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     const struct fragment_pipeline *fragment_pipe;
 331 };
 332
 333 /* GL locking for state handlers is done by the caller. */
 334 static BOOL need_rel_addr_const(const struct arb_vshader_private *shader_data,
 335         const struct wined3d_shader_reg_maps *reg_maps, const struct wined3d_gl_info *gl_info)
 336 {
 337     if (shader_data->rel_offset) return TRUE;
 338     if (!reg_maps->usesmova) return FALSE;
 339     return !gl_info->supported[NV_VERTEX_PROGRAM2_OPTION];
 340 }
 341
 342 /* Returns TRUE if result.clip from GL_NV_vertex_program2 should be used and FALSE otherwise */
 343 static inline BOOL use_nv_clip(const struct wined3d_gl_info *gl_info)
 344 {
 345     return gl_info->supported[NV_VERTEX_PROGRAM2_OPTION]
 346             && !(gl_info->quirks & WINED3D_QUIRK_NV_CLIP_BROKEN);
 347 }
 348
 349 static BOOL need_helper_const(const struct arb_vshader_private *shader_data,
 350         const struct wined3d_shader_reg_maps *reg_maps, const struct wined3d_gl_info *gl_info)
 351 {
 352     if (need_rel_addr_const(shader_data, reg_maps, gl_info)) return TRUE;
 353     if (!gl_info->supported[NV_VERTEX_PROGRAM]) return TRUE; /* Need to init colors. */
 354     if (gl_info->quirks & WINED3D_QUIRK_ARB_VS_OFFSET_LIMIT) return TRUE; /* Load the immval offset. */
 355     if (gl_info->quirks & WINED3D_QUIRK_SET_TEXCOORD_W) return TRUE; /* Have to init texcoords. */
 356     if (!use_nv_clip(gl_info)) return TRUE; /* Init the clip texcoord */
 357     if (reg_maps->usesnrm) return TRUE; /* 0.0 */
 358     if (reg_maps->usespow) return TRUE; /* EPS, 0.0 and 1.0 */
 359     if (reg_maps->fog) return TRUE; /* Clamping fog coord, 0.0 and 1.0 */
 360     return FALSE;
 361 }
 362
 363 static unsigned int reserved_vs_const(const struct arb_vshader_private *shader_data,
 364         const struct wined3d_shader_reg_maps *reg_maps, const struct wined3d_gl_info *gl_info)
 365 {
 366     unsigned int ret = 1;
 367     /* We use one PARAM for the pos fixup, and in some cases one to load
 368      * some immediate values into the shader. */
 369     if (need_helper_const(shader_data, reg_maps, gl_info)) ++ret;
 370     if (need_rel_addr_const(shader_data, reg_maps, gl_info)) ++ret;
 371     return ret;
 372 }
 373
 374 /* Loads floating point constants into the currently set ARB_vertex/fragment_program.
 375  * When constant_list == NULL, it will load all the constants.
 376  *
 377  * @target_type should be either GL_VERTEX_PROGRAM_ARB (for vertex shaders)
 378  *  or GL_FRAGMENT_PROGRAM_ARB (for pixel shaders)
 379  */
 380 /* GL locking is done by the caller */
 381 static unsigned int shader_arb_load_constantsF(const struct wined3d_shader *shader,
 382         const struct wined3d_gl_info *gl_info, GLuint target_type, unsigned int max_constants,
 383         const float *constants, char *dirty_consts)
 384 {
 385     struct wined3d_shader_lconst *lconst;
 386     DWORD i, j;
 387     unsigned int ret;
 388
 389     if (TRACE_ON(d3d_constants))
 390     {
 391         for(i = 0; i < max_constants; i++) {
 392             if(!dirty_consts[i]) continue;
 393             TRACE_(d3d_constants)("Loading constants %i: %f, %f, %f, %f\n", i,
 394                         constants[i * 4 + 0], constants[i * 4 + 1],
 395                         constants[i * 4 + 2], constants[i * 4 + 3]);
 396         }
 397     }
 398
 399     i = 0;
 400
 401     /* In 1.X pixel shaders constants are implicitly clamped in the range [-1;1] */
 402     if (target_type == GL_FRAGMENT_PROGRAM_ARB && shader->reg_maps.shader_version.major == 1)
 403     {
 404         float lcl_const[4];
 405         /* ps 1.x supports only 8 constants, clamp only those. When switching between 1.x and higher
 406          * shaders, the first 8 constants are marked dirty for reload
 407          */
 408         for(; i < min(8, max_constants); i++) {
 409             if(!dirty_consts[i]) continue;
 410             dirty_consts[i] = 0;
 411
 412             j = 4 * i;
 413             if (constants[j + 0] > 1.0f) lcl_const[0] = 1.0f;
 414             else if (constants[j + 0] < -1.0f) lcl_const[0] = -1.0f;
 415             else lcl_const[0] = constants[j + 0];
 416
 417             if (constants[j + 1] > 1.0f) lcl_const[1] = 1.0f;
 418             else if (constants[j + 1] < -1.0f) lcl_const[1] = -1.0f;
 419             else lcl_const[1] = constants[j + 1];
 420
 421             if (constants[j + 2] > 1.0f) lcl_const[2] = 1.0f;
 422             else if (constants[j + 2] < -1.0f) lcl_const[2] = -1.0f;
 423             else lcl_const[2] = constants[j + 2];
 424
 425             if (constants[j + 3] > 1.0f) lcl_const[3] = 1.0f;
 426             else if (constants[j + 3] < -1.0f) lcl_const[3] = -1.0f;
 427             else lcl_const[3] = constants[j + 3];
 428
 429             GL_EXTCALL(glProgramEnvParameter4fvARB(target_type, i, lcl_const));
 430         }
 431
 432         /* If further constants are dirty, reload them without clamping.
 433          *
 434          * The alternative is not to touch them, but then we cannot reset the dirty constant count
 435          * to zero. That's bad for apps that only use PS 1.x shaders, because in that case the code
 436          * above would always re-check the first 8 constants since max_constant remains at the init
 437          * value
 438          */
 439     }
 440
 441     if (gl_info->supported[EXT_GPU_PROGRAM_PARAMETERS])
 442     {
 443         /* TODO: Benchmark if we're better of with finding the dirty constants ourselves,
 444          * or just reloading *all* constants at once
 445          *
 446         GL_EXTCALL(glProgramEnvParameters4fvEXT(target_type, i, max_constants, constants + (i * 4)));
 447          */
 448         for(; i < max_constants; i++) {
 449             if(!dirty_consts[i]) continue;
 450
 451             /* Find the next block of dirty constants */
 452             dirty_consts[i] = 0;
 453             j = i;
 454             for(i++; (i < max_constants) && dirty_consts[i]; i++) {
 455                 dirty_consts[i] = 0;
 456             }
 457
 458             GL_EXTCALL(glProgramEnvParameters4fvEXT(target_type, j, i - j, constants + (j * 4)));
 459         }
 460     } else {
 461         for(; i < max_constants; i++) {
 462             if(dirty_consts[i]) {
 463                 dirty_consts[i] = 0;
 464                 GL_EXTCALL(glProgramEnvParameter4fvARB(target_type, i, constants + (i * 4)));
 465             }
 466         }
 467     }
 468     checkGLcall("glProgramEnvParameter4fvARB()");
 469
 470     /* Load immediate constants */
 471     if (shader->load_local_constsF)
 472     {
 473         if (TRACE_ON(d3d_shader))
 474         {
 475             LIST_FOR_EACH_ENTRY(lconst, &shader->constantsF, struct wined3d_shader_lconst, entry)
 476             {
 477                 GLfloat* values = (GLfloat*)lconst->value;
 478                 TRACE_(d3d_constants)("Loading local constants %i: %f, %f, %f, %f\n", lconst->idx,
 479                         values[0], values[1], values[2], values[3]);
 480             }
 481         }
 482         /* Immediate constants are clamped for 1.X shaders at loading times */
 483         ret = 0;
 484         LIST_FOR_EACH_ENTRY(lconst, &shader->constantsF, struct wined3d_shader_lconst, entry)
 485         {
 486             dirty_consts[lconst->idx] = 1; /* Dirtify so the non-immediate constant overwrites it next time */
 487             ret = max(ret, lconst->idx + 1);
 488             GL_EXTCALL(glProgramEnvParameter4fvARB(target_type, lconst->idx, (GLfloat*)lconst->value));
 489         }
 490         checkGLcall("glProgramEnvParameter4fvARB()");
 491         return ret; /* The loaded immediate constants need reloading for the next shader */
 492     } else {
 493         return 0; /* No constants are dirty now */
 494     }
 495 }
 496
 497 /**
 498  * Loads the texture dimensions for NP2 fixup into the currently set ARB_[vertex/fragment]_programs.
 499  */
 500 static void shader_arb_load_np2fixup_constants(void *shader_priv,
 501         const struct wined3d_gl_info *gl_info, const struct wined3d_state *state)
 502 {
 503     const struct shader_arb_priv * priv = shader_priv;
 504
 505     /* NP2 texcoord fixup is (currently) only done for pixelshaders. */
 506     if (!use_ps(state)) return;
 507
 508     if (priv->compiled_fprog && priv->compiled_fprog->np2fixup_info.super.active) {
 509         const struct arb_ps_np2fixup_info* const fixup = &priv->compiled_fprog->np2fixup_info;
 510         UINT i;
 511         WORD active = fixup->super.active;
 512         GLfloat np2fixup_constants[4 * MAX_FRAGMENT_SAMPLERS];
 513
 514         for (i = 0; active; active >>= 1, ++i)
 515         {
 516             const struct wined3d_texture *tex = state->textures[i];
 517             const unsigned char idx = fixup->super.idx[i];
 518             GLfloat *tex_dim = &np2fixup_constants[(idx >> 1) * 4];
 519
 520             if (!(active & 1)) continue;
 521
 522             if (!tex) {
 523                 FIXME("Nonexistent texture is flagged for NP2 texcoord fixup\n");
 524                 continue;
 525             }
 526
 527             if (idx % 2)
 528             {
 529                 tex_dim[2] = tex->pow2_matrix[0];
 530                 tex_dim[3] = tex->pow2_matrix[5];
 531             }
 532             else
 533             {
 534                 tex_dim[0] = tex->pow2_matrix[0];
 535                 tex_dim[1] = tex->pow2_matrix[5];
 536             }
 537         }
 538
 539         for (i = 0; i < fixup->super.num_consts; ++i) {
 540             GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB,
 541                                                    fixup->offset + i, &np2fixup_constants[i * 4]));
 542         }
 543     }
 544 }
 545
 546 /* GL locking is done by the caller. */
 547 static void shader_arb_ps_local_constants(const struct arb_ps_compiled_shader *gl_shader,
 548         const struct wined3d_context *context, const struct wined3d_state *state, UINT rt_height)
 549 {
 550     const struct wined3d_gl_info *gl_info = context->gl_info;
 551     unsigned char i;
 552
 553     for(i = 0; i < gl_shader->numbumpenvmatconsts; i++)
 554     {
 555         int texunit = gl_shader->bumpenvmatconst[i].texunit;
 556
 557         /* The state manager takes care that this function is always called if the bump env matrix changes */
 558         const float *data = (const float *)&state->texture_states[texunit][WINED3D_TSS_BUMPENV_MAT00];
 559         GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB,
 560                 gl_shader->bumpenvmatconst[i].const_num, data));
 561
 562         if (gl_shader->luminanceconst[i].const_num != WINED3D_CONST_NUM_UNUSED)
 563         {
 564             /* WINED3D_TSS_BUMPENVLSCALE and WINED3D_TSS_BUMPENVLOFFSET are next to each other.
 565              * point gl to the scale, and load 4 floats. x = scale, y = offset, z and w are junk, we
 566              * don't care about them. The pointers are valid for sure because the stateblock is bigger.
 567              * (they're WINED3D_TSS_TEXTURETRANSFORMFLAGS and WINED3D_TSS_ADDRESSW, so most likely 0 or NaN
 568             */
 569             const float *scale = (const float *)&state->texture_states[texunit][WINED3D_TSS_BUMPENV_LSCALE];
 570             GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB,
 571                     gl_shader->luminanceconst[i].const_num, scale));
 572         }
 573     }
 574     checkGLcall("Load bumpmap consts");
 575
 576     if(gl_shader->ycorrection != WINED3D_CONST_NUM_UNUSED)
 577     {
 578         /* ycorrection.x: Backbuffer height(onscreen) or 0(offscreen).
 579         * ycorrection.y: -1.0(onscreen), 1.0(offscreen)
 580         * ycorrection.z: 1.0
 581         * ycorrection.w: 0.0
 582         */
 583         float val[4];
 584         val[0] = context->render_offscreen ? 0.0f : (float) rt_height;
 585         val[1] = context->render_offscreen ? 1.0f : -1.0f;
 586         val[2] = 1.0f;
 587         val[3] = 0.0f;
 588         GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, gl_shader->ycorrection, val));
 589         checkGLcall("y correction loading");
 590     }
 591
 592     if (!gl_shader->num_int_consts) return;
 593
 594     for(i = 0; i < MAX_CONST_I; i++)
 595     {
 596         if(gl_shader->int_consts[i] != WINED3D_CONST_NUM_UNUSED)
 597         {
 598             float val[4];
 599             val[0] = (float)state->ps_consts_i[4 * i];
 600             val[1] = (float)state->ps_consts_i[4 * i + 1];
 601             val[2] = (float)state->ps_consts_i[4 * i + 2];
 602             val[3] = -1.0f;
 603
 604             GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, gl_shader->int_consts[i], val));
 605         }
 606     }
 607     checkGLcall("Load ps int consts");
 608 }
 609
 610 /* GL locking is done by the caller. */
 611 static void shader_arb_vs_local_constants(const struct arb_vs_compiled_shader *gl_shader,
 612         const struct wined3d_context *context, const struct wined3d_state *state)
 613 {
 614     const struct wined3d_gl_info *gl_info = context->gl_info;
 615     float position_fixup[4];
 616     unsigned char i;
 617
 618     /* Upload the position fixup */
 619     shader_get_position_fixup(context, state, position_fixup);
 620     GL_EXTCALL(glProgramLocalParameter4fvARB(GL_VERTEX_PROGRAM_ARB, gl_shader->pos_fixup, position_fixup));
 621
 622     if (!gl_shader->num_int_consts) return;
 623
 624     for(i = 0; i < MAX_CONST_I; i++)
 625     {
 626         if(gl_shader->int_consts[i] != WINED3D_CONST_NUM_UNUSED)
 627         {
 628             float val[4];
 629             val[0] = (float)state->vs_consts_i[4 * i];
 630             val[1] = (float)state->vs_consts_i[4 * i + 1];
 631             val[2] = (float)state->vs_consts_i[4 * i + 2];
 632             val[3] = -1.0f;
 633
 634             GL_EXTCALL(glProgramLocalParameter4fvARB(GL_VERTEX_PROGRAM_ARB, gl_shader->int_consts[i], val));
 635         }
 636     }
 637     checkGLcall("Load vs int consts");
 638 }
 639
 640 static void shader_arb_select(const struct wined3d_context *context, enum wined3d_shader_mode vertex_mode,
 641         enum wined3d_shader_mode fragment_mode);
 642
 643 /**
 644  * Loads the app-supplied constants into the currently set ARB_[vertex/fragment]_programs.
 645  *
 646  * We only support float constants in ARB at the moment, so don't
 647  * worry about the Integers or Booleans
 648  */
 649 /* GL locking is done by the caller (state handler) */
 650 static void shader_arb_load_constants_internal(const struct wined3d_context *context,
 651         BOOL usePixelShader, BOOL useVertexShader, BOOL from_shader_select)
 652 {
 653     struct wined3d_device *device = context->swapchain->device;
 654     const struct wined3d_stateblock *stateblock = device->stateBlock;
 655     const struct wined3d_state *state = &stateblock->state;
 656     const struct wined3d_gl_info *gl_info = context->gl_info;
 657     struct shader_arb_priv *priv = device->shader_priv;
 658
 659     if (!from_shader_select)
 660     {
 661         const struct wined3d_shader *vshader = state->vertex_shader, *pshader = state->pixel_shader;
 662         if (vshader
 663                 && (stateblock->changed.vertexShaderConstantsB & vshader->reg_maps.boolean_constants
 664                 || (!gl_info->supported[NV_VERTEX_PROGRAM2_OPTION]
 665                 && (stateblock->changed.vertexShaderConstantsI
 666                 & vshader->reg_maps.integer_constants & ~vshader->reg_maps.local_int_consts))))
 667         {
 668             TRACE("bool/integer vertex shader constants potentially modified, forcing shader reselection.\n");
 669             shader_arb_select(context,
 670                     useVertexShader ? WINED3D_SHADER_MODE_SHADER : WINED3D_SHADER_MODE_FFP,
 671                     usePixelShader ? WINED3D_SHADER_MODE_SHADER : WINED3D_SHADER_MODE_FFP);
 672         }
 673         else if (pshader
 674                 && (stateblock->changed.pixelShaderConstantsB & pshader->reg_maps.boolean_constants
 675                 || (!gl_info->supported[NV_FRAGMENT_PROGRAM_OPTION]
 676                 && (stateblock->changed.pixelShaderConstantsI
 677                 & pshader->reg_maps.integer_constants & ~pshader->reg_maps.local_int_consts))))
 678         {
 679             TRACE("bool/integer pixel shader constants potentially modified, forcing shader reselection.\n");
 680             shader_arb_select(context,
 681                     useVertexShader ? WINED3D_SHADER_MODE_SHADER : WINED3D_SHADER_MODE_FFP,
 682                     usePixelShader ? WINED3D_SHADER_MODE_SHADER : WINED3D_SHADER_MODE_FFP);
 683         }
 684     }
 685
 686     if (context != priv->last_context)
 687     {
 688         memset(priv->vshader_const_dirty, 1,
 689                 sizeof(*priv->vshader_const_dirty) * device->d3d_vshader_constantF);
 690         priv->highest_dirty_vs_const = device->d3d_vshader_constantF;
 691
 692         memset(priv->pshader_const_dirty, 1,
 693                 sizeof(*priv->pshader_const_dirty) * device->d3d_pshader_constantF);
 694         priv->highest_dirty_ps_const = device->d3d_pshader_constantF;
 695
 696         priv->last_context = context;
 697     }
 698
 699     if (useVertexShader)
 700     {
 701         struct wined3d_shader *vshader = state->vertex_shader;
 702         const struct arb_vs_compiled_shader *gl_shader = priv->compiled_vprog;
 703
 704         /* Load DirectX 9 float constants for vertex shader */
 705         priv->highest_dirty_vs_const = shader_arb_load_constantsF(vshader, gl_info, GL_VERTEX_PROGRAM_ARB,
 706                 priv->highest_dirty_vs_const, state->vs_consts_f, priv->vshader_const_dirty);
 707         shader_arb_vs_local_constants(gl_shader, context, state);
 708     }
 709
 710     if (usePixelShader)
 711     {
 712         struct wined3d_shader *pshader = state->pixel_shader;
 713         const struct arb_ps_compiled_shader *gl_shader = priv->compiled_fprog;
 714         UINT rt_height = state->fb->render_targets[0]->resource.height;
 715
 716         /* Load DirectX 9 float constants for pixel shader */
 717         priv->highest_dirty_ps_const = shader_arb_load_constantsF(pshader, gl_info, GL_FRAGMENT_PROGRAM_ARB,
 718                 priv->highest_dirty_ps_const, state->ps_consts_f, priv->pshader_const_dirty);
 719         shader_arb_ps_local_constants(gl_shader, context, state, rt_height);
 720     }
 721 }
 722
 723 static void shader_arb_load_constants(const struct wined3d_context *context, BOOL ps, BOOL vs)
 724 {
 725     shader_arb_load_constants_internal(context, ps, vs, FALSE);
 726 }
 727
 728 static void shader_arb_update_float_vertex_constants(struct wined3d_device *device, UINT start, UINT count)
 729 {
 730     struct wined3d_context *context = context_get_current();
 731     struct shader_arb_priv *priv = device->shader_priv;
 732
 733     /* We don't want shader constant dirtification to be an O(contexts), so just dirtify the active
 734      * context. On a context switch the old context will be fully dirtified */
 735     if (!context || context->swapchain->device != device) return;
 736
 737     memset(priv->vshader_const_dirty + start, 1, sizeof(*priv->vshader_const_dirty) * count);
 738     priv->highest_dirty_vs_const = max(priv->highest_dirty_vs_const, start + count);
 739 }
 740
 741 static void shader_arb_update_float_pixel_constants(struct wined3d_device *device, UINT start, UINT count)
 742 {
 743     struct wined3d_context *context = context_get_current();
 744     struct shader_arb_priv *priv = device->shader_priv;
 745
 746     /* We don't want shader constant dirtification to be an O(contexts), so just dirtify the active
 747      * context. On a context switch the old context will be fully dirtified */
 748     if (!context || context->swapchain->device != device) return;
 749
 750     memset(priv->pshader_const_dirty + start, 1, sizeof(*priv->pshader_const_dirty) * count);
 751     priv->highest_dirty_ps_const = max(priv->highest_dirty_ps_const, start + count);
 752 }
 753
 754 static DWORD *local_const_mapping(const struct wined3d_shader *shader)
 755 {
 756     const struct wined3d_shader_lconst *lconst;
 757     DWORD *ret;
 758     DWORD idx = 0;
 759
 760     if (shader->load_local_constsF || list_empty(&shader->constantsF))
 761         return NULL;
 762
 763     ret = HeapAlloc(GetProcessHeap(), 0, sizeof(DWORD) * shader->limits.constant_float);
 764     if (!ret)
 765     {
 766         ERR("Out of memory\n");
 767         return NULL;
 768     }
 769
 770     LIST_FOR_EACH_ENTRY(lconst, &shader->constantsF, struct wined3d_shader_lconst, entry)
 771     {
 772         ret[lconst->idx] = idx++;
 773     }
 774     return ret;
 775 }
 776
 777 /* Generate the variable & register declarations for the ARB_vertex_program output target */
 778 static DWORD shader_generate_arb_declarations(const struct wined3d_shader *shader,
 779         const struct wined3d_shader_reg_maps *reg_maps, struct wined3d_shader_buffer *buffer,
 780         const struct wined3d_gl_info *gl_info, const DWORD *lconst_map,
 781         DWORD *num_clipplanes, const struct shader_arb_ctx_priv *ctx)
 782 {
 783     DWORD i, next_local = 0;
 784     char pshader = shader_is_pshader_version(reg_maps->shader_version.type);
 785     const struct wined3d_shader_lconst *lconst;
 786     unsigned max_constantsF;
 787     DWORD map;
 788
 789     /* In pixel shaders, all private constants are program local, we don't need anything
 790      * from program.env. Thus we can advertise the full set of constants in pixel shaders.
 791      * If we need a private constant the GL implementation will squeeze it in somewhere
 792      *
 793      * With vertex shaders we need the posFixup and on some GL implementations 4 helper
 794      * immediate values. The posFixup is loaded using program.env for now, so always
 795      * subtract one from the number of constants. If the shader uses indirect addressing,
 796      * account for the helper const too because we have to declare all available d3d constants
 797      * and don't know which are actually used.
 798      */
 799     if (pshader)
 800     {
 801         max_constantsF = gl_info->limits.arb_ps_native_constants;
 802         /* 24 is the minimum MAX_PROGRAM_ENV_PARAMETERS_ARB value. */
 803         if (max_constantsF < 24)
 804             max_constantsF = gl_info->limits.arb_ps_float_constants;
 805     }
 806     else
 807     {
 808         const struct arb_vshader_private *shader_data = shader->backend_data;
 809         max_constantsF = gl_info->limits.arb_vs_native_constants;
 810         /* 96 is the minimum MAX_PROGRAM_ENV_PARAMETERS_ARB value.
 811          * Also prevents max_constantsF from becoming less than 0 and
 812          * wrapping . */
 813         if (max_constantsF < 96)
 814             max_constantsF = gl_info->limits.arb_vs_float_constants;
 815
 816         if (reg_maps->usesrelconstF)
 817         {
 818             DWORD highest_constf = 0, clip_limit;
 819
 820             max_constantsF -= reserved_vs_const(shader_data, reg_maps, gl_info);
 821             max_constantsF -= count_bits(reg_maps->integer_constants);
 822
 823             for (i = 0; i < shader->limits.constant_float; ++i)
 824             {
 825                 DWORD idx = i >> 5;
 826                 DWORD shift = i & 0x1f;
 827                 if(reg_maps->constf[idx] & (1 << shift)) highest_constf = i;
 828             }
 829
 830             if(use_nv_clip(gl_info) && ctx->target_version >= NV2)
 831             {
 832                 if(ctx->cur_vs_args->super.clip_enabled)
 833                     clip_limit = gl_info->limits.clipplanes;
 834                 else
 835                     clip_limit = 0;
 836             }
 837             else
 838             {
 839                 unsigned int mask = ctx->cur_vs_args->clip.boolclip.clipplane_mask;
 840                 clip_limit = min(count_bits(mask), 4);
 841             }
 842             *num_clipplanes = min(clip_limit, max_constantsF - highest_constf - 1);
 843             max_constantsF -= *num_clipplanes;
 844             if(*num_clipplanes < clip_limit)
 845             {
 846                 WARN("Only %u clipplanes out of %u enabled\n", *num_clipplanes, gl_info->limits.clipplanes);
 847             }
 848         }
 849         else
 850         {
 851             if (ctx->target_version >= NV2) *num_clipplanes = gl_info->limits.clipplanes;
 852             else *num_clipplanes = min(gl_info->limits.clipplanes, 4);
 853         }
 854     }
 855
 856     for (i = 0, map = reg_maps->temporary; map; map >>= 1, ++i)
 857     {
 858         if (map & 1) shader_addline(buffer, "TEMP R%u;\n", i);
 859     }
 860
 861     for (i = 0, map = reg_maps->address; map; map >>= 1, ++i)
 862     {
 863         if (map & 1) shader_addline(buffer, "ADDRESS A%u;\n", i);
 864     }
 865
 866     if (pshader && reg_maps->shader_version.major == 1 && reg_maps->shader_version.minor <= 3)
 867     {
 868         for (i = 0, map = reg_maps->texcoord; map; map >>= 1, ++i)
 869         {
 870             if (map & 1) shader_addline(buffer, "TEMP T%u;\n", i);
 871         }
 872     }
 873
 874     /* Load local constants using the program-local space,
 875      * this avoids reloading them each time the shader is used
 876      */
 877     if (lconst_map)
 878     {
 879         LIST_FOR_EACH_ENTRY(lconst, &shader->constantsF, struct wined3d_shader_lconst, entry)
 880         {
 881             shader_addline(buffer, "PARAM C%u = program.local[%u];\n", lconst->idx,
 882                            lconst_map[lconst->idx]);
 883             next_local = max(next_local, lconst_map[lconst->idx] + 1);
 884         }
 885     }
 886
 887     /* After subtracting privately used constants from the hardware limit(they are loaded as
 888      * local constants), make sure the shader doesn't violate the env constant limit
 889      */
 890     if(pshader)
 891     {
 892         max_constantsF = min(max_constantsF, gl_info->limits.arb_ps_float_constants);
 893     }
 894     else
 895     {
 896         max_constantsF = min(max_constantsF, gl_info->limits.arb_vs_float_constants);
 897     }
 898
 899     /* Avoid declaring more constants than needed */
 900     max_constantsF = min(max_constantsF, shader->limits.constant_float);
 901
 902     /* we use the array-based constants array if the local constants are marked for loading,
 903      * because then we use indirect addressing, or when the local constant list is empty,
 904      * because then we don't know if we're using indirect addressing or not. If we're hardcoding
 905      * local constants do not declare the loaded constants as an array because ARB compilers usually
 906      * do not optimize unused constants away
 907      */
 908     if (reg_maps->usesrelconstF)
 909     {
 910         /* Need to PARAM the environment parameters (constants) so we can use relative addressing */
 911         shader_addline(buffer, "PARAM C[%d] = { program.env[0..%d] };\n",
 912                     max_constantsF, max_constantsF - 1);
 913     } else {
 914         for(i = 0; i < max_constantsF; i++) {
 915             DWORD idx, mask;
 916             idx = i >> 5;
 917             mask = 1 << (i & 0x1f);
 918             if (!shader_constant_is_local(shader, i) && (reg_maps->constf[idx] & mask))
 919             {
 920                 shader_addline(buffer, "PARAM C%d = program.env[%d];\n",i, i);
 921             }
 922         }
 923     }
 924
 925     return next_local;
 926 }
 927
 928 static const char * const shift_tab[] = {
 929     "dummy",     /*  0 (none) */
 930     "coefmul.x", /*  1 (x2)   */
 931     "coefmul.y", /*  2 (x4)   */
 932     "coefmul.z", /*  3 (x8)   */
 933     "coefmul.w", /*  4 (x16)  */
 934     "dummy",     /*  5 (x32)  */
 935     "dummy",     /*  6 (x64)  */
 936     "dummy",     /*  7 (x128) */
 937     "dummy",     /*  8 (d256) */
 938     "dummy",     /*  9 (d128) */
 939     "dummy",     /* 10 (d64)  */
 940     "dummy",     /* 11 (d32)  */
 941     "coefdiv.w", /* 12 (d16)  */
 942     "coefdiv.z", /* 13 (d8)   */
 943     "coefdiv.y", /* 14 (d4)   */
 944     "coefdiv.x"  /* 15 (d2)   */
 945 };
 946
 947 static void shader_arb_get_write_mask(const struct wined3d_shader_instruction *ins,
 948         const struct wined3d_shader_dst_param *dst, char *write_mask)
 949 {
 950     char *ptr = write_mask;
 951
 952     if (dst->write_mask != WINED3DSP_WRITEMASK_ALL)
 953     {
 954         *ptr++ = '.';
 955         if (dst->write_mask & WINED3DSP_WRITEMASK_0) *ptr++ = 'x';
 956         if (dst->write_mask & WINED3DSP_WRITEMASK_1) *ptr++ = 'y';
 957         if (dst->write_mask & WINED3DSP_WRITEMASK_2) *ptr++ = 'z';
 958         if (dst->write_mask & WINED3DSP_WRITEMASK_3) *ptr++ = 'w';
 959     }
 960
 961     *ptr = '\0';
 962 }
 963
 964 static void shader_arb_get_swizzle(const struct wined3d_shader_src_param *param, BOOL fixup, char *swizzle_str)
 965 {
 966     /* For registers of type WINED3DDECLTYPE_D3DCOLOR, data is stored as "bgra",
 967      * but addressed as "rgba". To fix this we need to swap the register's x
 968      * and z components. */
 969     const char *swizzle_chars = fixup ? "zyxw" : "xyzw";
 970     char *ptr = swizzle_str;
 971
 972     /* swizzle bits fields: wwzzyyxx */
 973     DWORD swizzle = param->swizzle;
 974     DWORD swizzle_x = swizzle & 0x03;
 975     DWORD swizzle_y = (swizzle >> 2) & 0x03;
 976     DWORD swizzle_z = (swizzle >> 4) & 0x03;
 977     DWORD swizzle_w = (swizzle >> 6) & 0x03;
 978
 979     /* If the swizzle is the default swizzle (ie, "xyzw"), we don't need to
 980      * generate a swizzle string. Unless we need to our own swizzling. */
 981     if (swizzle != WINED3DSP_NOSWIZZLE || fixup)
 982     {
 983         *ptr++ = '.';
 984         if (swizzle_x == swizzle_y && swizzle_x == swizzle_z && swizzle_x == swizzle_w) {
 985             *ptr++ = swizzle_chars[swizzle_x];
 986         } else {
 987             *ptr++ = swizzle_chars[swizzle_x];
 988             *ptr++ = swizzle_chars[swizzle_y];
 989             *ptr++ = swizzle_chars[swizzle_z];
 990             *ptr++ = swizzle_chars[swizzle_w];
 991         }
 992     }
 993
 994     *ptr = '\0';
 995 }
 996
 997 static void shader_arb_request_a0(const struct wined3d_shader_instruction *ins, const char *src)
 998 {
 999     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
1000     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1001
1002     if (!strcmp(priv->addr_reg, src)) return;
1003
1004     strcpy(priv->addr_reg, src);
1005     shader_addline(buffer, "ARL A0.x, %s;\n", src);
1006 }
1007
1008 static void shader_arb_get_src_param(const struct wined3d_shader_instruction *ins,
1009         const struct wined3d_shader_src_param *src, unsigned int tmpreg, char *outregstr);
1010
1011 static void shader_arb_get_register_name(const struct wined3d_shader_instruction *ins,
1012         const struct wined3d_shader_register *reg, char *register_name, BOOL *is_color)
1013 {
1014     /* oPos, oFog and oPts in D3D */
1015     static const char * const rastout_reg_names[] = {"TMP_OUT", "TMP_FOGCOORD", "result.pointsize"};
1016     const struct wined3d_shader *shader = ins->ctx->shader;
1017     const struct wined3d_shader_reg_maps *reg_maps = ins->ctx->reg_maps;
1018     BOOL pshader = shader_is_pshader_version(reg_maps->shader_version.type);
1019     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
1020
1021     *is_color = FALSE;
1022
1023     switch (reg->type)
1024     {
1025         case WINED3DSPR_TEMP:
1026             sprintf(register_name, "R%u", reg->idx[0].offset);
1027             break;
1028
1029         case WINED3DSPR_INPUT:
1030             if (pshader)
1031             {
1032                 if (reg_maps->shader_version.major < 3)
1033                 {
1034                     if (!reg->idx[0].offset)
1035                         strcpy(register_name, "fragment.color.primary");
1036                     else
1037                         strcpy(register_name, "fragment.color.secondary");
1038                 }
1039                 else
1040                 {
1041                     if (reg->idx[0].rel_addr)
1042                     {
1043                         char rel_reg[50];
1044                         shader_arb_get_src_param(ins, reg->idx[0].rel_addr, 0, rel_reg);
1045
1046                         if (!strcmp(rel_reg, "**aL_emul**"))
1047                         {
1048                             DWORD idx = ctx->aL + reg->idx[0].offset;
1049                             if(idx < MAX_REG_INPUT)
1050                             {
1051                                 strcpy(register_name, ctx->ps_input[idx]);
1052                             }
1053                             else
1054                             {
1055                                 ERR("Pixel shader input register out of bounds: %u\n", idx);
1056                                 sprintf(register_name, "out_of_bounds_%u", idx);
1057                             }
1058                         }
1059                         else if (reg_maps->input_registers & 0x0300)
1060                         {
1061                             /* There are two ways basically:
1062                              *
1063                              * 1) Use the unrolling code that is used for loop emulation and unroll the loop.
1064                              *    That means trouble if the loop also contains a breakc or if the control values
1065                              *    aren't local constants.
1066                              * 2) Generate an if block that checks if aL.y < 8, == 8 or == 9 and selects the
1067                              *    source dynamically. The trouble is that we cannot simply read aL.y because it
1068                              *    is an ADDRESS register. We could however push it, load .zw with a value and use
1069                              *    ADAC to load the condition code register and pop it again afterwards
1070                              */
1071                             FIXME("Relative input register addressing with more than 8 registers\n");
1072
1073                             /* This is better than nothing for now */
1074                             sprintf(register_name, "fragment.texcoord[%s + %u]", rel_reg, reg->idx[0].offset);
1075                         }
1076                         else if(ctx->cur_ps_args->super.vp_mode != vertexshader)
1077                         {
1078                             /* This is problematic because we'd have to consult the ctx->ps_input strings
1079                              * for where to find the varying. Some may be "0.0", others can be texcoords or
1080                              * colors. This needs either a pipeline replacement to make the vertex shader feed
1081                              * proper varyings, or loop unrolling
1082                              *
1083                              * For now use the texcoords and hope for the best
1084                              */
1085                             FIXME("Non-vertex shader varying input with indirect addressing\n");
1086                             sprintf(register_name, "fragment.texcoord[%s + %u]", rel_reg, reg->idx[0].offset);
1087                         }
1088                         else
1089                         {
1090                             /* D3D supports indirect addressing only with aL in loop registers. The loop instruction
1091                              * pulls GL_NV_fragment_program2 in
1092                              */
1093                             sprintf(register_name, "fragment.texcoord[%s + %u]", rel_reg, reg->idx[0].offset);
1094                         }
1095                     }
1096                     else
1097                     {
1098                         if (reg->idx[0].offset < MAX_REG_INPUT)
1099                         {
1100                             strcpy(register_name, ctx->ps_input[reg->idx[0].offset]);
1101                         }
1102                         else
1103                         {
1104                             ERR("Pixel shader input register out of bounds: %u\n", reg->idx[0].offset);
1105                             sprintf(register_name, "out_of_bounds_%u", reg->idx[0].offset);
1106                         }
1107                     }
1108                 }
1109             }
1110             else
1111             {
1112                 if (ctx->cur_vs_args->super.swizzle_map & (1 << reg->idx[0].offset))
1113                     *is_color = TRUE;
1114                 sprintf(register_name, "vertex.attrib[%u]", reg->idx[0].offset);
1115             }
1116             break;
1117
1118         case WINED3DSPR_CONST:
1119             if (!pshader && reg->idx[0].rel_addr)
1120             {
1121                 const struct arb_vshader_private *shader_data = shader->backend_data;
1122                 UINT rel_offset = shader_data->rel_offset;
1123                 BOOL aL = FALSE;
1124                 char rel_reg[50];
1125                 if (reg_maps->shader_version.major < 2)
1126                 {
1127                     sprintf(rel_reg, "A0.x");
1128                 }
1129                 else
1130                 {
1131                     shader_arb_get_src_param(ins, reg->idx[0].rel_addr, 0, rel_reg);
1132                     if (ctx->target_version == ARB)
1133                     {
1134                         if (!strcmp(rel_reg, "**aL_emul**"))
1135                         {
1136                             aL = TRUE;
1137                         } else {
1138                             shader_arb_request_a0(ins, rel_reg);
1139                             sprintf(rel_reg, "A0.x");
1140                         }
1141                     }
1142                 }
1143                 if (aL)
1144                     sprintf(register_name, "C[%u]", ctx->aL + reg->idx[0].offset);
1145                 else if (reg->idx[0].offset >= rel_offset)
1146                     sprintf(register_name, "C[%s + %u]", rel_reg, reg->idx[0].offset - rel_offset);
1147                 else
1148                     sprintf(register_name, "C[%s - %u]", rel_reg, rel_offset - reg->idx[0].offset);
1149             }
1150             else
1151             {
1152                 if (reg_maps->usesrelconstF)
1153                     sprintf(register_name, "C[%u]", reg->idx[0].offset);
1154                 else
1155                     sprintf(register_name, "C%u", reg->idx[0].offset);
1156             }
1157             break;
1158
1159         case WINED3DSPR_TEXTURE: /* case WINED3DSPR_ADDR: */
1160             if (pshader)
1161             {
1162                 if (reg_maps->shader_version.major == 1
1163                         && reg_maps->shader_version.minor <= 3)
1164                     /* In ps <= 1.3, Tx is a temporary register as destination
1165                      * to all instructions, and as source to most instructions.
