7131      *
7132      * Reading the U and V values is harder. We have to avoid filtering horizontally,
7133      * because that would mix the U and V values of one pixel or two adjacent pixels.
7134      * Thus floor the texture coordinate and add 0.5 to get an unfiltered read,
7135      * regardless of the filtering setting. Vertical filtering works automatically
7136      * though - the U and V values of two rows are mixed nicely.
7137      *
7138      * Apart of avoiding filtering issues, the code has to know which value it just
7139      * read, and where it can find the other one. To determine this, it checks if
7140      * it sampled an even or odd pixel, and shifts the 2nd read accordingly.
7141      *
7142      * Handling horizontal filtering of U and V values requires reading a 2nd pair
7143      * of pixels, extracting U and V and mixing them. This is not implemented yet.
7144      *
7145      * An alternative implementation idea is to load the texture as A8R8G8B8 texture,
7146      * with width / 2. This way one read gives all 3 values, finding U and V is easy
7147      * in an unfiltered situation. Finding the luminance on the other hand requires
7148      * finding out if it is an odd or even pixel. The real drawback of this approach
7149      * is filtering. This would have to be emulated completely in the shader, reading
7150      * up two 2 packed pixels in up to 2 rows and interpolating both horizontally and
7151      * vertically. Beyond that it would require adjustments to the texture handling
7152      * code to deal with the width scaling
7153      */
7154     shader_addline(&buffer, "!!ARBfp1.0\n");
7155     shader_addline(&buffer, "TEMP luminance;\n");
7156     shader_addline(&buffer, "TEMP temp;\n");
7157     shader_addline(&buffer, "TEMP chroma;\n");
7158     shader_addline(&buffer, "TEMP texcrd;\n");
7159     shader_addline(&buffer, "TEMP texcrd2;\n");
7160     shader_addline(&buffer, "PARAM coef = {1.0, 0.5, 2.0, 0.25};\n");
7161     shader_addline(&buffer, "PARAM yuv_coef = {1.403, 0.344, 0.714, 1.770};\n");
7162     shader_addline(&buffer, "PARAM size = program.local[0];\n");
7163
7164     switch (yuv_fixup)
7165     {
7166         case COMPLEX_FIXUP_UYVY:
7167         case COMPLEX_FIXUP_YUY2:
7168             if (!gen_planar_yuv_read(&buffer, yuv_fixup, textype, &luminance_component))
7169             {
7170                 shader_buffer_free(&buffer);
7171                 return 0;
7172             }
7173             break;
7174
7175         case COMPLEX_FIXUP_YV12:
7176             if (!gen_yv12_read(&buffer, textype, &luminance_component))
7177             {
7178                 shader_buffer_free(&buffer);
7179                 return 0;
7180             }
7181             break;
7182
7183         default:
7184             FIXME("Unsupported YUV fixup %#x\n", yuv_fixup);
7185             shader_buffer_free(&buffer);
7186             return 0;
7187     }
7188
7189     /* Calculate the final result. Formula is taken from
7190      * http://www.fourcc.org/fccyvrgb.php. Note that the chroma
7191      * ranges from -0.5 to 0.5
7192      */
7193     shader_addline(&buffer, "SUB chroma.xy, chroma, coef.y;\n");
7194
7195     shader_addline(&buffer, "MAD result.color.x, chroma.x, yuv_coef.x, luminance.%c;\n", luminance_component);
7196     shader_addline(&buffer, "MAD temp.x, -chroma.y, yuv_coef.y, luminance.%c;\n", luminance_component);
7197     shader_addline(&buffer, "MAD result.color.y, -chroma.x, yuv_coef.z, temp.x;\n");
7198     shader_addline(&buffer, "MAD result.color.z, chroma.y, yuv_coef.w, luminance.%c;\n", luminance_component);
7199     shader_addline(&buffer, "END\n");
7200
7201     ENTER_GL();
