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
  32 #include <math.h>
  33 #include <stdio.h>
  34
  35 #include "wined3d_private.h"
  36
  37 WINE_DEFAULT_DEBUG_CHANNEL(d3d_shader);
  38 WINE_DECLARE_DEBUG_CHANNEL(d3d_constants);
  39 WINE_DECLARE_DEBUG_CHANNEL(d3d_caps);
  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     BOOL                            has_signature_idx;
 289     DWORD                           input_signature_idx;
 290     DWORD                           clipplane_emulation;
 291     BOOL                            clamp_consts;
 292 };
 293
 294 struct arb_vshader_private {
 295     struct arb_vs_compiled_shader   *gl_shaders;
 296     UINT                            num_gl_shaders, shader_array_size;
 297     UINT rel_offset;
 298 };
 299
 300 struct shader_arb_priv
 301 {
 302     GLuint                  current_vprogram_id;
 303     GLuint                  current_fprogram_id;
 304     const struct arb_ps_compiled_shader *compiled_fprog;
 305     const struct arb_vs_compiled_shader *compiled_vprog;
 306     GLuint                  depth_blt_vprogram_id;
 307     GLuint                  depth_blt_fprogram_id_full[tex_type_count];
 308     GLuint                  depth_blt_fprogram_id_masked[tex_type_count];
 309     BOOL                    use_arbfp_fixed_func;
 310     struct wine_rb_tree     fragment_shaders;
 311     BOOL                    last_ps_const_clamped;
 312     BOOL                    last_vs_color_unclamp;
 313
 314     struct wine_rb_tree     signature_tree;
 315     DWORD ps_sig_number;
 316 };
 317
 318 /* GL locking for state handlers is done by the caller. */
 319 static BOOL need_rel_addr_const(const struct arb_vshader_private *shader_data,
 320         const struct wined3d_shader_reg_maps *reg_maps, const struct wined3d_gl_info *gl_info)
 321 {
 322     if (shader_data->rel_offset) return TRUE;
 323     if (!reg_maps->usesmova) return FALSE;
 324     return !gl_info->supported[NV_VERTEX_PROGRAM2_OPTION];
 325 }
 326
 327 /* Returns TRUE if result.clip from GL_NV_vertex_program2 should be used and FALSE otherwise */
 328 static inline BOOL use_nv_clip(const struct wined3d_gl_info *gl_info)
 329 {
 330     return gl_info->supported[NV_VERTEX_PROGRAM2_OPTION]
 331             && !(gl_info->quirks & WINED3D_QUIRK_NV_CLIP_BROKEN);
 332 }
 333
 334 static BOOL need_helper_const(const struct arb_vshader_private *shader_data,
 335         const struct wined3d_shader_reg_maps *reg_maps, const struct wined3d_gl_info *gl_info)
 336 {
 337     if (need_rel_addr_const(shader_data, reg_maps, gl_info)) return TRUE;
 338     if (!gl_info->supported[NV_VERTEX_PROGRAM]) return TRUE; /* Need to init colors. */
 339     if (gl_info->quirks & WINED3D_QUIRK_ARB_VS_OFFSET_LIMIT) return TRUE; /* Load the immval offset. */
 340     if (gl_info->quirks & WINED3D_QUIRK_SET_TEXCOORD_W) return TRUE; /* Have to init texcoords. */
 341     if (!use_nv_clip(gl_info)) return TRUE; /* Init the clip texcoord */
 342     if (reg_maps->usesnrm) return TRUE; /* 0.0 */
 343     if (reg_maps->usespow) return TRUE; /* EPS, 0.0 and 1.0 */
 344     return FALSE;
 345 }
 346
 347 static unsigned int reserved_vs_const(const struct arb_vshader_private *shader_data,
 348         const struct wined3d_shader_reg_maps *reg_maps, const struct wined3d_gl_info *gl_info)
 349 {
 350     unsigned int ret = 1;
 351     /* We use one PARAM for the pos fixup, and in some cases one to load
 352      * some immediate values into the shader. */
 353     if (need_helper_const(shader_data, reg_maps, gl_info)) ++ret;
 354     if (need_rel_addr_const(shader_data, reg_maps, gl_info)) ++ret;
 355     return ret;
 356 }
 357
 358 /* Loads floating point constants into the currently set ARB_vertex/fragment_program.
 359  * When constant_list == NULL, it will load all the constants.
 360  *
 361  * @target_type should be either GL_VERTEX_PROGRAM_ARB (for vertex shaders)
 362  *  or GL_FRAGMENT_PROGRAM_ARB (for pixel shaders)
 363  */
 364 /* GL locking is done by the caller */
 365 static unsigned int shader_arb_load_constantsF(struct wined3d_shader *shader, const struct wined3d_gl_info *gl_info,
 366         GLuint target_type, unsigned int max_constants, const float *constants, char *dirty_consts)
 367 {
 368     local_constant* lconst;
 369     DWORD i, j;
 370     unsigned int ret;
 371
 372     if (TRACE_ON(d3d_constants))
 373     {
 374         for(i = 0; i < max_constants; i++) {
 375             if(!dirty_consts[i]) continue;
 376             TRACE_(d3d_constants)("Loading constants %i: %f, %f, %f, %f\n", i,
 377                         constants[i * 4 + 0], constants[i * 4 + 1],
 378                         constants[i * 4 + 2], constants[i * 4 + 3]);
 379         }
 380     }
 381
 382     i = 0;
 383
 384     /* In 1.X pixel shaders constants are implicitly clamped in the range [-1;1] */
 385     if (target_type == GL_FRAGMENT_PROGRAM_ARB && shader->reg_maps.shader_version.major == 1)
 386     {
 387         float lcl_const[4];
 388         /* ps 1.x supports only 8 constants, clamp only those. When switching between 1.x and higher
 389          * shaders, the first 8 constants are marked dirty for reload
 390          */
 391         for(; i < min(8, max_constants); i++) {
 392             if(!dirty_consts[i]) continue;
 393             dirty_consts[i] = 0;
 394
 395             j = 4 * i;
 396             if (constants[j + 0] > 1.0f) lcl_const[0] = 1.0f;
 397             else if (constants[j + 0] < -1.0f) lcl_const[0] = -1.0f;
 398             else lcl_const[0] = constants[j + 0];
 399
 400             if (constants[j + 1] > 1.0f) lcl_const[1] = 1.0f;
 401             else if (constants[j + 1] < -1.0f) lcl_const[1] = -1.0f;
 402             else lcl_const[1] = constants[j + 1];
 403
 404             if (constants[j + 2] > 1.0f) lcl_const[2] = 1.0f;
 405             else if (constants[j + 2] < -1.0f) lcl_const[2] = -1.0f;
 406             else lcl_const[2] = constants[j + 2];
 407
 408             if (constants[j + 3] > 1.0f) lcl_const[3] = 1.0f;
 409             else if (constants[j + 3] < -1.0f) lcl_const[3] = -1.0f;
 410             else lcl_const[3] = constants[j + 3];
 411
 412             GL_EXTCALL(glProgramEnvParameter4fvARB(target_type, i, lcl_const));
 413         }
 414
 415         /* If further constants are dirty, reload them without clamping.
 416          *
 417          * The alternative is not to touch them, but then we cannot reset the dirty constant count
 418          * to zero. That's bad for apps that only use PS 1.x shaders, because in that case the code
 419          * above would always re-check the first 8 constants since max_constant remains at the init
 420          * value
 421          */
 422     }
 423
 424     if (gl_info->supported[EXT_GPU_PROGRAM_PARAMETERS])
 425     {
 426         /* TODO: Benchmark if we're better of with finding the dirty constants ourselves,
 427          * or just reloading *all* constants at once
 428          *
 429         GL_EXTCALL(glProgramEnvParameters4fvEXT(target_type, i, max_constants, constants + (i * 4)));
 430          */
 431         for(; i < max_constants; i++) {
 432             if(!dirty_consts[i]) continue;
 433
 434             /* Find the next block of dirty constants */
 435             dirty_consts[i] = 0;
 436             j = i;
 437             for(i++; (i < max_constants) && dirty_consts[i]; i++) {
 438                 dirty_consts[i] = 0;
 439             }
 440
 441             GL_EXTCALL(glProgramEnvParameters4fvEXT(target_type, j, i - j, constants + (j * 4)));
 442         }
 443     } else {
 444         for(; i < max_constants; i++) {
 445             if(dirty_consts[i]) {
 446                 dirty_consts[i] = 0;
 447                 GL_EXTCALL(glProgramEnvParameter4fvARB(target_type, i, constants + (i * 4)));
 448             }
 449         }
 450     }
 451     checkGLcall("glProgramEnvParameter4fvARB()");
 452
 453     /* Load immediate constants */
 454     if (shader->load_local_constsF)
 455     {
 456         if (TRACE_ON(d3d_shader))
 457         {
 458             LIST_FOR_EACH_ENTRY(lconst, &shader->constantsF, local_constant, entry)
 459             {
 460                 GLfloat* values = (GLfloat*)lconst->value;
 461                 TRACE_(d3d_constants)("Loading local constants %i: %f, %f, %f, %f\n", lconst->idx,
 462                         values[0], values[1], values[2], values[3]);
 463             }
 464         }
 465         /* Immediate constants are clamped for 1.X shaders at loading times */
 466         ret = 0;
 467         LIST_FOR_EACH_ENTRY(lconst, &shader->constantsF, local_constant, entry)
 468         {
 469             dirty_consts[lconst->idx] = 1; /* Dirtify so the non-immediate constant overwrites it next time */
 470             ret = max(ret, lconst->idx + 1);
 471             GL_EXTCALL(glProgramEnvParameter4fvARB(target_type, lconst->idx, (GLfloat*)lconst->value));
 472         }
 473         checkGLcall("glProgramEnvParameter4fvARB()");
 474         return ret; /* The loaded immediate constants need reloading for the next shader */
 475     } else {
 476         return 0; /* No constants are dirty now */
 477     }
 478 }
 479
 480 /**
 481  * Loads the texture dimensions for NP2 fixup into the currently set ARB_[vertex/fragment]_programs.
 482  */
 483 static void shader_arb_load_np2fixup_constants(void *shader_priv,
 484         const struct wined3d_gl_info *gl_info, const struct wined3d_state *state)
 485 {
 486     const struct shader_arb_priv * priv = shader_priv;
 487
 488     /* NP2 texcoord fixup is (currently) only done for pixelshaders. */
 489     if (!use_ps(state)) return;
 490
 491     if (priv->compiled_fprog && priv->compiled_fprog->np2fixup_info.super.active) {
 492         const struct arb_ps_np2fixup_info* const fixup = &priv->compiled_fprog->np2fixup_info;
 493         UINT i;
 494         WORD active = fixup->super.active;
 495         GLfloat np2fixup_constants[4 * MAX_FRAGMENT_SAMPLERS];
 496
 497         for (i = 0; active; active >>= 1, ++i)
 498         {
 499             const struct wined3d_texture *tex = state->textures[i];
 500             const unsigned char idx = fixup->super.idx[i];
 501             GLfloat *tex_dim = &np2fixup_constants[(idx >> 1) * 4];
 502
 503             if (!(active & 1)) continue;
 504
 505             if (!tex) {
 506                 FIXME("Nonexistent texture is flagged for NP2 texcoord fixup\n");
 507                 continue;
 508             }
 509
 510             if (idx % 2)
 511             {
 512                 tex_dim[2] = tex->pow2_matrix[0];
 513                 tex_dim[3] = tex->pow2_matrix[5];
 514             }
 515             else
 516             {
 517                 tex_dim[0] = tex->pow2_matrix[0];
 518                 tex_dim[1] = tex->pow2_matrix[5];
 519             }
 520         }
 521
 522         for (i = 0; i < fixup->super.num_consts; ++i) {
 523             GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB,
 524                                                    fixup->offset + i, &np2fixup_constants[i * 4]));
 525         }
 526     }
 527 }
 528
 529 /* GL locking is done by the caller. */
 530 static void shader_arb_ps_local_constants(const struct arb_ps_compiled_shader *gl_shader,
 531         const struct wined3d_context *context, const struct wined3d_state *state, float rt_height)
 532 {
 533     const struct wined3d_gl_info *gl_info = context->gl_info;
 534     unsigned char i;
 535
 536     for(i = 0; i < gl_shader->numbumpenvmatconsts; i++)
 537     {
 538         int texunit = gl_shader->bumpenvmatconst[i].texunit;
 539
 540         /* The state manager takes care that this function is always called if the bump env matrix changes */
 541         const float *data = (const float *)&state->texture_states[texunit][WINED3DTSS_BUMPENVMAT00];
 542         GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB,
 543                 gl_shader->bumpenvmatconst[i].const_num, data));
 544
 545         if (gl_shader->luminanceconst[i].const_num != WINED3D_CONST_NUM_UNUSED)
 546         {
 547             /* WINED3DTSS_BUMPENVLSCALE and WINED3DTSS_BUMPENVLOFFSET are next to each other.
 548              * point gl to the scale, and load 4 floats. x = scale, y = offset, z and w are junk, we
 549              * don't care about them. The pointers are valid for sure because the stateblock is bigger.
 550              * (they're WINED3DTSS_TEXTURETRANSFORMFLAGS and WINED3DTSS_ADDRESSW, so most likely 0 or NaN
 551             */
 552             const float *scale = (const float *)&state->texture_states[texunit][WINED3DTSS_BUMPENVLSCALE];
 553             GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB,
 554                     gl_shader->luminanceconst[i].const_num, scale));
 555         }
 556     }
 557     checkGLcall("Load bumpmap consts");
 558
 559     if(gl_shader->ycorrection != WINED3D_CONST_NUM_UNUSED)
 560     {
 561         /* ycorrection.x: Backbuffer height(onscreen) or 0(offscreen).
