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

   1 /*
   2  * Pixel and vertex shaders implementation using ARB_vertex_program
   3  * and ARB_fragment_program GL extensions.
   4  *
   5  * Copyright 2002-2003 Jason Edmeades
   6  * Copyright 2002-2003 Raphael Junqueira
   7  * Copyright 2004 Christian Costa
   8  * Copyright 2005 Oliver Stieber
   9  * Copyright 2006 Ivan Gyurdiev
  10  * Copyright 2006 Jason Green
  11  * Copyright 2006 Henri Verbeet
  12  * Copyright 2007-2008 Stefan Dösinger for CodeWeavers
  13  * Copyright 2009 Henri Verbeet for CodeWeavers
  14  *
  15  * This library is free software; you can redistribute it and/or
  16  * modify it under the terms of the GNU Lesser General Public
  17  * License as published by the Free Software Foundation; either
  18  * version 2.1 of the License, or (at your option) any later version.
  19  *
  20  * This library is distributed in the hope that it will be useful,
  21  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  22  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  23  * Lesser General Public License for more details.
  24  *
  25  * You should have received a copy of the GNU Lesser General Public
  26  * License along with this library; if not, write to the Free Software
  27  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301, USA
  28  */
  29
  30 #include "config.h"
  31 #include "wine/port.h"
  32
  33 #include <math.h>
  34 #include <stdio.h>
  35
  36 #include "wined3d_private.h"
  37
  38 WINE_DEFAULT_DEBUG_CHANNEL(d3d_shader);
  39 WINE_DECLARE_DEBUG_CHANNEL(d3d_constants);
  40 WINE_DECLARE_DEBUG_CHANNEL(d3d);
  41
  42 /* Extract a line. Note that this modifies the source string. */
  43 static char *get_line(char **ptr)
  44 {
  45     char *p, *q;
  46
  47     p = *ptr;
  48     if (!(q = strstr(p, "\n")))
  49     {
  50         if (!*p) return NULL;
  51         *ptr += strlen(p);
  52         return p;
  53     }
  54     *q = '\0';
  55     *ptr = q + 1;
  56
  57     return p;
  58 }
  59
  60 static void shader_arb_dump_program_source(const char *source)
  61 {
  62     ULONG source_size;
  63     char *ptr, *line, *tmp;
  64
  65     source_size = strlen(source) + 1;
  66     tmp = HeapAlloc(GetProcessHeap(), 0, source_size);
  67     if (!tmp)
  68     {
  69         ERR("Failed to allocate %u bytes for shader source.\n", source_size);
  70         return;
  71     }
  72     memcpy(tmp, source, source_size);
  73
  74     ptr = tmp;
  75     while ((line = get_line(&ptr))) FIXME("    %s\n", line);
  76     FIXME("\n");
  77
  78     HeapFree(GetProcessHeap(), 0, tmp);
  79 }
  80
  81 enum arb_helper_value
  82 {
  83     ARB_ZERO,
  84     ARB_ONE,
  85     ARB_TWO,
  86     ARB_0001,
  87     ARB_EPS,
  88
  89     ARB_VS_REL_OFFSET
  90 };
  91
  92 static const char *arb_get_helper_value(enum wined3d_shader_type shader, enum arb_helper_value value)
  93 {
  94     if (shader == WINED3D_SHADER_TYPE_GEOMETRY)
  95     {
  96         ERR("Geometry shaders are unsupported\n");
  97         return "bad";
  98     }
  99
 100     if (shader == WINED3D_SHADER_TYPE_PIXEL)
 101     {
 102         switch (value)
 103         {
 104             case ARB_ZERO: return "ps_helper_const.x";
 105             case ARB_ONE: return "ps_helper_const.y";
 106             case ARB_TWO: return "coefmul.x";
 107             case ARB_0001: return "ps_helper_const.xxxy";
 108             case ARB_EPS: return "ps_helper_const.z";
 109             default: break;
 110         }
 111     }
 112     else
 113     {
 114         switch (value)
 115         {
 116             case ARB_ZERO: return "helper_const.x";
 117             case ARB_ONE: return "helper_const.y";
 118             case ARB_TWO: return "helper_const.z";
 119             case ARB_EPS: return "helper_const.w";
 120             case ARB_0001: return "helper_const.xxxy";
 121             case ARB_VS_REL_OFFSET: return "rel_addr_const.y";
 122         }
 123     }
 124     FIXME("Unmanaged %s shader helper constant requested: %u\n",
 125           shader == WINED3D_SHADER_TYPE_PIXEL ? "pixel" : "vertex", value);
 126     switch (value)
 127     {
 128         case ARB_ZERO: return "0.0";
 129         case ARB_ONE: return "1.0";
 130         case ARB_TWO: return "2.0";
 131         case ARB_0001: return "{0.0, 0.0, 0.0, 1.0}";
 132         case ARB_EPS: return "1e-8";
 133         default: return "bad";
 134     }
 135 }
 136
 137 static inline BOOL ffp_clip_emul(const struct wined3d_state *state)
 138 {
 139     return state->lowest_disabled_stage < 7;
 140 }
 141
 142 /* ARB_program_shader private data */
 143
 144 struct control_frame
 145 {
 146     struct                          list entry;
 147     enum
 148     {
 149         IF,
 150         IFC,
 151         LOOP,
 152         REP
 153     } type;
 154     BOOL                            muting;
 155     BOOL                            outer_loop;
 156     union
 157     {
 158         unsigned int                loop;
 159         unsigned int                ifc;
 160     } no;
 161     struct wined3d_shader_loop_control loop_control;
 162     BOOL                            had_else;
 163 };
 164
 165 struct arb_ps_np2fixup_info
 166 {
 167     struct ps_np2fixup_info         super;
 168     /* For ARB we need a offset value:
 169      * With both GLSL and ARB mode the NP2 fixup information (the texture dimensions) are stored in a
 170      * consecutive way (GLSL uses a uniform array). Since ARB doesn't know the notion of a "standalone"
 171      * array we need an offset to the index inside the program local parameter array. */
 172     UINT                            offset;
 173 };
 174
 175 struct arb_ps_compile_args
 176 {
 177     struct ps_compile_args          super;
 178     WORD                            bools;
 179     WORD                            clip;  /* only a boolean, use a WORD for alignment */
 180     unsigned char                   loop_ctrl[MAX_CONST_I][3];
 181 };
 182
 183 struct stb_const_desc
 184 {
 185     unsigned char           texunit;
 186     UINT                    const_num;
 187 };
 188
 189 struct arb_ps_compiled_shader
 190 {
 191     struct arb_ps_compile_args      args;
 192     struct arb_ps_np2fixup_info     np2fixup_info;
 193     struct stb_const_desc           bumpenvmatconst[MAX_TEXTURES];
 194     struct stb_const_desc           luminanceconst[MAX_TEXTURES];
 195     UINT                            int_consts[MAX_CONST_I];
 196     GLuint                          prgId;
 197     UINT                            ycorrection;
 198     unsigned char                   numbumpenvmatconsts;
 199     char                            num_int_consts;
 200 };
 201
 202 struct arb_vs_compile_args
 203 {
 204     struct vs_compile_args          super;
 205     union
 206     {
 207         struct
 208         {
 209             WORD                    bools;
 210             unsigned char           clip_texcoord;
 211             unsigned char           clipplane_mask;
 212         }                           boolclip;
 213         DWORD                       boolclip_compare;
 214     } clip;
 215     DWORD                           ps_signature;
 216     union
 217     {
 218         unsigned char               samplers[4];
 219         DWORD                       samplers_compare;
 220     } vertex;
 221     unsigned char                   loop_ctrl[MAX_CONST_I][3];
 222 };
 223
 224 struct arb_vs_compiled_shader
 225 {
 226     struct arb_vs_compile_args      args;
 227     GLuint                          prgId;
 228     UINT                            int_consts[MAX_CONST_I];
 229     char                            num_int_consts;
 230     char                            need_color_unclamp;
 231     UINT                            pos_fixup;
 232 };
 233
 234 struct recorded_instruction
 235 {
 236     struct wined3d_shader_instruction ins;
 237     struct list entry;
 238 };
 239
 240 struct shader_arb_ctx_priv
 241 {
 242     char addr_reg[20];
 243     enum
 244     {
 245         /* plain GL_ARB_vertex_program or GL_ARB_fragment_program */
 246         ARB,
 247         /* GL_NV_vertex_progam2_option or GL_NV_fragment_program_option */
 248         NV2,
 249         /* GL_NV_vertex_program3 or GL_NV_fragment_program2 */
 250         NV3
 251     } target_version;
 252
 253     const struct arb_vs_compile_args    *cur_vs_args;
 254     const struct arb_ps_compile_args    *cur_ps_args;
 255     const struct arb_ps_compiled_shader *compiled_fprog;
 256     const struct arb_vs_compiled_shader *compiled_vprog;
 257     struct arb_ps_np2fixup_info         *cur_np2fixup_info;
 258     struct list                         control_frames;
 259     struct list                         record;
 260     BOOL                                recording;
 261     BOOL                                muted;
 262     unsigned int                        num_loops, loop_depth, num_ifcs;
 263     int                                 aL;
 264
 265     unsigned int                        vs_clipplanes;
 266     BOOL                                footer_written;
 267     BOOL                                in_main_func;
 268
 269     /* For 3.0 vertex shaders */
 270     const char                          *vs_output[MAX_REG_OUTPUT];
 271     /* For 2.x and earlier vertex shaders */
 272     const char                          *texcrd_output[8], *color_output[2], *fog_output;
 273
 274     /* 3.0 pshader input for compatibility with fixed function */
 275     const char                          *ps_input[MAX_REG_INPUT];
 276 };
 277
 278 struct ps_signature
 279 {
 280     struct wined3d_shader_signature_element *sig;
 281     DWORD                               idx;
 282     struct wine_rb_entry                entry;
 283 };
 284
 285 struct arb_pshader_private {
 286     struct arb_ps_compiled_shader   *gl_shaders;
 287     UINT                            num_gl_shaders, shader_array_size;
 288     DWORD                           input_signature_idx;
 289     DWORD                           clipplane_emulation;
 290     BOOL                            clamp_consts;
 291 };
 292
 293 struct arb_vshader_private {
 294     struct arb_vs_compiled_shader   *gl_shaders;
 295     UINT                            num_gl_shaders, shader_array_size;
 296     UINT rel_offset;
 297 };
 298
 299 struct shader_arb_priv
 300 {
 301     GLuint                  current_vprogram_id;
 302     GLuint                  current_fprogram_id;
 303     const struct arb_ps_compiled_shader *compiled_fprog;
 304     const struct arb_vs_compiled_shader *compiled_vprog;
 305     GLuint                  depth_blt_vprogram_id;
 306     GLuint                  depth_blt_fprogram_id_full[tex_type_count];
 307     GLuint                  depth_blt_fprogram_id_masked[tex_type_count];
 308     BOOL                    use_arbfp_fixed_func;
 309     struct wine_rb_tree     fragment_shaders;
 310     BOOL                    last_ps_const_clamped;
 311     BOOL                    last_vs_color_unclamp;
 312
 313     struct wine_rb_tree     signature_tree;
 314     DWORD ps_sig_number;
 315
 316     unsigned int highest_dirty_ps_const, highest_dirty_vs_const;
 317     char *vshader_const_dirty, *pshader_const_dirty;
 318     const struct wined3d_context *last_context;
 319 };
 320
 321 /* GL locking for state handlers is done by the caller. */
 322 static BOOL need_rel_addr_const(const struct arb_vshader_private *shader_data,
 323         const struct wined3d_shader_reg_maps *reg_maps, const struct wined3d_gl_info *gl_info)
 324 {
 325     if (shader_data->rel_offset) return TRUE;
 326     if (!reg_maps->usesmova) return FALSE;
 327     return !gl_info->supported[NV_VERTEX_PROGRAM2_OPTION];
 328 }
 329
 330 /* Returns TRUE if result.clip from GL_NV_vertex_program2 should be used and FALSE otherwise */
 331 static inline BOOL use_nv_clip(const struct wined3d_gl_info *gl_info)
 332 {
 333     return gl_info->supported[NV_VERTEX_PROGRAM2_OPTION]
 334             && !(gl_info->quirks & WINED3D_QUIRK_NV_CLIP_BROKEN);
 335 }
 336
 337 static BOOL need_helper_const(const struct arb_vshader_private *shader_data,
 338         const struct wined3d_shader_reg_maps *reg_maps, const struct wined3d_gl_info *gl_info)
 339 {
 340     if (need_rel_addr_const(shader_data, reg_maps, gl_info)) return TRUE;
 341     if (!gl_info->supported[NV_VERTEX_PROGRAM]) return TRUE; /* Need to init colors. */
 342     if (gl_info->quirks & WINED3D_QUIRK_ARB_VS_OFFSET_LIMIT) return TRUE; /* Load the immval offset. */
 343     if (gl_info->quirks & WINED3D_QUIRK_SET_TEXCOORD_W) return TRUE; /* Have to init texcoords. */
 344     if (!use_nv_clip(gl_info)) return TRUE; /* Init the clip texcoord */
 345     if (reg_maps->usesnrm) return TRUE; /* 0.0 */
 346     if (reg_maps->usespow) return TRUE; /* EPS, 0.0 and 1.0 */
 347     if (reg_maps->fog) return TRUE; /* Clamping fog coord, 0.0 and 1.0 */
 348     return FALSE;
 349 }
 350
 351 static unsigned int reserved_vs_const(const struct arb_vshader_private *shader_data,
 352         const struct wined3d_shader_reg_maps *reg_maps, const struct wined3d_gl_info *gl_info)
 353 {
 354     unsigned int ret = 1;
 355     /* We use one PARAM for the pos fixup, and in some cases one to load
 356      * some immediate values into the shader. */
 357     if (need_helper_const(shader_data, reg_maps, gl_info)) ++ret;
 358     if (need_rel_addr_const(shader_data, reg_maps, gl_info)) ++ret;
 359     return ret;
 360 }
 361
 362 /* Loads floating point constants into the currently set ARB_vertex/fragment_program.
 363  * When constant_list == NULL, it will load all the constants.
 364  *
 365  * @target_type should be either GL_VERTEX_PROGRAM_ARB (for vertex shaders)
 366  *  or GL_FRAGMENT_PROGRAM_ARB (for pixel shaders)
 367  */
 368 /* GL locking is done by the caller */
 369 static unsigned int shader_arb_load_constantsF(const struct wined3d_shader *shader,
 370         const struct wined3d_gl_info *gl_info, GLuint target_type, unsigned int max_constants,
 371         const float *constants, char *dirty_consts)
 372 {
 373     struct wined3d_shader_lconst *lconst;
 374     DWORD i, j;
 375     unsigned int ret;
 376
 377     if (TRACE_ON(d3d_constants))
 378     {
 379         for(i = 0; i < max_constants; i++) {
 380             if(!dirty_consts[i]) continue;
 381             TRACE_(d3d_constants)("Loading constants %i: %f, %f, %f, %f\n", i,
 382                         constants[i * 4 + 0], constants[i * 4 + 1],
 383                         constants[i * 4 + 2], constants[i * 4 + 3]);
 384         }
 385     }
 386
 387     i = 0;
 388
 389     /* In 1.X pixel shaders constants are implicitly clamped in the range [-1;1] */
 390     if (target_type == GL_FRAGMENT_PROGRAM_ARB && shader->reg_maps.shader_version.major == 1)
 391     {
 392         float lcl_const[4];
 393         /* ps 1.x supports only 8 constants, clamp only those. When switching between 1.x and higher
 394          * shaders, the first 8 constants are marked dirty for reload
 395          */
 396         for(; i < min(8, max_constants); i++) {
 397             if(!dirty_consts[i]) continue;
 398             dirty_consts[i] = 0;
 399
 400             j = 4 * i;
 401             if (constants[j + 0] > 1.0f) lcl_const[0] = 1.0f;
 402             else if (constants[j + 0] < -1.0f) lcl_const[0] = -1.0f;
 403             else lcl_const[0] = constants[j + 0];
 404
 405             if (constants[j + 1] > 1.0f) lcl_const[1] = 1.0f;
 406             else if (constants[j + 1] < -1.0f) lcl_const[1] = -1.0f;
 407             else lcl_const[1] = constants[j + 1];
 408
 409             if (constants[j + 2] > 1.0f) lcl_const[2] = 1.0f;
 410             else if (constants[j + 2] < -1.0f) lcl_const[2] = -1.0f;
 411             else lcl_const[2] = constants[j + 2];
 412
 413             if (constants[j + 3] > 1.0f) lcl_const[3] = 1.0f;
 414             else if (constants[j + 3] < -1.0f) lcl_const[3] = -1.0f;
 415             else lcl_const[3] = constants[j + 3];
 416
 417             GL_EXTCALL(glProgramEnvParameter4fvARB(target_type, i, lcl_const));
 418         }
 419
 420         /* If further constants are dirty, reload them without clamping.
 421          *
 422          * The alternative is not to touch them, but then we cannot reset the dirty constant count
 423          * to zero. That's bad for apps that only use PS 1.x shaders, because in that case the code
 424          * above would always re-check the first 8 constants since max_constant remains at the init
 425          * value
 426          */
 427     }
 428
 429     if (gl_info->supported[EXT_GPU_PROGRAM_PARAMETERS])
 430     {
 431         /* TODO: Benchmark if we're better of with finding the dirty constants ourselves,
 432          * or just reloading *all* constants at once
 433          *
 434         GL_EXTCALL(glProgramEnvParameters4fvEXT(target_type, i, max_constants, constants + (i * 4)));
 435          */
 436         for(; i < max_constants; i++) {
 437             if(!dirty_consts[i]) continue;
 438
 439             /* Find the next block of dirty constants */
 440             dirty_consts[i] = 0;
 441             j = i;
 442             for(i++; (i < max_constants) && dirty_consts[i]; i++) {
 443                 dirty_consts[i] = 0;
 444             }
 445
 446             GL_EXTCALL(glProgramEnvParameters4fvEXT(target_type, j, i - j, constants + (j * 4)));
 447         }
 448     } else {
 449         for(; i < max_constants; i++) {
 450             if(dirty_consts[i]) {
 451                 dirty_consts[i] = 0;
 452                 GL_EXTCALL(glProgramEnvParameter4fvARB(target_type, i, constants + (i * 4)));
 453             }
 454         }
 455     }
 456     checkGLcall("glProgramEnvParameter4fvARB()");
 457
 458     /* Load immediate constants */
 459     if (shader->load_local_constsF)
 460     {
 461         if (TRACE_ON(d3d_shader))
 462         {
 463             LIST_FOR_EACH_ENTRY(lconst, &shader->constantsF, struct wined3d_shader_lconst, entry)
 464             {
 465                 GLfloat* values = (GLfloat*)lconst->value;
 466                 TRACE_(d3d_constants)("Loading local constants %i: %f, %f, %f, %f\n", lconst->idx,
 467                         values[0], values[1], values[2], values[3]);
 468             }
 469         }
 470         /* Immediate constants are clamped for 1.X shaders at loading times */
 471         ret = 0;
 472         LIST_FOR_EACH_ENTRY(lconst, &shader->constantsF, struct wined3d_shader_lconst, entry)
 473         {
 474             dirty_consts[lconst->idx] = 1; /* Dirtify so the non-immediate constant overwrites it next time */
 475             ret = max(ret, lconst->idx + 1);
 476             GL_EXTCALL(glProgramEnvParameter4fvARB(target_type, lconst->idx, (GLfloat*)lconst->value));
 477         }
 478         checkGLcall("glProgramEnvParameter4fvARB()");
 479         return ret; /* The loaded immediate constants need reloading for the next shader */
 480     } else {
 481         return 0; /* No constants are dirty now */
 482     }
 483 }
 484
 485 /**
 486  * Loads the texture dimensions for NP2 fixup into the currently set ARB_[vertex/fragment]_programs.
 487  */
 488 static void shader_arb_load_np2fixup_constants(void *shader_priv,
 489         const struct wined3d_gl_info *gl_info, const struct wined3d_state *state)
 490 {
 491     const struct shader_arb_priv * priv = shader_priv;
 492
 493     /* NP2 texcoord fixup is (currently) only done for pixelshaders. */
 494     if (!use_ps(state)) return;
 495
 496     if (priv->compiled_fprog && priv->compiled_fprog->np2fixup_info.super.active) {
 497         const struct arb_ps_np2fixup_info* const fixup = &priv->compiled_fprog->np2fixup_info;
 498         UINT i;
 499         WORD active = fixup->super.active;
 500         GLfloat np2fixup_constants[4 * MAX_FRAGMENT_SAMPLERS];
 501
 502         for (i = 0; active; active >>= 1, ++i)
 503         {
 504             const struct wined3d_texture *tex = state->textures[i];
 505             const unsigned char idx = fixup->super.idx[i];
 506             GLfloat *tex_dim = &np2fixup_constants[(idx >> 1) * 4];
 507
 508             if (!(active & 1)) continue;
 509
 510             if (!tex) {
 511                 FIXME("Nonexistent texture is flagged for NP2 texcoord fixup\n");
 512                 continue;
 513             }
 514
 515             if (idx % 2)
 516             {
 517                 tex_dim[2] = tex->pow2_matrix[0];
 518                 tex_dim[3] = tex->pow2_matrix[5];
 519             }
 520             else
 521             {
 522                 tex_dim[0] = tex->pow2_matrix[0];
 523                 tex_dim[1] = tex->pow2_matrix[5];
 524             }
 525         }
 526
 527         for (i = 0; i < fixup->super.num_consts; ++i) {
 528             GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB,
 529                                                    fixup->offset + i, &np2fixup_constants[i * 4]));
 530         }
 531     }
 532 }
 533
 534 /* GL locking is done by the caller. */
 535 static void shader_arb_ps_local_constants(const struct arb_ps_compiled_shader *gl_shader,
 536         const struct wined3d_context *context, const struct wined3d_state *state, UINT rt_height)
 537 {
 538     const struct wined3d_gl_info *gl_info = context->gl_info;
 539     unsigned char i;
 540
 541     for(i = 0; i < gl_shader->numbumpenvmatconsts; i++)
 542     {
 543         int texunit = gl_shader->bumpenvmatconst[i].texunit;
 544
 545         /* The state manager takes care that this function is always called if the bump env matrix changes */
 546         const float *data = (const float *)&state->texture_states[texunit][WINED3D_TSS_BUMPENV_MAT00];
 547         GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB,
 548                 gl_shader->bumpenvmatconst[i].const_num, data));
 549
 550         if (gl_shader->luminanceconst[i].const_num != WINED3D_CONST_NUM_UNUSED)
 551         {
 552             /* WINED3D_TSS_BUMPENVLSCALE and WINED3D_TSS_BUMPENVLOFFSET are next to each other.
 553              * point gl to the scale, and load 4 floats. x = scale, y = offset, z and w are junk, we
 554              * don't care about them. The pointers are valid for sure because the stateblock is bigger.
 555              * (they're WINED3D_TSS_TEXTURETRANSFORMFLAGS and WINED3D_TSS_ADDRESSW, so most likely 0 or NaN
 556             */
 557             const float *scale = (const float *)&state->texture_states[texunit][WINED3D_TSS_BUMPENV_LSCALE];
 558             GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB,
 559                     gl_shader->luminanceconst[i].const_num, scale));
 560         }
 561     }
 562     checkGLcall("Load bumpmap consts");
 563
 564     if(gl_shader->ycorrection != WINED3D_CONST_NUM_UNUSED)
 565     {
 566         /* ycorrection.x: Backbuffer height(onscreen) or 0(offscreen).
 567         * ycorrection.y: -1.0(onscreen), 1.0(offscreen)
 568         * ycorrection.z: 1.0
 569         * ycorrection.w: 0.0
 570         */
 571         float val[4];
 572         val[0] = context->render_offscreen ? 0.0f : (float) rt_height;
 573         val[1] = context->render_offscreen ? 1.0f : -1.0f;
 574         val[2] = 1.0f;
 575         val[3] = 0.0f;
 576         GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, gl_shader->ycorrection, val));
 577         checkGLcall("y correction loading");
 578     }
 579
 580     if (!gl_shader->num_int_consts) return;
 581
 582     for(i = 0; i < MAX_CONST_I; i++)
 583     {
 584         if(gl_shader->int_consts[i] != WINED3D_CONST_NUM_UNUSED)
 585         {
 586             float val[4];
 587             val[0] = (float)state->ps_consts_i[4 * i];
 588             val[1] = (float)state->ps_consts_i[4 * i + 1];
 589             val[2] = (float)state->ps_consts_i[4 * i + 2];
 590             val[3] = -1.0f;
 591
 592             GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, gl_shader->int_consts[i], val));
 593         }
 594     }
 595     checkGLcall("Load ps int consts");
 596 }
 597
 598 /* GL locking is done by the caller. */
 599 static void shader_arb_vs_local_constants(const struct arb_vs_compiled_shader *gl_shader,
 600         const struct wined3d_context *context, const struct wined3d_state *state)
 601 {
 602     const struct wined3d_gl_info *gl_info = context->gl_info;
 603     float position_fixup[4];
 604     unsigned char i;
 605
 606     /* Upload the position fixup */
 607     shader_get_position_fixup(context, state, position_fixup);
 608     GL_EXTCALL(glProgramLocalParameter4fvARB(GL_VERTEX_PROGRAM_ARB, gl_shader->pos_fixup, position_fixup));
 609
 610     if (!gl_shader->num_int_consts) return;
 611
 612     for(i = 0; i < MAX_CONST_I; i++)
 613     {
 614         if(gl_shader->int_consts[i] != WINED3D_CONST_NUM_UNUSED)
 615         {
 616             float val[4];
 617             val[0] = (float)state->vs_consts_i[4 * i];
 618             val[1] = (float)state->vs_consts_i[4 * i + 1];
 619             val[2] = (float)state->vs_consts_i[4 * i + 2];
 620             val[3] = -1.0f;
 621
 622             GL_EXTCALL(glProgramLocalParameter4fvARB(GL_VERTEX_PROGRAM_ARB, gl_shader->int_consts[i], val));
 623         }
 624     }
 625     checkGLcall("Load vs int consts");
 626 }
 627
 628 static void shader_arb_select(const struct wined3d_context *context, BOOL usePS, BOOL useVS);
 629
 630 /**
 631  * Loads the app-supplied constants into the currently set ARB_[vertex/fragment]_programs.
 632  *
 633  * We only support float constants in ARB at the moment, so don't
 634  * worry about the Integers or Booleans
 635  */
 636 /* GL locking is done by the caller (state handler) */
 637 static void shader_arb_load_constants_internal(const struct wined3d_context *context,
 638         BOOL usePixelShader, BOOL useVertexShader, BOOL from_shader_select)
 639 {
 640     struct wined3d_device *device = context->swapchain->device;
 641     const struct wined3d_stateblock *stateblock = device->stateBlock;
 642     const struct wined3d_state *state = &stateblock->state;
 643     const struct wined3d_gl_info *gl_info = context->gl_info;
 644     struct shader_arb_priv *priv = device->shader_priv;
 645
 646     if (!from_shader_select)
 647     {
 648         const struct wined3d_shader *vshader = state->vertex_shader, *pshader = state->pixel_shader;
 649         if (vshader
 650                 && (stateblock->changed.vertexShaderConstantsB & vshader->reg_maps.boolean_constants
 651                 || (!gl_info->supported[NV_VERTEX_PROGRAM2_OPTION]
 652                 && (stateblock->changed.vertexShaderConstantsI
 653                 & vshader->reg_maps.integer_constants & ~vshader->reg_maps.local_int_consts))))
 654         {
 655             TRACE("bool/integer vertex shader constants potentially modified, forcing shader reselection.\n");
 656             shader_arb_select(context, usePixelShader, useVertexShader);
 657         }
 658         else if (pshader
 659                 && (stateblock->changed.pixelShaderConstantsB & pshader->reg_maps.boolean_constants
 660                 || (!gl_info->supported[NV_FRAGMENT_PROGRAM_OPTION]
 661                 && (stateblock->changed.pixelShaderConstantsI
 662                 & pshader->reg_maps.integer_constants & ~pshader->reg_maps.local_int_consts))))
 663         {
 664             TRACE("bool/integer pixel shader constants potentially modified, forcing shader reselection.\n");
 665             shader_arb_select(context, usePixelShader, useVertexShader);
 666         }
 667     }
 668
 669     if (context != priv->last_context)
 670     {
 671         memset(priv->vshader_const_dirty, 1,
 672                 sizeof(*priv->vshader_const_dirty) * device->d3d_vshader_constantF);
 673         priv->highest_dirty_vs_const = device->d3d_vshader_constantF;
 674
 675         memset(priv->pshader_const_dirty, 1,
 676                 sizeof(*priv->pshader_const_dirty) * device->d3d_pshader_constantF);
 677         priv->highest_dirty_ps_const = device->d3d_pshader_constantF;
 678
 679         priv->last_context = context;
 680     }
 681
 682     if (useVertexShader)
 683     {
 684         struct wined3d_shader *vshader = state->vertex_shader;
 685         const struct arb_vs_compiled_shader *gl_shader = priv->compiled_vprog;
 686
 687         /* Load DirectX 9 float constants for vertex shader */
 688         priv->highest_dirty_vs_const = shader_arb_load_constantsF(vshader, gl_info, GL_VERTEX_PROGRAM_ARB,
 689                 priv->highest_dirty_vs_const, state->vs_consts_f, priv->vshader_const_dirty);
 690         shader_arb_vs_local_constants(gl_shader, context, state);
 691     }
 692
 693     if (usePixelShader)
 694     {
 695         struct wined3d_shader *pshader = state->pixel_shader;
 696         const struct arb_ps_compiled_shader *gl_shader = priv->compiled_fprog;
 697         UINT rt_height = state->fb->render_targets[0]->resource.height;
 698
 699         /* Load DirectX 9 float constants for pixel shader */
 700         priv->highest_dirty_ps_const = shader_arb_load_constantsF(pshader, gl_info, GL_FRAGMENT_PROGRAM_ARB,
 701                 priv->highest_dirty_ps_const, state->ps_consts_f, priv->pshader_const_dirty);
 702         shader_arb_ps_local_constants(gl_shader, context, state, rt_height);
 703     }
 704 }
 705
 706 static void shader_arb_load_constants(const struct wined3d_context *context, BOOL ps, BOOL vs)
 707 {
 708     shader_arb_load_constants_internal(context, ps, vs, FALSE);
 709 }
 710
 711 static void shader_arb_update_float_vertex_constants(struct wined3d_device *device, UINT start, UINT count)
 712 {
 713     struct wined3d_context *context = context_get_current();
 714     struct shader_arb_priv *priv = device->shader_priv;
 715
 716     /* We don't want shader constant dirtification to be an O(contexts), so just dirtify the active
 717      * context. On a context switch the old context will be fully dirtified */
 718     if (!context || context->swapchain->device != device) return;
 719
 720     memset(priv->vshader_const_dirty + start, 1, sizeof(*priv->vshader_const_dirty) * count);
 721     priv->highest_dirty_vs_const = max(priv->highest_dirty_vs_const, start + count);
 722 }
 723
 724 static void shader_arb_update_float_pixel_constants(struct wined3d_device *device, UINT start, UINT count)
 725 {
 726     struct wined3d_context *context = context_get_current();
 727     struct shader_arb_priv *priv = device->shader_priv;
 728
 729     /* We don't want shader constant dirtification to be an O(contexts), so just dirtify the active
 730      * context. On a context switch the old context will be fully dirtified */
 731     if (!context || context->swapchain->device != device) return;
 732
 733     memset(priv->pshader_const_dirty + start, 1, sizeof(*priv->pshader_const_dirty) * count);
 734     priv->highest_dirty_ps_const = max(priv->highest_dirty_ps_const, start + count);
 735 }
 736
 737 static DWORD *local_const_mapping(const struct wined3d_shader *shader)
 738 {
 739     const struct wined3d_shader_lconst *lconst;
 740     DWORD *ret;
 741     DWORD idx = 0;
 742
 743     if (shader->load_local_constsF || list_empty(&shader->constantsF))
 744         return NULL;
 745
 746     ret = HeapAlloc(GetProcessHeap(), 0, sizeof(DWORD) * shader->limits.constant_float);
 747     if (!ret)
 748     {
 749         ERR("Out of memory\n");
 750         return NULL;
 751     }
 752
 753     LIST_FOR_EACH_ENTRY(lconst, &shader->constantsF, struct wined3d_shader_lconst, entry)
 754     {
 755         ret[lconst->idx] = idx++;
 756     }
 757     return ret;
 758 }
 759
 760 /* Generate the variable & register declarations for the ARB_vertex_program output target */
 761 static DWORD shader_generate_arb_declarations(const struct wined3d_shader *shader,
 762         const struct wined3d_shader_reg_maps *reg_maps, struct wined3d_shader_buffer *buffer,
 763         const struct wined3d_gl_info *gl_info, const DWORD *lconst_map,
 764         DWORD *num_clipplanes, const struct shader_arb_ctx_priv *ctx)
 765 {
 766     DWORD i, next_local = 0;
 767     char pshader = shader_is_pshader_version(reg_maps->shader_version.type);
 768     const struct wined3d_shader_lconst *lconst;
 769     unsigned max_constantsF;
 770     DWORD map;
 771
 772     /* In pixel shaders, all private constants are program local, we don't need anything
 773      * from program.env. Thus we can advertise the full set of constants in pixel shaders.
 774      * If we need a private constant the GL implementation will squeeze it in somewhere
 775      *
 776      * With vertex shaders we need the posFixup and on some GL implementations 4 helper
 777      * immediate values. The posFixup is loaded using program.env for now, so always
 778      * subtract one from the number of constants. If the shader uses indirect addressing,
 779      * account for the helper const too because we have to declare all available d3d constants
 780      * and don't know which are actually used.
 781      */
 782     if (pshader)
 783     {
 784         max_constantsF = gl_info->limits.arb_ps_native_constants;
 785         /* 24 is the minimum MAX_PROGRAM_ENV_PARAMETERS_ARB value. */
 786         if (max_constantsF < 24)
 787             max_constantsF = gl_info->limits.arb_ps_float_constants;
 788     }
 789     else
 790     {
 791         const struct arb_vshader_private *shader_data = shader->backend_data;
 792         max_constantsF = gl_info->limits.arb_vs_native_constants;
 793         /* 96 is the minimum MAX_PROGRAM_ENV_PARAMETERS_ARB value.
 794          * Also prevents max_constantsF from becoming less than 0 and
 795          * wrapping . */
 796         if (max_constantsF < 96)
 797             max_constantsF = gl_info->limits.arb_vs_float_constants;
 798
 799         if (reg_maps->usesrelconstF)
 800         {
 801             DWORD highest_constf = 0, clip_limit;
 802
 803             max_constantsF -= reserved_vs_const(shader_data, reg_maps, gl_info);
 804             max_constantsF -= count_bits(reg_maps->integer_constants);
 805
 806             for (i = 0; i < shader->limits.constant_float; ++i)
 807             {
 808                 DWORD idx = i >> 5;
 809                 DWORD shift = i & 0x1f;
 810                 if(reg_maps->constf[idx] & (1 << shift)) highest_constf = i;
 811             }
 812
 813             if(use_nv_clip(gl_info) && ctx->target_version >= NV2)
 814             {
 815                 if(ctx->cur_vs_args->super.clip_enabled)
 816                     clip_limit = gl_info->limits.clipplanes;
 817                 else
 818                     clip_limit = 0;
 819             }
 820             else
 821             {
 822                 unsigned int mask = ctx->cur_vs_args->clip.boolclip.clipplane_mask;
 823                 clip_limit = min(count_bits(mask), 4);
 824             }
 825             *num_clipplanes = min(clip_limit, max_constantsF - highest_constf - 1);
 826             max_constantsF -= *num_clipplanes;
 827             if(*num_clipplanes < clip_limit)
 828             {
 829                 WARN("Only %u clipplanes out of %u enabled\n", *num_clipplanes, gl_info->limits.clipplanes);
 830             }
 831         }
 832         else
 833         {
 834             if (ctx->target_version >= NV2) *num_clipplanes = gl_info->limits.clipplanes;
 835             else *num_clipplanes = min(gl_info->limits.clipplanes, 4);
 836         }
 837     }
 838
 839     for (i = 0, map = reg_maps->temporary; map; map >>= 1, ++i)
 840     {
 841         if (map & 1) shader_addline(buffer, "TEMP R%u;\n", i);
 842     }
 843
 844     for (i = 0, map = reg_maps->address; map; map >>= 1, ++i)
 845     {
 846         if (map & 1) shader_addline(buffer, "ADDRESS A%u;\n", i);
 847     }
 848
 849     if (pshader && reg_maps->shader_version.major == 1 && reg_maps->shader_version.minor <= 3)
 850     {
 851         for (i = 0, map = reg_maps->texcoord; map; map >>= 1, ++i)
 852         {
 853             if (map & 1) shader_addline(buffer, "TEMP T%u;\n", i);
 854         }
 855     }
 856
 857     /* Load local constants using the program-local space,
 858      * this avoids reloading them each time the shader is used
 859      */
 860     if (lconst_map)
 861     {
 862         LIST_FOR_EACH_ENTRY(lconst, &shader->constantsF, struct wined3d_shader_lconst, entry)
 863         {
 864             shader_addline(buffer, "PARAM C%u = program.local[%u];\n", lconst->idx,
 865                            lconst_map[lconst->idx]);
 866             next_local = max(next_local, lconst_map[lconst->idx] + 1);
 867         }
 868     }
 869
 870     /* After subtracting privately used constants from the hardware limit(they are loaded as
 871      * local constants), make sure the shader doesn't violate the env constant limit
 872      */
 873     if(pshader)
 874     {
 875         max_constantsF = min(max_constantsF, gl_info->limits.arb_ps_float_constants);
 876     }
 877     else
 878     {
 879         max_constantsF = min(max_constantsF, gl_info->limits.arb_vs_float_constants);
 880     }
 881
 882     /* Avoid declaring more constants than needed */
 883     max_constantsF = min(max_constantsF, shader->limits.constant_float);
 884
 885     /* we use the array-based constants array if the local constants are marked for loading,
 886      * because then we use indirect addressing, or when the local constant list is empty,
 887      * because then we don't know if we're using indirect addressing or not. If we're hardcoding
 888      * local constants do not declare the loaded constants as an array because ARB compilers usually
 889      * do not optimize unused constants away
 890      */
 891     if (reg_maps->usesrelconstF)
 892     {
 893         /* Need to PARAM the environment parameters (constants) so we can use relative addressing */
 894         shader_addline(buffer, "PARAM C[%d] = { program.env[0..%d] };\n",
 895                     max_constantsF, max_constantsF - 1);
 896     } else {
 897         for(i = 0; i < max_constantsF; i++) {
 898             DWORD idx, mask;
 899             idx = i >> 5;
 900             mask = 1 << (i & 0x1f);
 901             if (!shader_constant_is_local(shader, i) && (reg_maps->constf[idx] & mask))
 902             {
 903                 shader_addline(buffer, "PARAM C%d = program.env[%d];\n",i, i);
 904             }
 905         }
 906     }
 907
 908     return next_local;
 909 }
 910
 911 static const char * const shift_tab[] = {
 912     "dummy",     /*  0 (none) */
 913     "coefmul.x", /*  1 (x2)   */
 914     "coefmul.y", /*  2 (x4)   */
 915     "coefmul.z", /*  3 (x8)   */
 916     "coefmul.w", /*  4 (x16)  */
 917     "dummy",     /*  5 (x32)  */
 918     "dummy",     /*  6 (x64)  */
 919     "dummy",     /*  7 (x128) */
 920     "dummy",     /*  8 (d256) */
 921     "dummy",     /*  9 (d128) */
 922     "dummy",     /* 10 (d64)  */
 923     "dummy",     /* 11 (d32)  */
 924     "coefdiv.w", /* 12 (d16)  */
 925     "coefdiv.z", /* 13 (d8)   */
 926     "coefdiv.y", /* 14 (d4)   */
 927     "coefdiv.x"  /* 15 (d2)   */
 928 };
 929
 930 static void shader_arb_get_write_mask(const struct wined3d_shader_instruction *ins,
 931         const struct wined3d_shader_dst_param *dst, char *write_mask)
 932 {
 933     char *ptr = write_mask;
 934
 935     if (dst->write_mask != WINED3DSP_WRITEMASK_ALL)
 936     {
 937         *ptr++ = '.';
 938         if (dst->write_mask & WINED3DSP_WRITEMASK_0) *ptr++ = 'x';
 939         if (dst->write_mask & WINED3DSP_WRITEMASK_1) *ptr++ = 'y';
 940         if (dst->write_mask & WINED3DSP_WRITEMASK_2) *ptr++ = 'z';
 941         if (dst->write_mask & WINED3DSP_WRITEMASK_3) *ptr++ = 'w';
 942     }
 943
 944     *ptr = '\0';
 945 }
 946
 947 static void shader_arb_get_swizzle(const struct wined3d_shader_src_param *param, BOOL fixup, char *swizzle_str)
 948 {
 949     /* For registers of type WINED3DDECLTYPE_D3DCOLOR, data is stored as "bgra",
 950      * but addressed as "rgba". To fix this we need to swap the register's x
 951      * and z components. */
 952     const char *swizzle_chars = fixup ? "zyxw" : "xyzw";
 953     char *ptr = swizzle_str;
 954
 955     /* swizzle bits fields: wwzzyyxx */
 956     DWORD swizzle = param->swizzle;
 957     DWORD swizzle_x = swizzle & 0x03;
 958     DWORD swizzle_y = (swizzle >> 2) & 0x03;
 959     DWORD swizzle_z = (swizzle >> 4) & 0x03;
 960     DWORD swizzle_w = (swizzle >> 6) & 0x03;
 961
 962     /* If the swizzle is the default swizzle (ie, "xyzw"), we don't need to
 963      * generate a swizzle string. Unless we need to our own swizzling. */
 964     if (swizzle != WINED3DSP_NOSWIZZLE || fixup)
 965     {
 966         *ptr++ = '.';
 967         if (swizzle_x == swizzle_y && swizzle_x == swizzle_z && swizzle_x == swizzle_w) {
 968             *ptr++ = swizzle_chars[swizzle_x];
 969         } else {
 970             *ptr++ = swizzle_chars[swizzle_x];
 971             *ptr++ = swizzle_chars[swizzle_y];
 972             *ptr++ = swizzle_chars[swizzle_z];
 973             *ptr++ = swizzle_chars[swizzle_w];
 974         }
 975     }
 976
 977     *ptr = '\0';
 978 }
 979
 980 static void shader_arb_request_a0(const struct wined3d_shader_instruction *ins, const char *src)
 981 {
 982     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
 983     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
 984
 985     if (!strcmp(priv->addr_reg, src)) return;
 986
 987     strcpy(priv->addr_reg, src);
 988     shader_addline(buffer, "ARL A0.x, %s;\n", src);
 989 }
 990
 991 static void shader_arb_get_src_param(const struct wined3d_shader_instruction *ins,
 992         const struct wined3d_shader_src_param *src, unsigned int tmpreg, char *outregstr);
 993
 994 static void shader_arb_get_register_name(const struct wined3d_shader_instruction *ins,
 995         const struct wined3d_shader_register *reg, char *register_name, BOOL *is_color)
 996 {
 997     /* oPos, oFog and oPts in D3D */
 998     static const char * const rastout_reg_names[] = {"TMP_OUT", "TMP_FOGCOORD", "result.pointsize"};
 999     const struct wined3d_shader *shader = ins->ctx->shader;
1000     const struct wined3d_shader_reg_maps *reg_maps = ins->ctx->reg_maps;
1001     BOOL pshader = shader_is_pshader_version(reg_maps->shader_version.type);
1002     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
1003
1004     *is_color = FALSE;
1005
1006     switch (reg->type)
1007     {
1008         case WINED3DSPR_TEMP:
1009             sprintf(register_name, "R%u", reg->idx);
1010             break;
1011
1012         case WINED3DSPR_INPUT:
1013             if (pshader)
1014             {
1015                 if (reg_maps->shader_version.major < 3)
1016                 {
1017                     if (!reg->idx) strcpy(register_name, "fragment.color.primary");
1018                     else strcpy(register_name, "fragment.color.secondary");
1019                 }
1020                 else
1021                 {
1022                     if(reg->rel_addr)
1023                     {
1024                         char rel_reg[50];
1025                         shader_arb_get_src_param(ins, reg->rel_addr, 0, rel_reg);
1026
1027                         if (!strcmp(rel_reg, "**aL_emul**"))
1028                         {
1029                             DWORD idx = ctx->aL + reg->idx;
1030                             if(idx < MAX_REG_INPUT)
1031                             {
1032                                 strcpy(register_name, ctx->ps_input[idx]);
1033                             }
1034                             else
1035                             {
1036                                 ERR("Pixel shader input register out of bounds: %u\n", idx);
1037                                 sprintf(register_name, "out_of_bounds_%u", idx);
1038                             }
1039                         }
1040                         else if (reg_maps->input_registers & 0x0300)
1041                         {
1042                             /* There are two ways basically:
1043                              *
1044                              * 1) Use the unrolling code that is used for loop emulation and unroll the loop.