7202     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
7203             strlen(buffer.buffer), buffer.buffer));
7204     checkGLcall("glProgramStringARB()");
7205
7206     gl_info->gl_ops.gl.p_glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
7207     if (pos != -1)
7208     {
7209         FIXME("Fragment program error at position %d: %s\n\n", pos,
7210               debugstr_a((const char *)gl_info->gl_ops.gl.p_glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
7211         shader_arb_dump_program_source(buffer.buffer);
7212     }
7213     else
7214     {
7215         GLint native;
7216
7217         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
7218         checkGLcall("glGetProgramivARB()");
7219         if (!native) WARN("Program exceeds native resource limits.\n");
7220     }
7221
7222     shader_buffer_free(&buffer);
7223     LEAVE_GL();
7224
7225     switch (yuv_fixup)
7226     {
7227         case COMPLEX_FIXUP_YUY2:
7228             if (textype == GL_TEXTURE_RECTANGLE_ARB) priv->yuy2_rect_shader = shader;
7229             else priv->yuy2_2d_shader = shader;
7230             break;
7231
7232         case COMPLEX_FIXUP_UYVY:
7233             if (textype == GL_TEXTURE_RECTANGLE_ARB) priv->uyvy_rect_shader = shader;
7234             else priv->uyvy_2d_shader = shader;
7235             break;
7236
7237         case COMPLEX_FIXUP_YV12:
7238             if (textype == GL_TEXTURE_RECTANGLE_ARB) priv->yv12_rect_shader = shader;
7239             else priv->yv12_2d_shader = shader;
7240             break;
7241         default:
7242             ERR("Unsupported complex fixup: %d\n", yuv_fixup);
7243     }
7244
7245     return shader;
7246 }
7247
7248 /* Context activation is done by the caller. */
7249 static HRESULT arbfp_blit_set(void *blit_priv, struct wined3d_context *context, const struct wined3d_surface *surface)
7250 {
7251     GLenum shader;
7252     float size[4] = {(float) surface->pow2Width, (float) surface->pow2Height, 1.0f, 1.0f};
7253     struct arbfp_blit_priv *priv = blit_priv;
7254     enum complex_fixup fixup;
7255     const struct wined3d_gl_info *gl_info = context->gl_info;
7256     GLenum textype;
7257
7258     if (surface->container.type == WINED3D_CONTAINER_TEXTURE)
7259         textype = surface->container.u.texture->target;
7260     else
7261         textype = surface->texture_target;
7262
7263     if (surface->flags & SFLAG_CONVERTED)
7264     {
7265         ENTER_GL();
7266         gl_info->gl_ops.gl.p_glEnable(textype);
7267         checkGLcall("glEnable(textype)");
7268         LEAVE_GL();
7269         return WINED3D_OK;
7270     }
7271
7272     if (!is_complex_fixup(surface->resource.format->color_fixup))
7273     {
7274         TRACE("Fixup:\n");
7275         dump_color_fixup_desc(surface->resource.format->color_fixup);
7276         /* Don't bother setting up a shader for unconverted formats */
7277         ENTER_GL();
7278         gl_info->gl_ops.gl.p_glEnable(textype);
7279         checkGLcall("glEnable(textype)");
7280         LEAVE_GL();
7281         return WINED3D_OK;
7282     }
7283
7284     fixup = get_complex_fixup(surface->resource.format->color_fixup);
7285
7286     switch(fixup)
7287     {
7288         case COMPLEX_FIXUP_YUY2:
7289             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->yuy2_rect_shader : priv->yuy2_2d_shader;
7290             break;
7291
7292         case COMPLEX_FIXUP_UYVY:
7293             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->uyvy_rect_shader : priv->uyvy_2d_shader;
7294             break;
7295
7296         case COMPLEX_FIXUP_YV12:
7297             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->yv12_rect_shader : priv->yv12_2d_shader;