1166                      * For some instructions it is the texcoord input. Those
1167                      * instructions know about the special use. */
1168                     sprintf(register_name, "T%u", reg->idx[0].offset);
1169                 else
1170                     /* In ps 1.4 and 2.x Tx is always a (read-only) varying. */
1171                     sprintf(register_name, "fragment.texcoord[%u]", reg->idx[0].offset);
1172             }
1173             else
1174             {
1175                 if (reg_maps->shader_version.major == 1 || ctx->target_version >= NV2)
1176                     sprintf(register_name, "A%u", reg->idx[0].offset);
1177                 else
1178                     sprintf(register_name, "A%u_SHADOW", reg->idx[0].offset);
1179             }
1180             break;
1181
1182         case WINED3DSPR_COLOROUT:
1183             if (ctx->cur_ps_args->super.srgb_correction && !reg->idx[0].offset)
1184             {
1185                 strcpy(register_name, "TMP_COLOR");
1186             }
1187             else
1188             {
1189                 if (ctx->cur_ps_args->super.srgb_correction)
1190                     FIXME("sRGB correction on higher render targets.\n");
1191                 if (reg_maps->rt_mask > 1)
1192                     sprintf(register_name, "result.color[%u]", reg->idx[0].offset);
1193                 else
1194                     strcpy(register_name, "result.color");
1195             }
1196             break;
1197
1198         case WINED3DSPR_RASTOUT:
1199             if (reg->idx[0].offset == 1)
1200                 sprintf(register_name, "%s", ctx->fog_output);
1201             else
1202                 sprintf(register_name, "%s", rastout_reg_names[reg->idx[0].offset]);
1203             break;
1204
1205         case WINED3DSPR_DEPTHOUT:
1206             strcpy(register_name, "result.depth");
1207             break;
1208
1209         case WINED3DSPR_ATTROUT:
1210         /* case WINED3DSPR_OUTPUT: */
1211             if (pshader)
1212                 sprintf(register_name, "oD[%u]", reg->idx[0].offset);
1213             else
1214                 strcpy(register_name, ctx->color_output[reg->idx[0].offset]);
1215             break;
1216
1217         case WINED3DSPR_TEXCRDOUT:
1218             if (pshader)
1219                 sprintf(register_name, "oT[%u]", reg->idx[0].offset);
1220             else if (reg_maps->shader_version.major < 3)
1221                 strcpy(register_name, ctx->texcrd_output[reg->idx[0].offset]);
1222             else
1223                 strcpy(register_name, ctx->vs_output[reg->idx[0].offset]);
1224             break;
1225
1226         case WINED3DSPR_LOOP:
1227             if(ctx->target_version >= NV2)
1228             {
1229                 /* Pshader has an implicitly declared loop index counter A0.x that cannot be renamed */
1230                 if(pshader) sprintf(register_name, "A0.x");
1231                 else sprintf(register_name, "aL.y");
1232             }
1233             else
1234             {
1235                 /* Unfortunately this code cannot return the value of ctx->aL here. An immediate value
1236                  * would be valid, but if aL is used for indexing(its only use), there's likely an offset,
1237                  * thus the result would be something like C[15 + 30], which is not valid in the ARB program
1238                  * grammar. So return a marker for the emulated aL and intercept it in constant and varying
1239                  * indexing
1240                  */
1241                 sprintf(register_name, "**aL_emul**");
1242             }
1243
1244             break;
1245
1246         case WINED3DSPR_CONSTINT:
1247             sprintf(register_name, "I%u", reg->idx[0].offset);
1248             break;
1249
1250         case WINED3DSPR_MISCTYPE:
1251             if (!reg->idx[0].offset)
1252                 sprintf(register_name, "vpos");
1253             else if (reg->idx[0].offset == 1)
1254                 sprintf(register_name, "fragment.facing.x");
1255             else
1256                 FIXME("Unknown MISCTYPE register index %u.\n", reg->idx[0].offset);
1257             break;
1258
1259         default:
1260             FIXME("Unhandled register type %#x[%u].\n", reg->type, reg->idx[0].offset);
1261             sprintf(register_name, "unrecognized_register[%u]", reg->idx[0].offset);
1262             break;
1263     }
1264 }
1265
1266 static void shader_arb_get_dst_param(const struct wined3d_shader_instruction *ins,
1267         const struct wined3d_shader_dst_param *wined3d_dst, char *str)
1268 {
1269     char register_name[255];
1270     char write_mask[6];
1271     BOOL is_color;
1272
1273     shader_arb_get_register_name(ins, &wined3d_dst->reg, register_name, &is_color);
1274     strcpy(str, register_name);
1275
1276     shader_arb_get_write_mask(ins, wined3d_dst, write_mask);
1277     strcat(str, write_mask);
1278 }
1279
1280 static const char *shader_arb_get_fixup_swizzle(enum fixup_channel_source channel_source)
1281 {
1282     switch(channel_source)
1283     {
1284         case CHANNEL_SOURCE_ZERO: return "0";
1285         case CHANNEL_SOURCE_ONE: return "1";
1286         case CHANNEL_SOURCE_X: return "x";
1287         case CHANNEL_SOURCE_Y: return "y";
1288         case CHANNEL_SOURCE_Z: return "z";
1289         case CHANNEL_SOURCE_W: return "w";
1290         default:
1291             FIXME("Unhandled channel source %#x\n", channel_source);
1292             return "undefined";
1293     }
1294 }
1295
1296 static void gen_color_correction(struct wined3d_shader_buffer *buffer, const char *reg,
1297         DWORD dst_mask, const char *one, const char *two, struct color_fixup_desc fixup)
1298 {
1299     DWORD mask;
1300
1301     if (is_complex_fixup(fixup))
1302     {
1303         enum complex_fixup complex_fixup = get_complex_fixup(fixup);
1304         FIXME("Complex fixup (%#x) not supported\n", complex_fixup);
1305         return;
1306     }
1307
1308     mask = 0;
1309     if (fixup.x_source != CHANNEL_SOURCE_X) mask |= WINED3DSP_WRITEMASK_0;
1310     if (fixup.y_source != CHANNEL_SOURCE_Y) mask |= WINED3DSP_WRITEMASK_1;
1311     if (fixup.z_source != CHANNEL_SOURCE_Z) mask |= WINED3DSP_WRITEMASK_2;
1312     if (fixup.w_source != CHANNEL_SOURCE_W) mask |= WINED3DSP_WRITEMASK_3;
1313     mask &= dst_mask;
1314
1315     if (mask)
1316     {
1317         shader_addline(buffer, "SWZ %s, %s, %s, %s, %s, %s;\n", reg, reg,
1318                 shader_arb_get_fixup_swizzle(fixup.x_source), shader_arb_get_fixup_swizzle(fixup.y_source),
1319                 shader_arb_get_fixup_swizzle(fixup.z_source), shader_arb_get_fixup_swizzle(fixup.w_source));
1320     }
1321
1322     mask = 0;
1323     if (fixup.x_sign_fixup) mask |= WINED3DSP_WRITEMASK_0;
1324     if (fixup.y_sign_fixup) mask |= WINED3DSP_WRITEMASK_1;
1325     if (fixup.z_sign_fixup) mask |= WINED3DSP_WRITEMASK_2;
1326     if (fixup.w_sign_fixup) mask |= WINED3DSP_WRITEMASK_3;
1327     mask &= dst_mask;
1328
1329     if (mask)
1330     {
1331         char reg_mask[6];
1332         char *ptr = reg_mask;
1333
1334         if (mask != WINED3DSP_WRITEMASK_ALL)
1335         {
1336             *ptr++ = '.';
1337             if (mask & WINED3DSP_WRITEMASK_0) *ptr++ = 'x';
1338             if (mask & WINED3DSP_WRITEMASK_1) *ptr++ = 'y';
1339             if (mask & WINED3DSP_WRITEMASK_2) *ptr++ = 'z';
1340             if (mask & WINED3DSP_WRITEMASK_3) *ptr++ = 'w';
1341         }
1342         *ptr = '\0';
1343
1344         shader_addline(buffer, "MAD %s%s, %s, %s, -%s;\n", reg, reg_mask, reg, two, one);
1345     }
1346 }
1347
1348 static const char *shader_arb_get_modifier(const struct wined3d_shader_instruction *ins)
1349 {
1350     DWORD mod;
1351     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
1352     if (!ins->dst_count) return "";
1353
1354     mod = ins->dst[0].modifiers;
1355
1356     /* Silently ignore PARTIALPRECISION if its not supported */
1357     if(priv->target_version == ARB) mod &= ~WINED3DSPDM_PARTIALPRECISION;
1358
1359     if(mod & WINED3DSPDM_MSAMPCENTROID)
1360     {
1361         FIXME("Unhandled modifier WINED3DSPDM_MSAMPCENTROID\n");
1362         mod &= ~WINED3DSPDM_MSAMPCENTROID;
1363     }
1364
1365     switch(mod)
1366     {
1367         case WINED3DSPDM_SATURATE | WINED3DSPDM_PARTIALPRECISION:
1368             return "H_SAT";
1369
1370         case WINED3DSPDM_SATURATE:
1371             return "_SAT";
1372
1373         case WINED3DSPDM_PARTIALPRECISION:
1374             return "H";
1375
1376         case 0:
1377             return "";
1378
1379         default:
1380             FIXME("Unknown modifiers 0x%08x\n", mod);
1381             return "";
1382     }
1383 }
1384
1385 #define TEX_PROJ        0x1
1386 #define TEX_BIAS        0x2
1387 #define TEX_LOD         0x4
1388 #define TEX_DERIV       0x10
1389
1390 static void shader_hw_sample(const struct wined3d_shader_instruction *ins, DWORD sampler_idx,
1391         const char *dst_str, const char *coord_reg, WORD flags, const char *dsx, const char *dsy)
1392 {
1393     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1394     DWORD sampler_type = ins->ctx->reg_maps->sampler_type[sampler_idx];
1395     const struct wined3d_shader *shader = ins->ctx->shader;
1396     const struct wined3d_texture *texture;
1397     const char *tex_type;
1398     BOOL np2_fixup = FALSE;
1399     struct wined3d_device *device = shader->device;
1400     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
1401     const char *mod;
1402     BOOL pshader = shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type);
1403
1404     /* D3D vertex shader sampler IDs are vertex samplers(0-3), not global d3d samplers */
1405     if(!pshader) sampler_idx += MAX_FRAGMENT_SAMPLERS;
1406
1407     switch(sampler_type) {
1408         case WINED3DSTT_1D:
1409             tex_type = "1D";
1410             break;
1411
1412         case WINED3DSTT_2D:
1413             texture = device->stateBlock->state.textures[sampler_idx];
1414             if (texture && texture->target == GL_TEXTURE_RECTANGLE_ARB)
1415             {
1416                 tex_type = "RECT";
1417             } else {
1418                 tex_type = "2D";
1419             }
1420             if (shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type))
1421             {
1422                 if (priv->cur_np2fixup_info->super.active & (1 << sampler_idx))
1423                 {
1424                     if (flags) FIXME("Only ordinary sampling from NP2 textures is supported.\n");
1425                     else np2_fixup = TRUE;
1426                 }
1427             }
1428             break;
1429
1430         case WINED3DSTT_VOLUME:
1431             tex_type = "3D";
1432             break;
1433
1434         case WINED3DSTT_CUBE:
1435             tex_type = "CUBE";
1436             break;
1437
1438         default:
1439             ERR("Unexpected texture type %d\n", sampler_type);
1440             tex_type = "";
1441     }
1442
1443     /* TEX, TXL, TXD and TXP do not support the "H" modifier,
1444      * so don't use shader_arb_get_modifier
1445      */
1446     if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE) mod = "_SAT";
1447     else mod = "";
1448
1449     /* Fragment samplers always have indentity mapping */
1450     if(sampler_idx >= MAX_FRAGMENT_SAMPLERS)
1451     {
1452         sampler_idx = priv->cur_vs_args->vertex.samplers[sampler_idx - MAX_FRAGMENT_SAMPLERS];
1453     }
1454
1455     if (flags & TEX_DERIV)
1456     {
1457         if(flags & TEX_PROJ) FIXME("Projected texture sampling with custom derivatives\n");
1458         if(flags & TEX_BIAS) FIXME("Biased texture sampling with custom derivatives\n");
1459         shader_addline(buffer, "TXD%s %s, %s, %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg,
1460                        dsx, dsy,sampler_idx, tex_type);
1461     }
1462     else if(flags & TEX_LOD)
1463     {
1464         if(flags & TEX_PROJ) FIXME("Projected texture sampling with explicit lod\n");
1465         if(flags & TEX_BIAS) FIXME("Biased texture sampling with explicit lod\n");
1466         shader_addline(buffer, "TXL%s %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg,
1467                        sampler_idx, tex_type);
1468     }
1469     else if (flags & TEX_BIAS)
1470     {
1471         /* Shouldn't be possible, but let's check for it */
1472         if(flags & TEX_PROJ) FIXME("Biased and Projected texture sampling\n");
1473         /* TXB takes the 4th component of the source vector automatically, as d3d. Nothing more to do */
1474         shader_addline(buffer, "TXB%s %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg, sampler_idx, tex_type);
1475     }
1476     else if (flags & TEX_PROJ)
1477     {
1478         shader_addline(buffer, "TXP%s %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg, sampler_idx, tex_type);
1479     }
1480     else
1481     {
1482         if (np2_fixup)
1483         {
1484             const unsigned char idx = priv->cur_np2fixup_info->super.idx[sampler_idx];
1485             shader_addline(buffer, "MUL TA, np2fixup[%u].%s, %s;\n", idx >> 1,
1486                            (idx % 2) ? "zwxy" : "xyzw", coord_reg);
1487
1488             shader_addline(buffer, "TEX%s %s, TA, texture[%u], %s;\n", mod, dst_str, sampler_idx, tex_type);
1489         }
1490         else
1491             shader_addline(buffer, "TEX%s %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg, sampler_idx, tex_type);
1492     }
1493
1494     if (pshader)
1495     {
1496         gen_color_correction(buffer, dst_str, ins->dst[0].write_mask,
1497                 arb_get_helper_value(WINED3D_SHADER_TYPE_PIXEL, ARB_ONE),
1498                 arb_get_helper_value(WINED3D_SHADER_TYPE_PIXEL, ARB_TWO),
1499                 priv->cur_ps_args->super.color_fixup[sampler_idx]);
1500     }
1501 }
1502
1503 static void shader_arb_get_src_param(const struct wined3d_shader_instruction *ins,
1504         const struct wined3d_shader_src_param *src, unsigned int tmpreg, char *outregstr)
1505 {
1506     /* Generate a line that does the input modifier computation and return the input register to use */
1507     BOOL is_color = FALSE;
1508     char regstr[256];
1509     char swzstr[20];
1510     int insert_line;
1511     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1512     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
1513     const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
1514     const char *two = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_TWO);
1515
1516     /* Assume a new line will be added */
1517     insert_line = 1;
1518
1519     /* Get register name */
1520     shader_arb_get_register_name(ins, &src->reg, regstr, &is_color);
1521     shader_arb_get_swizzle(src, is_color, swzstr);
1522
1523     switch (src->modifiers)
1524     {
1525     case WINED3DSPSM_NONE:
1526         sprintf(outregstr, "%s%s", regstr, swzstr);
1527         insert_line = 0;
1528         break;
1529     case WINED3DSPSM_NEG:
1530         sprintf(outregstr, "-%s%s", regstr, swzstr);
1531         insert_line = 0;
1532         break;
1533     case WINED3DSPSM_BIAS:
1534         shader_addline(buffer, "ADD T%c, %s, -coefdiv.x;\n", 'A' + tmpreg, regstr);
1535         break;
1536     case WINED3DSPSM_BIASNEG:
1537         shader_addline(buffer, "ADD T%c, -%s, coefdiv.x;\n", 'A' + tmpreg, regstr);
1538         break;
1539     case WINED3DSPSM_SIGN:
1540         shader_addline(buffer, "MAD T%c, %s, %s, -%s;\n", 'A' + tmpreg, regstr, two, one);
1541         break;
1542     case WINED3DSPSM_SIGNNEG:
1543         shader_addline(buffer, "MAD T%c, %s, -%s, %s;\n", 'A' + tmpreg, regstr, two, one);
1544         break;
1545     case WINED3DSPSM_COMP:
1546         shader_addline(buffer, "SUB T%c, %s, %s;\n", 'A' + tmpreg, one, regstr);
1547         break;
1548     case WINED3DSPSM_X2:
1549         shader_addline(buffer, "ADD T%c, %s, %s;\n", 'A' + tmpreg, regstr, regstr);
1550         break;
1551     case WINED3DSPSM_X2NEG:
1552         shader_addline(buffer, "ADD T%c, -%s, -%s;\n", 'A' + tmpreg, regstr, regstr);
1553         break;
1554     case WINED3DSPSM_DZ:
1555         shader_addline(buffer, "RCP T%c, %s.z;\n", 'A' + tmpreg, regstr);
1556         shader_addline(buffer, "MUL T%c, %s, T%c;\n", 'A' + tmpreg, regstr, 'A' + tmpreg);
1557         break;
1558     case WINED3DSPSM_DW:
1559         shader_addline(buffer, "RCP T%c, %s.w;\n", 'A' + tmpreg, regstr);
1560         shader_addline(buffer, "MUL T%c, %s, T%c;\n", 'A' + tmpreg, regstr, 'A' + tmpreg);
1561         break;
1562     case WINED3DSPSM_ABS:
1563         if(ctx->target_version >= NV2) {
1564             sprintf(outregstr, "|%s%s|", regstr, swzstr);
1565             insert_line = 0;
1566         } else {
1567             shader_addline(buffer, "ABS T%c, %s;\n", 'A' + tmpreg, regstr);
1568         }
1569         break;
1570     case WINED3DSPSM_ABSNEG:
1571         if(ctx->target_version >= NV2) {
1572             sprintf(outregstr, "-|%s%s|", regstr, swzstr);
1573         } else {
1574             shader_addline(buffer, "ABS T%c, %s;\n", 'A' + tmpreg, regstr);
1575             sprintf(outregstr, "-T%c%s", 'A' + tmpreg, swzstr);
1576         }
1577         insert_line = 0;
1578         break;
1579     default:
1580         sprintf(outregstr, "%s%s", regstr, swzstr);
1581         insert_line = 0;
1582     }
1583
1584     /* Return modified or original register, with swizzle */
1585     if (insert_line)
1586         sprintf(outregstr, "T%c%s", 'A' + tmpreg, swzstr);
1587 }
1588
1589 static void pshader_hw_bem(const struct wined3d_shader_instruction *ins)
1590 {
1591     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1592     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1593     DWORD sampler_code = dst->reg.idx[0].offset;
1594     char dst_name[50];
1595     char src_name[2][50];
1596
1597     shader_arb_get_dst_param(ins, dst, dst_name);
1598
1599     /* Sampling the perturbation map in Tsrc was done already, including the signedness correction if needed
1600      *
1601      * Keep in mind that src_name[1] can be "TB" and src_name[0] can be "TA" because modifiers like _x2 are valid
1602      * with bem. So delay loading the first parameter until after the perturbation calculation which needs two
1603      * temps is done.
1604      */
1605     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1606     shader_addline(buffer, "SWZ TA, bumpenvmat%d, x, z, 0, 0;\n", sampler_code);
1607     shader_addline(buffer, "DP3 TC.r, TA, %s;\n", src_name[1]);
1608     shader_addline(buffer, "SWZ TA, bumpenvmat%d, y, w, 0, 0;\n", sampler_code);
1609     shader_addline(buffer, "DP3 TC.g, TA, %s;\n", src_name[1]);
1610
1611     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
1612     shader_addline(buffer, "ADD %s, %s, TC;\n", dst_name, src_name[0]);
1613 }
1614
1615 static DWORD negate_modifiers(DWORD mod, char *extra_char)
1616 {
1617     *extra_char = ' ';
1618     switch(mod)
1619     {
1620         case WINED3DSPSM_NONE:      return WINED3DSPSM_NEG;
1621         case WINED3DSPSM_NEG:       return WINED3DSPSM_NONE;
1622         case WINED3DSPSM_BIAS:      return WINED3DSPSM_BIASNEG;
1623         case WINED3DSPSM_BIASNEG:   return WINED3DSPSM_BIAS;
1624         case WINED3DSPSM_SIGN:      return WINED3DSPSM_SIGNNEG;
1625         case WINED3DSPSM_SIGNNEG:   return WINED3DSPSM_SIGN;
1626         case WINED3DSPSM_COMP:      *extra_char = '-'; return WINED3DSPSM_COMP;
1627         case WINED3DSPSM_X2:        return WINED3DSPSM_X2NEG;
1628         case WINED3DSPSM_X2NEG:     return WINED3DSPSM_X2;
1629         case WINED3DSPSM_DZ:        *extra_char = '-'; return WINED3DSPSM_DZ;
1630         case WINED3DSPSM_DW:        *extra_char = '-'; return WINED3DSPSM_DW;
1631         case WINED3DSPSM_ABS:       return WINED3DSPSM_ABSNEG;
1632         case WINED3DSPSM_ABSNEG:    return WINED3DSPSM_ABS;
1633     }
1634     FIXME("Unknown modifier %u\n", mod);
1635     return mod;
1636 }
1637
1638 static void pshader_hw_cnd(const struct wined3d_shader_instruction *ins)
1639 {
1640     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1641     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1642     char dst_name[50];
1643     char src_name[3][50];
1644     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
1645             ins->ctx->reg_maps->shader_version.minor);
1646
1647     shader_arb_get_dst_param(ins, dst, dst_name);
1648     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1649
1650     /* The coissue flag changes the semantic of the cnd instruction in <= 1.3 shaders */
1651     if (shader_version <= WINED3D_SHADER_VERSION(1, 3) && ins->coissue)
1652     {
1653         shader_addline(buffer, "MOV%s %s, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name[1]);
1654     }
1655     else
1656     {
1657         struct wined3d_shader_src_param src0_copy = ins->src[0];
1658         char extra_neg;
1659
1660         /* src0 may have a negate srcmod set, so we can't blindly add "-" to the name */
1661         src0_copy.modifiers = negate_modifiers(src0_copy.modifiers, &extra_neg);
1662
1663         shader_arb_get_src_param(ins, &src0_copy, 0, src_name[0]);
1664         shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
1665         shader_addline(buffer, "ADD TA, %c%s, coefdiv.x;\n", extra_neg, src_name[0]);
1666         shader_addline(buffer, "CMP%s %s, TA, %s, %s;\n", shader_arb_get_modifier(ins),
1667                 dst_name, src_name[1], src_name[2]);
1668     }
1669 }
1670
1671 static void pshader_hw_cmp(const struct wined3d_shader_instruction *ins)
1672 {
1673     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1674     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1675     char dst_name[50];
1676     char src_name[3][50];
1677
1678     shader_arb_get_dst_param(ins, dst, dst_name);
1679
1680     /* Generate input register names (with modifiers) */
1681     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
1682     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1683     shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
1684
1685     shader_addline(buffer, "CMP%s %s, %s, %s, %s;\n", shader_arb_get_modifier(ins),
1686             dst_name, src_name[0], src_name[2], src_name[1]);
1687 }
1688
1689 /** Process the WINED3DSIO_DP2ADD instruction in ARB.
1690  * dst = dot2(src0, src1) + src2 */
1691 static void pshader_hw_dp2add(const struct wined3d_shader_instruction *ins)
1692 {
1693     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1694     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1695     char dst_name[50];
1696     char src_name[3][50];
1697     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
1698
1699     shader_arb_get_dst_param(ins, dst, dst_name);
1700     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
1701     shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
1702
1703     if(ctx->target_version >= NV3)
1704     {
1705         /* GL_NV_fragment_program2 has a 1:1 matching instruction */
1706         shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1707         shader_addline(buffer, "DP2A%s %s, %s, %s, %s;\n", shader_arb_get_modifier(ins),
1708                        dst_name, src_name[0], src_name[1], src_name[2]);
1709     }
1710     else if(ctx->target_version >= NV2)
1711     {
1712         /* dst.x = src2.?, src0.x, src1.x + src0.y * src1.y
1713          * dst.y = src2.?, src0.x, src1.z + src0.y * src1.w
1714          * dst.z = src2.?, src0.x, src1.x + src0.y * src1.y
1715          * dst.z = src2.?, src0.x, src1.z + src0.y * src1.w
1716          *
1717          * Make sure that src1.zw = src1.xy, then we get a classic dp2add
1718          *
1719          * .xyxy and other swizzles that we could get with this are not valid in
1720          * plain ARBfp, but luckily the NV extension grammar lifts this limitation.
1721          */
1722         struct wined3d_shader_src_param tmp_param = ins->src[1];
1723         DWORD swizzle = tmp_param.swizzle & 0xf; /* Selects .xy */
1724         tmp_param.swizzle = swizzle | (swizzle << 4); /* Creates .xyxy */
1725
1726         shader_arb_get_src_param(ins, &tmp_param, 1, src_name[1]);
1727
1728         shader_addline(buffer, "X2D%s %s, %s, %s, %s;\n", shader_arb_get_modifier(ins),
1729                        dst_name, src_name[2], src_name[0], src_name[1]);
1730     }
1731     else
1732     {
1733         shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1734         /* 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]
1735         * src_name[0] can be TA, but TA is a private temp for modifiers, so it is save to overwrite
1736         */
1737         shader_addline(buffer, "MOV TA, %s;\n", src_name[0]);
1738         shader_addline(buffer, "MOV TA.z, 0.0;\n");
1739         shader_addline(buffer, "DP3 TA, TA, %s;\n", src_name[1]);
1740         shader_addline(buffer, "ADD%s %s, TA, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name[2]);
1741     }
1742 }
1743
1744 /* Map the opcode 1-to-1 to the GL code */
1745 static void shader_hw_map2gl(const struct wined3d_shader_instruction *ins)
1746 {
1747     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1748     const char *instruction;
1749     char arguments[256], dst_str[50];
1750     unsigned int i;
1751     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1752
1753     switch (ins->handler_idx)
1754     {
1755         case WINED3DSIH_ABS: instruction = "ABS"; break;
1756         case WINED3DSIH_ADD: instruction = "ADD"; break;
1757         case WINED3DSIH_CRS: instruction = "XPD"; break;
1758         case WINED3DSIH_DP3: instruction = "DP3"; break;
1759         case WINED3DSIH_DP4: instruction = "DP4"; break;
1760         case WINED3DSIH_DST: instruction = "DST"; break;
1761         case WINED3DSIH_FRC: instruction = "FRC"; break;
1762         case WINED3DSIH_LIT: instruction = "LIT"; break;
1763         case WINED3DSIH_LRP: instruction = "LRP"; break;
1764         case WINED3DSIH_MAD: instruction = "MAD"; break;
1765         case WINED3DSIH_MAX: instruction = "MAX"; break;
1766         case WINED3DSIH_MIN: instruction = "MIN"; break;
1767         case WINED3DSIH_MOV: instruction = "MOV"; break;
1768         case WINED3DSIH_MUL: instruction = "MUL"; break;
1769         case WINED3DSIH_SGE: instruction = "SGE"; break;
1770         case WINED3DSIH_SLT: instruction = "SLT"; break;
1771         case WINED3DSIH_SUB: instruction = "SUB"; break;
1772         case WINED3DSIH_MOVA:instruction = "ARR"; break;
1773         case WINED3DSIH_DSX: instruction = "DDX"; break;
1774         default: instruction = "";
1775             FIXME("Unhandled opcode %#x\n", ins->handler_idx);
1776             break;
1777     }
1778
1779     /* Note that shader_arb_add_dst_param() adds spaces. */
1780     arguments[0] = '\0';
1781     shader_arb_get_dst_param(ins, dst, dst_str);
1782     for (i = 0; i < ins->src_count; ++i)
1783     {
1784         char operand[100];
1785         strcat(arguments, ", ");
1786         shader_arb_get_src_param(ins, &ins->src[i], i, operand);
1787         strcat(arguments, operand);
1788     }
1789     shader_addline(buffer, "%s%s %s%s;\n", instruction, shader_arb_get_modifier(ins), dst_str, arguments);
1790 }
1791
1792 static void shader_hw_nop(const struct wined3d_shader_instruction *ins) {}
1793
1794 static void shader_hw_mov(const struct wined3d_shader_instruction *ins)
1795 {
1796     const struct wined3d_shader *shader = ins->ctx->shader;
1797     const struct wined3d_shader_reg_maps *reg_maps = ins->ctx->reg_maps;
1798     BOOL pshader = shader_is_pshader_version(reg_maps->shader_version.type);
1799     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
1800     const char *zero = arb_get_helper_value(reg_maps->shader_version.type, ARB_ZERO);
1801     const char *one = arb_get_helper_value(reg_maps->shader_version.type, ARB_ONE);
1802     const char *two = arb_get_helper_value(reg_maps->shader_version.type, ARB_TWO);
1803
1804     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1805     char src0_param[256];
1806
1807     if (ins->handler_idx == WINED3DSIH_MOVA)
1808     {
1809         const struct arb_vshader_private *shader_data = shader->backend_data;
1810         char write_mask[6];
1811         const char *offset = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_VS_REL_OFFSET);
1812
1813         if(ctx->target_version >= NV2) {
1814             shader_hw_map2gl(ins);
1815             return;
1816         }
1817         shader_arb_get_src_param(ins, &ins->src[0], 0, src0_param);
1818         shader_arb_get_write_mask(ins, &ins->dst[0], write_mask);
1819
1820         /* This implements the mova formula used in GLSL. The first two instructions
1821          * prepare the sign() part. Note that it is fine to have my_sign(0.0) = 1.0
1822          * in this case:
1823          * mova A0.x, 0.0
1824          *
1825          * A0.x = arl(floor(abs(0.0) + 0.5) * 1.0) = floor(0.5) = 0.0 since arl does a floor
1826          *
1827          * The ARL is performed when A0 is used - the requested component is read from A0_SHADOW into
1828          * A0.x. We can use the overwritten component of A0_shadow as temporary storage for the sign.
1829          */
1830         shader_addline(buffer, "SGE A0_SHADOW%s, %s, %s;\n", write_mask, src0_param, zero);
1831         shader_addline(buffer, "MAD A0_SHADOW%s, A0_SHADOW, %s, -%s;\n", write_mask, two, one);
1832
1833         shader_addline(buffer, "ABS TA%s, %s;\n", write_mask, src0_param);
1834         shader_addline(buffer, "ADD TA%s, TA, rel_addr_const.x;\n", write_mask);
1835         shader_addline(buffer, "FLR TA%s, TA;\n", write_mask);
1836         if (shader_data->rel_offset)
1837         {
1838             shader_addline(buffer, "ADD TA%s, TA, %s;\n", write_mask, offset);
1839         }
1840         shader_addline(buffer, "MUL A0_SHADOW%s, TA, A0_SHADOW;\n", write_mask);
1841
1842         ((struct shader_arb_ctx_priv *)ins->ctx->backend_data)->addr_reg[0] = '\0';
1843     }
1844     else if (reg_maps->shader_version.major == 1
1845           && !shader_is_pshader_version(reg_maps->shader_version.type)
1846           && ins->dst[0].reg.type == WINED3DSPR_ADDR)
1847     {
1848         const struct arb_vshader_private *shader_data = shader->backend_data;
1849         src0_param[0] = '\0';
1850
1851         if (shader_data->rel_offset)
1852         {
1853             const char *offset = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_VS_REL_OFFSET);
1854             shader_arb_get_src_param(ins, &ins->src[0], 0, src0_param);
1855             shader_addline(buffer, "ADD TA.x, %s, %s;\n", src0_param, offset);
1856             shader_addline(buffer, "ARL A0.x, TA.x;\n");
1857         }
1858         else
1859         {
1860             /* Apple's ARB_vertex_program implementation does not accept an ARL source argument
1861              * with more than one component. Thus replicate the first source argument over all
1862              * 4 components. For example, .xyzw -> .x (or better: .xxxx), .zwxy -> .z, etc) */
1863             struct wined3d_shader_src_param tmp_src = ins->src[0];
1864             tmp_src.swizzle = (tmp_src.swizzle & 0x3) * 0x55;
1865             shader_arb_get_src_param(ins, &tmp_src, 0, src0_param);
1866             shader_addline(buffer, "ARL A0.x, %s;\n", src0_param);
1867         }
1868     }
1869     else if (ins->dst[0].reg.type == WINED3DSPR_COLOROUT && !ins->dst[0].reg.idx[0].offset && pshader)
1870     {
1871         if (ctx->cur_ps_args->super.srgb_correction && shader->u.ps.color0_mov)
1872         {
1873             shader_addline(buffer, "#mov handled in srgb write code\n");
1874             return;
1875         }
1876         shader_hw_map2gl(ins);
1877     }
1878     else
1879     {
1880         shader_hw_map2gl(ins);
1881     }
1882 }
1883
1884 static void pshader_hw_texkill(const struct wined3d_shader_instruction *ins)
1885 {
1886     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1887     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1888     char reg_dest[40];
1889
1890     /* No swizzles are allowed in d3d's texkill. PS 1.x ignores the 4th component as documented,
1891      * but >= 2.0 honors it (undocumented, but tested by the d3d9 testsuite)
1892      */
1893     shader_arb_get_dst_param(ins, dst, reg_dest);
1894
1895     if (ins->ctx->reg_maps->shader_version.major >= 2)
1896     {
1897         const char *kilsrc = "TA";
1898         BOOL is_color;
1899
1900         shader_arb_get_register_name(ins, &dst->reg, reg_dest, &is_color);
1901         if(dst->write_mask == WINED3DSP_WRITEMASK_ALL)
1902         {
1903             kilsrc = reg_dest;
1904         }
1905         else
1906         {
1907             /* Sigh. KIL doesn't support swizzles/writemasks. KIL passes a writemask, but ".xy" for example
1908              * is not valid as a swizzle in ARB (needs ".xyyy"). Use SWZ to load the register properly, and set
1909              * masked out components to 0(won't kill)
1910              */
1911             char x = '0', y = '0', z = '0', w = '0';
1912             if(dst->write_mask & WINED3DSP_WRITEMASK_0) x = 'x';
1913             if(dst->write_mask & WINED3DSP_WRITEMASK_1) y = 'y';
1914             if(dst->write_mask & WINED3DSP_WRITEMASK_2) z = 'z';
1915             if(dst->write_mask & WINED3DSP_WRITEMASK_3) w = 'w';
1916             shader_addline(buffer, "SWZ TA, %s, %c, %c, %c, %c;\n", reg_dest, x, y, z, w);
1917         }
1918         shader_addline(buffer, "KIL %s;\n", kilsrc);
1919     }
1920     else
1921     {
1922         /* ARB fp doesn't like swizzles on the parameter of the KIL instruction. To mask the 4th component,
1923          * copy the register into our general purpose TMP variable, overwrite .w and pass TMP to KIL
1924          *
1925          * ps_1_3 shaders use the texcoord incarnation of the Tx register. ps_1_4 shaders can use the same,
1926          * or pass in any temporary register(in shader phase 2)
1927          */
1928         if (ins->ctx->reg_maps->shader_version.minor <= 3)
1929             sprintf(reg_dest, "fragment.texcoord[%u]", dst->reg.idx[0].offset);
1930         else
1931             shader_arb_get_dst_param(ins, dst, reg_dest);
1932         shader_addline(buffer, "SWZ TA, %s, x, y, z, 1;\n", reg_dest);
1933         shader_addline(buffer, "KIL TA;\n");
1934     }
1935 }
1936
1937 static void pshader_hw_tex(const struct wined3d_shader_instruction *ins)
1938 {
1939     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
1940     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1941     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
1942             ins->ctx->reg_maps->shader_version.minor);
1943     struct wined3d_shader_src_param src;
1944
1945     char reg_dest[40];
1946     char reg_coord[40];
1947     DWORD reg_sampler_code;
1948     WORD myflags = 0;
1949     BOOL swizzle_coord = FALSE;
1950
1951     /* All versions have a destination register */
1952     shader_arb_get_dst_param(ins, dst, reg_dest);
1953
1954     /* 1.0-1.4: Use destination register number as texture code.
1955        2.0+: Use provided sampler number as texure code. */
1956     if (shader_version < WINED3D_SHADER_VERSION(2,0))
1957         reg_sampler_code = dst->reg.idx[0].offset;
1958     else
1959         reg_sampler_code = ins->src[1].reg.idx[0].offset;
1960
1961     /* 1.0-1.3: Use the texcoord varying.
1962        1.4+: Use provided coordinate source register. */
1963     if (shader_version < WINED3D_SHADER_VERSION(1,4))
1964         sprintf(reg_coord, "fragment.texcoord[%u]", reg_sampler_code);
1965     else {
1966         /* TEX is the only instruction that can handle DW and DZ natively */
1967         src = ins->src[0];
1968         if(src.modifiers == WINED3DSPSM_DW) src.modifiers = WINED3DSPSM_NONE;
1969         if(src.modifiers == WINED3DSPSM_DZ) src.modifiers = WINED3DSPSM_NONE;
1970         shader_arb_get_src_param(ins, &src, 0, reg_coord);
1971     }
1972
1973     /* projection flag:
1974      * 1.1, 1.2, 1.3: Use WINED3D_TSS_TEXTURETRANSFORMFLAGS
1975      * 1.4: Use WINED3DSPSM_DZ or WINED3DSPSM_DW on src[0]
1976      * 2.0+: Use WINED3DSI_TEXLD_PROJECT on the opcode
1977      */
1978     if (shader_version < WINED3D_SHADER_VERSION(1,4))
1979     {
1980         DWORD flags = 0;
1981         if (reg_sampler_code < MAX_TEXTURES)
1982             flags = priv->cur_ps_args->super.tex_transform >> reg_sampler_code * WINED3D_PSARGS_TEXTRANSFORM_SHIFT;
1983         if (flags & WINED3D_PSARGS_PROJECTED)
1984         {
1985             myflags |= TEX_PROJ;
1986             if ((flags & ~WINED3D_PSARGS_PROJECTED) == WINED3D_TTFF_COUNT3)
1987                 swizzle_coord = TRUE;
1988         }
1989     }
1990     else if (shader_version < WINED3D_SHADER_VERSION(2,0))
1991     {
1992         enum wined3d_shader_src_modifier src_mod = ins->src[0].modifiers;
1993         if (src_mod == WINED3DSPSM_DZ)
1994         {
1995             swizzle_coord = TRUE;
1996             myflags |= TEX_PROJ;
1997         } else if(src_mod == WINED3DSPSM_DW) {
1998             myflags |= TEX_PROJ;
1999         }
2000     } else {
2001         if (ins->flags & WINED3DSI_TEXLD_PROJECT) myflags |= TEX_PROJ;
2002         if (ins->flags & WINED3DSI_TEXLD_BIAS) myflags |= TEX_BIAS;
2003     }
2004
2005     if (swizzle_coord)
2006     {
2007         /* TXP cannot handle DZ natively, so move the z coordinate to .w.
2008          * reg_coord is a read-only varying register, so we need a temp reg */
2009         shader_addline(ins->ctx->buffer, "SWZ TA, %s, x, y, z, z;\n", reg_coord);
2010         strcpy(reg_coord, "TA");
2011     }
2012
2013     shader_hw_sample(ins, reg_sampler_code, reg_dest, reg_coord, myflags, NULL, NULL);
2014 }
2015
2016 static void pshader_hw_texcoord(const struct wined3d_shader_instruction *ins)
2017 {
2018     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2019     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2020     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
2021             ins->ctx->reg_maps->shader_version.minor);
2022     char dst_str[50];
2023
2024     if (shader_version < WINED3D_SHADER_VERSION(1,4))
2025     {
2026         DWORD reg = dst->reg.idx[0].offset;
2027
2028         shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2029         shader_addline(buffer, "MOV_SAT %s, fragment.texcoord[%u];\n", dst_str, reg);
2030     } else {
2031         char reg_src[40];
2032
2033         shader_arb_get_src_param(ins, &ins->src[0], 0, reg_src);
2034         shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2035         shader_addline(buffer, "MOV %s, %s;\n", dst_str, reg_src);
2036    }
2037 }
2038
2039 static void pshader_hw_texreg2ar(const struct wined3d_shader_instruction *ins)
2040 {
2041      struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2042      DWORD flags = 0;
2043
2044      DWORD reg1 = ins->dst[0].reg.idx[0].offset;
2045      char dst_str[50];
2046      char src_str[50];
2047
2048      /* Note that texreg2ar treats Tx as a temporary register, not as a varying */
2049      shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2050      shader_arb_get_src_param(ins, &ins->src[0], 0, src_str);
2051      /* Move .x first in case src_str is "TA" */
2052      shader_addline(buffer, "MOV TA.y, %s.x;\n", src_str);
2053      shader_addline(buffer, "MOV TA.x, %s.w;\n", src_str);
2054      if (reg1 < MAX_TEXTURES)
2055      {
2056          struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2057          flags = priv->cur_ps_args->super.tex_transform >> reg1 * WINED3D_PSARGS_TEXTRANSFORM_SHIFT;
2058      }
2059      shader_hw_sample(ins, reg1, dst_str, "TA", flags & WINED3D_PSARGS_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2060 }
2061
2062 static void pshader_hw_texreg2gb(const struct wined3d_shader_instruction *ins)
2063 {
2064      struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2065
2066      DWORD reg1 = ins->dst[0].reg.idx[0].offset;
2067      char dst_str[50];
2068      char src_str[50];
2069
2070      /* Note that texreg2gb treats Tx as a temporary register, not as a varying */
2071      shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2072      shader_arb_get_src_param(ins, &ins->src[0], 0, src_str);
2073      shader_addline(buffer, "MOV TA.x, %s.y;\n", src_str);
2074      shader_addline(buffer, "MOV TA.y, %s.z;\n", src_str);
2075      shader_hw_sample(ins, reg1, dst_str, "TA", 0, NULL, NULL);
2076 }
2077
2078 static void pshader_hw_texreg2rgb(const struct wined3d_shader_instruction *ins)
2079 {
2080     DWORD reg1 = ins->dst[0].reg.idx[0].offset;
2081     char dst_str[50];
2082     char src_str[50];
2083
2084     /* Note that texreg2rg treats Tx as a temporary register, not as a varying */
2085     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2086     shader_arb_get_src_param(ins, &ins->src[0], 0, src_str);
2087     shader_hw_sample(ins, reg1, dst_str, src_str, 0, NULL, NULL);
2088 }
2089
2090 static void pshader_hw_texbem(const struct wined3d_shader_instruction *ins)
2091 {
2092     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2093     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2094     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2095     char reg_coord[40], dst_reg[50], src_reg[50];
2096     DWORD reg_dest_code;
2097
2098     /* All versions have a destination register. The Tx where the texture coordinates come
2099      * from is the varying incarnation of the texture register
2100      */
2101     reg_dest_code = dst->reg.idx[0].offset;
2102     shader_arb_get_dst_param(ins, &ins->dst[0], dst_reg);
2103     shader_arb_get_src_param(ins, &ins->src[0], 0, src_reg);
2104     sprintf(reg_coord, "fragment.texcoord[%u]", reg_dest_code);
2105
2106     /* Sampling the perturbation map in Tsrc was done already, including the signedness correction if needed
2107      * The Tx in which the perturbation map is stored is the tempreg incarnation of the texture register
2108      *
2109      * GL_NV_fragment_program_option could handle this in one instruction via X2D:
2110      * X2D TA.xy, fragment.texcoord, T%u, bumpenvmat%u.xzyw
2111      *
2112      * However, the NV extensions are never enabled for <= 2.0 shaders because of the performance penalty that
2113      * comes with it, and texbem is an 1.x only instruction. No 1.x instruction forces us to enable the NV
2114      * extension.
2115      */
2116     shader_addline(buffer, "SWZ TB, bumpenvmat%d, x, z, 0, 0;\n", reg_dest_code);
2117     shader_addline(buffer, "DP3 TA.x, TB, %s;\n", src_reg);
2118     shader_addline(buffer, "SWZ TB, bumpenvmat%d, y, w, 0, 0;\n", reg_dest_code);
2119     shader_addline(buffer, "DP3 TA.y, TB, %s;\n", src_reg);
2120
2121     /* with projective textures, texbem only divides the static texture coord, not the displacement,
2122      * so we can't let the GL handle this.