1045                              *    That means trouble if the loop also contains a breakc or if the control values
1046                              *    aren't local constants.
1047                              * 2) Generate an if block that checks if aL.y < 8, == 8 or == 9 and selects the
1048                              *    source dynamically. The trouble is that we cannot simply read aL.y because it
1049                              *    is an ADDRESS register. We could however push it, load .zw with a value and use
1050                              *    ADAC to load the condition code register and pop it again afterwards
1051                              */
1052                             FIXME("Relative input register addressing with more than 8 registers\n");
1053
1054                             /* This is better than nothing for now */
1055                             sprintf(register_name, "fragment.texcoord[%s + %u]", rel_reg, reg->idx);
1056                         }
1057                         else if(ctx->cur_ps_args->super.vp_mode != vertexshader)
1058                         {
1059                             /* This is problematic because we'd have to consult the ctx->ps_input strings
1060                              * for where to find the varying. Some may be "0.0", others can be texcoords or
1061                              * colors. This needs either a pipeline replacement to make the vertex shader feed
1062                              * proper varyings, or loop unrolling
1063                              *
1064                              * For now use the texcoords and hope for the best
1065                              */
1066                             FIXME("Non-vertex shader varying input with indirect addressing\n");
1067                             sprintf(register_name, "fragment.texcoord[%s + %u]", rel_reg, reg->idx);
1068                         }
1069                         else
1070                         {
1071                             /* D3D supports indirect addressing only with aL in loop registers. The loop instruction
1072                              * pulls GL_NV_fragment_program2 in
1073                              */
1074                             sprintf(register_name, "fragment.texcoord[%s + %u]", rel_reg, reg->idx);
1075                         }
1076                     }
1077                     else
1078                     {
1079                         if(reg->idx < MAX_REG_INPUT)
1080                         {
1081                             strcpy(register_name, ctx->ps_input[reg->idx]);
1082                         }
1083                         else
1084                         {
1085                             ERR("Pixel shader input register out of bounds: %u\n", reg->idx);
1086                             sprintf(register_name, "out_of_bounds_%u", reg->idx);
1087                         }
1088                     }
1089                 }
1090             }
1091             else
1092             {
1093                 if (ctx->cur_vs_args->super.swizzle_map & (1 << reg->idx)) *is_color = TRUE;
1094                 sprintf(register_name, "vertex.attrib[%u]", reg->idx);
1095             }
1096             break;
1097
1098         case WINED3DSPR_CONST:
1099             if (!pshader && reg->rel_addr)
1100             {
1101                 const struct arb_vshader_private *shader_data = shader->backend_data;
1102                 UINT rel_offset = shader_data->rel_offset;
1103                 BOOL aL = FALSE;
1104                 char rel_reg[50];
1105                 if (reg_maps->shader_version.major < 2)
1106                 {
1107                     sprintf(rel_reg, "A0.x");
1108                 } else {
1109                     shader_arb_get_src_param(ins, reg->rel_addr, 0, rel_reg);
1110                     if(ctx->target_version == ARB) {
1111                         if (!strcmp(rel_reg, "**aL_emul**"))
1112                         {
1113                             aL = TRUE;
1114                         } else {
1115                             shader_arb_request_a0(ins, rel_reg);
1116                             sprintf(rel_reg, "A0.x");
1117                         }
1118                     }
1119                 }
1120                 if(aL)
1121                     sprintf(register_name, "C[%u]", ctx->aL + reg->idx);
1122                 else if (reg->idx >= rel_offset)
1123                     sprintf(register_name, "C[%s + %u]", rel_reg, reg->idx - rel_offset);
1124                 else
1125                     sprintf(register_name, "C[%s - %u]", rel_reg, rel_offset - reg->idx);
1126             }
1127             else
1128             {
1129                 if (reg_maps->usesrelconstF)
1130                     sprintf(register_name, "C[%u]", reg->idx);
1131                 else
1132                     sprintf(register_name, "C%u", reg->idx);
1133             }
1134             break;
1135
1136         case WINED3DSPR_TEXTURE: /* case WINED3DSPR_ADDR: */
1137             if (pshader)
1138             {
1139                 if (reg_maps->shader_version.major == 1
1140                         && reg_maps->shader_version.minor <= 3)
1141                 {
1142                     /* In ps <= 1.3, Tx is a temporary register as destination to all instructions,
1143                      * and as source to most instructions. For some instructions it is the texcoord
1144                      * input. Those instructions know about the special use
1145                      */
1146                     sprintf(register_name, "T%u", reg->idx);
1147                 } else {
1148                     /* in ps 1.4 and 2.x Tx is always a (read-only) varying */
1149                     sprintf(register_name, "fragment.texcoord[%u]", reg->idx);
1150                 }
1151             }
1152             else
1153             {
1154                 if (reg_maps->shader_version.major == 1 || ctx->target_version >= NV2)
1155                 {
1156                     sprintf(register_name, "A%u", reg->idx);
1157                 }
1158                 else
1159                 {
1160                     sprintf(register_name, "A%u_SHADOW", reg->idx);
1161                 }
1162             }
1163             break;
1164
1165         case WINED3DSPR_COLOROUT:
1166             if (ctx->cur_ps_args->super.srgb_correction && !reg->idx)
1167             {
1168                 strcpy(register_name, "TMP_COLOR");
1169             }
1170             else
1171             {
1172                 if(ctx->cur_ps_args->super.srgb_correction) FIXME("sRGB correction on higher render targets\n");
1173                 if (reg_maps->rt_mask > 1)
1174                 {
1175                     sprintf(register_name, "result.color[%u]", reg->idx);
1176                 }
1177                 else
1178                 {
1179                     strcpy(register_name, "result.color");
1180                 }
1181             }
1182             break;
1183
1184         case WINED3DSPR_RASTOUT:
1185             if(reg->idx == 1) sprintf(register_name, "%s", ctx->fog_output);
1186             else sprintf(register_name, "%s", rastout_reg_names[reg->idx]);
1187             break;
1188
1189         case WINED3DSPR_DEPTHOUT:
1190             strcpy(register_name, "result.depth");
1191             break;
1192
1193         case WINED3DSPR_ATTROUT:
1194         /* case WINED3DSPR_OUTPUT: */
1195             if (pshader) sprintf(register_name, "oD[%u]", reg->idx);
1196             else strcpy(register_name, ctx->color_output[reg->idx]);
1197             break;
1198
1199         case WINED3DSPR_TEXCRDOUT:
1200             if (pshader)
1201             {
1202                 sprintf(register_name, "oT[%u]", reg->idx);
1203             }
1204             else
1205             {
1206                 if (reg_maps->shader_version.major < 3)
1207                 {
1208                     strcpy(register_name, ctx->texcrd_output[reg->idx]);
1209                 }
1210                 else
1211                 {
1212                     strcpy(register_name, ctx->vs_output[reg->idx]);
1213                 }
1214             }
1215             break;
1216
1217         case WINED3DSPR_LOOP:
1218             if(ctx->target_version >= NV2)
1219             {
1220                 /* Pshader has an implicitly declared loop index counter A0.x that cannot be renamed */
1221                 if(pshader) sprintf(register_name, "A0.x");
1222                 else sprintf(register_name, "aL.y");
1223             }
1224             else
1225             {
1226                 /* Unfortunately this code cannot return the value of ctx->aL here. An immediate value
1227                  * would be valid, but if aL is used for indexing(its only use), there's likely an offset,
1228                  * thus the result would be something like C[15 + 30], which is not valid in the ARB program
1229                  * grammar. So return a marker for the emulated aL and intercept it in constant and varying
1230                  * indexing
1231                  */
1232                 sprintf(register_name, "**aL_emul**");
1233             }
1234
1235             break;
1236
1237         case WINED3DSPR_CONSTINT:
1238             sprintf(register_name, "I%u", reg->idx);
1239             break;
1240
1241         case WINED3DSPR_MISCTYPE:
1242             if (!reg->idx)
1243             {
1244                 sprintf(register_name, "vpos");
1245             }
1246             else if(reg->idx == 1)
1247             {
1248                 sprintf(register_name, "fragment.facing.x");
1249             }
1250             else
1251             {
1252                 FIXME("Unknown MISCTYPE register index %u\n", reg->idx);
1253             }
1254             break;
1255
1256         default:
1257             FIXME("Unhandled register type %#x[%u]\n", reg->type, reg->idx);
1258             sprintf(register_name, "unrecognized_register[%u]", reg->idx);
1259             break;
1260     }
1261 }
1262
1263 static void shader_arb_get_dst_param(const struct wined3d_shader_instruction *ins,
1264         const struct wined3d_shader_dst_param *wined3d_dst, char *str)
1265 {
1266     char register_name[255];
1267     char write_mask[6];
1268     BOOL is_color;
1269
1270     shader_arb_get_register_name(ins, &wined3d_dst->reg, register_name, &is_color);
1271     strcpy(str, register_name);
1272
1273     shader_arb_get_write_mask(ins, wined3d_dst, write_mask);
1274     strcat(str, write_mask);
1275 }
1276
1277 static const char *shader_arb_get_fixup_swizzle(enum fixup_channel_source channel_source)
1278 {
1279     switch(channel_source)
1280     {
1281         case CHANNEL_SOURCE_ZERO: return "0";
1282         case CHANNEL_SOURCE_ONE: return "1";
1283         case CHANNEL_SOURCE_X: return "x";
1284         case CHANNEL_SOURCE_Y: return "y";
1285         case CHANNEL_SOURCE_Z: return "z";
1286         case CHANNEL_SOURCE_W: return "w";
1287         default:
1288             FIXME("Unhandled channel source %#x\n", channel_source);
1289             return "undefined";
1290     }
1291 }
1292
1293 static void gen_color_correction(struct wined3d_shader_buffer *buffer, const char *reg,
1294         DWORD dst_mask, const char *one, const char *two, struct color_fixup_desc fixup)
1295 {
1296     DWORD mask;
1297
1298     if (is_complex_fixup(fixup))
1299     {
1300         enum complex_fixup complex_fixup = get_complex_fixup(fixup);
1301         FIXME("Complex fixup (%#x) not supported\n", complex_fixup);
1302         return;
1303     }
1304
1305     mask = 0;
1306     if (fixup.x_source != CHANNEL_SOURCE_X) mask |= WINED3DSP_WRITEMASK_0;
1307     if (fixup.y_source != CHANNEL_SOURCE_Y) mask |= WINED3DSP_WRITEMASK_1;
1308     if (fixup.z_source != CHANNEL_SOURCE_Z) mask |= WINED3DSP_WRITEMASK_2;
1309     if (fixup.w_source != CHANNEL_SOURCE_W) mask |= WINED3DSP_WRITEMASK_3;
1310     mask &= dst_mask;
1311
1312     if (mask)
1313     {
1314         shader_addline(buffer, "SWZ %s, %s, %s, %s, %s, %s;\n", reg, reg,
1315                 shader_arb_get_fixup_swizzle(fixup.x_source), shader_arb_get_fixup_swizzle(fixup.y_source),
1316                 shader_arb_get_fixup_swizzle(fixup.z_source), shader_arb_get_fixup_swizzle(fixup.w_source));
1317     }
1318
1319     mask = 0;
1320     if (fixup.x_sign_fixup) mask |= WINED3DSP_WRITEMASK_0;
1321     if (fixup.y_sign_fixup) mask |= WINED3DSP_WRITEMASK_1;
1322     if (fixup.z_sign_fixup) mask |= WINED3DSP_WRITEMASK_2;
1323     if (fixup.w_sign_fixup) mask |= WINED3DSP_WRITEMASK_3;
1324     mask &= dst_mask;
1325
1326     if (mask)
1327     {
1328         char reg_mask[6];
1329         char *ptr = reg_mask;
1330
1331         if (mask != WINED3DSP_WRITEMASK_ALL)
1332         {
1333             *ptr++ = '.';
1334             if (mask & WINED3DSP_WRITEMASK_0) *ptr++ = 'x';
1335             if (mask & WINED3DSP_WRITEMASK_1) *ptr++ = 'y';
1336             if (mask & WINED3DSP_WRITEMASK_2) *ptr++ = 'z';
1337             if (mask & WINED3DSP_WRITEMASK_3) *ptr++ = 'w';
1338         }
1339         *ptr = '\0';
1340
1341         shader_addline(buffer, "MAD %s%s, %s, %s, -%s;\n", reg, reg_mask, reg, two, one);
1342     }
1343 }
1344
1345 static const char *shader_arb_get_modifier(const struct wined3d_shader_instruction *ins)
1346 {
1347     DWORD mod;
1348     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
1349     if (!ins->dst_count) return "";
1350
1351     mod = ins->dst[0].modifiers;
1352
1353     /* Silently ignore PARTIALPRECISION if its not supported */
1354     if(priv->target_version == ARB) mod &= ~WINED3DSPDM_PARTIALPRECISION;
1355
1356     if(mod & WINED3DSPDM_MSAMPCENTROID)
1357     {
1358         FIXME("Unhandled modifier WINED3DSPDM_MSAMPCENTROID\n");
1359         mod &= ~WINED3DSPDM_MSAMPCENTROID;
1360     }
1361
1362     switch(mod)
1363     {
1364         case WINED3DSPDM_SATURATE | WINED3DSPDM_PARTIALPRECISION:
1365             return "H_SAT";
1366
1367         case WINED3DSPDM_SATURATE:
1368             return "_SAT";
1369
1370         case WINED3DSPDM_PARTIALPRECISION:
1371             return "H";
1372
1373         case 0:
1374             return "";
1375
1376         default:
1377             FIXME("Unknown modifiers 0x%08x\n", mod);
1378             return "";
1379     }
1380 }
1381
1382 #define TEX_PROJ        0x1
1383 #define TEX_BIAS        0x2
1384 #define TEX_LOD         0x4
1385 #define TEX_DERIV       0x10
1386
1387 static void shader_hw_sample(const struct wined3d_shader_instruction *ins, DWORD sampler_idx,
1388         const char *dst_str, const char *coord_reg, WORD flags, const char *dsx, const char *dsy)
1389 {
1390     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1391     DWORD sampler_type = ins->ctx->reg_maps->sampler_type[sampler_idx];
1392     const struct wined3d_shader *shader = ins->ctx->shader;
1393     const struct wined3d_texture *texture;
1394     const char *tex_type;
1395     BOOL np2_fixup = FALSE;
1396     struct wined3d_device *device = shader->device;
1397     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
1398     const char *mod;
1399     BOOL pshader = shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type);
1400
1401     /* D3D vertex shader sampler IDs are vertex samplers(0-3), not global d3d samplers */
1402     if(!pshader) sampler_idx += MAX_FRAGMENT_SAMPLERS;
1403
1404     switch(sampler_type) {
1405         case WINED3DSTT_1D:
1406             tex_type = "1D";
1407             break;
1408
1409         case WINED3DSTT_2D:
1410             texture = device->stateBlock->state.textures[sampler_idx];
1411             if (texture && texture->target == GL_TEXTURE_RECTANGLE_ARB)
1412             {
1413                 tex_type = "RECT";
1414             } else {
1415                 tex_type = "2D";
1416             }
1417             if (shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type))
1418             {
1419                 if (priv->cur_np2fixup_info->super.active & (1 << sampler_idx))
1420                 {
1421                     if (flags) FIXME("Only ordinary sampling from NP2 textures is supported.\n");
1422                     else np2_fixup = TRUE;
1423                 }
1424             }
1425             break;
1426
1427         case WINED3DSTT_VOLUME:
1428             tex_type = "3D";
1429             break;
1430
1431         case WINED3DSTT_CUBE:
1432             tex_type = "CUBE";
1433             break;
1434
1435         default:
1436             ERR("Unexpected texture type %d\n", sampler_type);
1437             tex_type = "";
1438     }
1439
1440     /* TEX, TXL, TXD and TXP do not support the "H" modifier,
1441      * so don't use shader_arb_get_modifier
1442      */
1443     if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE) mod = "_SAT";
1444     else mod = "";
1445
1446     /* Fragment samplers always have indentity mapping */
1447     if(sampler_idx >= MAX_FRAGMENT_SAMPLERS)
1448     {
1449         sampler_idx = priv->cur_vs_args->vertex.samplers[sampler_idx - MAX_FRAGMENT_SAMPLERS];
1450     }
1451
1452     if (flags & TEX_DERIV)
1453     {
1454         if(flags & TEX_PROJ) FIXME("Projected texture sampling with custom derivatives\n");
1455         if(flags & TEX_BIAS) FIXME("Biased texture sampling with custom derivatives\n");
1456         shader_addline(buffer, "TXD%s %s, %s, %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg,
1457                        dsx, dsy,sampler_idx, tex_type);
1458     }
1459     else if(flags & TEX_LOD)
1460     {
1461         if(flags & TEX_PROJ) FIXME("Projected texture sampling with explicit lod\n");
1462         if(flags & TEX_BIAS) FIXME("Biased texture sampling with explicit lod\n");
1463         shader_addline(buffer, "TXL%s %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg,
1464                        sampler_idx, tex_type);
1465     }
1466     else if (flags & TEX_BIAS)
1467     {
1468         /* Shouldn't be possible, but let's check for it */
1469         if(flags & TEX_PROJ) FIXME("Biased and Projected texture sampling\n");
1470         /* TXB takes the 4th component of the source vector automatically, as d3d. Nothing more to do */
1471         shader_addline(buffer, "TXB%s %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg, sampler_idx, tex_type);
1472     }
1473     else if (flags & TEX_PROJ)
1474     {
1475         shader_addline(buffer, "TXP%s %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg, sampler_idx, tex_type);
1476     }
1477     else
1478     {
1479         if (np2_fixup)
1480         {
1481             const unsigned char idx = priv->cur_np2fixup_info->super.idx[sampler_idx];
1482             shader_addline(buffer, "MUL TA, np2fixup[%u].%s, %s;\n", idx >> 1,
1483                            (idx % 2) ? "zwxy" : "xyzw", coord_reg);
1484
1485             shader_addline(buffer, "TEX%s %s, TA, texture[%u], %s;\n", mod, dst_str, sampler_idx, tex_type);
1486         }
1487         else
1488             shader_addline(buffer, "TEX%s %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg, sampler_idx, tex_type);
1489     }
1490
1491     if (pshader)
1492     {
1493         gen_color_correction(buffer, dst_str, ins->dst[0].write_mask,
1494                 arb_get_helper_value(WINED3D_SHADER_TYPE_PIXEL, ARB_ONE),
1495                 arb_get_helper_value(WINED3D_SHADER_TYPE_PIXEL, ARB_TWO),
1496                 priv->cur_ps_args->super.color_fixup[sampler_idx]);
1497     }
1498 }
1499
1500 static void shader_arb_get_src_param(const struct wined3d_shader_instruction *ins,
1501         const struct wined3d_shader_src_param *src, unsigned int tmpreg, char *outregstr)
1502 {
1503     /* Generate a line that does the input modifier computation and return the input register to use */
1504     BOOL is_color = FALSE;
1505     char regstr[256];
1506     char swzstr[20];
1507     int insert_line;
1508     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1509     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
1510     const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
1511     const char *two = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_TWO);
1512
1513     /* Assume a new line will be added */
1514     insert_line = 1;
1515
1516     /* Get register name */
1517     shader_arb_get_register_name(ins, &src->reg, regstr, &is_color);
1518     shader_arb_get_swizzle(src, is_color, swzstr);
1519
1520     switch (src->modifiers)
1521     {
1522     case WINED3DSPSM_NONE:
1523         sprintf(outregstr, "%s%s", regstr, swzstr);
1524         insert_line = 0;
1525         break;
1526     case WINED3DSPSM_NEG:
1527         sprintf(outregstr, "-%s%s", regstr, swzstr);
1528         insert_line = 0;
1529         break;
1530     case WINED3DSPSM_BIAS:
1531         shader_addline(buffer, "ADD T%c, %s, -coefdiv.x;\n", 'A' + tmpreg, regstr);
1532         break;
1533     case WINED3DSPSM_BIASNEG:
1534         shader_addline(buffer, "ADD T%c, -%s, coefdiv.x;\n", 'A' + tmpreg, regstr);
1535         break;
1536     case WINED3DSPSM_SIGN:
1537         shader_addline(buffer, "MAD T%c, %s, %s, -%s;\n", 'A' + tmpreg, regstr, two, one);
1538         break;
1539     case WINED3DSPSM_SIGNNEG:
1540         shader_addline(buffer, "MAD T%c, %s, -%s, %s;\n", 'A' + tmpreg, regstr, two, one);
1541         break;
1542     case WINED3DSPSM_COMP:
1543         shader_addline(buffer, "SUB T%c, %s, %s;\n", 'A' + tmpreg, one, regstr);
1544         break;
1545     case WINED3DSPSM_X2:
1546         shader_addline(buffer, "ADD T%c, %s, %s;\n", 'A' + tmpreg, regstr, regstr);
1547         break;
1548     case WINED3DSPSM_X2NEG:
1549         shader_addline(buffer, "ADD T%c, -%s, -%s;\n", 'A' + tmpreg, regstr, regstr);
1550         break;
1551     case WINED3DSPSM_DZ:
1552         shader_addline(buffer, "RCP T%c, %s.z;\n", 'A' + tmpreg, regstr);
1553         shader_addline(buffer, "MUL T%c, %s, T%c;\n", 'A' + tmpreg, regstr, 'A' + tmpreg);
1554         break;
1555     case WINED3DSPSM_DW:
1556         shader_addline(buffer, "RCP T%c, %s.w;\n", 'A' + tmpreg, regstr);
1557         shader_addline(buffer, "MUL T%c, %s, T%c;\n", 'A' + tmpreg, regstr, 'A' + tmpreg);
1558         break;
1559     case WINED3DSPSM_ABS:
1560         if(ctx->target_version >= NV2) {
1561             sprintf(outregstr, "|%s%s|", regstr, swzstr);
1562             insert_line = 0;
1563         } else {
1564             shader_addline(buffer, "ABS T%c, %s;\n", 'A' + tmpreg, regstr);
1565         }
1566         break;
1567     case WINED3DSPSM_ABSNEG:
1568         if(ctx->target_version >= NV2) {
1569             sprintf(outregstr, "-|%s%s|", regstr, swzstr);
1570         } else {
1571             shader_addline(buffer, "ABS T%c, %s;\n", 'A' + tmpreg, regstr);
1572             sprintf(outregstr, "-T%c%s", 'A' + tmpreg, swzstr);
1573         }
1574         insert_line = 0;
1575         break;
1576     default:
1577         sprintf(outregstr, "%s%s", regstr, swzstr);
1578         insert_line = 0;
1579     }
1580
1581     /* Return modified or original register, with swizzle */
1582     if (insert_line)
1583         sprintf(outregstr, "T%c%s", 'A' + tmpreg, swzstr);
1584 }
1585
1586 static void pshader_hw_bem(const struct wined3d_shader_instruction *ins)
1587 {
1588     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1589     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1590     char dst_name[50];
1591     char src_name[2][50];
1592     DWORD sampler_code = dst->reg.idx;
1593
1594     shader_arb_get_dst_param(ins, dst, dst_name);
1595
1596     /* Sampling the perturbation map in Tsrc was done already, including the signedness correction if needed
1597      *
1598      * Keep in mind that src_name[1] can be "TB" and src_name[0] can be "TA" because modifiers like _x2 are valid
1599      * with bem. So delay loading the first parameter until after the perturbation calculation which needs two
1600      * temps is done.
1601      */
1602     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1603     shader_addline(buffer, "SWZ TA, bumpenvmat%d, x, z, 0, 0;\n", sampler_code);
1604     shader_addline(buffer, "DP3 TC.r, TA, %s;\n", src_name[1]);
1605     shader_addline(buffer, "SWZ TA, bumpenvmat%d, y, w, 0, 0;\n", sampler_code);
1606     shader_addline(buffer, "DP3 TC.g, TA, %s;\n", src_name[1]);
1607
1608     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
1609     shader_addline(buffer, "ADD %s, %s, TC;\n", dst_name, src_name[0]);
1610 }
1611
1612 static DWORD negate_modifiers(DWORD mod, char *extra_char)
1613 {
1614     *extra_char = ' ';
1615     switch(mod)
1616     {
1617         case WINED3DSPSM_NONE:      return WINED3DSPSM_NEG;
1618         case WINED3DSPSM_NEG:       return WINED3DSPSM_NONE;
1619         case WINED3DSPSM_BIAS:      return WINED3DSPSM_BIASNEG;
1620         case WINED3DSPSM_BIASNEG:   return WINED3DSPSM_BIAS;
1621         case WINED3DSPSM_SIGN:      return WINED3DSPSM_SIGNNEG;
1622         case WINED3DSPSM_SIGNNEG:   return WINED3DSPSM_SIGN;
1623         case WINED3DSPSM_COMP:      *extra_char = '-'; return WINED3DSPSM_COMP;
1624         case WINED3DSPSM_X2:        return WINED3DSPSM_X2NEG;
1625         case WINED3DSPSM_X2NEG:     return WINED3DSPSM_X2;
1626         case WINED3DSPSM_DZ:        *extra_char = '-'; return WINED3DSPSM_DZ;
1627         case WINED3DSPSM_DW:        *extra_char = '-'; return WINED3DSPSM_DW;
1628         case WINED3DSPSM_ABS:       return WINED3DSPSM_ABSNEG;
1629         case WINED3DSPSM_ABSNEG:    return WINED3DSPSM_ABS;
1630     }
1631     FIXME("Unknown modifier %u\n", mod);
1632     return mod;
1633 }
1634
1635 static void pshader_hw_cnd(const struct wined3d_shader_instruction *ins)
1636 {
1637     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1638     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1639     char dst_name[50];
1640     char src_name[3][50];
1641     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
1642             ins->ctx->reg_maps->shader_version.minor);
1643
1644     shader_arb_get_dst_param(ins, dst, dst_name);
1645     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1646
1647     /* The coissue flag changes the semantic of the cnd instruction in <= 1.3 shaders */
1648     if (shader_version <= WINED3D_SHADER_VERSION(1, 3) && ins->coissue)
1649     {
1650         shader_addline(buffer, "MOV%s %s, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name[1]);
1651     }
1652     else
1653     {
1654         struct wined3d_shader_src_param src0_copy = ins->src[0];
1655         char extra_neg;
1656
1657         /* src0 may have a negate srcmod set, so we can't blindly add "-" to the name */
1658         src0_copy.modifiers = negate_modifiers(src0_copy.modifiers, &extra_neg);
1659
1660         shader_arb_get_src_param(ins, &src0_copy, 0, src_name[0]);
1661         shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
1662         shader_addline(buffer, "ADD TA, %c%s, coefdiv.x;\n", extra_neg, src_name[0]);
1663         shader_addline(buffer, "CMP%s %s, TA, %s, %s;\n", shader_arb_get_modifier(ins),
1664                 dst_name, src_name[1], src_name[2]);
1665     }
1666 }
1667
1668 static void pshader_hw_cmp(const struct wined3d_shader_instruction *ins)
1669 {
1670     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1671     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1672     char dst_name[50];
1673     char src_name[3][50];
1674
1675     shader_arb_get_dst_param(ins, dst, dst_name);
1676
1677     /* Generate input register names (with modifiers) */
1678     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
1679     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1680     shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
1681
1682     shader_addline(buffer, "CMP%s %s, %s, %s, %s;\n", shader_arb_get_modifier(ins),
1683             dst_name, src_name[0], src_name[2], src_name[1]);
1684 }
1685
1686 /** Process the WINED3DSIO_DP2ADD instruction in ARB.
1687  * dst = dot2(src0, src1) + src2 */
1688 static void pshader_hw_dp2add(const struct wined3d_shader_instruction *ins)
1689 {
1690     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1691     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1692     char dst_name[50];
1693     char src_name[3][50];
1694     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
1695
1696     shader_arb_get_dst_param(ins, dst, dst_name);
1697     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
1698     shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
1699
1700     if(ctx->target_version >= NV3)
1701     {
1702         /* GL_NV_fragment_program2 has a 1:1 matching instruction */
1703         shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1704         shader_addline(buffer, "DP2A%s %s, %s, %s, %s;\n", shader_arb_get_modifier(ins),
1705                        dst_name, src_name[0], src_name[1], src_name[2]);
1706     }
1707     else if(ctx->target_version >= NV2)
1708     {
1709         /* dst.x = src2.?, src0.x, src1.x + src0.y * src1.y
1710          * dst.y = src2.?, src0.x, src1.z + src0.y * src1.w
1711          * dst.z = src2.?, src0.x, src1.x + src0.y * src1.y
1712          * dst.z = src2.?, src0.x, src1.z + src0.y * src1.w
1713          *
1714          * Make sure that src1.zw = src1.xy, then we get a classic dp2add
1715          *
1716          * .xyxy and other swizzles that we could get with this are not valid in
1717          * plain ARBfp, but luckily the NV extension grammar lifts this limitation.
1718          */
1719         struct wined3d_shader_src_param tmp_param = ins->src[1];
1720         DWORD swizzle = tmp_param.swizzle & 0xf; /* Selects .xy */
1721         tmp_param.swizzle = swizzle | (swizzle << 4); /* Creates .xyxy */
1722
1723         shader_arb_get_src_param(ins, &tmp_param, 1, src_name[1]);
1724
1725         shader_addline(buffer, "X2D%s %s, %s, %s, %s;\n", shader_arb_get_modifier(ins),
1726                        dst_name, src_name[2], src_name[0], src_name[1]);
1727     }
1728     else
1729     {
1730         shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1731         /* 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]
1732         * src_name[0] can be TA, but TA is a private temp for modifiers, so it is save to overwrite
1733         */
1734         shader_addline(buffer, "MOV TA, %s;\n", src_name[0]);
1735         shader_addline(buffer, "MOV TA.z, 0.0;\n");
1736         shader_addline(buffer, "DP3 TA, TA, %s;\n", src_name[1]);
1737         shader_addline(buffer, "ADD%s %s, TA, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name[2]);
1738     }
1739 }
1740
1741 /* Map the opcode 1-to-1 to the GL code */
1742 static void shader_hw_map2gl(const struct wined3d_shader_instruction *ins)
1743 {
1744     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1745     const char *instruction;
1746     char arguments[256], dst_str[50];
1747     unsigned int i;
1748     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1749
1750     switch (ins->handler_idx)
1751     {
1752         case WINED3DSIH_ABS: instruction = "ABS"; break;
1753         case WINED3DSIH_ADD: instruction = "ADD"; break;
1754         case WINED3DSIH_CRS: instruction = "XPD"; break;
1755         case WINED3DSIH_DP3: instruction = "DP3"; break;
1756         case WINED3DSIH_DP4: instruction = "DP4"; break;
1757         case WINED3DSIH_DST: instruction = "DST"; break;
1758         case WINED3DSIH_FRC: instruction = "FRC"; break;
1759         case WINED3DSIH_LIT: instruction = "LIT"; break;
1760         case WINED3DSIH_LRP: instruction = "LRP"; break;
1761         case WINED3DSIH_MAD: instruction = "MAD"; break;
1762         case WINED3DSIH_MAX: instruction = "MAX"; break;
1763         case WINED3DSIH_MIN: instruction = "MIN"; break;
1764         case WINED3DSIH_MOV: instruction = "MOV"; break;
1765         case WINED3DSIH_MUL: instruction = "MUL"; break;
1766         case WINED3DSIH_SGE: instruction = "SGE"; break;
1767         case WINED3DSIH_SLT: instruction = "SLT"; break;
1768         case WINED3DSIH_SUB: instruction = "SUB"; break;
1769         case WINED3DSIH_MOVA:instruction = "ARR"; break;
1770         case WINED3DSIH_DSX: instruction = "DDX"; break;
1771         default: instruction = "";
1772             FIXME("Unhandled opcode %#x\n", ins->handler_idx);
1773             break;
1774     }
1775
1776     /* Note that shader_arb_add_dst_param() adds spaces. */
1777     arguments[0] = '\0';
1778     shader_arb_get_dst_param(ins, dst, dst_str);
1779     for (i = 0; i < ins->src_count; ++i)
1780     {
1781         char operand[100];
1782         strcat(arguments, ", ");
1783         shader_arb_get_src_param(ins, &ins->src[i], i, operand);
1784         strcat(arguments, operand);
1785     }
1786     shader_addline(buffer, "%s%s %s%s;\n", instruction, shader_arb_get_modifier(ins), dst_str, arguments);
1787 }
1788
1789 static void shader_hw_nop(const struct wined3d_shader_instruction *ins) {}
1790
1791 static void shader_hw_mov(const struct wined3d_shader_instruction *ins)
1792 {
1793     const struct wined3d_shader *shader = ins->ctx->shader;
1794     const struct wined3d_shader_reg_maps *reg_maps = ins->ctx->reg_maps;
1795     BOOL pshader = shader_is_pshader_version(reg_maps->shader_version.type);
1796     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
1797     const char *zero = arb_get_helper_value(reg_maps->shader_version.type, ARB_ZERO);
1798     const char *one = arb_get_helper_value(reg_maps->shader_version.type, ARB_ONE);
1799     const char *two = arb_get_helper_value(reg_maps->shader_version.type, ARB_TWO);
1800
1801     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1802     char src0_param[256];
1803
1804     if (ins->handler_idx == WINED3DSIH_MOVA)
1805     {
1806         const struct arb_vshader_private *shader_data = shader->backend_data;
1807         char write_mask[6];
1808         const char *offset = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_VS_REL_OFFSET);
1809
1810         if(ctx->target_version >= NV2) {
1811             shader_hw_map2gl(ins);
1812             return;
1813         }
1814         shader_arb_get_src_param(ins, &ins->src[0], 0, src0_param);
1815         shader_arb_get_write_mask(ins, &ins->dst[0], write_mask);
1816
1817         /* This implements the mova formula used in GLSL. The first two instructions
1818          * prepare the sign() part. Note that it is fine to have my_sign(0.0) = 1.0
1819          * in this case:
1820          * mova A0.x, 0.0
1821          *
1822          * A0.x = arl(floor(abs(0.0) + 0.5) * 1.0) = floor(0.5) = 0.0 since arl does a floor
1823          *
1824          * The ARL is performed when A0 is used - the requested component is read from A0_SHADOW into
1825          * A0.x. We can use the overwritten component of A0_shadow as temporary storage for the sign.
1826          */
1827         shader_addline(buffer, "SGE A0_SHADOW%s, %s, %s;\n", write_mask, src0_param, zero);
1828         shader_addline(buffer, "MAD A0_SHADOW%s, A0_SHADOW, %s, -%s;\n", write_mask, two, one);
1829
1830         shader_addline(buffer, "ABS TA%s, %s;\n", write_mask, src0_param);
1831         shader_addline(buffer, "ADD TA%s, TA, rel_addr_const.x;\n", write_mask);
1832         shader_addline(buffer, "FLR TA%s, TA;\n", write_mask);
1833         if (shader_data->rel_offset)
1834         {
1835             shader_addline(buffer, "ADD TA%s, TA, %s;\n", write_mask, offset);
1836         }
1837         shader_addline(buffer, "MUL A0_SHADOW%s, TA, A0_SHADOW;\n", write_mask);
1838
1839         ((struct shader_arb_ctx_priv *)ins->ctx->backend_data)->addr_reg[0] = '\0';
1840     }
1841     else if (reg_maps->shader_version.major == 1
1842           && !shader_is_pshader_version(reg_maps->shader_version.type)
1843           && ins->dst[0].reg.type == WINED3DSPR_ADDR)
1844     {
1845         const struct arb_vshader_private *shader_data = shader->backend_data;
1846         src0_param[0] = '\0';
1847
1848         if (shader_data->rel_offset)
1849         {
1850             const char *offset = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_VS_REL_OFFSET);
1851             shader_arb_get_src_param(ins, &ins->src[0], 0, src0_param);
1852             shader_addline(buffer, "ADD TA.x, %s, %s;\n", src0_param, offset);
1853             shader_addline(buffer, "ARL A0.x, TA.x;\n");
1854         }
1855         else
1856         {
1857             /* Apple's ARB_vertex_program implementation does not accept an ARL source argument
1858              * with more than one component. Thus replicate the first source argument over all
1859              * 4 components. For example, .xyzw -> .x (or better: .xxxx), .zwxy -> .z, etc) */
1860             struct wined3d_shader_src_param tmp_src = ins->src[0];
1861             tmp_src.swizzle = (tmp_src.swizzle & 0x3) * 0x55;
1862             shader_arb_get_src_param(ins, &tmp_src, 0, src0_param);
1863             shader_addline(buffer, "ARL A0.x, %s;\n", src0_param);
1864         }
1865     }
1866     else if (ins->dst[0].reg.type == WINED3DSPR_COLOROUT && !ins->dst[0].reg.idx && pshader)
1867     {
1868         if (ctx->cur_ps_args->super.srgb_correction && shader->u.ps.color0_mov)
1869         {
1870             shader_addline(buffer, "#mov handled in srgb write code\n");
1871             return;
1872         }
1873         shader_hw_map2gl(ins);
1874     }
1875     else
1876     {
1877         shader_hw_map2gl(ins);
1878     }
1879 }
1880
1881 static void pshader_hw_texkill(const struct wined3d_shader_instruction *ins)
1882 {
1883     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1884     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1885     char reg_dest[40];
1886
1887     /* No swizzles are allowed in d3d's texkill. PS 1.x ignores the 4th component as documented,
1888      * but >= 2.0 honors it (undocumented, but tested by the d3d9 testsuite)
1889      */
1890     shader_arb_get_dst_param(ins, dst, reg_dest);
1891
1892     if (ins->ctx->reg_maps->shader_version.major >= 2)
1893     {
1894         const char *kilsrc = "TA";
1895         BOOL is_color;
1896
1897         shader_arb_get_register_name(ins, &dst->reg, reg_dest, &is_color);
1898         if(dst->write_mask == WINED3DSP_WRITEMASK_ALL)
1899         {
1900             kilsrc = reg_dest;
1901         }
1902         else
1903         {
1904             /* Sigh. KIL doesn't support swizzles/writemasks. KIL passes a writemask, but ".xy" for example
1905              * is not valid as a swizzle in ARB (needs ".xyyy"). Use SWZ to load the register properly, and set
1906              * masked out components to 0(won't kill)
1907              */
1908             char x = '0', y = '0', z = '0', w = '0';
1909             if(dst->write_mask & WINED3DSP_WRITEMASK_0) x = 'x';
1910             if(dst->write_mask & WINED3DSP_WRITEMASK_1) y = 'y';
1911             if(dst->write_mask & WINED3DSP_WRITEMASK_2) z = 'z';
1912             if(dst->write_mask & WINED3DSP_WRITEMASK_3) w = 'w';
1913             shader_addline(buffer, "SWZ TA, %s, %c, %c, %c, %c;\n", reg_dest, x, y, z, w);
1914         }
1915         shader_addline(buffer, "KIL %s;\n", kilsrc);
1916     } else {
1917         /* ARB fp doesn't like swizzles on the parameter of the KIL instruction. To mask the 4th component,
1918          * copy the register into our general purpose TMP variable, overwrite .w and pass TMP to KIL
1919          *
1920          * ps_1_3 shaders use the texcoord incarnation of the Tx register. ps_1_4 shaders can use the same,
1921          * or pass in any temporary register(in shader phase 2)
1922          */
1923         if(ins->ctx->reg_maps->shader_version.minor <= 3) {
1924             sprintf(reg_dest, "fragment.texcoord[%u]", dst->reg.idx);
1925         } else {
1926             shader_arb_get_dst_param(ins, dst, reg_dest);
1927         }
1928         shader_addline(buffer, "SWZ TA, %s, x, y, z, 1;\n", reg_dest);
1929         shader_addline(buffer, "KIL TA;\n");
1930     }
1931 }
1932
1933 static void pshader_hw_tex(const struct wined3d_shader_instruction *ins)
1934 {
1935     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
1936     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1937     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
1938             ins->ctx->reg_maps->shader_version.minor);
1939     struct wined3d_shader_src_param src;
1940
1941     char reg_dest[40];
1942     char reg_coord[40];
1943     DWORD reg_sampler_code;
1944     WORD myflags = 0;
1945     BOOL swizzle_coord = FALSE;
1946
1947     /* All versions have a destination register */
1948     shader_arb_get_dst_param(ins, dst, reg_dest);
1949
1950     /* 1.0-1.4: Use destination register number as texture code.
1951        2.0+: Use provided sampler number as texure code. */
1952     if (shader_version < WINED3D_SHADER_VERSION(2,0))
1953         reg_sampler_code = dst->reg.idx;
1954     else
1955         reg_sampler_code = ins->src[1].reg.idx;
1956
1957     /* 1.0-1.3: Use the texcoord varying.