 562         * ycorrection.y: -1.0(onscreen), 1.0(offscreen)
 563         * ycorrection.z: 1.0
 564         * ycorrection.w: 0.0
 565         */
 566         float val[4];
 567         val[0] = context->render_offscreen ? 0.0f : rt_height;
 568         val[1] = context->render_offscreen ? 1.0f : -1.0f;
 569         val[2] = 1.0f;
 570         val[3] = 0.0f;
 571         GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, gl_shader->ycorrection, val));
 572         checkGLcall("y correction loading");
 573     }
 574
 575     if (!gl_shader->num_int_consts) return;
 576
 577     for(i = 0; i < MAX_CONST_I; i++)
 578     {
 579         if(gl_shader->int_consts[i] != WINED3D_CONST_NUM_UNUSED)
 580         {
 581             float val[4];
 582             val[0] = (float)state->ps_consts_i[4 * i];
 583             val[1] = (float)state->ps_consts_i[4 * i + 1];
 584             val[2] = (float)state->ps_consts_i[4 * i + 2];
 585             val[3] = -1.0f;
 586
 587             GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, gl_shader->int_consts[i], val));
 588         }
 589     }
 590     checkGLcall("Load ps int consts");
 591 }
 592
 593 /* GL locking is done by the caller. */
 594 static void shader_arb_vs_local_constants(const struct arb_vs_compiled_shader *gl_shader,
 595         const struct wined3d_context *context, const struct wined3d_state *state)
 596 {
 597     const struct wined3d_gl_info *gl_info = context->gl_info;
 598     float position_fixup[4];
 599     unsigned char i;
 600
 601     /* Upload the position fixup */
 602     shader_get_position_fixup(context, state, position_fixup);
 603     GL_EXTCALL(glProgramLocalParameter4fvARB(GL_VERTEX_PROGRAM_ARB, gl_shader->pos_fixup, position_fixup));
 604
 605     if (!gl_shader->num_int_consts) return;
 606
 607     for(i = 0; i < MAX_CONST_I; i++)
 608     {
 609         if(gl_shader->int_consts[i] != WINED3D_CONST_NUM_UNUSED)
 610         {
 611             float val[4];
 612             val[0] = (float)state->vs_consts_i[4 * i];
 613             val[1] = (float)state->vs_consts_i[4 * i + 1];
 614             val[2] = (float)state->vs_consts_i[4 * i + 2];
 615             val[3] = -1.0f;
 616
 617             GL_EXTCALL(glProgramLocalParameter4fvARB(GL_VERTEX_PROGRAM_ARB, gl_shader->int_consts[i], val));
 618         }
 619     }
 620     checkGLcall("Load vs int consts");
 621 }
 622
 623 /**
 624  * Loads the app-supplied constants into the currently set ARB_[vertex/fragment]_programs.
 625  *
 626  * We only support float constants in ARB at the moment, so don't
 627  * worry about the Integers or Booleans
 628  */
 629 /* GL locking is done by the caller (state handler) */
 630 static void shader_arb_load_constants(const struct wined3d_context *context, char usePixelShader, char useVertexShader)
 631 {
 632     IWineD3DDeviceImpl *device = context->swapchain->device;
 633     struct wined3d_stateblock *stateBlock = device->stateBlock;
 634     const struct wined3d_gl_info *gl_info = context->gl_info;
 635     struct shader_arb_priv *priv = device->shader_priv;
 636
 637     if (useVertexShader)
 638     {
 639         struct wined3d_shader *vshader = stateBlock->state.vertex_shader;
 640         const struct arb_vs_compiled_shader *gl_shader = priv->compiled_vprog;
 641
 642         /* Load DirectX 9 float constants for vertex shader */
 643         device->highest_dirty_vs_const = shader_arb_load_constantsF(vshader, gl_info, GL_VERTEX_PROGRAM_ARB,
 644                 device->highest_dirty_vs_const, stateBlock->state.vs_consts_f, context->vshader_const_dirty);
 645         shader_arb_vs_local_constants(gl_shader, context, &stateBlock->state);
 646     }
 647
 648     if (usePixelShader)
 649     {
 650         struct wined3d_shader *pshader = stateBlock->state.pixel_shader;
 651         const struct arb_ps_compiled_shader *gl_shader = priv->compiled_fprog;
 652         float rt_height = device->render_targets[0]->resource.height;
 653
 654         /* Load DirectX 9 float constants for pixel shader */
 655         device->highest_dirty_ps_const = shader_arb_load_constantsF(pshader, gl_info, GL_FRAGMENT_PROGRAM_ARB,
 656                 device->highest_dirty_ps_const, stateBlock->state.ps_consts_f, context->pshader_const_dirty);
 657         shader_arb_ps_local_constants(gl_shader, context, &stateBlock->state, rt_height);
 658     }
 659 }
 660
 661 static void shader_arb_update_float_vertex_constants(IWineD3DDeviceImpl *device, UINT start, UINT count)
 662 {
 663     struct wined3d_context *context = context_get_current();
 664
 665     /* We don't want shader constant dirtification to be an O(contexts), so just dirtify the active
 666      * context. On a context switch the old context will be fully dirtified */
 667     if (!context || context->swapchain->device != device) return;
 668
 669     memset(context->vshader_const_dirty + start, 1, sizeof(*context->vshader_const_dirty) * count);
 670     device->highest_dirty_vs_const = max(device->highest_dirty_vs_const, start + count);
 671 }
 672
 673 static void shader_arb_update_float_pixel_constants(IWineD3DDeviceImpl *device, UINT start, UINT count)
 674 {
 675     struct wined3d_context *context = context_get_current();
 676
 677     /* We don't want shader constant dirtification to be an O(contexts), so just dirtify the active
 678      * context. On a context switch the old context will be fully dirtified */
 679     if (!context || context->swapchain->device != device) return;
 680
 681     memset(context->pshader_const_dirty + start, 1, sizeof(*context->pshader_const_dirty) * count);
 682     device->highest_dirty_ps_const = max(device->highest_dirty_ps_const, start + count);
 683 }
 684
 685 static DWORD *local_const_mapping(const struct wined3d_shader *shader)
 686 {
 687     DWORD *ret;
 688     DWORD idx = 0;
 689     const local_constant *lconst;
 690
 691     if (shader->load_local_constsF || list_empty(&shader->constantsF))
 692         return NULL;
 693
 694     ret = HeapAlloc(GetProcessHeap(), 0, sizeof(DWORD) * shader->limits.constant_float);
 695     if (!ret)
 696     {
 697         ERR("Out of memory\n");
 698         return NULL;
 699     }
 700
 701     LIST_FOR_EACH_ENTRY(lconst, &shader->constantsF, local_constant, entry)
 702     {
 703         ret[lconst->idx] = idx++;
 704     }
 705     return ret;
 706 }
 707
 708 /* Generate the variable & register declarations for the ARB_vertex_program output target */
 709 static DWORD shader_generate_arb_declarations(struct wined3d_shader *shader,
 710         const struct wined3d_shader_reg_maps *reg_maps, struct wined3d_shader_buffer *buffer,
 711         const struct wined3d_gl_info *gl_info, DWORD *lconst_map,
 712         DWORD *num_clipplanes, struct shader_arb_ctx_priv *ctx)
 713 {
 714     DWORD i, next_local = 0;
 715     char pshader = shader_is_pshader_version(reg_maps->shader_version.type);
 716     unsigned max_constantsF;
 717     const local_constant *lconst;
 718     DWORD map;
 719
 720     /* In pixel shaders, all private constants are program local, we don't need anything
 721      * from program.env. Thus we can advertise the full set of constants in pixel shaders.
 722      * If we need a private constant the GL implementation will squeeze it in somewhere
 723      *
 724      * With vertex shaders we need the posFixup and on some GL implementations 4 helper
 725      * immediate values. The posFixup is loaded using program.env for now, so always
 726      * subtract one from the number of constants. If the shader uses indirect addressing,
 727      * account for the helper const too because we have to declare all availabke d3d constants
 728      * and don't know which are actually used.
 729      */
 730     if (pshader)
 731     {
 732         max_constantsF = gl_info->limits.arb_ps_native_constants;
 733         /* 24 is the minimum MAX_PROGRAM_ENV_PARAMETERS_ARB value. */
 734         if (max_constantsF < 24)
 735             max_constantsF = gl_info->limits.arb_ps_float_constants;
 736     }
 737     else
 738     {
 739         const struct arb_vshader_private *shader_data = shader->backend_data;
 740         max_constantsF = gl_info->limits.arb_vs_native_constants;
 741         /* 96 is the minimum MAX_PROGRAM_ENV_PARAMETERS_ARB value.
 742          * Also prevents max_constantsF from becoming less than 0 and
 743          * wrapping . */
 744         if (max_constantsF < 96)
 745             max_constantsF = gl_info->limits.arb_vs_float_constants;
 746
 747         if (reg_maps->usesrelconstF)
 748         {
 749             DWORD highest_constf = 0, clip_limit;
 750
 751             max_constantsF -= reserved_vs_const(shader_data, reg_maps, gl_info);
 752             max_constantsF -= count_bits(reg_maps->integer_constants);
 753
 754             for (i = 0; i < shader->limits.constant_float; ++i)
 755             {
 756                 DWORD idx = i >> 5;
 757                 DWORD shift = i & 0x1f;
 758                 if(reg_maps->constf[idx] & (1 << shift)) highest_constf = i;
 759             }
 760
 761             if(use_nv_clip(gl_info) && ctx->target_version >= NV2)
 762             {
 763                 if(ctx->cur_vs_args->super.clip_enabled)
 764                     clip_limit = gl_info->limits.clipplanes;
 765                 else
 766                     clip_limit = 0;
 767             }
 768             else
 769             {
 770                 unsigned int mask = ctx->cur_vs_args->clip.boolclip.clipplane_mask;
 771                 clip_limit = min(count_bits(mask), 4);
 772             }
 773             *num_clipplanes = min(clip_limit, max_constantsF - highest_constf - 1);
 774             max_constantsF -= *num_clipplanes;
 775             if(*num_clipplanes < clip_limit)
 776             {
 777                 WARN("Only %u clipplanes out of %u enabled\n", *num_clipplanes, gl_info->limits.clipplanes);
 778             }
 779         }
 780         else
 781         {
 782             if (ctx->target_version >= NV2) *num_clipplanes = gl_info->limits.clipplanes;
 783             else *num_clipplanes = min(gl_info->limits.clipplanes, 4);
 784         }
 785     }
 786
 787     for (i = 0, map = reg_maps->temporary; map; map >>= 1, ++i)
 788     {
 789         if (map & 1) shader_addline(buffer, "TEMP R%u;\n", i);
 790     }
 791
 792     for (i = 0, map = reg_maps->address; map; map >>= 1, ++i)
 793     {
 794         if (map & 1) shader_addline(buffer, "ADDRESS A%u;\n", i);
 795     }
 796
 797     if (pshader && reg_maps->shader_version.major == 1 && reg_maps->shader_version.minor <= 3)
 798     {
 799         for (i = 0, map = reg_maps->texcoord; map; map >>= 1, ++i)
 800         {
 801             if (map & 1) shader_addline(buffer, "TEMP T%u;\n", i);
 802         }
 803     }
 804
 805     /* Load local constants using the program-local space,
 806      * this avoids reloading them each time the shader is used
 807      */
 808     if (lconst_map)
 809     {
 810         LIST_FOR_EACH_ENTRY(lconst, &shader->constantsF, local_constant, entry)
 811         {
 812             shader_addline(buffer, "PARAM C%u = program.local[%u];\n", lconst->idx,
 813                            lconst_map[lconst->idx]);
 814             next_local = max(next_local, lconst_map[lconst->idx] + 1);
 815         }
 816     }
 817
 818     /* After subtracting privately used constants from the hardware limit(they are loaded as
 819      * local constants), make sure the shader doesn't violate the env constant limit
 820      */
 821     if(pshader)
 822     {
 823         max_constantsF = min(max_constantsF, gl_info->limits.arb_ps_float_constants);
 824     }
 825     else
 826     {
 827         max_constantsF = min(max_constantsF, gl_info->limits.arb_vs_float_constants);
 828     }
 829
 830     /* Avoid declaring more constants than needed */
 831     max_constantsF = min(max_constantsF, shader->limits.constant_float);
 832
 833     /* we use the array-based constants array if the local constants are marked for loading,
 834      * because then we use indirect addressing, or when the local constant list is empty,
 835      * because then we don't know if we're using indirect addressing or not. If we're hardcoding
 836      * local constants do not declare the loaded constants as an array because ARB compilers usually
 837      * do not optimize unused constants away
 838      */
 839     if (reg_maps->usesrelconstF)
 840     {
 841         /* Need to PARAM the environment parameters (constants) so we can use relative addressing */
 842         shader_addline(buffer, "PARAM C[%d] = { program.env[0..%d] };\n",
 843                     max_constantsF, max_constantsF - 1);
 844     } else {
 845         for(i = 0; i < max_constantsF; i++) {
 846             DWORD idx, mask;
 847             idx = i >> 5;
 848             mask = 1 << (i & 0x1f);
 849             if (!shader_constant_is_local(shader, i) && (reg_maps->constf[idx] & mask))
 850             {
 851                 shader_addline(buffer, "PARAM C%d = program.env[%d];\n",i, i);
 852             }
 853         }
 854     }
 855
 856     return next_local;
 857 }
 858
 859 static const char * const shift_tab[] = {
 860     "dummy",     /*  0 (none) */
 861     "coefmul.x", /*  1 (x2)   */
 862     "coefmul.y", /*  2 (x4)   */
 863     "coefmul.z", /*  3 (x8)   */
 864     "coefmul.w", /*  4 (x16)  */
 865     "dummy",     /*  5 (x32)  */
 866     "dummy",     /*  6 (x64)  */
 867     "dummy",     /*  7 (x128) */
 868     "dummy",     /*  8 (d256) */
 869     "dummy",     /*  9 (d128) */
 870     "dummy",     /* 10 (d64)  */
 871     "dummy",     /* 11 (d32)  */
 872     "coefdiv.w", /* 12 (d16)  */
 873     "coefdiv.z", /* 13 (d8)   */
 874     "coefdiv.y", /* 14 (d4)   */
 875     "coefdiv.x"  /* 15 (d2)   */
 876 };
 877
 878 static void shader_arb_get_write_mask(const struct wined3d_shader_instruction *ins,
 879         const struct wined3d_shader_dst_param *dst, char *write_mask)
 880 {
 881     char *ptr = write_mask;
 882
 883     if (dst->write_mask != WINED3DSP_WRITEMASK_ALL)
 884     {
 885         *ptr++ = '.';
 886         if (dst->write_mask & WINED3DSP_WRITEMASK_0) *ptr++ = 'x';
 887         if (dst->write_mask & WINED3DSP_WRITEMASK_1) *ptr++ = 'y';