2123      */
2124     if ((priv->cur_ps_args->super.tex_transform >> reg_dest_code * WINED3D_PSARGS_TEXTRANSFORM_SHIFT)
2125             & WINED3D_PSARGS_PROJECTED)
2126     {
2127         shader_addline(buffer, "RCP TB.w, %s.w;\n", reg_coord);
2128         shader_addline(buffer, "MUL TB.xy, %s, TB.w;\n", reg_coord);
2129         shader_addline(buffer, "ADD TA.xy, TA, TB;\n");
2130     } else {
2131         shader_addline(buffer, "ADD TA.xy, TA, %s;\n", reg_coord);
2132     }
2133
2134     shader_hw_sample(ins, reg_dest_code, dst_reg, "TA", 0, NULL, NULL);
2135
2136     if (ins->handler_idx == WINED3DSIH_TEXBEML)
2137     {
2138         /* No src swizzles are allowed, so this is ok */
2139         shader_addline(buffer, "MAD TA, %s.z, luminance%d.x, luminance%d.y;\n",
2140                        src_reg, reg_dest_code, reg_dest_code);
2141         shader_addline(buffer, "MUL %s, %s, TA;\n", dst_reg, dst_reg);
2142     }
2143 }
2144
2145 static void pshader_hw_texm3x2pad(const struct wined3d_shader_instruction *ins)
2146 {
2147     DWORD reg = ins->dst[0].reg.idx[0].offset;
2148     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2149     char src0_name[50], dst_name[50];
2150     BOOL is_color;
2151     struct wined3d_shader_register tmp_reg = ins->dst[0].reg;
2152
2153     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2154     /* The next instruction will be a texm3x2tex or texm3x2depth that writes to the uninitialized
2155      * T<reg+1> register. Use this register to store the calculated vector
2156      */
2157     tmp_reg.idx[0].offset = reg + 1;
2158     shader_arb_get_register_name(ins, &tmp_reg, dst_name, &is_color);
2159     shader_addline(buffer, "DP3 %s.x, fragment.texcoord[%u], %s;\n", dst_name, reg, src0_name);
2160 }
2161
2162 static void pshader_hw_texm3x2tex(const struct wined3d_shader_instruction *ins)
2163 {
2164     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2165     DWORD flags;
2166     DWORD reg = ins->dst[0].reg.idx[0].offset;
2167     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2168     char dst_str[50];
2169     char src0_name[50];
2170     char dst_reg[50];
2171     BOOL is_color;
2172
2173     /* We know that we're writing to the uninitialized T<reg> register, so use it for temporary storage */
2174     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_reg, &is_color);
2175
2176     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2177     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2178     shader_addline(buffer, "DP3 %s.y, fragment.texcoord[%u], %s;\n", dst_reg, reg, src0_name);
2179     flags = reg < MAX_TEXTURES ? priv->cur_ps_args->super.tex_transform >> reg * WINED3D_PSARGS_TEXTRANSFORM_SHIFT : 0;
2180     shader_hw_sample(ins, reg, dst_str, dst_reg, flags & WINED3D_PSARGS_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2181 }
2182
2183 static void pshader_hw_texm3x3pad(const struct wined3d_shader_instruction *ins)
2184 {
2185     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
2186     DWORD reg = ins->dst[0].reg.idx[0].offset;
2187     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2188     char src0_name[50], dst_name[50];
2189     struct wined3d_shader_register tmp_reg = ins->dst[0].reg;
2190     BOOL is_color;
2191
2192     /* There are always 2 texm3x3pad instructions followed by one texm3x3[tex,vspec, ...] instruction, with
2193      * incrementing ins->dst[0].register_idx numbers. So the pad instruction already knows the final destination
2194      * register, and this register is uninitialized(otherwise the assembler complains that it is 'redeclared')
2195      */
2196     tmp_reg.idx[0].offset = reg + 2 - tex_mx->current_row;
2197     shader_arb_get_register_name(ins, &tmp_reg, dst_name, &is_color);
2198
2199     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2200     shader_addline(buffer, "DP3 %s.%c, fragment.texcoord[%u], %s;\n",
2201                    dst_name, 'x' + tex_mx->current_row, reg, src0_name);
2202     tex_mx->texcoord_w[tex_mx->current_row++] = reg;
2203 }
2204
2205 static void pshader_hw_texm3x3tex(const struct wined3d_shader_instruction *ins)
2206 {
2207     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2208     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
2209     DWORD flags;
2210     DWORD reg = ins->dst[0].reg.idx[0].offset;
2211     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2212     char dst_str[50];
2213     char src0_name[50], dst_name[50];
2214     BOOL is_color;
2215
2216     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2217     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2218     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_name, reg, src0_name);
2219
2220     /* Sample the texture using the calculated coordinates */
2221     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2222     flags = reg < MAX_TEXTURES ? priv->cur_ps_args->super.tex_transform >> reg * WINED3D_PSARGS_TEXTRANSFORM_SHIFT : 0;
2223     shader_hw_sample(ins, reg, dst_str, dst_name, flags & WINED3D_PSARGS_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2224     tex_mx->current_row = 0;
2225 }
2226
2227 static void pshader_hw_texm3x3vspec(const struct wined3d_shader_instruction *ins)
2228 {
2229     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2230     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
2231     DWORD flags;
2232     DWORD reg = ins->dst[0].reg.idx[0].offset;
2233     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2234     char dst_str[50];
2235     char src0_name[50];
2236     char dst_reg[50];
2237     BOOL is_color;
2238
2239     /* Get the dst reg without writemask strings. We know this register is uninitialized, so we can use all
2240      * components for temporary data storage
2241      */
2242     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_reg, &is_color);
2243     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2244     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_reg, reg, src0_name);
2245
2246     /* Construct the eye-ray vector from w coordinates */
2247     shader_addline(buffer, "MOV TB.x, fragment.texcoord[%u].w;\n", tex_mx->texcoord_w[0]);
2248     shader_addline(buffer, "MOV TB.y, fragment.texcoord[%u].w;\n", tex_mx->texcoord_w[1]);
2249     shader_addline(buffer, "MOV TB.z, fragment.texcoord[%u].w;\n", reg);
2250
2251     /* Calculate reflection vector
2252      */
2253     shader_addline(buffer, "DP3 %s.w, %s, TB;\n", dst_reg, dst_reg);
2254     /* The .w is ignored when sampling, so I can use TB.w to calculate dot(N, N) */
2255     shader_addline(buffer, "DP3 TB.w, %s, %s;\n", dst_reg, dst_reg);
2256     shader_addline(buffer, "RCP TB.w, TB.w;\n");
2257     shader_addline(buffer, "MUL %s.w, %s.w, TB.w;\n", dst_reg, dst_reg);
2258     shader_addline(buffer, "MUL %s, %s.w, %s;\n", dst_reg, dst_reg, dst_reg);
2259     shader_addline(buffer, "MAD %s, coefmul.x, %s, -TB;\n", dst_reg, dst_reg);
2260
2261     /* Sample the texture using the calculated coordinates */
2262     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2263     flags = reg < MAX_TEXTURES ? priv->cur_ps_args->super.tex_transform >> reg * WINED3D_PSARGS_TEXTRANSFORM_SHIFT : 0;
2264     shader_hw_sample(ins, reg, dst_str, dst_reg, flags & WINED3D_PSARGS_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2265     tex_mx->current_row = 0;
2266 }
2267
2268 static void pshader_hw_texm3x3spec(const struct wined3d_shader_instruction *ins)
2269 {
2270     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2271     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
2272     DWORD flags;
2273     DWORD reg = ins->dst[0].reg.idx[0].offset;
2274     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2275     char dst_str[50];
2276     char src0_name[50];
2277     char src1_name[50];
2278     char dst_reg[50];
2279     BOOL is_color;
2280
2281     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2282     shader_arb_get_src_param(ins, &ins->src[0], 1, src1_name);
2283     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_reg, &is_color);
2284     /* Note: dst_reg.xy is input here, generated by two texm3x3pad instructions */
2285     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_reg, reg, src0_name);
2286
2287     /* Calculate reflection vector.
2288      *
2289      *                   dot(N, E)
2290      * dst_reg.xyz = 2 * --------- * N - E
2291      *                   dot(N, N)
2292      *
2293      * Which normalizes the normal vector
2294      */
2295     shader_addline(buffer, "DP3 %s.w, %s, %s;\n", dst_reg, dst_reg, src1_name);
2296     shader_addline(buffer, "DP3 TC.w, %s, %s;\n", dst_reg, dst_reg);
2297     shader_addline(buffer, "RCP TC.w, TC.w;\n");
2298     shader_addline(buffer, "MUL %s.w, %s.w, TC.w;\n", dst_reg, dst_reg);
2299     shader_addline(buffer, "MUL %s, %s.w, %s;\n", dst_reg, dst_reg, dst_reg);
2300     shader_addline(buffer, "MAD %s, coefmul.x, %s, -%s;\n", dst_reg, dst_reg, src1_name);
2301
2302     /* Sample the texture using the calculated coordinates */
2303     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2304     flags = reg < MAX_TEXTURES ? priv->cur_ps_args->super.tex_transform >> reg * WINED3D_PSARGS_TEXTRANSFORM_SHIFT : 0;
2305     shader_hw_sample(ins, reg, dst_str, dst_reg, flags & WINED3D_PSARGS_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2306     tex_mx->current_row = 0;
2307 }
2308
2309 static void pshader_hw_texdepth(const struct wined3d_shader_instruction *ins)
2310 {
2311     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2312     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2313     char dst_name[50];
2314     const char *zero = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ZERO);
2315     const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
2316
2317     /* texdepth has an implicit destination, the fragment depth value. It's only parameter,
2318      * which is essentially an input, is the destination register because it is the first
2319      * parameter. According to the msdn, this must be register r5, but let's keep it more flexible
2320      * here(writemasks/swizzles are not valid on texdepth)
2321      */
2322     shader_arb_get_dst_param(ins, dst, dst_name);
2323
2324     /* According to the msdn, the source register(must be r5) is unusable after
2325      * the texdepth instruction, so we're free to modify it
2326      */
2327     shader_addline(buffer, "MIN %s.y, %s.y, %s;\n", dst_name, dst_name, one);
2328
2329     /* How to deal with the special case dst_name.g == 0? if r != 0, then
2330      * the r * (1 / 0) will give infinity, which is clamped to 1.0, the correct
2331      * result. But if r = 0.0, then 0 * inf = 0, which is incorrect.
2332      */
2333     shader_addline(buffer, "RCP %s.y, %s.y;\n", dst_name, dst_name);
2334     shader_addline(buffer, "MUL TA.x, %s.x, %s.y;\n", dst_name, dst_name);
2335     shader_addline(buffer, "MIN TA.x, TA.x, %s;\n", one);
2336     shader_addline(buffer, "MAX result.depth, TA.x, %s;\n", zero);
2337 }
2338
2339 /** Process the WINED3DSIO_TEXDP3TEX instruction in ARB:
2340  * Take a 3-component dot product of the TexCoord[dstreg] and src,
2341  * then perform a 1D texture lookup from stage dstregnum, place into dst. */
2342 static void pshader_hw_texdp3tex(const struct wined3d_shader_instruction *ins)
2343 {
2344     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2345     DWORD sampler_idx = ins->dst[0].reg.idx[0].offset;
2346     char src0[50];
2347     char dst_str[50];
2348
2349     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2350     shader_addline(buffer, "MOV TB, 0.0;\n");
2351     shader_addline(buffer, "DP3 TB.x, fragment.texcoord[%u], %s;\n", sampler_idx, src0);
2352
2353     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2354     shader_hw_sample(ins, sampler_idx, dst_str, "TB", 0 /* Only one coord, can't be projected */, NULL, NULL);
2355 }
2356
2357 /** Process the WINED3DSIO_TEXDP3 instruction in ARB:
2358  * Take a 3-component dot product of the TexCoord[dstreg] and src. */
2359 static void pshader_hw_texdp3(const struct wined3d_shader_instruction *ins)
2360 {
2361     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2362     char src0[50];
2363     char dst_str[50];
2364     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2365
2366     /* Handle output register */
2367     shader_arb_get_dst_param(ins, dst, dst_str);
2368     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2369     shader_addline(buffer, "DP3 %s, fragment.texcoord[%u], %s;\n", dst_str, dst->reg.idx[0].offset, src0);
2370 }
2371
2372 /** Process the WINED3DSIO_TEXM3X3 instruction in ARB
2373  * Perform the 3rd row of a 3x3 matrix multiply */
2374 static void pshader_hw_texm3x3(const struct wined3d_shader_instruction *ins)
2375 {
2376     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2377     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2378     char dst_str[50], dst_name[50];
2379     char src0[50];
2380     BOOL is_color;
2381
2382     shader_arb_get_dst_param(ins, dst, dst_str);
2383     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2384     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2385     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_name, dst->reg.idx[0].offset, src0);
2386     shader_addline(buffer, "MOV %s, %s;\n", dst_str, dst_name);
2387 }
2388
2389 /** Process the WINED3DSIO_TEXM3X2DEPTH instruction in ARB:
2390  * Last row of a 3x2 matrix multiply, use the result to calculate the depth:
2391  * Calculate tmp0.y = TexCoord[dstreg] . src.xyz;  (tmp0.x has already been calculated)
2392  * depth = (tmp0.y == 0.0) ? 1.0 : tmp0.x / tmp0.y
2393  */
2394 static void pshader_hw_texm3x2depth(const struct wined3d_shader_instruction *ins)
2395 {
2396     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2397     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2398     char src0[50], dst_name[50];
2399     BOOL is_color;
2400     const char *zero = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ZERO);
2401     const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
2402
2403     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2404     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2405     shader_addline(buffer, "DP3 %s.y, fragment.texcoord[%u], %s;\n", dst_name, dst->reg.idx[0].offset, src0);
2406
2407     /* How to deal with the special case dst_name.g == 0? if r != 0, then
2408      * the r * (1 / 0) will give infinity, which is clamped to 1.0, the correct
2409      * result. But if r = 0.0, then 0 * inf = 0, which is incorrect.
2410      */
2411     shader_addline(buffer, "RCP %s.y, %s.y;\n", dst_name, dst_name);
2412     shader_addline(buffer, "MUL %s.x, %s.x, %s.y;\n", dst_name, dst_name, dst_name);
2413     shader_addline(buffer, "MIN %s.x, %s.x, %s;\n", dst_name, dst_name, one);
2414     shader_addline(buffer, "MAX result.depth, %s.x, %s;\n", dst_name, zero);
2415 }
2416
2417 /** Handles transforming all WINED3DSIO_M?x? opcodes for
2418     Vertex/Pixel shaders to ARB_vertex_program codes */
2419 static void shader_hw_mnxn(const struct wined3d_shader_instruction *ins)
2420 {
2421     int i;
2422     int nComponents = 0;
2423     struct wined3d_shader_dst_param tmp_dst = {{0}};
2424     struct wined3d_shader_src_param tmp_src[2] = {{{0}}};
2425     struct wined3d_shader_instruction tmp_ins;
2426
2427     memset(&tmp_ins, 0, sizeof(tmp_ins));
2428
2429     /* Set constants for the temporary argument */
2430     tmp_ins.ctx = ins->ctx;
2431     tmp_ins.dst_count = 1;
2432     tmp_ins.dst = &tmp_dst;
2433     tmp_ins.src_count = 2;
2434     tmp_ins.src = tmp_src;
2435
2436     switch(ins->handler_idx)
2437     {
2438         case WINED3DSIH_M4x4:
2439             nComponents = 4;
2440             tmp_ins.handler_idx = WINED3DSIH_DP4;
2441             break;
2442         case WINED3DSIH_M4x3:
2443             nComponents = 3;
2444             tmp_ins.handler_idx = WINED3DSIH_DP4;
2445             break;
2446         case WINED3DSIH_M3x4:
2447             nComponents = 4;
2448             tmp_ins.handler_idx = WINED3DSIH_DP3;
2449             break;
2450         case WINED3DSIH_M3x3:
2451             nComponents = 3;
2452             tmp_ins.handler_idx = WINED3DSIH_DP3;
2453             break;
2454         case WINED3DSIH_M3x2:
2455             nComponents = 2;
2456             tmp_ins.handler_idx = WINED3DSIH_DP3;
2457             break;
2458         default:
2459             FIXME("Unhandled opcode %#x\n", ins->handler_idx);
2460             break;
2461     }
2462
2463     tmp_dst = ins->dst[0];
2464     tmp_src[0] = ins->src[0];
2465     tmp_src[1] = ins->src[1];
2466     for (i = 0; i < nComponents; ++i)
2467     {
2468         tmp_dst.write_mask = WINED3DSP_WRITEMASK_0 << i;
2469         shader_hw_map2gl(&tmp_ins);
2470         ++tmp_src[1].reg.idx[0].offset;
2471     }
2472 }
2473
2474 static void shader_hw_rcp(const struct wined3d_shader_instruction *ins)
2475 {
2476     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2477
2478     char dst[50];
2479     char src[50];
2480
2481     shader_arb_get_dst_param(ins, &ins->dst[0], dst); /* Destination */
2482     shader_arb_get_src_param(ins, &ins->src[0], 0, src);
2483     if (ins->src[0].swizzle == WINED3DSP_NOSWIZZLE)
2484     {
2485         /* Dx sdk says .x is used if no swizzle is given, but our test shows that
2486          * .w is used
2487          */
2488         strcat(src, ".w");
2489     }
2490
2491     shader_addline(buffer, "RCP%s %s, %s;\n", shader_arb_get_modifier(ins), dst, src);
2492 }
2493
2494 static void shader_hw_scalar_op(const struct wined3d_shader_instruction *ins)
2495 {
2496     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2497     const char *instruction;
2498
2499     char dst[50];
2500     char src[50];
2501
2502     switch(ins->handler_idx)
2503     {
2504         case WINED3DSIH_RSQ:  instruction = "RSQ"; break;
2505         case WINED3DSIH_RCP:  instruction = "RCP"; break;
2506         case WINED3DSIH_EXP:  instruction = "EX2"; break;
2507         case WINED3DSIH_EXPP: instruction = "EXP"; break;
2508         default: instruction = "";
2509             FIXME("Unhandled opcode %#x\n", ins->handler_idx);
2510             break;
2511     }
2512
2513     shader_arb_get_dst_param(ins, &ins->dst[0], dst); /* Destination */
2514     shader_arb_get_src_param(ins, &ins->src[0], 0, src);
2515     if (ins->src[0].swizzle == WINED3DSP_NOSWIZZLE)
2516     {
2517         /* Dx sdk says .x is used if no swizzle is given, but our test shows that
2518          * .w is used
2519          */
2520         strcat(src, ".w");
2521     }
2522
2523     shader_addline(buffer, "%s%s %s, %s;\n", instruction, shader_arb_get_modifier(ins), dst, src);
2524 }
2525
2526 static void shader_hw_nrm(const struct wined3d_shader_instruction *ins)
2527 {
2528     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2529     char dst_name[50];
2530     char src_name[50];
2531     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2532     BOOL pshader = shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type);
2533     const char *zero = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ZERO);
2534
2535     shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2536     shader_arb_get_src_param(ins, &ins->src[0], 1 /* Use TB */, src_name);
2537
2538     /* In D3D, NRM of a vector with length zero returns zero. Catch this situation, as
2539      * otherwise NRM or RSQ would return NaN */
2540     if(pshader && priv->target_version >= NV3)
2541     {
2542         /* GL_NV_fragment_program2's NRM needs protection against length zero vectors too
2543          *
2544          * TODO: Find out if DP3+NRM+MOV is really faster than DP3+RSQ+MUL
2545          */
2546         shader_addline(buffer, "DP3C TA, %s, %s;\n", src_name, src_name);
2547         shader_addline(buffer, "NRM%s %s, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name);
2548         shader_addline(buffer, "MOV %s (EQ), %s;\n", dst_name, zero);
2549     }
2550     else if(priv->target_version >= NV2)
2551     {
2552         shader_addline(buffer, "DP3C TA.x, %s, %s;\n", src_name, src_name);
2553         shader_addline(buffer, "RSQ TA.x (NE), TA.x;\n");
2554         shader_addline(buffer, "MUL%s %s, %s, TA.x;\n", shader_arb_get_modifier(ins), dst_name,
2555                        src_name);
2556     }
2557     else
2558     {
2559         const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
2560
2561         shader_addline(buffer, "DP3 TA.x, %s, %s;\n", src_name, src_name);
2562         /* Pass any non-zero value to RSQ if the input vector has a length of zero. The
2563          * RSQ result doesn't matter, as long as multiplying it by 0 returns 0.
2564          */
2565         shader_addline(buffer, "SGE TA.y, -TA.x, %s;\n", zero);
2566         shader_addline(buffer, "MAD TA.x, %s, TA.y, TA.x;\n", one);
2567
2568         shader_addline(buffer, "RSQ TA.x, TA.x;\n");
2569         /* dst.w = src[0].w * 1 / (src.x^2 + src.y^2 + src.z^2)^(1/2) according to msdn*/
2570         shader_addline(buffer, "MUL%s %s, %s, TA.x;\n", shader_arb_get_modifier(ins), dst_name,
2571                     src_name);
2572     }
2573 }
2574
2575 static void shader_hw_lrp(const struct wined3d_shader_instruction *ins)
2576 {
2577     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2578     char dst_name[50];
2579     char src_name[3][50];
2580
2581     /* ARB_fragment_program has a convenient LRP instruction */
2582     if(shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type)) {
2583         shader_hw_map2gl(ins);
2584         return;
2585     }
2586
2587     shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2588     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
2589     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
2590     shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
2591
2592     shader_addline(buffer, "SUB TA, %s, %s;\n", src_name[1], src_name[2]);
2593     shader_addline(buffer, "MAD%s %s, %s, TA, %s;\n", shader_arb_get_modifier(ins),
2594                    dst_name, src_name[0], src_name[2]);
2595 }
2596
2597 static void shader_hw_sincos(const struct wined3d_shader_instruction *ins)
2598 {
2599     /* This instruction exists in ARB, but the d3d instruction takes two extra parameters which
2600      * must contain fixed constants. So we need a separate function to filter those constants and
2601      * can't use map2gl
2602      */
2603     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2604     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2605     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2606     char dst_name[50];
2607     char src_name0[50], src_name1[50], src_name2[50];
2608     BOOL is_color;
2609
2610     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name0);
2611     if(shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type)) {
2612         shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2613         /* No modifiers are supported on SCS */
2614         shader_addline(buffer, "SCS %s, %s;\n", dst_name, src_name0);
2615
2616         if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE)
2617         {
2618             shader_arb_get_register_name(ins, &dst->reg, src_name0, &is_color);
2619             shader_addline(buffer, "MOV_SAT %s, %s;\n", dst_name, src_name0);
2620         }
2621     } else if(priv->target_version >= NV2) {
2622         shader_arb_get_register_name(ins, &dst->reg, dst_name, &is_color);
2623
2624         /* Sincos writemask must be .x, .y or .xy */
2625         if(dst->write_mask & WINED3DSP_WRITEMASK_0)
2626             shader_addline(buffer, "COS%s %s.x, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name0);
2627         if(dst->write_mask & WINED3DSP_WRITEMASK_1)
2628             shader_addline(buffer, "SIN%s %s.y, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name0);
2629     } else {
2630         /* Approximate sine and cosine with a taylor series, as per math textbook. The application passes 8
2631          * helper constants(D3DSINCOSCONST1 and D3DSINCOSCONST2) in src1 and src2.
2632          *
2633          * sin(x) = x - x^3/3! + x^5/5! - x^7/7! + ...
2634          * cos(x) = 1 - x^2/2! + x^4/4! - x^6/6! + ...
2635          *
2636          * The constants we get are:
2637          *
2638          *  +1   +1,     -1     -1     +1      +1      -1       -1
2639          *      ---- ,  ---- , ---- , ----- , ----- , ----- , ------
2640          *      1!*2    2!*4   3!*8   4!*16   5!*32   6!*64   7!*128
2641          *
2642          * If used with x^2, x^3, x^4 etc they calculate sin(x/2) and cos(x/2):
2643          *
2644          * (x/2)^2 = x^2 / 4
2645          * (x/2)^3 = x^3 / 8
2646          * (x/2)^4 = x^4 / 16
2647          * (x/2)^5 = x^5 / 32
2648          * etc
2649          *
2650          * To get the final result:
2651          * sin(x) = 2 * sin(x/2) * cos(x/2)
2652          * cos(x) = cos(x/2)^2 - sin(x/2)^2
2653          * (from sin(x+y) and cos(x+y) rules)
2654          *
2655          * As per MSDN, dst.z is undefined after the operation, and so is
2656          * dst.x and dst.y if they're masked out by the writemask. Ie
2657          * sincos dst.y, src1, c0, c1
2658          * returns the sine in dst.y. dst.x and dst.z are undefined, dst.w is not touched. The assembler
2659          * vsa.exe also stops with an error if the dest register is the same register as the source
2660          * register. This means we can use dest.xyz as temporary storage. The assembler vsa.exe output also
2661          * indicates that sincos consumes 8 instruction slots in vs_2_0(and, strangely, in vs_3_0).
2662          */
2663         shader_arb_get_src_param(ins, &ins->src[1], 1, src_name1);
2664         shader_arb_get_src_param(ins, &ins->src[2], 2, src_name2);
2665         shader_arb_get_register_name(ins, &dst->reg, dst_name, &is_color);
2666
2667         shader_addline(buffer, "MUL %s.x, %s, %s;\n", dst_name, src_name0, src_name0);  /* x ^ 2 */
2668         shader_addline(buffer, "MUL TA.y, %s.x, %s;\n", dst_name, src_name0);           /* x ^ 3 */
2669         shader_addline(buffer, "MUL %s.y, TA.y, %s;\n", dst_name, src_name0);           /* x ^ 4 */
2670         shader_addline(buffer, "MUL TA.z, %s.y, %s;\n", dst_name, src_name0);           /* x ^ 5 */
2671         shader_addline(buffer, "MUL %s.z, TA.z, %s;\n", dst_name, src_name0);           /* x ^ 6 */
2672         shader_addline(buffer, "MUL TA.w, %s.z, %s;\n", dst_name, src_name0);           /* x ^ 7 */
2673
2674         /* sin(x/2)
2675          *
2676          * Unfortunately we don't get the constants in a DP4-capable form. Is there a way to
2677          * properly merge that with MULs in the code above?
2678          * The swizzles .yz and xw however fit into the .yzxw swizzle added to ps_2_0. Maybe
2679          * we can merge the sine and cosine MAD rows to calculate them together.
2680          */
2681         shader_addline(buffer, "MUL TA.x, %s, %s.w;\n", src_name0, src_name2); /* x^1, +1/(1!*2) */
2682         shader_addline(buffer, "MAD TA.x, TA.y, %s.x, TA.x;\n", src_name2); /* -1/(3!*8) */
2683         shader_addline(buffer, "MAD TA.x, TA.z, %s.w, TA.x;\n", src_name1); /* +1/(5!*32) */
2684         shader_addline(buffer, "MAD TA.x, TA.w, %s.x, TA.x;\n", src_name1); /* -1/(7!*128) */
2685
2686         /* cos(x/2) */
2687         shader_addline(buffer, "MAD TA.y, %s.x, %s.y, %s.z;\n", dst_name, src_name2, src_name2); /* -1/(2!*4), +1.0 */
2688         shader_addline(buffer, "MAD TA.y, %s.y, %s.z, TA.y;\n", dst_name, src_name1); /* +1/(4!*16) */
2689         shader_addline(buffer, "MAD TA.y, %s.z, %s.y, TA.y;\n", dst_name, src_name1); /* -1/(6!*64) */
2690
2691         if(dst->write_mask & WINED3DSP_WRITEMASK_0) {
2692             /* cos x */
2693             shader_addline(buffer, "MUL TA.z, TA.y, TA.y;\n");
2694             shader_addline(buffer, "MAD %s.x, -TA.x, TA.x, TA.z;\n", dst_name);
2695         }
2696         if(dst->write_mask & WINED3DSP_WRITEMASK_1) {
2697             /* sin x */
2698             shader_addline(buffer, "MUL %s.y, TA.x, TA.y;\n", dst_name);
2699             shader_addline(buffer, "ADD %s.y, %s.y, %s.y;\n", dst_name, dst_name, dst_name);
2700         }
2701     }
2702 }
2703
2704 static void shader_hw_sgn(const struct wined3d_shader_instruction *ins)
2705 {
2706     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2707     char dst_name[50];
2708     char src_name[50];
2709     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
2710
2711     shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2712     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name);
2713
2714     /* SGN is only valid in vertex shaders */
2715     if(ctx->target_version >= NV2) {
2716         shader_addline(buffer, "SSG%s %s, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name);
2717         return;
2718     }
2719
2720     /* If SRC > 0.0, -SRC < SRC = TRUE, otherwise false.
2721      * if SRC < 0.0,  SRC < -SRC = TRUE. If neither is true, src = 0.0
2722      */
2723     if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE) {
2724         shader_addline(buffer, "SLT %s, -%s, %s;\n", dst_name, src_name, src_name);
2725     } else {
2726         /* src contains TA? Write to the dest first. This won't overwrite our destination.
2727          * Then use TA, and calculate the final result
2728          *
2729          * Not reading from TA? Store the first result in TA to avoid overwriting the
2730          * destination if src reg = dst reg
2731          */
2732         if(strstr(src_name, "TA"))
2733         {
2734             shader_addline(buffer, "SLT %s,  %s, -%s;\n", dst_name, src_name, src_name);
2735             shader_addline(buffer, "SLT TA, -%s, %s;\n", src_name, src_name);
2736             shader_addline(buffer, "ADD %s, %s, -TA;\n", dst_name, dst_name);
2737         }
2738         else
2739         {
2740             shader_addline(buffer, "SLT TA, -%s, %s;\n", src_name, src_name);
2741             shader_addline(buffer, "SLT %s,  %s, -%s;\n", dst_name, src_name, src_name);
2742             shader_addline(buffer, "ADD %s, TA, -%s;\n", dst_name, dst_name);
2743         }
2744     }
2745 }
2746
2747 static void shader_hw_dsy(const struct wined3d_shader_instruction *ins)
2748 {
2749     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2750     char src[50];
2751     char dst[50];
2752     char dst_name[50];
2753     BOOL is_color;
2754
2755     shader_arb_get_dst_param(ins, &ins->dst[0], dst);
2756     shader_arb_get_src_param(ins, &ins->src[0], 0, src);
2757     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2758
2759     shader_addline(buffer, "DDY %s, %s;\n", dst, src);
2760     shader_addline(buffer, "MUL%s %s, %s, ycorrection.y;\n", shader_arb_get_modifier(ins), dst, dst_name);
2761 }
2762
2763 static DWORD abs_modifier(DWORD mod, BOOL *need_abs)
2764 {
2765     *need_abs = FALSE;
2766
2767     switch(mod)
2768     {
2769         case WINED3DSPSM_NONE:      return WINED3DSPSM_ABS;
2770         case WINED3DSPSM_NEG:       return WINED3DSPSM_ABS;
2771         case WINED3DSPSM_BIAS:      *need_abs = TRUE; return WINED3DSPSM_BIAS;
2772         case WINED3DSPSM_BIASNEG:   *need_abs = TRUE; return WINED3DSPSM_BIASNEG;
2773         case WINED3DSPSM_SIGN:      *need_abs = TRUE; return WINED3DSPSM_SIGN;
2774         case WINED3DSPSM_SIGNNEG:   *need_abs = TRUE; return WINED3DSPSM_SIGNNEG;
2775         case WINED3DSPSM_COMP:      *need_abs = TRUE; return WINED3DSPSM_COMP;
2776         case WINED3DSPSM_X2:        *need_abs = TRUE; return WINED3DSPSM_X2;
2777         case WINED3DSPSM_X2NEG:     *need_abs = TRUE; return WINED3DSPSM_X2NEG;
2778         case WINED3DSPSM_DZ:        *need_abs = TRUE; return WINED3DSPSM_DZ;
2779         case WINED3DSPSM_DW:        *need_abs = TRUE; return WINED3DSPSM_DW;
2780         case WINED3DSPSM_ABS:       return WINED3DSPSM_ABS;
2781         case WINED3DSPSM_ABSNEG:    return WINED3DSPSM_ABS;
2782     }
2783     FIXME("Unknown modifier %u\n", mod);
2784     return mod;
2785 }
2786
2787 static void shader_hw_log(const struct wined3d_shader_instruction *ins)
2788 {
2789     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2790     char src0[50], dst[50];
2791     struct wined3d_shader_src_param src0_copy = ins->src[0];
2792     BOOL need_abs = FALSE;
2793     const char *instr;
2794
2795     switch(ins->handler_idx)
2796     {
2797         case WINED3DSIH_LOG:  instr = "LG2"; break;
2798         case WINED3DSIH_LOGP: instr = "LOG"; break;
2799         default:
2800             ERR("Unexpected instruction %d\n", ins->handler_idx);
2801             return;
2802     }
2803
2804     /* LOG and LOGP operate on the absolute value of the input */
2805     src0_copy.modifiers = abs_modifier(src0_copy.modifiers, &need_abs);
2806
2807     shader_arb_get_dst_param(ins, &ins->dst[0], dst);
2808     shader_arb_get_src_param(ins, &src0_copy, 0, src0);
2809
2810     if(need_abs)
2811     {
2812         shader_addline(buffer, "ABS TA, %s;\n", src0);
2813         shader_addline(buffer, "%s%s %s, TA;\n", instr, shader_arb_get_modifier(ins), dst);
2814     }
2815     else
2816     {
2817         shader_addline(buffer, "%s%s %s, %s;\n", instr, shader_arb_get_modifier(ins), dst, src0);
2818     }
2819 }
2820
2821 static void shader_hw_pow(const struct wined3d_shader_instruction *ins)
2822 {
2823     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2824     char src0[50], src1[50], dst[50];
2825     struct wined3d_shader_src_param src0_copy = ins->src[0];
2826     BOOL need_abs = FALSE;
2827     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2828     const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
2829
2830     /* POW operates on the absolute value of the input */
2831     src0_copy.modifiers = abs_modifier(src0_copy.modifiers, &need_abs);
2832
2833     shader_arb_get_dst_param(ins, &ins->dst[0], dst);
2834     shader_arb_get_src_param(ins, &src0_copy, 0, src0);
2835     shader_arb_get_src_param(ins, &ins->src[1], 1, src1);
2836
2837     if (need_abs)
2838         shader_addline(buffer, "ABS TA.x, %s;\n", src0);
2839     else
2840         shader_addline(buffer, "MOV TA.x, %s;\n", src0);
2841
2842     if (priv->target_version >= NV2)
2843     {
2844         shader_addline(buffer, "MOVC TA.y, %s;\n", src1);
2845         shader_addline(buffer, "POW%s %s, TA.x, TA.y;\n", shader_arb_get_modifier(ins), dst);
2846         shader_addline(buffer, "MOV %s (EQ.y), %s;\n", dst, one);
2847     }
2848     else
2849     {
2850         const char *zero = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ZERO);
2851         const char *flt_eps = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_EPS);
2852
2853         shader_addline(buffer, "ABS TA.y, %s;\n", src1);
2854         shader_addline(buffer, "SGE TA.y, -TA.y, %s;\n", zero);
2855         /* Possibly add flt_eps to avoid getting float special values */
2856         shader_addline(buffer, "MAD TA.z, TA.y, %s, %s;\n", flt_eps, src1);
2857         shader_addline(buffer, "POW%s TA.x, TA.x, TA.z;\n", shader_arb_get_modifier(ins));
2858         shader_addline(buffer, "MAD TA.x, -TA.x, TA.y, TA.x;\n");
2859         shader_addline(buffer, "MAD %s, TA.y, %s, TA.x;\n", dst, one);
2860     }
2861 }
2862
2863 static void shader_hw_loop(const struct wined3d_shader_instruction *ins)
2864 {
2865     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2866     char src_name[50];
2867     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2868
2869     /* src0 is aL */
2870     shader_arb_get_src_param(ins, &ins->src[1], 0, src_name);
2871
2872     if(vshader)
2873     {
2874         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2875         struct list *e = list_head(&priv->control_frames);
2876         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2877
2878         if(priv->loop_depth > 1) shader_addline(buffer, "PUSHA aL;\n");
2879         /* The constant loader makes sure to load -1 into iX.w */
2880         shader_addline(buffer, "ARLC aL, %s.xywz;\n", src_name);
2881         shader_addline(buffer, "BRA loop_%u_end (LE.x);\n", control_frame->no.loop);
2882         shader_addline(buffer, "loop_%u_start:\n", control_frame->no.loop);
2883     }
2884     else
2885     {
2886         shader_addline(buffer, "LOOP %s;\n", src_name);
2887     }
2888 }
2889
2890 static void shader_hw_rep(const struct wined3d_shader_instruction *ins)
2891 {
2892     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2893     char src_name[50];
2894     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2895
2896     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name);
2897
2898     /* The constant loader makes sure to load -1 into iX.w */
2899     if(vshader)
2900     {
2901         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2902         struct list *e = list_head(&priv->control_frames);
2903         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2904
2905         if(priv->loop_depth > 1) shader_addline(buffer, "PUSHA aL;\n");
2906
2907         shader_addline(buffer, "ARLC aL, %s.xywz;\n", src_name);
2908         shader_addline(buffer, "BRA loop_%u_end (LE.x);\n", control_frame->no.loop);
2909         shader_addline(buffer, "loop_%u_start:\n", control_frame->no.loop);
2910     }
2911     else
2912     {
2913         shader_addline(buffer, "REP %s;\n", src_name);
2914     }
2915 }
2916
2917 static void shader_hw_endloop(const struct wined3d_shader_instruction *ins)
2918 {
2919     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2920     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2921
2922     if(vshader)
2923     {
2924         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2925         struct list *e = list_head(&priv->control_frames);
2926         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2927
2928         shader_addline(buffer, "ARAC aL.xy, aL;\n");
2929         shader_addline(buffer, "BRA loop_%u_start (GT.x);\n", control_frame->no.loop);
2930         shader_addline(buffer, "loop_%u_end:\n", control_frame->no.loop);
2931
2932         if(priv->loop_depth > 1) shader_addline(buffer, "POPA aL;\n");
2933     }
2934     else
2935     {
2936         shader_addline(buffer, "ENDLOOP;\n");
2937     }
2938 }
2939
2940 static void shader_hw_endrep(const struct wined3d_shader_instruction *ins)
2941 {
2942     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2943     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2944
2945     if(vshader)
2946     {
2947         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2948         struct list *e = list_head(&priv->control_frames);
2949         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2950
2951         shader_addline(buffer, "ARAC aL.xy, aL;\n");
2952         shader_addline(buffer, "BRA loop_%u_start (GT.x);\n", control_frame->no.loop);
2953         shader_addline(buffer, "loop_%u_end:\n", control_frame->no.loop);
2954
2955         if(priv->loop_depth > 1) shader_addline(buffer, "POPA aL;\n");
2956     }
2957     else
2958     {
2959         shader_addline(buffer, "ENDREP;\n");
2960     }
2961 }
2962
2963 static const struct control_frame *find_last_loop(const struct shader_arb_ctx_priv *priv)
2964 {
2965     struct control_frame *control_frame;
2966
2967     LIST_FOR_EACH_ENTRY(control_frame, &priv->control_frames, struct control_frame, entry)
2968     {
2969         if(control_frame->type == LOOP || control_frame->type == REP) return control_frame;
2970     }
2971     ERR("Could not find loop for break\n");
2972     return NULL;
2973 }
2974
2975 static void shader_hw_break(const struct wined3d_shader_instruction *ins)
2976 {
2977     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2978     const struct control_frame *control_frame = find_last_loop(ins->ctx->backend_data);
2979     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2980
2981     if(vshader)
2982     {
2983         shader_addline(buffer, "BRA loop_%u_end;\n", control_frame->no.loop);
2984     }
2985     else
2986     {
2987         shader_addline(buffer, "BRK;\n");
2988     }
2989 }
2990
2991 static const char *get_compare(enum wined3d_shader_rel_op op)
2992 {
2993     switch (op)
2994     {
2995         case WINED3D_SHADER_REL_OP_GT: return "GT";
2996         case WINED3D_SHADER_REL_OP_EQ: return "EQ";
2997         case WINED3D_SHADER_REL_OP_GE: return "GE";
2998         case WINED3D_SHADER_REL_OP_LT: return "LT";
2999         case WINED3D_SHADER_REL_OP_NE: return "NE";
3000         case WINED3D_SHADER_REL_OP_LE: return "LE";
3001         default:
3002             FIXME("Unrecognized operator %#x.\n", op);
3003             return "(\?\?)";
3004     }
3005 }
3006
3007 static enum wined3d_shader_rel_op invert_compare(enum wined3d_shader_rel_op op)
3008 {
3009     switch (op)
3010     {
3011         case WINED3D_SHADER_REL_OP_GT: return WINED3D_SHADER_REL_OP_LE;
3012         case WINED3D_SHADER_REL_OP_EQ: return WINED3D_SHADER_REL_OP_NE;
3013         case WINED3D_SHADER_REL_OP_GE: return WINED3D_SHADER_REL_OP_LT;
3014         case WINED3D_SHADER_REL_OP_LT: return WINED3D_SHADER_REL_OP_GE;
3015         case WINED3D_SHADER_REL_OP_NE: return WINED3D_SHADER_REL_OP_EQ;
3016         case WINED3D_SHADER_REL_OP_LE: return WINED3D_SHADER_REL_OP_GT;
3017         default:
3018             FIXME("Unrecognized operator %#x.\n", op);
3019             return -1;
3020     }
3021 }
3022
3023 static void shader_hw_breakc(const struct wined3d_shader_instruction *ins)
3024 {
3025     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3026     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
3027     const struct control_frame *control_frame = find_last_loop(ins->ctx->backend_data);
3028     char src_name0[50];
3029     char src_name1[50];
3030     const char *comp = get_compare(ins->flags);
3031
3032     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name0);
3033     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name1);
3034
3035     if(vshader)
3036     {
3037         /* SUBC CC, src0, src1" works only in pixel shaders, so use TA to throw
3038          * away the subtraction result
3039          */
3040         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
3041         shader_addline(buffer, "BRA loop_%u_end (%s.x);\n", control_frame->no.loop, comp);
3042     }
3043     else
3044     {
3045         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
3046         shader_addline(buffer, "BRK (%s.x);\n", comp);
3047     }
3048 }
3049
3050 static void shader_hw_ifc(const struct wined3d_shader_instruction *ins)
3051 {
3052     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3053     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3054     struct list *e = list_head(&priv->control_frames);
3055     struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
3056     const char *comp;
3057     char src_name0[50];
3058     char src_name1[50];
3059     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
3060
3061     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name0);
3062     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name1);
3063
3064     if(vshader)
3065     {
3066         /* Invert the flag. We jump to the else label if the condition is NOT true */
3067         comp = get_compare(invert_compare(ins->flags));
3068         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
3069         shader_addline(buffer, "BRA ifc_%u_else (%s.x);\n", control_frame->no.ifc, comp);
3070     }
3071     else
3072     {
3073         comp = get_compare(ins->flags);
3074         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
3075         shader_addline(buffer, "IF %s.x;\n", comp);
3076     }
3077 }
3078
3079 static void shader_hw_else(const struct wined3d_shader_instruction *ins)
3080 {
3081     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3082     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3083     struct list *e = list_head(&priv->control_frames);
3084     struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
3085     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
3086
3087     if(vshader)
3088     {
3089         shader_addline(buffer, "BRA ifc_%u_endif;\n", control_frame->no.ifc);
3090         shader_addline(buffer, "ifc_%u_else:\n", control_frame->no.ifc);
3091         control_frame->had_else = TRUE;
3092     }
3093     else
3094     {
3095         shader_addline(buffer, "ELSE;\n");
3096     }
3097 }
3098
3099 static void shader_hw_endif(const struct wined3d_shader_instruction *ins)
3100 {
3101     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3102     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3103     struct list *e = list_head(&priv->control_frames);
3104     struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
3105     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
3106
3107     if(vshader)
3108     {
3109         if(control_frame->had_else)
3110         {
3111             shader_addline(buffer, "ifc_%u_endif:\n", control_frame->no.ifc);
3112         }
3113         else
3114         {
3115             shader_addline(buffer, "#No else branch. else is endif\n");
3116             shader_addline(buffer, "ifc_%u_else:\n", control_frame->no.ifc);
3117         }
3118     }
3119     else
3120     {
3121         shader_addline(buffer, "ENDIF;\n");
3122     }
3123 }
3124
3125 static void shader_hw_texldd(const struct wined3d_shader_instruction *ins)
3126 {
3127     DWORD sampler_idx = ins->src[1].reg.idx[0].offset;
3128     char reg_dest[40];
3129     char reg_src[3][40];
3130     WORD flags = TEX_DERIV;
3131
3132     shader_arb_get_dst_param(ins, &ins->dst[0], reg_dest);
3133     shader_arb_get_src_param(ins, &ins->src[0], 0, reg_src[0]);
3134     shader_arb_get_src_param(ins, &ins->src[2], 1, reg_src[1]);
3135     shader_arb_get_src_param(ins, &ins->src[3], 2, reg_src[2]);
3136
3137     if (ins->flags & WINED3DSI_TEXLD_PROJECT) flags |= TEX_PROJ;
3138     if (ins->flags & WINED3DSI_TEXLD_BIAS) flags |= TEX_BIAS;
3139
3140     shader_hw_sample(ins, sampler_idx, reg_dest, reg_src[0], flags, reg_src[1], reg_src[2]);
3141 }
3142
3143 static void shader_hw_texldl(const struct wined3d_shader_instruction *ins)
3144 {
3145     DWORD sampler_idx = ins->src[1].reg.idx[0].offset;
3146     char reg_dest[40];
3147     char reg_coord[40];
3148     WORD flags = TEX_LOD;
3149
3150     shader_arb_get_dst_param(ins, &ins->dst[0], reg_dest);
3151     shader_arb_get_src_param(ins, &ins->src[0], 0, reg_coord);
3152
3153     if (ins->flags & WINED3DSI_TEXLD_PROJECT) flags |= TEX_PROJ;
3154     if (ins->flags & WINED3DSI_TEXLD_BIAS) flags |= TEX_BIAS;
3155
3156     shader_hw_sample(ins, sampler_idx, reg_dest, reg_coord, flags, NULL, NULL);
3157 }
3158
3159 static void shader_hw_label(const struct wined3d_shader_instruction *ins)
3160 {
3161     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3162     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3163
3164     priv->in_main_func = FALSE;
3165     /* Call instructions activate the NV extensions, not labels and rets. If there is an uncalled
3166      * subroutine, don't generate a label that will make GL complain
3167      */
3168     if(priv->target_version == ARB) return;
3169
3170     shader_addline(buffer, "l%u:\n", ins->src[0].reg.idx[0].offset);
3171 }
3172
3173 static void vshader_add_footer(struct shader_arb_ctx_priv *priv_ctx,
3174         const struct arb_vshader_private *shader_data, const struct arb_vs_compile_args *args,
3175         const struct wined3d_shader_reg_maps *reg_maps, const struct wined3d_gl_info *gl_info,
3176         struct wined3d_shader_buffer *buffer)
3177 {
3178     unsigned int i;
3179
3180     /* The D3DRS_FOGTABLEMODE render state defines if the shader-generated fog coord is used
3181      * or if the fragment depth is used. If the fragment depth is used(FOGTABLEMODE != NONE),
3182      * the fog frag coord is thrown away. If the fog frag coord is used, but not written by
3183      * the shader, it is set to 0.0(fully fogged, since start = 1.0, end = 0.0)
3184      */
3185     if (args->super.fog_src == VS_FOG_Z)
3186     {
3187         shader_addline(buffer, "MOV result.fogcoord, TMP_OUT.z;\n");
3188     }
3189     else
3190     {
3191         if (!reg_maps->fog)
3192         {
3193             /* posFixup.x is always 1.0, so we can safely use it */
3194             shader_addline(buffer, "ADD result.fogcoord, posFixup.x, -posFixup.x;\n");
3195         }
3196         else
3197         {
3198             /* Clamp fogcoord */
3199             const char *zero = arb_get_helper_value(reg_maps->shader_version.type, ARB_ZERO);
3200             const char *one = arb_get_helper_value(reg_maps->shader_version.type, ARB_ONE);
3201
3202             shader_addline(buffer, "MIN TMP_FOGCOORD.x, TMP_FOGCOORD.x, %s;\n", one);
3203             shader_addline(buffer, "MAX result.fogcoord.x, TMP_FOGCOORD.x, %s;\n", zero);
3204         }
3205     }
3206
3207     /* Clipplanes are always stored without y inversion */
3208     if (use_nv_clip(gl_info) && priv_ctx->target_version >= NV2)
3209     {
3210         if (args->super.clip_enabled)
3211         {
3212             for (i = 0; i < priv_ctx->vs_clipplanes; i++)
3213             {
3214                 shader_addline(buffer, "DP4 result.clip[%u].x, TMP_OUT, state.clip[%u].plane;\n", i, i);
3215             }
3216         }
3217     }
3218     else if (args->clip.boolclip.clip_texcoord)
3219     {
3220         unsigned int cur_clip = 0;
3221         char component[4] = {'x', 'y', 'z', 'w'};
3222         const char *zero = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_ZERO);
3223
3224         for (i = 0; i < gl_info->limits.clipplanes; ++i)
3225         {
3226             if (args->clip.boolclip.clipplane_mask & (1 << i))
3227             {
3228                 shader_addline(buffer, "DP4 TA.%c, TMP_OUT, state.clip[%u].plane;\n",
3229                                component[cur_clip++], i);
3230             }
3231         }
3232         switch (cur_clip)
3233         {
3234             case 0:
3235                 shader_addline(buffer, "MOV TA, %s;\n", zero);
3236                 break;
3237             case 1:
3238                 shader_addline(buffer, "MOV TA.yzw, %s;\n", zero);
3239                 break;
3240             case 2:
3241                 shader_addline(buffer, "MOV TA.zw, %s;\n", zero);
3242                 break;
3243             case 3:
3244                 shader_addline(buffer, "MOV TA.w, %s;\n", zero);
3245                 break;
3246         }
3247         shader_addline(buffer, "MOV result.texcoord[%u], TA;\n",
3248                        args->clip.boolclip.clip_texcoord - 1);
3249     }
3250
3251     /* Write the final position.