1958        1.4+: Use provided coordinate source register. */
1959     if (shader_version < WINED3D_SHADER_VERSION(1,4))
1960         sprintf(reg_coord, "fragment.texcoord[%u]", reg_sampler_code);
1961     else {
1962         /* TEX is the only instruction that can handle DW and DZ natively */
1963         src = ins->src[0];
1964         if(src.modifiers == WINED3DSPSM_DW) src.modifiers = WINED3DSPSM_NONE;
1965         if(src.modifiers == WINED3DSPSM_DZ) src.modifiers = WINED3DSPSM_NONE;
1966         shader_arb_get_src_param(ins, &src, 0, reg_coord);
1967     }
1968
1969     /* projection flag:
1970      * 1.1, 1.2, 1.3: Use WINED3D_TSS_TEXTURETRANSFORMFLAGS
1971      * 1.4: Use WINED3DSPSM_DZ or WINED3DSPSM_DW on src[0]
1972      * 2.0+: Use WINED3DSI_TEXLD_PROJECT on the opcode
1973      */
1974     if (shader_version < WINED3D_SHADER_VERSION(1,4))
1975     {
1976         DWORD flags = 0;
1977         if (reg_sampler_code < MAX_TEXTURES)
1978             flags = priv->cur_ps_args->super.tex_transform >> reg_sampler_code * WINED3D_PSARGS_TEXTRANSFORM_SHIFT;
1979         if (flags & WINED3D_PSARGS_PROJECTED)
1980         {
1981             myflags |= TEX_PROJ;
1982             if ((flags & ~WINED3D_PSARGS_PROJECTED) == WINED3D_TTFF_COUNT3)
1983                 swizzle_coord = TRUE;
1984         }
1985     }
1986     else if (shader_version < WINED3D_SHADER_VERSION(2,0))
1987     {
1988         enum wined3d_shader_src_modifier src_mod = ins->src[0].modifiers;
1989         if (src_mod == WINED3DSPSM_DZ)
1990         {
1991             swizzle_coord = TRUE;
1992             myflags |= TEX_PROJ;
1993         } else if(src_mod == WINED3DSPSM_DW) {
1994             myflags |= TEX_PROJ;
1995         }
1996     } else {
1997         if (ins->flags & WINED3DSI_TEXLD_PROJECT) myflags |= TEX_PROJ;
1998         if (ins->flags & WINED3DSI_TEXLD_BIAS) myflags |= TEX_BIAS;
1999     }
2000
2001     if (swizzle_coord)
2002     {
2003         /* TXP cannot handle DZ natively, so move the z coordinate to .w.
2004          * reg_coord is a read-only varying register, so we need a temp reg */
2005         shader_addline(ins->ctx->buffer, "SWZ TA, %s, x, y, z, z;\n", reg_coord);
2006         strcpy(reg_coord, "TA");
2007     }
2008
2009     shader_hw_sample(ins, reg_sampler_code, reg_dest, reg_coord, myflags, NULL, NULL);
2010 }
2011
2012 static void pshader_hw_texcoord(const struct wined3d_shader_instruction *ins)
2013 {
2014     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2015     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2016     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
2017             ins->ctx->reg_maps->shader_version.minor);
2018     char dst_str[50];
2019
2020     if (shader_version < WINED3D_SHADER_VERSION(1,4))
2021     {
2022         DWORD reg = dst->reg.idx;
2023
2024         shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2025         shader_addline(buffer, "MOV_SAT %s, fragment.texcoord[%u];\n", dst_str, reg);
2026     } else {
2027         char reg_src[40];
2028
2029         shader_arb_get_src_param(ins, &ins->src[0], 0, reg_src);
2030         shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2031         shader_addline(buffer, "MOV %s, %s;\n", dst_str, reg_src);
2032    }
2033 }
2034
2035 static void pshader_hw_texreg2ar(const struct wined3d_shader_instruction *ins)
2036 {
2037      struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2038      DWORD flags = 0;
2039
2040      DWORD reg1 = ins->dst[0].reg.idx;
2041      char dst_str[50];
2042      char src_str[50];
2043
2044      /* Note that texreg2ar treats Tx as a temporary register, not as a varying */
2045      shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2046      shader_arb_get_src_param(ins, &ins->src[0], 0, src_str);
2047      /* Move .x first in case src_str is "TA" */
2048      shader_addline(buffer, "MOV TA.y, %s.x;\n", src_str);
2049      shader_addline(buffer, "MOV TA.x, %s.w;\n", src_str);
2050      if (reg1 < MAX_TEXTURES)
2051      {
2052          struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2053          flags = priv->cur_ps_args->super.tex_transform >> reg1 * WINED3D_PSARGS_TEXTRANSFORM_SHIFT;
2054      }
2055      shader_hw_sample(ins, reg1, dst_str, "TA", flags & WINED3D_PSARGS_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2056 }
2057
2058 static void pshader_hw_texreg2gb(const struct wined3d_shader_instruction *ins)
2059 {
2060      struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2061
2062      DWORD reg1 = ins->dst[0].reg.idx;
2063      char dst_str[50];
2064      char src_str[50];
2065
2066      /* Note that texreg2gb treats Tx as a temporary register, not as a varying */
2067      shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2068      shader_arb_get_src_param(ins, &ins->src[0], 0, src_str);
2069      shader_addline(buffer, "MOV TA.x, %s.y;\n", src_str);
2070      shader_addline(buffer, "MOV TA.y, %s.z;\n", src_str);
2071      shader_hw_sample(ins, reg1, dst_str, "TA", 0, NULL, NULL);
2072 }
2073
2074 static void pshader_hw_texreg2rgb(const struct wined3d_shader_instruction *ins)
2075 {
2076     DWORD reg1 = ins->dst[0].reg.idx;
2077     char dst_str[50];
2078     char src_str[50];
2079
2080     /* Note that texreg2rg treats Tx as a temporary register, not as a varying */
2081     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2082     shader_arb_get_src_param(ins, &ins->src[0], 0, src_str);
2083     shader_hw_sample(ins, reg1, dst_str, src_str, 0, NULL, NULL);
2084 }
2085
2086 static void pshader_hw_texbem(const struct wined3d_shader_instruction *ins)
2087 {
2088     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2089     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2090     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2091     char reg_coord[40], dst_reg[50], src_reg[50];
2092     DWORD reg_dest_code;
2093
2094     /* All versions have a destination register. The Tx where the texture coordinates come
2095      * from is the varying incarnation of the texture register
2096      */
2097     reg_dest_code = dst->reg.idx;
2098     shader_arb_get_dst_param(ins, &ins->dst[0], dst_reg);
2099     shader_arb_get_src_param(ins, &ins->src[0], 0, src_reg);
2100     sprintf(reg_coord, "fragment.texcoord[%u]", reg_dest_code);
2101
2102     /* Sampling the perturbation map in Tsrc was done already, including the signedness correction if needed
2103      * The Tx in which the perturbation map is stored is the tempreg incarnation of the texture register
2104      *
2105      * GL_NV_fragment_program_option could handle this in one instruction via X2D:
2106      * X2D TA.xy, fragment.texcoord, T%u, bumpenvmat%u.xzyw
2107      *
2108      * However, the NV extensions are never enabled for <= 2.0 shaders because of the performance penalty that
2109      * comes with it, and texbem is an 1.x only instruction. No 1.x instruction forces us to enable the NV
2110      * extension.
2111      */
2112     shader_addline(buffer, "SWZ TB, bumpenvmat%d, x, z, 0, 0;\n", reg_dest_code);
2113     shader_addline(buffer, "DP3 TA.x, TB, %s;\n", src_reg);
2114     shader_addline(buffer, "SWZ TB, bumpenvmat%d, y, w, 0, 0;\n", reg_dest_code);
2115     shader_addline(buffer, "DP3 TA.y, TB, %s;\n", src_reg);
2116
2117     /* with projective textures, texbem only divides the static texture coord, not the displacement,
2118      * so we can't let the GL handle this.
2119      */
2120     if ((priv->cur_ps_args->super.tex_transform >> reg_dest_code * WINED3D_PSARGS_TEXTRANSFORM_SHIFT)
2121             & WINED3D_PSARGS_PROJECTED)
2122     {
2123         shader_addline(buffer, "RCP TB.w, %s.w;\n", reg_coord);
2124         shader_addline(buffer, "MUL TB.xy, %s, TB.w;\n", reg_coord);
2125         shader_addline(buffer, "ADD TA.xy, TA, TB;\n");
2126     } else {
2127         shader_addline(buffer, "ADD TA.xy, TA, %s;\n", reg_coord);
2128     }
2129
2130     shader_hw_sample(ins, reg_dest_code, dst_reg, "TA", 0, NULL, NULL);
2131
2132     if (ins->handler_idx == WINED3DSIH_TEXBEML)
2133     {
2134         /* No src swizzles are allowed, so this is ok */
2135         shader_addline(buffer, "MAD TA, %s.z, luminance%d.x, luminance%d.y;\n",
2136                        src_reg, reg_dest_code, reg_dest_code);
2137         shader_addline(buffer, "MUL %s, %s, TA;\n", dst_reg, dst_reg);
2138     }
2139 }
2140
2141 static void pshader_hw_texm3x2pad(const struct wined3d_shader_instruction *ins)
2142 {
2143     DWORD reg = ins->dst[0].reg.idx;
2144     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2145     char src0_name[50], dst_name[50];
2146     BOOL is_color;
2147     struct wined3d_shader_register tmp_reg = ins->dst[0].reg;
2148
2149     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2150     /* The next instruction will be a texm3x2tex or texm3x2depth that writes to the uninitialized
2151      * T<reg+1> register. Use this register to store the calculated vector
2152      */
2153     tmp_reg.idx = reg + 1;
2154     shader_arb_get_register_name(ins, &tmp_reg, dst_name, &is_color);
2155     shader_addline(buffer, "DP3 %s.x, fragment.texcoord[%u], %s;\n", dst_name, reg, src0_name);
2156 }
2157
2158 static void pshader_hw_texm3x2tex(const struct wined3d_shader_instruction *ins)
2159 {
2160     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2161     DWORD flags;
2162     DWORD reg = ins->dst[0].reg.idx;
2163     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2164     char dst_str[50];
2165     char src0_name[50];
2166     char dst_reg[50];
2167     BOOL is_color;
2168
2169     /* We know that we're writing to the uninitialized T<reg> register, so use it for temporary storage */
2170     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_reg, &is_color);
2171
2172     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2173     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2174     shader_addline(buffer, "DP3 %s.y, fragment.texcoord[%u], %s;\n", dst_reg, reg, src0_name);
2175     flags = reg < MAX_TEXTURES ? priv->cur_ps_args->super.tex_transform >> reg * WINED3D_PSARGS_TEXTRANSFORM_SHIFT : 0;
2176     shader_hw_sample(ins, reg, dst_str, dst_reg, flags & WINED3D_PSARGS_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2177 }
2178
2179 static void pshader_hw_texm3x3pad(const struct wined3d_shader_instruction *ins)
2180 {
2181     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
2182     DWORD reg = ins->dst[0].reg.idx;
2183     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2184     char src0_name[50], dst_name[50];
2185     struct wined3d_shader_register tmp_reg = ins->dst[0].reg;
2186     BOOL is_color;
2187
2188     /* There are always 2 texm3x3pad instructions followed by one texm3x3[tex,vspec, ...] instruction, with
2189      * incrementing ins->dst[0].register_idx numbers. So the pad instruction already knows the final destination
2190      * register, and this register is uninitialized(otherwise the assembler complains that it is 'redeclared')
2191      */
2192     tmp_reg.idx = reg + 2 - tex_mx->current_row;
2193     shader_arb_get_register_name(ins, &tmp_reg, dst_name, &is_color);
2194
2195     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2196     shader_addline(buffer, "DP3 %s.%c, fragment.texcoord[%u], %s;\n",
2197                    dst_name, 'x' + tex_mx->current_row, reg, src0_name);
2198     tex_mx->texcoord_w[tex_mx->current_row++] = reg;
2199 }
2200
2201 static void pshader_hw_texm3x3tex(const struct wined3d_shader_instruction *ins)
2202 {
2203     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2204     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
2205     DWORD flags;
2206     DWORD reg = ins->dst[0].reg.idx;
2207     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2208     char dst_str[50];
2209     char src0_name[50], dst_name[50];
2210     BOOL is_color;
2211
2212     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2213     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2214     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_name, reg, src0_name);
2215
2216     /* Sample the texture using the calculated coordinates */
2217     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2218     flags = reg < MAX_TEXTURES ? priv->cur_ps_args->super.tex_transform >> reg * WINED3D_PSARGS_TEXTRANSFORM_SHIFT : 0;
2219     shader_hw_sample(ins, reg, dst_str, dst_name, flags & WINED3D_PSARGS_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2220     tex_mx->current_row = 0;
2221 }
2222
2223 static void pshader_hw_texm3x3vspec(const struct wined3d_shader_instruction *ins)
2224 {
2225     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2226     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
2227     DWORD flags;
2228     DWORD reg = ins->dst[0].reg.idx;
2229     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2230     char dst_str[50];
2231     char src0_name[50];
2232     char dst_reg[50];
2233     BOOL is_color;
2234
2235     /* Get the dst reg without writemask strings. We know this register is uninitialized, so we can use all
2236      * components for temporary data storage
2237      */
2238     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_reg, &is_color);
2239     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2240     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_reg, reg, src0_name);
2241
2242     /* Construct the eye-ray vector from w coordinates */
2243     shader_addline(buffer, "MOV TB.x, fragment.texcoord[%u].w;\n", tex_mx->texcoord_w[0]);
2244     shader_addline(buffer, "MOV TB.y, fragment.texcoord[%u].w;\n", tex_mx->texcoord_w[1]);
2245     shader_addline(buffer, "MOV TB.z, fragment.texcoord[%u].w;\n", reg);
2246
2247     /* Calculate reflection vector
2248      */
2249     shader_addline(buffer, "DP3 %s.w, %s, TB;\n", dst_reg, dst_reg);
2250     /* The .w is ignored when sampling, so I can use TB.w to calculate dot(N, N) */
2251     shader_addline(buffer, "DP3 TB.w, %s, %s;\n", dst_reg, dst_reg);
2252     shader_addline(buffer, "RCP TB.w, TB.w;\n");
2253     shader_addline(buffer, "MUL %s.w, %s.w, TB.w;\n", dst_reg, dst_reg);
2254     shader_addline(buffer, "MUL %s, %s.w, %s;\n", dst_reg, dst_reg, dst_reg);
2255     shader_addline(buffer, "MAD %s, coefmul.x, %s, -TB;\n", dst_reg, dst_reg);
2256
2257     /* Sample the texture using the calculated coordinates */
2258     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2259     flags = reg < MAX_TEXTURES ? priv->cur_ps_args->super.tex_transform >> reg * WINED3D_PSARGS_TEXTRANSFORM_SHIFT : 0;
2260     shader_hw_sample(ins, reg, dst_str, dst_reg, flags & WINED3D_PSARGS_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2261     tex_mx->current_row = 0;
2262 }
2263
2264 static void pshader_hw_texm3x3spec(const struct wined3d_shader_instruction *ins)
2265 {
2266     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2267     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
2268     DWORD flags;
2269     DWORD reg = ins->dst[0].reg.idx;
2270     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2271     char dst_str[50];
2272     char src0_name[50];
2273     char src1_name[50];
2274     char dst_reg[50];
2275     BOOL is_color;
2276
2277     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2278     shader_arb_get_src_param(ins, &ins->src[0], 1, src1_name);
2279     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_reg, &is_color);
2280     /* Note: dst_reg.xy is input here, generated by two texm3x3pad instructions */
2281     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_reg, reg, src0_name);
2282
2283     /* Calculate reflection vector.
2284      *
2285      *                   dot(N, E)
2286      * dst_reg.xyz = 2 * --------- * N - E
2287      *                   dot(N, N)
2288      *
2289      * Which normalizes the normal vector
2290      */
2291     shader_addline(buffer, "DP3 %s.w, %s, %s;\n", dst_reg, dst_reg, src1_name);
2292     shader_addline(buffer, "DP3 TC.w, %s, %s;\n", dst_reg, dst_reg);
2293     shader_addline(buffer, "RCP TC.w, TC.w;\n");
2294     shader_addline(buffer, "MUL %s.w, %s.w, TC.w;\n", dst_reg, dst_reg);
2295     shader_addline(buffer, "MUL %s, %s.w, %s;\n", dst_reg, dst_reg, dst_reg);
2296     shader_addline(buffer, "MAD %s, coefmul.x, %s, -%s;\n", dst_reg, dst_reg, src1_name);
2297
2298     /* Sample the texture using the calculated coordinates */
2299     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2300     flags = reg < MAX_TEXTURES ? priv->cur_ps_args->super.tex_transform >> reg * WINED3D_PSARGS_TEXTRANSFORM_SHIFT : 0;
2301     shader_hw_sample(ins, reg, dst_str, dst_reg, flags & WINED3D_PSARGS_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2302     tex_mx->current_row = 0;
2303 }
2304
2305 static void pshader_hw_texdepth(const struct wined3d_shader_instruction *ins)
2306 {
2307     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2308     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2309     char dst_name[50];
2310     const char *zero = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ZERO);
2311     const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
2312
2313     /* texdepth has an implicit destination, the fragment depth value. It's only parameter,
2314      * which is essentially an input, is the destination register because it is the first
2315      * parameter. According to the msdn, this must be register r5, but let's keep it more flexible
2316      * here(writemasks/swizzles are not valid on texdepth)
2317      */
2318     shader_arb_get_dst_param(ins, dst, dst_name);
2319
2320     /* According to the msdn, the source register(must be r5) is unusable after
2321      * the texdepth instruction, so we're free to modify it
2322      */
2323     shader_addline(buffer, "MIN %s.y, %s.y, %s;\n", dst_name, dst_name, one);
2324
2325     /* How to deal with the special case dst_name.g == 0? if r != 0, then
2326      * the r * (1 / 0) will give infinity, which is clamped to 1.0, the correct
2327      * result. But if r = 0.0, then 0 * inf = 0, which is incorrect.
2328      */
2329     shader_addline(buffer, "RCP %s.y, %s.y;\n", dst_name, dst_name);
2330     shader_addline(buffer, "MUL TA.x, %s.x, %s.y;\n", dst_name, dst_name);
2331     shader_addline(buffer, "MIN TA.x, TA.x, %s;\n", one);
2332     shader_addline(buffer, "MAX result.depth, TA.x, %s;\n", zero);
2333 }
2334
2335 /** Process the WINED3DSIO_TEXDP3TEX instruction in ARB:
2336  * Take a 3-component dot product of the TexCoord[dstreg] and src,
2337  * then perform a 1D texture lookup from stage dstregnum, place into dst. */
2338 static void pshader_hw_texdp3tex(const struct wined3d_shader_instruction *ins)
2339 {
2340     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2341     DWORD sampler_idx = ins->dst[0].reg.idx;
2342     char src0[50];
2343     char dst_str[50];
2344
2345     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2346     shader_addline(buffer, "MOV TB, 0.0;\n");
2347     shader_addline(buffer, "DP3 TB.x, fragment.texcoord[%u], %s;\n", sampler_idx, src0);
2348
2349     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2350     shader_hw_sample(ins, sampler_idx, dst_str, "TB", 0 /* Only one coord, can't be projected */, NULL, NULL);
2351 }
2352
2353 /** Process the WINED3DSIO_TEXDP3 instruction in ARB:
2354  * Take a 3-component dot product of the TexCoord[dstreg] and src. */
2355 static void pshader_hw_texdp3(const struct wined3d_shader_instruction *ins)
2356 {
2357     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2358     char src0[50];
2359     char dst_str[50];
2360     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2361
2362     /* Handle output register */
2363     shader_arb_get_dst_param(ins, dst, dst_str);
2364     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2365     shader_addline(buffer, "DP3 %s, fragment.texcoord[%u], %s;\n", dst_str, dst->reg.idx, src0);
2366 }
2367
2368 /** Process the WINED3DSIO_TEXM3X3 instruction in ARB
2369  * Perform the 3rd row of a 3x3 matrix multiply */
2370 static void pshader_hw_texm3x3(const struct wined3d_shader_instruction *ins)
2371 {
2372     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2373     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2374     char dst_str[50], dst_name[50];
2375     char src0[50];
2376     BOOL is_color;
2377
2378     shader_arb_get_dst_param(ins, dst, dst_str);
2379     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2380     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2381     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_name, dst->reg.idx, src0);
2382     shader_addline(buffer, "MOV %s, %s;\n", dst_str, dst_name);
2383 }
2384
2385 /** Process the WINED3DSIO_TEXM3X2DEPTH instruction in ARB:
2386  * Last row of a 3x2 matrix multiply, use the result to calculate the depth:
2387  * Calculate tmp0.y = TexCoord[dstreg] . src.xyz;  (tmp0.x has already been calculated)
2388  * depth = (tmp0.y == 0.0) ? 1.0 : tmp0.x / tmp0.y
2389  */
2390 static void pshader_hw_texm3x2depth(const struct wined3d_shader_instruction *ins)
2391 {
2392     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2393     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2394     char src0[50], dst_name[50];
2395     BOOL is_color;
2396     const char *zero = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ZERO);
2397     const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
2398
2399     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2400     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2401     shader_addline(buffer, "DP3 %s.y, fragment.texcoord[%u], %s;\n", dst_name, dst->reg.idx, src0);
2402
2403     /* How to deal with the special case dst_name.g == 0? if r != 0, then
2404      * the r * (1 / 0) will give infinity, which is clamped to 1.0, the correct
2405      * result. But if r = 0.0, then 0 * inf = 0, which is incorrect.
2406      */
2407     shader_addline(buffer, "RCP %s.y, %s.y;\n", dst_name, dst_name);
2408     shader_addline(buffer, "MUL %s.x, %s.x, %s.y;\n", dst_name, dst_name, dst_name);
2409     shader_addline(buffer, "MIN %s.x, %s.x, %s;\n", dst_name, dst_name, one);
2410     shader_addline(buffer, "MAX result.depth, %s.x, %s;\n", dst_name, zero);
2411 }
2412
2413 /** Handles transforming all WINED3DSIO_M?x? opcodes for
2414     Vertex/Pixel shaders to ARB_vertex_program codes */
2415 static void shader_hw_mnxn(const struct wined3d_shader_instruction *ins)
2416 {
2417     int i;
2418     int nComponents = 0;
2419     struct wined3d_shader_dst_param tmp_dst = {{0}};
2420     struct wined3d_shader_src_param tmp_src[2] = {{{0}}};
2421     struct wined3d_shader_instruction tmp_ins;
2422
2423     memset(&tmp_ins, 0, sizeof(tmp_ins));
2424
2425     /* Set constants for the temporary argument */
2426     tmp_ins.ctx = ins->ctx;
2427     tmp_ins.dst_count = 1;
2428     tmp_ins.dst = &tmp_dst;
2429     tmp_ins.src_count = 2;
2430     tmp_ins.src = tmp_src;
2431
2432     switch(ins->handler_idx)
2433     {
2434         case WINED3DSIH_M4x4:
2435             nComponents = 4;
2436             tmp_ins.handler_idx = WINED3DSIH_DP4;
2437             break;
2438         case WINED3DSIH_M4x3:
2439             nComponents = 3;
2440             tmp_ins.handler_idx = WINED3DSIH_DP4;
2441             break;
2442         case WINED3DSIH_M3x4:
2443             nComponents = 4;
2444             tmp_ins.handler_idx = WINED3DSIH_DP3;
2445             break;
2446         case WINED3DSIH_M3x3:
2447             nComponents = 3;
2448             tmp_ins.handler_idx = WINED3DSIH_DP3;
2449             break;
2450         case WINED3DSIH_M3x2:
2451             nComponents = 2;
2452             tmp_ins.handler_idx = WINED3DSIH_DP3;
2453             break;
2454         default:
2455             FIXME("Unhandled opcode %#x\n", ins->handler_idx);
2456             break;
2457     }
2458
2459     tmp_dst = ins->dst[0];
2460     tmp_src[0] = ins->src[0];
2461     tmp_src[1] = ins->src[1];
2462     for (i = 0; i < nComponents; i++) {
2463         tmp_dst.write_mask = WINED3DSP_WRITEMASK_0 << i;
2464         shader_hw_map2gl(&tmp_ins);
2465         ++tmp_src[1].reg.idx;
2466     }
2467 }
2468
2469 static void shader_hw_rcp(const struct wined3d_shader_instruction *ins)
2470 {
2471     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2472
2473     char dst[50];
2474     char src[50];
2475
2476     shader_arb_get_dst_param(ins, &ins->dst[0], dst); /* Destination */
2477     shader_arb_get_src_param(ins, &ins->src[0], 0, src);
2478     if (ins->src[0].swizzle == WINED3DSP_NOSWIZZLE)
2479     {
2480         /* Dx sdk says .x is used if no swizzle is given, but our test shows that
2481          * .w is used
2482          */
2483         strcat(src, ".w");
2484     }
2485
2486     shader_addline(buffer, "RCP%s %s, %s;\n", shader_arb_get_modifier(ins), dst, src);
2487 }
2488
2489 static void shader_hw_scalar_op(const struct wined3d_shader_instruction *ins)
2490 {
2491     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2492     const char *instruction;
2493
2494     char dst[50];
2495     char src[50];
2496
2497     switch(ins->handler_idx)
2498     {
2499         case WINED3DSIH_RSQ:  instruction = "RSQ"; break;
2500         case WINED3DSIH_RCP:  instruction = "RCP"; break;
2501         case WINED3DSIH_EXP:  instruction = "EX2"; break;
2502         case WINED3DSIH_EXPP: instruction = "EXP"; break;
2503         default: instruction = "";
2504             FIXME("Unhandled opcode %#x\n", ins->handler_idx);
2505             break;
2506     }
2507
2508     shader_arb_get_dst_param(ins, &ins->dst[0], dst); /* Destination */
2509     shader_arb_get_src_param(ins, &ins->src[0], 0, src);
2510     if (ins->src[0].swizzle == WINED3DSP_NOSWIZZLE)
2511     {
2512         /* Dx sdk says .x is used if no swizzle is given, but our test shows that
2513          * .w is used
2514          */
2515         strcat(src, ".w");
2516     }
2517
2518     shader_addline(buffer, "%s%s %s, %s;\n", instruction, shader_arb_get_modifier(ins), dst, src);
2519 }
2520
2521 static void shader_hw_nrm(const struct wined3d_shader_instruction *ins)
2522 {
2523     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2524     char dst_name[50];
2525     char src_name[50];
2526     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2527     BOOL pshader = shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type);
2528     const char *zero = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ZERO);
2529
2530     shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2531     shader_arb_get_src_param(ins, &ins->src[0], 1 /* Use TB */, src_name);
2532
2533     /* In D3D, NRM of a vector with length zero returns zero. Catch this situation, as
2534      * otherwise NRM or RSQ would return NaN */
2535     if(pshader && priv->target_version >= NV3)
2536     {
2537         /* GL_NV_fragment_program2's NRM needs protection against length zero vectors too
2538          *
2539          * TODO: Find out if DP3+NRM+MOV is really faster than DP3+RSQ+MUL
2540          */
2541         shader_addline(buffer, "DP3C TA, %s, %s;\n", src_name, src_name);
2542         shader_addline(buffer, "NRM%s %s, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name);
2543         shader_addline(buffer, "MOV %s (EQ), %s;\n", dst_name, zero);
2544     }
2545     else if(priv->target_version >= NV2)
2546     {
2547         shader_addline(buffer, "DP3C TA.x, %s, %s;\n", src_name, src_name);
2548         shader_addline(buffer, "RSQ TA.x (NE), TA.x;\n");
2549         shader_addline(buffer, "MUL%s %s, %s, TA.x;\n", shader_arb_get_modifier(ins), dst_name,
2550                        src_name);
2551     }
2552     else
2553     {
2554         const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
2555
2556         shader_addline(buffer, "DP3 TA.x, %s, %s;\n", src_name, src_name);
2557         /* Pass any non-zero value to RSQ if the input vector has a length of zero. The
2558          * RSQ result doesn't matter, as long as multiplying it by 0 returns 0.
2559          */
2560         shader_addline(buffer, "SGE TA.y, -TA.x, %s;\n", zero);
2561         shader_addline(buffer, "MAD TA.x, %s, TA.y, TA.x;\n", one);
2562
2563         shader_addline(buffer, "RSQ TA.x, TA.x;\n");
2564         /* dst.w = src[0].w * 1 / (src.x^2 + src.y^2 + src.z^2)^(1/2) according to msdn*/
2565         shader_addline(buffer, "MUL%s %s, %s, TA.x;\n", shader_arb_get_modifier(ins), dst_name,
2566                     src_name);
2567     }
2568 }
2569
2570 static void shader_hw_lrp(const struct wined3d_shader_instruction *ins)
2571 {
2572     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2573     char dst_name[50];
2574     char src_name[3][50];
2575
2576     /* ARB_fragment_program has a convenient LRP instruction */
2577     if(shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type)) {
2578         shader_hw_map2gl(ins);
2579         return;
2580     }
2581
2582     shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2583     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
2584     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
2585     shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
2586
2587     shader_addline(buffer, "SUB TA, %s, %s;\n", src_name[1], src_name[2]);
2588     shader_addline(buffer, "MAD%s %s, %s, TA, %s;\n", shader_arb_get_modifier(ins),
2589                    dst_name, src_name[0], src_name[2]);
2590 }
2591
2592 static void shader_hw_sincos(const struct wined3d_shader_instruction *ins)
2593 {
2594     /* This instruction exists in ARB, but the d3d instruction takes two extra parameters which
2595      * must contain fixed constants. So we need a separate function to filter those constants and
2596      * can't use map2gl
2597      */
2598     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2599     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2600     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2601     char dst_name[50];
2602     char src_name0[50], src_name1[50], src_name2[50];
2603     BOOL is_color;
2604
2605     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name0);
2606     if(shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type)) {
2607         shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2608         /* No modifiers are supported on SCS */
2609         shader_addline(buffer, "SCS %s, %s;\n", dst_name, src_name0);
2610
2611         if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE)
2612         {
2613             shader_arb_get_register_name(ins, &dst->reg, src_name0, &is_color);
2614             shader_addline(buffer, "MOV_SAT %s, %s;\n", dst_name, src_name0);
2615         }
2616     } else if(priv->target_version >= NV2) {
2617         shader_arb_get_register_name(ins, &dst->reg, dst_name, &is_color);
2618
2619         /* Sincos writemask must be .x, .y or .xy */
2620         if(dst->write_mask & WINED3DSP_WRITEMASK_0)
2621             shader_addline(buffer, "COS%s %s.x, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name0);
2622         if(dst->write_mask & WINED3DSP_WRITEMASK_1)
2623             shader_addline(buffer, "SIN%s %s.y, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name0);
2624     } else {
2625         /* Approximate sine and cosine with a taylor series, as per math textbook. The application passes 8
2626          * helper constants(D3DSINCOSCONST1 and D3DSINCOSCONST2) in src1 and src2.
2627          *
2628          * sin(x) = x - x^3/3! + x^5/5! - x^7/7! + ...
2629          * cos(x) = 1 - x^2/2! + x^4/4! - x^6/6! + ...
2630          *
2631          * The constants we get are:
2632          *
2633          *  +1   +1,     -1     -1     +1      +1      -1       -1
2634          *      ---- ,  ---- , ---- , ----- , ----- , ----- , ------
2635          *      1!*2    2!*4   3!*8   4!*16   5!*32   6!*64   7!*128
2636          *
2637          * If used with x^2, x^3, x^4 etc they calculate sin(x/2) and cos(x/2):
2638          *
2639          * (x/2)^2 = x^2 / 4
2640          * (x/2)^3 = x^3 / 8
2641          * (x/2)^4 = x^4 / 16
2642          * (x/2)^5 = x^5 / 32
2643          * etc
2644          *
2645          * To get the final result:
2646          * sin(x) = 2 * sin(x/2) * cos(x/2)
2647          * cos(x) = cos(x/2)^2 - sin(x/2)^2
2648          * (from sin(x+y) and cos(x+y) rules)
2649          *
2650          * As per MSDN, dst.z is undefined after the operation, and so is
2651          * dst.x and dst.y if they're masked out by the writemask. Ie
2652          * sincos dst.y, src1, c0, c1
2653          * returns the sine in dst.y. dst.x and dst.z are undefined, dst.w is not touched. The assembler
2654          * vsa.exe also stops with an error if the dest register is the same register as the source
2655          * register. This means we can use dest.xyz as temporary storage. The assembler vsa.exe output also
2656          * indicates that sincos consumes 8 instruction slots in vs_2_0(and, strangely, in vs_3_0).
2657          */
2658         shader_arb_get_src_param(ins, &ins->src[1], 1, src_name1);
2659         shader_arb_get_src_param(ins, &ins->src[2], 2, src_name2);
2660         shader_arb_get_register_name(ins, &dst->reg, dst_name, &is_color);
2661
2662         shader_addline(buffer, "MUL %s.x, %s, %s;\n", dst_name, src_name0, src_name0);  /* x ^ 2 */
2663         shader_addline(buffer, "MUL TA.y, %s.x, %s;\n", dst_name, src_name0);           /* x ^ 3 */
2664         shader_addline(buffer, "MUL %s.y, TA.y, %s;\n", dst_name, src_name0);           /* x ^ 4 */
2665         shader_addline(buffer, "MUL TA.z, %s.y, %s;\n", dst_name, src_name0);           /* x ^ 5 */
2666         shader_addline(buffer, "MUL %s.z, TA.z, %s;\n", dst_name, src_name0);           /* x ^ 6 */
2667         shader_addline(buffer, "MUL TA.w, %s.z, %s;\n", dst_name, src_name0);           /* x ^ 7 */
2668
2669         /* sin(x/2)
2670          *
2671          * Unfortunately we don't get the constants in a DP4-capable form. Is there a way to
2672          * properly merge that with MULs in the code above?
2673          * The swizzles .yz and xw however fit into the .yzxw swizzle added to ps_2_0. Maybe
2674          * we can merge the sine and cosine MAD rows to calculate them together.
2675          */
2676         shader_addline(buffer, "MUL TA.x, %s, %s.w;\n", src_name0, src_name2); /* x^1, +1/(1!*2) */
2677         shader_addline(buffer, "MAD TA.x, TA.y, %s.x, TA.x;\n", src_name2); /* -1/(3!*8) */
2678         shader_addline(buffer, "MAD TA.x, TA.z, %s.w, TA.x;\n", src_name1); /* +1/(5!*32) */
2679         shader_addline(buffer, "MAD TA.x, TA.w, %s.x, TA.x;\n", src_name1); /* -1/(7!*128) */
2680
2681         /* cos(x/2) */
2682         shader_addline(buffer, "MAD TA.y, %s.x, %s.y, %s.z;\n", dst_name, src_name2, src_name2); /* -1/(2!*4), +1.0 */
2683         shader_addline(buffer, "MAD TA.y, %s.y, %s.z, TA.y;\n", dst_name, src_name1); /* +1/(4!*16) */
2684         shader_addline(buffer, "MAD TA.y, %s.z, %s.y, TA.y;\n", dst_name, src_name1); /* -1/(6!*64) */
2685
2686         if(dst->write_mask & WINED3DSP_WRITEMASK_0) {
2687             /* cos x */
2688             shader_addline(buffer, "MUL TA.z, TA.y, TA.y;\n");
2689             shader_addline(buffer, "MAD %s.x, -TA.x, TA.x, TA.z;\n", dst_name);
2690         }
2691         if(dst->write_mask & WINED3DSP_WRITEMASK_1) {
2692             /* sin x */
2693             shader_addline(buffer, "MUL %s.y, TA.x, TA.y;\n", dst_name);
2694             shader_addline(buffer, "ADD %s.y, %s.y, %s.y;\n", dst_name, dst_name, dst_name);
2695         }
2696     }
2697 }
2698
2699 static void shader_hw_sgn(const struct wined3d_shader_instruction *ins)
2700 {
2701     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2702     char dst_name[50];
2703     char src_name[50];
2704     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
2705
2706     shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2707     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name);
2708
2709     /* SGN is only valid in vertex shaders */
2710     if(ctx->target_version >= NV2) {
2711         shader_addline(buffer, "SSG%s %s, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name);
2712         return;
2713     }
2714
2715     /* If SRC > 0.0, -SRC < SRC = TRUE, otherwise false.
2716      * if SRC < 0.0,  SRC < -SRC = TRUE. If neither is true, src = 0.0
2717      */
2718     if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE) {
2719         shader_addline(buffer, "SLT %s, -%s, %s;\n", dst_name, src_name, src_name);
2720     } else {
2721         /* src contains TA? Write to the dest first. This won't overwrite our destination.