 888         if (dst->write_mask & WINED3DSP_WRITEMASK_2) *ptr++ = 'z';
 889         if (dst->write_mask & WINED3DSP_WRITEMASK_3) *ptr++ = 'w';
 890     }
 891
 892     *ptr = '\0';
 893 }
 894
 895 static void shader_arb_get_swizzle(const struct wined3d_shader_src_param *param, BOOL fixup, char *swizzle_str)
 896 {
 897     /* For registers of type WINED3DDECLTYPE_D3DCOLOR, data is stored as "bgra",
 898      * but addressed as "rgba". To fix this we need to swap the register's x
 899      * and z components. */
 900     const char *swizzle_chars = fixup ? "zyxw" : "xyzw";
 901     char *ptr = swizzle_str;
 902
 903     /* swizzle bits fields: wwzzyyxx */
 904     DWORD swizzle = param->swizzle;
 905     DWORD swizzle_x = swizzle & 0x03;
 906     DWORD swizzle_y = (swizzle >> 2) & 0x03;
 907     DWORD swizzle_z = (swizzle >> 4) & 0x03;
 908     DWORD swizzle_w = (swizzle >> 6) & 0x03;
 909
 910     /* If the swizzle is the default swizzle (ie, "xyzw"), we don't need to
 911      * generate a swizzle string. Unless we need to our own swizzling. */
 912     if (swizzle != WINED3DSP_NOSWIZZLE || fixup)
 913     {
 914         *ptr++ = '.';
 915         if (swizzle_x == swizzle_y && swizzle_x == swizzle_z && swizzle_x == swizzle_w) {
 916             *ptr++ = swizzle_chars[swizzle_x];
 917         } else {
 918             *ptr++ = swizzle_chars[swizzle_x];
 919             *ptr++ = swizzle_chars[swizzle_y];
 920             *ptr++ = swizzle_chars[swizzle_z];
 921             *ptr++ = swizzle_chars[swizzle_w];
 922         }
 923     }
 924
 925     *ptr = '\0';
 926 }
 927
 928 static void shader_arb_request_a0(const struct wined3d_shader_instruction *ins, const char *src)
 929 {
 930     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
 931     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
 932
 933     if (!strcmp(priv->addr_reg, src)) return;
 934
 935     strcpy(priv->addr_reg, src);
 936     shader_addline(buffer, "ARL A0.x, %s;\n", src);
 937 }
 938
 939 static void shader_arb_get_src_param(const struct wined3d_shader_instruction *ins,
 940         const struct wined3d_shader_src_param *src, unsigned int tmpreg, char *outregstr);
 941
 942 static void shader_arb_get_register_name(const struct wined3d_shader_instruction *ins,
 943         const struct wined3d_shader_register *reg, char *register_name, BOOL *is_color)
 944 {
 945     /* oPos, oFog and oPts in D3D */
 946     static const char * const rastout_reg_names[] = {"TMP_OUT", "result.fogcoord", "result.pointsize"};
 947     struct wined3d_shader *shader = ins->ctx->shader;
 948     const struct wined3d_shader_reg_maps *reg_maps = ins->ctx->reg_maps;
 949     BOOL pshader = shader_is_pshader_version(reg_maps->shader_version.type);
 950     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
 951
 952     *is_color = FALSE;
 953
 954     switch (reg->type)
 955     {
 956         case WINED3DSPR_TEMP:
 957             sprintf(register_name, "R%u", reg->idx);
 958             break;
 959
 960         case WINED3DSPR_INPUT:
 961             if (pshader)
 962             {
 963                 if (reg_maps->shader_version.major < 3)
 964                 {
 965                     if (!reg->idx) strcpy(register_name, "fragment.color.primary");
 966                     else strcpy(register_name, "fragment.color.secondary");
 967                 }
 968                 else
 969                 {
 970                     if(reg->rel_addr)
 971                     {
 972                         char rel_reg[50];
 973                         shader_arb_get_src_param(ins, reg->rel_addr, 0, rel_reg);
 974
 975                         if (!strcmp(rel_reg, "**aL_emul**"))
 976                         {
 977                             DWORD idx = ctx->aL + reg->idx;
 978                             if(idx < MAX_REG_INPUT)
 979                             {
 980                                 strcpy(register_name, ctx->ps_input[idx]);
 981                             }
 982                             else
 983                             {
 984                                 ERR("Pixel shader input register out of bounds: %u\n", idx);
 985                                 sprintf(register_name, "out_of_bounds_%u", idx);
 986                             }
 987                         }
 988                         else if (reg_maps->input_registers & 0x0300)
 989                         {
 990                             /* There are two ways basically:
 991                              *
 992                              * 1) Use the unrolling code that is used for loop emulation and unroll the loop.
 993                              *    That means trouble if the loop also contains a breakc or if the control values
 994                              *    aren't local constants.
 995                              * 2) Generate an if block that checks if aL.y < 8, == 8 or == 9 and selects the
 996                              *    source dynamically. The trouble is that we cannot simply read aL.y because it
 997                              *    is an ADDRESS register. We could however push it, load .zw with a value and use
 998                              *    ADAC to load the condition code register and pop it again afterwards
 999                              */
1000                             FIXME("Relative input register addressing with more than 8 registers\n");
1001
1002                             /* This is better than nothing for now */
1003                             sprintf(register_name, "fragment.texcoord[%s + %u]", rel_reg, reg->idx);
1004                         }
1005                         else if(ctx->cur_ps_args->super.vp_mode != vertexshader)
1006                         {
1007                             /* This is problematic because we'd have to consult the ctx->ps_input strings
1008                              * for where to find the varying. Some may be "0.0", others can be texcoords or
1009                              * colors. This needs either a pipeline replacement to make the vertex shader feed
1010                              * proper varyings, or loop unrolling
1011                              *
1012                              * For now use the texcoords and hope for the best
1013                              */
1014                             FIXME("Non-vertex shader varying input with indirect addressing\n");
1015                             sprintf(register_name, "fragment.texcoord[%s + %u]", rel_reg, reg->idx);
1016                         }
1017                         else
1018                         {
1019                             /* D3D supports indirect addressing only with aL in loop registers. The loop instruction
1020                              * pulls GL_NV_fragment_program2 in
1021                              */
1022                             sprintf(register_name, "fragment.texcoord[%s + %u]", rel_reg, reg->idx);
1023                         }
1024                     }
1025                     else
1026                     {
1027                         if(reg->idx < MAX_REG_INPUT)
1028                         {
1029                             strcpy(register_name, ctx->ps_input[reg->idx]);
1030                         }
1031                         else
1032                         {
1033                             ERR("Pixel shader input register out of bounds: %u\n", reg->idx);
1034                             sprintf(register_name, "out_of_bounds_%u", reg->idx);
1035                         }
1036                     }
1037                 }
1038             }
1039             else
1040             {
1041                 if (ctx->cur_vs_args->super.swizzle_map & (1 << reg->idx)) *is_color = TRUE;
1042                 sprintf(register_name, "vertex.attrib[%u]", reg->idx);
1043             }
1044             break;
1045
1046         case WINED3DSPR_CONST:
1047             if (!pshader && reg->rel_addr)
1048             {
1049                 const struct arb_vshader_private *shader_data = shader->backend_data;
1050                 UINT rel_offset = shader_data->rel_offset;
1051                 BOOL aL = FALSE;
1052                 char rel_reg[50];
1053                 if (reg_maps->shader_version.major < 2)
1054                 {
1055                     sprintf(rel_reg, "A0.x");
1056                 } else {
1057                     shader_arb_get_src_param(ins, reg->rel_addr, 0, rel_reg);
1058                     if(ctx->target_version == ARB) {
1059                         if (!strcmp(rel_reg, "**aL_emul**"))
1060                         {
1061                             aL = TRUE;
1062                         } else {
1063                             shader_arb_request_a0(ins, rel_reg);
1064                             sprintf(rel_reg, "A0.x");
1065                         }
1066                     }
1067                 }
1068                 if(aL)
1069                     sprintf(register_name, "C[%u]", ctx->aL + reg->idx);
1070                 else if (reg->idx >= rel_offset)
1071                     sprintf(register_name, "C[%s + %u]", rel_reg, reg->idx - rel_offset);
1072                 else
1073                     sprintf(register_name, "C[%s - %u]", rel_reg, rel_offset - reg->idx);
1074             }
1075             else
1076             {
1077                 if (reg_maps->usesrelconstF)
1078                     sprintf(register_name, "C[%u]", reg->idx);
1079                 else
1080                     sprintf(register_name, "C%u", reg->idx);
1081             }
1082             break;
1083
1084         case WINED3DSPR_TEXTURE: /* case WINED3DSPR_ADDR: */
1085             if (pshader)
1086             {
1087                 if (reg_maps->shader_version.major == 1
1088                         && reg_maps->shader_version.minor <= 3)
1089                 {
1090                     /* In ps <= 1.3, Tx is a temporary register as destination to all instructions,
1091                      * and as source to most instructions. For some instructions it is the texcoord
1092                      * input. Those instructions know about the special use
1093                      */
1094                     sprintf(register_name, "T%u", reg->idx);
1095                 } else {
1096                     /* in ps 1.4 and 2.x Tx is always a (read-only) varying */
1097                     sprintf(register_name, "fragment.texcoord[%u]", reg->idx);
1098                 }
1099             }
1100             else
1101             {
1102                 if (reg_maps->shader_version.major == 1 || ctx->target_version >= NV2)
1103                 {
1104                     sprintf(register_name, "A%u", reg->idx);
1105                 }
1106                 else
1107                 {
1108                     sprintf(register_name, "A%u_SHADOW", reg->idx);
1109                 }
1110             }
1111             break;
1112
1113         case WINED3DSPR_COLOROUT:
1114             if (ctx->cur_ps_args->super.srgb_correction && !reg->idx)
1115             {
1116                 strcpy(register_name, "TMP_COLOR");
1117             }
1118             else
1119             {
1120                 if(ctx->cur_ps_args->super.srgb_correction) FIXME("sRGB correction on higher render targets\n");
1121                 if (reg_maps->highest_render_target > 0)
1122                 {
1123                     sprintf(register_name, "result.color[%u]", reg->idx);
1124                 }
1125                 else
1126                 {
1127                     strcpy(register_name, "result.color");
1128                 }
1129             }
1130             break;
1131
1132         case WINED3DSPR_RASTOUT:
1133             if(reg->idx == 1) sprintf(register_name, "%s", ctx->fog_output);
1134             else sprintf(register_name, "%s", rastout_reg_names[reg->idx]);
1135             break;
1136
1137         case WINED3DSPR_DEPTHOUT:
1138             strcpy(register_name, "result.depth");
1139             break;
1140
1141         case WINED3DSPR_ATTROUT:
1142         /* case WINED3DSPR_OUTPUT: */
1143             if (pshader) sprintf(register_name, "oD[%u]", reg->idx);
1144             else strcpy(register_name, ctx->color_output[reg->idx]);
1145             break;
1146
1147         case WINED3DSPR_TEXCRDOUT:
1148             if (pshader)
1149             {
1150                 sprintf(register_name, "oT[%u]", reg->idx);
1151             }
1152             else
1153             {
1154                 if (reg_maps->shader_version.major < 3)
1155                 {
1156                     strcpy(register_name, ctx->texcrd_output[reg->idx]);
1157                 }
1158                 else
1159                 {
1160                     strcpy(register_name, ctx->vs_output[reg->idx]);
1161                 }
1162             }
1163             break;
1164
1165         case WINED3DSPR_LOOP:
1166             if(ctx->target_version >= NV2)
1167             {
1168                 /* Pshader has an implicitly declared loop index counter A0.x that cannot be renamed */
1169                 if(pshader) sprintf(register_name, "A0.x");
1170                 else sprintf(register_name, "aL.y");
1171             }
1172             else
1173             {
1174                 /* Unfortunately this code cannot return the value of ctx->aL here. An immediate value
1175                  * would be valid, but if aL is used for indexing(its only use), there's likely an offset,
1176                  * thus the result would be something like C[15 + 30], which is not valid in the ARB program
1177                  * grammar. So return a marker for the emulated aL and intercept it in constant and varying
1178                  * indexing
1179                  */
1180                 sprintf(register_name, "**aL_emul**");
1181             }
1182
1183             break;
1184
1185         case WINED3DSPR_CONSTINT:
1186             sprintf(register_name, "I%u", reg->idx);
1187             break;
1188
1189         case WINED3DSPR_MISCTYPE:
1190             if (!reg->idx)
1191             {
1192                 sprintf(register_name, "vpos");
1193             }
1194             else if(reg->idx == 1)
1195             {
1196                 sprintf(register_name, "fragment.facing.x");
1197             }