3252      *
3253      * OpenGL coordinates specify the center of the pixel while d3d coords specify
3254      * the corner. The offsets are stored in z and w in posFixup. posFixup.y contains
3255      * 1.0 or -1.0 to turn the rendering upside down for offscreen rendering. PosFixup.x
3256      * contains 1.0 to allow a mad, but arb vs swizzles are too restricted for that.
3257      */
3258     shader_addline(buffer, "MUL TA, posFixup, TMP_OUT.w;\n");
3259     shader_addline(buffer, "ADD TMP_OUT.x, TMP_OUT.x, TA.z;\n");
3260     shader_addline(buffer, "MAD TMP_OUT.y, TMP_OUT.y, posFixup.y, TA.w;\n");
3261
3262     /* Z coord [0;1]->[-1;1] mapping, see comment in transform_projection in state.c
3263      * and the glsl equivalent
3264      */
3265     if (need_helper_const(shader_data, reg_maps, gl_info))
3266     {
3267         const char *two = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_TWO);
3268         shader_addline(buffer, "MAD TMP_OUT.z, TMP_OUT.z, %s, -TMP_OUT.w;\n", two);
3269     }
3270     else
3271     {
3272         shader_addline(buffer, "ADD TMP_OUT.z, TMP_OUT.z, TMP_OUT.z;\n");
3273         shader_addline(buffer, "ADD TMP_OUT.z, TMP_OUT.z, -TMP_OUT.w;\n");
3274     }
3275
3276     shader_addline(buffer, "MOV result.position, TMP_OUT;\n");
3277
3278     priv_ctx->footer_written = TRUE;
3279 }
3280
3281 static void shader_hw_ret(const struct wined3d_shader_instruction *ins)
3282 {
3283     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3284     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3285     const struct wined3d_shader *shader = ins->ctx->shader;
3286     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
3287
3288     if(priv->target_version == ARB) return;
3289
3290     if(vshader)
3291     {
3292         if (priv->in_main_func) vshader_add_footer(priv, shader->backend_data,
3293                 priv->cur_vs_args, ins->ctx->reg_maps, ins->ctx->gl_info, buffer);
3294     }
3295
3296     shader_addline(buffer, "RET;\n");
3297 }
3298
3299 static void shader_hw_call(const struct wined3d_shader_instruction *ins)
3300 {
3301     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3302     shader_addline(buffer, "CAL l%u;\n", ins->src[0].reg.idx[0].offset);
3303 }
3304
3305 /* GL locking is done by the caller */
3306 static GLuint create_arb_blt_vertex_program(const struct wined3d_gl_info *gl_info)
3307 {
3308     GLuint program_id = 0;
3309     GLint pos;
3310
3311     const char *blt_vprogram =
3312         "!!ARBvp1.0\n"
3313         "PARAM c[1] = { { 1, 0.5 } };\n"
3314         "MOV result.position, vertex.position;\n"
3315         "MOV result.color, c[0].x;\n"
3316         "MOV result.texcoord[0], vertex.texcoord[0];\n"
3317         "END\n";
3318
3319     GL_EXTCALL(glGenProgramsARB(1, &program_id));
3320     GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, program_id));
3321     GL_EXTCALL(glProgramStringARB(GL_VERTEX_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
3322             strlen(blt_vprogram), blt_vprogram));
3323     checkGLcall("glProgramStringARB()");
3324
3325     gl_info->gl_ops.gl.p_glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
3326     if (pos != -1)
3327     {
3328         FIXME("Vertex program error at position %d: %s\n\n", pos,
3329             debugstr_a((const char *)gl_info->gl_ops.gl.p_glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
3330         shader_arb_dump_program_source(blt_vprogram);
3331     }
3332     else
3333     {
3334         GLint native;
3335
3336         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
3337         checkGLcall("glGetProgramivARB()");
3338         if (!native) WARN("Program exceeds native resource limits.\n");
3339     }
3340
3341     return program_id;
3342 }
3343
3344 /* GL locking is done by the caller */
3345 static GLuint create_arb_blt_fragment_program(const struct wined3d_gl_info *gl_info,
3346         enum tex_types tex_type, BOOL masked)
3347 {
3348     GLuint program_id = 0;
3349     const char *fprogram;
3350     GLint pos;
3351
3352     static const char * const blt_fprograms_full[tex_type_count] =
3353     {
3354         /* tex_1d */
3355         NULL,
3356         /* tex_2d */
3357         "!!ARBfp1.0\n"
3358         "TEMP R0;\n"
3359         "TEX R0.x, fragment.texcoord[0], texture[0], 2D;\n"
3360         "MOV result.depth.z, R0.x;\n"
3361         "END\n",
3362         /* tex_3d */
3363         NULL,
3364         /* tex_cube */
3365         "!!ARBfp1.0\n"
3366         "TEMP R0;\n"
3367         "TEX R0.x, fragment.texcoord[0], texture[0], CUBE;\n"
3368         "MOV result.depth.z, R0.x;\n"
3369         "END\n",
3370         /* tex_rect */
3371         "!!ARBfp1.0\n"
3372         "TEMP R0;\n"
3373         "TEX R0.x, fragment.texcoord[0], texture[0], RECT;\n"
3374         "MOV result.depth.z, R0.x;\n"
3375         "END\n",
3376     };
3377
3378     static const char * const blt_fprograms_masked[tex_type_count] =
3379     {
3380         /* tex_1d */
3381         NULL,
3382         /* tex_2d */
3383         "!!ARBfp1.0\n"
3384         "PARAM mask = program.local[0];\n"
3385         "TEMP R0;\n"
3386         "SLT R0.xy, fragment.position, mask.zwzw;\n"
3387         "MUL R0.x, R0.x, R0.y;\n"
3388         "KIL -R0.x;\n"
3389         "TEX R0.x, fragment.texcoord[0], texture[0], 2D;\n"
3390         "MOV result.depth.z, R0.x;\n"
3391         "END\n",
3392         /* tex_3d */
3393         NULL,
3394         /* tex_cube */
3395         "!!ARBfp1.0\n"
3396         "PARAM mask = program.local[0];\n"
3397         "TEMP R0;\n"
3398         "SLT R0.xy, fragment.position, mask.zwzw;\n"
3399         "MUL R0.x, R0.x, R0.y;\n"
3400         "KIL -R0.x;\n"
3401         "TEX R0.x, fragment.texcoord[0], texture[0], CUBE;\n"
3402         "MOV result.depth.z, R0.x;\n"
3403         "END\n",
3404         /* tex_rect */
3405         "!!ARBfp1.0\n"
3406         "PARAM mask = program.local[0];\n"
3407         "TEMP R0;\n"
3408         "SLT R0.xy, fragment.position, mask.zwzw;\n"
3409         "MUL R0.x, R0.x, R0.y;\n"
3410         "KIL -R0.x;\n"
3411         "TEX R0.x, fragment.texcoord[0], texture[0], RECT;\n"
3412         "MOV result.depth.z, R0.x;\n"
3413         "END\n",
3414     };
3415
3416     fprogram = masked ? blt_fprograms_masked[tex_type] : blt_fprograms_full[tex_type];
3417     if (!fprogram)
3418     {
3419         FIXME("tex_type %#x not supported, falling back to tex_2d\n", tex_type);
3420         tex_type = tex_2d;
3421         fprogram = masked ? blt_fprograms_masked[tex_type] : blt_fprograms_full[tex_type];
3422     }
3423
3424     GL_EXTCALL(glGenProgramsARB(1, &program_id));
3425     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, program_id));
3426     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB, strlen(fprogram), fprogram));
3427     checkGLcall("glProgramStringARB()");
3428
3429     gl_info->gl_ops.gl.p_glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
3430     if (pos != -1)
3431     {
3432         FIXME("Fragment program error at position %d: %s\n\n", pos,
3433             debugstr_a((const char *)gl_info->gl_ops.gl.p_glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
3434         shader_arb_dump_program_source(fprogram);
3435     }
3436     else
3437     {
3438         GLint native;
3439
3440         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
3441         checkGLcall("glGetProgramivARB()");
3442         if (!native) WARN("Program exceeds native resource limits.\n");
3443     }
3444
3445     return program_id;
3446 }
3447
3448 static void arbfp_add_sRGB_correction(struct wined3d_shader_buffer *buffer, const char *fragcolor,
3449         const char *tmp1, const char *tmp2, const char *tmp3, const char *tmp4, BOOL condcode)
3450 {
3451     /* Perform sRGB write correction. See GLX_EXT_framebuffer_sRGB */
3452
3453     if(condcode)
3454     {
3455         /* Sigh. MOVC CC doesn't work, so use one of the temps as dummy dest */
3456         shader_addline(buffer, "SUBC %s, %s.x, srgb_consts1.y;\n", tmp1, fragcolor);
3457         /* Calculate the > 0.0031308 case */
3458         shader_addline(buffer, "POW %s.x (GE), %s.x, srgb_consts1.z;\n", fragcolor, fragcolor);
3459         shader_addline(buffer, "POW %s.y (GE), %s.y, srgb_consts1.z;\n", fragcolor, fragcolor);
3460         shader_addline(buffer, "POW %s.z (GE), %s.z, srgb_consts1.z;\n", fragcolor, fragcolor);
3461         shader_addline(buffer, "MUL %s.xyz (GE), %s, srgb_consts1.w;\n", fragcolor, fragcolor);
3462         shader_addline(buffer, "SUB %s.xyz (GE), %s, srgb_consts2.x;\n", fragcolor, fragcolor);
3463         /* Calculate the < case */
3464         shader_addline(buffer, "MUL %s.xyz (LT), srgb_consts1.x, %s;\n", fragcolor, fragcolor);
3465     }
3466     else
3467     {
3468         /* Calculate the > 0.0031308 case */
3469         shader_addline(buffer, "POW %s.x, %s.x, srgb_consts1.z;\n", tmp1, fragcolor);
3470         shader_addline(buffer, "POW %s.y, %s.y, srgb_consts1.z;\n", tmp1, fragcolor);
3471         shader_addline(buffer, "POW %s.z, %s.z, srgb_consts1.z;\n", tmp1, fragcolor);
3472         shader_addline(buffer, "MUL %s, %s, srgb_consts1.w;\n", tmp1, tmp1);
3473         shader_addline(buffer, "SUB %s, %s, srgb_consts2.x;\n", tmp1, tmp1);
3474         /* Calculate the < case */
3475         shader_addline(buffer, "MUL %s, srgb_consts1.x, %s;\n", tmp2, fragcolor);
3476         /* Get 1.0 / 0.0 masks for > 0.0031308 and < 0.0031308 */
3477         shader_addline(buffer, "SLT %s, srgb_consts1.y, %s;\n", tmp3, fragcolor);
3478         shader_addline(buffer, "SGE %s, srgb_consts1.y, %s;\n", tmp4, fragcolor);
3479         /* Store the components > 0.0031308 in the destination */
3480         shader_addline(buffer, "MUL %s.xyz, %s, %s;\n", fragcolor, tmp1, tmp3);
3481         /* Add the components that are < 0.0031308 */
3482         shader_addline(buffer, "MAD %s.xyz, %s, %s, %s;\n", fragcolor, tmp2, tmp4, fragcolor);
3483         /* Move everything into result.color at once. Nvidia hardware cannot handle partial
3484         * result.color writes(.rgb first, then .a), or handle overwriting already written
3485         * components. The assembler uses a temporary register in this case, which is usually
3486         * not allocated from one of our registers that were used earlier.
3487         */
3488     }
3489     /* [0.0;1.0] clamping. Not needed, this is done implicitly */
3490 }
3491
3492 static const DWORD *find_loop_control_values(const struct wined3d_shader *shader, DWORD idx)
3493 {
3494     const struct wined3d_shader_lconst *constant;
3495
3496     LIST_FOR_EACH_ENTRY(constant, &shader->constantsI, struct wined3d_shader_lconst, entry)
3497     {
3498         if (constant->idx == idx)
3499         {
3500             return constant->value;
3501         }
3502     }
3503     return NULL;
3504 }
3505
3506 static void init_ps_input(const struct wined3d_shader *shader,
3507         const struct arb_ps_compile_args *args, struct shader_arb_ctx_priv *priv)
3508 {
3509     static const char * const texcoords[8] =
3510     {
3511         "fragment.texcoord[0]", "fragment.texcoord[1]", "fragment.texcoord[2]", "fragment.texcoord[3]",
3512         "fragment.texcoord[4]", "fragment.texcoord[5]", "fragment.texcoord[6]", "fragment.texcoord[7]"
3513     };
3514     unsigned int i;
3515     const struct wined3d_shader_signature_element *sig = shader->input_signature;
3516     const char *semantic_name;
3517     DWORD semantic_idx;
3518
3519     switch(args->super.vp_mode)
3520     {
3521         case pretransformed:
3522         case fixedfunction:
3523             /* The pixelshader has to collect the varyings on its own. In any case properly load
3524              * color0 and color1. In the case of pretransformed vertices also load texcoords. Set
3525              * other attribs to 0.0.
3526              *
3527              * For fixedfunction this behavior is correct, according to the tests. For pretransformed
3528              * we'd either need a replacement shader that can load other attribs like BINORMAL, or
3529              * load the texcoord attrib pointers to match the pixel shader signature
3530              */
3531             for(i = 0; i < MAX_REG_INPUT; i++)
3532             {
3533                 semantic_name = sig[i].semantic_name;
3534                 semantic_idx = sig[i].semantic_idx;
3535                 if (!semantic_name) continue;
3536
3537                 if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_COLOR))
3538                 {
3539                     if (!semantic_idx) priv->ps_input[i] = "fragment.color.primary";
3540                     else if(semantic_idx == 1) priv->ps_input[i] = "fragment.color.secondary";
3541                     else priv->ps_input[i] = "0.0";
3542                 }
3543                 else if(args->super.vp_mode == fixedfunction)
3544                 {
3545                     priv->ps_input[i] = "0.0";
3546                 }
3547                 else if(shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_TEXCOORD))
3548                 {
3549                     if(semantic_idx < 8) priv->ps_input[i] = texcoords[semantic_idx];
3550                     else priv->ps_input[i] = "0.0";
3551                 }
3552                 else if(shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_FOG))
3553                 {
3554                     if (!semantic_idx) priv->ps_input[i] = "fragment.fogcoord";
3555                     else priv->ps_input[i] = "0.0";
3556                 }
3557                 else
3558                 {
3559                     priv->ps_input[i] = "0.0";
3560                 }
3561
3562                 TRACE("v%u, semantic %s%u is %s\n", i, semantic_name, semantic_idx, priv->ps_input[i]);
3563             }
3564             break;
3565
3566         case vertexshader:
3567             /* That one is easy. The vertex shaders provide v0-v7 in fragment.texcoord and v8 and v9 in
3568              * fragment.color
3569              */
3570             for(i = 0; i < 8; i++)
3571             {
3572                 priv->ps_input[i] = texcoords[i];
3573             }
3574             priv->ps_input[8] = "fragment.color.primary";
3575             priv->ps_input[9] = "fragment.color.secondary";
3576             break;
3577     }
3578 }
3579
3580 /* GL locking is done by the caller */
3581 static GLuint shader_arb_generate_pshader(const struct wined3d_shader *shader,
3582         const struct wined3d_gl_info *gl_info, struct wined3d_shader_buffer *buffer,
3583         const struct arb_ps_compile_args *args, struct arb_ps_compiled_shader *compiled)
3584 {
3585     const struct wined3d_shader_reg_maps *reg_maps = &shader->reg_maps;
3586     const struct wined3d_shader_lconst *lconst;
3587     const DWORD *function = shader->function;
3588     GLuint retval;
3589     char fragcolor[16];
3590     DWORD *lconst_map = local_const_mapping(shader), next_local;
3591     struct shader_arb_ctx_priv priv_ctx;
3592     BOOL dcl_td = FALSE;
3593     BOOL want_nv_prog = FALSE;
3594     struct arb_pshader_private *shader_priv = shader->backend_data;
3595     GLint errPos;
3596     DWORD map;
3597
3598     char srgbtmp[4][4];
3599     unsigned int i, found = 0;
3600
3601     for (i = 0, map = reg_maps->temporary; map; map >>= 1, ++i)
3602     {
3603         if (!(map & 1)
3604                 || (shader->u.ps.color0_mov && i == shader->u.ps.color0_reg)
3605                 || (reg_maps->shader_version.major < 2 && !i))
3606             continue;
3607
3608         sprintf(srgbtmp[found], "R%u", i);
3609         ++found;
3610         if (found == 4) break;
3611     }
3612
3613     switch(found) {
3614         case 0:
3615             sprintf(srgbtmp[0], "TA");
3616             sprintf(srgbtmp[1], "TB");
3617             sprintf(srgbtmp[2], "TC");
3618             sprintf(srgbtmp[3], "TD");
3619             dcl_td = TRUE;
3620             break;
3621         case 1:
3622             sprintf(srgbtmp[1], "TA");
3623             sprintf(srgbtmp[2], "TB");
3624             sprintf(srgbtmp[3], "TC");
3625             break;
3626         case 2:
3627             sprintf(srgbtmp[2], "TA");
3628             sprintf(srgbtmp[3], "TB");
3629             break;
3630         case 3:
3631             sprintf(srgbtmp[3], "TA");
3632             break;
3633         case 4:
3634             break;
3635     }
3636
3637     /*  Create the hw ARB shader */
3638     memset(&priv_ctx, 0, sizeof(priv_ctx));
3639     priv_ctx.cur_ps_args = args;
3640     priv_ctx.compiled_fprog = compiled;
3641     priv_ctx.cur_np2fixup_info = &compiled->np2fixup_info;
3642     init_ps_input(shader, args, &priv_ctx);
3643     list_init(&priv_ctx.control_frames);
3644
3645     /* Avoid enabling NV_fragment_program* if we do not need it.
3646      *
3647      * Enabling GL_NV_fragment_program_option causes the driver to occupy a temporary register,
3648      * and it slows down the shader execution noticeably(about 5%). Usually our instruction emulation
3649      * is faster than what we gain from using higher native instructions. There are some things though
3650      * that cannot be emulated. In that case enable the extensions.
3651      * If the extension is enabled, instruction handlers that support both ways will use it.
3652      *
3653      * Testing shows no performance difference between OPTION NV_fragment_program2 and NV_fragment_program.
3654      * So enable the best we can get.
3655      */
3656     if(reg_maps->usesdsx || reg_maps->usesdsy || reg_maps->loop_depth > 0 || reg_maps->usestexldd ||
3657        reg_maps->usestexldl || reg_maps->usesfacing || reg_maps->usesifc || reg_maps->usescall)
3658     {
3659         want_nv_prog = TRUE;
3660     }
3661
3662     shader_addline(buffer, "!!ARBfp1.0\n");
3663     if (want_nv_prog && gl_info->supported[NV_FRAGMENT_PROGRAM2])
3664     {
3665         shader_addline(buffer, "OPTION NV_fragment_program2;\n");
3666         priv_ctx.target_version = NV3;
3667     }
3668     else if (want_nv_prog && gl_info->supported[NV_FRAGMENT_PROGRAM_OPTION])
3669     {
3670         shader_addline(buffer, "OPTION NV_fragment_program;\n");
3671         priv_ctx.target_version = NV2;
3672     } else {
3673         if(want_nv_prog)
3674         {
3675             /* This is an error - either we're advertising the wrong shader version, or aren't enforcing some
3676              * limits properly
3677              */
3678             ERR("The shader requires instructions that are not available in plain GL_ARB_fragment_program\n");
3679             ERR("Try GLSL\n");
3680         }
3681         priv_ctx.target_version = ARB;
3682     }
3683
3684     if (reg_maps->rt_mask > 1)
3685     {
3686         shader_addline(buffer, "OPTION ARB_draw_buffers;\n");
3687     }
3688
3689     if (reg_maps->shader_version.major < 3)
3690     {
3691         switch(args->super.fog) {
3692             case FOG_OFF:
3693                 break;
3694             case FOG_LINEAR:
3695                 shader_addline(buffer, "OPTION ARB_fog_linear;\n");
3696                 break;
3697             case FOG_EXP:
3698                 shader_addline(buffer, "OPTION ARB_fog_exp;\n");
3699                 break;
3700             case FOG_EXP2:
3701                 shader_addline(buffer, "OPTION ARB_fog_exp2;\n");
3702                 break;
3703         }
3704     }
3705
3706     /* For now always declare the temps. At least the Nvidia assembler optimizes completely
3707      * unused temps away(but occupies them for the whole shader if they're used once). Always
3708      * declaring them avoids tricky bookkeeping work
3709      */
3710     shader_addline(buffer, "TEMP TA;\n");      /* Used for modifiers */
3711     shader_addline(buffer, "TEMP TB;\n");      /* Used for modifiers */
3712     shader_addline(buffer, "TEMP TC;\n");      /* Used for modifiers */
3713     if(dcl_td) shader_addline(buffer, "TEMP TD;\n"); /* Used for sRGB writing */
3714     shader_addline(buffer, "PARAM coefdiv = { 0.5, 0.25, 0.125, 0.0625 };\n");
3715     shader_addline(buffer, "PARAM coefmul = { 2, 4, 8, 16 };\n");
3716     shader_addline(buffer, "PARAM ps_helper_const = { 0.0, 1.0, %1.10f, 0.0 };\n", eps);
3717
3718     if (reg_maps->shader_version.major < 2)
3719     {
3720         strcpy(fragcolor, "R0");
3721     }
3722     else
3723     {
3724         if (args->super.srgb_correction)
3725         {
3726             if (shader->u.ps.color0_mov)
3727             {
3728                 sprintf(fragcolor, "R%u", shader->u.ps.color0_reg);
3729             }
3730             else
3731             {
3732                 shader_addline(buffer, "TEMP TMP_COLOR;\n");
3733                 strcpy(fragcolor, "TMP_COLOR");
3734             }
3735         } else {
3736             strcpy(fragcolor, "result.color");
3737         }
3738     }
3739
3740     if(args->super.srgb_correction) {
3741         shader_addline(buffer, "PARAM srgb_consts1 = {%f, %f, %f, %f};\n",
3742                        srgb_mul_low, srgb_cmp, srgb_pow, srgb_mul_high);
3743         shader_addline(buffer, "PARAM srgb_consts2 = {%f, %f, %f, %f};\n",
3744                        srgb_sub_high, 0.0, 0.0, 0.0);
3745     }
3746
3747     /* Base Declarations */
3748     next_local = shader_generate_arb_declarations(shader, reg_maps,
3749             buffer, gl_info, lconst_map, NULL, &priv_ctx);
3750
3751     for (i = 0, map = reg_maps->bumpmat; map; map >>= 1, ++i)
3752     {
3753         unsigned char bump_const;
3754
3755         if (!(map & 1)) continue;
3756
3757         bump_const = compiled->numbumpenvmatconsts;
3758         compiled->bumpenvmatconst[bump_const].const_num = WINED3D_CONST_NUM_UNUSED;
3759         compiled->bumpenvmatconst[bump_const].texunit = i;
3760         compiled->luminanceconst[bump_const].const_num = WINED3D_CONST_NUM_UNUSED;
3761         compiled->luminanceconst[bump_const].texunit = i;
3762
3763         /* We can fit the constants into the constant limit for sure because texbem, texbeml, bem and beml are only supported
3764          * in 1.x shaders, and GL_ARB_fragment_program has a constant limit of 24 constants. So in the worst case we're loading
3765          * 8 shader constants, 8 bump matrices and 8 luminance parameters and are perfectly fine. (No NP2 fixup on bumpmapped
3766          * textures due to conditional NP2 restrictions)
3767          *
3768          * Use local constants to load the bump env parameters, not program.env. This avoids collisions with d3d constants of
3769          * shaders in newer shader models. Since the bump env parameters have to share their space with NP2 fixup constants,
3770          * their location is shader dependent anyway and they cannot be loaded globally.
3771          */
3772         compiled->bumpenvmatconst[bump_const].const_num = next_local++;
3773         shader_addline(buffer, "PARAM bumpenvmat%d = program.local[%d];\n",
3774                        i, compiled->bumpenvmatconst[bump_const].const_num);
3775         compiled->numbumpenvmatconsts = bump_const + 1;
3776
3777         if (!(reg_maps->luminanceparams & (1 << i))) continue;
3778
3779         compiled->luminanceconst[bump_const].const_num = next_local++;
3780         shader_addline(buffer, "PARAM luminance%d = program.local[%d];\n",
3781                        i, compiled->luminanceconst[bump_const].const_num);
3782     }
3783
3784     for(i = 0; i < MAX_CONST_I; i++)
3785     {
3786         compiled->int_consts[i] = WINED3D_CONST_NUM_UNUSED;
3787         if (reg_maps->integer_constants & (1 << i) && priv_ctx.target_version >= NV2)
3788         {
3789             const DWORD *control_values = find_loop_control_values(shader, i);
3790
3791             if(control_values)
3792             {
3793                 shader_addline(buffer, "PARAM I%u = {%u, %u, %u, -1};\n", i,
3794                                 control_values[0], control_values[1], control_values[2]);
3795             }
3796             else
3797             {
3798                 compiled->int_consts[i] = next_local;
3799                 compiled->num_int_consts++;
3800                 shader_addline(buffer, "PARAM I%u = program.local[%u];\n", i, next_local++);
3801             }
3802         }
3803     }
3804
3805     if(reg_maps->vpos || reg_maps->usesdsy)
3806     {
3807         compiled->ycorrection = next_local;
3808         shader_addline(buffer, "PARAM ycorrection = program.local[%u];\n", next_local++);
3809
3810         if(reg_maps->vpos)
3811         {
3812             shader_addline(buffer, "TEMP vpos;\n");
3813             /* ycorrection.x: Backbuffer height(onscreen) or 0(offscreen).
3814              * ycorrection.y: -1.0(onscreen), 1.0(offscreen)
3815              * ycorrection.z: 1.0
3816              * ycorrection.w: 0.0
3817              */
3818             shader_addline(buffer, "MAD vpos, fragment.position, ycorrection.zyww, ycorrection.wxww;\n");
3819             shader_addline(buffer, "FLR vpos.xy, vpos;\n");
3820         }
3821     }
3822     else
3823     {
3824         compiled->ycorrection = WINED3D_CONST_NUM_UNUSED;
3825     }
3826
3827     /* Load constants to fixup NP2 texcoords if there are still free constants left:
3828      * Constants (texture dimensions) for the NP2 fixup are loaded as local program parameters. This will consume
3829      * at most 8 (MAX_FRAGMENT_SAMPLERS / 2) parameters, which is highly unlikely, since the application had to
3830      * use 16 NP2 textures at the same time. In case that we run out of constants the fixup is simply not
3831      * applied / activated. This will probably result in wrong rendering of the texture, but will save us from
3832      * shader compilation errors and the subsequent errors when drawing with this shader. */
3833     if (priv_ctx.cur_ps_args->super.np2_fixup) {
3834         unsigned char cur_fixup_sampler = 0;
3835
3836         struct arb_ps_np2fixup_info* const fixup = priv_ctx.cur_np2fixup_info;
3837         const WORD map = priv_ctx.cur_ps_args->super.np2_fixup;
3838         const UINT max_lconsts = gl_info->limits.arb_ps_local_constants;
3839
3840         fixup->offset = next_local;
3841         fixup->super.active = 0;
3842
3843         for (i = 0; i < MAX_FRAGMENT_SAMPLERS; ++i) {
3844             if (!(map & (1 << i))) continue;
3845
3846             if (fixup->offset + (cur_fixup_sampler >> 1) < max_lconsts) {
3847                 fixup->super.active |= (1 << i);
3848                 fixup->super.idx[i] = cur_fixup_sampler++;
3849             } else {
3850                 FIXME("No free constant found to load NP2 fixup data into shader. "
3851                       "Sampling from this texture will probably look wrong.\n");
3852                 break;
3853             }
3854         }
3855
3856         fixup->super.num_consts = (cur_fixup_sampler + 1) >> 1;
3857         if (fixup->super.num_consts) {
3858             shader_addline(buffer, "PARAM np2fixup[%u] = { program.env[%u..%u] };\n",
3859                            fixup->super.num_consts, fixup->offset, fixup->super.num_consts + fixup->offset - 1);
3860         }
3861     }
3862
3863     if (shader_priv->clipplane_emulation != ~0U && args->clip)
3864     {
3865         shader_addline(buffer, "KIL fragment.texcoord[%u];\n", shader_priv->clipplane_emulation);
3866     }
3867
3868     /* Base Shader Body */
3869     shader_generate_main(shader, buffer, reg_maps, function, &priv_ctx);
3870
3871     if(args->super.srgb_correction) {
3872         arbfp_add_sRGB_correction(buffer, fragcolor, srgbtmp[0], srgbtmp[1], srgbtmp[2], srgbtmp[3],
3873                                   priv_ctx.target_version >= NV2);
3874     }
3875
3876     if(strcmp(fragcolor, "result.color")) {
3877         shader_addline(buffer, "MOV result.color, %s;\n", fragcolor);
3878     }
3879     shader_addline(buffer, "END\n");
3880
3881     /* TODO: change to resource.glObjectHandle or something like that */
3882     GL_EXTCALL(glGenProgramsARB(1, &retval));
3883
3884     TRACE("Creating a hw pixel shader, prg=%d\n", retval);
3885     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, retval));
3886
3887     TRACE("Created hw pixel shader, prg=%d\n", retval);
3888     /* Create the program and check for errors */
3889     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
3890                buffer->bsize, buffer->buffer));
3891     checkGLcall("glProgramStringARB()");
3892
3893     gl_info->gl_ops.gl.p_glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &errPos);
3894     if (errPos != -1)
3895     {
3896         FIXME("HW PixelShader Error at position %d: %s\n\n",
3897               errPos, debugstr_a((const char *)gl_info->gl_ops.gl.p_glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
3898         shader_arb_dump_program_source(buffer->buffer);
3899         retval = 0;
3900     }
3901     else
3902     {
3903         GLint native;
3904
3905         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
3906         checkGLcall("glGetProgramivARB()");
3907         if (!native) WARN("Program exceeds native resource limits.\n");
3908     }
3909
3910     /* Load immediate constants */
3911     if (lconst_map)
3912     {
3913         LIST_FOR_EACH_ENTRY(lconst, &shader->constantsF, struct wined3d_shader_lconst, entry)
3914         {
3915             const float *value = (const float *)lconst->value;
3916             GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, lconst_map[lconst->idx], value));
3917             checkGLcall("glProgramLocalParameter4fvARB");
3918         }
3919         HeapFree(GetProcessHeap(), 0, lconst_map);
3920     }
3921
3922     return retval;
3923 }
3924
3925 static int compare_sig(const struct wined3d_shader_signature_element *sig1, const struct wined3d_shader_signature_element *sig2)
3926 {
3927     unsigned int i;
3928     int ret;
3929
3930     for(i = 0; i < MAX_REG_INPUT; i++)
3931     {
3932         if (!sig1[i].semantic_name || !sig2[i].semantic_name)
3933         {
3934             /* Compare pointers, not contents. One string is NULL(element does not exist), the other one is not NULL */
3935             if(sig1[i].semantic_name != sig2[i].semantic_name) return sig1[i].semantic_name < sig2[i].semantic_name ? -1 : 1;
3936             continue;
3937         }
3938
3939         if ((ret = strcmp(sig1[i].semantic_name, sig2[i].semantic_name))) return ret;
3940         if(sig1[i].semantic_idx    != sig2[i].semantic_idx)    return sig1[i].semantic_idx    < sig2[i].semantic_idx    ? -1 : 1;
3941         if(sig1[i].sysval_semantic != sig2[i].sysval_semantic) return sig1[i].sysval_semantic < sig2[i].sysval_semantic ? -1 : 1;
3942         if(sig1[i].component_type  != sig2[i].component_type)  return sig1[i].component_type  < sig2[i].component_type  ? -1 : 1;
3943         if(sig1[i].register_idx    != sig2[i].register_idx)    return sig1[i].register_idx    < sig2[i].register_idx    ? -1 : 1;
3944         if(sig1[i].mask            != sig2[i].mask)            return sig1[i].mask            < sig2[i].mask            ? -1 : 1;
3945     }
3946     return 0;
3947 }
3948
3949 static struct wined3d_shader_signature_element *clone_sig(const struct wined3d_shader_signature_element *sig)
3950 {
3951     struct wined3d_shader_signature_element *new;
3952     int i;
3953     char *name;
3954
3955     new = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*new) * MAX_REG_INPUT);
3956     for(i = 0; i < MAX_REG_INPUT; i++)
3957     {
3958         if (!sig[i].semantic_name) continue;
3959
3960         new[i] = sig[i];
3961         /* Clone the semantic string */
3962         name = HeapAlloc(GetProcessHeap(), 0, strlen(sig[i].semantic_name) + 1);
3963         strcpy(name, sig[i].semantic_name);
3964         new[i].semantic_name = name;
3965     }
3966     return new;
3967 }
3968
3969 static DWORD find_input_signature(struct shader_arb_priv *priv, const struct wined3d_shader_signature_element *sig)
3970 {
3971     struct wine_rb_entry *entry = wine_rb_get(&priv->signature_tree, sig);
3972     struct ps_signature *found_sig;
3973
3974     if (entry)
3975     {
3976         found_sig = WINE_RB_ENTRY_VALUE(entry, struct ps_signature, entry);
3977         TRACE("Found existing signature %u\n", found_sig->idx);
3978         return found_sig->idx;
3979     }
3980     found_sig = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*sig));
3981     found_sig->sig = clone_sig(sig);
3982     found_sig->idx = priv->ps_sig_number++;
3983     TRACE("New signature stored and assigned number %u\n", found_sig->idx);
3984     if(wine_rb_put(&priv->signature_tree, sig, &found_sig->entry) == -1)
3985     {
3986         ERR("Failed to insert program entry.\n");
3987     }
3988     return found_sig->idx;
3989 }
3990
3991 static void init_output_registers(const struct wined3d_shader *shader, DWORD sig_num,
3992         struct shader_arb_ctx_priv *priv_ctx, struct arb_vs_compiled_shader *compiled)
3993 {
3994     unsigned int i, j;
3995     static const char * const texcoords[8] =
3996     {
3997         "result.texcoord[0]", "result.texcoord[1]", "result.texcoord[2]", "result.texcoord[3]",
3998         "result.texcoord[4]", "result.texcoord[5]", "result.texcoord[6]", "result.texcoord[7]"
3999     };
4000     struct wined3d_device *device = shader->device;
4001     const struct wined3d_shader_signature_element *sig;
4002     const char *semantic_name;
4003     DWORD semantic_idx, reg_idx;
4004
4005     /* Write generic input varyings 0 to 7 to result.texcoord[], varying 8 to result.color.primary
4006      * and varying 9 to result.color.secondary
4007      */
4008     static const char * const decl_idx_to_string[MAX_REG_INPUT] =
4009     {
4010         "result.texcoord[0]", "result.texcoord[1]", "result.texcoord[2]", "result.texcoord[3]",
4011         "result.texcoord[4]", "result.texcoord[5]", "result.texcoord[6]", "result.texcoord[7]",
4012         "result.color.primary", "result.color.secondary"
4013     };
4014
4015     if(sig_num == ~0)
4016     {
4017         TRACE("Pixel shader uses builtin varyings\n");
4018         /* Map builtins to builtins */
4019         for(i = 0; i < 8; i++)
4020         {
4021             priv_ctx->texcrd_output[i] = texcoords[i];
4022         }
4023         priv_ctx->color_output[0] = "result.color.primary";
4024         priv_ctx->color_output[1] = "result.color.secondary";
4025         priv_ctx->fog_output = "TMP_FOGCOORD";
4026
4027         /* Map declared regs to builtins. Use "TA" to /dev/null unread output */
4028         for (i = 0; i < (sizeof(shader->output_signature) / sizeof(*shader->output_signature)); ++i)
4029         {
4030             semantic_name = shader->output_signature[i].semantic_name;
4031             if (!semantic_name) continue;
4032
4033             if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_POSITION))
4034             {
4035                 TRACE("o%u is TMP_OUT\n", i);
4036                 if (!shader->output_signature[i].semantic_idx) priv_ctx->vs_output[i] = "TMP_OUT";
4037                 else priv_ctx->vs_output[i] = "TA";
4038             }
4039             else if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_PSIZE))
4040             {
4041                 TRACE("o%u is result.pointsize\n", i);
4042                 if (!shader->output_signature[i].semantic_idx) priv_ctx->vs_output[i] = "result.pointsize";
4043                 else priv_ctx->vs_output[i] = "TA";
4044             }
4045             else if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_COLOR))
4046             {
4047                 TRACE("o%u is result.color.?, idx %u\n", i, shader->output_signature[i].semantic_idx);
4048                 if (!shader->output_signature[i].semantic_idx)
4049                     priv_ctx->vs_output[i] = "result.color.primary";
4050                 else if (shader->output_signature[i].semantic_idx == 1)
4051                     priv_ctx->vs_output[i] = "result.color.secondary";
4052                 else priv_ctx->vs_output[i] = "TA";
4053             }
4054             else if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_TEXCOORD))
4055             {
4056                 TRACE("o%u is %s\n", i, texcoords[shader->output_signature[i].semantic_idx]);
4057                 if (shader->output_signature[i].semantic_idx >= 8) priv_ctx->vs_output[i] = "TA";
4058                 else priv_ctx->vs_output[i] = texcoords[shader->output_signature[i].semantic_idx];
4059             }
4060             else if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_FOG))
4061             {
4062                 TRACE("o%u is result.fogcoord\n", i);
4063                 if (shader->output_signature[i].semantic_idx > 0) priv_ctx->vs_output[i] = "TA";
4064                 else priv_ctx->vs_output[i] = "result.fogcoord";
4065             }
4066             else
4067             {
4068                 priv_ctx->vs_output[i] = "TA";
4069             }
4070         }
4071         return;
4072     }
4073
4074     /* Instead of searching for the signature in the signature list, read the one from the current pixel shader.