2722          * Then use TA, and calculate the final result
2723          *
2724          * Not reading from TA? Store the first result in TA to avoid overwriting the
2725          * destination if src reg = dst reg
2726          */
2727         if(strstr(src_name, "TA"))
2728         {
2729             shader_addline(buffer, "SLT %s,  %s, -%s;\n", dst_name, src_name, src_name);
2730             shader_addline(buffer, "SLT TA, -%s, %s;\n", src_name, src_name);
2731             shader_addline(buffer, "ADD %s, %s, -TA;\n", dst_name, dst_name);
2732         }
2733         else
2734         {
2735             shader_addline(buffer, "SLT TA, -%s, %s;\n", src_name, src_name);
2736             shader_addline(buffer, "SLT %s,  %s, -%s;\n", dst_name, src_name, src_name);
2737             shader_addline(buffer, "ADD %s, TA, -%s;\n", dst_name, dst_name);
2738         }
2739     }
2740 }
2741
2742 static void shader_hw_dsy(const struct wined3d_shader_instruction *ins)
2743 {
2744     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2745     char src[50];
2746     char dst[50];
2747     char dst_name[50];
2748     BOOL is_color;
2749
2750     shader_arb_get_dst_param(ins, &ins->dst[0], dst);
2751     shader_arb_get_src_param(ins, &ins->src[0], 0, src);
2752     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2753
2754     shader_addline(buffer, "DDY %s, %s;\n", dst, src);
2755     shader_addline(buffer, "MUL%s %s, %s, ycorrection.y;\n", shader_arb_get_modifier(ins), dst, dst_name);
2756 }
2757
2758 static DWORD abs_modifier(DWORD mod, BOOL *need_abs)
2759 {
2760     *need_abs = FALSE;
2761
2762     switch(mod)
2763     {
2764         case WINED3DSPSM_NONE:      return WINED3DSPSM_ABS;
2765         case WINED3DSPSM_NEG:       return WINED3DSPSM_ABS;
2766         case WINED3DSPSM_BIAS:      *need_abs = TRUE; return WINED3DSPSM_BIAS;
2767         case WINED3DSPSM_BIASNEG:   *need_abs = TRUE; return WINED3DSPSM_BIASNEG;
2768         case WINED3DSPSM_SIGN:      *need_abs = TRUE; return WINED3DSPSM_SIGN;
2769         case WINED3DSPSM_SIGNNEG:   *need_abs = TRUE; return WINED3DSPSM_SIGNNEG;
2770         case WINED3DSPSM_COMP:      *need_abs = TRUE; return WINED3DSPSM_COMP;
2771         case WINED3DSPSM_X2:        *need_abs = TRUE; return WINED3DSPSM_X2;
2772         case WINED3DSPSM_X2NEG:     *need_abs = TRUE; return WINED3DSPSM_X2NEG;
2773         case WINED3DSPSM_DZ:        *need_abs = TRUE; return WINED3DSPSM_DZ;
2774         case WINED3DSPSM_DW:        *need_abs = TRUE; return WINED3DSPSM_DW;
2775         case WINED3DSPSM_ABS:       return WINED3DSPSM_ABS;
2776         case WINED3DSPSM_ABSNEG:    return WINED3DSPSM_ABS;
2777     }
2778     FIXME("Unknown modifier %u\n", mod);
2779     return mod;
2780 }
2781
2782 static void shader_hw_log(const struct wined3d_shader_instruction *ins)
2783 {
2784     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2785     char src0[50], dst[50];
2786     struct wined3d_shader_src_param src0_copy = ins->src[0];
2787     BOOL need_abs = FALSE;
2788     const char *instr;
2789
2790     switch(ins->handler_idx)
2791     {
2792         case WINED3DSIH_LOG:  instr = "LG2"; break;
2793         case WINED3DSIH_LOGP: instr = "LOG"; break;
2794         default:
2795             ERR("Unexpected instruction %d\n", ins->handler_idx);
2796             return;
2797     }
2798
2799     /* LOG and LOGP operate on the absolute value of the input */
2800     src0_copy.modifiers = abs_modifier(src0_copy.modifiers, &need_abs);
2801
2802     shader_arb_get_dst_param(ins, &ins->dst[0], dst);
2803     shader_arb_get_src_param(ins, &src0_copy, 0, src0);
2804
2805     if(need_abs)
2806     {
2807         shader_addline(buffer, "ABS TA, %s;\n", src0);
2808         shader_addline(buffer, "%s%s %s, TA;\n", instr, shader_arb_get_modifier(ins), dst);
2809     }
2810     else
2811     {
2812         shader_addline(buffer, "%s%s %s, %s;\n", instr, shader_arb_get_modifier(ins), dst, src0);
2813     }
2814 }
2815
2816 static void shader_hw_pow(const struct wined3d_shader_instruction *ins)
2817 {
2818     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2819     char src0[50], src1[50], dst[50];
2820     struct wined3d_shader_src_param src0_copy = ins->src[0];
2821     BOOL need_abs = FALSE;
2822     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2823     const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
2824
2825     /* POW operates on the absolute value of the input */
2826     src0_copy.modifiers = abs_modifier(src0_copy.modifiers, &need_abs);
2827
2828     shader_arb_get_dst_param(ins, &ins->dst[0], dst);
2829     shader_arb_get_src_param(ins, &src0_copy, 0, src0);
2830     shader_arb_get_src_param(ins, &ins->src[1], 1, src1);
2831
2832     if (need_abs)
2833         shader_addline(buffer, "ABS TA.x, %s;\n", src0);
2834     else
2835         shader_addline(buffer, "MOV TA.x, %s;\n", src0);
2836
2837     if (priv->target_version >= NV2)
2838     {
2839         shader_addline(buffer, "MOVC TA.y, %s;\n", src1);
2840         shader_addline(buffer, "POW%s %s, TA.x, TA.y;\n", shader_arb_get_modifier(ins), dst);
2841         shader_addline(buffer, "MOV %s (EQ.y), %s;\n", dst, one);
2842     }
2843     else
2844     {
2845         const char *zero = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ZERO);
2846         const char *flt_eps = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_EPS);
2847
2848         shader_addline(buffer, "ABS TA.y, %s;\n", src1);
2849         shader_addline(buffer, "SGE TA.y, -TA.y, %s;\n", zero);
2850         /* Possibly add flt_eps to avoid getting float special values */
2851         shader_addline(buffer, "MAD TA.z, TA.y, %s, %s;\n", flt_eps, src1);
2852         shader_addline(buffer, "POW%s TA.x, TA.x, TA.z;\n", shader_arb_get_modifier(ins));
2853         shader_addline(buffer, "MAD TA.x, -TA.x, TA.y, TA.x;\n");
2854         shader_addline(buffer, "MAD %s, TA.y, %s, TA.x;\n", dst, one);
2855     }
2856 }
2857
2858 static void shader_hw_loop(const struct wined3d_shader_instruction *ins)
2859 {
2860     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2861     char src_name[50];
2862     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2863
2864     /* src0 is aL */
2865     shader_arb_get_src_param(ins, &ins->src[1], 0, src_name);
2866
2867     if(vshader)
2868     {
2869         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2870         struct list *e = list_head(&priv->control_frames);
2871         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2872
2873         if(priv->loop_depth > 1) shader_addline(buffer, "PUSHA aL;\n");
2874         /* The constant loader makes sure to load -1 into iX.w */
2875         shader_addline(buffer, "ARLC aL, %s.xywz;\n", src_name);
2876         shader_addline(buffer, "BRA loop_%u_end (LE.x);\n", control_frame->no.loop);
2877         shader_addline(buffer, "loop_%u_start:\n", control_frame->no.loop);
2878     }
2879     else
2880     {
2881         shader_addline(buffer, "LOOP %s;\n", src_name);
2882     }
2883 }
2884
2885 static void shader_hw_rep(const struct wined3d_shader_instruction *ins)
2886 {
2887     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2888     char src_name[50];
2889     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2890
2891     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name);
2892
2893     /* The constant loader makes sure to load -1 into iX.w */
2894     if(vshader)
2895     {
2896         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2897         struct list *e = list_head(&priv->control_frames);
2898         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2899
2900         if(priv->loop_depth > 1) shader_addline(buffer, "PUSHA aL;\n");
2901
2902         shader_addline(buffer, "ARLC aL, %s.xywz;\n", src_name);
2903         shader_addline(buffer, "BRA loop_%u_end (LE.x);\n", control_frame->no.loop);
2904         shader_addline(buffer, "loop_%u_start:\n", control_frame->no.loop);
2905     }
2906     else
2907     {
2908         shader_addline(buffer, "REP %s;\n", src_name);
2909     }
2910 }
2911
2912 static void shader_hw_endloop(const struct wined3d_shader_instruction *ins)
2913 {
2914     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2915     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2916
2917     if(vshader)
2918     {
2919         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2920         struct list *e = list_head(&priv->control_frames);
2921         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2922
2923         shader_addline(buffer, "ARAC aL.xy, aL;\n");
2924         shader_addline(buffer, "BRA loop_%u_start (GT.x);\n", control_frame->no.loop);
2925         shader_addline(buffer, "loop_%u_end:\n", control_frame->no.loop);
2926
2927         if(priv->loop_depth > 1) shader_addline(buffer, "POPA aL;\n");
2928     }
2929     else
2930     {
2931         shader_addline(buffer, "ENDLOOP;\n");
2932     }
2933 }
2934
2935 static void shader_hw_endrep(const struct wined3d_shader_instruction *ins)
2936 {
2937     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2938     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2939
2940     if(vshader)
2941     {
2942         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2943         struct list *e = list_head(&priv->control_frames);
2944         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2945
2946         shader_addline(buffer, "ARAC aL.xy, aL;\n");
2947         shader_addline(buffer, "BRA loop_%u_start (GT.x);\n", control_frame->no.loop);
2948         shader_addline(buffer, "loop_%u_end:\n", control_frame->no.loop);
2949
2950         if(priv->loop_depth > 1) shader_addline(buffer, "POPA aL;\n");
2951     }
2952     else
2953     {
2954         shader_addline(buffer, "ENDREP;\n");
2955     }
2956 }
2957
2958 static const struct control_frame *find_last_loop(const struct shader_arb_ctx_priv *priv)
2959 {
2960     struct control_frame *control_frame;
2961
2962     LIST_FOR_EACH_ENTRY(control_frame, &priv->control_frames, struct control_frame, entry)
2963     {
2964         if(control_frame->type == LOOP || control_frame->type == REP) return control_frame;
2965     }
2966     ERR("Could not find loop for break\n");
2967     return NULL;
2968 }
2969
2970 static void shader_hw_break(const struct wined3d_shader_instruction *ins)
2971 {
2972     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2973     const struct control_frame *control_frame = find_last_loop(ins->ctx->backend_data);
2974     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2975
2976     if(vshader)
2977     {
2978         shader_addline(buffer, "BRA loop_%u_end;\n", control_frame->no.loop);
2979     }
2980     else
2981     {
2982         shader_addline(buffer, "BRK;\n");
2983     }
2984 }
2985
2986 static const char *get_compare(enum wined3d_shader_rel_op op)
2987 {
2988     switch (op)
2989     {
2990         case WINED3D_SHADER_REL_OP_GT: return "GT";
2991         case WINED3D_SHADER_REL_OP_EQ: return "EQ";
2992         case WINED3D_SHADER_REL_OP_GE: return "GE";
2993         case WINED3D_SHADER_REL_OP_LT: return "LT";
2994         case WINED3D_SHADER_REL_OP_NE: return "NE";
2995         case WINED3D_SHADER_REL_OP_LE: return "LE";
2996         default:
2997             FIXME("Unrecognized operator %#x.\n", op);
2998             return "(\?\?)";
2999     }
3000 }
3001
3002 static enum wined3d_shader_rel_op invert_compare(enum wined3d_shader_rel_op op)
3003 {
3004     switch (op)
3005     {
3006         case WINED3D_SHADER_REL_OP_GT: return WINED3D_SHADER_REL_OP_LE;
3007         case WINED3D_SHADER_REL_OP_EQ: return WINED3D_SHADER_REL_OP_NE;
3008         case WINED3D_SHADER_REL_OP_GE: return WINED3D_SHADER_REL_OP_LT;
3009         case WINED3D_SHADER_REL_OP_LT: return WINED3D_SHADER_REL_OP_GE;
3010         case WINED3D_SHADER_REL_OP_NE: return WINED3D_SHADER_REL_OP_EQ;
3011         case WINED3D_SHADER_REL_OP_LE: return WINED3D_SHADER_REL_OP_GT;
3012         default:
3013             FIXME("Unrecognized operator %#x.\n", op);
3014             return -1;
3015     }
3016 }
3017
3018 static void shader_hw_breakc(const struct wined3d_shader_instruction *ins)
3019 {
3020     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3021     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
3022     const struct control_frame *control_frame = find_last_loop(ins->ctx->backend_data);
3023     char src_name0[50];
3024     char src_name1[50];
3025     const char *comp = get_compare(ins->flags);
3026
3027     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name0);
3028     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name1);
3029
3030     if(vshader)
3031     {
3032         /* SUBC CC, src0, src1" works only in pixel shaders, so use TA to throw
3033          * away the subtraction result
3034          */
3035         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
3036         shader_addline(buffer, "BRA loop_%u_end (%s.x);\n", control_frame->no.loop, comp);
3037     }
3038     else
3039     {
3040         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
3041         shader_addline(buffer, "BRK (%s.x);\n", comp);
3042     }
3043 }
3044
3045 static void shader_hw_ifc(const struct wined3d_shader_instruction *ins)
3046 {
3047     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3048     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3049     struct list *e = list_head(&priv->control_frames);
3050     struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
3051     const char *comp;
3052     char src_name0[50];
3053     char src_name1[50];
3054     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
3055
3056     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name0);
3057     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name1);
3058
3059     if(vshader)
3060     {
3061         /* Invert the flag. We jump to the else label if the condition is NOT true */
3062         comp = get_compare(invert_compare(ins->flags));
3063         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
3064         shader_addline(buffer, "BRA ifc_%u_else (%s.x);\n", control_frame->no.ifc, comp);
3065     }
3066     else
3067     {
3068         comp = get_compare(ins->flags);
3069         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
3070         shader_addline(buffer, "IF %s.x;\n", comp);
3071     }
3072 }
3073
3074 static void shader_hw_else(const struct wined3d_shader_instruction *ins)
3075 {
3076     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3077     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3078     struct list *e = list_head(&priv->control_frames);
3079     struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
3080     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
3081
3082     if(vshader)
3083     {
3084         shader_addline(buffer, "BRA ifc_%u_endif;\n", control_frame->no.ifc);
3085         shader_addline(buffer, "ifc_%u_else:\n", control_frame->no.ifc);
3086         control_frame->had_else = TRUE;
3087     }
3088     else
3089     {
3090         shader_addline(buffer, "ELSE;\n");
3091     }
3092 }
3093
3094 static void shader_hw_endif(const struct wined3d_shader_instruction *ins)
3095 {
3096     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3097     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3098     struct list *e = list_head(&priv->control_frames);
3099     struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
3100     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
3101
3102     if(vshader)
3103     {
3104         if(control_frame->had_else)
3105         {
3106             shader_addline(buffer, "ifc_%u_endif:\n", control_frame->no.ifc);
3107         }
3108         else
3109         {
3110             shader_addline(buffer, "#No else branch. else is endif\n");
3111             shader_addline(buffer, "ifc_%u_else:\n", control_frame->no.ifc);
3112         }
3113     }
3114     else
3115     {
3116         shader_addline(buffer, "ENDIF;\n");
3117     }
3118 }
3119
3120 static void shader_hw_texldd(const struct wined3d_shader_instruction *ins)
3121 {
3122     DWORD sampler_idx = ins->src[1].reg.idx;
3123     char reg_dest[40];
3124     char reg_src[3][40];
3125     WORD flags = TEX_DERIV;
3126
3127     shader_arb_get_dst_param(ins, &ins->dst[0], reg_dest);
3128     shader_arb_get_src_param(ins, &ins->src[0], 0, reg_src[0]);
3129     shader_arb_get_src_param(ins, &ins->src[2], 1, reg_src[1]);
3130     shader_arb_get_src_param(ins, &ins->src[3], 2, reg_src[2]);
3131
3132     if (ins->flags & WINED3DSI_TEXLD_PROJECT) flags |= TEX_PROJ;
3133     if (ins->flags & WINED3DSI_TEXLD_BIAS) flags |= TEX_BIAS;
3134
3135     shader_hw_sample(ins, sampler_idx, reg_dest, reg_src[0], flags, reg_src[1], reg_src[2]);
3136 }
3137
3138 static void shader_hw_texldl(const struct wined3d_shader_instruction *ins)
3139 {
3140     DWORD sampler_idx = ins->src[1].reg.idx;
3141     char reg_dest[40];
3142     char reg_coord[40];
3143     WORD flags = TEX_LOD;
3144
3145     shader_arb_get_dst_param(ins, &ins->dst[0], reg_dest);
3146     shader_arb_get_src_param(ins, &ins->src[0], 0, reg_coord);
3147
3148     if (ins->flags & WINED3DSI_TEXLD_PROJECT) flags |= TEX_PROJ;
3149     if (ins->flags & WINED3DSI_TEXLD_BIAS) flags |= TEX_BIAS;
3150
3151     shader_hw_sample(ins, sampler_idx, reg_dest, reg_coord, flags, NULL, NULL);
3152 }
3153
3154 static void shader_hw_label(const struct wined3d_shader_instruction *ins)
3155 {
3156     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3157     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3158
3159     priv->in_main_func = FALSE;
3160     /* Call instructions activate the NV extensions, not labels and rets. If there is an uncalled
3161      * subroutine, don't generate a label that will make GL complain
3162      */
3163     if(priv->target_version == ARB) return;
3164
3165     shader_addline(buffer, "l%u:\n", ins->src[0].reg.idx);
3166 }
3167
3168 static void vshader_add_footer(struct shader_arb_ctx_priv *priv_ctx,
3169         const struct arb_vshader_private *shader_data, const struct arb_vs_compile_args *args,
3170         const struct wined3d_shader_reg_maps *reg_maps, const struct wined3d_gl_info *gl_info,
3171         struct wined3d_shader_buffer *buffer)
3172 {
3173     unsigned int i;
3174
3175     /* The D3DRS_FOGTABLEMODE render state defines if the shader-generated fog coord is used
3176      * or if the fragment depth is used. If the fragment depth is used(FOGTABLEMODE != NONE),
3177      * the fog frag coord is thrown away. If the fog frag coord is used, but not written by
3178      * the shader, it is set to 0.0(fully fogged, since start = 1.0, end = 0.0)
3179      */
3180     if (args->super.fog_src == VS_FOG_Z)
3181     {
3182         shader_addline(buffer, "MOV result.fogcoord, TMP_OUT.z;\n");
3183     }
3184     else
3185     {
3186         if (!reg_maps->fog)
3187         {
3188             /* posFixup.x is always 1.0, so we can safely use it */
3189             shader_addline(buffer, "ADD result.fogcoord, posFixup.x, -posFixup.x;\n");
3190         }
3191         else
3192         {
3193             /* Clamp fogcoord */
3194             const char *zero = arb_get_helper_value(reg_maps->shader_version.type, ARB_ZERO);
3195             const char *one = arb_get_helper_value(reg_maps->shader_version.type, ARB_ONE);
3196
3197             shader_addline(buffer, "MIN TMP_FOGCOORD.x, TMP_FOGCOORD.x, %s;\n", one);
3198             shader_addline(buffer, "MAX result.fogcoord.x, TMP_FOGCOORD.x, %s;\n", zero);
3199         }
3200     }
3201
3202     /* Clipplanes are always stored without y inversion */
3203     if (use_nv_clip(gl_info) && priv_ctx->target_version >= NV2)
3204     {
3205         if (args->super.clip_enabled)
3206         {
3207             for (i = 0; i < priv_ctx->vs_clipplanes; i++)
3208             {
3209                 shader_addline(buffer, "DP4 result.clip[%u].x, TMP_OUT, state.clip[%u].plane;\n", i, i);
3210             }
3211         }
3212     }
3213     else if (args->clip.boolclip.clip_texcoord)
3214     {
3215         unsigned int cur_clip = 0;
3216         char component[4] = {'x', 'y', 'z', 'w'};
3217         const char *zero = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_ZERO);
3218
3219         for (i = 0; i < gl_info->limits.clipplanes; ++i)
3220         {
3221             if (args->clip.boolclip.clipplane_mask & (1 << i))
3222             {
3223                 shader_addline(buffer, "DP4 TA.%c, TMP_OUT, state.clip[%u].plane;\n",
3224                                component[cur_clip++], i);
3225             }
3226         }
3227         switch (cur_clip)
3228         {
3229             case 0:
3230                 shader_addline(buffer, "MOV TA, %s;\n", zero);
3231                 break;
3232             case 1:
3233                 shader_addline(buffer, "MOV TA.yzw, %s;\n", zero);
3234                 break;
3235             case 2:
3236                 shader_addline(buffer, "MOV TA.zw, %s;\n", zero);
3237                 break;
3238             case 3:
3239                 shader_addline(buffer, "MOV TA.w, %s;\n", zero);
3240                 break;
3241         }
3242         shader_addline(buffer, "MOV result.texcoord[%u], TA;\n",
3243                        args->clip.boolclip.clip_texcoord - 1);
3244     }
3245
3246     /* Write the final position.
3247      *
3248      * OpenGL coordinates specify the center of the pixel while d3d coords specify
3249      * the corner. The offsets are stored in z and w in posFixup. posFixup.y contains
3250      * 1.0 or -1.0 to turn the rendering upside down for offscreen rendering. PosFixup.x
3251      * contains 1.0 to allow a mad, but arb vs swizzles are too restricted for that.
3252      */
3253     shader_addline(buffer, "MUL TA, posFixup, TMP_OUT.w;\n");
3254     shader_addline(buffer, "ADD TMP_OUT.x, TMP_OUT.x, TA.z;\n");
3255     shader_addline(buffer, "MAD TMP_OUT.y, TMP_OUT.y, posFixup.y, TA.w;\n");
3256
3257     /* Z coord [0;1]->[-1;1] mapping, see comment in transform_projection in state.c
3258      * and the glsl equivalent
3259      */
3260     if (need_helper_const(shader_data, reg_maps, gl_info))
3261     {
3262         const char *two = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_TWO);
3263         shader_addline(buffer, "MAD TMP_OUT.z, TMP_OUT.z, %s, -TMP_OUT.w;\n", two);
3264     }
3265     else
3266     {
3267         shader_addline(buffer, "ADD TMP_OUT.z, TMP_OUT.z, TMP_OUT.z;\n");
3268         shader_addline(buffer, "ADD TMP_OUT.z, TMP_OUT.z, -TMP_OUT.w;\n");
3269     }
3270
3271     shader_addline(buffer, "MOV result.position, TMP_OUT;\n");
3272
3273     priv_ctx->footer_written = TRUE;
3274 }
3275
3276 static void shader_hw_ret(const struct wined3d_shader_instruction *ins)
3277 {
3278     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3279     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3280     const struct wined3d_shader *shader = ins->ctx->shader;
3281     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
3282
3283     if(priv->target_version == ARB) return;
3284
3285     if(vshader)
3286     {
3287         if (priv->in_main_func) vshader_add_footer(priv, shader->backend_data,
3288                 priv->cur_vs_args, ins->ctx->reg_maps, ins->ctx->gl_info, buffer);
3289     }
3290
3291     shader_addline(buffer, "RET;\n");
3292 }
3293
3294 static void shader_hw_call(const struct wined3d_shader_instruction *ins)
3295 {
3296     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3297     shader_addline(buffer, "CAL l%u;\n", ins->src[0].reg.idx);
3298 }
3299
3300 /* GL locking is done by the caller */
3301 static GLuint create_arb_blt_vertex_program(const struct wined3d_gl_info *gl_info)
3302 {
3303     GLuint program_id = 0;
3304     GLint pos;
3305
3306     const char *blt_vprogram =
3307         "!!ARBvp1.0\n"
3308         "PARAM c[1] = { { 1, 0.5 } };\n"
3309         "MOV result.position, vertex.position;\n"
3310         "MOV result.color, c[0].x;\n"
3311         "MOV result.texcoord[0], vertex.texcoord[0];\n"
3312         "END\n";
3313
3314     GL_EXTCALL(glGenProgramsARB(1, &program_id));
3315     GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, program_id));
3316     GL_EXTCALL(glProgramStringARB(GL_VERTEX_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
3317             strlen(blt_vprogram), blt_vprogram));
3318     checkGLcall("glProgramStringARB()");
3319
3320     gl_info->gl_ops.gl.p_glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
3321     if (pos != -1)
3322     {
3323         FIXME("Vertex program error at position %d: %s\n\n", pos,
3324             debugstr_a((const char *)gl_info->gl_ops.gl.p_glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
3325         shader_arb_dump_program_source(blt_vprogram);
3326     }
3327     else
3328     {
3329         GLint native;
3330
3331         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
3332         checkGLcall("glGetProgramivARB()");
3333         if (!native) WARN("Program exceeds native resource limits.\n");
3334     }
3335
3336     return program_id;
3337 }
3338
3339 /* GL locking is done by the caller */
3340 static GLuint create_arb_blt_fragment_program(const struct wined3d_gl_info *gl_info,
3341         enum tex_types tex_type, BOOL masked)
3342 {
3343     GLuint program_id = 0;
3344     const char *fprogram;
3345     GLint pos;
3346
3347     static const char * const blt_fprograms_full[tex_type_count] =
3348     {
3349         /* tex_1d */
3350         NULL,
3351         /* tex_2d */
3352         "!!ARBfp1.0\n"
3353         "TEMP R0;\n"
3354         "TEX R0.x, fragment.texcoord[0], texture[0], 2D;\n"
3355         "MOV result.depth.z, R0.x;\n"
3356         "END\n",
3357         /* tex_3d */
3358         NULL,
3359         /* tex_cube */
3360         "!!ARBfp1.0\n"
3361         "TEMP R0;\n"
3362         "TEX R0.x, fragment.texcoord[0], texture[0], CUBE;\n"
3363         "MOV result.depth.z, R0.x;\n"
3364         "END\n",
3365         /* tex_rect */
3366         "!!ARBfp1.0\n"
3367         "TEMP R0;\n"
3368         "TEX R0.x, fragment.texcoord[0], texture[0], RECT;\n"
3369         "MOV result.depth.z, R0.x;\n"
3370         "END\n",
3371     };
3372
3373     static const char * const blt_fprograms_masked[tex_type_count] =
3374     {
3375         /* tex_1d */
3376         NULL,
3377         /* tex_2d */
3378         "!!ARBfp1.0\n"
3379         "PARAM mask = program.local[0];\n"
3380         "TEMP R0;\n"
3381         "SLT R0.xy, fragment.position, mask.zwzw;\n"
3382         "MUL R0.x, R0.x, R0.y;\n"
3383         "KIL -R0.x;\n"
3384         "TEX R0.x, fragment.texcoord[0], texture[0], 2D;\n"
3385         "MOV result.depth.z, R0.x;\n"
3386         "END\n",
3387         /* tex_3d */
3388         NULL,
3389         /* tex_cube */
3390         "!!ARBfp1.0\n"
3391         "PARAM mask = program.local[0];\n"
3392         "TEMP R0;\n"
3393         "SLT R0.xy, fragment.position, mask.zwzw;\n"
3394         "MUL R0.x, R0.x, R0.y;\n"
3395         "KIL -R0.x;\n"
3396         "TEX R0.x, fragment.texcoord[0], texture[0], CUBE;\n"
3397         "MOV result.depth.z, R0.x;\n"
3398         "END\n",
3399         /* tex_rect */
3400         "!!ARBfp1.0\n"
3401         "PARAM mask = program.local[0];\n"
3402         "TEMP R0;\n"
3403         "SLT R0.xy, fragment.position, mask.zwzw;\n"
3404         "MUL R0.x, R0.x, R0.y;\n"
3405         "KIL -R0.x;\n"
3406         "TEX R0.x, fragment.texcoord[0], texture[0], RECT;\n"
3407         "MOV result.depth.z, R0.x;\n"
3408         "END\n",
3409     };
3410
3411     fprogram = masked ? blt_fprograms_masked[tex_type] : blt_fprograms_full[tex_type];
3412     if (!fprogram)
3413     {
3414         FIXME("tex_type %#x not supported, falling back to tex_2d\n", tex_type);
3415         tex_type = tex_2d;
3416         fprogram = masked ? blt_fprograms_masked[tex_type] : blt_fprograms_full[tex_type];
3417     }
3418
3419     GL_EXTCALL(glGenProgramsARB(1, &program_id));
3420     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, program_id));
3421     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB, strlen(fprogram), fprogram));
3422     checkGLcall("glProgramStringARB()");
3423
3424     gl_info->gl_ops.gl.p_glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
3425     if (pos != -1)
3426     {
3427         FIXME("Fragment program error at position %d: %s\n\n", pos,
3428             debugstr_a((const char *)gl_info->gl_ops.gl.p_glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
3429         shader_arb_dump_program_source(fprogram);
3430     }
3431     else
3432     {
3433         GLint native;
3434
3435         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
3436         checkGLcall("glGetProgramivARB()");
3437         if (!native) WARN("Program exceeds native resource limits.\n");
3438     }
3439
3440     return program_id;
3441 }
3442
3443 static void arbfp_add_sRGB_correction(struct wined3d_shader_buffer *buffer, const char *fragcolor,
3444         const char *tmp1, const char *tmp2, const char *tmp3, const char *tmp4, BOOL condcode)
3445 {
3446     /* Perform sRGB write correction. See GLX_EXT_framebuffer_sRGB */
3447
3448     if(condcode)
3449     {
3450         /* Sigh. MOVC CC doesn't work, so use one of the temps as dummy dest */
3451         shader_addline(buffer, "SUBC %s, %s.x, srgb_consts1.y;\n", tmp1, fragcolor);
3452         /* Calculate the > 0.0031308 case */
3453         shader_addline(buffer, "POW %s.x (GE), %s.x, srgb_consts1.z;\n", fragcolor, fragcolor);
3454         shader_addline(buffer, "POW %s.y (GE), %s.y, srgb_consts1.z;\n", fragcolor, fragcolor);
3455         shader_addline(buffer, "POW %s.z (GE), %s.z, srgb_consts1.z;\n", fragcolor, fragcolor);
3456         shader_addline(buffer, "MUL %s.xyz (GE), %s, srgb_consts1.w;\n", fragcolor, fragcolor);
3457         shader_addline(buffer, "SUB %s.xyz (GE), %s, srgb_consts2.x;\n", fragcolor, fragcolor);
3458         /* Calculate the < case */
3459         shader_addline(buffer, "MUL %s.xyz (LT), srgb_consts1.x, %s;\n", fragcolor, fragcolor);
3460     }
3461     else
3462     {
3463         /* Calculate the > 0.0031308 case */
3464         shader_addline(buffer, "POW %s.x, %s.x, srgb_consts1.z;\n", tmp1, fragcolor);
3465         shader_addline(buffer, "POW %s.y, %s.y, srgb_consts1.z;\n", tmp1, fragcolor);
3466         shader_addline(buffer, "POW %s.z, %s.z, srgb_consts1.z;\n", tmp1, fragcolor);
3467         shader_addline(buffer, "MUL %s, %s, srgb_consts1.w;\n", tmp1, tmp1);
3468         shader_addline(buffer, "SUB %s, %s, srgb_consts2.x;\n", tmp1, tmp1);
3469         /* Calculate the < case */
3470         shader_addline(buffer, "MUL %s, srgb_consts1.x, %s;\n", tmp2, fragcolor);
3471         /* Get 1.0 / 0.0 masks for > 0.0031308 and < 0.0031308 */
3472         shader_addline(buffer, "SLT %s, srgb_consts1.y, %s;\n", tmp3, fragcolor);
3473         shader_addline(buffer, "SGE %s, srgb_consts1.y, %s;\n", tmp4, fragcolor);
3474         /* Store the components > 0.0031308 in the destination */
3475         shader_addline(buffer, "MUL %s.xyz, %s, %s;\n", fragcolor, tmp1, tmp3);
3476         /* Add the components that are < 0.0031308 */
3477         shader_addline(buffer, "MAD %s.xyz, %s, %s, %s;\n", fragcolor, tmp2, tmp4, fragcolor);
3478         /* Move everything into result.color at once. Nvidia hardware cannot handle partial
3479         * result.color writes(.rgb first, then .a), or handle overwriting already written
3480         * components. The assembler uses a temporary register in this case, which is usually
3481         * not allocated from one of our registers that were used earlier.
3482         */
3483     }
3484     /* [0.0;1.0] clamping. Not needed, this is done implicitly */
3485 }
3486
3487 static const DWORD *find_loop_control_values(const struct wined3d_shader *shader, DWORD idx)
3488 {
3489     const struct wined3d_shader_lconst *constant;
3490
3491     LIST_FOR_EACH_ENTRY(constant, &shader->constantsI, struct wined3d_shader_lconst, entry)
3492     {
3493         if (constant->idx == idx)
3494         {
3495             return constant->value;
3496         }
3497     }
3498     return NULL;
3499 }
3500
3501 static void init_ps_input(const struct wined3d_shader *shader,
3502         const struct arb_ps_compile_args *args, struct shader_arb_ctx_priv *priv)
3503 {
3504     static const char * const texcoords[8] =
3505     {
3506         "fragment.texcoord[0]", "fragment.texcoord[1]", "fragment.texcoord[2]", "fragment.texcoord[3]",
3507         "fragment.texcoord[4]", "fragment.texcoord[5]", "fragment.texcoord[6]", "fragment.texcoord[7]"
3508     };
3509     unsigned int i;
3510     const struct wined3d_shader_signature_element *sig = shader->input_signature;
3511     const char *semantic_name;
3512     DWORD semantic_idx;
3513
3514     switch(args->super.vp_mode)
3515     {
3516         case pretransformed:
3517         case fixedfunction:
3518             /* The pixelshader has to collect the varyings on its own. In any case properly load
3519              * color0 and color1. In the case of pretransformed vertices also load texcoords. Set
3520              * other attribs to 0.0.
3521              *
3522              * For fixedfunction this behavior is correct, according to the tests. For pretransformed
3523              * we'd either need a replacement shader that can load other attribs like BINORMAL, or
3524              * load the texcoord attrib pointers to match the pixel shader signature
3525              */
3526             for(i = 0; i < MAX_REG_INPUT; i++)
3527             {
3528                 semantic_name = sig[i].semantic_name;
3529                 semantic_idx = sig[i].semantic_idx;
3530                 if (!semantic_name) continue;
3531
3532                 if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_COLOR))
3533                 {
3534                     if (!semantic_idx) priv->ps_input[i] = "fragment.color.primary";
3535                     else if(semantic_idx == 1) priv->ps_input[i] = "fragment.color.secondary";
3536                     else priv->ps_input[i] = "0.0";
3537                 }
3538                 else if(args->super.vp_mode == fixedfunction)
3539                 {
3540                     priv->ps_input[i] = "0.0";
3541                 }
3542                 else if(shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_TEXCOORD))
3543                 {
3544                     if(semantic_idx < 8) priv->ps_input[i] = texcoords[semantic_idx];
3545                     else priv->ps_input[i] = "0.0";
3546                 }
3547                 else if(shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_FOG))
3548                 {
3549                     if (!semantic_idx) priv->ps_input[i] = "fragment.fogcoord";
3550                     else priv->ps_input[i] = "0.0";
3551                 }
3552                 else
3553                 {
3554                     priv->ps_input[i] = "0.0";
3555                 }
3556
3557                 TRACE("v%u, semantic %s%u is %s\n", i, semantic_name, semantic_idx, priv->ps_input[i]);
3558             }
3559             break;
3560
3561         case vertexshader:
3562             /* That one is easy. The vertex shaders provide v0-v7 in fragment.texcoord and v8 and v9 in
3563              * fragment.color
3564              */
3565             for(i = 0; i < 8; i++)
3566             {
3567                 priv->ps_input[i] = texcoords[i];
3568             }
3569             priv->ps_input[8] = "fragment.color.primary";
3570             priv->ps_input[9] = "fragment.color.secondary";
3571             break;
3572     }
3573 }
3574
3575 /* GL locking is done by the caller */
3576 static GLuint shader_arb_generate_pshader(const struct wined3d_shader *shader,
3577         const struct wined3d_gl_info *gl_info, struct wined3d_shader_buffer *buffer,
3578         const struct arb_ps_compile_args *args, struct arb_ps_compiled_shader *compiled)
3579 {
3580     const struct wined3d_shader_reg_maps *reg_maps = &shader->reg_maps;
3581     const struct wined3d_shader_lconst *lconst;
3582     const DWORD *function = shader->function;
3583     GLuint retval;
3584     char fragcolor[16];
3585     DWORD *lconst_map = local_const_mapping(shader), next_local;
3586     struct shader_arb_ctx_priv priv_ctx;
3587     BOOL dcl_td = FALSE;
3588     BOOL want_nv_prog = FALSE;
3589     struct arb_pshader_private *shader_priv = shader->backend_data;
3590     GLint errPos;
3591     DWORD map;
3592
3593     char srgbtmp[4][4];
3594     unsigned int i, found = 0;
3595
3596     for (i = 0, map = reg_maps->temporary; map; map >>= 1, ++i)
3597     {
3598         if (!(map & 1)
3599                 || (shader->u.ps.color0_mov && i == shader->u.ps.color0_reg)
3600                 || (reg_maps->shader_version.major < 2 && !i))
3601             continue;
3602
3603         sprintf(srgbtmp[found], "R%u", i);
3604         ++found;
3605         if (found == 4) break;
3606     }
3607
3608     switch(found) {
3609         case 0:
3610             sprintf(srgbtmp[0], "TA");
3611             sprintf(srgbtmp[1], "TB");
3612             sprintf(srgbtmp[2], "TC");
3613             sprintf(srgbtmp[3], "TD");
3614             dcl_td = TRUE;
3615             break;
3616         case 1:
3617             sprintf(srgbtmp[1], "TA");
3618             sprintf(srgbtmp[2], "TB");
3619             sprintf(srgbtmp[3], "TC");
3620             break;
3621         case 2:
3622             sprintf(srgbtmp[2], "TA");
3623             sprintf(srgbtmp[3], "TB");
3624             break;
3625         case 3:
3626             sprintf(srgbtmp[3], "TA");
3627             break;
3628         case 4:
3629             break;
3630     }
3631
3632     /*  Create the hw ARB shader */
3633     memset(&priv_ctx, 0, sizeof(priv_ctx));
3634     priv_ctx.cur_ps_args = args;
3635     priv_ctx.compiled_fprog = compiled;
3636     priv_ctx.cur_np2fixup_info = &compiled->np2fixup_info;
3637     init_ps_input(shader, args, &priv_ctx);
3638     list_init(&priv_ctx.control_frames);
3639
3640     /* Avoid enabling NV_fragment_program* if we do not need it.
3641      *
3642      * Enabling GL_NV_fragment_program_option causes the driver to occupy a temporary register,
3643      * and it slows down the shader execution noticeably(about 5%). Usually our instruction emulation
3644      * is faster than what we gain from using higher native instructions. There are some things though
3645      * that cannot be emulated. In that case enable the extensions.
3646      * If the extension is enabled, instruction handlers that support both ways will use it.
3647      *
3648      * Testing shows no performance difference between OPTION NV_fragment_program2 and NV_fragment_program.
3649      * So enable the best we can get.
3650      */
3651     if(reg_maps->usesdsx || reg_maps->usesdsy || reg_maps->loop_depth > 0 || reg_maps->usestexldd ||
3652        reg_maps->usestexldl || reg_maps->usesfacing || reg_maps->usesifc || reg_maps->usescall)
3653     {
3654         want_nv_prog = TRUE;
3655     }
3656
3657     shader_addline(buffer, "!!ARBfp1.0\n");
3658     if (want_nv_prog && gl_info->supported[NV_FRAGMENT_PROGRAM2])
3659     {
3660         shader_addline(buffer, "OPTION NV_fragment_program2;\n");
3661         priv_ctx.target_version = NV3;
3662     }
3663     else if (want_nv_prog && gl_info->supported[NV_FRAGMENT_PROGRAM_OPTION])
3664     {
3665         shader_addline(buffer, "OPTION NV_fragment_program;\n");
3666         priv_ctx.target_version = NV2;
3667     } else {
3668         if(want_nv_prog)
3669         {
3670             /* This is an error - either we're advertising the wrong shader version, or aren't enforcing some
3671              * limits properly
3672              */
3673             ERR("The shader requires instructions that are not available in plain GL_ARB_fragment_program\n");
3674             ERR("Try GLSL\n");
3675         }
3676         priv_ctx.target_version = ARB;
3677     }
3678
3679     if (reg_maps->rt_mask > 1)
3680     {
3681         shader_addline(buffer, "OPTION ARB_draw_buffers;\n");
3682     }
3683
3684     if (reg_maps->shader_version.major < 3)
3685     {
3686         switch(args->super.fog) {
3687             case FOG_OFF:
3688                 break;
3689             case FOG_LINEAR:
3690                 shader_addline(buffer, "OPTION ARB_fog_linear;\n");
3691                 break;
3692             case FOG_EXP:
3693                 shader_addline(buffer, "OPTION ARB_fog_exp;\n");
3694                 break;
3695             case FOG_EXP2:
3696                 shader_addline(buffer, "OPTION ARB_fog_exp2;\n");
3697                 break;
3698         }
3699     }
3700
3701     /* For now always declare the temps. At least the Nvidia assembler optimizes completely
3702      * unused temps away(but occupies them for the whole shader if they're used once). Always
3703      * declaring them avoids tricky bookkeeping work
3704      */
3705     shader_addline(buffer, "TEMP TA;\n");      /* Used for modifiers */
3706     shader_addline(buffer, "TEMP TB;\n");      /* Used for modifiers */
3707     shader_addline(buffer, "TEMP TC;\n");      /* Used for modifiers */
3708     if(dcl_td) shader_addline(buffer, "TEMP TD;\n"); /* Used for sRGB writing */
3709     shader_addline(buffer, "PARAM coefdiv = { 0.5, 0.25, 0.125, 0.0625 };\n");
3710     shader_addline(buffer, "PARAM coefmul = { 2, 4, 8, 16 };\n");
3711     shader_addline(buffer, "PARAM ps_helper_const = { 0.0, 1.0, %1.10f, 0.0 };\n", eps);
3712
3713     if (reg_maps->shader_version.major < 2)
3714     {
3715         strcpy(fragcolor, "R0");
3716     }
3717     else
3718     {
3719         if (args->super.srgb_correction)
3720         {
3721             if (shader->u.ps.color0_mov)
3722             {
3723                 sprintf(fragcolor, "R%u", shader->u.ps.color0_reg);
3724             }
3725             else
3726             {
3727                 shader_addline(buffer, "TEMP TMP_COLOR;\n");
3728                 strcpy(fragcolor, "TMP_COLOR");
3729             }
3730         } else {
3731             strcpy(fragcolor, "result.color");
3732         }
3733     }
3734
3735     if(args->super.srgb_correction) {
3736         shader_addline(buffer, "PARAM srgb_consts1 = {%f, %f, %f, %f};\n",
3737                        srgb_mul_low, srgb_cmp, srgb_pow, srgb_mul_high);
3738         shader_addline(buffer, "PARAM srgb_consts2 = {%f, %f, %f, %f};\n",
3739                        srgb_sub_high, 0.0, 0.0, 0.0);
3740     }
3741
3742     /* Base Declarations */
3743     next_local = shader_generate_arb_declarations(shader, reg_maps,
3744             buffer, gl_info, lconst_map, NULL, &priv_ctx);
3745
3746     for (i = 0, map = reg_maps->bumpmat; map; map >>= 1, ++i)
3747     {
3748         unsigned char bump_const;
3749
3750         if (!(map & 1)) continue;
3751
3752         bump_const = compiled->numbumpenvmatconsts;
3753         compiled->bumpenvmatconst[bump_const].const_num = WINED3D_CONST_NUM_UNUSED;
3754         compiled->bumpenvmatconst[bump_const].texunit = i;
3755         compiled->luminanceconst[bump_const].const_num = WINED3D_CONST_NUM_UNUSED;
3756         compiled->luminanceconst[bump_const].texunit = i;
3757
3758         /* We can fit the constants into the constant limit for sure because texbem, texbeml, bem and beml are only supported
3759          * in 1.x shaders, and GL_ARB_fragment_program has a constant limit of 24 constants. So in the worst case we're loading
3760          * 8 shader constants, 8 bump matrices and 8 luminance parameters and are perfectly fine. (No NP2 fixup on bumpmapped
3761          * textures due to conditional NP2 restrictions)
3762          *
3763          * Use local constants to load the bump env parameters, not program.env. This avoids collisions with d3d constants of
3764          * shaders in newer shader models. Since the bump env parameters have to share their space with NP2 fixup constants,
3765          * their location is shader dependent anyway and they cannot be loaded globally.
3766          */
3767         compiled->bumpenvmatconst[bump_const].const_num = next_local++;
3768         shader_addline(buffer, "PARAM bumpenvmat%d = program.local[%d];\n",
3769                        i, compiled->bumpenvmatconst[bump_const].const_num);
3770         compiled->numbumpenvmatconsts = bump_const + 1;
3771
3772         if (!(reg_maps->luminanceparams & (1 << i))) continue;
3773
3774         compiled->luminanceconst[bump_const].const_num = next_local++;
3775         shader_addline(buffer, "PARAM luminance%d = program.local[%d];\n",
3776                        i, compiled->luminanceconst[bump_const].const_num);
3777     }
3778
3779     for(i = 0; i < MAX_CONST_I; i++)
3780     {
3781         compiled->int_consts[i] = WINED3D_CONST_NUM_UNUSED;
3782         if (reg_maps->integer_constants & (1 << i) && priv_ctx.target_version >= NV2)
3783         {
3784             const DWORD *control_values = find_loop_control_values(shader, i);
3785
3786             if(control_values)
3787             {
3788                 shader_addline(buffer, "PARAM I%u = {%u, %u, %u, -1};\n", i,
3789                                 control_values[0], control_values[1], control_values[2]);
3790             }
3791             else
3792             {
3793                 compiled->int_consts[i] = next_local;
3794                 compiled->num_int_consts++;
3795                 shader_addline(buffer, "PARAM I%u = program.local[%u];\n", i, next_local++);
3796             }
3797         }
3798     }
3799
3800     if(reg_maps->vpos || reg_maps->usesdsy)
3801     {
3802         compiled->ycorrection = next_local;
3803         shader_addline(buffer, "PARAM ycorrection = program.local[%u];\n", next_local++);
3804
3805         if(reg_maps->vpos)
3806         {
3807             shader_addline(buffer, "TEMP vpos;\n");
3808             /* ycorrection.x: Backbuffer height(onscreen) or 0(offscreen).