1198             else
1199             {
1200                 FIXME("Unknown MISCTYPE register index %u\n", reg->idx);
1201             }
1202             break;
1203
1204         default:
1205             FIXME("Unhandled register type %#x[%u]\n", reg->type, reg->idx);
1206             sprintf(register_name, "unrecognized_register[%u]", reg->idx);
1207             break;
1208     }
1209 }
1210
1211 static void shader_arb_get_dst_param(const struct wined3d_shader_instruction *ins,
1212         const struct wined3d_shader_dst_param *wined3d_dst, char *str)
1213 {
1214     char register_name[255];
1215     char write_mask[6];
1216     BOOL is_color;
1217
1218     shader_arb_get_register_name(ins, &wined3d_dst->reg, register_name, &is_color);
1219     strcpy(str, register_name);
1220
1221     shader_arb_get_write_mask(ins, wined3d_dst, write_mask);
1222     strcat(str, write_mask);
1223 }
1224
1225 static const char *shader_arb_get_fixup_swizzle(enum fixup_channel_source channel_source)
1226 {
1227     switch(channel_source)
1228     {
1229         case CHANNEL_SOURCE_ZERO: return "0";
1230         case CHANNEL_SOURCE_ONE: return "1";
1231         case CHANNEL_SOURCE_X: return "x";
1232         case CHANNEL_SOURCE_Y: return "y";
1233         case CHANNEL_SOURCE_Z: return "z";
1234         case CHANNEL_SOURCE_W: return "w";
1235         default:
1236             FIXME("Unhandled channel source %#x\n", channel_source);
1237             return "undefined";
1238     }
1239 }
1240
1241 static void gen_color_correction(struct wined3d_shader_buffer *buffer, const char *reg,
1242         DWORD dst_mask, const char *one, const char *two, struct color_fixup_desc fixup)
1243 {
1244     DWORD mask;
1245
1246     if (is_complex_fixup(fixup))
1247     {
1248         enum complex_fixup complex_fixup = get_complex_fixup(fixup);
1249         FIXME("Complex fixup (%#x) not supported\n", complex_fixup);
1250         return;
1251     }
1252
1253     mask = 0;
1254     if (fixup.x_source != CHANNEL_SOURCE_X) mask |= WINED3DSP_WRITEMASK_0;
1255     if (fixup.y_source != CHANNEL_SOURCE_Y) mask |= WINED3DSP_WRITEMASK_1;
1256     if (fixup.z_source != CHANNEL_SOURCE_Z) mask |= WINED3DSP_WRITEMASK_2;
1257     if (fixup.w_source != CHANNEL_SOURCE_W) mask |= WINED3DSP_WRITEMASK_3;
1258     mask &= dst_mask;
1259
1260     if (mask)
1261     {
1262         shader_addline(buffer, "SWZ %s, %s, %s, %s, %s, %s;\n", reg, reg,
1263                 shader_arb_get_fixup_swizzle(fixup.x_source), shader_arb_get_fixup_swizzle(fixup.y_source),
1264                 shader_arb_get_fixup_swizzle(fixup.z_source), shader_arb_get_fixup_swizzle(fixup.w_source));
1265     }
1266
1267     mask = 0;
1268     if (fixup.x_sign_fixup) mask |= WINED3DSP_WRITEMASK_0;
1269     if (fixup.y_sign_fixup) mask |= WINED3DSP_WRITEMASK_1;
1270     if (fixup.z_sign_fixup) mask |= WINED3DSP_WRITEMASK_2;
1271     if (fixup.w_sign_fixup) mask |= WINED3DSP_WRITEMASK_3;
1272     mask &= dst_mask;
1273
1274     if (mask)
1275     {
1276         char reg_mask[6];
1277         char *ptr = reg_mask;
1278
1279         if (mask != WINED3DSP_WRITEMASK_ALL)
1280         {
1281             *ptr++ = '.';
1282             if (mask & WINED3DSP_WRITEMASK_0) *ptr++ = 'x';
1283             if (mask & WINED3DSP_WRITEMASK_1) *ptr++ = 'y';
1284             if (mask & WINED3DSP_WRITEMASK_2) *ptr++ = 'z';
1285             if (mask & WINED3DSP_WRITEMASK_3) *ptr++ = 'w';
1286         }
1287         *ptr = '\0';
1288
1289         shader_addline(buffer, "MAD %s%s, %s, %s, -%s;\n", reg, reg_mask, reg, two, one);
1290     }
1291 }
1292
1293 static const char *shader_arb_get_modifier(const struct wined3d_shader_instruction *ins)
1294 {
1295     DWORD mod;
1296     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
1297     if (!ins->dst_count) return "";
1298
1299     mod = ins->dst[0].modifiers;
1300
1301     /* Silently ignore PARTIALPRECISION if its not supported */
1302     if(priv->target_version == ARB) mod &= ~WINED3DSPDM_PARTIALPRECISION;
1303
1304     if(mod & WINED3DSPDM_MSAMPCENTROID)
1305     {
1306         FIXME("Unhandled modifier WINED3DSPDM_MSAMPCENTROID\n");
1307         mod &= ~WINED3DSPDM_MSAMPCENTROID;
1308     }
1309
1310     switch(mod)
1311     {
1312         case WINED3DSPDM_SATURATE | WINED3DSPDM_PARTIALPRECISION:
1313             return "H_SAT";
1314
1315         case WINED3DSPDM_SATURATE:
1316             return "_SAT";
1317
1318         case WINED3DSPDM_PARTIALPRECISION:
1319             return "H";
1320
1321         case 0:
1322             return "";
1323
1324         default:
1325             FIXME("Unknown modifiers 0x%08x\n", mod);
1326             return "";
1327     }
1328 }
1329
1330 #define TEX_PROJ        0x1
1331 #define TEX_BIAS        0x2
1332 #define TEX_LOD         0x4
1333 #define TEX_DERIV       0x10
1334
1335 static void shader_hw_sample(const struct wined3d_shader_instruction *ins, DWORD sampler_idx,
1336         const char *dst_str, const char *coord_reg, WORD flags, const char *dsx, const char *dsy)
1337 {
1338     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1339     DWORD sampler_type = ins->ctx->reg_maps->sampler_type[sampler_idx];
1340     const struct wined3d_texture *texture;
1341     const char *tex_type;
1342     BOOL np2_fixup = FALSE;
1343     struct wined3d_shader *shader = ins->ctx->shader;
1344     IWineD3DDeviceImpl *device = shader->device;
1345     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
1346     const char *mod;
1347     BOOL pshader = shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type);
1348
1349     /* D3D vertex shader sampler IDs are vertex samplers(0-3), not global d3d samplers */
1350     if(!pshader) sampler_idx += MAX_FRAGMENT_SAMPLERS;
1351
1352     switch(sampler_type) {
1353         case WINED3DSTT_1D:
1354             tex_type = "1D";
1355             break;
1356
1357         case WINED3DSTT_2D:
1358             texture = device->stateBlock->state.textures[sampler_idx];
1359             if (texture && texture->target == GL_TEXTURE_RECTANGLE_ARB)
1360             {
1361                 tex_type = "RECT";
1362             } else {
1363                 tex_type = "2D";
1364             }
1365             if (shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type))
1366             {
1367                 if (priv->cur_np2fixup_info->super.active & (1 << sampler_idx))
1368                 {
1369                     if (flags) FIXME("Only ordinary sampling from NP2 textures is supported.\n");
1370                     else np2_fixup = TRUE;
1371                 }
1372             }
1373             break;
1374
1375         case WINED3DSTT_VOLUME:
1376             tex_type = "3D";
1377             break;
1378
1379         case WINED3DSTT_CUBE:
1380             tex_type = "CUBE";
1381             break;
1382
1383         default:
1384             ERR("Unexpected texture type %d\n", sampler_type);
1385             tex_type = "";
1386     }
1387
1388     /* TEX, TXL, TXD and TXP do not support the "H" modifier,
1389      * so don't use shader_arb_get_modifier
1390      */
1391     if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE) mod = "_SAT";
1392     else mod = "";
1393
1394     /* Fragment samplers always have indentity mapping */
1395     if(sampler_idx >= MAX_FRAGMENT_SAMPLERS)
1396     {
1397         sampler_idx = priv->cur_vs_args->vertex.samplers[sampler_idx - MAX_FRAGMENT_SAMPLERS];
1398     }
1399
1400     if (flags & TEX_DERIV)
1401     {
1402         if(flags & TEX_PROJ) FIXME("Projected texture sampling with custom derivatives\n");
1403         if(flags & TEX_BIAS) FIXME("Biased texture sampling with custom derivatives\n");
1404         shader_addline(buffer, "TXD%s %s, %s, %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg,
1405                        dsx, dsy,sampler_idx, tex_type);
1406     }
1407     else if(flags & TEX_LOD)
1408     {
1409         if(flags & TEX_PROJ) FIXME("Projected texture sampling with explicit lod\n");
1410         if(flags & TEX_BIAS) FIXME("Biased texture sampling with explicit lod\n");
1411         shader_addline(buffer, "TXL%s %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg,
1412                        sampler_idx, tex_type);
1413     }
1414     else if (flags & TEX_BIAS)
1415     {
1416         /* Shouldn't be possible, but let's check for it */
1417         if(flags & TEX_PROJ) FIXME("Biased and Projected texture sampling\n");
1418         /* TXB takes the 4th component of the source vector automatically, as d3d. Nothing more to do */
1419         shader_addline(buffer, "TXB%s %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg, sampler_idx, tex_type);
1420     }
1421     else if (flags & TEX_PROJ)
1422     {
1423         shader_addline(buffer, "TXP%s %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg, sampler_idx, tex_type);
1424     }
1425     else
1426     {
1427         if (np2_fixup)
1428         {
1429             const unsigned char idx = priv->cur_np2fixup_info->super.idx[sampler_idx];
1430             shader_addline(buffer, "MUL TA, np2fixup[%u].%s, %s;\n", idx >> 1,
1431                            (idx % 2) ? "zwxy" : "xyzw", coord_reg);
1432
1433             shader_addline(buffer, "TEX%s %s, TA, texture[%u], %s;\n", mod, dst_str, sampler_idx, tex_type);
1434         }
1435         else
1436             shader_addline(buffer, "TEX%s %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg, sampler_idx, tex_type);
1437     }
1438
1439     if (pshader)
1440     {
1441         gen_color_correction(buffer, dst_str, ins->dst[0].write_mask,
1442                 arb_get_helper_value(WINED3D_SHADER_TYPE_PIXEL, ARB_ONE),
1443                 arb_get_helper_value(WINED3D_SHADER_TYPE_PIXEL, ARB_TWO),
1444                 priv->cur_ps_args->super.color_fixup[sampler_idx]);
1445     }
1446 }
1447
1448 static void shader_arb_get_src_param(const struct wined3d_shader_instruction *ins,
1449         const struct wined3d_shader_src_param *src, unsigned int tmpreg, char *outregstr)
1450 {
1451     /* Generate a line that does the input modifier computation and return the input register to use */
1452     BOOL is_color = FALSE;
1453     char regstr[256];
1454     char swzstr[20];
1455     int insert_line;
1456     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1457     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
1458     const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
1459     const char *two = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_TWO);
1460
1461     /* Assume a new line will be added */
1462     insert_line = 1;
1463
1464     /* Get register name */
1465     shader_arb_get_register_name(ins, &src->reg, regstr, &is_color);
1466     shader_arb_get_swizzle(src, is_color, swzstr);
1467
1468     switch (src->modifiers)
1469     {
1470     case WINED3DSPSM_NONE:
1471         sprintf(outregstr, "%s%s", regstr, swzstr);
1472         insert_line = 0;
1473         break;
1474     case WINED3DSPSM_NEG:
1475         sprintf(outregstr, "-%s%s", regstr, swzstr);
1476         insert_line = 0;
1477         break;
1478     case WINED3DSPSM_BIAS:
1479         shader_addline(buffer, "ADD T%c, %s, -coefdiv.x;\n", 'A' + tmpreg, regstr);
1480         break;
1481     case WINED3DSPSM_BIASNEG:
1482         shader_addline(buffer, "ADD T%c, -%s, coefdiv.x;\n", 'A' + tmpreg, regstr);
1483         break;
1484     case WINED3DSPSM_SIGN:
1485         shader_addline(buffer, "MAD T%c, %s, %s, -%s;\n", 'A' + tmpreg, regstr, two, one);
1486         break;
1487     case WINED3DSPSM_SIGNNEG:
1488         shader_addline(buffer, "MAD T%c, %s, -%s, %s;\n", 'A' + tmpreg, regstr, two, one);
1489         break;
1490     case WINED3DSPSM_COMP:
1491         shader_addline(buffer, "SUB T%c, %s, %s;\n", 'A' + tmpreg, one, regstr);
1492         break;
1493     case WINED3DSPSM_X2:
1494         shader_addline(buffer, "ADD T%c, %s, %s;\n", 'A' + tmpreg, regstr, regstr);
1495         break;
1496     case WINED3DSPSM_X2NEG:
1497         shader_addline(buffer, "ADD T%c, -%s, -%s;\n", 'A' + tmpreg, regstr, regstr);
1498         break;
1499     case WINED3DSPSM_DZ:
1500         shader_addline(buffer, "RCP T%c, %s.z;\n", 'A' + tmpreg, regstr);
1501         shader_addline(buffer, "MUL T%c, %s, T%c;\n", 'A' + tmpreg, regstr, 'A' + tmpreg);
1502         break;
1503     case WINED3DSPSM_DW:
1504         shader_addline(buffer, "RCP T%c, %s.w;\n", 'A' + tmpreg, regstr);
1505         shader_addline(buffer, "MUL T%c, %s, T%c;\n", 'A' + tmpreg, regstr, 'A' + tmpreg);
1506         break;
1507     case WINED3DSPSM_ABS:
1508         if(ctx->target_version >= NV2) {
1509             sprintf(outregstr, "|%s%s|", regstr, swzstr);
1510             insert_line = 0;
1511         } else {
1512             shader_addline(buffer, "ABS T%c, %s;\n", 'A' + tmpreg, regstr);
1513         }
1514         break;
1515     case WINED3DSPSM_ABSNEG:
1516         if(ctx->target_version >= NV2) {
1517             sprintf(outregstr, "-|%s%s|", regstr, swzstr);
1518         } else {
1519             shader_addline(buffer, "ABS T%c, %s;\n", 'A' + tmpreg, regstr);
1520             sprintf(outregstr, "-T%c%s", 'A' + tmpreg, swzstr);
1521         }
1522         insert_line = 0;
1523         break;
1524     default:
1525         sprintf(outregstr, "%s%s", regstr, swzstr);
1526         insert_line = 0;
1527     }
1528
1529     /* Return modified or original register, with swizzle */
1530     if (insert_line)
1531         sprintf(outregstr, "T%c%s", 'A' + tmpreg, swzstr);
1532 }
1533
1534 static void pshader_hw_bem(const struct wined3d_shader_instruction *ins)
1535 {
1536     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1537     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1538     char dst_name[50];
1539     char src_name[2][50];
1540     DWORD sampler_code = dst->reg.idx;
1541
1542     shader_arb_get_dst_param(ins, dst, dst_name);
1543
1544     /* Sampling the perturbation map in Tsrc was done already, including the signedness correction if needed
1545      *
1546      * Keep in mind that src_name[1] can be "TB" and src_name[0] can be "TA" because modifiers like _x2 are valid
1547      * with bem. So delay loading the first parameter until after the perturbation calculation which needs two
1548      * temps is done.