4075      * Its maybe not the shader where the signature came from, but it is the same signature and faster to find
4076      */
4077     sig = device->stateBlock->state.pixel_shader->input_signature;
4078     TRACE("Pixel shader uses declared varyings\n");
4079
4080     /* Map builtin to declared. /dev/null the results by default to the TA temp reg */
4081     for(i = 0; i < 8; i++)
4082     {
4083         priv_ctx->texcrd_output[i] = "TA";
4084     }
4085     priv_ctx->color_output[0] = "TA";
4086     priv_ctx->color_output[1] = "TA";
4087     priv_ctx->fog_output = "TA";
4088
4089     for(i = 0; i < MAX_REG_INPUT; i++)
4090     {
4091         semantic_name = sig[i].semantic_name;
4092         semantic_idx = sig[i].semantic_idx;
4093         reg_idx = sig[i].register_idx;
4094         if (!semantic_name) continue;
4095
4096         /* If a declared input register is not written by builtin arguments, don't write to it.
4097          * GL_NV_vertex_program makes sure the input defaults to 0.0, which is correct with D3D
4098          *
4099          * Don't care about POSITION and PSIZE here - this is a builtin vertex shader, position goes
4100          * to TMP_OUT in any case
4101          */
4102         if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_TEXCOORD))
4103         {
4104             if (semantic_idx < 8)
4105                 priv_ctx->texcrd_output[semantic_idx] = decl_idx_to_string[reg_idx];
4106         }
4107         else if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_COLOR))
4108         {
4109             if (semantic_idx < 2)
4110                 priv_ctx->color_output[semantic_idx] = decl_idx_to_string[reg_idx];
4111         }
4112         else if(shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_FOG))
4113         {
4114             if (!semantic_idx)
4115                 priv_ctx->fog_output = decl_idx_to_string[reg_idx];
4116         }
4117         else
4118         {
4119             continue;
4120         }
4121
4122         if (!strcmp(decl_idx_to_string[reg_idx], "result.color.primary")
4123                 || !strcmp(decl_idx_to_string[reg_idx], "result.color.secondary"))
4124         {
4125             compiled->need_color_unclamp = TRUE;
4126         }
4127     }
4128
4129     /* Map declared to declared */
4130     for (i = 0; i < (sizeof(shader->output_signature) / sizeof(*shader->output_signature)); ++i)
4131     {
4132         /* Write unread output to TA to throw them away */
4133         priv_ctx->vs_output[i] = "TA";
4134         semantic_name = shader->output_signature[i].semantic_name;
4135         if (!semantic_name) continue;
4136
4137         if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_POSITION)
4138                 && !shader->output_signature[i].semantic_idx)
4139         {
4140             priv_ctx->vs_output[i] = "TMP_OUT";
4141             continue;
4142         }
4143         else if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_PSIZE)
4144                 && !shader->output_signature[i].semantic_idx)
4145         {
4146             priv_ctx->vs_output[i] = "result.pointsize";
4147             continue;
4148         }
4149
4150         for(j = 0; j < MAX_REG_INPUT; j++)
4151         {
4152             if (!sig[j].semantic_name) continue;
4153
4154             if (!strcmp(sig[j].semantic_name, semantic_name)
4155                     && sig[j].semantic_idx == shader->output_signature[i].semantic_idx)
4156             {
4157                 priv_ctx->vs_output[i] = decl_idx_to_string[sig[j].register_idx];
4158
4159                 if (!strcmp(priv_ctx->vs_output[i], "result.color.primary")
4160                         || !strcmp(priv_ctx->vs_output[i], "result.color.secondary"))
4161                 {
4162                     compiled->need_color_unclamp = TRUE;
4163                 }
4164             }
4165         }
4166     }
4167 }
4168
4169 /* GL locking is done by the caller */
4170 static GLuint shader_arb_generate_vshader(const struct wined3d_shader *shader,
4171         const struct wined3d_gl_info *gl_info, struct wined3d_shader_buffer *buffer,
4172         const struct arb_vs_compile_args *args, struct arb_vs_compiled_shader *compiled)
4173 {
4174     const struct arb_vshader_private *shader_data = shader->backend_data;
4175     const struct wined3d_shader_reg_maps *reg_maps = &shader->reg_maps;
4176     struct shader_arb_priv *priv = shader->device->shader_priv;
4177     const struct wined3d_shader_lconst *lconst;
4178     const DWORD *function = shader->function;
4179     GLuint ret;
4180     DWORD next_local, *lconst_map = local_const_mapping(shader);
4181     struct shader_arb_ctx_priv priv_ctx;
4182     unsigned int i;
4183     GLint errPos;
4184
4185     memset(&priv_ctx, 0, sizeof(priv_ctx));
4186     priv_ctx.cur_vs_args = args;
4187     list_init(&priv_ctx.control_frames);
4188     init_output_registers(shader, args->ps_signature, &priv_ctx, compiled);
4189
4190     /*  Create the hw ARB shader */
4191     shader_addline(buffer, "!!ARBvp1.0\n");
4192
4193     /* Always enable the NV extension if available. Unlike fragment shaders, there is no
4194      * mesurable performance penalty, and we can always make use of it for clipplanes.
4195      */
4196     if (gl_info->supported[NV_VERTEX_PROGRAM3])
4197     {
4198         shader_addline(buffer, "OPTION NV_vertex_program3;\n");
4199         priv_ctx.target_version = NV3;
4200         shader_addline(buffer, "ADDRESS aL;\n");
4201     }
4202     else if (gl_info->supported[NV_VERTEX_PROGRAM2_OPTION])
4203     {
4204         shader_addline(buffer, "OPTION NV_vertex_program2;\n");
4205         priv_ctx.target_version = NV2;
4206         shader_addline(buffer, "ADDRESS aL;\n");
4207     } else {
4208         priv_ctx.target_version = ARB;
4209     }
4210
4211     shader_addline(buffer, "TEMP TMP_OUT;\n");
4212     if (reg_maps->fog)
4213         shader_addline(buffer, "TEMP TMP_FOGCOORD;\n");
4214     if (need_helper_const(shader_data, reg_maps, gl_info))
4215     {
4216         shader_addline(buffer, "PARAM helper_const = { 0.0, 1.0, 2.0, %1.10f};\n", eps);
4217     }
4218     if (need_rel_addr_const(shader_data, reg_maps, gl_info))
4219     {
4220         shader_addline(buffer, "PARAM rel_addr_const = { 0.5, %d.0, 0.0, 0.0 };\n", shader_data->rel_offset);
4221         shader_addline(buffer, "TEMP A0_SHADOW;\n");
4222     }
4223
4224     shader_addline(buffer, "TEMP TA;\n");
4225     shader_addline(buffer, "TEMP TB;\n");
4226
4227     /* Base Declarations */
4228     next_local = shader_generate_arb_declarations(shader, reg_maps, buffer,
4229             gl_info, lconst_map, &priv_ctx.vs_clipplanes, &priv_ctx);
4230
4231     for(i = 0; i < MAX_CONST_I; i++)
4232     {
4233         compiled->int_consts[i] = WINED3D_CONST_NUM_UNUSED;
4234         if(reg_maps->integer_constants & (1 << i) && priv_ctx.target_version >= NV2)
4235         {
4236             const DWORD *control_values = find_loop_control_values(shader, i);
4237
4238             if(control_values)
4239             {
4240                 shader_addline(buffer, "PARAM I%u = {%u, %u, %u, -1};\n", i,
4241                                 control_values[0], control_values[1], control_values[2]);
4242             }
4243             else
4244             {
4245                 compiled->int_consts[i] = next_local;
4246                 compiled->num_int_consts++;
4247                 shader_addline(buffer, "PARAM I%u = program.local[%u];\n", i, next_local++);
4248             }
4249         }
4250     }
4251
4252     /* We need a constant to fixup the final position */
4253     shader_addline(buffer, "PARAM posFixup = program.local[%u];\n", next_local);
4254     compiled->pos_fixup = next_local++;
4255
4256     /* Initialize output parameters. GL_ARB_vertex_program does not require special initialization values
4257      * for output parameters. D3D in theory does not do that either, but some applications depend on a
4258      * proper initialization of the secondary color, and programs using the fixed function pipeline without
4259      * a replacement shader depend on the texcoord.w being set properly.
4260      *
4261      * GL_NV_vertex_program defines that all output values are initialized to {0.0, 0.0, 0.0, 1.0}. This
4262      * assertion is in effect even when using GL_ARB_vertex_program without any NV specific additions. So
4263      * skip this if NV_vertex_program is supported. Otherwise, initialize the secondary color. For the tex-
4264      * coords, we have a flag in the opengl caps. Many cards do not require the texcoord being set, and
4265      * this can eat a number of instructions, so skip it unless this cap is set as well
4266      */
4267     if (!gl_info->supported[NV_VERTEX_PROGRAM])
4268     {
4269         const char *color_init = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_0001);
4270         shader_addline(buffer, "MOV result.color.secondary, %s;\n", color_init);
4271
4272         if (gl_info->quirks & WINED3D_QUIRK_SET_TEXCOORD_W && !priv->fragment_pipe->ffp_proj_control)
4273         {
4274             int i;
4275             const char *one = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_ONE);
4276             for(i = 0; i < min(8, MAX_REG_TEXCRD); i++)
4277             {
4278                 if (reg_maps->texcoord_mask[i] && reg_maps->texcoord_mask[i] != WINED3DSP_WRITEMASK_ALL)
4279                     shader_addline(buffer, "MOV result.texcoord[%u].w, %s\n", i, one);
4280             }
4281         }
4282     }
4283
4284     /* The shader starts with the main function */
4285     priv_ctx.in_main_func = TRUE;
4286     /* Base Shader Body */
4287     shader_generate_main(shader, buffer, reg_maps, function, &priv_ctx);
4288
4289     if (!priv_ctx.footer_written) vshader_add_footer(&priv_ctx,
4290             shader_data, args, reg_maps, gl_info, buffer);
4291
4292     shader_addline(buffer, "END\n");
4293
4294     /* TODO: change to resource.glObjectHandle or something like that */
4295     GL_EXTCALL(glGenProgramsARB(1, &ret));
4296
4297     TRACE("Creating a hw vertex shader, prg=%d\n", ret);
4298     GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, ret));
4299
4300     TRACE("Created hw vertex shader, prg=%d\n", ret);
4301     /* Create the program and check for errors */
4302     GL_EXTCALL(glProgramStringARB(GL_VERTEX_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
4303                buffer->bsize, buffer->buffer));
4304     checkGLcall("glProgramStringARB()");
4305
4306     gl_info->gl_ops.gl.p_glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &errPos);
4307     if (errPos != -1)
4308     {
4309         FIXME("HW VertexShader Error at position %d: %s\n\n",
4310               errPos, debugstr_a((const char *)gl_info->gl_ops.gl.p_glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
4311         shader_arb_dump_program_source(buffer->buffer);
4312         ret = -1;
4313     }
4314     else
4315     {
4316         GLint native;
4317
4318         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
4319         checkGLcall("glGetProgramivARB()");
4320         if (!native) WARN("Program exceeds native resource limits.\n");
4321
4322         /* Load immediate constants */
4323         if (lconst_map)
4324         {
4325             LIST_FOR_EACH_ENTRY(lconst, &shader->constantsF, struct wined3d_shader_lconst, entry)
4326             {
4327                 const float *value = (const float *)lconst->value;
4328                 GL_EXTCALL(glProgramLocalParameter4fvARB(GL_VERTEX_PROGRAM_ARB, lconst_map[lconst->idx], value));
4329             }
4330         }
4331     }
4332     HeapFree(GetProcessHeap(), 0, lconst_map);
4333
4334     return ret;
4335 }
4336
4337 /* GL locking is done by the caller */
4338 static struct arb_ps_compiled_shader *find_arb_pshader(struct wined3d_shader *shader,
4339         const struct arb_ps_compile_args *args)
4340 {
4341     struct wined3d_device *device = shader->device;
4342     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4343     UINT i;
4344     DWORD new_size;
4345     struct arb_ps_compiled_shader *new_array;
4346     struct wined3d_shader_buffer buffer;
4347     struct arb_pshader_private *shader_data;
4348     GLuint ret;
4349
4350     if (!shader->backend_data)
4351     {
4352         struct shader_arb_priv *priv = device->shader_priv;
4353
4354         shader->backend_data = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data));
4355         shader_data = shader->backend_data;
4356         shader_data->clamp_consts = shader->reg_maps.shader_version.major == 1;
4357
4358         if (shader->reg_maps.shader_version.major < 3)
4359             shader_data->input_signature_idx = ~0;
4360         else
4361             shader_data->input_signature_idx = find_input_signature(priv, shader->input_signature);
4362
4363         TRACE("Shader got assigned input signature index %u\n", shader_data->input_signature_idx);
4364
4365         if (!device->vs_clipping)
4366             shader_data->clipplane_emulation = shader_find_free_input_register(&shader->reg_maps,
4367                     gl_info->limits.texture_stages - 1);
4368         else
4369             shader_data->clipplane_emulation = ~0U;
4370     }
4371     shader_data = shader->backend_data;
4372
4373     /* Usually we have very few GL shaders for each d3d shader(just 1 or maybe 2),
4374      * so a linear search is more performant than a hashmap or a binary search
4375      * (cache coherency etc)
4376      */
4377     for (i = 0; i < shader_data->num_gl_shaders; ++i)
4378     {
4379         if (!memcmp(&shader_data->gl_shaders[i].args, args, sizeof(*args)))
4380             return &shader_data->gl_shaders[i];
4381     }
4382
4383     TRACE("No matching GL shader found, compiling a new shader\n");
4384     if(shader_data->shader_array_size == shader_data->num_gl_shaders) {
4385         if (shader_data->num_gl_shaders)
4386         {
4387             new_size = shader_data->shader_array_size + max(1, shader_data->shader_array_size / 2);
4388             new_array = HeapReAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, shader_data->gl_shaders,
4389                                     new_size * sizeof(*shader_data->gl_shaders));
4390         } else {
4391             new_array = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data->gl_shaders));
4392             new_size = 1;
4393         }
4394
4395         if(!new_array) {
4396             ERR("Out of memory\n");
4397             return 0;
4398         }
4399         shader_data->gl_shaders = new_array;
4400         shader_data->shader_array_size = new_size;
4401     }
4402
4403     shader_data->gl_shaders[shader_data->num_gl_shaders].args = *args;
4404
4405     pixelshader_update_samplers(&shader->reg_maps, device->stateBlock->state.textures);
4406
4407     if (!shader_buffer_init(&buffer))
4408     {
4409         ERR("Failed to initialize shader buffer.\n");
4410         return 0;
4411     }
4412
4413     ret = shader_arb_generate_pshader(shader, gl_info, &buffer, args,
4414             &shader_data->gl_shaders[shader_data->num_gl_shaders]);
4415     shader_buffer_free(&buffer);
4416     shader_data->gl_shaders[shader_data->num_gl_shaders].prgId = ret;
4417
4418     return &shader_data->gl_shaders[shader_data->num_gl_shaders++];
4419 }
4420
4421 static inline BOOL vs_args_equal(const struct arb_vs_compile_args *stored, const struct arb_vs_compile_args *new,
4422                                  const DWORD use_map, BOOL skip_int) {
4423     if((stored->super.swizzle_map & use_map) != new->super.swizzle_map) return FALSE;
4424     if(stored->super.clip_enabled != new->super.clip_enabled) return FALSE;
4425     if(stored->super.fog_src != new->super.fog_src) return FALSE;
4426     if(stored->clip.boolclip_compare != new->clip.boolclip_compare) return FALSE;
4427     if(stored->ps_signature != new->ps_signature) return FALSE;
4428     if(stored->vertex.samplers_compare != new->vertex.samplers_compare) return FALSE;
4429     if(skip_int) return TRUE;
4430
4431     return !memcmp(stored->loop_ctrl, new->loop_ctrl, sizeof(stored->loop_ctrl));
4432 }
4433
4434 static struct arb_vs_compiled_shader *find_arb_vshader(struct wined3d_shader *shader,
4435         const struct arb_vs_compile_args *args)
4436 {
4437     struct wined3d_device *device = shader->device;
4438     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4439     DWORD use_map = device->strided_streams.use_map;
4440     UINT i;
4441     DWORD new_size;
4442     struct arb_vs_compiled_shader *new_array;
4443     struct wined3d_shader_buffer buffer;
4444     struct arb_vshader_private *shader_data;
4445     GLuint ret;
4446
4447     if (!shader->backend_data)
4448     {
4449         const struct wined3d_shader_reg_maps *reg_maps = &shader->reg_maps;
4450
4451         shader->backend_data = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data));
4452         shader_data = shader->backend_data;
4453
4454         if ((gl_info->quirks & WINED3D_QUIRK_ARB_VS_OFFSET_LIMIT)
4455                 && reg_maps->min_rel_offset <= reg_maps->max_rel_offset)
4456         {
4457             if (reg_maps->max_rel_offset - reg_maps->min_rel_offset > 127)
4458             {
4459                 FIXME("The difference between the minimum and maximum relative offset is > 127.\n");
4460                 FIXME("Which this OpenGL implementation does not support. Try using GLSL.\n");
4461                 FIXME("Min: %u, Max: %u.\n", reg_maps->min_rel_offset, reg_maps->max_rel_offset);
4462             }
4463             else if (reg_maps->max_rel_offset - reg_maps->min_rel_offset > 63)
4464                 shader_data->rel_offset = reg_maps->min_rel_offset + 63;
4465             else if (reg_maps->max_rel_offset > 63)
4466                 shader_data->rel_offset = reg_maps->min_rel_offset;
4467         }
4468     }
4469     shader_data = shader->backend_data;
4470
4471     /* Usually we have very few GL shaders for each d3d shader(just 1 or maybe 2),
4472      * so a linear search is more performant than a hashmap or a binary search
4473      * (cache coherency etc)
4474      */
4475     for(i = 0; i < shader_data->num_gl_shaders; i++) {
4476         if (vs_args_equal(&shader_data->gl_shaders[i].args, args,
4477                 use_map, gl_info->supported[NV_VERTEX_PROGRAM2_OPTION]))
4478         {
4479             return &shader_data->gl_shaders[i];
4480         }
4481     }
4482
4483     TRACE("No matching GL shader found, compiling a new shader\n");
4484
4485     if(shader_data->shader_array_size == shader_data->num_gl_shaders) {
4486         if (shader_data->num_gl_shaders)
4487         {
4488             new_size = shader_data->shader_array_size + max(1, shader_data->shader_array_size / 2);
4489             new_array = HeapReAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, shader_data->gl_shaders,
4490                                     new_size * sizeof(*shader_data->gl_shaders));
4491         } else {
4492             new_array = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data->gl_shaders));
4493             new_size = 1;
4494         }
4495
4496         if(!new_array) {
4497             ERR("Out of memory\n");
4498             return 0;
4499         }
4500         shader_data->gl_shaders = new_array;
4501         shader_data->shader_array_size = new_size;
4502     }
4503
4504     shader_data->gl_shaders[shader_data->num_gl_shaders].args = *args;
4505
4506     if (!shader_buffer_init(&buffer))
4507     {
4508         ERR("Failed to initialize shader buffer.\n");
4509         return 0;
4510     }
4511
4512     ret = shader_arb_generate_vshader(shader, gl_info, &buffer, args,
4513             &shader_data->gl_shaders[shader_data->num_gl_shaders]);
4514     shader_buffer_free(&buffer);
4515     shader_data->gl_shaders[shader_data->num_gl_shaders].prgId = ret;
4516
4517     return &shader_data->gl_shaders[shader_data->num_gl_shaders++];
4518 }
4519
4520 static void find_arb_ps_compile_args(const struct wined3d_state *state,
4521         const struct wined3d_shader *shader, struct arb_ps_compile_args *args)
4522 {
4523     struct wined3d_device *device = shader->device;
4524     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4525     int i;
4526     WORD int_skip;
4527
4528     find_ps_compile_args(state, shader, &args->super);
4529
4530     /* This forces all local boolean constants to 1 to make them stateblock independent */
4531     args->bools = shader->reg_maps.local_bool_consts;
4532
4533     for(i = 0; i < MAX_CONST_B; i++)
4534     {
4535         if (state->ps_consts_b[i])
4536             args->bools |= ( 1 << i);
4537     }
4538
4539     /* Only enable the clip plane emulation KIL if at least one clipplane is enabled. The KIL instruction
4540      * is quite expensive because it forces the driver to disable early Z discards. It is cheaper to
4541      * duplicate the shader than have a no-op KIL instruction in every shader
4542      */
4543     if (!device->vs_clipping && use_vs(state)
4544             && state->render_states[WINED3D_RS_CLIPPING]
4545             && state->render_states[WINED3D_RS_CLIPPLANEENABLE])
4546         args->clip = 1;
4547     else
4548         args->clip = 0;
4549
4550     /* Skip if unused or local, or supported natively */
4551     int_skip = ~shader->reg_maps.integer_constants | shader->reg_maps.local_int_consts;
4552     if (int_skip == 0xffff || gl_info->supported[NV_FRAGMENT_PROGRAM_OPTION])
4553     {
4554         memset(args->loop_ctrl, 0, sizeof(args->loop_ctrl));
4555         return;
4556     }
4557
4558     for(i = 0; i < MAX_CONST_I; i++)
4559     {
4560         if(int_skip & (1 << i))
4561         {
4562             args->loop_ctrl[i][0] = 0;
4563             args->loop_ctrl[i][1] = 0;
4564             args->loop_ctrl[i][2] = 0;
4565         }
4566         else
4567         {
4568             args->loop_ctrl[i][0] = state->ps_consts_i[i * 4];
4569             args->loop_ctrl[i][1] = state->ps_consts_i[i * 4 + 1];
4570             args->loop_ctrl[i][2] = state->ps_consts_i[i * 4 + 2];
4571         }
4572     }
4573 }
4574
4575 static void find_arb_vs_compile_args(const struct wined3d_state *state,
4576         const struct wined3d_shader *shader, struct arb_vs_compile_args *args)
4577 {
4578     struct wined3d_device *device = shader->device;
4579     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4580     int i;
4581     WORD int_skip;
4582
4583     find_vs_compile_args(state, shader, &args->super);
4584
4585     args->clip.boolclip_compare = 0;
4586     if (use_ps(state))
4587     {
4588         const struct wined3d_shader *ps = state->pixel_shader;
4589         const struct arb_pshader_private *shader_priv = ps->backend_data;
4590         args->ps_signature = shader_priv->input_signature_idx;
4591
4592         args->clip.boolclip.clip_texcoord = shader_priv->clipplane_emulation + 1;
4593     }
4594     else
4595     {
4596         args->ps_signature = ~0;
4597         if (!device->vs_clipping && device->adapter->fragment_pipe == &arbfp_fragment_pipeline)
4598         {
4599             args->clip.boolclip.clip_texcoord = ffp_clip_emul(state) ? gl_info->limits.texture_stages : 0;
4600         }
4601         /* Otherwise: Setting boolclip_compare set clip_texcoord to 0 */
4602     }
4603
4604     if (args->clip.boolclip.clip_texcoord)
4605     {
4606         if (state->render_states[WINED3D_RS_CLIPPING])
4607             args->clip.boolclip.clipplane_mask = (unsigned char)state->render_states[WINED3D_RS_CLIPPLANEENABLE];
4608         /* clipplane_mask was set to 0 by setting boolclip_compare to 0 */
4609     }
4610
4611     /* This forces all local boolean constants to 1 to make them stateblock independent */
4612     args->clip.boolclip.bools = shader->reg_maps.local_bool_consts;
4613     /* TODO: Figure out if it would be better to store bool constants as bitmasks in the stateblock */
4614     for(i = 0; i < MAX_CONST_B; i++)
4615     {
4616         if (state->vs_consts_b[i])
4617             args->clip.boolclip.bools |= ( 1 << i);
4618     }
4619
4620     args->vertex.samplers[0] = device->texUnitMap[MAX_FRAGMENT_SAMPLERS + 0];
4621     args->vertex.samplers[1] = device->texUnitMap[MAX_FRAGMENT_SAMPLERS + 1];
4622     args->vertex.samplers[2] = device->texUnitMap[MAX_FRAGMENT_SAMPLERS + 2];
4623     args->vertex.samplers[3] = 0;
4624
4625     /* Skip if unused or local */
4626     int_skip = ~shader->reg_maps.integer_constants | shader->reg_maps.local_int_consts;
4627     /* This is about flow control, not clipping. */
4628     if (int_skip == 0xffff || gl_info->supported[NV_VERTEX_PROGRAM2_OPTION])
4629     {
4630         memset(args->loop_ctrl, 0, sizeof(args->loop_ctrl));
4631         return;
4632     }
4633
4634     for(i = 0; i < MAX_CONST_I; i++)
4635     {
4636         if(int_skip & (1 << i))
4637         {
4638             args->loop_ctrl[i][0] = 0;
4639             args->loop_ctrl[i][1] = 0;
4640             args->loop_ctrl[i][2] = 0;
4641         }
4642         else
4643         {
4644             args->loop_ctrl[i][0] = state->vs_consts_i[i * 4];
4645             args->loop_ctrl[i][1] = state->vs_consts_i[i * 4 + 1];
4646             args->loop_ctrl[i][2] = state->vs_consts_i[i * 4 + 2];
4647         }
4648     }
4649 }
4650
4651 /* GL locking is done by the caller */
4652 static void shader_arb_select(const struct wined3d_context *context, enum wined3d_shader_mode vertex_mode,
4653         enum wined3d_shader_mode fragment_mode)
4654 {
4655     struct wined3d_device *device = context->swapchain->device;
4656     struct shader_arb_priv *priv = device->shader_priv;
4657     const struct wined3d_gl_info *gl_info = context->gl_info;
4658     const struct wined3d_state *state = &device->stateBlock->state;
4659     int i;
4660
4661     /* Deal with pixel shaders first so the vertex shader arg function has the input signature ready */
4662     if (fragment_mode == WINED3D_SHADER_MODE_SHADER)
4663     {
4664         struct wined3d_shader *ps = state->pixel_shader;
4665         struct arb_ps_compile_args compile_args;
4666         struct arb_ps_compiled_shader *compiled;
4667
4668         TRACE("Using pixel shader %p.\n", ps);
4669         find_arb_ps_compile_args(state, ps, &compile_args);
4670         compiled = find_arb_pshader(ps, &compile_args);
4671         priv->current_fprogram_id = compiled->prgId;
4672         priv->compiled_fprog = compiled;
4673
4674         /* Bind the fragment program */
4675         GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, priv->current_fprogram_id));
4676         checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, priv->current_fprogram_id);");
4677
4678         if (!priv->use_arbfp_fixed_func)
4679         {
4680             priv->fragment_pipe->enable_extension(gl_info, FALSE);
4681
4682             /* Enable OpenGL fragment programs. */
4683             gl_info->gl_ops.gl.p_glEnable(GL_FRAGMENT_PROGRAM_ARB);
4684             checkGLcall("glEnable(GL_FRAGMENT_PROGRAM_ARB);");
4685         }
4686         TRACE("(%p) : Bound fragment program %u and enabled GL_FRAGMENT_PROGRAM_ARB\n",
4687                 device, priv->current_fprogram_id);
4688
4689         /* Pixel Shader 1.x constants are clamped to [-1;1], Pixel Shader 2.0 constants are not. If switching between
4690          * a 1.x and newer shader, reload the first 8 constants
4691          */
4692         if (priv->last_ps_const_clamped != ((struct arb_pshader_private *)ps->backend_data)->clamp_consts)
4693         {
4694             priv->last_ps_const_clamped = ((struct arb_pshader_private *)ps->backend_data)->clamp_consts;
4695             priv->highest_dirty_ps_const = max(priv->highest_dirty_ps_const, 8);
4696             for(i = 0; i < 8; i++)
4697             {
4698                 priv->pshader_const_dirty[i] = 1;
4699             }
4700             /* Also takes care of loading local constants */
4701             shader_arb_load_constants_internal(context, TRUE, FALSE, TRUE);
4702         }
4703         else
4704         {
4705             UINT rt_height = state->fb->render_targets[0]->resource.height;
4706             shader_arb_ps_local_constants(compiled, context, state, rt_height);
4707         }
4708
4709         /* Force constant reloading for the NP2 fixup (see comment in shader_glsl_select for more info) */
4710         if (compiled->np2fixup_info.super.active)
4711             shader_arb_load_np2fixup_constants(priv, gl_info, state);
4712     }
4713     else
4714     {
4715         if (gl_info->supported[ARB_FRAGMENT_PROGRAM] && !priv->use_arbfp_fixed_func)
4716         {
4717             /* Disable only if we're not using arbfp fixed function fragment
4718              * processing. If this is used, keep GL_FRAGMENT_PROGRAM_ARB
4719              * enabled, and the fixed function pipeline will bind the fixed
4720              * function replacement shader. */
4721             gl_info->gl_ops.gl.p_glDisable(GL_FRAGMENT_PROGRAM_ARB);
4722             checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
4723             priv->current_fprogram_id = 0;
4724         }
4725         priv->fragment_pipe->enable_extension(gl_info, fragment_mode == WINED3D_SHADER_MODE_FFP);
4726     }
4727
4728     if (vertex_mode == WINED3D_SHADER_MODE_SHADER)
4729     {
4730         struct wined3d_shader *vs = state->vertex_shader;
4731         struct arb_vs_compile_args compile_args;
4732         struct arb_vs_compiled_shader *compiled;
4733
4734         TRACE("Using vertex shader %p\n", vs);
4735         find_arb_vs_compile_args(state, vs, &compile_args);
4736         compiled = find_arb_vshader(vs, &compile_args);
4737         priv->current_vprogram_id = compiled->prgId;
4738         priv->compiled_vprog = compiled;
4739
4740         /* Bind the vertex program */
4741         GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->current_vprogram_id));
4742         checkGLcall("glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->current_vprogram_id);");
4743
4744         /* Enable OpenGL vertex programs */
4745         gl_info->gl_ops.gl.p_glEnable(GL_VERTEX_PROGRAM_ARB);
4746         checkGLcall("glEnable(GL_VERTEX_PROGRAM_ARB);");
4747         TRACE("(%p) : Bound vertex program %u and enabled GL_VERTEX_PROGRAM_ARB\n", device, priv->current_vprogram_id);
4748         shader_arb_vs_local_constants(compiled, context, state);
4749
4750         if(priv->last_vs_color_unclamp != compiled->need_color_unclamp) {
4751             priv->last_vs_color_unclamp = compiled->need_color_unclamp;
4752
4753             if (gl_info->supported[ARB_COLOR_BUFFER_FLOAT])
4754             {
4755                 GL_EXTCALL(glClampColorARB(GL_CLAMP_VERTEX_COLOR_ARB, !compiled->need_color_unclamp));
4756                 checkGLcall("glClampColorARB");
4757             } else {
4758                 FIXME("vertex color clamp needs to be changed, but extension not supported.\n");
4759             }
4760         }
4761     }
4762     else if (gl_info->supported[ARB_VERTEX_PROGRAM])
4763     {
4764         priv->current_vprogram_id = 0;
4765         gl_info->gl_ops.gl.p_glDisable(GL_VERTEX_PROGRAM_ARB);
4766         checkGLcall("glDisable(GL_VERTEX_PROGRAM_ARB)");
4767     }
4768 }
4769
4770 /* GL locking is done by the caller */
4771 static void shader_arb_select_depth_blt(void *shader_priv, const struct wined3d_gl_info *gl_info,
4772         enum tex_types tex_type, const SIZE *ds_mask_size)
4773 {
4774     const float mask[] = {0.0f, 0.0f, (float)ds_mask_size->cx, (float)ds_mask_size->cy};
4775     BOOL masked = ds_mask_size->cx && ds_mask_size->cy;
4776     struct shader_arb_priv *priv = shader_priv;
4777     GLuint *blt_fprogram;
4778
4779     if (!priv->depth_blt_vprogram_id) priv->depth_blt_vprogram_id = create_arb_blt_vertex_program(gl_info);
4780     GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->depth_blt_vprogram_id));
4781     gl_info->gl_ops.gl.p_glEnable(GL_VERTEX_PROGRAM_ARB);
4782
4783     blt_fprogram = masked ? &priv->depth_blt_fprogram_id_masked[tex_type] : &priv->depth_blt_fprogram_id_full[tex_type];
4784     if (!*blt_fprogram) *blt_fprogram = create_arb_blt_fragment_program(gl_info, tex_type, masked);
4785     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, *blt_fprogram));
4786     if (masked) GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, 0, mask));
4787     gl_info->gl_ops.gl.p_glEnable(GL_FRAGMENT_PROGRAM_ARB);
4788 }
4789
4790 /* GL locking is done by the caller */
4791 static void shader_arb_deselect_depth_blt(void *shader_priv, const struct wined3d_gl_info *gl_info)
4792 {
4793     struct shader_arb_priv *priv = shader_priv;
4794
4795     if (priv->current_vprogram_id) {
4796         GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->current_vprogram_id));
4797         checkGLcall("glBindProgramARB(GL_VERTEX_PROGRAM_ARB, vertexShader->prgId);");
4798
4799         TRACE("Bound vertex program %u and enabled GL_VERTEX_PROGRAM_ARB.\n", priv->current_vprogram_id);
4800     }
4801     else
4802     {
4803         gl_info->gl_ops.gl.p_glDisable(GL_VERTEX_PROGRAM_ARB);
4804         checkGLcall("glDisable(GL_VERTEX_PROGRAM_ARB)");
4805     }
4806
4807     if (priv->current_fprogram_id) {
4808         GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, priv->current_fprogram_id));
4809         checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, pixelShader->prgId);");
4810
4811         TRACE("Bound fragment program %u and enabled GL_FRAGMENT_PROGRAM_ARB.\n", priv->current_fprogram_id);
4812     }
4813     else if(!priv->use_arbfp_fixed_func)
4814     {
4815         gl_info->gl_ops.gl.p_glDisable(GL_FRAGMENT_PROGRAM_ARB);
4816         checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
4817     }
4818 }
4819
4820 static void shader_arb_destroy(struct wined3d_shader *shader)
4821 {
4822     struct wined3d_device *device = shader->device;
4823     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4824
4825     if (shader_is_pshader_version(shader->reg_maps.shader_version.type))
4826     {
4827         struct arb_pshader_private *shader_data = shader->backend_data;
4828         UINT i;
4829
4830         if(!shader_data) return; /* This can happen if a shader was never compiled */
4831
4832         if (shader_data->num_gl_shaders)
4833         {
4834             struct wined3d_context *context = context_acquire(device, NULL);
4835
4836             ENTER_GL();
4837             for (i = 0; i < shader_data->num_gl_shaders; ++i)
4838             {
4839                 GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId));
4840                 checkGLcall("GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId))");
4841             }
4842             LEAVE_GL();
4843
4844             context_release(context);
4845         }
4846
4847         HeapFree(GetProcessHeap(), 0, shader_data->gl_shaders);
4848         HeapFree(GetProcessHeap(), 0, shader_data);
4849         shader->backend_data = NULL;
4850     }
4851     else
4852     {
4853         struct arb_vshader_private *shader_data = shader->backend_data;
4854         UINT i;
4855
4856         if(!shader_data) return; /* This can happen if a shader was never compiled */
4857
4858         if (shader_data->num_gl_shaders)
4859         {
4860             struct wined3d_context *context = context_acquire(device, NULL);
4861
4862             ENTER_GL();
4863             for (i = 0; i < shader_data->num_gl_shaders; ++i)
4864             {
4865                 GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId));
4866                 checkGLcall("GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId))");
4867             }
4868             LEAVE_GL();
4869
4870             context_release(context);
4871         }
4872
4873         HeapFree(GetProcessHeap(), 0, shader_data->gl_shaders);
4874         HeapFree(GetProcessHeap(), 0, shader_data);
4875         shader->backend_data = NULL;
4876     }
4877 }
4878
4879 static int sig_tree_compare(const void *key, const struct wine_rb_entry *entry)
4880 {
4881     struct ps_signature *e = WINE_RB_ENTRY_VALUE(entry, struct ps_signature, entry);
4882     return compare_sig(key, e->sig);
4883 }
4884
4885 static const struct wine_rb_functions sig_tree_functions =
4886 {
4887     wined3d_rb_alloc,
4888     wined3d_rb_realloc,
4889     wined3d_rb_free,
4890     sig_tree_compare
4891 };
4892
4893 static HRESULT shader_arb_alloc(struct wined3d_device *device, const struct fragment_pipeline *fragment_pipe)
4894 {
4895     struct shader_arb_priv *priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*priv));
4896     void *fragment_priv;
4897
4898     if (!(fragment_priv = fragment_pipe->alloc_private(&arb_program_shader_backend, priv)))
4899     {
4900         ERR("Failed to initialize fragment pipe.\n");
4901         HeapFree(GetProcessHeap(), 0, priv);
4902         return E_FAIL;
4903     }
4904
4905     priv->vshader_const_dirty = HeapAlloc(GetProcessHeap(), 0,
4906             sizeof(*priv->vshader_const_dirty) * device->d3d_vshader_constantF);
4907     if (!priv->vshader_const_dirty)
4908         goto fail;
4909     memset(priv->vshader_const_dirty, 1,
4910            sizeof(*priv->vshader_const_dirty) * device->d3d_vshader_constantF);
4911
4912     priv->pshader_const_dirty = HeapAlloc(GetProcessHeap(), 0,
4913             sizeof(*priv->pshader_const_dirty) * device->d3d_pshader_constantF);
4914     if (!priv->pshader_const_dirty)
4915         goto fail;
4916     memset(priv->pshader_const_dirty, 1,
4917             sizeof(*priv->pshader_const_dirty) * device->d3d_pshader_constantF);
4918
4919     if(wine_rb_init(&priv->signature_tree, &sig_tree_functions) == -1)
4920     {
4921         ERR("RB tree init failed\n");
4922         goto fail;
4923     }
4924     device->fragment_priv = fragment_priv;
4925     priv->fragment_pipe = fragment_pipe;
4926     device->shader_priv = priv;
4927     return WINED3D_OK;
4928
4929 fail:
4930     HeapFree(GetProcessHeap(), 0, priv->pshader_const_dirty);
4931     HeapFree(GetProcessHeap(), 0, priv->vshader_const_dirty);
4932     fragment_pipe->free_private(device);
4933     HeapFree(GetProcessHeap(), 0, priv);
4934     return E_OUTOFMEMORY;
4935 }
4936
4937 static void release_signature(struct wine_rb_entry *entry, void *context)
4938 {
4939     struct ps_signature *sig = WINE_RB_ENTRY_VALUE(entry, struct ps_signature, entry);
4940     int i;
4941     for(i = 0; i < MAX_REG_INPUT; i++)
4942     {
4943         HeapFree(GetProcessHeap(), 0, (char *) sig->sig[i].semantic_name);
4944     }
4945     HeapFree(GetProcessHeap(), 0, sig->sig);
4946     HeapFree(GetProcessHeap(), 0, sig);
4947 }
4948
4949 /* Context activation is done by the caller. */
4950 static void shader_arb_free(struct wined3d_device *device)
4951 {
4952     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4953     struct shader_arb_priv *priv = device->shader_priv;
4954     int i;
4955
4956     ENTER_GL();
4957     if(priv->depth_blt_vprogram_id) {
4958         GL_EXTCALL(glDeleteProgramsARB(1, &priv->depth_blt_vprogram_id));
4959     }
4960     for (i = 0; i < tex_type_count; ++i)
4961     {
4962         if (priv->depth_blt_fprogram_id_full[i])
4963         {
4964             GL_EXTCALL(glDeleteProgramsARB(1, &priv->depth_blt_fprogram_id_full[i]));
4965         }
4966         if (priv->depth_blt_fprogram_id_masked[i])
4967         {
4968             GL_EXTCALL(glDeleteProgramsARB(1, &priv->depth_blt_fprogram_id_masked[i]));
4969         }
4970     }
4971     LEAVE_GL();
4972
4973     wine_rb_destroy(&priv->signature_tree, release_signature, NULL);
4974     HeapFree(GetProcessHeap(), 0, priv->pshader_const_dirty);
4975     HeapFree(GetProcessHeap(), 0, priv->vshader_const_dirty);
4976     priv->fragment_pipe->free_private(device);
4977     HeapFree(GetProcessHeap(), 0, device->shader_priv);
4978 }
4979
4980 static void shader_arb_context_destroyed(void *shader_priv, const struct wined3d_context *context)
4981 {
4982     struct shader_arb_priv *priv = shader_priv;
4983
4984     if (priv->last_context == context)
4985         priv->last_context = NULL;
4986 }
4987
4988 static void shader_arb_get_caps(const struct wined3d_gl_info *gl_info, struct shader_caps *caps)
4989 {
4990     if (gl_info->supported[ARB_VERTEX_PROGRAM])
4991     {
4992         DWORD vs_consts;
4993
4994         /* 96 is the minimum allowed value of MAX_PROGRAM_ENV_PARAMETERS_ARB
4995          * for vertex programs. If the native limit is less than that it's
4996          * not very useful, and e.g. Mesa swrast returns 0, probably to
4997          * indicate it's a software implementation. */
4998         if (gl_info->limits.arb_vs_native_constants < 96)
4999             vs_consts = gl_info->limits.arb_vs_float_constants;
5000         else
5001             vs_consts = min(gl_info->limits.arb_vs_float_constants, gl_info->limits.arb_vs_native_constants);
5002
5003         if (gl_info->supported[NV_VERTEX_PROGRAM3])
5004         {
5005             caps->vs_version = 3;
5006             TRACE("Hardware vertex shader version 3.0 enabled (NV_VERTEX_PROGRAM3)\n");
5007         }
5008         else if (vs_consts >= 256)
5009         {
5010             /* Shader Model 2.0 requires at least 256 vertex shader constants */
5011             caps->vs_version = 2;
5012             TRACE("Hardware vertex shader version 2.0 enabled (ARB_PROGRAM)\n");
5013         }
5014         else
5015         {
5016             caps->vs_version = 1;
5017             TRACE("Hardware vertex shader version 1.1 enabled (ARB_PROGRAM)\n");
5018         }
5019         caps->vs_uniform_count = vs_consts;
5020     }
5021     else
5022     {
5023         caps->vs_version = 0;
5024         caps->vs_uniform_count = 0;
5025     }
5026
5027     caps->gs_version = 0;
5028
5029     if (gl_info->supported[ARB_FRAGMENT_PROGRAM])
5030     {
5031         DWORD ps_consts;
5032
5033         /* Similar as above for vertex programs, but the minimum for fragment
5034          * programs is 24. */
5035         if (gl_info->limits.arb_ps_native_constants < 24)
5036             ps_consts = gl_info->limits.arb_ps_float_constants;
5037         else
5038             ps_consts = min(gl_info->limits.arb_ps_float_constants, gl_info->limits.arb_ps_native_constants);
5039
5040         if (gl_info->supported[NV_FRAGMENT_PROGRAM2])
5041         {
5042             caps->ps_version = 3;
5043             TRACE("Hardware pixel shader version 3.0 enabled (NV_FRAGMENT_PROGRAM2)\n");