3809              * ycorrection.y: -1.0(onscreen), 1.0(offscreen)
3810              * ycorrection.z: 1.0
3811              * ycorrection.w: 0.0
3812              */
3813             shader_addline(buffer, "MAD vpos, fragment.position, ycorrection.zyww, ycorrection.wxww;\n");
3814             shader_addline(buffer, "FLR vpos.xy, vpos;\n");
3815         }
3816     }
3817     else
3818     {
3819         compiled->ycorrection = WINED3D_CONST_NUM_UNUSED;
3820     }
3821
3822     /* Load constants to fixup NP2 texcoords if there are still free constants left:
3823      * Constants (texture dimensions) for the NP2 fixup are loaded as local program parameters. This will consume
3824      * at most 8 (MAX_FRAGMENT_SAMPLERS / 2) parameters, which is highly unlikely, since the application had to
3825      * use 16 NP2 textures at the same time. In case that we run out of constants the fixup is simply not
3826      * applied / activated. This will probably result in wrong rendering of the texture, but will save us from
3827      * shader compilation errors and the subsequent errors when drawing with this shader. */
3828     if (priv_ctx.cur_ps_args->super.np2_fixup) {
3829         unsigned char cur_fixup_sampler = 0;
3830
3831         struct arb_ps_np2fixup_info* const fixup = priv_ctx.cur_np2fixup_info;
3832         const WORD map = priv_ctx.cur_ps_args->super.np2_fixup;
3833         const UINT max_lconsts = gl_info->limits.arb_ps_local_constants;
3834
3835         fixup->offset = next_local;
3836         fixup->super.active = 0;
3837
3838         for (i = 0; i < MAX_FRAGMENT_SAMPLERS; ++i) {
3839             if (!(map & (1 << i))) continue;
3840
3841             if (fixup->offset + (cur_fixup_sampler >> 1) < max_lconsts) {
3842                 fixup->super.active |= (1 << i);
3843                 fixup->super.idx[i] = cur_fixup_sampler++;
3844             } else {
3845                 FIXME("No free constant found to load NP2 fixup data into shader. "
3846                       "Sampling from this texture will probably look wrong.\n");
3847                 break;
3848             }
3849         }
3850
3851         fixup->super.num_consts = (cur_fixup_sampler + 1) >> 1;
3852         if (fixup->super.num_consts) {
3853             shader_addline(buffer, "PARAM np2fixup[%u] = { program.env[%u..%u] };\n",
3854                            fixup->super.num_consts, fixup->offset, fixup->super.num_consts + fixup->offset - 1);
3855         }
3856     }
3857
3858     if (shader_priv->clipplane_emulation != ~0U && args->clip)
3859     {
3860         shader_addline(buffer, "KIL fragment.texcoord[%u];\n", shader_priv->clipplane_emulation);
3861     }
3862
3863     /* Base Shader Body */
3864     shader_generate_main(shader, buffer, reg_maps, function, &priv_ctx);
3865
3866     if(args->super.srgb_correction) {
3867         arbfp_add_sRGB_correction(buffer, fragcolor, srgbtmp[0], srgbtmp[1], srgbtmp[2], srgbtmp[3],
3868                                   priv_ctx.target_version >= NV2);
3869     }
3870
3871     if(strcmp(fragcolor, "result.color")) {
3872         shader_addline(buffer, "MOV result.color, %s;\n", fragcolor);
3873     }
3874     shader_addline(buffer, "END\n");
3875
3876     /* TODO: change to resource.glObjectHandle or something like that */
3877     GL_EXTCALL(glGenProgramsARB(1, &retval));
3878
3879     TRACE("Creating a hw pixel shader, prg=%d\n", retval);
3880     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, retval));
3881
3882     TRACE("Created hw pixel shader, prg=%d\n", retval);
3883     /* Create the program and check for errors */
3884     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
3885                buffer->bsize, buffer->buffer));
3886     checkGLcall("glProgramStringARB()");
3887
3888     gl_info->gl_ops.gl.p_glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &errPos);
3889     if (errPos != -1)
3890     {
3891         FIXME("HW PixelShader Error at position %d: %s\n\n",
3892               errPos, debugstr_a((const char *)gl_info->gl_ops.gl.p_glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
3893         shader_arb_dump_program_source(buffer->buffer);
3894         retval = 0;
3895     }
3896     else
3897     {
3898         GLint native;
3899
3900         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
3901         checkGLcall("glGetProgramivARB()");
3902         if (!native) WARN("Program exceeds native resource limits.\n");
3903     }
3904
3905     /* Load immediate constants */
3906     if (lconst_map)
3907     {
3908         LIST_FOR_EACH_ENTRY(lconst, &shader->constantsF, struct wined3d_shader_lconst, entry)
3909         {
3910             const float *value = (const float *)lconst->value;
3911             GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, lconst_map[lconst->idx], value));
3912             checkGLcall("glProgramLocalParameter4fvARB");
3913         }
3914         HeapFree(GetProcessHeap(), 0, lconst_map);
3915     }
3916
3917     return retval;
3918 }
3919
3920 static int compare_sig(const struct wined3d_shader_signature_element *sig1, const struct wined3d_shader_signature_element *sig2)
3921 {
3922     unsigned int i;
3923     int ret;
3924
3925     for(i = 0; i < MAX_REG_INPUT; i++)
3926     {
3927         if (!sig1[i].semantic_name || !sig2[i].semantic_name)
3928         {
3929             /* Compare pointers, not contents. One string is NULL(element does not exist), the other one is not NULL */
3930             if(sig1[i].semantic_name != sig2[i].semantic_name) return sig1[i].semantic_name < sig2[i].semantic_name ? -1 : 1;
3931             continue;
3932         }
3933
3934         if ((ret = strcmp(sig1[i].semantic_name, sig2[i].semantic_name))) return ret;
3935         if(sig1[i].semantic_idx    != sig2[i].semantic_idx)    return sig1[i].semantic_idx    < sig2[i].semantic_idx    ? -1 : 1;
3936         if(sig1[i].sysval_semantic != sig2[i].sysval_semantic) return sig1[i].sysval_semantic < sig2[i].sysval_semantic ? -1 : 1;
3937         if(sig1[i].component_type  != sig2[i].component_type)  return sig1[i].component_type  < sig2[i].component_type  ? -1 : 1;
3938         if(sig1[i].register_idx    != sig2[i].register_idx)    return sig1[i].register_idx    < sig2[i].register_idx    ? -1 : 1;
3939         if(sig1[i].mask            != sig2[i].mask)            return sig1[i].mask            < sig2[i].mask            ? -1 : 1;
3940     }
3941     return 0;
3942 }
3943
3944 static struct wined3d_shader_signature_element *clone_sig(const struct wined3d_shader_signature_element *sig)
3945 {
3946     struct wined3d_shader_signature_element *new;
3947     int i;
3948     char *name;
3949
3950     new = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*new) * MAX_REG_INPUT);
3951     for(i = 0; i < MAX_REG_INPUT; i++)
3952     {
3953         if (!sig[i].semantic_name) continue;
3954
3955         new[i] = sig[i];
3956         /* Clone the semantic string */
3957         name = HeapAlloc(GetProcessHeap(), 0, strlen(sig[i].semantic_name) + 1);
3958         strcpy(name, sig[i].semantic_name);
3959         new[i].semantic_name = name;
3960     }
3961     return new;
3962 }
3963
3964 static DWORD find_input_signature(struct shader_arb_priv *priv, const struct wined3d_shader_signature_element *sig)
3965 {
3966     struct wine_rb_entry *entry = wine_rb_get(&priv->signature_tree, sig);
3967     struct ps_signature *found_sig;
3968
3969     if (entry)
3970     {
3971         found_sig = WINE_RB_ENTRY_VALUE(entry, struct ps_signature, entry);
3972         TRACE("Found existing signature %u\n", found_sig->idx);
3973         return found_sig->idx;
3974     }
3975     found_sig = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*sig));
3976     found_sig->sig = clone_sig(sig);
3977     found_sig->idx = priv->ps_sig_number++;
3978     TRACE("New signature stored and assigned number %u\n", found_sig->idx);
3979     if(wine_rb_put(&priv->signature_tree, sig, &found_sig->entry) == -1)
3980     {
3981         ERR("Failed to insert program entry.\n");
3982     }
3983     return found_sig->idx;
3984 }
3985
3986 static void init_output_registers(const struct wined3d_shader *shader, DWORD sig_num,
3987         struct shader_arb_ctx_priv *priv_ctx, struct arb_vs_compiled_shader *compiled)
3988 {
3989     unsigned int i, j;
3990     static const char * const texcoords[8] =
3991     {
3992         "result.texcoord[0]", "result.texcoord[1]", "result.texcoord[2]", "result.texcoord[3]",
3993         "result.texcoord[4]", "result.texcoord[5]", "result.texcoord[6]", "result.texcoord[7]"
3994     };
3995     struct wined3d_device *device = shader->device;
3996     const struct wined3d_shader_signature_element *sig;
3997     const char *semantic_name;
3998     DWORD semantic_idx, reg_idx;
3999
4000     /* Write generic input varyings 0 to 7 to result.texcoord[], varying 8 to result.color.primary
4001      * and varying 9 to result.color.secondary
4002      */
4003     static const char * const decl_idx_to_string[MAX_REG_INPUT] =
4004     {
4005         "result.texcoord[0]", "result.texcoord[1]", "result.texcoord[2]", "result.texcoord[3]",
4006         "result.texcoord[4]", "result.texcoord[5]", "result.texcoord[6]", "result.texcoord[7]",
4007         "result.color.primary", "result.color.secondary"
4008     };
4009
4010     if(sig_num == ~0)
4011     {
4012         TRACE("Pixel shader uses builtin varyings\n");
4013         /* Map builtins to builtins */
4014         for(i = 0; i < 8; i++)
4015         {
4016             priv_ctx->texcrd_output[i] = texcoords[i];
4017         }
4018         priv_ctx->color_output[0] = "result.color.primary";
4019         priv_ctx->color_output[1] = "result.color.secondary";
4020         priv_ctx->fog_output = "TMP_FOGCOORD";
4021
4022         /* Map declared regs to builtins. Use "TA" to /dev/null unread output */
4023         for (i = 0; i < (sizeof(shader->output_signature) / sizeof(*shader->output_signature)); ++i)
4024         {
4025             semantic_name = shader->output_signature[i].semantic_name;
4026             if (!semantic_name) continue;
4027
4028             if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_POSITION))
4029             {
4030                 TRACE("o%u is TMP_OUT\n", i);
4031                 if (!shader->output_signature[i].semantic_idx) priv_ctx->vs_output[i] = "TMP_OUT";
4032                 else priv_ctx->vs_output[i] = "TA";
4033             }
4034             else if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_PSIZE))
4035             {
4036                 TRACE("o%u is result.pointsize\n", i);
4037                 if (!shader->output_signature[i].semantic_idx) priv_ctx->vs_output[i] = "result.pointsize";
4038                 else priv_ctx->vs_output[i] = "TA";
4039             }
4040             else if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_COLOR))
4041             {
4042                 TRACE("o%u is result.color.?, idx %u\n", i, shader->output_signature[i].semantic_idx);
4043                 if (!shader->output_signature[i].semantic_idx)
4044                     priv_ctx->vs_output[i] = "result.color.primary";
4045                 else if (shader->output_signature[i].semantic_idx == 1)
4046                     priv_ctx->vs_output[i] = "result.color.secondary";
4047                 else priv_ctx->vs_output[i] = "TA";
4048             }
4049             else if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_TEXCOORD))
4050             {
4051                 TRACE("o%u is %s\n", i, texcoords[shader->output_signature[i].semantic_idx]);
4052                 if (shader->output_signature[i].semantic_idx >= 8) priv_ctx->vs_output[i] = "TA";
4053                 else priv_ctx->vs_output[i] = texcoords[shader->output_signature[i].semantic_idx];
4054             }
4055             else if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_FOG))
4056             {
4057                 TRACE("o%u is result.fogcoord\n", i);
4058                 if (shader->output_signature[i].semantic_idx > 0) priv_ctx->vs_output[i] = "TA";
4059                 else priv_ctx->vs_output[i] = "result.fogcoord";
4060             }
4061             else
4062             {
4063                 priv_ctx->vs_output[i] = "TA";
4064             }
4065         }
4066         return;
4067     }
4068
4069     /* Instead of searching for the signature in the signature list, read the one from the current pixel shader.
4070      * Its maybe not the shader where the signature came from, but it is the same signature and faster to find
4071      */
4072     sig = device->stateBlock->state.pixel_shader->input_signature;
4073     TRACE("Pixel shader uses declared varyings\n");
4074
4075     /* Map builtin to declared. /dev/null the results by default to the TA temp reg */
4076     for(i = 0; i < 8; i++)
4077     {
4078         priv_ctx->texcrd_output[i] = "TA";
4079     }
4080     priv_ctx->color_output[0] = "TA";
4081     priv_ctx->color_output[1] = "TA";
4082     priv_ctx->fog_output = "TA";
4083
4084     for(i = 0; i < MAX_REG_INPUT; i++)
4085     {
4086         semantic_name = sig[i].semantic_name;
4087         semantic_idx = sig[i].semantic_idx;
4088         reg_idx = sig[i].register_idx;
4089         if (!semantic_name) continue;
4090
4091         /* If a declared input register is not written by builtin arguments, don't write to it.
4092          * GL_NV_vertex_program makes sure the input defaults to 0.0, which is correct with D3D
4093          *
4094          * Don't care about POSITION and PSIZE here - this is a builtin vertex shader, position goes
4095          * to TMP_OUT in any case
4096          */
4097         if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_TEXCOORD))
4098         {
4099             if (semantic_idx < 8)
4100                 priv_ctx->texcrd_output[semantic_idx] = decl_idx_to_string[reg_idx];
4101         }
4102         else if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_COLOR))
4103         {
4104             if (semantic_idx < 2)
4105                 priv_ctx->color_output[semantic_idx] = decl_idx_to_string[reg_idx];
4106         }
4107         else if(shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_FOG))
4108         {
4109             if (!semantic_idx)
4110                 priv_ctx->fog_output = decl_idx_to_string[reg_idx];
4111         }
4112         else
4113         {
4114             continue;
4115         }
4116
4117         if (!strcmp(decl_idx_to_string[reg_idx], "result.color.primary")
4118                 || !strcmp(decl_idx_to_string[reg_idx], "result.color.secondary"))
4119         {
4120             compiled->need_color_unclamp = TRUE;
4121         }
4122     }
4123
4124     /* Map declared to declared */
4125     for (i = 0; i < (sizeof(shader->output_signature) / sizeof(*shader->output_signature)); ++i)
4126     {
4127         /* Write unread output to TA to throw them away */
4128         priv_ctx->vs_output[i] = "TA";
4129         semantic_name = shader->output_signature[i].semantic_name;
4130         if (!semantic_name) continue;
4131
4132         if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_POSITION)
4133                 && !shader->output_signature[i].semantic_idx)
4134         {
4135             priv_ctx->vs_output[i] = "TMP_OUT";
4136             continue;
4137         }
4138         else if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_PSIZE)
4139                 && !shader->output_signature[i].semantic_idx)
4140         {
4141             priv_ctx->vs_output[i] = "result.pointsize";
4142             continue;
4143         }
4144
4145         for(j = 0; j < MAX_REG_INPUT; j++)
4146         {
4147             if (!sig[j].semantic_name) continue;
4148
4149             if (!strcmp(sig[j].semantic_name, semantic_name)
4150                     && sig[j].semantic_idx == shader->output_signature[i].semantic_idx)
4151             {
4152                 priv_ctx->vs_output[i] = decl_idx_to_string[sig[j].register_idx];
4153
4154                 if (!strcmp(priv_ctx->vs_output[i], "result.color.primary")
4155                         || !strcmp(priv_ctx->vs_output[i], "result.color.secondary"))
4156                 {
4157                     compiled->need_color_unclamp = TRUE;
4158                 }
4159             }
4160         }
4161     }
4162 }
4163
4164 /* GL locking is done by the caller */
4165 static GLuint shader_arb_generate_vshader(const struct wined3d_shader *shader,
4166         const struct wined3d_gl_info *gl_info, struct wined3d_shader_buffer *buffer,
4167         const struct arb_vs_compile_args *args, struct arb_vs_compiled_shader *compiled)
4168 {
4169     const struct arb_vshader_private *shader_data = shader->backend_data;
4170     const struct wined3d_shader_reg_maps *reg_maps = &shader->reg_maps;
4171     const struct wined3d_shader_lconst *lconst;
4172     const DWORD *function = shader->function;
4173     GLuint ret;
4174     DWORD next_local, *lconst_map = local_const_mapping(shader);
4175     struct shader_arb_ctx_priv priv_ctx;
4176     unsigned int i;
4177     GLint errPos;
4178
4179     memset(&priv_ctx, 0, sizeof(priv_ctx));
4180     priv_ctx.cur_vs_args = args;
4181     list_init(&priv_ctx.control_frames);
4182     init_output_registers(shader, args->ps_signature, &priv_ctx, compiled);
4183
4184     /*  Create the hw ARB shader */
4185     shader_addline(buffer, "!!ARBvp1.0\n");
4186
4187     /* Always enable the NV extension if available. Unlike fragment shaders, there is no
4188      * mesurable performance penalty, and we can always make use of it for clipplanes.
4189      */
4190     if (gl_info->supported[NV_VERTEX_PROGRAM3])
4191     {
4192         shader_addline(buffer, "OPTION NV_vertex_program3;\n");
4193         priv_ctx.target_version = NV3;
4194         shader_addline(buffer, "ADDRESS aL;\n");
4195     }
4196     else if (gl_info->supported[NV_VERTEX_PROGRAM2_OPTION])
4197     {
4198         shader_addline(buffer, "OPTION NV_vertex_program2;\n");
4199         priv_ctx.target_version = NV2;
4200         shader_addline(buffer, "ADDRESS aL;\n");
4201     } else {
4202         priv_ctx.target_version = ARB;
4203     }
4204
4205     shader_addline(buffer, "TEMP TMP_OUT;\n");
4206     if (reg_maps->fog)
4207         shader_addline(buffer, "TEMP TMP_FOGCOORD;\n");
4208     if (need_helper_const(shader_data, reg_maps, gl_info))
4209     {
4210         shader_addline(buffer, "PARAM helper_const = { 0.0, 1.0, 2.0, %1.10f};\n", eps);
4211     }
4212     if (need_rel_addr_const(shader_data, reg_maps, gl_info))
4213     {
4214         shader_addline(buffer, "PARAM rel_addr_const = { 0.5, %d.0, 0.0, 0.0 };\n", shader_data->rel_offset);
4215         shader_addline(buffer, "TEMP A0_SHADOW;\n");
4216     }
4217
4218     shader_addline(buffer, "TEMP TA;\n");
4219     shader_addline(buffer, "TEMP TB;\n");
4220
4221     /* Base Declarations */
4222     next_local = shader_generate_arb_declarations(shader, reg_maps, buffer,
4223             gl_info, lconst_map, &priv_ctx.vs_clipplanes, &priv_ctx);
4224
4225     for(i = 0; i < MAX_CONST_I; i++)
4226     {
4227         compiled->int_consts[i] = WINED3D_CONST_NUM_UNUSED;
4228         if(reg_maps->integer_constants & (1 << i) && priv_ctx.target_version >= NV2)
4229         {
4230             const DWORD *control_values = find_loop_control_values(shader, i);
4231
4232             if(control_values)
4233             {
4234                 shader_addline(buffer, "PARAM I%u = {%u, %u, %u, -1};\n", i,
4235                                 control_values[0], control_values[1], control_values[2]);
4236             }
4237             else
4238             {
4239                 compiled->int_consts[i] = next_local;
4240                 compiled->num_int_consts++;
4241                 shader_addline(buffer, "PARAM I%u = program.local[%u];\n", i, next_local++);
4242             }
4243         }
4244     }
4245
4246     /* We need a constant to fixup the final position */
4247     shader_addline(buffer, "PARAM posFixup = program.local[%u];\n", next_local);
4248     compiled->pos_fixup = next_local++;
4249
4250     /* Initialize output parameters. GL_ARB_vertex_program does not require special initialization values
4251      * for output parameters. D3D in theory does not do that either, but some applications depend on a
4252      * proper initialization of the secondary color, and programs using the fixed function pipeline without
4253      * a replacement shader depend on the texcoord.w being set properly.
4254      *
4255      * GL_NV_vertex_program defines that all output values are initialized to {0.0, 0.0, 0.0, 1.0}. This
4256      * assertion is in effect even when using GL_ARB_vertex_program without any NV specific additions. So
4257      * skip this if NV_vertex_program is supported. Otherwise, initialize the secondary color. For the tex-
4258      * coords, we have a flag in the opengl caps. Many cards do not require the texcoord being set, and
4259      * this can eat a number of instructions, so skip it unless this cap is set as well
4260      */
4261     if (!gl_info->supported[NV_VERTEX_PROGRAM])
4262     {
4263         struct wined3d_device *device = shader->device;
4264         const char *color_init = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_0001);
4265         shader_addline(buffer, "MOV result.color.secondary, %s;\n", color_init);
4266
4267         if (gl_info->quirks & WINED3D_QUIRK_SET_TEXCOORD_W && !device->frag_pipe->ffp_proj_control)
4268         {
4269             int i;
4270             const char *one = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_ONE);
4271             for(i = 0; i < min(8, MAX_REG_TEXCRD); i++)
4272             {
4273                 if (reg_maps->texcoord_mask[i] && reg_maps->texcoord_mask[i] != WINED3DSP_WRITEMASK_ALL)
4274                     shader_addline(buffer, "MOV result.texcoord[%u].w, %s\n", i, one);
4275             }
4276         }
4277     }
4278
4279     /* The shader starts with the main function */
4280     priv_ctx.in_main_func = TRUE;
4281     /* Base Shader Body */
4282     shader_generate_main(shader, buffer, reg_maps, function, &priv_ctx);
4283
4284     if (!priv_ctx.footer_written) vshader_add_footer(&priv_ctx,
4285             shader_data, args, reg_maps, gl_info, buffer);
4286
4287     shader_addline(buffer, "END\n");
4288
4289     /* TODO: change to resource.glObjectHandle or something like that */
4290     GL_EXTCALL(glGenProgramsARB(1, &ret));
4291
4292     TRACE("Creating a hw vertex shader, prg=%d\n", ret);
4293     GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, ret));
4294
4295     TRACE("Created hw vertex shader, prg=%d\n", ret);
4296     /* Create the program and check for errors */
4297     GL_EXTCALL(glProgramStringARB(GL_VERTEX_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
4298                buffer->bsize, buffer->buffer));
4299     checkGLcall("glProgramStringARB()");
4300
4301     gl_info->gl_ops.gl.p_glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &errPos);
4302     if (errPos != -1)
4303     {
4304         FIXME("HW VertexShader Error at position %d: %s\n\n",
4305               errPos, debugstr_a((const char *)gl_info->gl_ops.gl.p_glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
4306         shader_arb_dump_program_source(buffer->buffer);
4307         ret = -1;
4308     }
4309     else
4310     {
4311         GLint native;
4312
4313         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
4314         checkGLcall("glGetProgramivARB()");
4315         if (!native) WARN("Program exceeds native resource limits.\n");
4316
4317         /* Load immediate constants */
4318         if (lconst_map)
4319         {
4320             LIST_FOR_EACH_ENTRY(lconst, &shader->constantsF, struct wined3d_shader_lconst, entry)
4321             {
4322                 const float *value = (const float *)lconst->value;
4323                 GL_EXTCALL(glProgramLocalParameter4fvARB(GL_VERTEX_PROGRAM_ARB, lconst_map[lconst->idx], value));
4324             }
4325         }
4326     }
4327     HeapFree(GetProcessHeap(), 0, lconst_map);
4328
4329     return ret;
4330 }
4331
4332 /* GL locking is done by the caller */
4333 static struct arb_ps_compiled_shader *find_arb_pshader(struct wined3d_shader *shader,
4334         const struct arb_ps_compile_args *args)
4335 {
4336     struct wined3d_device *device = shader->device;
4337     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4338     UINT i;
4339     DWORD new_size;
4340     struct arb_ps_compiled_shader *new_array;
4341     struct wined3d_shader_buffer buffer;
4342     struct arb_pshader_private *shader_data;
4343     GLuint ret;
4344
4345     if (!shader->backend_data)
4346     {
4347         struct shader_arb_priv *priv = device->shader_priv;
4348
4349         shader->backend_data = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data));
4350         shader_data = shader->backend_data;
4351         shader_data->clamp_consts = shader->reg_maps.shader_version.major == 1;
4352
4353         if (shader->reg_maps.shader_version.major < 3)
4354             shader_data->input_signature_idx = ~0;
4355         else
4356             shader_data->input_signature_idx = find_input_signature(priv, shader->input_signature);
4357
4358         TRACE("Shader got assigned input signature index %u\n", shader_data->input_signature_idx);
4359
4360         if (!device->vs_clipping)
4361             shader_data->clipplane_emulation = shader_find_free_input_register(&shader->reg_maps,
4362                     gl_info->limits.texture_stages - 1);
4363         else
4364             shader_data->clipplane_emulation = ~0U;
4365     }
4366     shader_data = shader->backend_data;
4367
4368     /* Usually we have very few GL shaders for each d3d shader(just 1 or maybe 2),
4369      * so a linear search is more performant than a hashmap or a binary search
4370      * (cache coherency etc)
4371      */
4372     for (i = 0; i < shader_data->num_gl_shaders; ++i)
4373     {
4374         if (!memcmp(&shader_data->gl_shaders[i].args, args, sizeof(*args)))
4375             return &shader_data->gl_shaders[i];
4376     }
4377
4378     TRACE("No matching GL shader found, compiling a new shader\n");
4379     if(shader_data->shader_array_size == shader_data->num_gl_shaders) {
4380         if (shader_data->num_gl_shaders)
4381         {
4382             new_size = shader_data->shader_array_size + max(1, shader_data->shader_array_size / 2);
4383             new_array = HeapReAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, shader_data->gl_shaders,
4384                                     new_size * sizeof(*shader_data->gl_shaders));
4385         } else {
4386             new_array = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data->gl_shaders));
4387             new_size = 1;
4388         }
4389
4390         if(!new_array) {
4391             ERR("Out of memory\n");
4392             return 0;
4393         }
4394         shader_data->gl_shaders = new_array;
4395         shader_data->shader_array_size = new_size;
4396     }
4397
4398     shader_data->gl_shaders[shader_data->num_gl_shaders].args = *args;
4399
4400     pixelshader_update_samplers(&shader->reg_maps, device->stateBlock->state.textures);
4401
4402     if (!shader_buffer_init(&buffer))
4403     {
4404         ERR("Failed to initialize shader buffer.\n");
4405         return 0;
4406     }
4407
4408     ret = shader_arb_generate_pshader(shader, gl_info, &buffer, args,
4409             &shader_data->gl_shaders[shader_data->num_gl_shaders]);
4410     shader_buffer_free(&buffer);
4411     shader_data->gl_shaders[shader_data->num_gl_shaders].prgId = ret;
4412
4413     return &shader_data->gl_shaders[shader_data->num_gl_shaders++];
4414 }
4415
4416 static inline BOOL vs_args_equal(const struct arb_vs_compile_args *stored, const struct arb_vs_compile_args *new,
4417                                  const DWORD use_map, BOOL skip_int) {
4418     if((stored->super.swizzle_map & use_map) != new->super.swizzle_map) return FALSE;
4419     if(stored->super.clip_enabled != new->super.clip_enabled) return FALSE;
4420     if(stored->super.fog_src != new->super.fog_src) return FALSE;
4421     if(stored->clip.boolclip_compare != new->clip.boolclip_compare) return FALSE;
4422     if(stored->ps_signature != new->ps_signature) return FALSE;
4423     if(stored->vertex.samplers_compare != new->vertex.samplers_compare) return FALSE;
4424     if(skip_int) return TRUE;
4425
4426     return !memcmp(stored->loop_ctrl, new->loop_ctrl, sizeof(stored->loop_ctrl));
4427 }
4428
4429 static struct arb_vs_compiled_shader *find_arb_vshader(struct wined3d_shader *shader,
4430         const struct arb_vs_compile_args *args)
4431 {
4432     struct wined3d_device *device = shader->device;
4433     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4434     DWORD use_map = device->strided_streams.use_map;
4435     UINT i;
4436     DWORD new_size;
4437     struct arb_vs_compiled_shader *new_array;
4438     struct wined3d_shader_buffer buffer;
4439     struct arb_vshader_private *shader_data;
4440     GLuint ret;
4441
4442     if (!shader->backend_data)
4443     {
4444         const struct wined3d_shader_reg_maps *reg_maps = &shader->reg_maps;
4445
4446         shader->backend_data = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data));
4447         shader_data = shader->backend_data;
4448
4449         if ((gl_info->quirks & WINED3D_QUIRK_ARB_VS_OFFSET_LIMIT)
4450                 && reg_maps->min_rel_offset <= reg_maps->max_rel_offset)
4451         {
4452             if (reg_maps->max_rel_offset - reg_maps->min_rel_offset > 127)
4453             {
4454                 FIXME("The difference between the minimum and maximum relative offset is > 127.\n");
4455                 FIXME("Which this OpenGL implementation does not support. Try using GLSL.\n");
4456                 FIXME("Min: %u, Max: %u.\n", reg_maps->min_rel_offset, reg_maps->max_rel_offset);
4457             }
4458             else if (reg_maps->max_rel_offset - reg_maps->min_rel_offset > 63)
4459                 shader_data->rel_offset = reg_maps->min_rel_offset + 63;
4460             else if (reg_maps->max_rel_offset > 63)
4461                 shader_data->rel_offset = reg_maps->min_rel_offset;
4462         }
4463     }
4464     shader_data = shader->backend_data;
4465
4466     /* Usually we have very few GL shaders for each d3d shader(just 1 or maybe 2),
4467      * so a linear search is more performant than a hashmap or a binary search
4468      * (cache coherency etc)
4469      */
4470     for(i = 0; i < shader_data->num_gl_shaders; i++) {
4471         if (vs_args_equal(&shader_data->gl_shaders[i].args, args,
4472                 use_map, gl_info->supported[NV_VERTEX_PROGRAM2_OPTION]))
4473         {
4474             return &shader_data->gl_shaders[i];
4475         }
4476     }
4477
4478     TRACE("No matching GL shader found, compiling a new shader\n");
4479
4480     if(shader_data->shader_array_size == shader_data->num_gl_shaders) {
4481         if (shader_data->num_gl_shaders)
4482         {
4483             new_size = shader_data->shader_array_size + max(1, shader_data->shader_array_size / 2);
4484             new_array = HeapReAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, shader_data->gl_shaders,
4485                                     new_size * sizeof(*shader_data->gl_shaders));
4486         } else {
4487             new_array = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data->gl_shaders));
4488             new_size = 1;
4489         }
4490
4491         if(!new_array) {
4492             ERR("Out of memory\n");
4493             return 0;
4494         }
4495         shader_data->gl_shaders = new_array;
4496         shader_data->shader_array_size = new_size;
4497     }
4498
4499     shader_data->gl_shaders[shader_data->num_gl_shaders].args = *args;
4500
4501     if (!shader_buffer_init(&buffer))
4502     {
4503         ERR("Failed to initialize shader buffer.\n");
4504         return 0;
4505     }
4506
4507     ret = shader_arb_generate_vshader(shader, gl_info, &buffer, args,
4508             &shader_data->gl_shaders[shader_data->num_gl_shaders]);
4509     shader_buffer_free(&buffer);
4510     shader_data->gl_shaders[shader_data->num_gl_shaders].prgId = ret;
4511
4512     return &shader_data->gl_shaders[shader_data->num_gl_shaders++];
4513 }
4514
4515 static void find_arb_ps_compile_args(const struct wined3d_state *state,
4516         const struct wined3d_shader *shader, struct arb_ps_compile_args *args)
4517 {
4518     struct wined3d_device *device = shader->device;
4519     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4520     int i;
4521     WORD int_skip;
4522
4523     find_ps_compile_args(state, shader, &args->super);
4524
4525     /* This forces all local boolean constants to 1 to make them stateblock independent */
4526     args->bools = shader->reg_maps.local_bool_consts;
4527
4528     for(i = 0; i < MAX_CONST_B; i++)
4529     {
4530         if (state->ps_consts_b[i])
4531             args->bools |= ( 1 << i);
4532     }
4533
4534     /* Only enable the clip plane emulation KIL if at least one clipplane is enabled. The KIL instruction
4535      * is quite expensive because it forces the driver to disable early Z discards. It is cheaper to
4536      * duplicate the shader than have a no-op KIL instruction in every shader
4537      */
4538     if (!device->vs_clipping && use_vs(state)
4539             && state->render_states[WINED3D_RS_CLIPPING]
4540             && state->render_states[WINED3D_RS_CLIPPLANEENABLE])
4541         args->clip = 1;
4542     else
4543         args->clip = 0;
4544
4545     /* Skip if unused or local, or supported natively */
4546     int_skip = ~shader->reg_maps.integer_constants | shader->reg_maps.local_int_consts;
4547     if (int_skip == 0xffff || gl_info->supported[NV_FRAGMENT_PROGRAM_OPTION])
4548     {
4549         memset(args->loop_ctrl, 0, sizeof(args->loop_ctrl));
4550         return;
4551     }
4552
4553     for(i = 0; i < MAX_CONST_I; i++)
4554     {
4555         if(int_skip & (1 << i))
4556         {
4557             args->loop_ctrl[i][0] = 0;
4558             args->loop_ctrl[i][1] = 0;
4559             args->loop_ctrl[i][2] = 0;
4560         }
4561         else
4562         {
4563             args->loop_ctrl[i][0] = state->ps_consts_i[i * 4];
4564             args->loop_ctrl[i][1] = state->ps_consts_i[i * 4 + 1];
4565             args->loop_ctrl[i][2] = state->ps_consts_i[i * 4 + 2];
4566         }
4567     }
4568 }
4569
4570 static void find_arb_vs_compile_args(const struct wined3d_state *state,
4571         const struct wined3d_shader *shader, struct arb_vs_compile_args *args)
4572 {
4573     struct wined3d_device *device = shader->device;
4574     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4575     int i;
4576     WORD int_skip;
4577
4578     find_vs_compile_args(state, shader, &args->super);
4579
4580     args->clip.boolclip_compare = 0;
4581     if (use_ps(state))
4582     {
4583         const struct wined3d_shader *ps = state->pixel_shader;
4584         const struct arb_pshader_private *shader_priv = ps->backend_data;
4585         args->ps_signature = shader_priv->input_signature_idx;
4586
4587         args->clip.boolclip.clip_texcoord = shader_priv->clipplane_emulation + 1;
4588     }
4589     else
4590     {
4591         args->ps_signature = ~0;
4592         if (!device->vs_clipping && device->adapter->fragment_pipe == &arbfp_fragment_pipeline)
4593         {
4594             args->clip.boolclip.clip_texcoord = ffp_clip_emul(state) ? gl_info->limits.texture_stages : 0;
4595         }
4596         /* Otherwise: Setting boolclip_compare set clip_texcoord to 0 */
4597     }
4598
4599     if (args->clip.boolclip.clip_texcoord)
4600     {
4601         if (state->render_states[WINED3D_RS_CLIPPING])
4602             args->clip.boolclip.clipplane_mask = (unsigned char)state->render_states[WINED3D_RS_CLIPPLANEENABLE];
4603         /* clipplane_mask was set to 0 by setting boolclip_compare to 0 */
4604     }
4605
4606     /* This forces all local boolean constants to 1 to make them stateblock independent */
4607     args->clip.boolclip.bools = shader->reg_maps.local_bool_consts;
4608     /* TODO: Figure out if it would be better to store bool constants as bitmasks in the stateblock */
4609     for(i = 0; i < MAX_CONST_B; i++)
4610     {
4611         if (state->vs_consts_b[i])
4612             args->clip.boolclip.bools |= ( 1 << i);
4613     }
4614
4615     args->vertex.samplers[0] = device->texUnitMap[MAX_FRAGMENT_SAMPLERS + 0];
4616     args->vertex.samplers[1] = device->texUnitMap[MAX_FRAGMENT_SAMPLERS + 1];
4617     args->vertex.samplers[2] = device->texUnitMap[MAX_FRAGMENT_SAMPLERS + 2];
4618     args->vertex.samplers[3] = 0;
4619
4620     /* Skip if unused or local */
4621     int_skip = ~shader->reg_maps.integer_constants | shader->reg_maps.local_int_consts;
4622     /* This is about flow control, not clipping. */
4623     if (int_skip == 0xffff || gl_info->supported[NV_VERTEX_PROGRAM2_OPTION])
4624     {
4625         memset(args->loop_ctrl, 0, sizeof(args->loop_ctrl));
4626         return;
4627     }
4628
4629     for(i = 0; i < MAX_CONST_I; i++)
4630     {
4631         if(int_skip & (1 << i))
4632         {
4633             args->loop_ctrl[i][0] = 0;
4634             args->loop_ctrl[i][1] = 0;
4635             args->loop_ctrl[i][2] = 0;
4636         }
4637         else
4638         {
4639             args->loop_ctrl[i][0] = state->vs_consts_i[i * 4];
4640             args->loop_ctrl[i][1] = state->vs_consts_i[i * 4 + 1];
4641             args->loop_ctrl[i][2] = state->vs_consts_i[i * 4 + 2];
4642         }
4643     }
4644 }
4645
4646 /* GL locking is done by the caller */
4647 static void shader_arb_select(const struct wined3d_context *context, BOOL usePS, BOOL useVS)
4648 {
4649     struct wined3d_device *device = context->swapchain->device;
4650     struct shader_arb_priv *priv = device->shader_priv;
4651     const struct wined3d_gl_info *gl_info = context->gl_info;
4652     const struct wined3d_state *state = &device->stateBlock->state;
4653     int i;
4654
4655     /* Deal with pixel shaders first so the vertex shader arg function has the input signature ready */
4656     if (usePS)
4657     {
4658         struct wined3d_shader *ps = state->pixel_shader;
4659         struct arb_ps_compile_args compile_args;
4660         struct arb_ps_compiled_shader *compiled;
4661
4662         TRACE("Using pixel shader %p.\n", ps);
4663         find_arb_ps_compile_args(state, ps, &compile_args);
4664         compiled = find_arb_pshader(ps, &compile_args);
4665         priv->current_fprogram_id = compiled->prgId;
4666         priv->compiled_fprog = compiled;
4667
4668         /* Bind the fragment program */
4669         GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, priv->current_fprogram_id));
4670         checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, priv->current_fprogram_id);");
4671
4672         if (!priv->use_arbfp_fixed_func)
4673         {
4674             /* Enable OpenGL fragment programs. */
4675             gl_info->gl_ops.gl.p_glEnable(GL_FRAGMENT_PROGRAM_ARB);
4676             checkGLcall("glEnable(GL_FRAGMENT_PROGRAM_ARB);");
4677         }
4678         TRACE("(%p) : Bound fragment program %u and enabled GL_FRAGMENT_PROGRAM_ARB\n",
4679                 device, priv->current_fprogram_id);
4680
4681         /* Pixel Shader 1.x constants are clamped to [-1;1], Pixel Shader 2.0 constants are not. If switching between
4682          * a 1.x and newer shader, reload the first 8 constants
4683          */
4684         if (priv->last_ps_const_clamped != ((struct arb_pshader_private *)ps->backend_data)->clamp_consts)
4685         {
4686             priv->last_ps_const_clamped = ((struct arb_pshader_private *)ps->backend_data)->clamp_consts;
4687             priv->highest_dirty_ps_const = max(priv->highest_dirty_ps_const, 8);
4688             for(i = 0; i < 8; i++)
4689             {
4690                 priv->pshader_const_dirty[i] = 1;
4691             }
4692             /* Also takes care of loading local constants */
4693             shader_arb_load_constants_internal(context, TRUE, FALSE, TRUE);
4694         }
4695         else