1549      */
1550     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1551     shader_addline(buffer, "SWZ TA, bumpenvmat%d, x, z, 0, 0;\n", sampler_code);
1552     shader_addline(buffer, "DP3 TC.r, TA, %s;\n", src_name[1]);
1553     shader_addline(buffer, "SWZ TA, bumpenvmat%d, y, w, 0, 0;\n", sampler_code);
1554     shader_addline(buffer, "DP3 TC.g, TA, %s;\n", src_name[1]);
1555
1556     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
1557     shader_addline(buffer, "ADD %s, %s, TC;\n", dst_name, src_name[0]);
1558 }
1559
1560 static DWORD negate_modifiers(DWORD mod, char *extra_char)
1561 {
1562     *extra_char = ' ';
1563     switch(mod)
1564     {
1565         case WINED3DSPSM_NONE:      return WINED3DSPSM_NEG;
1566         case WINED3DSPSM_NEG:       return WINED3DSPSM_NONE;
1567         case WINED3DSPSM_BIAS:      return WINED3DSPSM_BIASNEG;
1568         case WINED3DSPSM_BIASNEG:   return WINED3DSPSM_BIAS;
1569         case WINED3DSPSM_SIGN:      return WINED3DSPSM_SIGNNEG;
1570         case WINED3DSPSM_SIGNNEG:   return WINED3DSPSM_SIGN;
1571         case WINED3DSPSM_COMP:      *extra_char = '-'; return WINED3DSPSM_COMP;
1572         case WINED3DSPSM_X2:        return WINED3DSPSM_X2NEG;
1573         case WINED3DSPSM_X2NEG:     return WINED3DSPSM_X2;
1574         case WINED3DSPSM_DZ:        *extra_char = '-'; return WINED3DSPSM_DZ;
1575         case WINED3DSPSM_DW:        *extra_char = '-'; return WINED3DSPSM_DW;
1576         case WINED3DSPSM_ABS:       return WINED3DSPSM_ABSNEG;
1577         case WINED3DSPSM_ABSNEG:    return WINED3DSPSM_ABS;
1578     }
1579     FIXME("Unknown modifier %u\n", mod);
1580     return mod;
1581 }
1582
1583 static void pshader_hw_cnd(const struct wined3d_shader_instruction *ins)
1584 {
1585     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1586     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1587     char dst_name[50];
1588     char src_name[3][50];
1589     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
1590             ins->ctx->reg_maps->shader_version.minor);
1591     BOOL is_color;
1592
1593     shader_arb_get_dst_param(ins, dst, dst_name);
1594     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1595
1596     /* The coissue flag changes the semantic of the cnd instruction in <= 1.3 shaders */
1597     if (shader_version <= WINED3D_SHADER_VERSION(1, 3) && ins->coissue)
1598     {
1599         shader_addline(buffer, "MOV%s %s, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name[1]);
1600     } else {
1601         struct wined3d_shader_src_param src0_copy = ins->src[0];
1602         char extra_neg;
1603
1604         /* src0 may have a negate srcmod set, so we can't blindly add "-" to the name */
1605         src0_copy.modifiers = negate_modifiers(src0_copy.modifiers, &extra_neg);
1606
1607         shader_arb_get_src_param(ins, &src0_copy, 0, src_name[0]);
1608         shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
1609         shader_addline(buffer, "ADD TA, %c%s, coefdiv.x;\n", extra_neg, src_name[0]);
1610         /* No modifiers supported on CMP */
1611         shader_addline(buffer, "CMP %s, TA, %s, %s;\n", dst_name, src_name[1], src_name[2]);
1612
1613         /* _SAT on CMP doesn't make much sense, but it is not a pure NOP */
1614         if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE)
1615         {
1616             shader_arb_get_register_name(ins, &dst->reg, src_name[0], &is_color);
1617             shader_addline(buffer, "MOV_SAT %s, %s;\n", dst_name, dst_name);
1618         }
1619     }
1620 }
1621
1622 static void pshader_hw_cmp(const struct wined3d_shader_instruction *ins)
1623 {
1624     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1625     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1626     char dst_name[50];
1627     char src_name[3][50];
1628     BOOL is_color;
1629
1630     shader_arb_get_dst_param(ins, dst, dst_name);
1631
1632     /* Generate input register names (with modifiers) */
1633     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
1634     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1635     shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
1636
1637     /* No modifiers are supported on CMP */
1638     shader_addline(buffer, "CMP %s, %s, %s, %s;\n", dst_name,
1639                    src_name[0], src_name[2], src_name[1]);
1640
1641     if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE)
1642     {
1643         shader_arb_get_register_name(ins, &dst->reg, src_name[0], &is_color);
1644         shader_addline(buffer, "MOV_SAT %s, %s;\n", dst_name, src_name[0]);
1645     }
1646 }
1647
1648 /** Process the WINED3DSIO_DP2ADD instruction in ARB.
1649  * dst = dot2(src0, src1) + src2 */
1650 static void pshader_hw_dp2add(const struct wined3d_shader_instruction *ins)
1651 {
1652     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1653     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1654     char dst_name[50];
1655     char src_name[3][50];
1656     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
1657
1658     shader_arb_get_dst_param(ins, dst, dst_name);
1659     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
1660     shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
1661
1662     if(ctx->target_version >= NV3)
1663     {
1664         /* GL_NV_fragment_program2 has a 1:1 matching instruction */
1665         shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1666         shader_addline(buffer, "DP2A%s %s, %s, %s, %s;\n", shader_arb_get_modifier(ins),
1667                        dst_name, src_name[0], src_name[1], src_name[2]);
1668     }
1669     else if(ctx->target_version >= NV2)
1670     {
1671         /* dst.x = src2.?, src0.x, src1.x + src0.y * src1.y
1672          * dst.y = src2.?, src0.x, src1.z + src0.y * src1.w
1673          * dst.z = src2.?, src0.x, src1.x + src0.y * src1.y
1674          * dst.z = src2.?, src0.x, src1.z + src0.y * src1.w
1675          *
1676          * Make sure that src1.zw = src1.xy, then we get a classic dp2add
1677          *
1678          * .xyxy and other swizzles that we could get with this are not valid in
1679          * plain ARBfp, but luckily the NV extension grammar lifts this limitation.
1680          */
1681         struct wined3d_shader_src_param tmp_param = ins->src[1];
1682         DWORD swizzle = tmp_param.swizzle & 0xf; /* Selects .xy */
1683         tmp_param.swizzle = swizzle | (swizzle << 4); /* Creates .xyxy */
1684
1685         shader_arb_get_src_param(ins, &tmp_param, 1, src_name[1]);
1686
1687         shader_addline(buffer, "X2D%s %s, %s, %s, %s;\n", shader_arb_get_modifier(ins),
1688                        dst_name, src_name[2], src_name[0], src_name[1]);
1689     }
1690     else
1691     {
1692         shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1693         /* 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]
1694         * src_name[0] can be TA, but TA is a private temp for modifiers, so it is save to overwrite
1695         */
1696         shader_addline(buffer, "MOV TA, %s;\n", src_name[0]);
1697         shader_addline(buffer, "MOV TA.z, 0.0;\n");
1698         shader_addline(buffer, "DP3 TA, TA, %s;\n", src_name[1]);
1699         shader_addline(buffer, "ADD%s %s, TA, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name[2]);
1700     }
1701 }
1702
1703 /* Map the opcode 1-to-1 to the GL code */
1704 static void shader_hw_map2gl(const struct wined3d_shader_instruction *ins)
1705 {
1706     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1707     const char *instruction;
1708     char arguments[256], dst_str[50];
1709     unsigned int i;
1710     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1711
1712     switch (ins->handler_idx)
1713     {
1714         case WINED3DSIH_ABS: instruction = "ABS"; break;
1715         case WINED3DSIH_ADD: instruction = "ADD"; break;
1716         case WINED3DSIH_CRS: instruction = "XPD"; break;
1717         case WINED3DSIH_DP3: instruction = "DP3"; break;
1718         case WINED3DSIH_DP4: instruction = "DP4"; break;
1719         case WINED3DSIH_DST: instruction = "DST"; break;
1720         case WINED3DSIH_FRC: instruction = "FRC"; break;
1721         case WINED3DSIH_LIT: instruction = "LIT"; break;
1722         case WINED3DSIH_LRP: instruction = "LRP"; break;
1723         case WINED3DSIH_MAD: instruction = "MAD"; break;
1724         case WINED3DSIH_MAX: instruction = "MAX"; break;
1725         case WINED3DSIH_MIN: instruction = "MIN"; break;
1726         case WINED3DSIH_MOV: instruction = "MOV"; break;
1727         case WINED3DSIH_MUL: instruction = "MUL"; break;
1728         case WINED3DSIH_SGE: instruction = "SGE"; break;
1729         case WINED3DSIH_SLT: instruction = "SLT"; break;
1730         case WINED3DSIH_SUB: instruction = "SUB"; break;
1731         case WINED3DSIH_MOVA:instruction = "ARR"; break;
1732         case WINED3DSIH_DSX: instruction = "DDX"; break;
1733         default: instruction = "";
1734             FIXME("Unhandled opcode %#x\n", ins->handler_idx);
1735             break;
1736     }
1737
1738     /* Note that shader_arb_add_dst_param() adds spaces. */
1739     arguments[0] = '\0';
1740     shader_arb_get_dst_param(ins, dst, dst_str);
1741     for (i = 0; i < ins->src_count; ++i)
1742     {
1743         char operand[100];
1744         strcat(arguments, ", ");
1745         shader_arb_get_src_param(ins, &ins->src[i], i, operand);
1746         strcat(arguments, operand);
1747     }
1748     shader_addline(buffer, "%s%s %s%s;\n", instruction, shader_arb_get_modifier(ins), dst_str, arguments);
1749 }
1750
1751 static void shader_hw_nop(const struct wined3d_shader_instruction *ins)
1752 {
1753     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1754     shader_addline(buffer, "NOP;\n");
1755 }
1756
1757 static void shader_hw_mov(const struct wined3d_shader_instruction *ins)
1758 {
1759     struct wined3d_shader *shader = ins->ctx->shader;
1760     const struct wined3d_shader_reg_maps *reg_maps = ins->ctx->reg_maps;
1761     BOOL pshader = shader_is_pshader_version(reg_maps->shader_version.type);
1762     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
1763     const char *zero = arb_get_helper_value(reg_maps->shader_version.type, ARB_ZERO);
1764     const char *one = arb_get_helper_value(reg_maps->shader_version.type, ARB_ONE);
1765     const char *two = arb_get_helper_value(reg_maps->shader_version.type, ARB_TWO);
1766
1767     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1768     char src0_param[256];
1769
1770     if (ins->handler_idx == WINED3DSIH_MOVA)
1771     {
1772         const struct arb_vshader_private *shader_data = shader->backend_data;
1773         char write_mask[6];
1774         const char *offset = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_VS_REL_OFFSET);
1775
1776         if(ctx->target_version >= NV2) {
1777             shader_hw_map2gl(ins);
1778             return;
1779         }
1780         shader_arb_get_src_param(ins, &ins->src[0], 0, src0_param);
1781         shader_arb_get_write_mask(ins, &ins->dst[0], write_mask);
1782
1783         /* This implements the mova formula used in GLSL. The first two instructions
1784          * prepare the sign() part. Note that it is fine to have my_sign(0.0) = 1.0
1785          * in this case:
1786          * mova A0.x, 0.0
1787          *
1788          * A0.x = arl(floor(abs(0.0) + 0.5) * 1.0) = floor(0.5) = 0.0 since arl does a floor
1789          *
1790          * The ARL is performed when A0 is used - the requested component is read from A0_SHADOW into
1791          * A0.x. We can use the overwritten component of A0_shadow as temporary storage for the sign.