5044         }
5045         else if (ps_consts >= 32)
5046         {
5047             /* Shader Model 2.0 requires at least 32 pixel shader constants */
5048             caps->ps_version = 2;
5049             TRACE("Hardware pixel shader version 2.0 enabled (ARB_PROGRAM)\n");
5050         }
5051         else
5052         {
5053             caps->ps_version = 1;
5054             TRACE("Hardware pixel shader version 1.4 enabled (ARB_PROGRAM)\n");
5055         }
5056         caps->ps_uniform_count = ps_consts;
5057         caps->ps_1x_max_value = 8.0f;
5058     }
5059     else
5060     {
5061         caps->ps_version = 0;
5062         caps->ps_uniform_count = 0;
5063         caps->ps_1x_max_value = 0.0f;
5064     }
5065
5066     caps->vs_clipping = use_nv_clip(gl_info);
5067 }
5068
5069 static BOOL shader_arb_color_fixup_supported(struct color_fixup_desc fixup)
5070 {
5071     if (TRACE_ON(d3d_shader) && TRACE_ON(d3d))
5072     {
5073         TRACE("Checking support for color_fixup:\n");
5074         dump_color_fixup_desc(fixup);
5075     }
5076
5077     /* We support everything except complex conversions. */
5078     if (!is_complex_fixup(fixup))
5079     {
5080         TRACE("[OK]\n");
5081         return TRUE;
5082     }
5083
5084     TRACE("[FAILED]\n");
5085     return FALSE;
5086 }
5087
5088 static void shader_arb_add_instruction_modifiers(const struct wined3d_shader_instruction *ins) {
5089     DWORD shift;
5090     char write_mask[20], regstr[50];
5091     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
5092     BOOL is_color = FALSE;
5093     const struct wined3d_shader_dst_param *dst;
5094
5095     if (!ins->dst_count) return;
5096
5097     dst = &ins->dst[0];
5098     shift = dst->shift;
5099     if (!shift) return; /* Saturate alone is handled by the instructions */
5100
5101     shader_arb_get_write_mask(ins, dst, write_mask);
5102     shader_arb_get_register_name(ins, &dst->reg, regstr, &is_color);
5103
5104     /* Generate a line that does the output modifier computation
5105      * FIXME: _SAT vs shift? _SAT alone is already handled in the instructions, if this
5106      * maps problems in e.g. _d4_sat modify shader_arb_get_modifier
5107      */
5108     shader_addline(buffer, "MUL%s %s%s, %s, %s;\n", shader_arb_get_modifier(ins),
5109                    regstr, write_mask, regstr, shift_tab[shift]);
5110 }
5111
5112 static const SHADER_HANDLER shader_arb_instruction_handler_table[WINED3DSIH_TABLE_SIZE] =
5113 {
5114     /* WINED3DSIH_ABS                   */ shader_hw_map2gl,
5115     /* WINED3DSIH_ADD                   */ shader_hw_map2gl,
5116     /* WINED3DSIH_AND                   */ NULL,
5117     /* WINED3DSIH_BEM                   */ pshader_hw_bem,
5118     /* WINED3DSIH_BREAK                 */ shader_hw_break,
5119     /* WINED3DSIH_BREAKC                */ shader_hw_breakc,
5120     /* WINED3DSIH_BREAKP                */ NULL,
5121     /* WINED3DSIH_CALL                  */ shader_hw_call,
5122     /* WINED3DSIH_CALLNZ                */ NULL,
5123     /* WINED3DSIH_CMP                   */ pshader_hw_cmp,
5124     /* WINED3DSIH_CND                   */ pshader_hw_cnd,
5125     /* WINED3DSIH_CRS                   */ shader_hw_map2gl,
5126     /* WINED3DSIH_CUT                   */ NULL,
5127     /* WINED3DSIH_DCL                   */ shader_hw_nop,
5128     /* WINED3DSIH_DCL_CONSTANT_BUFFER   */ shader_hw_nop,
5129     /* WINED3DSIH_DCL_INPUT_PRIMITIVE   */ shader_hw_nop,
5130     /* WINED3DSIH_DCL_OUTPUT_TOPOLOGY   */ shader_hw_nop,
5131     /* WINED3DSIH_DCL_VERTICES_OUT      */ shader_hw_nop,
5132     /* WINED3DSIH_DEF                   */ shader_hw_nop,
5133     /* WINED3DSIH_DEFB                  */ shader_hw_nop,
5134     /* WINED3DSIH_DEFI                  */ shader_hw_nop,
5135     /* WINED3DSIH_DIV                   */ NULL,
5136     /* WINED3DSIH_DP2ADD                */ pshader_hw_dp2add,
5137     /* WINED3DSIH_DP3                   */ shader_hw_map2gl,
5138     /* WINED3DSIH_DP4                   */ shader_hw_map2gl,
5139     /* WINED3DSIH_DST                   */ shader_hw_map2gl,
5140     /* WINED3DSIH_DSX                   */ shader_hw_map2gl,
5141     /* WINED3DSIH_DSY                   */ shader_hw_dsy,
5142     /* WINED3DSIH_ELSE                  */ shader_hw_else,
5143     /* WINED3DSIH_EMIT                  */ NULL,
5144     /* WINED3DSIH_ENDIF                 */ shader_hw_endif,
5145     /* WINED3DSIH_ENDLOOP               */ shader_hw_endloop,
5146     /* WINED3DSIH_ENDREP                */ shader_hw_endrep,
5147     /* WINED3DSIH_EQ                    */ NULL,
5148     /* WINED3DSIH_EXP                   */ shader_hw_scalar_op,
5149     /* WINED3DSIH_EXPP                  */ shader_hw_scalar_op,
5150     /* WINED3DSIH_FRC                   */ shader_hw_map2gl,
5151     /* WINED3DSIH_FTOI                  */ NULL,
5152     /* WINED3DSIH_GE                    */ NULL,
5153     /* WINED3DSIH_IADD                  */ NULL,
5154     /* WINED3DSIH_IEQ                   */ NULL,
5155     /* WINED3DSIH_IF                    */ NULL /* Hardcoded into the shader */,
5156     /* WINED3DSIH_IFC                   */ shader_hw_ifc,
5157     /* WINED3DSIH_IGE                   */ NULL,
5158     /* WINED3DSIH_IMUL                  */ NULL,
5159     /* WINED3DSIH_ITOF                  */ NULL,
5160     /* WINED3DSIH_LABEL                 */ shader_hw_label,
5161     /* WINED3DSIH_LD                    */ NULL,
5162     /* WINED3DSIH_LIT                   */ shader_hw_map2gl,
5163     /* WINED3DSIH_LOG                   */ shader_hw_log,
5164     /* WINED3DSIH_LOGP                  */ shader_hw_log,
5165     /* WINED3DSIH_LOOP                  */ shader_hw_loop,
5166     /* WINED3DSIH_LRP                   */ shader_hw_lrp,
5167     /* WINED3DSIH_LT                    */ NULL,
5168     /* WINED3DSIH_M3x2                  */ shader_hw_mnxn,
5169     /* WINED3DSIH_M3x3                  */ shader_hw_mnxn,
5170     /* WINED3DSIH_M3x4                  */ shader_hw_mnxn,
5171     /* WINED3DSIH_M4x3                  */ shader_hw_mnxn,
5172     /* WINED3DSIH_M4x4                  */ shader_hw_mnxn,
5173     /* WINED3DSIH_MAD                   */ shader_hw_map2gl,
5174     /* WINED3DSIH_MAX                   */ shader_hw_map2gl,
5175     /* WINED3DSIH_MIN                   */ shader_hw_map2gl,
5176     /* WINED3DSIH_MOV                   */ shader_hw_mov,
5177     /* WINED3DSIH_MOVA                  */ shader_hw_mov,
5178     /* WINED3DSIH_MOVC                  */ NULL,
5179     /* WINED3DSIH_MUL                   */ shader_hw_map2gl,
5180     /* WINED3DSIH_NOP                   */ shader_hw_nop,
5181     /* WINED3DSIH_NRM                   */ shader_hw_nrm,
5182     /* WINED3DSIH_PHASE                 */ shader_hw_nop,
5183     /* WINED3DSIH_POW                   */ shader_hw_pow,
5184     /* WINED3DSIH_RCP                   */ shader_hw_rcp,
5185     /* WINED3DSIH_REP                   */ shader_hw_rep,
5186     /* WINED3DSIH_RET                   */ shader_hw_ret,
5187     /* WINED3DSIH_ROUND_NI              */ NULL,
5188     /* WINED3DSIH_RSQ                   */ shader_hw_scalar_op,
5189     /* WINED3DSIH_SAMPLE                */ NULL,
5190     /* WINED3DSIH_SAMPLE_GRAD           */ NULL,
5191     /* WINED3DSIH_SAMPLE_LOD            */ NULL,
5192     /* WINED3DSIH_SETP                  */ NULL,
5193     /* WINED3DSIH_SGE                   */ shader_hw_map2gl,
5194     /* WINED3DSIH_SGN                   */ shader_hw_sgn,
5195     /* WINED3DSIH_SINCOS                */ shader_hw_sincos,
5196     /* WINED3DSIH_SLT                   */ shader_hw_map2gl,
5197     /* WINED3DSIH_SQRT                  */ NULL,
5198     /* WINED3DSIH_SUB                   */ shader_hw_map2gl,
5199     /* WINED3DSIH_TEX                   */ pshader_hw_tex,
5200     /* WINED3DSIH_TEXBEM                */ pshader_hw_texbem,
5201     /* WINED3DSIH_TEXBEML               */ pshader_hw_texbem,
5202     /* WINED3DSIH_TEXCOORD              */ pshader_hw_texcoord,
5203     /* WINED3DSIH_TEXDEPTH              */ pshader_hw_texdepth,
5204     /* WINED3DSIH_TEXDP3                */ pshader_hw_texdp3,
5205     /* WINED3DSIH_TEXDP3TEX             */ pshader_hw_texdp3tex,
5206     /* WINED3DSIH_TEXKILL               */ pshader_hw_texkill,
5207     /* WINED3DSIH_TEXLDD                */ shader_hw_texldd,
5208     /* WINED3DSIH_TEXLDL                */ shader_hw_texldl,
5209     /* WINED3DSIH_TEXM3x2DEPTH          */ pshader_hw_texm3x2depth,
5210     /* WINED3DSIH_TEXM3x2PAD            */ pshader_hw_texm3x2pad,
5211     /* WINED3DSIH_TEXM3x2TEX            */ pshader_hw_texm3x2tex,
5212     /* WINED3DSIH_TEXM3x3               */ pshader_hw_texm3x3,
5213     /* WINED3DSIH_TEXM3x3DIFF           */ NULL,
5214     /* WINED3DSIH_TEXM3x3PAD            */ pshader_hw_texm3x3pad,
5215     /* WINED3DSIH_TEXM3x3SPEC           */ pshader_hw_texm3x3spec,
5216     /* WINED3DSIH_TEXM3x3TEX            */ pshader_hw_texm3x3tex,
5217     /* WINED3DSIH_TEXM3x3VSPEC          */ pshader_hw_texm3x3vspec,
5218     /* WINED3DSIH_TEXREG2AR             */ pshader_hw_texreg2ar,
5219     /* WINED3DSIH_TEXREG2GB             */ pshader_hw_texreg2gb,
5220     /* WINED3DSIH_TEXREG2RGB            */ pshader_hw_texreg2rgb,
5221     /* WINED3DSIH_UDIV                  */ NULL,
5222     /* WINED3DSIH_USHR                  */ NULL,
5223     /* WINED3DSIH_UTOF                  */ NULL,
5224     /* WINED3DSIH_XOR                   */ NULL,
5225 };
5226
5227 static BOOL get_bool_const(const struct wined3d_shader_instruction *ins,
5228         const struct wined3d_shader *shader, DWORD idx)
5229 {
5230     const struct wined3d_shader_reg_maps *reg_maps = ins->ctx->reg_maps;
5231     BOOL vshader = shader_is_vshader_version(reg_maps->shader_version.type);
5232     const struct wined3d_shader_lconst *constant;
5233     WORD bools = 0;
5234     WORD flag = (1 << idx);
5235     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
5236
5237     if (reg_maps->local_bool_consts & flag)
5238     {
5239         /* What good is a if(bool) with a hardcoded local constant? I don't know, but handle it */
5240         LIST_FOR_EACH_ENTRY(constant, &shader->constantsB, struct wined3d_shader_lconst, entry)
5241         {
5242             if (constant->idx == idx)
5243             {
5244                 return constant->value[0];
5245             }
5246         }
5247         ERR("Local constant not found\n");
5248         return FALSE;
5249     }
5250     else
5251     {
5252         if(vshader) bools = priv->cur_vs_args->clip.boolclip.bools;
5253         else bools = priv->cur_ps_args->bools;
5254         return bools & flag;
5255     }
5256 }
5257
5258 static void get_loop_control_const(const struct wined3d_shader_instruction *ins,
5259         const struct wined3d_shader *shader, UINT idx, struct wined3d_shader_loop_control *loop_control)
5260 {
5261     const struct wined3d_shader_reg_maps *reg_maps = ins->ctx->reg_maps;
5262     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
5263
5264     /* Integer constants can either be a local constant, or they can be stored in the shader
5265      * type specific compile args. */
5266     if (reg_maps->local_int_consts & (1 << idx))
5267     {
5268         const struct wined3d_shader_lconst *constant;
5269
5270         LIST_FOR_EACH_ENTRY(constant, &shader->constantsI, struct wined3d_shader_lconst, entry)
5271         {
5272             if (constant->idx == idx)
5273             {
5274                 loop_control->count = constant->value[0];
5275                 loop_control->start = constant->value[1];
5276                 /* Step is signed. */
5277                 loop_control->step = (int)constant->value[2];
5278                 return;
5279             }
5280         }
5281         /* If this happens the flag was set incorrectly */
5282         ERR("Local constant not found\n");
5283         loop_control->count = 0;
5284         loop_control->start = 0;
5285         loop_control->step = 0;
5286         return;
5287     }
5288
5289     switch (reg_maps->shader_version.type)
5290     {
5291         case WINED3D_SHADER_TYPE_VERTEX:
5292             /* Count and aL start value are unsigned */
5293             loop_control->count = priv->cur_vs_args->loop_ctrl[idx][0];
5294             loop_control->start = priv->cur_vs_args->loop_ctrl[idx][1];
5295             /* Step is signed. */
5296             loop_control->step = ((char)priv->cur_vs_args->loop_ctrl[idx][2]);
5297             break;
5298
5299         case WINED3D_SHADER_TYPE_PIXEL:
5300             loop_control->count = priv->cur_ps_args->loop_ctrl[idx][0];
5301             loop_control->start = priv->cur_ps_args->loop_ctrl[idx][1];
5302             loop_control->step = ((char)priv->cur_ps_args->loop_ctrl[idx][2]);
5303             break;
5304
5305         default:
5306             FIXME("Unhandled shader type %#x.\n", reg_maps->shader_version.type);
5307             break;
5308     }
5309 }
5310
5311 static void record_instruction(struct list *list, const struct wined3d_shader_instruction *ins)
5312 {
5313     unsigned int i;
5314     struct wined3d_shader_dst_param *dst_param = NULL;
5315     struct wined3d_shader_src_param *src_param = NULL, *rel_addr = NULL;
5316     struct recorded_instruction *rec = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*rec));
5317     if(!rec)
5318     {
5319         ERR("Out of memory\n");
5320         return;
5321     }
5322
5323     rec->ins = *ins;
5324     dst_param = HeapAlloc(GetProcessHeap(), 0, sizeof(*dst_param));
5325     if(!dst_param) goto free;
5326     *dst_param = *ins->dst;
5327     if (ins->dst->reg.idx[0].rel_addr)
5328     {
5329         rel_addr = HeapAlloc(GetProcessHeap(), 0, sizeof(*rel_addr));
5330         if (!rel_addr)
5331             goto free;
5332         *rel_addr = *ins->dst->reg.idx[0].rel_addr;
5333         dst_param->reg.idx[0].rel_addr = rel_addr;
5334     }
5335     rec->ins.dst = dst_param;
5336
5337     src_param = HeapAlloc(GetProcessHeap(), 0, sizeof(*src_param) * ins->src_count);
5338     if (!src_param)
5339         goto free;
5340     for (i = 0; i < ins->src_count; ++i)
5341     {
5342         src_param[i] = ins->src[i];
5343         if (ins->src[i].reg.idx[0].rel_addr)
5344         {
5345             rel_addr = HeapAlloc(GetProcessHeap(), 0, sizeof(*rel_addr));
5346             if (!rel_addr)
5347                 goto free;
5348             *rel_addr = *ins->src[i].reg.idx[0].rel_addr;
5349             src_param[i].reg.idx[0].rel_addr = rel_addr;
5350         }
5351     }
5352     rec->ins.src = src_param;
5353     list_add_tail(list, &rec->entry);
5354     return;
5355
5356 free:
5357     ERR("Out of memory\n");
5358     if(dst_param)
5359     {
5360         HeapFree(GetProcessHeap(), 0, (void *)dst_param->reg.idx[0].rel_addr);
5361         HeapFree(GetProcessHeap(), 0, dst_param);
5362     }
5363     if(src_param)
5364     {
5365         for(i = 0; i < ins->src_count; i++)
5366         {
5367             HeapFree(GetProcessHeap(), 0, (void *)src_param[i].reg.idx[0].rel_addr);
5368         }
5369         HeapFree(GetProcessHeap(), 0, src_param);
5370     }
5371     HeapFree(GetProcessHeap(), 0, rec);
5372 }
5373
5374 static void free_recorded_instruction(struct list *list)
5375 {
5376     struct recorded_instruction *rec_ins, *entry2;
5377     unsigned int i;
5378
5379     LIST_FOR_EACH_ENTRY_SAFE(rec_ins, entry2, list, struct recorded_instruction, entry)
5380     {
5381         list_remove(&rec_ins->entry);
5382         if (rec_ins->ins.dst)
5383         {
5384             HeapFree(GetProcessHeap(), 0, (void *)rec_ins->ins.dst->reg.idx[0].rel_addr);
5385             HeapFree(GetProcessHeap(), 0, (void *)rec_ins->ins.dst);
5386         }
5387         if (rec_ins->ins.src)
5388         {
5389             for (i = 0; i < rec_ins->ins.src_count; ++i)
5390             {
5391                 HeapFree(GetProcessHeap(), 0, (void *)rec_ins->ins.src[i].reg.idx[0].rel_addr);
5392             }
5393             HeapFree(GetProcessHeap(), 0, (void *)rec_ins->ins.src);
5394         }
5395         HeapFree(GetProcessHeap(), 0, rec_ins);
5396     }
5397 }
5398
5399 static void shader_arb_handle_instruction(const struct wined3d_shader_instruction *ins) {
5400     SHADER_HANDLER hw_fct;
5401     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
5402     const struct wined3d_shader *shader = ins->ctx->shader;
5403     struct control_frame *control_frame;
5404     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
5405     BOOL bool_const;
5406
5407     if(ins->handler_idx == WINED3DSIH_LOOP || ins->handler_idx == WINED3DSIH_REP)
5408     {
5409         control_frame = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*control_frame));
5410         list_add_head(&priv->control_frames, &control_frame->entry);
5411
5412         if(ins->handler_idx == WINED3DSIH_LOOP) control_frame->type = LOOP;
5413         if(ins->handler_idx == WINED3DSIH_REP) control_frame->type = REP;
5414
5415         if(priv->target_version >= NV2)
5416         {
5417             control_frame->no.loop = priv->num_loops++;
5418             priv->loop_depth++;
5419         }
5420         else
5421         {
5422             /* Don't bother recording when we're in a not used if branch */
5423             if(priv->muted)
5424             {
5425                 return;
5426             }
5427
5428             if(!priv->recording)
5429             {
5430                 list_init(&priv->record);
5431                 priv->recording = TRUE;
5432                 control_frame->outer_loop = TRUE;
5433                 get_loop_control_const(ins, shader, ins->src[0].reg.idx[0].offset, &control_frame->loop_control);
5434                 return; /* Instruction is handled */
5435             }
5436             /* Record this loop in the outer loop's recording */
5437         }
5438     }
5439     else if(ins->handler_idx == WINED3DSIH_ENDLOOP || ins->handler_idx == WINED3DSIH_ENDREP)
5440     {
5441         if(priv->target_version >= NV2)
5442         {
5443             /* Nothing to do. The control frame is popped after the HW instr handler */
5444         }
5445         else
5446         {
5447             struct list *e = list_head(&priv->control_frames);
5448             control_frame = LIST_ENTRY(e, struct control_frame, entry);
5449             list_remove(&control_frame->entry);
5450
5451             if(control_frame->outer_loop)
5452             {
5453                 unsigned int iteration;
5454                 int aL = 0;
5455                 struct list copy;
5456
5457                 /* Turn off recording before playback */
5458                 priv->recording = FALSE;
5459
5460                 /* Move the recorded instructions to a separate list and get them out of the private data
5461                  * structure. If there are nested loops, the shader_arb_handle_instruction below will
5462                  * be recorded again, thus priv->record might be overwritten
5463                  */
5464                 list_init(&copy);
5465                 list_move_tail(&copy, &priv->record);
5466                 list_init(&priv->record);
5467
5468                 if(ins->handler_idx == WINED3DSIH_ENDLOOP)
5469                 {
5470                     shader_addline(buffer, "#unrolling loop: %u iterations, aL=%u, inc %d\n",
5471                                    control_frame->loop_control.count, control_frame->loop_control.start,
5472                                    control_frame->loop_control.step);
5473                     aL = control_frame->loop_control.start;
5474                 }
5475                 else
5476                 {
5477                     shader_addline(buffer, "#unrolling rep: %u iterations\n", control_frame->loop_control.count);
5478                 }
5479
5480                 for (iteration = 0; iteration < control_frame->loop_control.count; ++iteration)
5481                 {
5482                     struct recorded_instruction *rec_ins;
5483                     if(ins->handler_idx == WINED3DSIH_ENDLOOP)
5484                     {
5485                         priv->aL = aL;
5486                         shader_addline(buffer, "#Iteration %u, aL=%d\n", iteration, aL);
5487                     }
5488                     else
5489                     {
5490                         shader_addline(buffer, "#Iteration %u\n", iteration);
5491                     }
5492
5493                     LIST_FOR_EACH_ENTRY(rec_ins, &copy, struct recorded_instruction, entry)
5494                     {
5495                         shader_arb_handle_instruction(&rec_ins->ins);
5496                     }
5497
5498                     if(ins->handler_idx == WINED3DSIH_ENDLOOP)
5499                     {
5500                         aL += control_frame->loop_control.step;
5501                     }
5502                 }
5503                 shader_addline(buffer, "#end loop/rep\n");
5504
5505                 free_recorded_instruction(&copy);
5506                 HeapFree(GetProcessHeap(), 0, control_frame);
5507                 return; /* Instruction is handled */
5508             }
5509             else
5510             {
5511                 /* This is a nested loop. Proceed to the normal recording function */
5512                 HeapFree(GetProcessHeap(), 0, control_frame);
5513             }
5514         }
5515     }
5516
5517     if(priv->recording)
5518     {
5519         record_instruction(&priv->record, ins);
5520         return;
5521     }
5522
5523     /* boolean if */
5524     if(ins->handler_idx == WINED3DSIH_IF)
5525     {
5526         control_frame = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*control_frame));
5527         list_add_head(&priv->control_frames, &control_frame->entry);
5528         control_frame->type = IF;
5529
5530         bool_const = get_bool_const(ins, shader, ins->src[0].reg.idx[0].offset);
5531         if (ins->src[0].modifiers == WINED3DSPSM_NOT)
5532             bool_const = !bool_const;
5533         if (!priv->muted && !bool_const)
5534         {
5535             shader_addline(buffer, "#if(FALSE){\n");
5536             priv->muted = TRUE;
5537             control_frame->muting = TRUE;
5538         }
5539         else shader_addline(buffer, "#if(TRUE) {\n");
5540
5541         return; /* Instruction is handled */
5542     }
5543     else if(ins->handler_idx == WINED3DSIH_IFC)
5544     {
5545         /* IF(bool) and if_cond(a, b) use the same ELSE and ENDIF tokens */
5546         control_frame = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*control_frame));
5547         control_frame->type = IFC;
5548         control_frame->no.ifc = priv->num_ifcs++;
5549         list_add_head(&priv->control_frames, &control_frame->entry);
5550     }
5551     else if(ins->handler_idx == WINED3DSIH_ELSE)
5552     {
5553         struct list *e = list_head(&priv->control_frames);
5554         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5555
5556         if(control_frame->type == IF)
5557         {
5558             shader_addline(buffer, "#} else {\n");
5559             if(!priv->muted && !control_frame->muting)
5560             {
5561                 priv->muted = TRUE;
5562                 control_frame->muting = TRUE;
5563             }
5564             else if(control_frame->muting) priv->muted = FALSE;
5565             return; /* Instruction is handled. */
5566         }
5567         /* In case of an ifc, generate a HW shader instruction */
5568     }
5569     else if(ins->handler_idx == WINED3DSIH_ENDIF)
5570     {
5571         struct list *e = list_head(&priv->control_frames);
5572         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5573
5574         if(control_frame->type == IF)
5575         {
5576             shader_addline(buffer, "#} endif\n");
5577             if(control_frame->muting) priv->muted = FALSE;
5578             list_remove(&control_frame->entry);
5579             HeapFree(GetProcessHeap(), 0, control_frame);
5580             return; /* Instruction is handled */
5581         }
5582     }
5583
5584     if(priv->muted) return;
5585
5586     /* Select handler */
5587     hw_fct = shader_arb_instruction_handler_table[ins->handler_idx];
5588
5589     /* Unhandled opcode */
5590     if (!hw_fct)
5591     {
5592         FIXME("Backend can't handle opcode %#x\n", ins->handler_idx);
5593         return;
5594     }
5595     hw_fct(ins);
5596
5597     if(ins->handler_idx == WINED3DSIH_ENDLOOP || ins->handler_idx == WINED3DSIH_ENDREP)
5598     {
5599         struct list *e = list_head(&priv->control_frames);
5600         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5601         list_remove(&control_frame->entry);
5602         HeapFree(GetProcessHeap(), 0, control_frame);
5603         priv->loop_depth--;
5604     }
5605     else if(ins->handler_idx == WINED3DSIH_ENDIF)
5606     {
5607         /* Non-ifc ENDIFs don't reach that place because of the return in the if block above */
5608         struct list *e = list_head(&priv->control_frames);
5609         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5610         list_remove(&control_frame->entry);
5611         HeapFree(GetProcessHeap(), 0, control_frame);
5612     }
5613
5614
5615     shader_arb_add_instruction_modifiers(ins);
5616 }
5617
5618 static BOOL shader_arb_has_ffp_proj_control(void *shader_priv)
5619 {
5620     struct shader_arb_priv *priv = shader_priv;
5621
5622     return priv->fragment_pipe->ffp_proj_control;
5623 }
5624
5625 const struct wined3d_shader_backend_ops arb_program_shader_backend =
5626 {
5627     shader_arb_handle_instruction,
5628     shader_arb_select,
5629     shader_arb_select_depth_blt,
5630     shader_arb_deselect_depth_blt,
5631     shader_arb_update_float_vertex_constants,
5632     shader_arb_update_float_pixel_constants,
5633     shader_arb_load_constants,
5634     shader_arb_load_np2fixup_constants,
5635     shader_arb_destroy,
5636     shader_arb_alloc,
5637     shader_arb_free,
5638     shader_arb_context_destroyed,
5639     shader_arb_get_caps,
5640     shader_arb_color_fixup_supported,
5641     shader_arb_has_ffp_proj_control,
5642 };
5643
5644 /* ARB_fragment_program fixed function pipeline replacement definitions */
5645 #define ARB_FFP_CONST_TFACTOR           0
5646 #define ARB_FFP_CONST_SPECULAR_ENABLE   ((ARB_FFP_CONST_TFACTOR) + 1)
5647 #define ARB_FFP_CONST_CONSTANT(i)       ((ARB_FFP_CONST_SPECULAR_ENABLE) + 1 + i)
5648 #define ARB_FFP_CONST_BUMPMAT(i)        ((ARB_FFP_CONST_CONSTANT(7)) + 1 + i)
5649 #define ARB_FFP_CONST_LUMINANCE(i)      ((ARB_FFP_CONST_BUMPMAT(7)) + 1 + i)
5650
5651 struct arbfp_ffp_desc
5652 {
5653     struct ffp_frag_desc parent;
5654     GLuint shader;
5655 };
5656
5657 /* Context activation and GL locking are done by the caller. */
5658 static void arbfp_enable(const struct wined3d_gl_info *gl_info, BOOL enable)
5659 {
5660     if (enable)
5661     {
5662         gl_info->gl_ops.gl.p_glEnable(GL_FRAGMENT_PROGRAM_ARB);
5663         checkGLcall("glEnable(GL_FRAGMENT_PROGRAM_ARB)");
5664     }
5665     else
5666     {
5667         gl_info->gl_ops.gl.p_glDisable(GL_FRAGMENT_PROGRAM_ARB);
5668         checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
5669     }
5670 }
5671
5672 static void *arbfp_alloc(const struct wined3d_shader_backend_ops *shader_backend, void *shader_priv)
5673 {
5674     struct shader_arb_priv *priv;
5675
5676     /* Share private data between the shader backend and the pipeline
5677      * replacement, if both are the arb implementation. This is needed to
5678      * figure out whether ARBfp should be disabled if no pixel shader is bound
5679      * or not. */
5680     if (shader_backend == &arb_program_shader_backend)
5681         priv = shader_priv;
5682     else if (!(priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*priv))))
5683         return NULL;
5684
5685     if (wine_rb_init(&priv->fragment_shaders, &wined3d_ffp_frag_program_rb_functions) == -1)
5686     {
5687         ERR("Failed to initialize rbtree.\n");
5688         if (priv != shader_priv)
5689             HeapFree(GetProcessHeap(), 0, priv);
5690         return NULL;
5691     }
5692     priv->use_arbfp_fixed_func = TRUE;
5693
5694     return priv;
5695 }
5696
5697 /* Context activation is done by the caller. */
5698 static void arbfp_free_ffpshader(struct wine_rb_entry *entry, void *context)
5699 {
5700     const struct wined3d_gl_info *gl_info = context;
5701     struct arbfp_ffp_desc *entry_arb = WINE_RB_ENTRY_VALUE(entry, struct arbfp_ffp_desc, parent.entry);
5702
5703     ENTER_GL();
5704     GL_EXTCALL(glDeleteProgramsARB(1, &entry_arb->shader));
5705     checkGLcall("glDeleteProgramsARB(1, &entry_arb->shader)");
5706     HeapFree(GetProcessHeap(), 0, entry_arb);
5707     LEAVE_GL();
5708 }
5709
5710 /* Context activation is done by the caller. */
5711 static void arbfp_free(struct wined3d_device *device)
5712 {
5713     struct shader_arb_priv *priv = device->fragment_priv;
5714
5715     wine_rb_destroy(&priv->fragment_shaders, arbfp_free_ffpshader, &device->adapter->gl_info);
5716     priv->use_arbfp_fixed_func = FALSE;
5717
5718     if (device->shader_backend != &arb_program_shader_backend)
5719     {
5720         HeapFree(GetProcessHeap(), 0, device->fragment_priv);
5721     }
5722 }
5723
5724 static void arbfp_get_caps(const struct wined3d_gl_info *gl_info, struct fragment_caps *caps)
5725 {
5726     caps->PrimitiveMiscCaps = WINED3DPMISCCAPS_TSSARGTEMP;
5727     caps->TextureOpCaps =  WINED3DTEXOPCAPS_DISABLE                     |
5728                            WINED3DTEXOPCAPS_SELECTARG1                  |
5729                            WINED3DTEXOPCAPS_SELECTARG2                  |
5730                            WINED3DTEXOPCAPS_MODULATE4X                  |
5731                            WINED3DTEXOPCAPS_MODULATE2X                  |
5732                            WINED3DTEXOPCAPS_MODULATE                    |
5733                            WINED3DTEXOPCAPS_ADDSIGNED2X                 |
5734                            WINED3DTEXOPCAPS_ADDSIGNED                   |
5735                            WINED3DTEXOPCAPS_ADD                         |
5736                            WINED3DTEXOPCAPS_SUBTRACT                    |
5737                            WINED3DTEXOPCAPS_ADDSMOOTH                   |
5738                            WINED3DTEXOPCAPS_BLENDCURRENTALPHA           |
5739                            WINED3DTEXOPCAPS_BLENDFACTORALPHA            |
5740                            WINED3DTEXOPCAPS_BLENDTEXTUREALPHA           |
5741                            WINED3DTEXOPCAPS_BLENDDIFFUSEALPHA           |
5742                            WINED3DTEXOPCAPS_BLENDTEXTUREALPHAPM         |
5743                            WINED3DTEXOPCAPS_MODULATEALPHA_ADDCOLOR      |
5744                            WINED3DTEXOPCAPS_MODULATECOLOR_ADDALPHA      |
5745                            WINED3DTEXOPCAPS_MODULATEINVCOLOR_ADDALPHA   |
5746                            WINED3DTEXOPCAPS_MODULATEINVALPHA_ADDCOLOR   |
5747                            WINED3DTEXOPCAPS_DOTPRODUCT3                 |
5748                            WINED3DTEXOPCAPS_MULTIPLYADD                 |
5749                            WINED3DTEXOPCAPS_LERP                        |
5750                            WINED3DTEXOPCAPS_BUMPENVMAP                  |
5751                            WINED3DTEXOPCAPS_BUMPENVMAPLUMINANCE;
5752
5753     /* TODO: Implement WINED3DTEXOPCAPS_PREMODULATE */
5754
5755     caps->MaxTextureBlendStages   = 8;
5756     caps->MaxSimultaneousTextures = min(gl_info->limits.fragment_samplers, 8);
5757 }
5758
5759 static void state_texfactor_arbfp(struct wined3d_context *context,
5760         const struct wined3d_state *state, DWORD state_id)
5761 {
5762     struct wined3d_device *device = context->swapchain->device;
5763     const struct wined3d_gl_info *gl_info = context->gl_info;
5764     float col[4];
5765
5766     /* Don't load the parameter if we're using an arbfp pixel shader,
5767      * otherwise we'll overwrite application provided constants. */
5768     if (device->shader_backend == &arb_program_shader_backend)
5769     {
5770         struct shader_arb_priv *priv;
5771
5772         if (use_ps(state)) return;
5773
5774         priv = device->shader_priv;
5775         priv->pshader_const_dirty[ARB_FFP_CONST_TFACTOR] = 1;
5776         priv->highest_dirty_ps_const = max(priv->highest_dirty_ps_const, ARB_FFP_CONST_TFACTOR + 1);
5777     }
5778
5779     D3DCOLORTOGLFLOAT4(state->render_states[WINED3D_RS_TEXTUREFACTOR], col);
5780     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_TFACTOR, col));
5781     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_TFACTOR, col)");
5782 }
5783
5784 static void state_arb_specularenable(struct wined3d_context *context,
5785         const struct wined3d_state *state, DWORD state_id)
5786 {
5787     struct wined3d_device *device = context->swapchain->device;
5788     const struct wined3d_gl_info *gl_info = context->gl_info;
5789     float col[4];
5790
5791     /* Don't load the parameter if we're using an arbfp pixel shader, otherwise we'll overwrite
5792      * application provided constants
5793      */
5794     if (device->shader_backend == &arb_program_shader_backend)
5795     {
5796         struct shader_arb_priv *priv;
5797
5798         if (use_ps(state)) return;
5799
5800         priv = device->shader_priv;
5801         priv->pshader_const_dirty[ARB_FFP_CONST_SPECULAR_ENABLE] = 1;
5802         priv->highest_dirty_ps_const = max(priv->highest_dirty_ps_const, ARB_FFP_CONST_SPECULAR_ENABLE + 1);
5803     }
5804
5805     if (state->render_states[WINED3D_RS_SPECULARENABLE])
5806     {
5807         /* The specular color has no alpha */
5808         col[0] = 1.0f; col[1] = 1.0f;
5809         col[2] = 1.0f; col[3] = 0.0f;
5810     } else {
5811         col[0] = 0.0f; col[1] = 0.0f;
5812         col[2] = 0.0f; col[3] = 0.0f;
5813     }
5814     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_SPECULAR_ENABLE, col));
5815     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_SPECULAR_ENABLE, col)");
5816 }
5817
5818 static void set_bumpmat_arbfp(struct wined3d_context *context, const struct wined3d_state *state, DWORD state_id)
5819 {
5820     DWORD stage = (state_id - STATE_TEXTURESTAGE(0, 0)) / (WINED3D_HIGHEST_TEXTURE_STATE + 1);
5821     struct wined3d_device *device = context->swapchain->device;
5822     const struct wined3d_gl_info *gl_info = context->gl_info;
5823     float mat[2][2];
5824
5825     if (use_ps(state))
5826     {
5827         /* The pixel shader has to know the bump env matrix. Do a constants
5828          * update. */
5829         if (stage && (state->pixel_shader->reg_maps.bumpmat & (1 << stage)))
5830             context->load_constants = 1;
5831
5832         if(device->shader_backend == &arb_program_shader_backend) {
5833             /* Exit now, don't set the bumpmat below, otherwise we may overwrite pixel shader constants */
5834             return;
5835         }
5836     }
5837     else if (device->shader_backend == &arb_program_shader_backend)
5838     {
5839         struct shader_arb_priv *priv = device->shader_priv;
5840         priv->pshader_const_dirty[ARB_FFP_CONST_BUMPMAT(stage)] = 1;
5841         priv->highest_dirty_ps_const = max(priv->highest_dirty_ps_const, ARB_FFP_CONST_BUMPMAT(stage) + 1);
5842     }
5843
5844     mat[0][0] = *((float *)&state->texture_states[stage][WINED3D_TSS_BUMPENV_MAT00]);
5845     mat[0][1] = *((float *)&state->texture_states[stage][WINED3D_TSS_BUMPENV_MAT01]);
5846     mat[1][0] = *((float *)&state->texture_states[stage][WINED3D_TSS_BUMPENV_MAT10]);
5847     mat[1][1] = *((float *)&state->texture_states[stage][WINED3D_TSS_BUMPENV_MAT11]);
5848
5849     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_BUMPMAT(stage), &mat[0][0]));
5850     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_BUMPMAT(stage), &mat[0][0])");
5851 }
5852
5853 static void tex_bumpenvlum_arbfp(struct wined3d_context *context,
5854         const struct wined3d_state *state, DWORD state_id)
5855 {
5856     DWORD stage = (state_id - STATE_TEXTURESTAGE(0, 0)) / (WINED3D_HIGHEST_TEXTURE_STATE + 1);
5857     struct wined3d_device *device = context->swapchain->device;
5858     const struct wined3d_gl_info *gl_info = context->gl_info;
5859     float param[4];
5860
5861     if (use_ps(state))
5862     {
5863         /* The pixel shader has to know the luminance offset. Do a constants
5864          * update. */
5865         if (stage && (state->pixel_shader->reg_maps.luminanceparams & (1 << stage)))
5866             context->load_constants = 1;
5867
5868         if(device->shader_backend == &arb_program_shader_backend) {
5869             /* Exit now, don't set the bumpmat below, otherwise we may overwrite pixel shader constants */
5870             return;
5871         }
5872     }
5873     else if (device->shader_backend == &arb_program_shader_backend)
5874     {
5875         struct shader_arb_priv *priv = device->shader_priv;
5876         priv->pshader_const_dirty[ARB_FFP_CONST_LUMINANCE(stage)] = 1;
5877         priv->highest_dirty_ps_const = max(priv->highest_dirty_ps_const, ARB_FFP_CONST_LUMINANCE(stage) + 1);
5878     }
5879
5880     param[0] = *((float *)&state->texture_states[stage][WINED3D_TSS_BUMPENV_LSCALE]);
5881     param[1] = *((float *)&state->texture_states[stage][WINED3D_TSS_BUMPENV_LOFFSET]);
5882     param[2] = 0.0f;
5883     param[3] = 0.0f;
5884
5885     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_LUMINANCE(stage), param));
5886     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_LUMINANCE(stage), param)");
5887 }
5888
5889 static const char *get_argreg(struct wined3d_shader_buffer *buffer, DWORD argnum, unsigned int stage, DWORD arg)
5890 {
5891     const char *ret;
5892
5893     if(arg == ARG_UNUSED) return "unused"; /* This is the marker for unused registers */
5894
5895     switch(arg & WINED3DTA_SELECTMASK) {
5896         case WINED3DTA_DIFFUSE:
5897             ret = "fragment.color.primary"; break;
5898
5899         case WINED3DTA_CURRENT:
5900             if (!stage) ret = "fragment.color.primary";
5901             else ret = "ret";
5902             break;
5903
5904         case WINED3DTA_TEXTURE:
5905             switch(stage) {
5906                 case 0: ret = "tex0"; break;
5907                 case 1: ret = "tex1"; break;
5908                 case 2: ret = "tex2"; break;
5909                 case 3: ret = "tex3"; break;
5910                 case 4: ret = "tex4"; break;
5911                 case 5: ret = "tex5"; break;
5912                 case 6: ret = "tex6"; break;
5913                 case 7: ret = "tex7"; break;
5914                 default: ret = "unknown texture";
5915             }
5916             break;
5917
5918         case WINED3DTA_TFACTOR:
5919             ret = "tfactor"; break;
5920
5921         case WINED3DTA_SPECULAR:
5922             ret = "fragment.color.secondary"; break;
5923
5924         case WINED3DTA_TEMP:
5925             ret = "tempreg"; break;
5926
5927         case WINED3DTA_CONSTANT:
5928             FIXME("Implement perstage constants\n");
5929             switch(stage) {
5930                 case 0: ret = "const0"; break;
5931                 case 1: ret = "const1"; break;
5932                 case 2: ret = "const2"; break;
5933                 case 3: ret = "const3"; break;
5934                 case 4: ret = "const4"; break;
5935                 case 5: ret = "const5"; break;
5936                 case 6: ret = "const6"; break;
5937                 case 7: ret = "const7"; break;
5938                 default: ret = "unknown constant";
5939             }
5940             break;
5941
5942         default:
5943             return "unknown";
5944     }
5945
5946     if(arg & WINED3DTA_COMPLEMENT) {
5947         shader_addline(buffer, "SUB arg%u, const.x, %s;\n", argnum, ret);
5948         if(argnum == 0) ret = "arg0";
5949         if(argnum == 1) ret = "arg1";
5950         if(argnum == 2) ret = "arg2";
5951     }
5952     if(arg & WINED3DTA_ALPHAREPLICATE) {
5953         shader_addline(buffer, "MOV arg%u, %s.w;\n", argnum, ret);
5954         if(argnum == 0) ret = "arg0";
5955         if(argnum == 1) ret = "arg1";
5956         if(argnum == 2) ret = "arg2";
5957     }
5958     return ret;
5959 }
5960
5961 static void gen_ffp_instr(struct wined3d_shader_buffer *buffer, unsigned int stage, BOOL color,
5962         BOOL alpha, DWORD dst, DWORD op, DWORD dw_arg0, DWORD dw_arg1, DWORD dw_arg2)
5963 {
5964     const char *dstmask, *dstreg, *arg0, *arg1, *arg2;
5965     unsigned int mul = 1;
5966
5967     if(color && alpha) dstmask = "";
5968     else if(color) dstmask = ".xyz";
5969     else dstmask = ".w";
5970
5971     if(dst == tempreg) dstreg = "tempreg";
5972     else dstreg = "ret";
5973
5974     arg0 = get_argreg(buffer, 0, stage, dw_arg0);
5975     arg1 = get_argreg(buffer, 1, stage, dw_arg1);
5976     arg2 = get_argreg(buffer, 2, stage, dw_arg2);
5977
5978     switch (op)
5979     {
5980         case WINED3D_TOP_DISABLE:
5981             if (!stage)
5982                 shader_addline(buffer, "MOV %s%s, fragment.color.primary;\n", dstreg, dstmask);
5983             break;
5984
5985         case WINED3D_TOP_SELECT_ARG2:
5986             arg1 = arg2;
5987             /* FALLTHROUGH */
5988         case WINED3D_TOP_SELECT_ARG1:
5989             shader_addline(buffer, "MOV %s%s, %s;\n", dstreg, dstmask, arg1);
5990             break;
5991
5992         case WINED3D_TOP_MODULATE_4X:
5993             mul = 2;
5994             /* FALLTHROUGH */
5995         case WINED3D_TOP_MODULATE_2X:
5996             mul *= 2;
5997             /* FALLTHROUGH */
5998         case WINED3D_TOP_MODULATE:
5999             shader_addline(buffer, "MUL %s%s, %s, %s;\n", dstreg, dstmask, arg1, arg2);
6000             break;
6001
6002         case WINED3D_TOP_ADD_SIGNED_2X:
6003             mul = 2;
6004             /* FALLTHROUGH */
6005         case WINED3D_TOP_ADD_SIGNED:
6006             shader_addline(buffer, "SUB arg2, %s, const.w;\n", arg2);
6007             arg2 = "arg2";
6008             /* FALLTHROUGH */
6009         case WINED3D_TOP_ADD:
6010             shader_addline(buffer, "ADD_SAT %s%s, %s, %s;\n", dstreg, dstmask, arg1, arg2);
6011             break;
6012
6013         case WINED3D_TOP_SUBTRACT:
6014             shader_addline(buffer, "SUB_SAT %s%s, %s, %s;\n", dstreg, dstmask, arg1, arg2);