4696         {
4697             UINT rt_height = state->fb->render_targets[0]->resource.height;
4698             shader_arb_ps_local_constants(compiled, context, state, rt_height);
4699         }
4700
4701         /* Force constant reloading for the NP2 fixup (see comment in shader_glsl_select for more info) */
4702         if (compiled->np2fixup_info.super.active)
4703             shader_arb_load_np2fixup_constants(priv, gl_info, state);
4704     }
4705     else if (gl_info->supported[ARB_FRAGMENT_PROGRAM] && !priv->use_arbfp_fixed_func)
4706     {
4707         /* Disable only if we're not using arbfp fixed function fragment processing. If this is used,
4708         * keep GL_FRAGMENT_PROGRAM_ARB enabled, and the fixed function pipeline will bind the fixed function
4709         * replacement shader
4710         */
4711         gl_info->gl_ops.gl.p_glDisable(GL_FRAGMENT_PROGRAM_ARB);
4712         checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
4713         priv->current_fprogram_id = 0;
4714     }
4715
4716     if (useVS)
4717     {
4718         struct wined3d_shader *vs = state->vertex_shader;
4719         struct arb_vs_compile_args compile_args;
4720         struct arb_vs_compiled_shader *compiled;
4721
4722         TRACE("Using vertex shader %p\n", vs);
4723         find_arb_vs_compile_args(state, vs, &compile_args);
4724         compiled = find_arb_vshader(vs, &compile_args);
4725         priv->current_vprogram_id = compiled->prgId;
4726         priv->compiled_vprog = compiled;
4727
4728         /* Bind the vertex program */
4729         GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->current_vprogram_id));
4730         checkGLcall("glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->current_vprogram_id);");
4731
4732         /* Enable OpenGL vertex programs */
4733         gl_info->gl_ops.gl.p_glEnable(GL_VERTEX_PROGRAM_ARB);
4734         checkGLcall("glEnable(GL_VERTEX_PROGRAM_ARB);");
4735         TRACE("(%p) : Bound vertex program %u and enabled GL_VERTEX_PROGRAM_ARB\n", device, priv->current_vprogram_id);
4736         shader_arb_vs_local_constants(compiled, context, state);
4737
4738         if(priv->last_vs_color_unclamp != compiled->need_color_unclamp) {
4739             priv->last_vs_color_unclamp = compiled->need_color_unclamp;
4740
4741             if (gl_info->supported[ARB_COLOR_BUFFER_FLOAT])
4742             {
4743                 GL_EXTCALL(glClampColorARB(GL_CLAMP_VERTEX_COLOR_ARB, !compiled->need_color_unclamp));
4744                 checkGLcall("glClampColorARB");
4745             } else {
4746                 FIXME("vertex color clamp needs to be changed, but extension not supported.\n");
4747             }
4748         }
4749     }
4750     else if (gl_info->supported[ARB_VERTEX_PROGRAM])
4751     {
4752         priv->current_vprogram_id = 0;
4753         gl_info->gl_ops.gl.p_glDisable(GL_VERTEX_PROGRAM_ARB);
4754         checkGLcall("glDisable(GL_VERTEX_PROGRAM_ARB)");
4755     }
4756 }
4757
4758 /* GL locking is done by the caller */
4759 static void shader_arb_select_depth_blt(void *shader_priv, const struct wined3d_gl_info *gl_info,
4760         enum tex_types tex_type, const SIZE *ds_mask_size)
4761 {
4762     const float mask[] = {0.0f, 0.0f, (float)ds_mask_size->cx, (float)ds_mask_size->cy};
4763     BOOL masked = ds_mask_size->cx && ds_mask_size->cy;
4764     struct shader_arb_priv *priv = shader_priv;
4765     GLuint *blt_fprogram;
4766
4767     if (!priv->depth_blt_vprogram_id) priv->depth_blt_vprogram_id = create_arb_blt_vertex_program(gl_info);
4768     GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->depth_blt_vprogram_id));
4769     gl_info->gl_ops.gl.p_glEnable(GL_VERTEX_PROGRAM_ARB);
4770
4771     blt_fprogram = masked ? &priv->depth_blt_fprogram_id_masked[tex_type] : &priv->depth_blt_fprogram_id_full[tex_type];
4772     if (!*blt_fprogram) *blt_fprogram = create_arb_blt_fragment_program(gl_info, tex_type, masked);
4773     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, *blt_fprogram));
4774     if (masked) GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, 0, mask));
4775     gl_info->gl_ops.gl.p_glEnable(GL_FRAGMENT_PROGRAM_ARB);
4776 }
4777
4778 /* GL locking is done by the caller */
4779 static void shader_arb_deselect_depth_blt(void *shader_priv, const struct wined3d_gl_info *gl_info)
4780 {
4781     struct shader_arb_priv *priv = shader_priv;
4782
4783     if (priv->current_vprogram_id) {
4784         GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->current_vprogram_id));
4785         checkGLcall("glBindProgramARB(GL_VERTEX_PROGRAM_ARB, vertexShader->prgId);");
4786
4787         TRACE("Bound vertex program %u and enabled GL_VERTEX_PROGRAM_ARB.\n", priv->current_vprogram_id);
4788     }
4789     else
4790     {
4791         gl_info->gl_ops.gl.p_glDisable(GL_VERTEX_PROGRAM_ARB);
4792         checkGLcall("glDisable(GL_VERTEX_PROGRAM_ARB)");
4793     }
4794
4795     if (priv->current_fprogram_id) {
4796         GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, priv->current_fprogram_id));
4797         checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, pixelShader->prgId);");
4798
4799         TRACE("Bound fragment program %u and enabled GL_FRAGMENT_PROGRAM_ARB.\n", priv->current_fprogram_id);
4800     }
4801     else if(!priv->use_arbfp_fixed_func)
4802     {
4803         gl_info->gl_ops.gl.p_glDisable(GL_FRAGMENT_PROGRAM_ARB);
4804         checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
4805     }
4806 }
4807
4808 static void shader_arb_destroy(struct wined3d_shader *shader)
4809 {
4810     struct wined3d_device *device = shader->device;
4811     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4812
4813     if (shader_is_pshader_version(shader->reg_maps.shader_version.type))
4814     {
4815         struct arb_pshader_private *shader_data = shader->backend_data;
4816         UINT i;
4817
4818         if(!shader_data) return; /* This can happen if a shader was never compiled */
4819
4820         if (shader_data->num_gl_shaders)
4821         {
4822             struct wined3d_context *context = context_acquire(device, NULL);
4823
4824             ENTER_GL();
4825             for (i = 0; i < shader_data->num_gl_shaders; ++i)
4826             {
4827                 GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId));
4828                 checkGLcall("GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId))");
4829             }
4830             LEAVE_GL();
4831
4832             context_release(context);
4833         }
4834
4835         HeapFree(GetProcessHeap(), 0, shader_data->gl_shaders);
4836         HeapFree(GetProcessHeap(), 0, shader_data);
4837         shader->backend_data = NULL;
4838     }
4839     else
4840     {
4841         struct arb_vshader_private *shader_data = shader->backend_data;
4842         UINT i;
4843
4844         if(!shader_data) return; /* This can happen if a shader was never compiled */
4845
4846         if (shader_data->num_gl_shaders)
4847         {
4848             struct wined3d_context *context = context_acquire(device, NULL);
4849
4850             ENTER_GL();
4851             for (i = 0; i < shader_data->num_gl_shaders; ++i)
4852             {
4853                 GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId));
4854                 checkGLcall("GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId))");
4855             }
4856             LEAVE_GL();
4857
4858             context_release(context);
4859         }
4860
4861         HeapFree(GetProcessHeap(), 0, shader_data->gl_shaders);
4862         HeapFree(GetProcessHeap(), 0, shader_data);
4863         shader->backend_data = NULL;
4864     }
4865 }
4866
4867 static int sig_tree_compare(const void *key, const struct wine_rb_entry *entry)
4868 {
4869     struct ps_signature *e = WINE_RB_ENTRY_VALUE(entry, struct ps_signature, entry);
4870     return compare_sig(key, e->sig);
4871 }
4872
4873 static const struct wine_rb_functions sig_tree_functions =
4874 {
4875     wined3d_rb_alloc,
4876     wined3d_rb_realloc,
4877     wined3d_rb_free,
4878     sig_tree_compare
4879 };
4880
4881 static HRESULT shader_arb_alloc(struct wined3d_device *device)
4882 {
4883     struct shader_arb_priv *priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*priv));
4884
4885     priv->vshader_const_dirty = HeapAlloc(GetProcessHeap(), 0,
4886             sizeof(*priv->vshader_const_dirty) * device->d3d_vshader_constantF);
4887     if (!priv->vshader_const_dirty)
4888         goto fail;
4889     memset(priv->vshader_const_dirty, 1,
4890            sizeof(*priv->vshader_const_dirty) * device->d3d_vshader_constantF);
4891
4892     priv->pshader_const_dirty = HeapAlloc(GetProcessHeap(), 0,
4893             sizeof(*priv->pshader_const_dirty) * device->d3d_pshader_constantF);
4894     if (!priv->pshader_const_dirty)
4895         goto fail;
4896     memset(priv->pshader_const_dirty, 1,
4897             sizeof(*priv->pshader_const_dirty) * device->d3d_pshader_constantF);
4898
4899     if(wine_rb_init(&priv->signature_tree, &sig_tree_functions) == -1)
4900     {
4901         ERR("RB tree init failed\n");
4902         goto fail;
4903     }
4904     device->shader_priv = priv;
4905     return WINED3D_OK;
4906
4907 fail:
4908     HeapFree(GetProcessHeap(), 0, priv->pshader_const_dirty);
4909     HeapFree(GetProcessHeap(), 0, priv->vshader_const_dirty);
4910     HeapFree(GetProcessHeap(), 0, priv);
4911     return E_OUTOFMEMORY;
4912 }
4913
4914 static void release_signature(struct wine_rb_entry *entry, void *context)
4915 {
4916     struct ps_signature *sig = WINE_RB_ENTRY_VALUE(entry, struct ps_signature, entry);
4917     int i;
4918     for(i = 0; i < MAX_REG_INPUT; i++)
4919     {
4920         HeapFree(GetProcessHeap(), 0, (char *) sig->sig[i].semantic_name);
4921     }
4922     HeapFree(GetProcessHeap(), 0, sig->sig);
4923     HeapFree(GetProcessHeap(), 0, sig);
4924 }
4925
4926 /* Context activation is done by the caller. */
4927 static void shader_arb_free(struct wined3d_device *device)
4928 {
4929     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4930     struct shader_arb_priv *priv = device->shader_priv;
4931     int i;
4932
4933     ENTER_GL();
4934     if(priv->depth_blt_vprogram_id) {
4935         GL_EXTCALL(glDeleteProgramsARB(1, &priv->depth_blt_vprogram_id));
4936     }
4937     for (i = 0; i < tex_type_count; ++i)
4938     {
4939         if (priv->depth_blt_fprogram_id_full[i])
4940         {
4941             GL_EXTCALL(glDeleteProgramsARB(1, &priv->depth_blt_fprogram_id_full[i]));
4942         }
4943         if (priv->depth_blt_fprogram_id_masked[i])
4944         {
4945             GL_EXTCALL(glDeleteProgramsARB(1, &priv->depth_blt_fprogram_id_masked[i]));
4946         }
4947     }
4948     LEAVE_GL();
4949
4950     wine_rb_destroy(&priv->signature_tree, release_signature, NULL);
4951     HeapFree(GetProcessHeap(), 0, priv->pshader_const_dirty);
4952     HeapFree(GetProcessHeap(), 0, priv->vshader_const_dirty);
4953     HeapFree(GetProcessHeap(), 0, device->shader_priv);
4954 }
4955
4956 static void shader_arb_context_destroyed(void *shader_priv, const struct wined3d_context *context)
4957 {
4958     struct shader_arb_priv *priv = shader_priv;
4959
4960     if (priv->last_context == context)
4961         priv->last_context = NULL;
4962 }
4963
4964 static void shader_arb_get_caps(const struct wined3d_gl_info *gl_info, struct shader_caps *caps)
4965 {
4966     if (gl_info->supported[ARB_VERTEX_PROGRAM])
4967     {
4968         DWORD vs_consts;
4969
4970         /* 96 is the minimum allowed value of MAX_PROGRAM_ENV_PARAMETERS_ARB
4971          * for vertex programs. If the native limit is less than that it's
4972          * not very useful, and e.g. Mesa swrast returns 0, probably to
4973          * indicate it's a software implementation. */
4974         if (gl_info->limits.arb_vs_native_constants < 96)
4975             vs_consts = gl_info->limits.arb_vs_float_constants;
4976         else
4977             vs_consts = min(gl_info->limits.arb_vs_float_constants, gl_info->limits.arb_vs_native_constants);
4978
4979         if (gl_info->supported[NV_VERTEX_PROGRAM3])
4980         {
4981             caps->VertexShaderVersion = 3;
4982             TRACE("Hardware vertex shader version 3.0 enabled (NV_VERTEX_PROGRAM3)\n");
4983         }
4984         else if (vs_consts >= 256)
4985         {
4986             /* Shader Model 2.0 requires at least 256 vertex shader constants */
4987             caps->VertexShaderVersion = 2;
4988             TRACE("Hardware vertex shader version 2.0 enabled (ARB_PROGRAM)\n");
4989         }
4990         else
4991         {
4992             caps->VertexShaderVersion = 1;
4993             TRACE("Hardware vertex shader version 1.1 enabled (ARB_PROGRAM)\n");
4994         }
4995         caps->MaxVertexShaderConst = vs_consts;
4996     }
4997     else
4998     {
4999         caps->VertexShaderVersion = 0;
5000         caps->MaxVertexShaderConst = 0;
5001     }
5002
5003     if (gl_info->supported[ARB_FRAGMENT_PROGRAM])
5004     {
5005         DWORD ps_consts;
5006
5007         /* Similar as above for vertex programs, but the minimum for fragment
5008          * programs is 24. */
5009         if (gl_info->limits.arb_ps_native_constants < 24)
5010             ps_consts = gl_info->limits.arb_ps_float_constants;
5011         else
5012             ps_consts = min(gl_info->limits.arb_ps_float_constants, gl_info->limits.arb_ps_native_constants);
5013
5014         if (gl_info->supported[NV_FRAGMENT_PROGRAM2])
5015         {
5016             caps->PixelShaderVersion = 3;
5017             TRACE("Hardware pixel shader version 3.0 enabled (NV_FRAGMENT_PROGRAM2)\n");
5018         }
5019         else if (ps_consts >= 32)
5020         {
5021             /* Shader Model 2.0 requires at least 32 pixel shader constants */
5022             caps->PixelShaderVersion = 2;
5023             TRACE("Hardware pixel shader version 2.0 enabled (ARB_PROGRAM)\n");
5024         }
5025         else
5026         {
5027             caps->PixelShaderVersion = 1;
5028             TRACE("Hardware pixel shader version 1.4 enabled (ARB_PROGRAM)\n");
5029         }
5030         caps->PixelShader1xMaxValue = 8.0f;
5031         caps->MaxPixelShaderConst = ps_consts;
5032     }
5033     else
5034     {
5035         caps->PixelShaderVersion = 0;
5036         caps->PixelShader1xMaxValue = 0.0f;
5037         caps->MaxPixelShaderConst = 0;
5038     }
5039
5040     caps->VSClipping = use_nv_clip(gl_info);
5041 }
5042
5043 static BOOL shader_arb_color_fixup_supported(struct color_fixup_desc fixup)
5044 {
5045     if (TRACE_ON(d3d_shader) && TRACE_ON(d3d))
5046     {
5047         TRACE("Checking support for color_fixup:\n");
5048         dump_color_fixup_desc(fixup);
5049     }
5050
5051     /* We support everything except complex conversions. */
5052     if (!is_complex_fixup(fixup))
5053     {
5054         TRACE("[OK]\n");
5055         return TRUE;
5056     }
5057
5058     TRACE("[FAILED]\n");
5059     return FALSE;
5060 }
5061
5062 static void shader_arb_add_instruction_modifiers(const struct wined3d_shader_instruction *ins) {
5063     DWORD shift;
5064     char write_mask[20], regstr[50];
5065     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
5066     BOOL is_color = FALSE;
5067     const struct wined3d_shader_dst_param *dst;
5068
5069     if (!ins->dst_count) return;
5070
5071     dst = &ins->dst[0];
5072     shift = dst->shift;
5073     if (!shift) return; /* Saturate alone is handled by the instructions */
5074
5075     shader_arb_get_write_mask(ins, dst, write_mask);
5076     shader_arb_get_register_name(ins, &dst->reg, regstr, &is_color);
5077
5078     /* Generate a line that does the output modifier computation
5079      * FIXME: _SAT vs shift? _SAT alone is already handled in the instructions, if this
5080      * maps problems in e.g. _d4_sat modify shader_arb_get_modifier
5081      */
5082     shader_addline(buffer, "MUL%s %s%s, %s, %s;\n", shader_arb_get_modifier(ins),
5083                    regstr, write_mask, regstr, shift_tab[shift]);
5084 }
5085
5086 static const SHADER_HANDLER shader_arb_instruction_handler_table[WINED3DSIH_TABLE_SIZE] =
5087 {
5088     /* WINED3DSIH_ABS                   */ shader_hw_map2gl,
5089     /* WINED3DSIH_ADD                   */ shader_hw_map2gl,
5090     /* WINED3DSIH_AND                   */ NULL,
5091     /* WINED3DSIH_BEM                   */ pshader_hw_bem,
5092     /* WINED3DSIH_BREAK                 */ shader_hw_break,
5093     /* WINED3DSIH_BREAKC                */ shader_hw_breakc,
5094     /* WINED3DSIH_BREAKP                */ NULL,
5095     /* WINED3DSIH_CALL                  */ shader_hw_call,
5096     /* WINED3DSIH_CALLNZ                */ NULL,
5097     /* WINED3DSIH_CMP                   */ pshader_hw_cmp,
5098     /* WINED3DSIH_CND                   */ pshader_hw_cnd,
5099     /* WINED3DSIH_CRS                   */ shader_hw_map2gl,
5100     /* WINED3DSIH_CUT                   */ NULL,
5101     /* WINED3DSIH_DCL                   */ shader_hw_nop,
5102     /* WINED3DSIH_DCL_INPUT_PRIMITIVE   */ shader_hw_nop,
5103     /* WINED3DSIH_DCL_OUTPUT_TOPOLOGY   */ shader_hw_nop,
5104     /* WINED3DSIH_DCL_VERTICES_OUT      */ shader_hw_nop,
5105     /* WINED3DSIH_DEF                   */ shader_hw_nop,
5106     /* WINED3DSIH_DEFB                  */ shader_hw_nop,
5107     /* WINED3DSIH_DEFI                  */ shader_hw_nop,
5108     /* WINED3DSIH_DIV                   */ NULL,
5109     /* WINED3DSIH_DP2ADD                */ pshader_hw_dp2add,
5110     /* WINED3DSIH_DP3                   */ shader_hw_map2gl,
5111     /* WINED3DSIH_DP4                   */ shader_hw_map2gl,
5112     /* WINED3DSIH_DST                   */ shader_hw_map2gl,
5113     /* WINED3DSIH_DSX                   */ shader_hw_map2gl,
5114     /* WINED3DSIH_DSY                   */ shader_hw_dsy,
5115     /* WINED3DSIH_ELSE                  */ shader_hw_else,
5116     /* WINED3DSIH_EMIT                  */ NULL,
5117     /* WINED3DSIH_ENDIF                 */ shader_hw_endif,
5118     /* WINED3DSIH_ENDLOOP               */ shader_hw_endloop,
5119     /* WINED3DSIH_ENDREP                */ shader_hw_endrep,
5120     /* WINED3DSIH_EQ                    */ NULL,
5121     /* WINED3DSIH_EXP                   */ shader_hw_scalar_op,
5122     /* WINED3DSIH_EXPP                  */ shader_hw_scalar_op,
5123     /* WINED3DSIH_FRC                   */ shader_hw_map2gl,
5124     /* WINED3DSIH_FTOI                  */ NULL,
5125     /* WINED3DSIH_GE                    */ NULL,
5126     /* WINED3DSIH_IADD                  */ NULL,
5127     /* WINED3DSIH_IEQ                   */ NULL,
5128     /* WINED3DSIH_IF                    */ NULL /* Hardcoded into the shader */,
5129     /* WINED3DSIH_IFC                   */ shader_hw_ifc,
5130     /* WINED3DSIH_IGE                   */ NULL,
5131     /* WINED3DSIH_IMUL                  */ NULL,
5132     /* WINED3DSIH_ITOF                  */ NULL,
5133     /* WINED3DSIH_LABEL                 */ shader_hw_label,
5134     /* WINED3DSIH_LD                    */ NULL,
5135     /* WINED3DSIH_LIT                   */ shader_hw_map2gl,
5136     /* WINED3DSIH_LOG                   */ shader_hw_log,
5137     /* WINED3DSIH_LOGP                  */ shader_hw_log,
5138     /* WINED3DSIH_LOOP                  */ shader_hw_loop,
5139     /* WINED3DSIH_LRP                   */ shader_hw_lrp,
5140     /* WINED3DSIH_LT                    */ NULL,
5141     /* WINED3DSIH_M3x2                  */ shader_hw_mnxn,
5142     /* WINED3DSIH_M3x3                  */ shader_hw_mnxn,
5143     /* WINED3DSIH_M3x4                  */ shader_hw_mnxn,
5144     /* WINED3DSIH_M4x3                  */ shader_hw_mnxn,
5145     /* WINED3DSIH_M4x4                  */ shader_hw_mnxn,
5146     /* WINED3DSIH_MAD                   */ shader_hw_map2gl,
5147     /* WINED3DSIH_MAX                   */ shader_hw_map2gl,
5148     /* WINED3DSIH_MIN                   */ shader_hw_map2gl,
5149     /* WINED3DSIH_MOV                   */ shader_hw_mov,
5150     /* WINED3DSIH_MOVA                  */ shader_hw_mov,
5151     /* WINED3DSIH_MOVC                  */ NULL,
5152     /* WINED3DSIH_MUL                   */ shader_hw_map2gl,
5153     /* WINED3DSIH_NOP                   */ shader_hw_nop,
5154     /* WINED3DSIH_NRM                   */ shader_hw_nrm,
5155     /* WINED3DSIH_PHASE                 */ shader_hw_nop,
5156     /* WINED3DSIH_POW                   */ shader_hw_pow,
5157     /* WINED3DSIH_RCP                   */ shader_hw_rcp,
5158     /* WINED3DSIH_REP                   */ shader_hw_rep,
5159     /* WINED3DSIH_RET                   */ shader_hw_ret,
5160     /* WINED3DSIH_ROUND_NI              */ NULL,
5161     /* WINED3DSIH_RSQ                   */ shader_hw_scalar_op,
5162     /* WINED3DSIH_SAMPLE                */ NULL,
5163     /* WINED3DSIH_SAMPLE_GRAD           */ NULL,
5164     /* WINED3DSIH_SAMPLE_LOD            */ NULL,
5165     /* WINED3DSIH_SETP                  */ NULL,
5166     /* WINED3DSIH_SGE                   */ shader_hw_map2gl,
5167     /* WINED3DSIH_SGN                   */ shader_hw_sgn,
5168     /* WINED3DSIH_SINCOS                */ shader_hw_sincos,
5169     /* WINED3DSIH_SLT                   */ shader_hw_map2gl,
5170     /* WINED3DSIH_SQRT                  */ NULL,
5171     /* WINED3DSIH_SUB                   */ shader_hw_map2gl,
5172     /* WINED3DSIH_TEX                   */ pshader_hw_tex,
5173     /* WINED3DSIH_TEXBEM                */ pshader_hw_texbem,
5174     /* WINED3DSIH_TEXBEML               */ pshader_hw_texbem,
5175     /* WINED3DSIH_TEXCOORD              */ pshader_hw_texcoord,
5176     /* WINED3DSIH_TEXDEPTH              */ pshader_hw_texdepth,
5177     /* WINED3DSIH_TEXDP3                */ pshader_hw_texdp3,
5178     /* WINED3DSIH_TEXDP3TEX             */ pshader_hw_texdp3tex,
5179     /* WINED3DSIH_TEXKILL               */ pshader_hw_texkill,
5180     /* WINED3DSIH_TEXLDD                */ shader_hw_texldd,
5181     /* WINED3DSIH_TEXLDL                */ shader_hw_texldl,
5182     /* WINED3DSIH_TEXM3x2DEPTH          */ pshader_hw_texm3x2depth,
5183     /* WINED3DSIH_TEXM3x2PAD            */ pshader_hw_texm3x2pad,
5184     /* WINED3DSIH_TEXM3x2TEX            */ pshader_hw_texm3x2tex,
5185     /* WINED3DSIH_TEXM3x3               */ pshader_hw_texm3x3,
5186     /* WINED3DSIH_TEXM3x3DIFF           */ NULL,
5187     /* WINED3DSIH_TEXM3x3PAD            */ pshader_hw_texm3x3pad,
5188     /* WINED3DSIH_TEXM3x3SPEC           */ pshader_hw_texm3x3spec,
5189     /* WINED3DSIH_TEXM3x3TEX            */ pshader_hw_texm3x3tex,
5190     /* WINED3DSIH_TEXM3x3VSPEC          */ pshader_hw_texm3x3vspec,
5191     /* WINED3DSIH_TEXREG2AR             */ pshader_hw_texreg2ar,
5192     /* WINED3DSIH_TEXREG2GB             */ pshader_hw_texreg2gb,
5193     /* WINED3DSIH_TEXREG2RGB            */ pshader_hw_texreg2rgb,
5194     /* WINED3DSIH_UDIV                  */ NULL,
5195     /* WINED3DSIH_USHR                  */ NULL,
5196     /* WINED3DSIH_UTOF                  */ NULL,
5197     /* WINED3DSIH_XOR                   */ NULL,
5198 };
5199
5200 static BOOL get_bool_const(const struct wined3d_shader_instruction *ins,
5201         const struct wined3d_shader *shader, DWORD idx)
5202 {
5203     const struct wined3d_shader_reg_maps *reg_maps = ins->ctx->reg_maps;
5204     BOOL vshader = shader_is_vshader_version(reg_maps->shader_version.type);
5205     const struct wined3d_shader_lconst *constant;
5206     WORD bools = 0;
5207     WORD flag = (1 << idx);
5208     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
5209
5210     if (reg_maps->local_bool_consts & flag)
5211     {
5212         /* What good is a if(bool) with a hardcoded local constant? I don't know, but handle it */
5213         LIST_FOR_EACH_ENTRY(constant, &shader->constantsB, struct wined3d_shader_lconst, entry)
5214         {
5215             if (constant->idx == idx)
5216             {
5217                 return constant->value[0];
5218             }
5219         }
5220         ERR("Local constant not found\n");
5221         return FALSE;
5222     }
5223     else
5224     {
5225         if(vshader) bools = priv->cur_vs_args->clip.boolclip.bools;
5226         else bools = priv->cur_ps_args->bools;
5227         return bools & flag;
5228     }
5229 }
5230
5231 static void get_loop_control_const(const struct wined3d_shader_instruction *ins,
5232         const struct wined3d_shader *shader, UINT idx, struct wined3d_shader_loop_control *loop_control)
5233 {
5234     const struct wined3d_shader_reg_maps *reg_maps = ins->ctx->reg_maps;
5235     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
5236
5237     /* Integer constants can either be a local constant, or they can be stored in the shader
5238      * type specific compile args. */
5239     if (reg_maps->local_int_consts & (1 << idx))
5240     {
5241         const struct wined3d_shader_lconst *constant;
5242
5243         LIST_FOR_EACH_ENTRY(constant, &shader->constantsI, struct wined3d_shader_lconst, entry)
5244         {
5245             if (constant->idx == idx)
5246             {
5247                 loop_control->count = constant->value[0];
5248                 loop_control->start = constant->value[1];
5249                 /* Step is signed. */
5250                 loop_control->step = (int)constant->value[2];
5251                 return;
5252             }
5253         }
5254         /* If this happens the flag was set incorrectly */
5255         ERR("Local constant not found\n");
5256         loop_control->count = 0;
5257         loop_control->start = 0;
5258         loop_control->step = 0;
5259         return;
5260     }
5261
5262     switch (reg_maps->shader_version.type)
5263     {
5264         case WINED3D_SHADER_TYPE_VERTEX:
5265             /* Count and aL start value are unsigned */
5266             loop_control->count = priv->cur_vs_args->loop_ctrl[idx][0];
5267             loop_control->start = priv->cur_vs_args->loop_ctrl[idx][1];
5268             /* Step is signed. */
5269             loop_control->step = ((char)priv->cur_vs_args->loop_ctrl[idx][2]);
5270             break;
5271
5272         case WINED3D_SHADER_TYPE_PIXEL:
5273             loop_control->count = priv->cur_ps_args->loop_ctrl[idx][0];
5274             loop_control->start = priv->cur_ps_args->loop_ctrl[idx][1];
5275             loop_control->step = ((char)priv->cur_ps_args->loop_ctrl[idx][2]);
5276             break;
5277
5278         default:
5279             FIXME("Unhandled shader type %#x.\n", reg_maps->shader_version.type);
5280             break;
5281     }
5282 }
5283
5284 static void record_instruction(struct list *list, const struct wined3d_shader_instruction *ins)
5285 {
5286     unsigned int i;
5287     struct wined3d_shader_dst_param *dst_param = NULL;
5288     struct wined3d_shader_src_param *src_param = NULL, *rel_addr = NULL;
5289     struct recorded_instruction *rec = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*rec));
5290     if(!rec)
5291     {
5292         ERR("Out of memory\n");
5293         return;
5294     }
5295
5296     rec->ins = *ins;
5297     dst_param = HeapAlloc(GetProcessHeap(), 0, sizeof(*dst_param));
5298     if(!dst_param) goto free;
5299     *dst_param = *ins->dst;
5300     if(ins->dst->reg.rel_addr)
5301     {
5302         rel_addr = HeapAlloc(GetProcessHeap(), 0, sizeof(*dst_param->reg.rel_addr));
5303         if(!rel_addr) goto free;
5304         *rel_addr = *ins->dst->reg.rel_addr;
5305         dst_param->reg.rel_addr = rel_addr;
5306     }
5307     rec->ins.dst = dst_param;
5308
5309     src_param = HeapAlloc(GetProcessHeap(), 0, sizeof(*src_param) * ins->src_count);
5310     if(!src_param) goto free;
5311     for(i = 0; i < ins->src_count; i++)
5312     {
5313         src_param[i] = ins->src[i];
5314         if(ins->src[i].reg.rel_addr)
5315         {
5316             rel_addr = HeapAlloc(GetProcessHeap(), 0, sizeof(*rel_addr));
5317             if(!rel_addr) goto free;
5318             *rel_addr = *ins->src[i].reg.rel_addr;
5319             src_param[i].reg.rel_addr = rel_addr;
5320         }
5321     }
5322     rec->ins.src = src_param;
5323     list_add_tail(list, &rec->entry);
5324     return;
5325
5326 free:
5327     ERR("Out of memory\n");
5328     if(dst_param)
5329     {
5330         HeapFree(GetProcessHeap(), 0, (void *) dst_param->reg.rel_addr);
5331         HeapFree(GetProcessHeap(), 0, dst_param);
5332     }
5333     if(src_param)
5334     {
5335         for(i = 0; i < ins->src_count; i++)
5336         {
5337             HeapFree(GetProcessHeap(), 0, (void *) src_param[i].reg.rel_addr);
5338         }
5339         HeapFree(GetProcessHeap(), 0, src_param);
5340     }
5341     HeapFree(GetProcessHeap(), 0, rec);
5342 }
5343
5344 static void free_recorded_instruction(struct list *list)
5345 {
5346     struct recorded_instruction *rec_ins, *entry2;
5347     unsigned int i;
5348
5349     LIST_FOR_EACH_ENTRY_SAFE(rec_ins, entry2, list, struct recorded_instruction, entry)
5350     {
5351         list_remove(&rec_ins->entry);
5352         if(rec_ins->ins.dst)
5353         {
5354             HeapFree(GetProcessHeap(), 0, (void *) rec_ins->ins.dst->reg.rel_addr);
5355             HeapFree(GetProcessHeap(), 0, (void *) rec_ins->ins.dst);
5356         }
5357         if(rec_ins->ins.src)
5358         {
5359             for(i = 0; i < rec_ins->ins.src_count; i++)
5360             {
5361                 HeapFree(GetProcessHeap(), 0, (void *) rec_ins->ins.src[i].reg.rel_addr);
5362             }
5363             HeapFree(GetProcessHeap(), 0, (void *) rec_ins->ins.src);
5364         }
5365         HeapFree(GetProcessHeap(), 0, rec_ins);
5366     }
5367 }
5368
5369 static void shader_arb_handle_instruction(const struct wined3d_shader_instruction *ins) {
5370     SHADER_HANDLER hw_fct;
5371     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
5372     const struct wined3d_shader *shader = ins->ctx->shader;
5373     struct control_frame *control_frame;
5374     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
5375     BOOL bool_const;
5376
5377     if(ins->handler_idx == WINED3DSIH_LOOP || ins->handler_idx == WINED3DSIH_REP)
5378     {
5379         control_frame = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*control_frame));
5380         list_add_head(&priv->control_frames, &control_frame->entry);
5381
5382         if(ins->handler_idx == WINED3DSIH_LOOP) control_frame->type = LOOP;
5383         if(ins->handler_idx == WINED3DSIH_REP) control_frame->type = REP;
5384
5385         if(priv->target_version >= NV2)
5386         {
5387             control_frame->no.loop = priv->num_loops++;
5388             priv->loop_depth++;
5389         }
5390         else
5391         {
5392             /* Don't bother recording when we're in a not used if branch */
5393             if(priv->muted)
5394             {
5395                 return;
5396             }
5397
5398             if(!priv->recording)
5399             {
5400                 list_init(&priv->record);
5401                 priv->recording = TRUE;
5402                 control_frame->outer_loop = TRUE;
5403                 get_loop_control_const(ins, shader, ins->src[0].reg.idx, &control_frame->loop_control);
5404                 return; /* Instruction is handled */
5405             }
5406             /* Record this loop in the outer loop's recording */
5407         }
5408     }
5409     else if(ins->handler_idx == WINED3DSIH_ENDLOOP || ins->handler_idx == WINED3DSIH_ENDREP)
5410     {
5411         if(priv->target_version >= NV2)
5412         {
5413             /* Nothing to do. The control frame is popped after the HW instr handler */
5414         }
5415         else
5416         {
5417             struct list *e = list_head(&priv->control_frames);
5418             control_frame = LIST_ENTRY(e, struct control_frame, entry);
5419             list_remove(&control_frame->entry);
5420
5421             if(control_frame->outer_loop)
5422             {
5423                 unsigned int iteration;
5424                 int aL = 0;
5425                 struct list copy;
5426
5427                 /* Turn off recording before playback */
5428                 priv->recording = FALSE;
5429
5430                 /* Move the recorded instructions to a separate list and get them out of the private data
5431                  * structure. If there are nested loops, the shader_arb_handle_instruction below will
5432                  * be recorded again, thus priv->record might be overwritten
5433                  */
5434                 list_init(&copy);
5435                 list_move_tail(&copy, &priv->record);
5436                 list_init(&priv->record);
5437
5438                 if(ins->handler_idx == WINED3DSIH_ENDLOOP)
5439                 {
5440                     shader_addline(buffer, "#unrolling loop: %u iterations, aL=%u, inc %d\n",
5441                                    control_frame->loop_control.count, control_frame->loop_control.start,
5442                                    control_frame->loop_control.step);
5443                     aL = control_frame->loop_control.start;
5444                 }
5445                 else
5446                 {
5447                     shader_addline(buffer, "#unrolling rep: %u iterations\n", control_frame->loop_control.count);
5448                 }
5449
5450                 for (iteration = 0; iteration < control_frame->loop_control.count; ++iteration)
5451                 {
5452                     struct recorded_instruction *rec_ins;
5453                     if(ins->handler_idx == WINED3DSIH_ENDLOOP)
5454                     {
5455                         priv->aL = aL;
5456                         shader_addline(buffer, "#Iteration %u, aL=%d\n", iteration, aL);
5457                     }
5458                     else
5459                     {
5460                         shader_addline(buffer, "#Iteration %u\n", iteration);
5461                     }
5462
5463                     LIST_FOR_EACH_ENTRY(rec_ins, &copy, struct recorded_instruction, entry)
5464                     {
5465                         shader_arb_handle_instruction(&rec_ins->ins);
5466                     }
5467
5468                     if(ins->handler_idx == WINED3DSIH_ENDLOOP)
5469                     {
5470                         aL += control_frame->loop_control.step;
5471                     }
5472                 }
5473                 shader_addline(buffer, "#end loop/rep\n");
5474
5475                 free_recorded_instruction(&copy);
5476                 HeapFree(GetProcessHeap(), 0, control_frame);
5477                 return; /* Instruction is handled */
5478             }
5479             else
5480             {
5481                 /* This is a nested loop. Proceed to the normal recording function */
5482                 HeapFree(GetProcessHeap(), 0, control_frame);
5483             }
5484         }
5485     }
5486
5487     if(priv->recording)
5488     {
5489         record_instruction(&priv->record, ins);
5490         return;
5491     }
5492
5493     /* boolean if */
5494     if(ins->handler_idx == WINED3DSIH_IF)
5495     {
5496         control_frame = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*control_frame));
5497         list_add_head(&priv->control_frames, &control_frame->entry);
5498         control_frame->type = IF;
5499
5500         bool_const = get_bool_const(ins, shader, ins->src[0].reg.idx);
5501         if(ins->src[0].modifiers == WINED3DSPSM_NOT) bool_const = !bool_const;
5502         if (!priv->muted && !bool_const)
5503         {
5504             shader_addline(buffer, "#if(FALSE){\n");
5505             priv->muted = TRUE;
5506             control_frame->muting = TRUE;
5507         }
5508         else shader_addline(buffer, "#if(TRUE) {\n");
5509
5510         return; /* Instruction is handled */
5511     }
5512     else if(ins->handler_idx == WINED3DSIH_IFC)
5513     {
5514         /* IF(bool) and if_cond(a, b) use the same ELSE and ENDIF tokens */
5515         control_frame = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*control_frame));
5516         control_frame->type = IFC;
5517         control_frame->no.ifc = priv->num_ifcs++;
5518         list_add_head(&priv->control_frames, &control_frame->entry);
5519     }
5520     else if(ins->handler_idx == WINED3DSIH_ELSE)
5521     {
5522         struct list *e = list_head(&priv->control_frames);
5523         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5524
5525         if(control_frame->type == IF)
5526         {
5527             shader_addline(buffer, "#} else {\n");
5528             if(!priv->muted && !control_frame->muting)
5529             {
5530                 priv->muted = TRUE;
5531                 control_frame->muting = TRUE;
5532             }
5533             else if(control_frame->muting) priv->muted = FALSE;
5534             return; /* Instruction is handled. */
5535         }
5536         /* In case of an ifc, generate a HW shader instruction */
5537     }
5538     else if(ins->handler_idx == WINED3DSIH_ENDIF)
5539     {
5540         struct list *e = list_head(&priv->control_frames);
5541         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5542
5543         if(control_frame->type == IF)
5544         {
5545             shader_addline(buffer, "#} endif\n");
5546             if(control_frame->muting) priv->muted = FALSE;
5547             list_remove(&control_frame->entry);
5548             HeapFree(GetProcessHeap(), 0, control_frame);
5549             return; /* Instruction is handled */
5550         }
5551     }
5552
5553     if(priv->muted) return;
5554
5555     /* Select handler */
5556     hw_fct = shader_arb_instruction_handler_table[ins->handler_idx];
5557
5558     /* Unhandled opcode */
5559     if (!hw_fct)
5560     {
5561         FIXME("Backend can't handle opcode %#x\n", ins->handler_idx);
5562         return;
5563     }
5564     hw_fct(ins);
5565
5566     if(ins->handler_idx == WINED3DSIH_ENDLOOP || ins->handler_idx == WINED3DSIH_ENDREP)
5567     {
5568         struct list *e = list_head(&priv->control_frames);
5569         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5570         list_remove(&control_frame->entry);
5571         HeapFree(GetProcessHeap(), 0, control_frame);
5572         priv->loop_depth--;
5573     }
5574     else if(ins->handler_idx == WINED3DSIH_ENDIF)
5575     {
5576         /* Non-ifc ENDIFs don't reach that place because of the return in the if block above */
5577         struct list *e = list_head(&priv->control_frames);
5578         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5579         list_remove(&control_frame->entry);
5580         HeapFree(GetProcessHeap(), 0, control_frame);
5581     }
5582
5583
5584     shader_arb_add_instruction_modifiers(ins);
5585 }
5586
5587 const struct wined3d_shader_backend_ops arb_program_shader_backend =
5588 {
5589     shader_arb_handle_instruction,
5590     shader_arb_select,
5591     shader_arb_select_depth_blt,
5592     shader_arb_deselect_depth_blt,
5593     shader_arb_update_float_vertex_constants,
5594     shader_arb_update_float_pixel_constants,
5595     shader_arb_load_constants,
5596     shader_arb_load_np2fixup_constants,
5597     shader_arb_destroy,
5598     shader_arb_alloc,
5599     shader_arb_free,
5600     shader_arb_context_destroyed,
5601     shader_arb_get_caps,
5602     shader_arb_color_fixup_supported,
5603 };
5604
5605 /* ARB_fragment_program fixed function pipeline replacement definitions */
5606 #define ARB_FFP_CONST_TFACTOR           0
5607 #define ARB_FFP_CONST_SPECULAR_ENABLE   ((ARB_FFP_CONST_TFACTOR) + 1)
5608 #define ARB_FFP_CONST_CONSTANT(i)       ((ARB_FFP_CONST_SPECULAR_ENABLE) + 1 + i)
5609 #define ARB_FFP_CONST_BUMPMAT(i)        ((ARB_FFP_CONST_CONSTANT(7)) + 1 + i)
5610 #define ARB_FFP_CONST_LUMINANCE(i)      ((ARB_FFP_CONST_BUMPMAT(7)) + 1 + i)
5611
5612 struct arbfp_ffp_desc
5613 {
5614     struct ffp_frag_desc parent;
5615     GLuint shader;
5616     unsigned int num_textures_used;
5617 };
5618
5619 /* Context activation and GL locking are done by the caller. */
5620 static void arbfp_enable(const struct wined3d_gl_info *gl_info, BOOL enable)
5621 {
5622     if (enable)
5623     {
5624         gl_info->gl_ops.gl.p_glEnable(GL_FRAGMENT_PROGRAM_ARB);