1792          */
1793         shader_addline(buffer, "SGE A0_SHADOW%s, %s, %s;\n", write_mask, src0_param, zero);
1794         shader_addline(buffer, "MAD A0_SHADOW%s, A0_SHADOW, %s, -%s;\n", write_mask, two, one);
1795
1796         shader_addline(buffer, "ABS TA%s, %s;\n", write_mask, src0_param);
1797         shader_addline(buffer, "ADD TA%s, TA, rel_addr_const.x;\n", write_mask);
1798         shader_addline(buffer, "FLR TA%s, TA;\n", write_mask);
1799         if (shader_data->rel_offset)
1800         {
1801             shader_addline(buffer, "ADD TA%s, TA, %s;\n", write_mask, offset);
1802         }
1803         shader_addline(buffer, "MUL A0_SHADOW%s, TA, A0_SHADOW;\n", write_mask);
1804
1805         ((struct shader_arb_ctx_priv *)ins->ctx->backend_data)->addr_reg[0] = '\0';
1806     }
1807     else if (reg_maps->shader_version.major == 1
1808           && !shader_is_pshader_version(reg_maps->shader_version.type)
1809           && ins->dst[0].reg.type == WINED3DSPR_ADDR)
1810     {
1811         const struct arb_vshader_private *shader_data = shader->backend_data;
1812         src0_param[0] = '\0';
1813
1814         if (shader_data->rel_offset)
1815         {
1816             const char *offset = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_VS_REL_OFFSET);
1817             shader_arb_get_src_param(ins, &ins->src[0], 0, src0_param);
1818             shader_addline(buffer, "ADD TA.x, %s, %s;\n", src0_param, offset);
1819             shader_addline(buffer, "ARL A0.x, TA.x;\n");
1820         }
1821         else
1822         {
1823             /* Apple's ARB_vertex_program implementation does not accept an ARL source argument
1824              * with more than one component. Thus replicate the first source argument over all
1825              * 4 components. For example, .xyzw -> .x (or better: .xxxx), .zwxy -> .z, etc) */
1826             struct wined3d_shader_src_param tmp_src = ins->src[0];
1827             tmp_src.swizzle = (tmp_src.swizzle & 0x3) * 0x55;
1828             shader_arb_get_src_param(ins, &tmp_src, 0, src0_param);
1829             shader_addline(buffer, "ARL A0.x, %s;\n", src0_param);
1830         }
1831     }
1832     else if (ins->dst[0].reg.type == WINED3DSPR_COLOROUT && !ins->dst[0].reg.idx && pshader)
1833     {
1834         if (ctx->cur_ps_args->super.srgb_correction && shader->u.ps.color0_mov)
1835         {
1836             shader_addline(buffer, "#mov handled in srgb write code\n");
1837             return;
1838         }
1839         shader_hw_map2gl(ins);
1840     }
1841     else
1842     {
1843         shader_hw_map2gl(ins);
1844     }
1845 }
1846
1847 static void pshader_hw_texkill(const struct wined3d_shader_instruction *ins)
1848 {
1849     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1850     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1851     char reg_dest[40];
1852
1853     /* No swizzles are allowed in d3d's texkill. PS 1.x ignores the 4th component as documented,
1854      * but >= 2.0 honors it(undocumented, but tested by the d3d9 testsuit)
1855      */
1856     shader_arb_get_dst_param(ins, dst, reg_dest);
1857
1858     if (ins->ctx->reg_maps->shader_version.major >= 2)
1859     {
1860         const char *kilsrc = "TA";
1861         BOOL is_color;
1862
1863         shader_arb_get_register_name(ins, &dst->reg, reg_dest, &is_color);
1864         if(dst->write_mask == WINED3DSP_WRITEMASK_ALL)
1865         {
1866             kilsrc = reg_dest;
1867         }
1868         else
1869         {
1870             /* Sigh. KIL doesn't support swizzles/writemasks. KIL passes a writemask, but ".xy" for example
1871              * is not valid as a swizzle in ARB (needs ".xyyy"). Use SWZ to load the register properly, and set
1872              * masked out components to 0(won't kill)
1873              */
1874             char x = '0', y = '0', z = '0', w = '0';
1875             if(dst->write_mask & WINED3DSP_WRITEMASK_0) x = 'x';
1876             if(dst->write_mask & WINED3DSP_WRITEMASK_1) y = 'y';
1877             if(dst->write_mask & WINED3DSP_WRITEMASK_2) z = 'z';
1878             if(dst->write_mask & WINED3DSP_WRITEMASK_3) w = 'w';
1879             shader_addline(buffer, "SWZ TA, %s, %c, %c, %c, %c;\n", reg_dest, x, y, z, w);
1880         }
1881         shader_addline(buffer, "KIL %s;\n", kilsrc);
1882     } else {
1883         /* ARB fp doesn't like swizzles on the parameter of the KIL instruction. To mask the 4th component,
1884          * copy the register into our general purpose TMP variable, overwrite .w and pass TMP to KIL
1885          *
1886          * ps_1_3 shaders use the texcoord incarnation of the Tx register. ps_1_4 shaders can use the same,
1887          * or pass in any temporary register(in shader phase 2)
1888          */
1889         if(ins->ctx->reg_maps->shader_version.minor <= 3) {
1890             sprintf(reg_dest, "fragment.texcoord[%u]", dst->reg.idx);
1891         } else {
1892             shader_arb_get_dst_param(ins, dst, reg_dest);
1893         }
1894         shader_addline(buffer, "SWZ TA, %s, x, y, z, 1;\n", reg_dest);
1895         shader_addline(buffer, "KIL TA;\n");
1896     }
1897 }
1898
1899 static void pshader_hw_tex(const struct wined3d_shader_instruction *ins)
1900 {
1901     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
1902     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1903     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
1904             ins->ctx->reg_maps->shader_version.minor);
1905     struct wined3d_shader_src_param src;
1906
1907     char reg_dest[40];
1908     char reg_coord[40];
1909     DWORD reg_sampler_code;
1910     WORD myflags = 0;
1911
1912     /* All versions have a destination register */
1913     shader_arb_get_dst_param(ins, dst, reg_dest);
1914
1915     /* 1.0-1.4: Use destination register number as texture code.
1916        2.0+: Use provided sampler number as texure code. */
1917     if (shader_version < WINED3D_SHADER_VERSION(2,0))
1918         reg_sampler_code = dst->reg.idx;
1919     else
1920         reg_sampler_code = ins->src[1].reg.idx;
1921
1922     /* 1.0-1.3: Use the texcoord varying.
1923        1.4+: Use provided coordinate source register. */
1924     if (shader_version < WINED3D_SHADER_VERSION(1,4))
1925         sprintf(reg_coord, "fragment.texcoord[%u]", reg_sampler_code);
1926     else {
1927         /* TEX is the only instruction that can handle DW and DZ natively */
1928         src = ins->src[0];
1929         if(src.modifiers == WINED3DSPSM_DW) src.modifiers = WINED3DSPSM_NONE;
1930         if(src.modifiers == WINED3DSPSM_DZ) src.modifiers = WINED3DSPSM_NONE;
1931         shader_arb_get_src_param(ins, &src, 0, reg_coord);
1932     }
1933
1934     /* projection flag:
1935      * 1.1, 1.2, 1.3: Use WINED3DTSS_TEXTURETRANSFORMFLAGS
1936      * 1.4: Use WINED3DSPSM_DZ or WINED3DSPSM_DW on src[0]
1937      * 2.0+: Use WINED3DSI_TEXLD_PROJECT on the opcode
1938      */
1939     if (shader_version < WINED3D_SHADER_VERSION(1,4))
1940     {
1941         DWORD flags = 0;
1942         if (reg_sampler_code < MAX_TEXTURES)
1943             flags = priv->cur_ps_args->super.tex_transform >> reg_sampler_code * WINED3D_PSARGS_TEXTRANSFORM_SHIFT;
1944         if (flags & WINED3D_PSARGS_PROJECTED)
1945             myflags |= TEX_PROJ;
1946     }
1947     else if (shader_version < WINED3D_SHADER_VERSION(2,0))
1948     {
1949         DWORD src_mod = ins->src[0].modifiers;
1950         if (src_mod == WINED3DSPSM_DZ) {
1951             /* TXP cannot handle DZ natively, so move the z coordinate to .w. reg_coord is a read-only
1952              * varying register, so we need a temp reg
1953              */
1954             shader_addline(ins->ctx->buffer, "SWZ TA, %s, x, y, z, z;\n", reg_coord);
1955             strcpy(reg_coord, "TA");
1956             myflags |= TEX_PROJ;
1957         } else if(src_mod == WINED3DSPSM_DW) {
1958             myflags |= TEX_PROJ;
1959         }
1960     } else {
1961         if (ins->flags & WINED3DSI_TEXLD_PROJECT) myflags |= TEX_PROJ;
1962         if (ins->flags & WINED3DSI_TEXLD_BIAS) myflags |= TEX_BIAS;
1963     }
1964     shader_hw_sample(ins, reg_sampler_code, reg_dest, reg_coord, myflags, NULL, NULL);
1965 }
1966
1967 static void pshader_hw_texcoord(const struct wined3d_shader_instruction *ins)
1968 {
1969     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1970     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1971     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
1972             ins->ctx->reg_maps->shader_version.minor);
1973     char dst_str[50];
1974
1975     if (shader_version < WINED3D_SHADER_VERSION(1,4))
1976     {
1977         DWORD reg = dst->reg.idx;
1978
1979         shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
1980         shader_addline(buffer, "MOV_SAT %s, fragment.texcoord[%u];\n", dst_str, reg);
1981     } else {
1982         char reg_src[40];
1983
1984         shader_arb_get_src_param(ins, &ins->src[0], 0, reg_src);
1985         shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
1986         shader_addline(buffer, "MOV %s, %s;\n", dst_str, reg_src);
1987    }
1988 }
1989
1990 static void pshader_hw_texreg2ar(const struct wined3d_shader_instruction *ins)
1991 {
1992      struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1993      DWORD flags = 0;
1994
1995      DWORD reg1 = ins->dst[0].reg.idx;
1996      char dst_str[50];
1997      char src_str[50];
1998
1999      /* Note that texreg2ar treats Tx as a temporary register, not as a varying */
2000      shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2001      shader_arb_get_src_param(ins, &ins->src[0], 0, src_str);
2002      /* Move .x first in case src_str is "TA" */
2003      shader_addline(buffer, "MOV TA.y, %s.x;\n", src_str);
2004      shader_addline(buffer, "MOV TA.x, %s.w;\n", src_str);
2005      if (reg1 < MAX_TEXTURES)
2006      {
2007          struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2008          flags = priv->cur_ps_args->super.tex_transform >> reg1 * WINED3D_PSARGS_TEXTRANSFORM_SHIFT;
2009      }
2010      shader_hw_sample(ins, reg1, dst_str, "TA", flags & WINED3D_PSARGS_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2011 }
2012
2013 static void pshader_hw_texreg2gb(const struct wined3d_shader_instruction *ins)
2014 {
2015      struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2016
2017      DWORD reg1 = ins->dst[0].reg.idx;
2018      char dst_str[50];
2019      char src_str[50];
2020
2021      /* Note that texreg2gb treats Tx as a temporary register, not as a varying */
2022      shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2023      shader_arb_get_src_param(ins, &ins->src[0], 0, src_str);
2024      shader_addline(buffer, "MOV TA.x, %s.y;\n", src_str);
2025      shader_addline(buffer, "MOV TA.y, %s.z;\n", src_str);
2026      shader_hw_sample(ins, reg1, dst_str, "TA", 0, NULL, NULL);
2027 }
2028
2029 static void pshader_hw_texreg2rgb(const struct wined3d_shader_instruction *ins)
2030 {
2031     DWORD reg1 = ins->dst[0].reg.idx;
2032     char dst_str[50];
2033     char src_str[50];
2034
2035     /* Note that texreg2rg treats Tx as a temporary register, not as a varying */
2036     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2037     shader_arb_get_src_param(ins, &ins->src[0], 0, src_str);
2038     shader_hw_sample(ins, reg1, dst_str, src_str, 0, NULL, NULL);
2039 }
2040
2041 static void pshader_hw_texbem(const struct wined3d_shader_instruction *ins)
2042 {
2043     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2044     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2045     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2046     char reg_coord[40], dst_reg[50], src_reg[50];
2047     DWORD reg_dest_code;
2048
2049     /* All versions have a destination register. The Tx where the texture coordinates come
2050      * from is the varying incarnation of the texture register
2051      */
2052     reg_dest_code = dst->reg.idx;
2053     shader_arb_get_dst_param(ins, &ins->dst[0], dst_reg);
2054     shader_arb_get_src_param(ins, &ins->src[0], 0, src_reg);
2055     sprintf(reg_coord, "fragment.texcoord[%u]", reg_dest_code);
2056
2057     /* Sampling the perturbation map in Tsrc was done already, including the signedness correction if needed
2058      * The Tx in which the perturbation map is stored is the tempreg incarnation of the texture register
2059      *
2060      * GL_NV_fragment_program_option could handle this in one instruction via X2D:
2061      * X2D TA.xy, fragment.texcoord, T%u, bumpenvmat%u.xzyw
2062      *
2063      * However, the NV extensions are never enabled for <= 2.0 shaders because of the performance penalty that
2064      * comes with it, and texbem is an 1.x only instruction. No 1.x instruction forces us to enable the NV
2065      * extension.
2066      */
2067     shader_addline(buffer, "SWZ TB, bumpenvmat%d, x, z, 0, 0;\n", reg_dest_code);
2068     shader_addline(buffer, "DP3 TA.x, TB, %s;\n", src_reg);
2069     shader_addline(buffer, "SWZ TB, bumpenvmat%d, y, w, 0, 0;\n", reg_dest_code);
2070     shader_addline(buffer, "DP3 TA.y, TB, %s;\n", src_reg);
2071
2072     /* with projective textures, texbem only divides the static texture coord, not the displacement,
2073      * so we can't let the GL handle this.