6015             break;
6016
6017         case WINED3D_TOP_ADD_SMOOTH:
6018             shader_addline(buffer, "SUB arg1, const.x, %s;\n", arg1);
6019             shader_addline(buffer, "MAD_SAT %s%s, arg1, %s, %s;\n", dstreg, dstmask, arg2, arg1);
6020             break;
6021
6022         case WINED3D_TOP_BLEND_CURRENT_ALPHA:
6023             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_CURRENT);
6024             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
6025             break;
6026         case WINED3D_TOP_BLEND_FACTOR_ALPHA:
6027             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_TFACTOR);
6028             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
6029             break;
6030         case WINED3D_TOP_BLEND_TEXTURE_ALPHA:
6031             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_TEXTURE);
6032             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
6033             break;
6034         case WINED3D_TOP_BLEND_DIFFUSE_ALPHA:
6035             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_DIFFUSE);
6036             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
6037             break;
6038
6039         case WINED3D_TOP_BLEND_TEXTURE_ALPHA_PM:
6040             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_TEXTURE);
6041             shader_addline(buffer, "SUB arg0.w, const.x, %s.w;\n", arg0);
6042             shader_addline(buffer, "MAD_SAT %s%s, %s, arg0.w, %s;\n", dstreg, dstmask, arg2, arg1);
6043             break;
6044
6045         /* D3DTOP_PREMODULATE ???? */
6046
6047         case WINED3D_TOP_MODULATE_INVALPHA_ADD_COLOR:
6048             shader_addline(buffer, "SUB arg0.w, const.x, %s;\n", arg1);
6049             shader_addline(buffer, "MAD_SAT %s%s, arg0.w, %s, %s;\n", dstreg, dstmask, arg2, arg1);
6050             break;
6051         case WINED3D_TOP_MODULATE_ALPHA_ADD_COLOR:
6052             shader_addline(buffer, "MAD_SAT %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg1, arg2, arg1);
6053             break;
6054         case WINED3D_TOP_MODULATE_INVCOLOR_ADD_ALPHA:
6055             shader_addline(buffer, "SUB arg0, const.x, %s;\n", arg1);
6056             shader_addline(buffer, "MAD_SAT %s%s, arg0, %s, %s.w;\n", dstreg, dstmask, arg2, arg1);
6057             break;
6058         case WINED3D_TOP_MODULATE_COLOR_ADD_ALPHA:
6059             shader_addline(buffer, "MAD_SAT %s%s, %s, %s, %s.w;\n", dstreg, dstmask, arg1, arg2, arg1);
6060             break;
6061
6062         case WINED3D_TOP_DOTPRODUCT3:
6063             mul = 4;
6064             shader_addline(buffer, "SUB arg1, %s, const.w;\n", arg1);
6065             shader_addline(buffer, "SUB arg2, %s, const.w;\n", arg2);
6066             shader_addline(buffer, "DP3_SAT %s%s, arg1, arg2;\n", dstreg, dstmask);
6067             break;
6068
6069         case WINED3D_TOP_MULTIPLY_ADD:
6070             shader_addline(buffer, "MAD_SAT %s%s, %s, %s, %s;\n", dstreg, dstmask, arg1, arg2, arg0);
6071             break;
6072
6073         case WINED3D_TOP_LERP:
6074             /* The msdn is not quite right here */
6075             shader_addline(buffer, "LRP %s%s, %s, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
6076             break;
6077
6078         case WINED3D_TOP_BUMPENVMAP:
6079         case WINED3D_TOP_BUMPENVMAP_LUMINANCE:
6080             /* Those are handled in the first pass of the shader(generation pass 1 and 2) already */
6081             break;
6082
6083         default:
6084             FIXME("Unhandled texture op %08x\n", op);
6085     }
6086
6087     if (mul == 2)
6088         shader_addline(buffer, "MUL_SAT %s%s, %s, const.y;\n", dstreg, dstmask, dstreg);
6089     else if (mul == 4)
6090         shader_addline(buffer, "MUL_SAT %s%s, %s, const.z;\n", dstreg, dstmask, dstreg);
6091 }
6092
6093 static GLuint gen_arbfp_ffp_shader(const struct ffp_frag_settings *settings, const struct wined3d_gl_info *gl_info)
6094 {
6095     unsigned int stage;
6096     struct wined3d_shader_buffer buffer;
6097     BOOL tex_read[MAX_TEXTURES] = {FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE};
6098     BOOL bump_used[MAX_TEXTURES] = {FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE};
6099     BOOL luminance_used[MAX_TEXTURES] = {FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE};
6100     UINT lowest_disabled_stage;
6101     const char *textype;
6102     const char *instr, *sat;
6103     char colorcor_dst[8];
6104     GLuint ret;
6105     DWORD arg0, arg1, arg2;
6106     BOOL tempreg_used = FALSE, tfactor_used = FALSE;
6107     BOOL op_equal;
6108     const char *final_combiner_src = "ret";
6109     GLint pos;
6110
6111     /* Find out which textures are read */
6112     for (stage = 0; stage < MAX_TEXTURES; ++stage)
6113     {
6114         if (settings->op[stage].cop == WINED3D_TOP_DISABLE)
6115             break;
6116         arg0 = settings->op[stage].carg0 & WINED3DTA_SELECTMASK;
6117         arg1 = settings->op[stage].carg1 & WINED3DTA_SELECTMASK;
6118         arg2 = settings->op[stage].carg2 & WINED3DTA_SELECTMASK;
6119         if(arg0 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
6120         if(arg1 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
6121         if(arg2 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
6122
6123         if (settings->op[stage].cop == WINED3D_TOP_BLEND_TEXTURE_ALPHA)
6124             tex_read[stage] = TRUE;
6125         if (settings->op[stage].cop == WINED3D_TOP_BLEND_TEXTURE_ALPHA_PM)
6126             tex_read[stage] = TRUE;
6127         if (settings->op[stage].cop == WINED3D_TOP_BUMPENVMAP)
6128         {
6129             bump_used[stage] = TRUE;
6130             tex_read[stage] = TRUE;
6131         }
6132         if (settings->op[stage].cop == WINED3D_TOP_BUMPENVMAP_LUMINANCE)
6133         {
6134             bump_used[stage] = TRUE;
6135             tex_read[stage] = TRUE;
6136             luminance_used[stage] = TRUE;
6137         }
6138         else if (settings->op[stage].cop == WINED3D_TOP_BLEND_FACTOR_ALPHA)
6139         {
6140             tfactor_used = TRUE;
6141         }
6142
6143         if(arg0 == WINED3DTA_TFACTOR || arg1 == WINED3DTA_TFACTOR || arg2 == WINED3DTA_TFACTOR) {
6144             tfactor_used = TRUE;
6145         }
6146
6147         if(settings->op[stage].dst == tempreg) tempreg_used = TRUE;
6148         if(arg0 == WINED3DTA_TEMP || arg1 == WINED3DTA_TEMP || arg2 == WINED3DTA_TEMP) {
6149             tempreg_used = TRUE;
6150         }
6151
6152         if (settings->op[stage].aop == WINED3D_TOP_DISABLE)
6153             continue;
6154         arg0 = settings->op[stage].aarg0 & WINED3DTA_SELECTMASK;
6155         arg1 = settings->op[stage].aarg1 & WINED3DTA_SELECTMASK;
6156         arg2 = settings->op[stage].aarg2 & WINED3DTA_SELECTMASK;
6157         if(arg0 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
6158         if(arg1 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
6159         if(arg2 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
6160
6161         if(arg0 == WINED3DTA_TEMP || arg1 == WINED3DTA_TEMP || arg2 == WINED3DTA_TEMP) {
6162             tempreg_used = TRUE;
6163         }
6164         if(arg0 == WINED3DTA_TFACTOR || arg1 == WINED3DTA_TFACTOR || arg2 == WINED3DTA_TFACTOR) {
6165             tfactor_used = TRUE;
6166         }
6167     }
6168     lowest_disabled_stage = stage;
6169
6170     /* Shader header */
6171     if (!shader_buffer_init(&buffer))
6172     {
6173         ERR("Failed to initialize shader buffer.\n");
6174         return 0;
6175     }
6176
6177     shader_addline(&buffer, "!!ARBfp1.0\n");
6178
6179     switch(settings->fog) {
6180         case FOG_OFF:                                                         break;
6181         case FOG_LINEAR: shader_addline(&buffer, "OPTION ARB_fog_linear;\n"); break;
6182         case FOG_EXP:    shader_addline(&buffer, "OPTION ARB_fog_exp;\n");    break;
6183         case FOG_EXP2:   shader_addline(&buffer, "OPTION ARB_fog_exp2;\n");   break;
6184         default: FIXME("Unexpected fog setting %d\n", settings->fog);
6185     }
6186
6187     shader_addline(&buffer, "PARAM const = {1, 2, 4, 0.5};\n");
6188     shader_addline(&buffer, "TEMP TMP;\n");
6189     shader_addline(&buffer, "TEMP ret;\n");
6190     if(tempreg_used || settings->sRGB_write) shader_addline(&buffer, "TEMP tempreg;\n");
6191     shader_addline(&buffer, "TEMP arg0;\n");
6192     shader_addline(&buffer, "TEMP arg1;\n");
6193     shader_addline(&buffer, "TEMP arg2;\n");
6194     for(stage = 0; stage < MAX_TEXTURES; stage++) {
6195         if(!tex_read[stage]) continue;
6196         shader_addline(&buffer, "TEMP tex%u;\n", stage);
6197         if(!bump_used[stage]) continue;
6198         shader_addline(&buffer, "PARAM bumpmat%u = program.env[%u];\n", stage, ARB_FFP_CONST_BUMPMAT(stage));
6199         if(!luminance_used[stage]) continue;
6200         shader_addline(&buffer, "PARAM luminance%u = program.env[%u];\n", stage, ARB_FFP_CONST_LUMINANCE(stage));
6201     }
6202     if(tfactor_used) {
6203         shader_addline(&buffer, "PARAM tfactor = program.env[%u];\n", ARB_FFP_CONST_TFACTOR);
6204     }
6205         shader_addline(&buffer, "PARAM specular_enable = program.env[%u];\n", ARB_FFP_CONST_SPECULAR_ENABLE);
6206
6207     if(settings->sRGB_write) {
6208         shader_addline(&buffer, "PARAM srgb_consts1 = {%f, %f, %f, %f};\n",
6209                        srgb_mul_low, srgb_cmp, srgb_pow, srgb_mul_high);
6210         shader_addline(&buffer, "PARAM srgb_consts2 = {%f, %f, %f, %f};\n",
6211                        srgb_sub_high, 0.0, 0.0, 0.0);
6212     }
6213
6214     if (lowest_disabled_stage < 7 && settings->emul_clipplanes)
6215         shader_addline(&buffer, "KIL fragment.texcoord[7];\n");
6216
6217     /* Generate texture sampling instructions) */
6218     for (stage = 0; stage < MAX_TEXTURES && settings->op[stage].cop != WINED3D_TOP_DISABLE; ++stage)
6219     {
6220         if (!tex_read[stage])
6221             continue;
6222
6223         switch(settings->op[stage].tex_type) {
6224             case tex_1d:                    textype = "1D";     break;
6225             case tex_2d:                    textype = "2D";     break;
6226             case tex_3d:                    textype = "3D";     break;
6227             case tex_cube:                  textype = "CUBE";   break;
6228             case tex_rect:                  textype = "RECT";   break;
6229             default: textype = "unexpected_textype";   break;
6230         }
6231
6232         if (settings->op[stage].cop == WINED3D_TOP_BUMPENVMAP
6233                 || settings->op[stage].cop == WINED3D_TOP_BUMPENVMAP_LUMINANCE)
6234             sat = "";
6235         else
6236             sat = "_SAT";
6237
6238         if(settings->op[stage].projected == proj_none) {
6239             instr = "TEX";
6240         } else if(settings->op[stage].projected == proj_count4 ||
6241                   settings->op[stage].projected == proj_count3) {
6242             instr = "TXP";
6243         } else {
6244             FIXME("Unexpected projection mode %d\n", settings->op[stage].projected);
6245             instr = "TXP";
6246         }
6247
6248         if (stage > 0
6249                 && (settings->op[stage - 1].cop == WINED3D_TOP_BUMPENVMAP
6250                 || settings->op[stage - 1].cop == WINED3D_TOP_BUMPENVMAP_LUMINANCE))
6251         {
6252             shader_addline(&buffer, "SWZ arg1, bumpmat%u, x, z, 0, 0;\n", stage - 1);
6253             shader_addline(&buffer, "DP3 ret.x, arg1, tex%u;\n", stage - 1);
6254             shader_addline(&buffer, "SWZ arg1, bumpmat%u, y, w, 0, 0;\n", stage - 1);
6255             shader_addline(&buffer, "DP3 ret.y, arg1, tex%u;\n", stage - 1);
6256
6257             /* with projective textures, texbem only divides the static texture coord, not the displacement,
6258              * so multiply the displacement with the dividing parameter before passing it to TXP
6259              */
6260             if (settings->op[stage].projected != proj_none) {
6261                 if(settings->op[stage].projected == proj_count4) {
6262                     shader_addline(&buffer, "MOV ret.w, fragment.texcoord[%u].w;\n", stage);
6263                     shader_addline(&buffer, "MUL ret.xyz, ret, fragment.texcoord[%u].w, fragment.texcoord[%u];\n", stage, stage);
6264                 } else {
6265                     shader_addline(&buffer, "MOV ret.w, fragment.texcoord[%u].z;\n", stage);
6266                     shader_addline(&buffer, "MAD ret.xyz, ret, fragment.texcoord[%u].z, fragment.texcoord[%u];\n", stage, stage);
6267                 }
6268             } else {
6269                 shader_addline(&buffer, "ADD ret, ret, fragment.texcoord[%u];\n", stage);
6270             }
6271
6272             shader_addline(&buffer, "%s%s tex%u, ret, texture[%u], %s;\n",
6273                     instr, sat, stage, stage, textype);
6274             if (settings->op[stage - 1].cop == WINED3D_TOP_BUMPENVMAP_LUMINANCE)
6275             {
6276                 shader_addline(&buffer, "MAD_SAT ret.x, tex%u.z, luminance%u.x, luminance%u.y;\n",
6277                                stage - 1, stage - 1, stage - 1);
6278                 shader_addline(&buffer, "MUL tex%u, tex%u, ret.x;\n", stage, stage);
6279             }
6280         } else if(settings->op[stage].projected == proj_count3) {
6281             shader_addline(&buffer, "MOV ret, fragment.texcoord[%u];\n", stage);
6282             shader_addline(&buffer, "MOV ret.w, ret.z;\n");
6283             shader_addline(&buffer, "%s%s tex%u, ret, texture[%u], %s;\n",
6284                             instr, sat, stage, stage, textype);
6285         } else {
6286             shader_addline(&buffer, "%s%s tex%u, fragment.texcoord[%u], texture[%u], %s;\n",
6287                             instr, sat, stage, stage, stage, textype);
6288         }
6289
6290         sprintf(colorcor_dst, "tex%u", stage);
6291         gen_color_correction(&buffer, colorcor_dst, WINED3DSP_WRITEMASK_ALL, "const.x", "const.y",
6292                 settings->op[stage].color_fixup);
6293     }
6294
6295     /* Generate the main shader */
6296     for (stage = 0; stage < MAX_TEXTURES; ++stage)
6297     {
6298         if (settings->op[stage].cop == WINED3D_TOP_DISABLE)
6299         {
6300             if (!stage)
6301                 final_combiner_src = "fragment.color.primary";
6302             break;
6303         }
6304
6305         if (settings->op[stage].cop == WINED3D_TOP_SELECT_ARG1
6306                 && settings->op[stage].aop == WINED3D_TOP_SELECT_ARG1)
6307             op_equal = settings->op[stage].carg1 == settings->op[stage].aarg1;
6308         else if (settings->op[stage].cop == WINED3D_TOP_SELECT_ARG1
6309                 && settings->op[stage].aop == WINED3D_TOP_SELECT_ARG2)
6310             op_equal = settings->op[stage].carg1 == settings->op[stage].aarg2;
6311         else if (settings->op[stage].cop == WINED3D_TOP_SELECT_ARG2
6312                 && settings->op[stage].aop == WINED3D_TOP_SELECT_ARG1)
6313             op_equal = settings->op[stage].carg2 == settings->op[stage].aarg1;
6314         else if (settings->op[stage].cop == WINED3D_TOP_SELECT_ARG2
6315                 && settings->op[stage].aop == WINED3D_TOP_SELECT_ARG2)
6316             op_equal = settings->op[stage].carg2 == settings->op[stage].aarg2;
6317         else
6318             op_equal = settings->op[stage].aop   == settings->op[stage].cop
6319                     && settings->op[stage].carg0 == settings->op[stage].aarg0
6320                     && settings->op[stage].carg1 == settings->op[stage].aarg1
6321                     && settings->op[stage].carg2 == settings->op[stage].aarg2;
6322
6323         if (settings->op[stage].aop == WINED3D_TOP_DISABLE)
6324         {
6325             gen_ffp_instr(&buffer, stage, TRUE, FALSE, settings->op[stage].dst,
6326                           settings->op[stage].cop, settings->op[stage].carg0,
6327                           settings->op[stage].carg1, settings->op[stage].carg2);
6328             if (!stage)
6329                 shader_addline(&buffer, "MOV ret.w, fragment.color.primary.w;\n");
6330         }
6331         else if (op_equal)
6332         {
6333             gen_ffp_instr(&buffer, stage, TRUE, TRUE, settings->op[stage].dst,
6334                           settings->op[stage].cop, settings->op[stage].carg0,
6335                           settings->op[stage].carg1, settings->op[stage].carg2);
6336         } else {
6337             gen_ffp_instr(&buffer, stage, TRUE, FALSE, settings->op[stage].dst,
6338                           settings->op[stage].cop, settings->op[stage].carg0,
6339                           settings->op[stage].carg1, settings->op[stage].carg2);
6340             gen_ffp_instr(&buffer, stage, FALSE, TRUE, settings->op[stage].dst,
6341                           settings->op[stage].aop, settings->op[stage].aarg0,
6342                           settings->op[stage].aarg1, settings->op[stage].aarg2);
6343         }
6344     }
6345
6346     if(settings->sRGB_write) {
6347         shader_addline(&buffer, "MAD ret, fragment.color.secondary, specular_enable, %s;\n", final_combiner_src);
6348         arbfp_add_sRGB_correction(&buffer, "ret", "arg0", "arg1", "arg2", "tempreg", FALSE);
6349         shader_addline(&buffer, "MOV result.color, ret;\n");
6350     } else {
6351         shader_addline(&buffer, "MAD result.color, fragment.color.secondary, specular_enable, %s;\n", final_combiner_src);
6352     }
6353
6354     /* Footer */
6355     shader_addline(&buffer, "END\n");
6356
6357     /* Generate the shader */
6358     GL_EXTCALL(glGenProgramsARB(1, &ret));
6359     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, ret));
6360     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
6361             strlen(buffer.buffer), buffer.buffer));
6362     checkGLcall("glProgramStringARB()");
6363
6364     gl_info->gl_ops.gl.p_glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
6365     if (pos != -1)
6366     {
6367         FIXME("Fragment program error at position %d: %s\n\n", pos,
6368               debugstr_a((const char *)gl_info->gl_ops.gl.p_glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
6369         shader_arb_dump_program_source(buffer.buffer);
6370     }
6371     else
6372     {
6373         GLint native;
6374
6375         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
6376         checkGLcall("glGetProgramivARB()");
6377         if (!native) WARN("Program exceeds native resource limits.\n");
6378     }
6379
6380     shader_buffer_free(&buffer);
6381     return ret;
6382 }
6383
6384 static void fragment_prog_arbfp(struct wined3d_context *context, const struct wined3d_state *state, DWORD state_id)
6385 {
6386     const struct wined3d_device *device = context->swapchain->device;
6387     const struct wined3d_gl_info *gl_info = context->gl_info;
6388     struct shader_arb_priv *priv = device->fragment_priv;
6389     BOOL use_pshader = use_ps(state);
6390     struct ffp_frag_settings settings;
6391     const struct arbfp_ffp_desc *desc;
6392     unsigned int i;
6393
6394     TRACE("context %p, state %p, state_id %#x.\n", context, state, state_id);
6395
6396     if (isStateDirty(context, STATE_RENDER(WINED3D_RS_FOGENABLE)))
6397     {
6398         if (!use_pshader && device->shader_backend == &arb_program_shader_backend && context->last_was_pshader)
6399         {
6400             /* Reload fixed function constants since they collide with the
6401              * pixel shader constants. */
6402             for (i = 0; i < MAX_TEXTURES; ++i)
6403             {
6404                 set_bumpmat_arbfp(context, state, STATE_TEXTURESTAGE(i, WINED3D_TSS_BUMPENV_MAT00));
6405             }
6406             state_texfactor_arbfp(context, state, STATE_RENDER(WINED3D_RS_TEXTUREFACTOR));
6407             state_arb_specularenable(context, state, STATE_RENDER(WINED3D_RS_SPECULARENABLE));
6408         }
6409         else if (use_pshader)
6410         {
6411             context->select_shader = 1;
6412         }
6413         return;
6414     }
6415
6416     if (!use_pshader)
6417     {
6418         /* Find or create a shader implementing the fixed function pipeline
6419          * settings, then activate it. */
6420         gen_ffp_frag_op(device, state, &settings, FALSE);
6421         desc = (const struct arbfp_ffp_desc *)find_ffp_frag_shader(&priv->fragment_shaders, &settings);
6422         if(!desc) {
6423             struct arbfp_ffp_desc *new_desc = HeapAlloc(GetProcessHeap(), 0, sizeof(*new_desc));
6424             if (!new_desc)
6425             {
6426                 ERR("Out of memory\n");
6427                 return;
6428             }
6429
6430             memcpy(&new_desc->parent.settings, &settings, sizeof(settings));
6431             new_desc->shader = gen_arbfp_ffp_shader(&settings, gl_info);
6432             add_ffp_frag_shader(&priv->fragment_shaders, &new_desc->parent);
6433             TRACE("Allocated fixed function replacement shader descriptor %p\n", new_desc);
6434             desc = new_desc;
6435         }
6436
6437         /* Now activate the replacement program. GL_FRAGMENT_PROGRAM_ARB is already active (however, note the
6438          * comment above the shader_select call below). If e.g. GLSL is active, the shader_select call will
6439          * deactivate it.
6440          */
6441         GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, desc->shader));
6442         checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, desc->shader)");
6443         priv->current_fprogram_id = desc->shader;
6444
6445         if (device->shader_backend == &arb_program_shader_backend && context->last_was_pshader)
6446         {
6447             /* Reload fixed function constants since they collide with the
6448              * pixel shader constants. */
6449             for (i = 0; i < MAX_TEXTURES; ++i)
6450             {
6451                 set_bumpmat_arbfp(context, state, STATE_TEXTURESTAGE(i, WINED3D_TSS_BUMPENV_MAT00));
6452             }
6453             state_texfactor_arbfp(context, state, STATE_RENDER(WINED3D_RS_TEXTUREFACTOR));
6454             state_arb_specularenable(context, state, STATE_RENDER(WINED3D_RS_SPECULARENABLE));
6455         }
6456         context->last_was_pshader = FALSE;
6457     } else {
6458         context->last_was_pshader = TRUE;
6459     }
6460
6461     context->select_shader = 1;
6462     context->load_constants = 1;
6463 }
6464
6465 /* We can't link the fog states to the fragment state directly since the
6466  * vertex pipeline links them to FOGENABLE. A different linking in different
6467  * pipeline parts can't be expressed in the combined state table, so we need
6468  * to handle that with a forwarding function. The other invisible side effect
6469  * is that changing the fog start and fog end (which links to FOGENABLE in
6470  * vertex) results in the fragment_prog_arbfp function being called because
6471  * FOGENABLE is dirty, which calls this function here. */
6472 static void state_arbfp_fog(struct wined3d_context *context, const struct wined3d_state *state, DWORD state_id)
6473 {
6474     enum fogsource new_source;
6475
6476     TRACE("context %p, state %p, state_id %#x.\n", context, state, state_id);
6477
6478     if (!isStateDirty(context, STATE_PIXELSHADER))
6479         fragment_prog_arbfp(context, state, state_id);
6480
6481     if (!state->render_states[WINED3D_RS_FOGENABLE])
6482         return;
6483
6484     if (state->render_states[WINED3D_RS_FOGTABLEMODE] == WINED3D_FOG_NONE)
6485     {
6486         if (use_vs(state))
6487         {
6488             new_source = FOGSOURCE_VS;
6489         }
6490         else
6491         {
6492             if (state->render_states[WINED3D_RS_FOGVERTEXMODE] == WINED3D_FOG_NONE || context->last_was_rhw)
6493                 new_source = FOGSOURCE_COORD;
6494             else
6495                 new_source = FOGSOURCE_FFP;
6496         }
6497     }
6498     else
6499     {
6500         new_source = FOGSOURCE_FFP;
6501     }
6502
6503     if (new_source != context->fog_source)
6504     {
6505         context->fog_source = new_source;
6506         state_fogstartend(context, state, STATE_RENDER(WINED3D_RS_FOGSTART));
6507     }
6508 }
6509
6510 static void textransform(struct wined3d_context *context, const struct wined3d_state *state, DWORD state_id)
6511 {
6512     if (!isStateDirty(context, STATE_PIXELSHADER))
6513         fragment_prog_arbfp(context, state, state_id);
6514 }
6515
6516 static void state_srgbwrite(struct wined3d_context *context, const struct wined3d_state *state, DWORD state_id)
6517 {
6518     const struct wined3d_gl_info *gl_info = context->gl_info;
6519     const struct wined3d_surface *rt = state->fb->render_targets[0];
6520
6521     TRACE("context %p, state %p, state_id %#x.\n", context, state, state_id);
6522
6523     if (state->render_states[WINED3D_RS_SRGBWRITEENABLE]
6524             && rt->resource.format->flags & WINED3DFMT_FLAG_SRGB_WRITE)
6525         gl_info->gl_ops.gl.p_glEnable(GL_FRAMEBUFFER_SRGB);
6526     else
6527         gl_info->gl_ops.gl.p_glDisable(GL_FRAMEBUFFER_SRGB);
6528 }
6529
6530 static const struct StateEntryTemplate arbfp_fragmentstate_template[] =
6531 {
6532     {STATE_RENDER(WINED3D_RS_TEXTUREFACTOR),              { STATE_RENDER(WINED3D_RS_TEXTUREFACTOR),             state_texfactor_arbfp   }, WINED3D_GL_EXT_NONE             },
6533     {STATE_TEXTURESTAGE(0, WINED3D_TSS_COLOR_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6534     {STATE_TEXTURESTAGE(0, WINED3D_TSS_COLOR_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6535     {STATE_TEXTURESTAGE(0, WINED3D_TSS_COLOR_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6536     {STATE_TEXTURESTAGE(0, WINED3D_TSS_COLOR_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6537     {STATE_TEXTURESTAGE(0, WINED3D_TSS_ALPHA_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6538     {STATE_TEXTURESTAGE(0, WINED3D_TSS_ALPHA_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6539     {STATE_TEXTURESTAGE(0, WINED3D_TSS_ALPHA_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6540     {STATE_TEXTURESTAGE(0, WINED3D_TSS_ALPHA_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6541     {STATE_TEXTURESTAGE(0, WINED3D_TSS_RESULT_ARG),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6542     {STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6543     {STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6544     {STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6545     {STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6546     {STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6547     {STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6548     {STATE_TEXTURESTAGE(1, WINED3D_TSS_COLOR_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6549     {STATE_TEXTURESTAGE(1, WINED3D_TSS_COLOR_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6550     {STATE_TEXTURESTAGE(1, WINED3D_TSS_COLOR_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6551     {STATE_TEXTURESTAGE(1, WINED3D_TSS_COLOR_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6552     {STATE_TEXTURESTAGE(1, WINED3D_TSS_ALPHA_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6553     {STATE_TEXTURESTAGE(1, WINED3D_TSS_ALPHA_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6554     {STATE_TEXTURESTAGE(1, WINED3D_TSS_ALPHA_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6555     {STATE_TEXTURESTAGE(1, WINED3D_TSS_ALPHA_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6556     {STATE_TEXTURESTAGE(1, WINED3D_TSS_RESULT_ARG),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6557     {STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6558     {STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6559     {STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6560     {STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6561     {STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6562     {STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6563     {STATE_TEXTURESTAGE(2, WINED3D_TSS_COLOR_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6564     {STATE_TEXTURESTAGE(2, WINED3D_TSS_COLOR_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6565     {STATE_TEXTURESTAGE(2, WINED3D_TSS_COLOR_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6566     {STATE_TEXTURESTAGE(2, WINED3D_TSS_COLOR_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6567     {STATE_TEXTURESTAGE(2, WINED3D_TSS_ALPHA_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6568     {STATE_TEXTURESTAGE(2, WINED3D_TSS_ALPHA_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6569     {STATE_TEXTURESTAGE(2, WINED3D_TSS_ALPHA_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6570     {STATE_TEXTURESTAGE(2, WINED3D_TSS_ALPHA_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6571     {STATE_TEXTURESTAGE(2, WINED3D_TSS_RESULT_ARG),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6572     {STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6573     {STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6574     {STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6575     {STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6576     {STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6577     {STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6578     {STATE_TEXTURESTAGE(3, WINED3D_TSS_COLOR_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6579     {STATE_TEXTURESTAGE(3, WINED3D_TSS_COLOR_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6580     {STATE_TEXTURESTAGE(3, WINED3D_TSS_COLOR_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6581     {STATE_TEXTURESTAGE(3, WINED3D_TSS_COLOR_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6582     {STATE_TEXTURESTAGE(3, WINED3D_TSS_ALPHA_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6583     {STATE_TEXTURESTAGE(3, WINED3D_TSS_ALPHA_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6584     {STATE_TEXTURESTAGE(3, WINED3D_TSS_ALPHA_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6585     {STATE_TEXTURESTAGE(3, WINED3D_TSS_ALPHA_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6586     {STATE_TEXTURESTAGE(3, WINED3D_TSS_RESULT_ARG),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6587     {STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6588     {STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6589     {STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6590     {STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6591     {STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6592     {STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6593     {STATE_TEXTURESTAGE(4, WINED3D_TSS_COLOR_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6594     {STATE_TEXTURESTAGE(4, WINED3D_TSS_COLOR_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6595     {STATE_TEXTURESTAGE(4, WINED3D_TSS_COLOR_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6596     {STATE_TEXTURESTAGE(4, WINED3D_TSS_COLOR_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6597     {STATE_TEXTURESTAGE(4, WINED3D_TSS_ALPHA_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6598     {STATE_TEXTURESTAGE(4, WINED3D_TSS_ALPHA_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6599     {STATE_TEXTURESTAGE(4, WINED3D_TSS_ALPHA_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6600     {STATE_TEXTURESTAGE(4, WINED3D_TSS_ALPHA_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6601     {STATE_TEXTURESTAGE(4, WINED3D_TSS_RESULT_ARG),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6602     {STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6603     {STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6604     {STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6605     {STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6606     {STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6607     {STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6608     {STATE_TEXTURESTAGE(5, WINED3D_TSS_COLOR_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6609     {STATE_TEXTURESTAGE(5, WINED3D_TSS_COLOR_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6610     {STATE_TEXTURESTAGE(5, WINED3D_TSS_COLOR_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6611     {STATE_TEXTURESTAGE(5, WINED3D_TSS_COLOR_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6612     {STATE_TEXTURESTAGE(5, WINED3D_TSS_ALPHA_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6613     {STATE_TEXTURESTAGE(5, WINED3D_TSS_ALPHA_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6614     {STATE_TEXTURESTAGE(5, WINED3D_TSS_ALPHA_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6615     {STATE_TEXTURESTAGE(5, WINED3D_TSS_ALPHA_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6616     {STATE_TEXTURESTAGE(5, WINED3D_TSS_RESULT_ARG),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6617     {STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6618     {STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6619     {STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6620     {STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6621     {STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6622     {STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6623     {STATE_TEXTURESTAGE(6, WINED3D_TSS_COLOR_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6624     {STATE_TEXTURESTAGE(6, WINED3D_TSS_COLOR_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6625     {STATE_TEXTURESTAGE(6, WINED3D_TSS_COLOR_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6626     {STATE_TEXTURESTAGE(6, WINED3D_TSS_COLOR_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6627     {STATE_TEXTURESTAGE(6, WINED3D_TSS_ALPHA_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6628     {STATE_TEXTURESTAGE(6, WINED3D_TSS_ALPHA_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6629     {STATE_TEXTURESTAGE(6, WINED3D_TSS_ALPHA_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6630     {STATE_TEXTURESTAGE(6, WINED3D_TSS_ALPHA_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6631     {STATE_TEXTURESTAGE(6, WINED3D_TSS_RESULT_ARG),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6632     {STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6633     {STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6634     {STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6635     {STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6636     {STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6637     {STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6638     {STATE_TEXTURESTAGE(7, WINED3D_TSS_COLOR_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6639     {STATE_TEXTURESTAGE(7, WINED3D_TSS_COLOR_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6640     {STATE_TEXTURESTAGE(7, WINED3D_TSS_COLOR_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6641     {STATE_TEXTURESTAGE(7, WINED3D_TSS_COLOR_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6642     {STATE_TEXTURESTAGE(7, WINED3D_TSS_ALPHA_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6643     {STATE_TEXTURESTAGE(7, WINED3D_TSS_ALPHA_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6644     {STATE_TEXTURESTAGE(7, WINED3D_TSS_ALPHA_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6645     {STATE_TEXTURESTAGE(7, WINED3D_TSS_ALPHA_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6646     {STATE_TEXTURESTAGE(7, WINED3D_TSS_RESULT_ARG),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6647     {STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6648     {STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6649     {STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6650     {STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6651     {STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6652     {STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6653     {STATE_PIXELSHADER,                                   { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6654     {STATE_RENDER(WINED3D_RS_FOGENABLE),                  { STATE_RENDER(WINED3D_RS_FOGENABLE),                 state_arbfp_fog         }, WINED3D_GL_EXT_NONE             },
6655     {STATE_RENDER(WINED3D_RS_FOGTABLEMODE),               { STATE_RENDER(WINED3D_RS_FOGENABLE),                 NULL                    }, WINED3D_GL_EXT_NONE             },
6656     {STATE_RENDER(WINED3D_RS_FOGVERTEXMODE),              { STATE_RENDER(WINED3D_RS_FOGENABLE),                 NULL                    }, WINED3D_GL_EXT_NONE             },
6657     {STATE_RENDER(WINED3D_RS_FOGSTART),                   { STATE_RENDER(WINED3D_RS_FOGSTART),                  state_fogstartend       }, WINED3D_GL_EXT_NONE             },
6658     {STATE_RENDER(WINED3D_RS_FOGEND),                     { STATE_RENDER(WINED3D_RS_FOGSTART),                  NULL                    }, WINED3D_GL_EXT_NONE             },
6659     {STATE_RENDER(WINED3D_RS_SRGBWRITEENABLE),            { STATE_RENDER(WINED3D_RS_SRGBWRITEENABLE),           state_srgbwrite         }, ARB_FRAMEBUFFER_SRGB            },
6660     {STATE_RENDER(WINED3D_RS_SRGBWRITEENABLE),            { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6661     {STATE_RENDER(WINED3D_RS_FOGCOLOR),                   { STATE_RENDER(WINED3D_RS_FOGCOLOR),                  state_fogcolor          }, WINED3D_GL_EXT_NONE             },
6662     {STATE_RENDER(WINED3D_RS_FOGDENSITY),                 { STATE_RENDER(WINED3D_RS_FOGDENSITY),                state_fogdensity        }, WINED3D_GL_EXT_NONE             },
6663     {STATE_TEXTURESTAGE(0,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(0, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6664     {STATE_TEXTURESTAGE(1,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(1, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6665     {STATE_TEXTURESTAGE(2,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(2, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6666     {STATE_TEXTURESTAGE(3,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(3, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6667     {STATE_TEXTURESTAGE(4,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(4, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6668     {STATE_TEXTURESTAGE(5,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(5, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6669     {STATE_TEXTURESTAGE(6,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(6, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6670     {STATE_TEXTURESTAGE(7,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(7, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6671     {STATE_RENDER(WINED3D_RS_SPECULARENABLE),             { STATE_RENDER(WINED3D_RS_SPECULARENABLE),            state_arb_specularenable}, WINED3D_GL_EXT_NONE             },
6672     {0 /* Terminate */,                                   { 0,                                                  0                       }, WINED3D_GL_EXT_NONE             },
6673 };
6674
6675 const struct fragment_pipeline arbfp_fragment_pipeline = {
6676     arbfp_enable,
6677     arbfp_get_caps,
6678     arbfp_alloc,
6679     arbfp_free,
6680     shader_arb_color_fixup_supported,
6681     arbfp_fragmentstate_template,
6682     TRUE /* We can disable projected textures */
6683 };
6684
6685 struct arbfp_blit_priv {
6686     GLenum yuy2_rect_shader, yuy2_2d_shader;
6687     GLenum uyvy_rect_shader, uyvy_2d_shader;
6688     GLenum yv12_rect_shader, yv12_2d_shader;
6689     GLenum p8_rect_shader, p8_2d_shader;
6690     GLuint palette_texture;
6691 };
6692
6693 static HRESULT arbfp_blit_alloc(struct wined3d_device *device)
6694 {
6695     device->blit_priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(struct arbfp_blit_priv));
6696     if(!device->blit_priv) {
6697         ERR("Out of memory\n");
6698         return E_OUTOFMEMORY;
6699     }
6700     return WINED3D_OK;
6701 }
6702
6703 /* Context activation is done by the caller. */
6704 static void arbfp_blit_free(struct wined3d_device *device)
6705 {
6706     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
6707     struct arbfp_blit_priv *priv = device->blit_priv;
6708
6709     ENTER_GL();
6710     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yuy2_rect_shader));
6711     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yuy2_2d_shader));
6712     GL_EXTCALL(glDeleteProgramsARB(1, &priv->uyvy_rect_shader));
6713     GL_EXTCALL(glDeleteProgramsARB(1, &priv->uyvy_2d_shader));
6714     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yv12_rect_shader));
6715     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yv12_2d_shader));
6716     GL_EXTCALL(glDeleteProgramsARB(1, &priv->p8_rect_shader));
6717     GL_EXTCALL(glDeleteProgramsARB(1, &priv->p8_2d_shader));
6718     checkGLcall("Delete yuv and p8 programs");
6719
6720     if (priv->palette_texture)
6721         gl_info->gl_ops.gl.p_glDeleteTextures(1, &priv->palette_texture);
6722     LEAVE_GL();
6723
6724     HeapFree(GetProcessHeap(), 0, device->blit_priv);
6725     device->blit_priv = NULL;
6726 }
6727
6728 static BOOL gen_planar_yuv_read(struct wined3d_shader_buffer *buffer, enum complex_fixup fixup,
6729         GLenum textype, char *luminance)
6730 {
6731     char chroma;
6732     const char *tex, *texinstr;
6733
6734     if (fixup == COMPLEX_FIXUP_UYVY) {
6735         chroma = 'x';
6736         *luminance = 'w';
6737     } else {
6738         chroma = 'w';
6739         *luminance = 'x';
6740     }
6741     switch(textype) {
6742         case GL_TEXTURE_2D:             tex = "2D";     texinstr = "TXP"; break;
6743         case GL_TEXTURE_RECTANGLE_ARB:  tex = "RECT";   texinstr = "TEX"; break;
6744         default:
6745             /* This is more tricky than just replacing the texture type - we have to navigate
6746              * properly in the texture to find the correct chroma values
6747              */
6748             FIXME("Implement yuv correction for non-2d, non-rect textures\n");
6749             return FALSE;
6750     }
6751
6752     /* First we have to read the chroma values. This means we need at least two pixels(no filtering),
6753      * or 4 pixels(with filtering). To get the unmodified chromas, we have to rid ourselves of the
6754      * filtering when we sample the texture.