5625         checkGLcall("glEnable(GL_FRAGMENT_PROGRAM_ARB)");
5626     }
5627     else
5628     {
5629         gl_info->gl_ops.gl.p_glDisable(GL_FRAGMENT_PROGRAM_ARB);
5630         checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
5631     }
5632 }
5633
5634 static HRESULT arbfp_alloc(struct wined3d_device *device)
5635 {
5636     struct shader_arb_priv *priv;
5637     /* Share private data between the shader backend and the pipeline replacement, if both
5638      * are the arb implementation. This is needed to figure out whether ARBfp should be disabled
5639      * if no pixel shader is bound or not
5640      */
5641     if (device->shader_backend == &arb_program_shader_backend)
5642     {
5643         device->fragment_priv = device->shader_priv;
5644     }
5645     else
5646     {
5647         device->fragment_priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(struct shader_arb_priv));
5648         if (!device->fragment_priv) return E_OUTOFMEMORY;
5649     }
5650     priv = device->fragment_priv;
5651     if (wine_rb_init(&priv->fragment_shaders, &wined3d_ffp_frag_program_rb_functions) == -1)
5652     {
5653         ERR("Failed to initialize rbtree.\n");
5654         HeapFree(GetProcessHeap(), 0, device->fragment_priv);
5655         return E_OUTOFMEMORY;
5656     }
5657     priv->use_arbfp_fixed_func = TRUE;
5658     return WINED3D_OK;
5659 }
5660
5661 /* Context activation is done by the caller. */
5662 static void arbfp_free_ffpshader(struct wine_rb_entry *entry, void *context)
5663 {
5664     const struct wined3d_gl_info *gl_info = context;
5665     struct arbfp_ffp_desc *entry_arb = WINE_RB_ENTRY_VALUE(entry, struct arbfp_ffp_desc, parent.entry);
5666
5667     ENTER_GL();
5668     GL_EXTCALL(glDeleteProgramsARB(1, &entry_arb->shader));
5669     checkGLcall("glDeleteProgramsARB(1, &entry_arb->shader)");
5670     HeapFree(GetProcessHeap(), 0, entry_arb);
5671     LEAVE_GL();
5672 }
5673
5674 /* Context activation is done by the caller. */
5675 static void arbfp_free(struct wined3d_device *device)
5676 {
5677     struct shader_arb_priv *priv = device->fragment_priv;
5678
5679     wine_rb_destroy(&priv->fragment_shaders, arbfp_free_ffpshader, &device->adapter->gl_info);
5680     priv->use_arbfp_fixed_func = FALSE;
5681
5682     if (device->shader_backend != &arb_program_shader_backend)
5683     {
5684         HeapFree(GetProcessHeap(), 0, device->fragment_priv);
5685     }
5686 }
5687
5688 static void arbfp_get_caps(const struct wined3d_gl_info *gl_info, struct fragment_caps *caps)
5689 {
5690     caps->PrimitiveMiscCaps = WINED3DPMISCCAPS_TSSARGTEMP;
5691     caps->TextureOpCaps =  WINED3DTEXOPCAPS_DISABLE                     |
5692                            WINED3DTEXOPCAPS_SELECTARG1                  |
5693                            WINED3DTEXOPCAPS_SELECTARG2                  |
5694                            WINED3DTEXOPCAPS_MODULATE4X                  |
5695                            WINED3DTEXOPCAPS_MODULATE2X                  |
5696                            WINED3DTEXOPCAPS_MODULATE                    |
5697                            WINED3DTEXOPCAPS_ADDSIGNED2X                 |
5698                            WINED3DTEXOPCAPS_ADDSIGNED                   |
5699                            WINED3DTEXOPCAPS_ADD                         |
5700                            WINED3DTEXOPCAPS_SUBTRACT                    |
5701                            WINED3DTEXOPCAPS_ADDSMOOTH                   |
5702                            WINED3DTEXOPCAPS_BLENDCURRENTALPHA           |
5703                            WINED3DTEXOPCAPS_BLENDFACTORALPHA            |
5704                            WINED3DTEXOPCAPS_BLENDTEXTUREALPHA           |
5705                            WINED3DTEXOPCAPS_BLENDDIFFUSEALPHA           |
5706                            WINED3DTEXOPCAPS_BLENDTEXTUREALPHAPM         |
5707                            WINED3DTEXOPCAPS_MODULATEALPHA_ADDCOLOR      |
5708                            WINED3DTEXOPCAPS_MODULATECOLOR_ADDALPHA      |
5709                            WINED3DTEXOPCAPS_MODULATEINVCOLOR_ADDALPHA   |
5710                            WINED3DTEXOPCAPS_MODULATEINVALPHA_ADDCOLOR   |
5711                            WINED3DTEXOPCAPS_DOTPRODUCT3                 |
5712                            WINED3DTEXOPCAPS_MULTIPLYADD                 |
5713                            WINED3DTEXOPCAPS_LERP                        |
5714                            WINED3DTEXOPCAPS_BUMPENVMAP                  |
5715                            WINED3DTEXOPCAPS_BUMPENVMAPLUMINANCE;
5716
5717     /* TODO: Implement WINED3DTEXOPCAPS_PREMODULATE */
5718
5719     caps->MaxTextureBlendStages   = 8;
5720     caps->MaxSimultaneousTextures = min(gl_info->limits.fragment_samplers, 8);
5721 }
5722
5723 static void state_texfactor_arbfp(struct wined3d_context *context,
5724         const struct wined3d_state *state, DWORD state_id)
5725 {
5726     struct wined3d_device *device = context->swapchain->device;
5727     const struct wined3d_gl_info *gl_info = context->gl_info;
5728     float col[4];
5729
5730     /* Don't load the parameter if we're using an arbfp pixel shader,
5731      * otherwise we'll overwrite application provided constants. */
5732     if (device->shader_backend == &arb_program_shader_backend)
5733     {
5734         struct shader_arb_priv *priv;
5735
5736         if (use_ps(state)) return;
5737
5738         priv = device->shader_priv;
5739         priv->pshader_const_dirty[ARB_FFP_CONST_TFACTOR] = 1;
5740         priv->highest_dirty_ps_const = max(priv->highest_dirty_ps_const, ARB_FFP_CONST_TFACTOR + 1);
5741     }
5742
5743     D3DCOLORTOGLFLOAT4(state->render_states[WINED3D_RS_TEXTUREFACTOR], col);
5744     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_TFACTOR, col));
5745     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_TFACTOR, col)");
5746 }
5747
5748 static void state_arb_specularenable(struct wined3d_context *context,
5749         const struct wined3d_state *state, DWORD state_id)
5750 {
5751     struct wined3d_device *device = context->swapchain->device;
5752     const struct wined3d_gl_info *gl_info = context->gl_info;
5753     float col[4];
5754
5755     /* Don't load the parameter if we're using an arbfp pixel shader, otherwise we'll overwrite
5756      * application provided constants
5757      */
5758     if (device->shader_backend == &arb_program_shader_backend)
5759     {
5760         struct shader_arb_priv *priv;
5761
5762         if (use_ps(state)) return;
5763
5764         priv = device->shader_priv;
5765         priv->pshader_const_dirty[ARB_FFP_CONST_SPECULAR_ENABLE] = 1;
5766         priv->highest_dirty_ps_const = max(priv->highest_dirty_ps_const, ARB_FFP_CONST_SPECULAR_ENABLE + 1);
5767     }
5768
5769     if (state->render_states[WINED3D_RS_SPECULARENABLE])
5770     {
5771         /* The specular color has no alpha */
5772         col[0] = 1.0f; col[1] = 1.0f;
5773         col[2] = 1.0f; col[3] = 0.0f;
5774     } else {
5775         col[0] = 0.0f; col[1] = 0.0f;
5776         col[2] = 0.0f; col[3] = 0.0f;
5777     }
5778     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_SPECULAR_ENABLE, col));
5779     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_SPECULAR_ENABLE, col)");
5780 }
5781
5782 static void set_bumpmat_arbfp(struct wined3d_context *context, const struct wined3d_state *state, DWORD state_id)
5783 {
5784     DWORD stage = (state_id - STATE_TEXTURESTAGE(0, 0)) / (WINED3D_HIGHEST_TEXTURE_STATE + 1);
5785     struct wined3d_device *device = context->swapchain->device;
5786     const struct wined3d_gl_info *gl_info = context->gl_info;
5787     float mat[2][2];
5788
5789     if (use_ps(state))
5790     {
5791         if (stage && (state->pixel_shader->reg_maps.bumpmat & (1 << stage)))
5792         {
5793             /* The pixel shader has to know the bump env matrix. Do a constants update if it isn't scheduled
5794              * anyway
5795              */
5796             if (!isStateDirty(context, STATE_PIXELSHADERCONSTANT))
5797                 context_apply_state(context, state, STATE_PIXELSHADERCONSTANT);
5798         }
5799
5800         if(device->shader_backend == &arb_program_shader_backend) {
5801             /* Exit now, don't set the bumpmat below, otherwise we may overwrite pixel shader constants */
5802             return;
5803         }
5804     }
5805     else if (device->shader_backend == &arb_program_shader_backend)
5806     {
5807         struct shader_arb_priv *priv = device->shader_priv;
5808         priv->pshader_const_dirty[ARB_FFP_CONST_BUMPMAT(stage)] = 1;
5809         priv->highest_dirty_ps_const = max(priv->highest_dirty_ps_const, ARB_FFP_CONST_BUMPMAT(stage) + 1);
5810     }
5811
5812     mat[0][0] = *((float *)&state->texture_states[stage][WINED3D_TSS_BUMPENV_MAT00]);
5813     mat[0][1] = *((float *)&state->texture_states[stage][WINED3D_TSS_BUMPENV_MAT01]);
5814     mat[1][0] = *((float *)&state->texture_states[stage][WINED3D_TSS_BUMPENV_MAT10]);
5815     mat[1][1] = *((float *)&state->texture_states[stage][WINED3D_TSS_BUMPENV_MAT11]);
5816
5817     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_BUMPMAT(stage), &mat[0][0]));
5818     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_BUMPMAT(stage), &mat[0][0])");
5819 }
5820
5821 static void tex_bumpenvlum_arbfp(struct wined3d_context *context,
5822         const struct wined3d_state *state, DWORD state_id)
5823 {
5824     DWORD stage = (state_id - STATE_TEXTURESTAGE(0, 0)) / (WINED3D_HIGHEST_TEXTURE_STATE + 1);
5825     struct wined3d_device *device = context->swapchain->device;
5826     const struct wined3d_gl_info *gl_info = context->gl_info;
5827     float param[4];
5828
5829     if (use_ps(state))
5830     {
5831         if (stage && (state->pixel_shader->reg_maps.luminanceparams & (1 << stage)))
5832         {
5833             /* The pixel shader has to know the luminance offset. Do a constants update if it
5834              * isn't scheduled anyway
5835              */
5836             if (!isStateDirty(context, STATE_PIXELSHADERCONSTANT))
5837                 context_apply_state(context, state, STATE_PIXELSHADERCONSTANT);
5838         }
5839
5840         if(device->shader_backend == &arb_program_shader_backend) {
5841             /* Exit now, don't set the bumpmat below, otherwise we may overwrite pixel shader constants */
5842             return;
5843         }
5844     }
5845     else if (device->shader_backend == &arb_program_shader_backend)
5846     {
5847         struct shader_arb_priv *priv = device->shader_priv;
5848         priv->pshader_const_dirty[ARB_FFP_CONST_LUMINANCE(stage)] = 1;
5849         priv->highest_dirty_ps_const = max(priv->highest_dirty_ps_const, ARB_FFP_CONST_LUMINANCE(stage) + 1);
5850     }
5851
5852     param[0] = *((float *)&state->texture_states[stage][WINED3D_TSS_BUMPENV_LSCALE]);
5853     param[1] = *((float *)&state->texture_states[stage][WINED3D_TSS_BUMPENV_LOFFSET]);
5854     param[2] = 0.0f;
5855     param[3] = 0.0f;
5856
5857     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_LUMINANCE(stage), param));
5858     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_LUMINANCE(stage), param)");
5859 }
5860
5861 static const char *get_argreg(struct wined3d_shader_buffer *buffer, DWORD argnum, unsigned int stage, DWORD arg)
5862 {
5863     const char *ret;
5864
5865     if(arg == ARG_UNUSED) return "unused"; /* This is the marker for unused registers */
5866
5867     switch(arg & WINED3DTA_SELECTMASK) {
5868         case WINED3DTA_DIFFUSE:
5869             ret = "fragment.color.primary"; break;
5870
5871         case WINED3DTA_CURRENT:
5872             if (!stage) ret = "fragment.color.primary";
5873             else ret = "ret";
5874             break;
5875
5876         case WINED3DTA_TEXTURE:
5877             switch(stage) {
5878                 case 0: ret = "tex0"; break;
5879                 case 1: ret = "tex1"; break;
5880                 case 2: ret = "tex2"; break;
5881                 case 3: ret = "tex3"; break;
5882                 case 4: ret = "tex4"; break;
5883                 case 5: ret = "tex5"; break;
5884                 case 6: ret = "tex6"; break;
5885                 case 7: ret = "tex7"; break;
5886                 default: ret = "unknown texture";
5887             }
5888             break;
5889
5890         case WINED3DTA_TFACTOR:
5891             ret = "tfactor"; break;
5892
5893         case WINED3DTA_SPECULAR:
5894             ret = "fragment.color.secondary"; break;
5895
5896         case WINED3DTA_TEMP:
5897             ret = "tempreg"; break;
5898
5899         case WINED3DTA_CONSTANT:
5900             FIXME("Implement perstage constants\n");
5901             switch(stage) {
5902                 case 0: ret = "const0"; break;
5903                 case 1: ret = "const1"; break;
5904                 case 2: ret = "const2"; break;
5905                 case 3: ret = "const3"; break;
5906                 case 4: ret = "const4"; break;
5907                 case 5: ret = "const5"; break;
5908                 case 6: ret = "const6"; break;
5909                 case 7: ret = "const7"; break;
5910                 default: ret = "unknown constant";
5911             }
5912             break;
5913
5914         default:
5915             return "unknown";
5916     }
5917
5918     if(arg & WINED3DTA_COMPLEMENT) {
5919         shader_addline(buffer, "SUB arg%u, const.x, %s;\n", argnum, ret);
5920         if(argnum == 0) ret = "arg0";
5921         if(argnum == 1) ret = "arg1";
5922         if(argnum == 2) ret = "arg2";
5923     }
5924     if(arg & WINED3DTA_ALPHAREPLICATE) {
5925         shader_addline(buffer, "MOV arg%u, %s.w;\n", argnum, ret);
5926         if(argnum == 0) ret = "arg0";
5927         if(argnum == 1) ret = "arg1";
5928         if(argnum == 2) ret = "arg2";
5929     }
5930     return ret;
5931 }
5932
5933 static void gen_ffp_instr(struct wined3d_shader_buffer *buffer, unsigned int stage, BOOL color,
5934         BOOL alpha, DWORD dst, DWORD op, DWORD dw_arg0, DWORD dw_arg1, DWORD dw_arg2)
5935 {
5936     const char *dstmask, *dstreg, *arg0, *arg1, *arg2;
5937     unsigned int mul = 1;
5938     BOOL mul_final_dest = FALSE;
5939
5940     if(color && alpha) dstmask = "";
5941     else if(color) dstmask = ".xyz";
5942     else dstmask = ".w";
5943
5944     if(dst == tempreg) dstreg = "tempreg";
5945     else dstreg = "ret";
5946
5947     arg0 = get_argreg(buffer, 0, stage, dw_arg0);
5948     arg1 = get_argreg(buffer, 1, stage, dw_arg1);
5949     arg2 = get_argreg(buffer, 2, stage, dw_arg2);
5950
5951     switch (op)
5952     {
5953         case WINED3D_TOP_DISABLE:
5954             if (!stage)
5955                 shader_addline(buffer, "MOV %s%s, fragment.color.primary;\n", dstreg, dstmask);
5956             break;
5957
5958         case WINED3D_TOP_SELECT_ARG2:
5959             arg1 = arg2;
5960             /* FALLTHROUGH */
5961         case WINED3D_TOP_SELECT_ARG1:
5962             shader_addline(buffer, "MOV %s%s, %s;\n", dstreg, dstmask, arg1);
5963             break;
5964
5965         case WINED3D_TOP_MODULATE_4X:
5966             mul = 2;
5967             /* FALLTHROUGH */
5968         case WINED3D_TOP_MODULATE_2X:
5969             mul *= 2;
5970             if (!strcmp(dstreg, "result.color"))
5971             {
5972                 dstreg = "ret";
5973                 mul_final_dest = TRUE;
5974             }
5975             /* FALLTHROUGH */
5976         case WINED3D_TOP_MODULATE:
5977             shader_addline(buffer, "MUL %s%s, %s, %s;\n", dstreg, dstmask, arg1, arg2);
5978             break;
5979
5980         case WINED3D_TOP_ADD_SIGNED_2X:
5981             mul = 2;
5982             if (!strcmp(dstreg, "result.color"))
5983             {
5984                 dstreg = "ret";
5985                 mul_final_dest = TRUE;
5986             }
5987             /* FALLTHROUGH */
5988         case WINED3D_TOP_ADD_SIGNED:
5989             shader_addline(buffer, "SUB arg2, %s, const.w;\n", arg2);
5990             arg2 = "arg2";
5991             /* FALLTHROUGH */
5992         case WINED3D_TOP_ADD:
5993             shader_addline(buffer, "ADD_SAT %s%s, %s, %s;\n", dstreg, dstmask, arg1, arg2);
5994             break;
5995
5996         case WINED3D_TOP_SUBTRACT:
5997             shader_addline(buffer, "SUB_SAT %s%s, %s, %s;\n", dstreg, dstmask, arg1, arg2);
5998             break;
5999
6000         case WINED3D_TOP_ADD_SMOOTH:
6001             shader_addline(buffer, "SUB arg1, const.x, %s;\n", arg1);
6002             shader_addline(buffer, "MAD_SAT %s%s, arg1, %s, %s;\n", dstreg, dstmask, arg2, arg1);
6003             break;
6004
6005         case WINED3D_TOP_BLEND_CURRENT_ALPHA:
6006             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_CURRENT);
6007             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
6008             break;
6009         case WINED3D_TOP_BLEND_FACTOR_ALPHA:
6010             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_TFACTOR);
6011             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
6012             break;
6013         case WINED3D_TOP_BLEND_TEXTURE_ALPHA:
6014             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_TEXTURE);
6015             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
6016             break;
6017         case WINED3D_TOP_BLEND_DIFFUSE_ALPHA:
6018             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_DIFFUSE);
6019             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
6020             break;
6021
6022         case WINED3D_TOP_BLEND_TEXTURE_ALPHA_PM:
6023             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_TEXTURE);
6024             shader_addline(buffer, "SUB arg0.w, const.x, %s.w;\n", arg0);
6025             shader_addline(buffer, "MAD_SAT %s%s, %s, arg0.w, %s;\n", dstreg, dstmask, arg2, arg1);
6026             break;
6027
6028         /* D3DTOP_PREMODULATE ???? */
6029
6030         case WINED3D_TOP_MODULATE_INVALPHA_ADD_COLOR:
6031             shader_addline(buffer, "SUB arg0.w, const.x, %s;\n", arg1);
6032             shader_addline(buffer, "MAD_SAT %s%s, arg0.w, %s, %s;\n", dstreg, dstmask, arg2, arg1);
6033             break;
6034         case WINED3D_TOP_MODULATE_ALPHA_ADD_COLOR:
6035             shader_addline(buffer, "MAD_SAT %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg1, arg2, arg1);
6036             break;
6037         case WINED3D_TOP_MODULATE_INVCOLOR_ADD_ALPHA:
6038             shader_addline(buffer, "SUB arg0, const.x, %s;\n", arg1);
6039             shader_addline(buffer, "MAD_SAT %s%s, arg0, %s, %s.w;\n", dstreg, dstmask, arg2, arg1);
6040             break;
6041         case WINED3D_TOP_MODULATE_COLOR_ADD_ALPHA:
6042             shader_addline(buffer, "MAD_SAT %s%s, %s, %s, %s.w;\n", dstreg, dstmask, arg1, arg2, arg1);
6043             break;
6044
6045         case WINED3D_TOP_DOTPRODUCT3:
6046             mul = 4;
6047             if (!strcmp(dstreg, "result.color"))
6048             {
6049                 dstreg = "ret";
6050                 mul_final_dest = TRUE;
6051             }
6052             shader_addline(buffer, "SUB arg1, %s, const.w;\n", arg1);
6053             shader_addline(buffer, "SUB arg2, %s, const.w;\n", arg2);
6054             shader_addline(buffer, "DP3_SAT %s%s, arg1, arg2;\n", dstreg, dstmask);
6055             break;
6056
6057         case WINED3D_TOP_MULTIPLY_ADD:
6058             shader_addline(buffer, "MAD_SAT %s%s, %s, %s, %s;\n", dstreg, dstmask, arg1, arg2, arg0);
6059             break;
6060
6061         case WINED3D_TOP_LERP:
6062             /* The msdn is not quite right here */
6063             shader_addline(buffer, "LRP %s%s, %s, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
6064             break;
6065
6066         case WINED3D_TOP_BUMPENVMAP:
6067         case WINED3D_TOP_BUMPENVMAP_LUMINANCE:
6068             /* Those are handled in the first pass of the shader(generation pass 1 and 2) already */
6069             break;
6070
6071         default:
6072             FIXME("Unhandled texture op %08x\n", op);
6073     }
6074
6075     if(mul == 2) {
6076         shader_addline(buffer, "MUL_SAT %s%s, %s, const.y;\n", mul_final_dest ? "result.color" : dstreg, dstmask, dstreg);
6077     } else if(mul == 4) {
6078         shader_addline(buffer, "MUL_SAT %s%s, %s, const.z;\n", mul_final_dest ? "result.color" : dstreg, dstmask, dstreg);
6079     }
6080 }
6081
6082 static GLuint gen_arbfp_ffp_shader(const struct ffp_frag_settings *settings, const struct wined3d_gl_info *gl_info)
6083 {
6084     unsigned int stage;
6085     struct wined3d_shader_buffer buffer;
6086     BOOL tex_read[MAX_TEXTURES] = {FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE};
6087     BOOL bump_used[MAX_TEXTURES] = {FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE};
6088     BOOL luminance_used[MAX_TEXTURES] = {FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE};
6089     UINT lowest_disabled_stage;
6090     const char *textype;
6091     const char *instr, *sat;
6092     char colorcor_dst[8];
6093     GLuint ret;
6094     DWORD arg0, arg1, arg2;
6095     BOOL tempreg_used = FALSE, tfactor_used = FALSE;
6096     BOOL op_equal;
6097     const char *final_combiner_src = "ret";
6098     GLint pos;
6099
6100     /* Find out which textures are read */
6101     for (stage = 0; stage < MAX_TEXTURES; ++stage)
6102     {
6103         if (settings->op[stage].cop == WINED3D_TOP_DISABLE)
6104             break;
6105         arg0 = settings->op[stage].carg0 & WINED3DTA_SELECTMASK;
6106         arg1 = settings->op[stage].carg1 & WINED3DTA_SELECTMASK;
6107         arg2 = settings->op[stage].carg2 & WINED3DTA_SELECTMASK;
6108         if(arg0 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
6109         if(arg1 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
6110         if(arg2 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
6111
6112         if (settings->op[stage].cop == WINED3D_TOP_BLEND_TEXTURE_ALPHA)
6113             tex_read[stage] = TRUE;
6114         if (settings->op[stage].cop == WINED3D_TOP_BLEND_TEXTURE_ALPHA_PM)
6115             tex_read[stage] = TRUE;
6116         if (settings->op[stage].cop == WINED3D_TOP_BUMPENVMAP)
6117         {
6118             bump_used[stage] = TRUE;
6119             tex_read[stage] = TRUE;
6120         }
6121         if (settings->op[stage].cop == WINED3D_TOP_BUMPENVMAP_LUMINANCE)
6122         {
6123             bump_used[stage] = TRUE;
6124             tex_read[stage] = TRUE;
6125             luminance_used[stage] = TRUE;
6126         }
6127         else if (settings->op[stage].cop == WINED3D_TOP_BLEND_FACTOR_ALPHA)
6128         {
6129             tfactor_used = TRUE;
6130         }
6131
6132         if(arg0 == WINED3DTA_TFACTOR || arg1 == WINED3DTA_TFACTOR || arg2 == WINED3DTA_TFACTOR) {
6133             tfactor_used = TRUE;
6134         }
6135
6136         if(settings->op[stage].dst == tempreg) tempreg_used = TRUE;
6137         if(arg0 == WINED3DTA_TEMP || arg1 == WINED3DTA_TEMP || arg2 == WINED3DTA_TEMP) {
6138             tempreg_used = TRUE;
6139         }
6140
6141         if (settings->op[stage].aop == WINED3D_TOP_DISABLE)
6142             continue;
6143         arg0 = settings->op[stage].aarg0 & WINED3DTA_SELECTMASK;
6144         arg1 = settings->op[stage].aarg1 & WINED3DTA_SELECTMASK;
6145         arg2 = settings->op[stage].aarg2 & WINED3DTA_SELECTMASK;
6146         if(arg0 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
6147         if(arg1 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
6148         if(arg2 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
6149
6150         if(arg0 == WINED3DTA_TEMP || arg1 == WINED3DTA_TEMP || arg2 == WINED3DTA_TEMP) {
6151             tempreg_used = TRUE;
6152         }
6153         if(arg0 == WINED3DTA_TFACTOR || arg1 == WINED3DTA_TFACTOR || arg2 == WINED3DTA_TFACTOR) {
6154             tfactor_used = TRUE;
6155         }
6156     }
6157     lowest_disabled_stage = stage;
6158
6159     /* Shader header */
6160     if (!shader_buffer_init(&buffer))
6161     {
6162         ERR("Failed to initialize shader buffer.\n");
6163         return 0;
6164     }
6165
6166     shader_addline(&buffer, "!!ARBfp1.0\n");
6167
6168     switch(settings->fog) {
6169         case FOG_OFF:                                                         break;
6170         case FOG_LINEAR: shader_addline(&buffer, "OPTION ARB_fog_linear;\n"); break;
6171         case FOG_EXP:    shader_addline(&buffer, "OPTION ARB_fog_exp;\n");    break;
6172         case FOG_EXP2:   shader_addline(&buffer, "OPTION ARB_fog_exp2;\n");   break;
6173         default: FIXME("Unexpected fog setting %d\n", settings->fog);
6174     }
6175
6176     shader_addline(&buffer, "PARAM const = {1, 2, 4, 0.5};\n");
6177     shader_addline(&buffer, "TEMP TMP;\n");
6178     shader_addline(&buffer, "TEMP ret;\n");
6179     if(tempreg_used || settings->sRGB_write) shader_addline(&buffer, "TEMP tempreg;\n");
6180     shader_addline(&buffer, "TEMP arg0;\n");
6181     shader_addline(&buffer, "TEMP arg1;\n");
6182     shader_addline(&buffer, "TEMP arg2;\n");
6183     for(stage = 0; stage < MAX_TEXTURES; stage++) {
6184         if(!tex_read[stage]) continue;
6185         shader_addline(&buffer, "TEMP tex%u;\n", stage);
6186         if(!bump_used[stage]) continue;
6187         shader_addline(&buffer, "PARAM bumpmat%u = program.env[%u];\n", stage, ARB_FFP_CONST_BUMPMAT(stage));
6188         if(!luminance_used[stage]) continue;
6189         shader_addline(&buffer, "PARAM luminance%u = program.env[%u];\n", stage, ARB_FFP_CONST_LUMINANCE(stage));
6190     }
6191     if(tfactor_used) {
6192         shader_addline(&buffer, "PARAM tfactor = program.env[%u];\n", ARB_FFP_CONST_TFACTOR);
6193     }
6194         shader_addline(&buffer, "PARAM specular_enable = program.env[%u];\n", ARB_FFP_CONST_SPECULAR_ENABLE);
6195
6196     if(settings->sRGB_write) {
6197         shader_addline(&buffer, "PARAM srgb_consts1 = {%f, %f, %f, %f};\n",
6198                        srgb_mul_low, srgb_cmp, srgb_pow, srgb_mul_high);
6199         shader_addline(&buffer, "PARAM srgb_consts2 = {%f, %f, %f, %f};\n",
6200                        srgb_sub_high, 0.0, 0.0, 0.0);
6201     }
6202
6203     if (lowest_disabled_stage < 7 && settings->emul_clipplanes)
6204         shader_addline(&buffer, "KIL fragment.texcoord[7];\n");
6205
6206     /* Generate texture sampling instructions) */
6207     for (stage = 0; stage < MAX_TEXTURES && settings->op[stage].cop != WINED3D_TOP_DISABLE; ++stage)
6208     {
6209         if (!tex_read[stage])
6210             continue;
6211
6212         switch(settings->op[stage].tex_type) {
6213             case tex_1d:                    textype = "1D";     break;
6214             case tex_2d:                    textype = "2D";     break;
6215             case tex_3d:                    textype = "3D";     break;
6216             case tex_cube:                  textype = "CUBE";   break;
6217             case tex_rect:                  textype = "RECT";   break;
6218             default: textype = "unexpected_textype";   break;
6219         }
6220
6221         if (settings->op[stage].cop == WINED3D_TOP_BUMPENVMAP
6222                 || settings->op[stage].cop == WINED3D_TOP_BUMPENVMAP_LUMINANCE)
6223             sat = "";
6224         else
6225             sat = "_SAT";
6226
6227         if(settings->op[stage].projected == proj_none) {
6228             instr = "TEX";
6229         } else if(settings->op[stage].projected == proj_count4 ||
6230                   settings->op[stage].projected == proj_count3) {
6231             instr = "TXP";
6232         } else {
6233             FIXME("Unexpected projection mode %d\n", settings->op[stage].projected);
6234             instr = "TXP";
6235         }
6236
6237         if (stage > 0
6238                 && (settings->op[stage - 1].cop == WINED3D_TOP_BUMPENVMAP
6239                 || settings->op[stage - 1].cop == WINED3D_TOP_BUMPENVMAP_LUMINANCE))
6240         {
6241             shader_addline(&buffer, "SWZ arg1, bumpmat%u, x, z, 0, 0;\n", stage - 1);
6242             shader_addline(&buffer, "DP3 ret.x, arg1, tex%u;\n", stage - 1);
6243             shader_addline(&buffer, "SWZ arg1, bumpmat%u, y, w, 0, 0;\n", stage - 1);
6244             shader_addline(&buffer, "DP3 ret.y, arg1, tex%u;\n", stage - 1);
6245
6246             /* with projective textures, texbem only divides the static texture coord, not the displacement,
6247              * so multiply the displacement with the dividing parameter before passing it to TXP
6248              */
6249             if (settings->op[stage].projected != proj_none) {
6250                 if(settings->op[stage].projected == proj_count4) {
6251                     shader_addline(&buffer, "MOV ret.w, fragment.texcoord[%u].w;\n", stage);
6252                     shader_addline(&buffer, "MUL ret.xyz, ret, fragment.texcoord[%u].w, fragment.texcoord[%u];\n", stage, stage);
6253                 } else {
6254                     shader_addline(&buffer, "MOV ret.w, fragment.texcoord[%u].z;\n", stage);
6255                     shader_addline(&buffer, "MAD ret.xyz, ret, fragment.texcoord[%u].z, fragment.texcoord[%u];\n", stage, stage);
6256                 }
6257             } else {
6258                 shader_addline(&buffer, "ADD ret, ret, fragment.texcoord[%u];\n", stage);
6259             }
6260
6261             shader_addline(&buffer, "%s%s tex%u, ret, texture[%u], %s;\n",
6262                     instr, sat, stage, stage, textype);
6263             if (settings->op[stage - 1].cop == WINED3D_TOP_BUMPENVMAP_LUMINANCE)
6264             {
6265                 shader_addline(&buffer, "MAD_SAT ret.x, tex%u.z, luminance%u.x, luminance%u.y;\n",
6266                                stage - 1, stage - 1, stage - 1);
6267                 shader_addline(&buffer, "MUL tex%u, tex%u, ret.x;\n", stage, stage);
6268             }
6269         } else if(settings->op[stage].projected == proj_count3) {
6270             shader_addline(&buffer, "MOV ret, fragment.texcoord[%u];\n", stage);
6271             shader_addline(&buffer, "MOV ret.w, ret.z;\n");
6272             shader_addline(&buffer, "%s%s tex%u, ret, texture[%u], %s;\n",
6273                             instr, sat, stage, stage, textype);
6274         } else {
6275             shader_addline(&buffer, "%s%s tex%u, fragment.texcoord[%u], texture[%u], %s;\n",
6276                             instr, sat, stage, stage, stage, textype);
6277         }
6278
6279         sprintf(colorcor_dst, "tex%u", stage);
6280         gen_color_correction(&buffer, colorcor_dst, WINED3DSP_WRITEMASK_ALL, "const.x", "const.y",
6281                 settings->op[stage].color_fixup);
6282     }
6283
6284     /* Generate the main shader */
6285     for (stage = 0; stage < MAX_TEXTURES; ++stage)
6286     {
6287         if (settings->op[stage].cop == WINED3D_TOP_DISABLE)
6288         {
6289             if (!stage)
6290                 final_combiner_src = "fragment.color.primary";
6291             break;
6292         }
6293
6294         if (settings->op[stage].cop == WINED3D_TOP_SELECT_ARG1
6295                 && settings->op[stage].aop == WINED3D_TOP_SELECT_ARG1)
6296             op_equal = settings->op[stage].carg1 == settings->op[stage].aarg1;
6297         else if (settings->op[stage].cop == WINED3D_TOP_SELECT_ARG1
6298                 && settings->op[stage].aop == WINED3D_TOP_SELECT_ARG2)
6299             op_equal = settings->op[stage].carg1 == settings->op[stage].aarg2;
6300         else if (settings->op[stage].cop == WINED3D_TOP_SELECT_ARG2
6301                 && settings->op[stage].aop == WINED3D_TOP_SELECT_ARG1)
6302             op_equal = settings->op[stage].carg2 == settings->op[stage].aarg1;
6303         else if (settings->op[stage].cop == WINED3D_TOP_SELECT_ARG2
6304                 && settings->op[stage].aop == WINED3D_TOP_SELECT_ARG2)
6305             op_equal = settings->op[stage].carg2 == settings->op[stage].aarg2;
6306         else
6307             op_equal = settings->op[stage].aop   == settings->op[stage].cop
6308                     && settings->op[stage].carg0 == settings->op[stage].aarg0
6309                     && settings->op[stage].carg1 == settings->op[stage].aarg1
6310                     && settings->op[stage].carg2 == settings->op[stage].aarg2;
6311
6312         if (settings->op[stage].aop == WINED3D_TOP_DISABLE)
6313         {
6314             gen_ffp_instr(&buffer, stage, TRUE, FALSE, settings->op[stage].dst,
6315                           settings->op[stage].cop, settings->op[stage].carg0,
6316                           settings->op[stage].carg1, settings->op[stage].carg2);
6317             if (!stage)
6318                 shader_addline(&buffer, "MOV ret.w, fragment.color.primary.w;\n");
6319         }
6320         else if (op_equal)
6321         {
6322             gen_ffp_instr(&buffer, stage, TRUE, TRUE, settings->op[stage].dst,
6323                           settings->op[stage].cop, settings->op[stage].carg0,
6324                           settings->op[stage].carg1, settings->op[stage].carg2);
6325         } else {
6326             gen_ffp_instr(&buffer, stage, TRUE, FALSE, settings->op[stage].dst,
6327                           settings->op[stage].cop, settings->op[stage].carg0,
6328                           settings->op[stage].carg1, settings->op[stage].carg2);
6329             gen_ffp_instr(&buffer, stage, FALSE, TRUE, settings->op[stage].dst,
6330                           settings->op[stage].aop, settings->op[stage].aarg0,
6331                           settings->op[stage].aarg1, settings->op[stage].aarg2);
6332         }
6333     }
6334
6335     if(settings->sRGB_write) {
6336         shader_addline(&buffer, "MAD ret, fragment.color.secondary, specular_enable, %s;\n", final_combiner_src);
6337         arbfp_add_sRGB_correction(&buffer, "ret", "arg0", "arg1", "arg2", "tempreg", FALSE);
6338         shader_addline(&buffer, "MOV result.color, ret;\n");
6339     } else {
6340         shader_addline(&buffer, "MAD result.color, fragment.color.secondary, specular_enable, %s;\n", final_combiner_src);
6341     }
6342
6343     /* Footer */
6344     shader_addline(&buffer, "END\n");
6345
6346     /* Generate the shader */
6347     GL_EXTCALL(glGenProgramsARB(1, &ret));
6348     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, ret));
6349     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
6350             strlen(buffer.buffer), buffer.buffer));
6351     checkGLcall("glProgramStringARB()");
6352
6353     gl_info->gl_ops.gl.p_glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
6354     if (pos != -1)
6355     {
6356         FIXME("Fragment program error at position %d: %s\n\n", pos,
6357               debugstr_a((const char *)gl_info->gl_ops.gl.p_glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
6358         shader_arb_dump_program_source(buffer.buffer);
6359     }
6360     else
6361     {
6362         GLint native;
6363
6364         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
6365         checkGLcall("glGetProgramivARB()");
6366         if (!native) WARN("Program exceeds native resource limits.\n");
6367     }
6368
6369     shader_buffer_free(&buffer);
6370     return ret;
6371 }
6372
6373 static void fragment_prog_arbfp(struct wined3d_context *context, const struct wined3d_state *state, DWORD state_id)
6374 {
6375     const struct wined3d_device *device = context->swapchain->device;
6376     const struct wined3d_gl_info *gl_info = context->gl_info;
6377     struct shader_arb_priv *priv = device->fragment_priv;
6378     BOOL use_vshader = use_vs(state);
6379     BOOL use_pshader = use_ps(state);
6380     struct ffp_frag_settings settings;
6381     const struct arbfp_ffp_desc *desc;
6382     unsigned int i;
6383
6384     TRACE("context %p, state %p, state_id %#x.\n", context, state, state_id);
6385
6386     if (isStateDirty(context, STATE_RENDER(WINED3D_RS_FOGENABLE)))
6387     {
6388         if (!use_pshader && device->shader_backend == &arb_program_shader_backend && context->last_was_pshader)
6389         {
6390             /* Reload fixed function constants since they collide with the
6391              * pixel shader constants. */
6392             for (i = 0; i < MAX_TEXTURES; ++i)
6393             {
6394                 set_bumpmat_arbfp(context, state, STATE_TEXTURESTAGE(i, WINED3D_TSS_BUMPENV_MAT00));
6395             }
6396             state_texfactor_arbfp(context, state, STATE_RENDER(WINED3D_RS_TEXTUREFACTOR));
6397             state_arb_specularenable(context, state, STATE_RENDER(WINED3D_RS_SPECULARENABLE));
6398         }
6399         else if (use_pshader && !isStateDirty(context, context->state_table[STATE_VSHADER].representative))
6400         {
6401             device->shader_backend->shader_select(context, use_pshader, use_vshader);
6402         }
6403         return;
6404     }
6405
6406     if (!use_pshader)
6407     {
6408         /* Find or create a shader implementing the fixed function pipeline
6409          * settings, then activate it. */
6410         gen_ffp_frag_op(device, state, &settings, FALSE);
6411         desc = (const struct arbfp_ffp_desc *)find_ffp_frag_shader(&priv->fragment_shaders, &settings);
6412         if(!desc) {
6413             struct arbfp_ffp_desc *new_desc = HeapAlloc(GetProcessHeap(), 0, sizeof(*new_desc));
6414             if (!new_desc)
6415             {
6416                 ERR("Out of memory\n");
6417                 return;
6418             }
6419             new_desc->num_textures_used = 0;
6420             for (i = 0; i < gl_info->limits.texture_stages; ++i)
6421             {
6422                 if (settings.op[i].cop == WINED3D_TOP_DISABLE)
6423                     break;
6424                 new_desc->num_textures_used = i;
6425             }
6426
6427             memcpy(&new_desc->parent.settings, &settings, sizeof(settings));
6428             new_desc->shader = gen_arbfp_ffp_shader(&settings, gl_info);
6429             add_ffp_frag_shader(&priv->fragment_shaders, &new_desc->parent);
6430             TRACE("Allocated fixed function replacement shader descriptor %p\n", new_desc);
6431             desc = new_desc;
6432         }
6433
6434         /* Now activate the replacement program. GL_FRAGMENT_PROGRAM_ARB is already active (however, note the
6435          * comment above the shader_select call below). If e.g. GLSL is active, the shader_select call will
6436          * deactivate it.
6437          */
6438         GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, desc->shader));
6439         checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, desc->shader)");
6440         priv->current_fprogram_id = desc->shader;
6441
6442         if (device->shader_backend == &arb_program_shader_backend && context->last_was_pshader)
6443         {
6444             /* Reload fixed function constants since they collide with the
6445              * pixel shader constants. */
6446             for (i = 0; i < MAX_TEXTURES; ++i)
6447             {
6448                 set_bumpmat_arbfp(context, state, STATE_TEXTURESTAGE(i, WINED3D_TSS_BUMPENV_MAT00));
6449             }
6450             state_texfactor_arbfp(context, state, STATE_RENDER(WINED3D_RS_TEXTUREFACTOR));
6451             state_arb_specularenable(context, state, STATE_RENDER(WINED3D_RS_SPECULARENABLE));
6452         }
6453         context->last_was_pshader = FALSE;
6454     } else {
6455         context->last_was_pshader = TRUE;
6456     }
6457
6458     /* Finally, select the shader. If a pixel shader is used, it will be set and enabled by the shader backend.
6459      * If this shader backend is arbfp(most likely), then it will simply overwrite the last fixed function
6460      * replacement shader. If the shader backend is not ARB, it currently is important that the opengl implementation
6461      * type overwrites GL_ARB_fragment_program. This is currently the case with GLSL. If we really want to use
6462      * atifs or nvrc pixel shaders with arb fragment programs we'd have to disable GL_FRAGMENT_PROGRAM_ARB here
6463      *
6464      * Don't call shader_select if the vertex shader is dirty, because it will be called later on by the vertex
6465      * shader handler.