2074      */
2075     if ((priv->cur_ps_args->super.tex_transform >> reg_dest_code * WINED3D_PSARGS_TEXTRANSFORM_SHIFT)
2076             & WINED3D_PSARGS_PROJECTED)
2077     {
2078         shader_addline(buffer, "RCP TB.w, %s.w;\n", reg_coord);
2079         shader_addline(buffer, "MUL TB.xy, %s, TB.w;\n", reg_coord);
2080         shader_addline(buffer, "ADD TA.xy, TA, TB;\n");
2081     } else {
2082         shader_addline(buffer, "ADD TA.xy, TA, %s;\n", reg_coord);
2083     }
2084
2085     shader_hw_sample(ins, reg_dest_code, dst_reg, "TA", 0, NULL, NULL);
2086
2087     if (ins->handler_idx == WINED3DSIH_TEXBEML)
2088     {
2089         /* No src swizzles are allowed, so this is ok */
2090         shader_addline(buffer, "MAD TA, %s.z, luminance%d.x, luminance%d.y;\n",
2091                        src_reg, reg_dest_code, reg_dest_code);
2092         shader_addline(buffer, "MUL %s, %s, TA;\n", dst_reg, dst_reg);
2093     }
2094 }
2095
2096 static void pshader_hw_texm3x2pad(const struct wined3d_shader_instruction *ins)
2097 {
2098     DWORD reg = ins->dst[0].reg.idx;
2099     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2100     char src0_name[50], dst_name[50];
2101     BOOL is_color;
2102     struct wined3d_shader_register tmp_reg = ins->dst[0].reg;
2103
2104     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2105     /* The next instruction will be a texm3x2tex or texm3x2depth that writes to the uninitialized
2106      * T<reg+1> register. Use this register to store the calculated vector
2107      */
2108     tmp_reg.idx = reg + 1;
2109     shader_arb_get_register_name(ins, &tmp_reg, dst_name, &is_color);
2110     shader_addline(buffer, "DP3 %s.x, fragment.texcoord[%u], %s;\n", dst_name, reg, src0_name);
2111 }
2112
2113 static void pshader_hw_texm3x2tex(const struct wined3d_shader_instruction *ins)
2114 {
2115     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2116     DWORD flags;
2117     DWORD reg = ins->dst[0].reg.idx;
2118     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2119     char dst_str[50];
2120     char src0_name[50];
2121     char dst_reg[50];
2122     BOOL is_color;
2123
2124     /* We know that we're writing to the uninitialized T<reg> register, so use it for temporary storage */
2125     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_reg, &is_color);
2126
2127     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2128     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2129     shader_addline(buffer, "DP3 %s.y, fragment.texcoord[%u], %s;\n", dst_reg, reg, src0_name);
2130     flags = reg < MAX_TEXTURES ? priv->cur_ps_args->super.tex_transform >> reg * WINED3D_PSARGS_TEXTRANSFORM_SHIFT : 0;
2131     shader_hw_sample(ins, reg, dst_str, dst_reg, flags & WINED3DTTFF_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2132 }
2133
2134 static void pshader_hw_texm3x3pad(const struct wined3d_shader_instruction *ins)
2135 {
2136     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
2137     DWORD reg = ins->dst[0].reg.idx;
2138     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2139     char src0_name[50], dst_name[50];
2140     struct wined3d_shader_register tmp_reg = ins->dst[0].reg;
2141     BOOL is_color;
2142
2143     /* There are always 2 texm3x3pad instructions followed by one texm3x3[tex,vspec, ...] instruction, with
2144      * incrementing ins->dst[0].register_idx numbers. So the pad instruction already knows the final destination
2145      * register, and this register is uninitialized(otherwise the assembler complains that it is 'redeclared')
2146      */
2147     tmp_reg.idx = reg + 2 - tex_mx->current_row;
2148     shader_arb_get_register_name(ins, &tmp_reg, dst_name, &is_color);
2149
2150     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2151     shader_addline(buffer, "DP3 %s.%c, fragment.texcoord[%u], %s;\n",
2152                    dst_name, 'x' + tex_mx->current_row, reg, src0_name);
2153     tex_mx->texcoord_w[tex_mx->current_row++] = reg;
2154 }
2155
2156 static void pshader_hw_texm3x3tex(const struct wined3d_shader_instruction *ins)
2157 {
2158     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2159     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
2160     DWORD flags;
2161     DWORD reg = ins->dst[0].reg.idx;
2162     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2163     char dst_str[50];
2164     char src0_name[50], dst_name[50];
2165     BOOL is_color;
2166
2167     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2168     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2169     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_name, reg, src0_name);
2170
2171     /* Sample the texture using the calculated coordinates */
2172     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2173     flags = reg < MAX_TEXTURES ? priv->cur_ps_args->super.tex_transform >> reg * WINED3D_PSARGS_TEXTRANSFORM_SHIFT : 0;
2174     shader_hw_sample(ins, reg, dst_str, dst_name, flags & WINED3DTTFF_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2175     tex_mx->current_row = 0;
2176 }
2177
2178 static void pshader_hw_texm3x3vspec(const struct wined3d_shader_instruction *ins)
2179 {
2180     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2181     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
2182     DWORD flags;
2183     DWORD reg = ins->dst[0].reg.idx;
2184     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2185     char dst_str[50];
2186     char src0_name[50];
2187     char dst_reg[50];
2188     BOOL is_color;
2189
2190     /* Get the dst reg without writemask strings. We know this register is uninitialized, so we can use all
2191      * components for temporary data storage
2192      */
2193     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_reg, &is_color);
2194     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2195     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_reg, reg, src0_name);
2196
2197     /* Construct the eye-ray vector from w coordinates */
2198     shader_addline(buffer, "MOV TB.x, fragment.texcoord[%u].w;\n", tex_mx->texcoord_w[0]);
2199     shader_addline(buffer, "MOV TB.y, fragment.texcoord[%u].w;\n", tex_mx->texcoord_w[1]);
2200     shader_addline(buffer, "MOV TB.z, fragment.texcoord[%u].w;\n", reg);
2201
2202     /* Calculate reflection vector
2203      */
2204     shader_addline(buffer, "DP3 %s.w, %s, TB;\n", dst_reg, dst_reg);
2205     /* The .w is ignored when sampling, so I can use TB.w to calculate dot(N, N) */
2206     shader_addline(buffer, "DP3 TB.w, %s, %s;\n", dst_reg, dst_reg);
2207     shader_addline(buffer, "RCP TB.w, TB.w;\n");
2208     shader_addline(buffer, "MUL %s.w, %s.w, TB.w;\n", dst_reg, dst_reg);
2209     shader_addline(buffer, "MUL %s, %s.w, %s;\n", dst_reg, dst_reg, dst_reg);
2210     shader_addline(buffer, "MAD %s, coefmul.x, %s, -TB;\n", dst_reg, dst_reg);
2211
2212     /* Sample the texture using the calculated coordinates */
2213     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2214     flags = reg < MAX_TEXTURES ? priv->cur_ps_args->super.tex_transform >> reg * WINED3D_PSARGS_TEXTRANSFORM_SHIFT : 0;
2215     shader_hw_sample(ins, reg, dst_str, dst_reg, flags & WINED3DTTFF_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2216     tex_mx->current_row = 0;
2217 }
2218
2219 static void pshader_hw_texm3x3spec(const struct wined3d_shader_instruction *ins)
2220 {
2221     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2222     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
2223     DWORD flags;
2224     DWORD reg = ins->dst[0].reg.idx;
2225     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2226     char dst_str[50];
2227     char src0_name[50];
2228     char src1_name[50];
2229     char dst_reg[50];
2230     BOOL is_color;
2231
2232     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2233     shader_arb_get_src_param(ins, &ins->src[0], 1, src1_name);
2234     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_reg, &is_color);
2235     /* Note: dst_reg.xy is input here, generated by two texm3x3pad instructions */
2236     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_reg, reg, src0_name);
2237
2238     /* Calculate reflection vector.
2239      *
2240      *                   dot(N, E)
2241      * dst_reg.xyz = 2 * --------- * N - E
2242      *                   dot(N, N)
2243      *
2244      * Which normalizes the normal vector
2245      */
2246     shader_addline(buffer, "DP3 %s.w, %s, %s;\n", dst_reg, dst_reg, src1_name);
2247     shader_addline(buffer, "DP3 TC.w, %s, %s;\n", dst_reg, dst_reg);
2248     shader_addline(buffer, "RCP TC.w, TC.w;\n");
2249     shader_addline(buffer, "MUL %s.w, %s.w, TC.w;\n", dst_reg, dst_reg);
2250     shader_addline(buffer, "MUL %s, %s.w, %s;\n", dst_reg, dst_reg, dst_reg);
2251     shader_addline(buffer, "MAD %s, coefmul.x, %s, -%s;\n", dst_reg, dst_reg, src1_name);
2252
2253     /* Sample the texture using the calculated coordinates */
2254     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2255     flags = reg < MAX_TEXTURES ? priv->cur_ps_args->super.tex_transform >> reg * WINED3D_PSARGS_TEXTRANSFORM_SHIFT : 0;
2256     shader_hw_sample(ins, reg, dst_str, dst_reg, flags & WINED3DTTFF_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2257     tex_mx->current_row = 0;
2258 }
2259
2260 static void pshader_hw_texdepth(const struct wined3d_shader_instruction *ins)
2261 {
2262     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2263     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2264     char dst_name[50];
2265     const char *zero = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ZERO);
2266     const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
2267
2268     /* texdepth has an implicit destination, the fragment depth value. It's only parameter,
2269      * which is essentially an input, is the destination register because it is the first
2270      * parameter. According to the msdn, this must be register r5, but let's keep it more flexible
2271      * here(writemasks/swizzles are not valid on texdepth)
2272      */
2273     shader_arb_get_dst_param(ins, dst, dst_name);
2274
2275     /* According to the msdn, the source register(must be r5) is unusable after
2276      * the texdepth instruction, so we're free to modify it
2277      */
2278     shader_addline(buffer, "MIN %s.y, %s.y, %s;\n", dst_name, dst_name, one);
2279
2280     /* How to deal with the special case dst_name.g == 0? if r != 0, then
2281      * the r * (1 / 0) will give infinity, which is clamped to 1.0, the correct
2282      * result. But if r = 0.0, then 0 * inf = 0, which is incorrect.
2283      */
2284     shader_addline(buffer, "RCP %s.y, %s.y;\n", dst_name, dst_name);
2285     shader_addline(buffer, "MUL TA.x, %s.x, %s.y;\n", dst_name, dst_name);
2286     shader_addline(buffer, "MIN TA.x, TA.x, %s;\n", one);
2287     shader_addline(buffer, "MAX result.depth, TA.x, %s;\n", zero);
2288 }
2289
2290 /** Process the WINED3DSIO_TEXDP3TEX instruction in ARB:
2291  * Take a 3-component dot product of the TexCoord[dstreg] and src,
2292  * then perform a 1D texture lookup from stage dstregnum, place into dst. */
2293 static void pshader_hw_texdp3tex(const struct wined3d_shader_instruction *ins)
2294 {
2295     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2296     DWORD sampler_idx = ins->dst[0].reg.idx;
2297     char src0[50];
2298     char dst_str[50];
2299
2300     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2301     shader_addline(buffer, "MOV TB, 0.0;\n");
2302     shader_addline(buffer, "DP3 TB.x, fragment.texcoord[%u], %s;\n", sampler_idx, src0);
2303
2304     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2305     shader_hw_sample(ins, sampler_idx, dst_str, "TB", 0 /* Only one coord, can't be projected */, NULL, NULL);
2306 }
2307
2308 /** Process the WINED3DSIO_TEXDP3 instruction in ARB:
2309  * Take a 3-component dot product of the TexCoord[dstreg] and src. */
2310 static void pshader_hw_texdp3(const struct wined3d_shader_instruction *ins)
2311 {
2312     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2313     char src0[50];
2314     char dst_str[50];
2315     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2316
2317     /* Handle output register */
2318     shader_arb_get_dst_param(ins, dst, dst_str);
2319     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2320     shader_addline(buffer, "DP3 %s, fragment.texcoord[%u], %s;\n", dst_str, dst->reg.idx, src0);
2321 }
2322
2323 /** Process the WINED3DSIO_TEXM3X3 instruction in ARB
2324  * Perform the 3rd row of a 3x3 matrix multiply */
2325 static void pshader_hw_texm3x3(const struct wined3d_shader_instruction *ins)
2326 {
2327     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2328     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2329     char dst_str[50], dst_name[50];
2330     char src0[50];
2331     BOOL is_color;
2332
2333     shader_arb_get_dst_param(ins, dst, dst_str);
2334     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2335     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2336     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_name, dst->reg.idx, src0);
2337     shader_addline(buffer, "MOV %s, %s;\n", dst_str, dst_name);
2338 }
2339
2340 /** Process the WINED3DSIO_TEXM3X2DEPTH instruction in ARB:
2341  * Last row of a 3x2 matrix multiply, use the result to calculate the depth:
2342  * Calculate tmp0.y = TexCoord[dstreg] . src.xyz;  (tmp0.x has already been calculated)
2343  * depth = (tmp0.y == 0.0) ? 1.0 : tmp0.x / tmp0.y
2344  */
2345 static void pshader_hw_texm3x2depth(const struct wined3d_shader_instruction *ins)
2346 {
2347     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2348     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2349     char src0[50], dst_name[50];
2350     BOOL is_color;
2351     const char *zero = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ZERO);
2352     const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
2353
2354     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2355     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2356     shader_addline(buffer, "DP3 %s.y, fragment.texcoord[%u], %s;\n", dst_name, dst->reg.idx, src0);
2357
2358     /* How to deal with the special case dst_name.g == 0? if r != 0, then
2359      * the r * (1 / 0) will give infinity, which is clamped to 1.0, the correct
2360      * result. But if r = 0.0, then 0 * inf = 0, which is incorrect.