6755      *
6756      * These are the rules for reading the chroma:
6757      *
6758      * Even pixel: Cr
6759      * Even pixel: U
6760      * Odd pixel: V
6761      *
6762      * So we have to get the sampling x position in non-normalized coordinates in integers
6763      */
6764     if(textype != GL_TEXTURE_RECTANGLE_ARB) {
6765         shader_addline(buffer, "MUL texcrd.xy, fragment.texcoord[0], size.x;\n");
6766         shader_addline(buffer, "MOV texcrd.w, size.x;\n");
6767     } else {
6768         shader_addline(buffer, "MOV texcrd, fragment.texcoord[0];\n");
6769     }
6770     /* We must not allow filtering between pixel x and x+1, this would mix U and V
6771      * Vertical filtering is ok. However, bear in mind that the pixel center is at
6772      * 0.5, so add 0.5.
6773      */
6774     shader_addline(buffer, "FLR texcrd.x, texcrd.x;\n");
6775     shader_addline(buffer, "ADD texcrd.x, texcrd.x, coef.y;\n");
6776
6777     /* Divide the x coordinate by 0.5 and get the fraction. This gives 0.25 and 0.75 for the
6778      * even and odd pixels respectively
6779      */
6780     shader_addline(buffer, "MUL texcrd2, texcrd, coef.y;\n");
6781     shader_addline(buffer, "FRC texcrd2, texcrd2;\n");
6782
6783     /* Sample Pixel 1 */
6784     shader_addline(buffer, "%s luminance, texcrd, texture[0], %s;\n", texinstr, tex);
6785
6786     /* Put the value into either of the chroma values */
6787     shader_addline(buffer, "SGE temp.x, texcrd2.x, coef.y;\n");
6788     shader_addline(buffer, "MUL chroma.x, luminance.%c, temp.x;\n", chroma);
6789     shader_addline(buffer, "SLT temp.x, texcrd2.x, coef.y;\n");
6790     shader_addline(buffer, "MUL chroma.y, luminance.%c, temp.x;\n", chroma);
6791
6792     /* Sample pixel 2. If we read an even pixel(SLT above returned 1), sample
6793      * the pixel right to the current one. Otherwise, sample the left pixel.
6794      * Bias and scale the SLT result to -1;1 and add it to the texcrd.x.
6795      */
6796     shader_addline(buffer, "MAD temp.x, temp.x, coef.z, -coef.x;\n");
6797     shader_addline(buffer, "ADD texcrd.x, texcrd, temp.x;\n");
6798     shader_addline(buffer, "%s luminance, texcrd, texture[0], %s;\n", texinstr, tex);
6799
6800     /* Put the value into the other chroma */
6801     shader_addline(buffer, "SGE temp.x, texcrd2.x, coef.y;\n");
6802     shader_addline(buffer, "MAD chroma.y, luminance.%c, temp.x, chroma.y;\n", chroma);
6803     shader_addline(buffer, "SLT temp.x, texcrd2.x, coef.y;\n");
6804     shader_addline(buffer, "MAD chroma.x, luminance.%c, temp.x, chroma.x;\n", chroma);
6805
6806     /* TODO: If filtering is enabled, sample a 2nd pair of pixels left or right of
6807      * the current one and lerp the two U and V values
6808      */
6809
6810     /* This gives the correctly filtered luminance value */
6811     shader_addline(buffer, "TEX luminance, fragment.texcoord[0], texture[0], %s;\n", tex);
6812
6813     return TRUE;
6814 }
6815
6816 static BOOL gen_yv12_read(struct wined3d_shader_buffer *buffer, GLenum textype, char *luminance)
6817 {
6818     const char *tex;
6819
6820     switch(textype) {
6821         case GL_TEXTURE_2D:             tex = "2D";     break;
6822         case GL_TEXTURE_RECTANGLE_ARB:  tex = "RECT";   break;
6823         default:
6824             FIXME("Implement yv12 correction for non-2d, non-rect textures\n");
6825             return FALSE;
6826     }
6827
6828     /* YV12 surfaces contain a WxH sized luminance plane, followed by a (W/2)x(H/2)
6829      * V and a (W/2)x(H/2) U plane, each with 8 bit per pixel. So the effective
6830      * bitdepth is 12 bits per pixel. Since the U and V planes have only half the
6831      * pitch of the luminance plane, the packing into the gl texture is a bit
6832      * unfortunate. If the whole texture is interpreted as luminance data it looks
6833      * approximately like this:
6834      *
6835      *        +----------------------------------+----
6836      *        |                                  |
6837      *        |                                  |
6838      *        |                                  |
6839      *        |                                  |
6840      *        |                                  |   2
6841      *        |            LUMINANCE             |   -
6842      *        |                                  |   3
6843      *        |                                  |
6844      *        |                                  |
6845      *        |                                  |
6846      *        |                                  |
6847      *        +----------------+-----------------+----
6848      *        |                |                 |
6849      *        |  U even rows   |  U odd rows     |
6850      *        |                |                 |   1
6851      *        +----------------+------------------   -
6852      *        |                |                 |   3
6853      *        |  V even rows   |  V odd rows     |
6854      *        |                |                 |
6855      *        +----------------+-----------------+----
6856      *        |                |                 |
6857      *        |     0.5        |       0.5       |
6858      *
6859      * So it appears as if there are 4 chroma images, but in fact the odd rows
6860      * in the chroma images are in the same row as the even ones. So its is
6861      * kinda tricky to read
6862      *
6863      * When reading from rectangle textures, keep in mind that the input y coordinates
6864      * go from 0 to d3d_height, whereas the opengl texture height is 1.5 * d3d_height
6865      */
6866     shader_addline(buffer, "PARAM yv12_coef = {%f, %f, %f, %f};\n",
6867             2.0f / 3.0f, 1.0f / 6.0f, (2.0f / 3.0f) + (1.0f / 6.0f), 1.0f / 3.0f);
6868
6869     shader_addline(buffer, "MOV texcrd, fragment.texcoord[0];\n");
6870     /* the chroma planes have only half the width */
6871     shader_addline(buffer, "MUL texcrd.x, texcrd.x, coef.y;\n");
6872
6873     /* The first value is between 2/3 and 5/6th of the texture's height, so scale+bias
6874      * the coordinate. Also read the right side of the image when reading odd lines
6875      *
6876      * Don't forget to clamp the y values in into the range, otherwise we'll get filtering
6877      * bleeding
6878      */
6879     if(textype == GL_TEXTURE_2D) {
6880
6881         shader_addline(buffer, "RCP chroma.w, size.y;\n");
6882
6883         shader_addline(buffer, "MUL texcrd2.y, texcrd.y, size.y;\n");
6884
6885         shader_addline(buffer, "FLR texcrd2.y, texcrd2.y;\n");
6886         shader_addline(buffer, "MAD texcrd.y, texcrd.y, yv12_coef.y, yv12_coef.x;\n");
6887
6888         /* Read odd lines from the right side(add size * 0.5 to the x coordinate */
6889         shader_addline(buffer, "ADD texcrd2.x, texcrd2.y, yv12_coef.y;\n"); /* To avoid 0.5 == 0.5 comparisons */
6890         shader_addline(buffer, "FRC texcrd2.x, texcrd2.x;\n");
6891         shader_addline(buffer, "SGE texcrd2.x, texcrd2.x, coef.y;\n");
6892         shader_addline(buffer, "MAD texcrd.x, texcrd2.x, coef.y, texcrd.x;\n");
6893
6894         /* clamp, keep the half pixel origin in mind */
6895         shader_addline(buffer, "MAD temp.y, coef.y, chroma.w, yv12_coef.x;\n");
6896         shader_addline(buffer, "MAX texcrd.y, temp.y, texcrd.y;\n");
6897         shader_addline(buffer, "MAD temp.y, -coef.y, chroma.w, yv12_coef.z;\n");
6898         shader_addline(buffer, "MIN texcrd.y, temp.y, texcrd.y;\n");
6899     } else {
6900         /* Read from [size - size+size/4] */
6901         shader_addline(buffer, "FLR texcrd.y, texcrd.y;\n");
6902         shader_addline(buffer, "MAD texcrd.y, texcrd.y, coef.w, size.y;\n");
6903
6904         /* Read odd lines from the right side(add size * 0.5 to the x coordinate */
6905         shader_addline(buffer, "ADD texcrd2.x, texcrd.y, yv12_coef.y;\n"); /* To avoid 0.5 == 0.5 comparisons */
6906         shader_addline(buffer, "FRC texcrd2.x, texcrd2.x;\n");
6907         shader_addline(buffer, "SGE texcrd2.x, texcrd2.x, coef.y;\n");
6908         shader_addline(buffer, "MUL texcrd2.x, texcrd2.x, size.x;\n");
6909         shader_addline(buffer, "MAD texcrd.x, texcrd2.x, coef.y, texcrd.x;\n");
6910
6911         /* Make sure to read exactly from the pixel center */
6912         shader_addline(buffer, "FLR texcrd.y, texcrd.y;\n");
6913         shader_addline(buffer, "ADD texcrd.y, texcrd.y, coef.y;\n");
6914
6915         /* Clamp */
6916         shader_addline(buffer, "MAD temp.y, size.y, coef.w, size.y;\n");
6917         shader_addline(buffer, "ADD temp.y, temp.y, -coef.y;\n");
6918         shader_addline(buffer, "MIN texcrd.y, temp.y, texcrd.y;\n");
6919         shader_addline(buffer, "ADD temp.y, size.y, -coef.y;\n");
6920         shader_addline(buffer, "MAX texcrd.y, temp.y, texcrd.y;\n");
6921     }
6922     /* Read the texture, put the result into the output register */
6923     shader_addline(buffer, "TEX temp, texcrd, texture[0], %s;\n", tex);
6924     shader_addline(buffer, "MOV chroma.x, temp.w;\n");
6925
6926     /* The other chroma value is 1/6th of the texture lower, from 5/6th to 6/6th
6927      * No need to clamp because we're just reusing the already clamped value from above
6928      */
6929     if(textype == GL_TEXTURE_2D) {
6930         shader_addline(buffer, "ADD texcrd.y, texcrd.y, yv12_coef.y;\n");
6931     } else {
6932         shader_addline(buffer, "MAD texcrd.y, size.y, coef.w, texcrd.y;\n");
6933     }
6934     shader_addline(buffer, "TEX temp, texcrd, texture[0], %s;\n", tex);
6935     shader_addline(buffer, "MOV chroma.y, temp.w;\n");
6936
6937     /* Sample the luminance value. It is in the top 2/3rd of the texture, so scale the y coordinate.
6938      * Clamp the y coordinate to prevent the chroma values from bleeding into the sampled luminance
6939      * values due to filtering
6940      */
6941     shader_addline(buffer, "MOV texcrd, fragment.texcoord[0];\n");
6942     if(textype == GL_TEXTURE_2D) {
6943         /* Multiply the y coordinate by 2/3 and clamp it */
6944         shader_addline(buffer, "MUL texcrd.y, texcrd.y, yv12_coef.x;\n");
6945         shader_addline(buffer, "MAD temp.y, -coef.y, chroma.w, yv12_coef.x;\n");
6946         shader_addline(buffer, "MIN texcrd.y, temp.y, texcrd.y;\n");
6947         shader_addline(buffer, "TEX luminance, texcrd, texture[0], %s;\n", tex);
6948     } else {
6949         /* Reading from texture_rectangles is pretty straightforward, just use the unmodified
6950          * texture coordinate. It is still a good idea to clamp it though, since the opengl texture
6951          * is bigger
6952          */
6953         shader_addline(buffer, "ADD temp.x, size.y, -coef.y;\n");
6954         shader_addline(buffer, "MIN texcrd.y, texcrd.y, size.x;\n");
6955         shader_addline(buffer, "TEX luminance, texcrd, texture[0], %s;\n", tex);
6956     }
6957     *luminance = 'a';
6958
6959     return TRUE;
6960 }
6961
6962 static GLuint gen_p8_shader(struct arbfp_blit_priv *priv,
6963         const struct wined3d_gl_info *gl_info, GLenum textype)
6964 {
6965     GLenum shader;
6966     struct wined3d_shader_buffer buffer;
6967     GLint pos;
6968
6969     /* Shader header */
6970     if (!shader_buffer_init(&buffer))
6971     {
6972         ERR("Failed to initialize shader buffer.\n");
6973         return 0;
6974     }
6975
6976     ENTER_GL();
6977     GL_EXTCALL(glGenProgramsARB(1, &shader));
6978     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader));
6979     LEAVE_GL();
6980     if(!shader) {
6981         shader_buffer_free(&buffer);
6982         return 0;
6983     }
6984
6985     shader_addline(&buffer, "!!ARBfp1.0\n");
6986     shader_addline(&buffer, "TEMP index;\n");
6987
6988     /* { 255/256, 0.5/255*255/256, 0, 0 } */
6989     shader_addline(&buffer, "PARAM constants = { 0.996, 0.00195, 0, 0 };\n");
6990
6991     /* The alpha-component contains the palette index */
6992     if(textype == GL_TEXTURE_RECTANGLE_ARB)
6993         shader_addline(&buffer, "TXP index, fragment.texcoord[0], texture[0], RECT;\n");
6994     else
6995         shader_addline(&buffer, "TEX index, fragment.texcoord[0], texture[0], 2D;\n");
6996
6997     /* Scale the index by 255/256 and add a bias of '0.5' in order to sample in the middle */
6998     shader_addline(&buffer, "MAD index.a, index.a, constants.x, constants.y;\n");
6999
7000     /* Use the alpha-component as an index in the palette to get the final color */
7001     shader_addline(&buffer, "TEX result.color, index.a, texture[1], 1D;\n");
7002     shader_addline(&buffer, "END\n");
7003
7004     ENTER_GL();
7005     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
7006             strlen(buffer.buffer), buffer.buffer));
7007     checkGLcall("glProgramStringARB()");
7008
7009     gl_info->gl_ops.gl.p_glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
7010     if (pos != -1)
7011     {
7012         FIXME("Fragment program error at position %d: %s\n\n", pos,
7013               debugstr_a((const char *)gl_info->gl_ops.gl.p_glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
7014         shader_arb_dump_program_source(buffer.buffer);
7015     }
7016
7017     if (textype == GL_TEXTURE_RECTANGLE_ARB)
7018         priv->p8_rect_shader = shader;
7019     else
7020         priv->p8_2d_shader = shader;
7021
7022     shader_buffer_free(&buffer);
7023     LEAVE_GL();
7024
7025     return shader;
7026 }
7027
7028 /* Context activation is done by the caller. */
7029 static void upload_palette(const struct wined3d_surface *surface, struct wined3d_context *context)
7030 {
7031     BYTE table[256][4];
7032     struct wined3d_device *device = surface->resource.device;
7033     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
7034     struct arbfp_blit_priv *priv = device->blit_priv;
7035     BOOL colorkey = (surface->CKeyFlags & WINEDDSD_CKSRCBLT) != 0;
7036
7037     d3dfmt_p8_init_palette(surface, table, colorkey);
7038
7039     ENTER_GL();
7040
7041     if (gl_info->supported[APPLE_CLIENT_STORAGE])
7042     {
7043         gl_info->gl_ops.gl.p_glPixelStorei(GL_UNPACK_CLIENT_STORAGE_APPLE, GL_FALSE);
7044         checkGLcall("glPixelStorei(GL_UNPACK_CLIENT_STORAGE_APPLE, GL_FALSE)");
7045     }
7046
7047     if (!priv->palette_texture)
7048         gl_info->gl_ops.gl.p_glGenTextures(1, &priv->palette_texture);
7049
7050     GL_EXTCALL(glActiveTextureARB(GL_TEXTURE1));
7051     gl_info->gl_ops.gl.p_glBindTexture(GL_TEXTURE_1D, priv->palette_texture);
7052
7053     gl_info->gl_ops.gl.p_glTexEnvi(GL_TEXTURE_ENV, GL_TEXTURE_ENV_MODE, GL_REPLACE);
7054
7055     gl_info->gl_ops.gl.p_glTexParameteri(GL_TEXTURE_1D, GL_TEXTURE_MIN_FILTER, GL_NEAREST);
7056     /* Make sure we have discrete color levels. */
7057     gl_info->gl_ops.gl.p_glTexParameteri(GL_TEXTURE_1D, GL_TEXTURE_MAG_FILTER, GL_NEAREST);
7058     gl_info->gl_ops.gl.p_glTexParameteri(GL_TEXTURE_1D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
7059     /* Upload the palette */
7060     /* TODO: avoid unneeded uploads in the future by adding some SFLAG_PALETTE_DIRTY mechanism */
7061     gl_info->gl_ops.gl.p_glTexImage1D(GL_TEXTURE_1D, 0, GL_RGBA, 256, 0, GL_RGBA, GL_UNSIGNED_BYTE, table);
7062
7063     if (gl_info->supported[APPLE_CLIENT_STORAGE])
7064     {
7065         gl_info->gl_ops.gl.p_glPixelStorei(GL_UNPACK_CLIENT_STORAGE_APPLE, GL_TRUE);
7066         checkGLcall("glPixelStorei(GL_UNPACK_CLIENT_STORAGE_APPLE, GL_TRUE)");
7067     }
7068
7069     /* Switch back to unit 0 in which the 2D texture will be stored. */
7070     context_active_texture(context, gl_info, 0);
7071     LEAVE_GL();
7072 }
7073
7074 /* Context activation is done by the caller. */
7075 static GLuint gen_yuv_shader(struct arbfp_blit_priv *priv, const struct wined3d_gl_info *gl_info,
7076         enum complex_fixup yuv_fixup, GLenum textype)
7077 {
7078     GLenum shader;
7079     struct wined3d_shader_buffer buffer;
7080     char luminance_component;
7081     GLint pos;
7082
7083     /* Shader header */
7084     if (!shader_buffer_init(&buffer))
7085     {
7086         ERR("Failed to initialize shader buffer.\n");
7087         return 0;
7088     }
7089
7090     ENTER_GL();
7091     GL_EXTCALL(glGenProgramsARB(1, &shader));
7092     checkGLcall("GL_EXTCALL(glGenProgramsARB(1, &shader))");
7093     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader));
7094     checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader)");
7095     LEAVE_GL();
7096     if(!shader) {
7097         shader_buffer_free(&buffer);
7098         return 0;
7099     }
7100
7101     /* The YUY2 and UYVY formats contain two pixels packed into a 32 bit macropixel,
7102      * giving effectively 16 bit per pixel. The color consists of a luminance(Y) and
7103      * two chroma(U and V) values. Each macropixel has two luminance values, one for
7104      * each single pixel it contains, and one U and one V value shared between both
7105      * pixels.
7106      *
7107      * The data is loaded into an A8L8 texture. With YUY2, the luminance component
7108      * contains the luminance and alpha the chroma. With UYVY it is vice versa. Thus
7109      * take the format into account when generating the read swizzles
7110      *
7111      * Reading the Y value is straightforward - just sample the texture. The hardware
7112      * takes care of filtering in the horizontal and vertical direction.
7113      *
7114      * Reading the U and V values is harder. We have to avoid filtering horizontally,
7115      * because that would mix the U and V values of one pixel or two adjacent pixels.
7116      * Thus floor the texture coordinate and add 0.5 to get an unfiltered read,
7117      * regardless of the filtering setting. Vertical filtering works automatically
7118      * though - the U and V values of two rows are mixed nicely.
7119      *
7120      * Apart of avoiding filtering issues, the code has to know which value it just
7121      * read, and where it can find the other one. To determine this, it checks if
7122      * it sampled an even or odd pixel, and shifts the 2nd read accordingly.
7123      *
7124      * Handling horizontal filtering of U and V values requires reading a 2nd pair
7125      * of pixels, extracting U and V and mixing them. This is not implemented yet.
7126      *
7127      * An alternative implementation idea is to load the texture as A8R8G8B8 texture,
7128      * with width / 2. This way one read gives all 3 values, finding U and V is easy
7129      * in an unfiltered situation. Finding the luminance on the other hand requires
7130      * finding out if it is an odd or even pixel. The real drawback of this approach
7131      * is filtering. This would have to be emulated completely in the shader, reading
7132      * up two 2 packed pixels in up to 2 rows and interpolating both horizontally and
7133      * vertically. Beyond that it would require adjustments to the texture handling
7134      * code to deal with the width scaling
7135      */
7136     shader_addline(&buffer, "!!ARBfp1.0\n");
7137     shader_addline(&buffer, "TEMP luminance;\n");
7138     shader_addline(&buffer, "TEMP temp;\n");
7139     shader_addline(&buffer, "TEMP chroma;\n");
7140     shader_addline(&buffer, "TEMP texcrd;\n");
7141     shader_addline(&buffer, "TEMP texcrd2;\n");
7142     shader_addline(&buffer, "PARAM coef = {1.0, 0.5, 2.0, 0.25};\n");
7143     shader_addline(&buffer, "PARAM yuv_coef = {1.403, 0.344, 0.714, 1.770};\n");
7144     shader_addline(&buffer, "PARAM size = program.local[0];\n");
7145
7146     switch (yuv_fixup)
7147     {
7148         case COMPLEX_FIXUP_UYVY:
7149         case COMPLEX_FIXUP_YUY2:
7150             if (!gen_planar_yuv_read(&buffer, yuv_fixup, textype, &luminance_component))
7151             {
7152                 shader_buffer_free(&buffer);
7153                 return 0;
7154             }
7155             break;
7156
7157         case COMPLEX_FIXUP_YV12:
7158             if (!gen_yv12_read(&buffer, textype, &luminance_component))
7159             {
7160                 shader_buffer_free(&buffer);
7161                 return 0;
7162             }
7163             break;
7164
7165         default:
7166             FIXME("Unsupported YUV fixup %#x\n", yuv_fixup);
7167             shader_buffer_free(&buffer);
7168             return 0;
7169     }
7170
7171     /* Calculate the final result. Formula is taken from
7172      * http://www.fourcc.org/fccyvrgb.php. Note that the chroma
7173      * ranges from -0.5 to 0.5
7174      */
7175     shader_addline(&buffer, "SUB chroma.xy, chroma, coef.y;\n");
7176
7177     shader_addline(&buffer, "MAD result.color.x, chroma.x, yuv_coef.x, luminance.%c;\n", luminance_component);
7178     shader_addline(&buffer, "MAD temp.x, -chroma.y, yuv_coef.y, luminance.%c;\n", luminance_component);
7179     shader_addline(&buffer, "MAD result.color.y, -chroma.x, yuv_coef.z, temp.x;\n");
7180     shader_addline(&buffer, "MAD result.color.z, chroma.y, yuv_coef.w, luminance.%c;\n", luminance_component);
7181     shader_addline(&buffer, "END\n");
7182
7183     ENTER_GL();
7184     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
7185             strlen(buffer.buffer), buffer.buffer));
7186     checkGLcall("glProgramStringARB()");
7187
7188     gl_info->gl_ops.gl.p_glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
7189     if (pos != -1)
7190     {
7191         FIXME("Fragment program error at position %d: %s\n\n", pos,
7192               debugstr_a((const char *)gl_info->gl_ops.gl.p_glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
7193         shader_arb_dump_program_source(buffer.buffer);
7194     }
7195     else
7196     {
7197         GLint native;
7198
7199         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
7200         checkGLcall("glGetProgramivARB()");
7201         if (!native) WARN("Program exceeds native resource limits.\n");
7202     }
7203
7204     shader_buffer_free(&buffer);
7205     LEAVE_GL();
7206
7207     switch (yuv_fixup)
7208     {
7209         case COMPLEX_FIXUP_YUY2:
7210             if (textype == GL_TEXTURE_RECTANGLE_ARB) priv->yuy2_rect_shader = shader;
7211             else priv->yuy2_2d_shader = shader;
7212             break;
7213
7214         case COMPLEX_FIXUP_UYVY:
7215             if (textype == GL_TEXTURE_RECTANGLE_ARB) priv->uyvy_rect_shader = shader;
7216             else priv->uyvy_2d_shader = shader;
7217             break;
7218
7219         case COMPLEX_FIXUP_YV12:
7220             if (textype == GL_TEXTURE_RECTANGLE_ARB) priv->yv12_rect_shader = shader;
7221             else priv->yv12_2d_shader = shader;
7222             break;
7223         default:
7224             ERR("Unsupported complex fixup: %d\n", yuv_fixup);
7225     }
7226
7227     return shader;
7228 }
7229
7230 /* Context activation is done by the caller. */
7231 static HRESULT arbfp_blit_set(void *blit_priv, struct wined3d_context *context, const struct wined3d_surface *surface)
7232 {
7233     GLenum shader;
7234     float size[4] = {(float) surface->pow2Width, (float) surface->pow2Height, 1.0f, 1.0f};
7235     struct arbfp_blit_priv *priv = blit_priv;
7236     enum complex_fixup fixup;
7237     const struct wined3d_gl_info *gl_info = context->gl_info;
7238     GLenum textype;
7239
7240     if (surface->container.type == WINED3D_CONTAINER_TEXTURE)
7241         textype = surface->container.u.texture->target;
7242     else
7243         textype = surface->texture_target;
7244
7245     if (surface->flags & SFLAG_CONVERTED)
7246     {
7247         ENTER_GL();
7248         gl_info->gl_ops.gl.p_glEnable(textype);
7249         checkGLcall("glEnable(textype)");
7250         LEAVE_GL();
7251         return WINED3D_OK;
7252     }
7253
7254     if (!is_complex_fixup(surface->resource.format->color_fixup))
7255     {
7256         TRACE("Fixup:\n");
7257         dump_color_fixup_desc(surface->resource.format->color_fixup);
7258         /* Don't bother setting up a shader for unconverted formats */
7259         ENTER_GL();
7260         gl_info->gl_ops.gl.p_glEnable(textype);
7261         checkGLcall("glEnable(textype)");
7262         LEAVE_GL();
7263         return WINED3D_OK;
7264     }
7265
7266     fixup = get_complex_fixup(surface->resource.format->color_fixup);
7267
7268     switch(fixup)
7269     {
7270         case COMPLEX_FIXUP_YUY2:
7271             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->yuy2_rect_shader : priv->yuy2_2d_shader;
7272             break;
7273
7274         case COMPLEX_FIXUP_UYVY:
7275             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->uyvy_rect_shader : priv->uyvy_2d_shader;
7276             break;
7277
7278         case COMPLEX_FIXUP_YV12:
7279             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->yv12_rect_shader : priv->yv12_2d_shader;
7280             break;
7281
7282         case COMPLEX_FIXUP_P8:
7283             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->p8_rect_shader : priv->p8_2d_shader;
7284             if (!shader) shader = gen_p8_shader(priv, gl_info, textype);
7285
7286             upload_palette(surface, context);
7287             break;
7288
7289         default:
7290             FIXME("Unsupported complex fixup %#x, not setting a shader\n", fixup);
7291             ENTER_GL();
7292             gl_info->gl_ops.gl.p_glEnable(textype);
7293             checkGLcall("glEnable(textype)");
7294             LEAVE_GL();
7295             return E_NOTIMPL;
7296     }
7297
7298     if (!shader) shader = gen_yuv_shader(priv, gl_info, fixup, textype);
7299
7300     ENTER_GL();
7301     gl_info->gl_ops.gl.p_glEnable(GL_FRAGMENT_PROGRAM_ARB);
7302     checkGLcall("glEnable(GL_FRAGMENT_PROGRAM_ARB)");
7303     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader));
7304     checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader)");
7305     GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, 0, size));
7306     checkGLcall("glProgramLocalParameter4fvARB");
7307     LEAVE_GL();
7308
7309     return WINED3D_OK;
7310 }
7311
7312 /* Context activation is done by the caller. */
7313 static void arbfp_blit_unset(const struct wined3d_gl_info *gl_info)
7314 {
7315     ENTER_GL();
7316     gl_info->gl_ops.gl.p_glDisable(GL_FRAGMENT_PROGRAM_ARB);
7317     checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
7318     gl_info->gl_ops.gl.p_glDisable(GL_TEXTURE_2D);
7319     checkGLcall("glDisable(GL_TEXTURE_2D)");
7320     if (gl_info->supported[ARB_TEXTURE_CUBE_MAP])
7321     {
7322         gl_info->gl_ops.gl.p_glDisable(GL_TEXTURE_CUBE_MAP_ARB);
7323         checkGLcall("glDisable(GL_TEXTURE_CUBE_MAP_ARB)");
7324     }
7325     if (gl_info->supported[ARB_TEXTURE_RECTANGLE])
7326     {
7327         gl_info->gl_ops.gl.p_glDisable(GL_TEXTURE_RECTANGLE_ARB);
7328         checkGLcall("glDisable(GL_TEXTURE_RECTANGLE_ARB)");
7329     }
7330     LEAVE_GL();
7331 }
7332
7333 static BOOL arbfp_blit_supported(const struct wined3d_gl_info *gl_info, enum wined3d_blit_op blit_op,
7334         const RECT *src_rect, DWORD src_usage, enum wined3d_pool src_pool, const struct wined3d_format *src_format,
7335         const RECT *dst_rect, DWORD dst_usage, enum wined3d_pool dst_pool, const struct wined3d_format *dst_format)
7336 {
7337     enum complex_fixup src_fixup;
7338
7339     if (!gl_info->supported[ARB_FRAGMENT_PROGRAM])
7340         return FALSE;
7341
7342     if (blit_op != WINED3D_BLIT_OP_COLOR_BLIT)
7343     {
7344         TRACE("Unsupported blit_op=%d\n", blit_op);
7345         return FALSE;
7346     }
7347
7348     if (src_pool == WINED3D_POOL_SYSTEM_MEM || dst_pool == WINED3D_POOL_SYSTEM_MEM)
7349         return FALSE;
7350
7351     src_fixup = get_complex_fixup(src_format->color_fixup);
7352     if (TRACE_ON(d3d_shader) && TRACE_ON(d3d))
7353     {
7354         TRACE("Checking support for fixup:\n");
7355         dump_color_fixup_desc(src_format->color_fixup);
7356     }
7357
7358     if (!is_identity_fixup(dst_format->color_fixup))
7359     {
7360         TRACE("Destination fixups are not supported\n");
7361         return FALSE;
7362     }
7363
7364     if (is_identity_fixup(src_format->color_fixup))
7365     {
7366         TRACE("[OK]\n");
7367         return TRUE;
7368     }
7369
7370      /* We only support YUV conversions. */
7371     if (!is_complex_fixup(src_format->color_fixup))
7372     {
7373         TRACE("[FAILED]\n");
7374         return FALSE;
7375     }
7376
7377     switch(src_fixup)
7378     {
7379         case COMPLEX_FIXUP_YUY2:
7380         case COMPLEX_FIXUP_UYVY:
7381         case COMPLEX_FIXUP_YV12:
7382         case COMPLEX_FIXUP_P8:
7383             TRACE("[OK]\n");
7384             return TRUE;
7385
7386         default:
7387             FIXME("Unsupported YUV fixup %#x\n", src_fixup);
7388             TRACE("[FAILED]\n");
7389             return FALSE;
7390     }
7391 }
7392
7393 HRESULT arbfp_blit_surface(struct wined3d_device *device, DWORD filter,
7394         struct wined3d_surface *src_surface, const RECT *src_rect_in,
7395         struct wined3d_surface *dst_surface, const RECT *dst_rect_in)
7396 {
7397     struct wined3d_context *context;
7398     RECT src_rect = *src_rect_in;
7399     RECT dst_rect = *dst_rect_in;
7400
7401     /* Now load the surface */
7402     if (wined3d_settings.offscreen_rendering_mode != ORM_FBO
7403             && (src_surface->flags & (SFLAG_INTEXTURE | SFLAG_INDRAWABLE)) == SFLAG_INDRAWABLE)
7404     {
7405         /* Without FBO blits transferring from the drawable to the texture is
7406          * expensive, because we have to flip the data in sysmem. Since we can
7407          * flip in the blitter, we don't actually need that flip anyway. So we
7408          * use the surface's texture as scratch texture, and flip the source
7409          * rectangle instead. */
7410         surface_load_fb_texture(src_surface, FALSE);
7411
7412         src_rect.top = src_surface->resource.height - src_rect.top;
7413         src_rect.bottom = src_surface->resource.height - src_rect.bottom;
7414     }
7415     else
7416         surface_internal_preload(src_surface, SRGB_RGB);
7417
7418     /* Activate the destination context, set it up for blitting */
7419     context = context_acquire(device, dst_surface);
7420     context_apply_blit_state(context, device);
7421
7422     if (!surface_is_offscreen(dst_surface))
7423         surface_translate_drawable_coords(dst_surface, context->win_handle, &dst_rect);
7424
7425     arbfp_blit_set(device->blit_priv, context, src_surface);
7426
7427     ENTER_GL();
7428
7429     /* Draw a textured quad */
7430     draw_textured_quad(src_surface, context, &src_rect, &dst_rect, filter);
7431
7432     LEAVE_GL();
7433
7434     /* Leave the opengl state valid for blitting */
7435     arbfp_blit_unset(context->gl_info);
7436
7437     if (wined3d_settings.strict_draw_ordering
7438             || (dst_surface->container.type == WINED3D_CONTAINER_SWAPCHAIN
7439             && (dst_surface->container.u.swapchain->front_buffer == dst_surface)))
7440         context->gl_info->gl_ops.gl.p_glFlush(); /* Flush to ensure ordering across contexts. */
7441
7442     context_release(context);
7443
7444     surface_modify_location(dst_surface, dst_surface->draw_binding, TRUE);
7445     return WINED3D_OK;
7446 }
7447
7448 /* Do not call while under the GL lock. */
7449 static HRESULT arbfp_blit_color_fill(struct wined3d_device *device, struct wined3d_surface *dst_surface,
7450         const RECT *dst_rect, const struct wined3d_color *color)
7451 {
7452     FIXME("Color filling not implemented by arbfp_blit\n");
7453     return WINED3DERR_INVALIDCALL;
7454 }
7455
7456 /* Do not call while under the GL lock. */
7457 static HRESULT arbfp_blit_depth_fill(struct wined3d_device *device,
7458         struct wined3d_surface *surface, const RECT *rect, float depth)
7459 {
7460     FIXME("Depth filling not implemented by arbfp_blit.\n");
7461     return WINED3DERR_INVALIDCALL;
7462 }
7463
7464 const struct blit_shader arbfp_blit = {
7465     arbfp_blit_alloc,
7466     arbfp_blit_free,
7467     arbfp_blit_set,
7468     arbfp_blit_unset,
7469     arbfp_blit_supported,
7470     arbfp_blit_color_fill,
7471     arbfp_blit_depth_fill,
7472 };