6466      */
6467     if (!isStateDirty(context, context->state_table[STATE_VSHADER].representative))
6468     {
6469         device->shader_backend->shader_select(context, use_pshader, use_vshader);
6470
6471         if (!isStateDirty(context, STATE_VERTEXSHADERCONSTANT) && (use_vshader || use_pshader))
6472             context_apply_state(context, state, STATE_VERTEXSHADERCONSTANT);
6473     }
6474     if (use_pshader)
6475         context_apply_state(context, state, STATE_PIXELSHADERCONSTANT);
6476 }
6477
6478 /* We can't link the fog states to the fragment state directly since the
6479  * vertex pipeline links them to FOGENABLE. A different linking in different
6480  * pipeline parts can't be expressed in the combined state table, so we need
6481  * to handle that with a forwarding function. The other invisible side effect
6482  * is that changing the fog start and fog end (which links to FOGENABLE in
6483  * vertex) results in the fragment_prog_arbfp function being called because
6484  * FOGENABLE is dirty, which calls this function here. */
6485 static void state_arbfp_fog(struct wined3d_context *context, const struct wined3d_state *state, DWORD state_id)
6486 {
6487     enum fogsource new_source;
6488
6489     TRACE("context %p, state %p, state_id %#x.\n", context, state, state_id);
6490
6491     if (!isStateDirty(context, STATE_PIXELSHADER))
6492         fragment_prog_arbfp(context, state, state_id);
6493
6494     if (!state->render_states[WINED3D_RS_FOGENABLE])
6495         return;
6496
6497     if (state->render_states[WINED3D_RS_FOGTABLEMODE] == WINED3D_FOG_NONE)
6498     {
6499         if (use_vs(state))
6500         {
6501             new_source = FOGSOURCE_VS;
6502         }
6503         else
6504         {
6505             if (state->render_states[WINED3D_RS_FOGVERTEXMODE] == WINED3D_FOG_NONE || context->last_was_rhw)
6506                 new_source = FOGSOURCE_COORD;
6507             else
6508                 new_source = FOGSOURCE_FFP;
6509         }
6510     }
6511     else
6512     {
6513         new_source = FOGSOURCE_FFP;
6514     }
6515
6516     if (new_source != context->fog_source)
6517     {
6518         context->fog_source = new_source;
6519         state_fogstartend(context, state, STATE_RENDER(WINED3D_RS_FOGSTART));
6520     }
6521 }
6522
6523 static void textransform(struct wined3d_context *context, const struct wined3d_state *state, DWORD state_id)
6524 {
6525     if (!isStateDirty(context, STATE_PIXELSHADER))
6526         fragment_prog_arbfp(context, state, state_id);
6527 }
6528
6529 static void state_srgbwrite(struct wined3d_context *context, const struct wined3d_state *state, DWORD state_id)
6530 {
6531     const struct wined3d_gl_info *gl_info = context->gl_info;
6532     const struct wined3d_surface *rt = state->fb->render_targets[0];
6533
6534     TRACE("context %p, state %p, state_id %#x.\n", context, state, state_id);
6535
6536     if (state->render_states[WINED3D_RS_SRGBWRITEENABLE]
6537             && rt->resource.format->flags & WINED3DFMT_FLAG_SRGB_WRITE)
6538         gl_info->gl_ops.gl.p_glEnable(GL_FRAMEBUFFER_SRGB);
6539     else
6540         gl_info->gl_ops.gl.p_glDisable(GL_FRAMEBUFFER_SRGB);
6541 }
6542
6543 static const struct StateEntryTemplate arbfp_fragmentstate_template[] =
6544 {
6545     {STATE_RENDER(WINED3D_RS_TEXTUREFACTOR),              { STATE_RENDER(WINED3D_RS_TEXTUREFACTOR),             state_texfactor_arbfp   }, WINED3D_GL_EXT_NONE             },
6546     {STATE_TEXTURESTAGE(0, WINED3D_TSS_COLOR_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6547     {STATE_TEXTURESTAGE(0, WINED3D_TSS_COLOR_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6548     {STATE_TEXTURESTAGE(0, WINED3D_TSS_COLOR_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6549     {STATE_TEXTURESTAGE(0, WINED3D_TSS_COLOR_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6550     {STATE_TEXTURESTAGE(0, WINED3D_TSS_ALPHA_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6551     {STATE_TEXTURESTAGE(0, WINED3D_TSS_ALPHA_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6552     {STATE_TEXTURESTAGE(0, WINED3D_TSS_ALPHA_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6553     {STATE_TEXTURESTAGE(0, WINED3D_TSS_ALPHA_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6554     {STATE_TEXTURESTAGE(0, WINED3D_TSS_RESULT_ARG),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6555     {STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6556     {STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6557     {STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6558     {STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6559     {STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6560     {STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6561     {STATE_TEXTURESTAGE(1, WINED3D_TSS_COLOR_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6562     {STATE_TEXTURESTAGE(1, WINED3D_TSS_COLOR_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6563     {STATE_TEXTURESTAGE(1, WINED3D_TSS_COLOR_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6564     {STATE_TEXTURESTAGE(1, WINED3D_TSS_COLOR_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6565     {STATE_TEXTURESTAGE(1, WINED3D_TSS_ALPHA_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6566     {STATE_TEXTURESTAGE(1, WINED3D_TSS_ALPHA_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6567     {STATE_TEXTURESTAGE(1, WINED3D_TSS_ALPHA_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6568     {STATE_TEXTURESTAGE(1, WINED3D_TSS_ALPHA_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6569     {STATE_TEXTURESTAGE(1, WINED3D_TSS_RESULT_ARG),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6570     {STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6571     {STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6572     {STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6573     {STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6574     {STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6575     {STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6576     {STATE_TEXTURESTAGE(2, WINED3D_TSS_COLOR_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6577     {STATE_TEXTURESTAGE(2, WINED3D_TSS_COLOR_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6578     {STATE_TEXTURESTAGE(2, WINED3D_TSS_COLOR_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6579     {STATE_TEXTURESTAGE(2, WINED3D_TSS_COLOR_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6580     {STATE_TEXTURESTAGE(2, WINED3D_TSS_ALPHA_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6581     {STATE_TEXTURESTAGE(2, WINED3D_TSS_ALPHA_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6582     {STATE_TEXTURESTAGE(2, WINED3D_TSS_ALPHA_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6583     {STATE_TEXTURESTAGE(2, WINED3D_TSS_ALPHA_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6584     {STATE_TEXTURESTAGE(2, WINED3D_TSS_RESULT_ARG),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6585     {STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6586     {STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6587     {STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6588     {STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6589     {STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6590     {STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6591     {STATE_TEXTURESTAGE(3, WINED3D_TSS_COLOR_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6592     {STATE_TEXTURESTAGE(3, WINED3D_TSS_COLOR_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6593     {STATE_TEXTURESTAGE(3, WINED3D_TSS_COLOR_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6594     {STATE_TEXTURESTAGE(3, WINED3D_TSS_COLOR_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6595     {STATE_TEXTURESTAGE(3, WINED3D_TSS_ALPHA_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6596     {STATE_TEXTURESTAGE(3, WINED3D_TSS_ALPHA_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6597     {STATE_TEXTURESTAGE(3, WINED3D_TSS_ALPHA_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6598     {STATE_TEXTURESTAGE(3, WINED3D_TSS_ALPHA_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6599     {STATE_TEXTURESTAGE(3, WINED3D_TSS_RESULT_ARG),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6600     {STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6601     {STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6602     {STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6603     {STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6604     {STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6605     {STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6606     {STATE_TEXTURESTAGE(4, WINED3D_TSS_COLOR_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6607     {STATE_TEXTURESTAGE(4, WINED3D_TSS_COLOR_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6608     {STATE_TEXTURESTAGE(4, WINED3D_TSS_COLOR_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6609     {STATE_TEXTURESTAGE(4, WINED3D_TSS_COLOR_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6610     {STATE_TEXTURESTAGE(4, WINED3D_TSS_ALPHA_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6611     {STATE_TEXTURESTAGE(4, WINED3D_TSS_ALPHA_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6612     {STATE_TEXTURESTAGE(4, WINED3D_TSS_ALPHA_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6613     {STATE_TEXTURESTAGE(4, WINED3D_TSS_ALPHA_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6614     {STATE_TEXTURESTAGE(4, WINED3D_TSS_RESULT_ARG),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6615     {STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6616     {STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6617     {STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6618     {STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6619     {STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6620     {STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6621     {STATE_TEXTURESTAGE(5, WINED3D_TSS_COLOR_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6622     {STATE_TEXTURESTAGE(5, WINED3D_TSS_COLOR_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6623     {STATE_TEXTURESTAGE(5, WINED3D_TSS_COLOR_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6624     {STATE_TEXTURESTAGE(5, WINED3D_TSS_COLOR_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6625     {STATE_TEXTURESTAGE(5, WINED3D_TSS_ALPHA_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6626     {STATE_TEXTURESTAGE(5, WINED3D_TSS_ALPHA_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6627     {STATE_TEXTURESTAGE(5, WINED3D_TSS_ALPHA_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6628     {STATE_TEXTURESTAGE(5, WINED3D_TSS_ALPHA_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6629     {STATE_TEXTURESTAGE(5, WINED3D_TSS_RESULT_ARG),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6630     {STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6631     {STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6632     {STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6633     {STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6634     {STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6635     {STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6636     {STATE_TEXTURESTAGE(6, WINED3D_TSS_COLOR_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6637     {STATE_TEXTURESTAGE(6, WINED3D_TSS_COLOR_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6638     {STATE_TEXTURESTAGE(6, WINED3D_TSS_COLOR_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6639     {STATE_TEXTURESTAGE(6, WINED3D_TSS_COLOR_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6640     {STATE_TEXTURESTAGE(6, WINED3D_TSS_ALPHA_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6641     {STATE_TEXTURESTAGE(6, WINED3D_TSS_ALPHA_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6642     {STATE_TEXTURESTAGE(6, WINED3D_TSS_ALPHA_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6643     {STATE_TEXTURESTAGE(6, WINED3D_TSS_ALPHA_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6644     {STATE_TEXTURESTAGE(6, WINED3D_TSS_RESULT_ARG),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6645     {STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6646     {STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6647     {STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6648     {STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6649     {STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6650     {STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6651     {STATE_TEXTURESTAGE(7, WINED3D_TSS_COLOR_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6652     {STATE_TEXTURESTAGE(7, WINED3D_TSS_COLOR_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6653     {STATE_TEXTURESTAGE(7, WINED3D_TSS_COLOR_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6654     {STATE_TEXTURESTAGE(7, WINED3D_TSS_COLOR_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6655     {STATE_TEXTURESTAGE(7, WINED3D_TSS_ALPHA_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6656     {STATE_TEXTURESTAGE(7, WINED3D_TSS_ALPHA_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6657     {STATE_TEXTURESTAGE(7, WINED3D_TSS_ALPHA_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6658     {STATE_TEXTURESTAGE(7, WINED3D_TSS_ALPHA_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6659     {STATE_TEXTURESTAGE(7, WINED3D_TSS_RESULT_ARG),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6660     {STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6661     {STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6662     {STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6663     {STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6664     {STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6665     {STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6666     {STATE_PIXELSHADER,                                   { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6667     {STATE_RENDER(WINED3D_RS_FOGENABLE),                  { STATE_RENDER(WINED3D_RS_FOGENABLE),                 state_arbfp_fog         }, WINED3D_GL_EXT_NONE             },
6668     {STATE_RENDER(WINED3D_RS_FOGTABLEMODE),               { STATE_RENDER(WINED3D_RS_FOGENABLE),                 NULL                    }, WINED3D_GL_EXT_NONE             },
6669     {STATE_RENDER(WINED3D_RS_FOGVERTEXMODE),              { STATE_RENDER(WINED3D_RS_FOGENABLE),                 NULL                    }, WINED3D_GL_EXT_NONE             },
6670     {STATE_RENDER(WINED3D_RS_FOGSTART),                   { STATE_RENDER(WINED3D_RS_FOGSTART),                  state_fogstartend       }, WINED3D_GL_EXT_NONE             },
6671     {STATE_RENDER(WINED3D_RS_FOGEND),                     { STATE_RENDER(WINED3D_RS_FOGSTART),                  NULL                    }, WINED3D_GL_EXT_NONE             },
6672     {STATE_RENDER(WINED3D_RS_SRGBWRITEENABLE),            { STATE_RENDER(WINED3D_RS_SRGBWRITEENABLE),           state_srgbwrite         }, ARB_FRAMEBUFFER_SRGB            },
6673     {STATE_RENDER(WINED3D_RS_SRGBWRITEENABLE),            { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6674     {STATE_RENDER(WINED3D_RS_FOGCOLOR),                   { STATE_RENDER(WINED3D_RS_FOGCOLOR),                  state_fogcolor          }, WINED3D_GL_EXT_NONE             },
6675     {STATE_RENDER(WINED3D_RS_FOGDENSITY),                 { STATE_RENDER(WINED3D_RS_FOGDENSITY),                state_fogdensity        }, WINED3D_GL_EXT_NONE             },
6676     {STATE_TEXTURESTAGE(0,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(0, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6677     {STATE_TEXTURESTAGE(1,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(1, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6678     {STATE_TEXTURESTAGE(2,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(2, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6679     {STATE_TEXTURESTAGE(3,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(3, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6680     {STATE_TEXTURESTAGE(4,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(4, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6681     {STATE_TEXTURESTAGE(5,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(5, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6682     {STATE_TEXTURESTAGE(6,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(6, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6683     {STATE_TEXTURESTAGE(7,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(7, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6684     {STATE_RENDER(WINED3D_RS_SPECULARENABLE),             { STATE_RENDER(WINED3D_RS_SPECULARENABLE),            state_arb_specularenable}, WINED3D_GL_EXT_NONE             },
6685     {0 /* Terminate */,                                   { 0,                                                  0                       }, WINED3D_GL_EXT_NONE             },
6686 };
6687
6688 const struct fragment_pipeline arbfp_fragment_pipeline = {
6689     arbfp_enable,
6690     arbfp_get_caps,
6691     arbfp_alloc,
6692     arbfp_free,
6693     shader_arb_color_fixup_supported,
6694     arbfp_fragmentstate_template,
6695     TRUE /* We can disable projected textures */
6696 };
6697
6698 struct arbfp_blit_priv {
6699     GLenum yuy2_rect_shader, yuy2_2d_shader;
6700     GLenum uyvy_rect_shader, uyvy_2d_shader;
6701     GLenum yv12_rect_shader, yv12_2d_shader;
6702     GLenum p8_rect_shader, p8_2d_shader;
6703     GLuint palette_texture;
6704 };
6705
6706 static HRESULT arbfp_blit_alloc(struct wined3d_device *device)
6707 {
6708     device->blit_priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(struct arbfp_blit_priv));
6709     if(!device->blit_priv) {
6710         ERR("Out of memory\n");
6711         return E_OUTOFMEMORY;
6712     }
6713     return WINED3D_OK;
6714 }
6715
6716 /* Context activation is done by the caller. */
6717 static void arbfp_blit_free(struct wined3d_device *device)
6718 {
6719     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
6720     struct arbfp_blit_priv *priv = device->blit_priv;
6721
6722     ENTER_GL();
6723     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yuy2_rect_shader));
6724     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yuy2_2d_shader));
6725     GL_EXTCALL(glDeleteProgramsARB(1, &priv->uyvy_rect_shader));
6726     GL_EXTCALL(glDeleteProgramsARB(1, &priv->uyvy_2d_shader));
6727     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yv12_rect_shader));
6728     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yv12_2d_shader));
6729     GL_EXTCALL(glDeleteProgramsARB(1, &priv->p8_rect_shader));
6730     GL_EXTCALL(glDeleteProgramsARB(1, &priv->p8_2d_shader));
6731     checkGLcall("Delete yuv and p8 programs");
6732
6733     if (priv->palette_texture)
6734         gl_info->gl_ops.gl.p_glDeleteTextures(1, &priv->palette_texture);
6735     LEAVE_GL();
6736
6737     HeapFree(GetProcessHeap(), 0, device->blit_priv);
6738     device->blit_priv = NULL;
6739 }
6740
6741 static BOOL gen_planar_yuv_read(struct wined3d_shader_buffer *buffer, enum complex_fixup fixup,
6742         GLenum textype, char *luminance)
6743 {
6744     char chroma;
6745     const char *tex, *texinstr;
6746
6747     if (fixup == COMPLEX_FIXUP_UYVY) {
6748         chroma = 'x';
6749         *luminance = 'w';
6750     } else {
6751         chroma = 'w';
6752         *luminance = 'x';
6753     }
6754     switch(textype) {
6755         case GL_TEXTURE_2D:             tex = "2D";     texinstr = "TXP"; break;
6756         case GL_TEXTURE_RECTANGLE_ARB:  tex = "RECT";   texinstr = "TEX"; break;
6757         default:
6758             /* This is more tricky than just replacing the texture type - we have to navigate
6759              * properly in the texture to find the correct chroma values
6760              */
6761             FIXME("Implement yuv correction for non-2d, non-rect textures\n");
6762             return FALSE;
6763     }
6764
6765     /* First we have to read the chroma values. This means we need at least two pixels(no filtering),
6766      * or 4 pixels(with filtering). To get the unmodified chromas, we have to rid ourselves of the
6767      * filtering when we sample the texture.
6768      *
6769      * These are the rules for reading the chroma:
6770      *
6771      * Even pixel: Cr
6772      * Even pixel: U
6773      * Odd pixel: V
6774      *
6775      * So we have to get the sampling x position in non-normalized coordinates in integers
6776      */
6777     if(textype != GL_TEXTURE_RECTANGLE_ARB) {
6778         shader_addline(buffer, "MUL texcrd.xy, fragment.texcoord[0], size.x;\n");
6779         shader_addline(buffer, "MOV texcrd.w, size.x;\n");
6780     } else {
6781         shader_addline(buffer, "MOV texcrd, fragment.texcoord[0];\n");
6782     }
6783     /* We must not allow filtering between pixel x and x+1, this would mix U and V
6784      * Vertical filtering is ok. However, bear in mind that the pixel center is at
6785      * 0.5, so add 0.5.
6786      */
6787     shader_addline(buffer, "FLR texcrd.x, texcrd.x;\n");
6788     shader_addline(buffer, "ADD texcrd.x, texcrd.x, coef.y;\n");
6789
6790     /* Divide the x coordinate by 0.5 and get the fraction. This gives 0.25 and 0.75 for the
6791      * even and odd pixels respectively
6792      */
6793     shader_addline(buffer, "MUL texcrd2, texcrd, coef.y;\n");
6794     shader_addline(buffer, "FRC texcrd2, texcrd2;\n");
6795
6796     /* Sample Pixel 1 */
6797     shader_addline(buffer, "%s luminance, texcrd, texture[0], %s;\n", texinstr, tex);
6798
6799     /* Put the value into either of the chroma values */
6800     shader_addline(buffer, "SGE temp.x, texcrd2.x, coef.y;\n");
6801     shader_addline(buffer, "MUL chroma.x, luminance.%c, temp.x;\n", chroma);
6802     shader_addline(buffer, "SLT temp.x, texcrd2.x, coef.y;\n");
6803     shader_addline(buffer, "MUL chroma.y, luminance.%c, temp.x;\n", chroma);
6804
6805     /* Sample pixel 2. If we read an even pixel(SLT above returned 1), sample
6806      * the pixel right to the current one. Otherwise, sample the left pixel.
6807      * Bias and scale the SLT result to -1;1 and add it to the texcrd.x.
6808      */
6809     shader_addline(buffer, "MAD temp.x, temp.x, coef.z, -coef.x;\n");
6810     shader_addline(buffer, "ADD texcrd.x, texcrd, temp.x;\n");
6811     shader_addline(buffer, "%s luminance, texcrd, texture[0], %s;\n", texinstr, tex);
6812
6813     /* Put the value into the other chroma */
6814     shader_addline(buffer, "SGE temp.x, texcrd2.x, coef.y;\n");
6815     shader_addline(buffer, "MAD chroma.y, luminance.%c, temp.x, chroma.y;\n", chroma);
6816     shader_addline(buffer, "SLT temp.x, texcrd2.x, coef.y;\n");
6817     shader_addline(buffer, "MAD chroma.x, luminance.%c, temp.x, chroma.x;\n", chroma);
6818
6819     /* TODO: If filtering is enabled, sample a 2nd pair of pixels left or right of
6820      * the current one and lerp the two U and V values
6821      */
6822
6823     /* This gives the correctly filtered luminance value */
6824     shader_addline(buffer, "TEX luminance, fragment.texcoord[0], texture[0], %s;\n", tex);
6825
6826     return TRUE;
6827 }
6828
6829 static BOOL gen_yv12_read(struct wined3d_shader_buffer *buffer, GLenum textype, char *luminance)
6830 {
6831     const char *tex;
6832
6833     switch(textype) {
6834         case GL_TEXTURE_2D:             tex = "2D";     break;
6835         case GL_TEXTURE_RECTANGLE_ARB:  tex = "RECT";   break;
6836         default:
6837             FIXME("Implement yv12 correction for non-2d, non-rect textures\n");
6838             return FALSE;
6839     }
6840
6841     /* YV12 surfaces contain a WxH sized luminance plane, followed by a (W/2)x(H/2)
6842      * V and a (W/2)x(H/2) U plane, each with 8 bit per pixel. So the effective
6843      * bitdepth is 12 bits per pixel. Since the U and V planes have only half the
6844      * pitch of the luminance plane, the packing into the gl texture is a bit
6845      * unfortunate. If the whole texture is interpreted as luminance data it looks
6846      * approximately like this:
6847      *
6848      *        +----------------------------------+----
6849      *        |                                  |
6850      *        |                                  |
6851      *        |                                  |
6852      *        |                                  |
6853      *        |                                  |   2
6854      *        |            LUMINANCE             |   -
6855      *        |                                  |   3
6856      *        |                                  |
6857      *        |                                  |
6858      *        |                                  |
6859      *        |                                  |
6860      *        +----------------+-----------------+----
6861      *        |                |                 |
6862      *        |  U even rows   |  U odd rows     |
6863      *        |                |                 |   1
6864      *        +----------------+------------------   -
6865      *        |                |                 |   3
6866      *        |  V even rows   |  V odd rows     |
6867      *        |                |                 |
6868      *        +----------------+-----------------+----
6869      *        |                |                 |
6870      *        |     0.5        |       0.5       |
6871      *
6872      * So it appears as if there are 4 chroma images, but in fact the odd rows
6873      * in the chroma images are in the same row as the even ones. So its is
6874      * kinda tricky to read
6875      *
6876      * When reading from rectangle textures, keep in mind that the input y coordinates
6877      * go from 0 to d3d_height, whereas the opengl texture height is 1.5 * d3d_height
6878      */
6879     shader_addline(buffer, "PARAM yv12_coef = {%f, %f, %f, %f};\n",
6880             2.0f / 3.0f, 1.0f / 6.0f, (2.0f / 3.0f) + (1.0f / 6.0f), 1.0f / 3.0f);
6881
6882     shader_addline(buffer, "MOV texcrd, fragment.texcoord[0];\n");
6883     /* the chroma planes have only half the width */
6884     shader_addline(buffer, "MUL texcrd.x, texcrd.x, coef.y;\n");
6885
6886     /* The first value is between 2/3 and 5/6th of the texture's height, so scale+bias
6887      * the coordinate. Also read the right side of the image when reading odd lines
6888      *
6889      * Don't forget to clamp the y values in into the range, otherwise we'll get filtering
6890      * bleeding
6891      */
6892     if(textype == GL_TEXTURE_2D) {
6893
6894         shader_addline(buffer, "RCP chroma.w, size.y;\n");
6895
6896         shader_addline(buffer, "MUL texcrd2.y, texcrd.y, size.y;\n");
6897
6898         shader_addline(buffer, "FLR texcrd2.y, texcrd2.y;\n");
6899         shader_addline(buffer, "MAD texcrd.y, texcrd.y, yv12_coef.y, yv12_coef.x;\n");
6900
6901         /* Read odd lines from the right side(add size * 0.5 to the x coordinate */
6902         shader_addline(buffer, "ADD texcrd2.x, texcrd2.y, yv12_coef.y;\n"); /* To avoid 0.5 == 0.5 comparisons */
6903         shader_addline(buffer, "FRC texcrd2.x, texcrd2.x;\n");
6904         shader_addline(buffer, "SGE texcrd2.x, texcrd2.x, coef.y;\n");
6905         shader_addline(buffer, "MAD texcrd.x, texcrd2.x, coef.y, texcrd.x;\n");
6906
6907         /* clamp, keep the half pixel origin in mind */
6908         shader_addline(buffer, "MAD temp.y, coef.y, chroma.w, yv12_coef.x;\n");
6909         shader_addline(buffer, "MAX texcrd.y, temp.y, texcrd.y;\n");
6910         shader_addline(buffer, "MAD temp.y, -coef.y, chroma.w, yv12_coef.z;\n");
6911         shader_addline(buffer, "MIN texcrd.y, temp.y, texcrd.y;\n");
6912     } else {
6913         /* Read from [size - size+size/4] */
6914         shader_addline(buffer, "FLR texcrd.y, texcrd.y;\n");
6915         shader_addline(buffer, "MAD texcrd.y, texcrd.y, coef.w, size.y;\n");
6916
6917         /* Read odd lines from the right side(add size * 0.5 to the x coordinate */
6918         shader_addline(buffer, "ADD texcrd2.x, texcrd.y, yv12_coef.y;\n"); /* To avoid 0.5 == 0.5 comparisons */
6919         shader_addline(buffer, "FRC texcrd2.x, texcrd2.x;\n");
6920         shader_addline(buffer, "SGE texcrd2.x, texcrd2.x, coef.y;\n");
6921         shader_addline(buffer, "MUL texcrd2.x, texcrd2.x, size.x;\n");
6922         shader_addline(buffer, "MAD texcrd.x, texcrd2.x, coef.y, texcrd.x;\n");
6923
6924         /* Make sure to read exactly from the pixel center */
6925         shader_addline(buffer, "FLR texcrd.y, texcrd.y;\n");
6926         shader_addline(buffer, "ADD texcrd.y, texcrd.y, coef.y;\n");
6927
6928         /* Clamp */
6929         shader_addline(buffer, "MAD temp.y, size.y, coef.w, size.y;\n");
6930         shader_addline(buffer, "ADD temp.y, temp.y, -coef.y;\n");
6931         shader_addline(buffer, "MIN texcrd.y, temp.y, texcrd.y;\n");
6932         shader_addline(buffer, "ADD temp.y, size.y, -coef.y;\n");
6933         shader_addline(buffer, "MAX texcrd.y, temp.y, texcrd.y;\n");
6934     }
6935     /* Read the texture, put the result into the output register */
6936     shader_addline(buffer, "TEX temp, texcrd, texture[0], %s;\n", tex);
6937     shader_addline(buffer, "MOV chroma.x, temp.w;\n");
6938
6939     /* The other chroma value is 1/6th of the texture lower, from 5/6th to 6/6th
6940      * No need to clamp because we're just reusing the already clamped value from above
6941      */
6942     if(textype == GL_TEXTURE_2D) {
6943         shader_addline(buffer, "ADD texcrd.y, texcrd.y, yv12_coef.y;\n");
6944     } else {
6945         shader_addline(buffer, "MAD texcrd.y, size.y, coef.w, texcrd.y;\n");
6946     }
6947     shader_addline(buffer, "TEX temp, texcrd, texture[0], %s;\n", tex);
6948     shader_addline(buffer, "MOV chroma.y, temp.w;\n");
6949
6950     /* Sample the luminance value. It is in the top 2/3rd of the texture, so scale the y coordinate.
6951      * Clamp the y coordinate to prevent the chroma values from bleeding into the sampled luminance
6952      * values due to filtering
6953      */
6954     shader_addline(buffer, "MOV texcrd, fragment.texcoord[0];\n");
6955     if(textype == GL_TEXTURE_2D) {
6956         /* Multiply the y coordinate by 2/3 and clamp it */
6957         shader_addline(buffer, "MUL texcrd.y, texcrd.y, yv12_coef.x;\n");
6958         shader_addline(buffer, "MAD temp.y, -coef.y, chroma.w, yv12_coef.x;\n");
6959         shader_addline(buffer, "MIN texcrd.y, temp.y, texcrd.y;\n");
6960         shader_addline(buffer, "TEX luminance, texcrd, texture[0], %s;\n", tex);
6961     } else {
6962         /* Reading from texture_rectangles is pretty straightforward, just use the unmodified
6963          * texture coordinate. It is still a good idea to clamp it though, since the opengl texture
6964          * is bigger
6965          */
6966         shader_addline(buffer, "ADD temp.x, size.y, -coef.y;\n");
6967         shader_addline(buffer, "MIN texcrd.y, texcrd.y, size.x;\n");
6968         shader_addline(buffer, "TEX luminance, texcrd, texture[0], %s;\n", tex);
6969     }
6970     *luminance = 'a';
6971
6972     return TRUE;
6973 }
6974
6975 static GLuint gen_p8_shader(struct arbfp_blit_priv *priv,
6976         const struct wined3d_gl_info *gl_info, GLenum textype)
6977 {
6978     GLenum shader;
6979     struct wined3d_shader_buffer buffer;
6980     GLint pos;
6981
6982     /* Shader header */
6983     if (!shader_buffer_init(&buffer))
6984     {
6985         ERR("Failed to initialize shader buffer.\n");
6986         return 0;
6987     }
6988
6989     ENTER_GL();
6990     GL_EXTCALL(glGenProgramsARB(1, &shader));
6991     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader));
6992     LEAVE_GL();
6993     if(!shader) {
6994         shader_buffer_free(&buffer);
6995         return 0;
6996     }
6997
6998     shader_addline(&buffer, "!!ARBfp1.0\n");
6999     shader_addline(&buffer, "TEMP index;\n");
7000
7001     /* { 255/256, 0.5/255*255/256, 0, 0 } */
7002     shader_addline(&buffer, "PARAM constants = { 0.996, 0.00195, 0, 0 };\n");
7003
7004     /* The alpha-component contains the palette index */
7005     if(textype == GL_TEXTURE_RECTANGLE_ARB)
7006         shader_addline(&buffer, "TXP index, fragment.texcoord[0], texture[0], RECT;\n");
7007     else
7008         shader_addline(&buffer, "TEX index, fragment.texcoord[0], texture[0], 2D;\n");
7009
7010     /* Scale the index by 255/256 and add a bias of '0.5' in order to sample in the middle */
7011     shader_addline(&buffer, "MAD index.a, index.a, constants.x, constants.y;\n");
7012
7013     /* Use the alpha-component as an index in the palette to get the final color */
7014     shader_addline(&buffer, "TEX result.color, index.a, texture[1], 1D;\n");
7015     shader_addline(&buffer, "END\n");
7016
7017     ENTER_GL();
7018     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
7019             strlen(buffer.buffer), buffer.buffer));
7020     checkGLcall("glProgramStringARB()");
7021
7022     gl_info->gl_ops.gl.p_glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
7023     if (pos != -1)
7024     {
7025         FIXME("Fragment program error at position %d: %s\n\n", pos,
7026               debugstr_a((const char *)gl_info->gl_ops.gl.p_glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
7027         shader_arb_dump_program_source(buffer.buffer);
7028     }
7029
7030     if (textype == GL_TEXTURE_RECTANGLE_ARB)
7031         priv->p8_rect_shader = shader;
7032     else
7033         priv->p8_2d_shader = shader;
7034
7035     shader_buffer_free(&buffer);
7036     LEAVE_GL();
7037
7038     return shader;
7039 }
7040
7041 /* Context activation is done by the caller. */
7042 static void upload_palette(const struct wined3d_surface *surface, struct wined3d_context *context)
7043 {
7044     BYTE table[256][4];
7045     struct wined3d_device *device = surface->resource.device;
7046     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
7047     struct arbfp_blit_priv *priv = device->blit_priv;
7048     BOOL colorkey = (surface->CKeyFlags & WINEDDSD_CKSRCBLT) != 0;
7049
7050     d3dfmt_p8_init_palette(surface, table, colorkey);
7051
7052     ENTER_GL();
7053
7054     if (gl_info->supported[APPLE_CLIENT_STORAGE])
7055     {
7056         gl_info->gl_ops.gl.p_glPixelStorei(GL_UNPACK_CLIENT_STORAGE_APPLE, GL_FALSE);
7057         checkGLcall("glPixelStorei(GL_UNPACK_CLIENT_STORAGE_APPLE, GL_FALSE)");
7058     }
7059
7060     if (!priv->palette_texture)
7061         gl_info->gl_ops.gl.p_glGenTextures(1, &priv->palette_texture);
7062
7063     GL_EXTCALL(glActiveTextureARB(GL_TEXTURE1));
7064     gl_info->gl_ops.gl.p_glBindTexture(GL_TEXTURE_1D, priv->palette_texture);
7065
7066     gl_info->gl_ops.gl.p_glTexEnvi(GL_TEXTURE_ENV, GL_TEXTURE_ENV_MODE, GL_REPLACE);
7067
7068     gl_info->gl_ops.gl.p_glTexParameteri(GL_TEXTURE_1D, GL_TEXTURE_MIN_FILTER, GL_NEAREST);
7069     /* Make sure we have discrete color levels. */
7070     gl_info->gl_ops.gl.p_glTexParameteri(GL_TEXTURE_1D, GL_TEXTURE_MAG_FILTER, GL_NEAREST);
7071     gl_info->gl_ops.gl.p_glTexParameteri(GL_TEXTURE_1D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
7072     /* Upload the palette */
7073     /* TODO: avoid unneeded uploads in the future by adding some SFLAG_PALETTE_DIRTY mechanism */
7074     gl_info->gl_ops.gl.p_glTexImage1D(GL_TEXTURE_1D, 0, GL_RGBA, 256, 0, GL_RGBA, GL_UNSIGNED_BYTE, table);
7075
7076     if (gl_info->supported[APPLE_CLIENT_STORAGE])
7077     {
7078         gl_info->gl_ops.gl.p_glPixelStorei(GL_UNPACK_CLIENT_STORAGE_APPLE, GL_TRUE);
7079         checkGLcall("glPixelStorei(GL_UNPACK_CLIENT_STORAGE_APPLE, GL_TRUE)");
7080     }
7081
7082     /* Switch back to unit 0 in which the 2D texture will be stored. */
7083     context_active_texture(context, gl_info, 0);
7084     LEAVE_GL();
7085 }
7086
7087 /* Context activation is done by the caller. */
7088 static GLuint gen_yuv_shader(struct arbfp_blit_priv *priv, const struct wined3d_gl_info *gl_info,
7089         enum complex_fixup yuv_fixup, GLenum textype)
7090 {
7091     GLenum shader;
7092     struct wined3d_shader_buffer buffer;
7093     char luminance_component;
7094     GLint pos;
7095
7096     /* Shader header */
7097     if (!shader_buffer_init(&buffer))
7098     {
7099         ERR("Failed to initialize shader buffer.\n");
7100         return 0;
7101     }
7102
7103     ENTER_GL();
7104     GL_EXTCALL(glGenProgramsARB(1, &shader));
7105     checkGLcall("GL_EXTCALL(glGenProgramsARB(1, &shader))");
7106     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader));
7107     checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader)");
7108     LEAVE_GL();
7109     if(!shader) {
7110         shader_buffer_free(&buffer);
7111         return 0;
7112     }
7113
7114     /* The YUY2 and UYVY formats contain two pixels packed into a 32 bit macropixel,
7115      * giving effectively 16 bit per pixel. The color consists of a luminance(Y) and
7116      * two chroma(U and V) values. Each macropixel has two luminance values, one for
7117      * each single pixel it contains, and one U and one V value shared between both
7118      * pixels.
7119      *
7120      * The data is loaded into an A8L8 texture. With YUY2, the luminance component
7121      * contains the luminance and alpha the chroma. With UYVY it is vice versa. Thus
7122      * take the format into account when generating the read swizzles
7123      *
7124      * Reading the Y value is straightforward - just sample the texture. The hardware
7125      * takes care of filtering in the horizontal and vertical direction.
7126      *
7127      * Reading the U and V values is harder. We have to avoid filtering horizontally,
7128      * because that would mix the U and V values of one pixel or two adjacent pixels.
7129      * Thus floor the texture coordinate and add 0.5 to get an unfiltered read,
7130      * regardless of the filtering setting. Vertical filtering works automatically
7131      * though - the U and V values of two rows are mixed nicely.
7132      *
7133      * Apart of avoiding filtering issues, the code has to know which value it just
7134      * read, and where it can find the other one. To determine this, it checks if
7135      * it sampled an even or odd pixel, and shifts the 2nd read accordingly.
7136      *
7137      * Handling horizontal filtering of U and V values requires reading a 2nd pair
7138      * of pixels, extracting U and V and mixing them. This is not implemented yet.
7139      *
7140      * An alternative implementation idea is to load the texture as A8R8G8B8 texture,
7141      * with width / 2. This way one read gives all 3 values, finding U and V is easy
7142      * in an unfiltered situation. Finding the luminance on the other hand requires
7143      * finding out if it is an odd or even pixel. The real drawback of this approach
7144      * is filtering. This would have to be emulated completely in the shader, reading
7145      * up two 2 packed pixels in up to 2 rows and interpolating both horizontally and
7146      * vertically. Beyond that it would require adjustments to the texture handling
7147      * code to deal with the width scaling
7148      */
7149     shader_addline(&buffer, "!!ARBfp1.0\n");
7150     shader_addline(&buffer, "TEMP luminance;\n");
7151     shader_addline(&buffer, "TEMP temp;\n");
7152     shader_addline(&buffer, "TEMP chroma;\n");
7153     shader_addline(&buffer, "TEMP texcrd;\n");
7154     shader_addline(&buffer, "TEMP texcrd2;\n");
7155     shader_addline(&buffer, "PARAM coef = {1.0, 0.5, 2.0, 0.25};\n");
7156     shader_addline(&buffer, "PARAM yuv_coef = {1.403, 0.344, 0.714, 1.770};\n");
7157     shader_addline(&buffer, "PARAM size = program.local[0];\n");
7158
7159     switch (yuv_fixup)
7160     {
7161         case COMPLEX_FIXUP_UYVY:
7162         case COMPLEX_FIXUP_YUY2:
7163             if (!gen_planar_yuv_read(&buffer, yuv_fixup, textype, &luminance_component))
7164             {
7165                 shader_buffer_free(&buffer);
7166                 return 0;
7167             }
7168             break;
7169
7170         case COMPLEX_FIXUP_YV12:
7171             if (!gen_yv12_read(&buffer, textype, &luminance_component))
7172             {
7173                 shader_buffer_free(&buffer);
7174                 return 0;
7175             }
7176             break;
7177
7178         default:
7179             FIXME("Unsupported YUV fixup %#x\n", yuv_fixup);
7180             shader_buffer_free(&buffer);
7181             return 0;
7182     }
7183
7184     /* Calculate the final result. Formula is taken from
7185      * http://www.fourcc.org/fccyvrgb.php. Note that the chroma
7186      * ranges from -0.5 to 0.5
7187      */
7188     shader_addline(&buffer, "SUB chroma.xy, chroma, coef.y;\n");
7189
7190     shader_addline(&buffer, "MAD result.color.x, chroma.x, yuv_coef.x, luminance.%c;\n", luminance_component);
7191     shader_addline(&buffer, "MAD temp.x, -chroma.y, yuv_coef.y, luminance.%c;\n", luminance_component);
7192     shader_addline(&buffer, "MAD result.color.y, -chroma.x, yuv_coef.z, temp.x;\n");
7193     shader_addline(&buffer, "MAD result.color.z, chroma.y, yuv_coef.w, luminance.%c;\n", luminance_component);
7194     shader_addline(&buffer, "END\n");
7195
7196     ENTER_GL();
7197     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
7198             strlen(buffer.buffer), buffer.buffer));
7199     checkGLcall("glProgramStringARB()");
7200
7201     gl_info->gl_ops.gl.p_glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
7202     if (pos != -1)
7203     {
7204         FIXME("Fragment program error at position %d: %s\n\n", pos,
7205               debugstr_a((const char *)gl_info->gl_ops.gl.p_glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
7206         shader_arb_dump_program_source(buffer.buffer);
7207     }
7208     else
7209     {
7210         GLint native;
7211
7212         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
7213         checkGLcall("glGetProgramivARB()");
7214         if (!native) WARN("Program exceeds native resource limits.\n");
7215     }
7216
7217     shader_buffer_free(&buffer);
7218     LEAVE_GL();
7219
7220     switch (yuv_fixup)
7221     {
7222         case COMPLEX_FIXUP_YUY2:
7223             if (textype == GL_TEXTURE_RECTANGLE_ARB) priv->yuy2_rect_shader = shader;
7224             else priv->yuy2_2d_shader = shader;
7225             break;
7226
7227         case COMPLEX_FIXUP_UYVY:
7228             if (textype == GL_TEXTURE_RECTANGLE_ARB) priv->uyvy_rect_shader = shader;
7229             else priv->uyvy_2d_shader = shader;
7230             break;
7231
7232         case COMPLEX_FIXUP_YV12:
7233             if (textype == GL_TEXTURE_RECTANGLE_ARB) priv->yv12_rect_shader = shader;
7234             else priv->yv12_2d_shader = shader;
7235             break;
7236         default:
7237             ERR("Unsupported complex fixup: %d\n", yuv_fixup);
7238     }
7239
7240     return shader;
7241 }
7242
7243 /* Context activation is done by the caller. */
7244 static HRESULT arbfp_blit_set(void *blit_priv, struct wined3d_context *context, const struct wined3d_surface *surface)
7245 {
7246     GLenum shader;
7247     float size[4] = {(float) surface->pow2Width, (float) surface->pow2Height, 1.0f, 1.0f};
7248     struct arbfp_blit_priv *priv = blit_priv;
7249     enum complex_fixup fixup;
7250     const struct wined3d_gl_info *gl_info = context->gl_info;
7251     GLenum textype;
7252
7253     if (surface->container.type == WINED3D_CONTAINER_TEXTURE)
7254         textype = surface->container.u.texture->target;
7255     else
7256         textype = surface->texture_target;
7257
7258     if (surface->flags & SFLAG_CONVERTED)
7259     {
7260         ENTER_GL();
7261         gl_info->gl_ops.gl.p_glEnable(textype);
7262         checkGLcall("glEnable(textype)");
7263         LEAVE_GL();
7264         return WINED3D_OK;
7265     }
7266
7267     if (!is_complex_fixup(surface->resource.format->color_fixup))
7268     {
7269         TRACE("Fixup:\n");
7270         dump_color_fixup_desc(surface->resource.format->color_fixup);
7271         /* Don't bother setting up a shader for unconverted formats */
7272         ENTER_GL();
7273         gl_info->gl_ops.gl.p_glEnable(textype);
7274         checkGLcall("glEnable(textype)");
7275         LEAVE_GL();
7276         return WINED3D_OK;
7277     }
7278
7279     fixup = get_complex_fixup(surface->resource.format->color_fixup);
7280
7281     switch(fixup)
7282     {
7283         case COMPLEX_FIXUP_YUY2:
7284             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->yuy2_rect_shader : priv->yuy2_2d_shader;
7285             break;
7286
7287         case COMPLEX_FIXUP_UYVY:
7288             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->uyvy_rect_shader : priv->uyvy_2d_shader;
7289             break;
7290
7291         case COMPLEX_FIXUP_YV12:
7292             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->yv12_rect_shader : priv->yv12_2d_shader;
7293             break;
7294
7295         case COMPLEX_FIXUP_P8:
7296             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->p8_rect_shader : priv->p8_2d_shader;
7297             if (!shader) shader = gen_p8_shader(priv, gl_info, textype);
7298
7299             upload_palette(surface, context);
7300             break;
7301
7302         default:
7303             FIXME("Unsupported complex fixup %#x, not setting a shader\n", fixup);
7304             ENTER_GL();
7305             gl_info->gl_ops.gl.p_glEnable(textype);
7306             checkGLcall("glEnable(textype)");
7307             LEAVE_GL();
7308             return E_NOTIMPL;
7309     }
7310
7311     if (!shader) shader = gen_yuv_shader(priv, gl_info, fixup, textype);
7312
7313     ENTER_GL();
7314     gl_info->gl_ops.gl.p_glEnable(GL_FRAGMENT_PROGRAM_ARB);
7315     checkGLcall("glEnable(GL_FRAGMENT_PROGRAM_ARB)");
7316     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader));
7317     checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader)");
7318     GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, 0, size));
7319     checkGLcall("glProgramLocalParameter4fvARB");
7320     LEAVE_GL();
7321
7322     return WINED3D_OK;
7323 }
7324
7325 /* Context activation is done by the caller. */
7326 static void arbfp_blit_unset(const struct wined3d_gl_info *gl_info)
7327 {
7328     ENTER_GL();
7329     gl_info->gl_ops.gl.p_glDisable(GL_FRAGMENT_PROGRAM_ARB);
7330     checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
7331     gl_info->gl_ops.gl.p_glDisable(GL_TEXTURE_2D);
7332     checkGLcall("glDisable(GL_TEXTURE_2D)");
7333     if (gl_info->supported[ARB_TEXTURE_CUBE_MAP])
7334     {
7335         gl_info->gl_ops.gl.p_glDisable(GL_TEXTURE_CUBE_MAP_ARB);
7336         checkGLcall("glDisable(GL_TEXTURE_CUBE_MAP_ARB)");
7337     }
7338     if (gl_info->supported[ARB_TEXTURE_RECTANGLE])
7339     {
7340         gl_info->gl_ops.gl.p_glDisable(GL_TEXTURE_RECTANGLE_ARB);
7341         checkGLcall("glDisable(GL_TEXTURE_RECTANGLE_ARB)");
7342     }
7343     LEAVE_GL();
7344 }
7345
7346 static BOOL arbfp_blit_supported(const struct wined3d_gl_info *gl_info, enum wined3d_blit_op blit_op,
7347         const RECT *src_rect, DWORD src_usage, enum wined3d_pool src_pool, const struct wined3d_format *src_format,
7348         const RECT *dst_rect, DWORD dst_usage, enum wined3d_pool dst_pool, const struct wined3d_format *dst_format)
7349 {
7350     enum complex_fixup src_fixup;
7351
7352     if (!gl_info->supported[ARB_FRAGMENT_PROGRAM])
7353         return FALSE;
7354
7355     if (blit_op != WINED3D_BLIT_OP_COLOR_BLIT)
7356     {
7357         TRACE("Unsupported blit_op=%d\n", blit_op);
7358         return FALSE;
7359     }
7360
7361     if (src_pool == WINED3D_POOL_SYSTEM_MEM || dst_pool == WINED3D_POOL_SYSTEM_MEM)
7362         return FALSE;
7363
7364     src_fixup = get_complex_fixup(src_format->color_fixup);
7365     if (TRACE_ON(d3d_shader) && TRACE_ON(d3d))
7366     {
7367         TRACE("Checking support for fixup:\n");
7368         dump_color_fixup_desc(src_format->color_fixup);
7369     }
7370
7371     if (!is_identity_fixup(dst_format->color_fixup))
7372     {
7373         TRACE("Destination fixups are not supported\n");
7374         return FALSE;
7375     }
7376
7377     if (is_identity_fixup(src_format->color_fixup))
7378     {
7379         TRACE("[OK]\n");
7380         return TRUE;
7381     }
7382
7383      /* We only support YUV conversions. */
7384     if (!is_complex_fixup(src_format->color_fixup))
7385     {
7386         TRACE("[FAILED]\n");
7387         return FALSE;
7388     }
7389
7390     switch(src_fixup)
7391     {
7392         case COMPLEX_FIXUP_YUY2:
7393         case COMPLEX_FIXUP_UYVY:
7394         case COMPLEX_FIXUP_YV12:
7395         case COMPLEX_FIXUP_P8:
7396             TRACE("[OK]\n");
7397             return TRUE;
7398
7399         default:
7400             FIXME("Unsupported YUV fixup %#x\n", src_fixup);
7401             TRACE("[FAILED]\n");
7402             return FALSE;
7403     }
7404 }
7405
7406 HRESULT arbfp_blit_surface(struct wined3d_device *device, DWORD filter,
7407         struct wined3d_surface *src_surface, const RECT *src_rect_in,
7408         struct wined3d_surface *dst_surface, const RECT *dst_rect_in)
7409 {
7410     struct wined3d_context *context;
7411     RECT src_rect = *src_rect_in;
7412     RECT dst_rect = *dst_rect_in;
7413
7414     /* Now load the surface */
7415     if (wined3d_settings.offscreen_rendering_mode != ORM_FBO
7416             && (src_surface->flags & (SFLAG_INTEXTURE | SFLAG_INDRAWABLE)) == SFLAG_INDRAWABLE)
7417     {
7418         /* Without FBO blits transferring from the drawable to the texture is
7419          * expensive, because we have to flip the data in sysmem. Since we can
7420          * flip in the blitter, we don't actually need that flip anyway. So we
7421          * use the surface's texture as scratch texture, and flip the source
7422          * rectangle instead. */
7423         surface_load_fb_texture(src_surface, FALSE);
7424
7425         src_rect.top = src_surface->resource.height - src_rect.top;
7426         src_rect.bottom = src_surface->resource.height - src_rect.bottom;
7427     }
7428     else
7429         surface_internal_preload(src_surface, SRGB_RGB);
7430
7431     /* Activate the destination context, set it up for blitting */
7432     context = context_acquire(device, dst_surface);
7433     context_apply_blit_state(context, device);
7434
7435     if (!surface_is_offscreen(dst_surface))
7436         surface_translate_drawable_coords(dst_surface, context->win_handle, &dst_rect);
7437
7438     arbfp_blit_set(device->blit_priv, context, src_surface);
7439
7440     ENTER_GL();
7441
7442     /* Draw a textured quad */
7443     draw_textured_quad(src_surface, context, &src_rect, &dst_rect, filter);
7444
7445     LEAVE_GL();
7446
7447     /* Leave the opengl state valid for blitting */
7448     arbfp_blit_unset(context->gl_info);
7449
7450     if (wined3d_settings.strict_draw_ordering
7451             || (dst_surface->container.type == WINED3D_CONTAINER_SWAPCHAIN
7452             && (dst_surface->container.u.swapchain->front_buffer == dst_surface)))
7453         context->gl_info->gl_ops.gl.p_glFlush(); /* Flush to ensure ordering across contexts. */
7454
7455     context_release(context);
7456
7457     surface_modify_location(dst_surface, dst_surface->draw_binding, TRUE);
7458     return WINED3D_OK;
7459 }
7460
7461 /* Do not call while under the GL lock. */
7462 static HRESULT arbfp_blit_color_fill(struct wined3d_device *device, struct wined3d_surface *dst_surface,
7463         const RECT *dst_rect, const struct wined3d_color *color)
7464 {
7465     FIXME("Color filling not implemented by arbfp_blit\n");
7466     return WINED3DERR_INVALIDCALL;
7467 }
7468
7469 /* Do not call while under the GL lock. */
7470 static HRESULT arbfp_blit_depth_fill(struct wined3d_device *device,
7471         struct wined3d_surface *surface, const RECT *rect, float depth)
7472 {
7473     FIXME("Depth filling not implemented by arbfp_blit.\n");
7474     return WINED3DERR_INVALIDCALL;
7475 }
7476
7477 const struct blit_shader arbfp_blit = {
7478     arbfp_blit_alloc,
7479     arbfp_blit_free,
7480     arbfp_blit_set,
7481     arbfp_blit_unset,
7482     arbfp_blit_supported,
7483     arbfp_blit_color_fill,
7484     arbfp_blit_depth_fill,
7485 };