7298             break;
7299
7300         case COMPLEX_FIXUP_P8:
7301             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->p8_rect_shader : priv->p8_2d_shader;
7302             if (!shader) shader = gen_p8_shader(priv, gl_info, textype);
7303
7304             upload_palette(surface, context);
7305             break;
7306
7307         default:
7308             FIXME("Unsupported complex fixup %#x, not setting a shader\n", fixup);
7309             ENTER_GL();
7310             gl_info->gl_ops.gl.p_glEnable(textype);
7311             checkGLcall("glEnable(textype)");
7312             LEAVE_GL();
7313             return E_NOTIMPL;
7314     }
7315
7316     if (!shader) shader = gen_yuv_shader(priv, gl_info, fixup, textype);
7317
7318     ENTER_GL();
7319     gl_info->gl_ops.gl.p_glEnable(GL_FRAGMENT_PROGRAM_ARB);
7320     checkGLcall("glEnable(GL_FRAGMENT_PROGRAM_ARB)");
7321     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader));
7322     checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader)");
7323     GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, 0, size));
7324     checkGLcall("glProgramLocalParameter4fvARB");
7325     LEAVE_GL();
7326
7327     return WINED3D_OK;
7328 }
7329
7330 /* Context activation is done by the caller. */
7331 static void arbfp_blit_unset(const struct wined3d_gl_info *gl_info)
7332 {
7333     ENTER_GL();
7334     gl_info->gl_ops.gl.p_glDisable(GL_FRAGMENT_PROGRAM_ARB);
7335     checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
7336     gl_info->gl_ops.gl.p_glDisable(GL_TEXTURE_2D);
7337     checkGLcall("glDisable(GL_TEXTURE_2D)");
7338     if (gl_info->supported[ARB_TEXTURE_CUBE_MAP])
7339     {
7340         gl_info->gl_ops.gl.p_glDisable(GL_TEXTURE_CUBE_MAP_ARB);
7341         checkGLcall("glDisable(GL_TEXTURE_CUBE_MAP_ARB)");
7342     }
7343     if (gl_info->supported[ARB_TEXTURE_RECTANGLE])
7344     {
7345         gl_info->gl_ops.gl.p_glDisable(GL_TEXTURE_RECTANGLE_ARB);
7346         checkGLcall("glDisable(GL_TEXTURE_RECTANGLE_ARB)");
7347     }
7348     LEAVE_GL();
7349 }
7350
7351 static BOOL arbfp_blit_supported(const struct wined3d_gl_info *gl_info, enum wined3d_blit_op blit_op,
7352         const RECT *src_rect, DWORD src_usage, enum wined3d_pool src_pool, const struct wined3d_format *src_format,
7353         const RECT *dst_rect, DWORD dst_usage, enum wined3d_pool dst_pool, const struct wined3d_format *dst_format)
7354 {
7355     enum complex_fixup src_fixup;
7356
7357     if (!gl_info->supported[ARB_FRAGMENT_PROGRAM])
7358         return FALSE;
7359
7360     if (blit_op != WINED3D_BLIT_OP_COLOR_BLIT)
7361     {
7362         TRACE("Unsupported blit_op=%d\n", blit_op);
7363         return FALSE;
7364     }
7365
7366     if (src_pool == WINED3D_POOL_SYSTEM_MEM || dst_pool == WINED3D_POOL_SYSTEM_MEM)
7367         return FALSE;
7368
7369     src_fixup = get_complex_fixup(src_format->color_fixup);
7370     if (TRACE_ON(d3d_shader) && TRACE_ON(d3d))
7371     {
7372         TRACE("Checking support for fixup:\n");
7373         dump_color_fixup_desc(src_format->color_fixup);
7374     }
7375
7376     if (!is_identity_fixup(dst_format->color_fixup))
7377     {
7378         TRACE("Destination fixups are not supported\n");
7379         return FALSE;
7380     }
7381
7382     if (is_identity_fixup(src_format->color_fixup))
7383     {
7384         TRACE("[OK]\n");
7385         return TRUE;
7386     }
7387
7388      /* We only support YUV conversions. */
7389     if (!is_complex_fixup(src_format->color_fixup))
7390     {
7391         TRACE("[FAILED]\n");
7392         return FALSE;
7393     }
7394
7395     switch(src_fixup)
7396     {
7397         case COMPLEX_FIXUP_YUY2:
7398         case COMPLEX_FIXUP_UYVY:
7399         case COMPLEX_FIXUP_YV12:
7400         case COMPLEX_FIXUP_P8:
7401             TRACE("[OK]\n");
7402             return TRUE;
7403
7404         default:
7405             FIXME("Unsupported YUV fixup %#x\n", src_fixup);
7406             TRACE("[FAILED]\n");
7407             return FALSE;
7408     }
7409 }
7410
7411 HRESULT arbfp_blit_surface(struct wined3d_device *device, DWORD filter,
7412         struct wined3d_surface *src_surface, const RECT *src_rect_in,
7413         struct wined3d_surface *dst_surface, const RECT *dst_rect_in)
7414 {
7415     struct wined3d_context *context;
7416     RECT src_rect = *src_rect_in;
7417     RECT dst_rect = *dst_rect_in;
7418
7419     /* Now load the surface */
7420     if (wined3d_settings.offscreen_rendering_mode != ORM_FBO
7421             && (src_surface->flags & (SFLAG_INTEXTURE | SFLAG_INDRAWABLE)) == SFLAG_INDRAWABLE)
7422     {
7423         /* Without FBO blits transferring from the drawable to the texture is
7424          * expensive, because we have to flip the data in sysmem. Since we can
7425          * flip in the blitter, we don't actually need that flip anyway. So we
7426          * use the surface's texture as scratch texture, and flip the source
7427          * rectangle instead. */
7428         surface_load_fb_texture(src_surface, FALSE);
7429
7430         src_rect.top = src_surface->resource.height - src_rect.top;
7431         src_rect.bottom = src_surface->resource.height - src_rect.bottom;
7432     }
7433     else
7434         surface_internal_preload(src_surface, SRGB_RGB);
7435
7436     /* Activate the destination context, set it up for blitting */
7437     context = context_acquire(device, dst_surface);
7438     context_apply_blit_state(context, device);
7439
7440     if (!surface_is_offscreen(dst_surface))
7441         surface_translate_drawable_coords(dst_surface, context->win_handle, &dst_rect);
7442
7443     arbfp_blit_set(device->blit_priv, context, src_surface);
7444
7445     ENTER_GL();
7446
7447     /* Draw a textured quad */
7448     draw_textured_quad(src_surface, context, &src_rect, &dst_rect, filter);
7449
7450     LEAVE_GL();
7451
7452     /* Leave the opengl state valid for blitting */
7453     arbfp_blit_unset(context->gl_info);
7454
7455     if (wined3d_settings.strict_draw_ordering
7456             || (dst_surface->container.type == WINED3D_CONTAINER_SWAPCHAIN
7457             && (dst_surface->container.u.swapchain->front_buffer == dst_surface)))
7458         context->gl_info->gl_ops.gl.p_glFlush(); /* Flush to ensure ordering across contexts. */
7459
7460     context_release(context);
7461
7462     surface_modify_location(dst_surface, dst_surface->draw_binding, TRUE);
7463     return WINED3D_OK;
7464 }
7465
7466 /* Do not call while under the GL lock. */
7467 static HRESULT arbfp_blit_color_fill(struct wined3d_device *device, struct wined3d_surface *dst_surface,
7468         const RECT *dst_rect, const struct wined3d_color *color)
7469 {
7470     FIXME("Color filling not implemented by arbfp_blit\n");
7471     return WINED3DERR_INVALIDCALL;
7472 }
7473
7474 /* Do not call while under the GL lock. */
7475 static HRESULT arbfp_blit_depth_fill(struct wined3d_device *device,
7476         struct wined3d_surface *surface, const RECT *rect, float depth)
7477 {
7478     FIXME("Depth filling not implemented by arbfp_blit.\n");
7479     return WINED3DERR_INVALIDCALL;
7480 }
7481
7482 const struct blit_shader arbfp_blit = {
7483     arbfp_blit_alloc,
7484     arbfp_blit_free,
7485     arbfp_blit_set,
7486     arbfp_blit_unset,
7487     arbfp_blit_supported,
7488     arbfp_blit_color_fill,
7489     arbfp_blit_depth_fill,
7490 };