2361      */
2362     shader_addline(buffer, "RCP %s.y, %s.y;\n", dst_name, dst_name);
2363     shader_addline(buffer, "MUL %s.x, %s.x, %s.y;\n", dst_name, dst_name, dst_name);
2364     shader_addline(buffer, "MIN %s.x, %s.x, %s;\n", dst_name, dst_name, one);
2365     shader_addline(buffer, "MAX result.depth, %s.x, %s;\n", dst_name, zero);
2366 }
2367
2368 /** Handles transforming all WINED3DSIO_M?x? opcodes for
2369     Vertex/Pixel shaders to ARB_vertex_program codes */
2370 static void shader_hw_mnxn(const struct wined3d_shader_instruction *ins)
2371 {
2372     int i;
2373     int nComponents = 0;
2374     struct wined3d_shader_dst_param tmp_dst = {{0}};
2375     struct wined3d_shader_src_param tmp_src[2] = {{{0}}};
2376     struct wined3d_shader_instruction tmp_ins;
2377
2378     memset(&tmp_ins, 0, sizeof(tmp_ins));
2379
2380     /* Set constants for the temporary argument */
2381     tmp_ins.ctx = ins->ctx;
2382     tmp_ins.dst_count = 1;
2383     tmp_ins.dst = &tmp_dst;
2384     tmp_ins.src_count = 2;
2385     tmp_ins.src = tmp_src;
2386
2387     switch(ins->handler_idx)
2388     {
2389         case WINED3DSIH_M4x4:
2390             nComponents = 4;
2391             tmp_ins.handler_idx = WINED3DSIH_DP4;
2392             break;
2393         case WINED3DSIH_M4x3:
2394             nComponents = 3;
2395             tmp_ins.handler_idx = WINED3DSIH_DP4;
2396             break;
2397         case WINED3DSIH_M3x4:
2398             nComponents = 4;
2399             tmp_ins.handler_idx = WINED3DSIH_DP3;
2400             break;
2401         case WINED3DSIH_M3x3:
2402             nComponents = 3;
2403             tmp_ins.handler_idx = WINED3DSIH_DP3;
2404             break;
2405         case WINED3DSIH_M3x2:
2406             nComponents = 2;
2407             tmp_ins.handler_idx = WINED3DSIH_DP3;
2408             break;
2409         default:
2410             FIXME("Unhandled opcode %#x\n", ins->handler_idx);
2411             break;
2412     }
2413
2414     tmp_dst = ins->dst[0];
2415     tmp_src[0] = ins->src[0];
2416     tmp_src[1] = ins->src[1];
2417     for (i = 0; i < nComponents; i++) {
2418         tmp_dst.write_mask = WINED3DSP_WRITEMASK_0 << i;
2419         shader_hw_map2gl(&tmp_ins);
2420         ++tmp_src[1].reg.idx;
2421     }
2422 }
2423
2424 static void shader_hw_rcp(const struct wined3d_shader_instruction *ins)
2425 {
2426     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2427
2428     char dst[50];
2429     char src[50];
2430
2431     shader_arb_get_dst_param(ins, &ins->dst[0], dst); /* Destination */
2432     shader_arb_get_src_param(ins, &ins->src[0], 0, src);
2433     if (ins->src[0].swizzle == WINED3DSP_NOSWIZZLE)
2434     {
2435         /* Dx sdk says .x is used if no swizzle is given, but our test shows that
2436          * .w is used
2437          */
2438         strcat(src, ".w");
2439     }
2440
2441     shader_addline(buffer, "RCP%s %s, %s;\n", shader_arb_get_modifier(ins), dst, src);
2442 }
2443
2444 static void shader_hw_scalar_op(const struct wined3d_shader_instruction *ins)
2445 {
2446     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2447     const char *instruction;
2448
2449     char dst[50];
2450     char src[50];
2451
2452     switch(ins->handler_idx)
2453     {
2454         case WINED3DSIH_RSQ:  instruction = "RSQ"; break;
2455         case WINED3DSIH_RCP:  instruction = "RCP"; break;
2456         case WINED3DSIH_EXP:  instruction = "EX2"; break;
2457         case WINED3DSIH_EXPP: instruction = "EXP"; break;
2458         default: instruction = "";
2459             FIXME("Unhandled opcode %#x\n", ins->handler_idx);
2460             break;
2461     }
2462
2463     shader_arb_get_dst_param(ins, &ins->dst[0], dst); /* Destination */
2464     shader_arb_get_src_param(ins, &ins->src[0], 0, src);
2465     if (ins->src[0].swizzle == WINED3DSP_NOSWIZZLE)
2466     {
2467         /* Dx sdk says .x is used if no swizzle is given, but our test shows that
2468          * .w is used
2469          */
2470         strcat(src, ".w");
2471     }
2472
2473     shader_addline(buffer, "%s%s %s, %s;\n", instruction, shader_arb_get_modifier(ins), dst, src);
2474 }
2475
2476 static void shader_hw_nrm(const struct wined3d_shader_instruction *ins)
2477 {
2478     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2479     char dst_name[50];
2480     char src_name[50];
2481     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2482     BOOL pshader = shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type);
2483     const char *zero = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ZERO);
2484
2485     shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2486     shader_arb_get_src_param(ins, &ins->src[0], 1 /* Use TB */, src_name);
2487
2488     /* In D3D, NRM of a vector with length zero returns zero. Catch this situation, as
2489      * otherwise NRM or RSQ would return NaN */
2490     if(pshader && priv->target_version >= NV3)
2491     {
2492         /* GL_NV_fragment_program2's NRM needs protection against length zero vectors too
2493          *
2494          * TODO: Find out if DP3+NRM+MOV is really faster than DP3+RSQ+MUL
2495          */
2496         shader_addline(buffer, "DP3C TA, %s, %s;\n", src_name, src_name);
2497         shader_addline(buffer, "NRM%s %s, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name);
2498         shader_addline(buffer, "MOV %s (EQ), %s;\n", dst_name, zero);
2499     }
2500     else if(priv->target_version >= NV2)
2501     {
2502         shader_addline(buffer, "DP3C TA.x, %s, %s;\n", src_name, src_name);
2503         shader_addline(buffer, "RSQ TA.x (NE), TA.x;\n");
2504         shader_addline(buffer, "MUL%s %s, %s, TA.x;\n", shader_arb_get_modifier(ins), dst_name,
2505                        src_name);
2506     }
2507     else
2508     {
2509         const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
2510
2511         shader_addline(buffer, "DP3 TA.x, %s, %s;\n", src_name, src_name);
2512         /* Pass any non-zero value to RSQ if the input vector has a length of zero. The
2513          * RSQ result doesn't matter, as long as multiplying it by 0 returns 0.
2514          */
2515         shader_addline(buffer, "SGE TA.y, -TA.x, %s;\n", zero);
2516         shader_addline(buffer, "MAD TA.x, %s, TA.y, TA.x;\n", one);
2517
2518         shader_addline(buffer, "RSQ TA.x, TA.x;\n");
2519         /* dst.w = src[0].w * 1 / (src.x^2 + src.y^2 + src.z^2)^(1/2) according to msdn*/
2520         shader_addline(buffer, "MUL%s %s, %s, TA.x;\n", shader_arb_get_modifier(ins), dst_name,
2521                     src_name);
2522     }
2523 }
2524
2525 static void shader_hw_lrp(const struct wined3d_shader_instruction *ins)
2526 {
2527     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2528     char dst_name[50];
2529     char src_name[3][50];
2530
2531     /* ARB_fragment_program has a convenient LRP instruction */
2532     if(shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type)) {
2533         shader_hw_map2gl(ins);
2534         return;
2535     }
2536
2537     shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2538     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
2539     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
2540     shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
2541
2542     shader_addline(buffer, "SUB TA, %s, %s;\n", src_name[1], src_name[2]);
2543     shader_addline(buffer, "MAD%s %s, %s, TA, %s;\n", shader_arb_get_modifier(ins),
2544                    dst_name, src_name[0], src_name[2]);
2545 }
2546
2547 static void shader_hw_sincos(const struct wined3d_shader_instruction *ins)
2548 {
2549     /* This instruction exists in ARB, but the d3d instruction takes two extra parameters which
2550      * must contain fixed constants. So we need a separate function to filter those constants and
2551      * can't use map2gl
2552      */
2553     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2554     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2555     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2556     char dst_name[50];
2557     char src_name0[50], src_name1[50], src_name2[50];
2558     BOOL is_color;
2559
2560     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name0);
2561     if(shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type)) {
2562         shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2563         /* No modifiers are supported on SCS */
2564         shader_addline(buffer, "SCS %s, %s;\n", dst_name, src_name0);
2565
2566         if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE)
2567         {
2568             shader_arb_get_register_name(ins, &dst->reg, src_name0, &is_color);
2569             shader_addline(buffer, "MOV_SAT %s, %s;\n", dst_name, src_name0);
2570         }
2571     } else if(priv->target_version >= NV2) {
2572         shader_arb_get_register_name(ins, &dst->reg, dst_name, &is_color);
2573
2574         /* Sincos writemask must be .x, .y or .xy */
2575         if(dst->write_mask & WINED3DSP_WRITEMASK_0)
2576             shader_addline(buffer, "COS%s %s.x, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name0);
2577         if(dst->write_mask & WINED3DSP_WRITEMASK_1)
2578             shader_addline(buffer, "SIN%s %s.y, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name0);
2579     } else {
2580         /* Approximate sine and cosine with a taylor series, as per math textbook. The application passes 8
2581          * helper constants(D3DSINCOSCONST1 and D3DSINCOSCONST2) in src1 and src2.
2582          *
2583          * sin(x) = x - x^3/3! + x^5/5! - x^7/7! + ...
2584          * cos(x) = 1 - x^2/2! + x^4/4! - x^6/6! + ...
2585          *
2586          * The constants we get are:
2587          *
2588          *  +1   +1,     -1     -1     +1      +1      -1       -1
2589          *      ---- ,  ---- , ---- , ----- , ----- , ----- , ------
2590          *      1!*2    2!*4   3!*8   4!*16   5!*32   6!*64   7!*128
2591          *
2592          * If used with x^2, x^3, x^4 etc they calculate sin(x/2) and cos(x/2):
2593          *
2594          * (x/2)^2 = x^2 / 4
2595          * (x/2)^3 = x^3 / 8
2596          * (x/2)^4 = x^4 / 16
2597          * (x/2)^5 = x^5 / 32
2598          * etc
2599          *
2600          * To get the final result:
2601          * sin(x) = 2 * sin(x/2) * cos(x/2)
2602          * cos(x) = cos(x/2)^2 - sin(x/2)^2
2603          * (from sin(x+y) and cos(x+y) rules)
2604          *
2605          * As per MSDN, dst.z is undefined after the operation, and so is
2606          * dst.x and dst.y if they're masked out by the writemask. Ie
2607          * sincos dst.y, src1, c0, c1
2608          * returns the sine in dst.y. dst.x and dst.z are undefined, dst.w is not touched. The assembler
2609          * vsa.exe also stops with an error if the dest register is the same register as the source
2610          * register. This means we can use dest.xyz as temporary storage. The assembler vsa.exe output also
2611          * indicates that sincos consumes 8 instruction slots in vs_2_0(and, strangely, in vs_3_0).
2612          */
2613         shader_arb_get_src_param(ins, &ins->src[1], 1, src_name1);
2614         shader_arb_get_src_param(ins, &ins->src[2], 2, src_name2);
2615         shader_arb_get_register_name(ins, &dst->reg, dst_name, &is_color);
2616
2617         shader_addline(buffer, "MUL %s.x, %s, %s;\n", dst_name, src_name0, src_name0);  /* x ^ 2 */
2618         shader_addline(buffer, "MUL TA.y, %s.x, %s;\n", dst_name, src_name0);           /* x ^ 3 */
2619         shader_addline(buffer, "MUL %s.y, TA.y, %s;\n", dst_name, src_name0);           /* x ^ 4 */
2620         shader_addline(buffer, "MUL TA.z, %s.y, %s;\n", dst_name, src_name0);           /* x ^ 5 */
2621         shader_addline(buffer, "MUL %s.z, TA.z, %s;\n", dst_name, src_name0);           /* x ^ 6 */
2622         shader_addline(buffer, "MUL TA.w, %s.z, %s;\n", dst_name, src_name0);           /* x ^ 7 */
2623
2624         /* sin(x/2)
2625          *
2626          * Unfortunately we don't get the constants in a DP4-capable form. Is there a way to
2627          * properly merge that with MULs in the code above?
2628          * The swizzles .yz and xw however fit into the .yzxw swizzle added to ps_2_0. Maybe
2629          * we can merge the sine and cosine MAD rows to calculate them together.
2630          */
2631         shader_addline(buffer, "MUL TA.x, %s, %s.w;\n", src_name0, src_name2); /* x^1, +1/(1!*2) */
2632         shader_addline(buffer, "MAD TA.x, TA.y, %s.x, TA.x;\n", src_name2); /* -1/(3!*8) */
2633         shader_addline(buffer, "MAD TA.x, TA.z, %s.w, TA.x;\n", src_name1); /* +1/(5!*32) */
2634         shader_addline(buffer, "MAD TA.x, TA.w, %s.x, TA.x;\n", src_name1); /* -1/(7!*128) */
2635
2636         /* cos(x/2) */
2637         shader_addline(buffer, "MAD TA.y, %s.x, %s.y, %s.z;\n", dst_name, src_name2, src_name2); /* -1/(2!*4), +1.0 */
2638         shader_addline(buffer, "MAD TA.y, %s.y, %s.z, TA.y;\n", dst_name, src_name1); /* +1/(4!*16) */
2639         shader_addline(buffer, "MAD TA.y, %s.z, %s.y, TA.y;\n", dst_name, src_name1); /* -1/(6!*64) */
2640
2641         if(dst->write_mask & WINED3DSP_WRITEMASK_0) {
2642             /* cos x */
2643             shader_addline(buffer, "MUL TA.z, TA.y, TA.y;\n");
2644             shader_addline(buffer, "MAD %s.x, -TA.x, TA.x, TA.z;\n", dst_name);
2645         }
2646         if(dst->write_mask & WINED3DSP_WRITEMASK_1) {
2647             /* sin x */
2648             shader_addline(buffer, "MUL %s.y, TA.x, TA.y;\n", dst_name);
2649             shader_addline(buffer, "ADD %s.y, %s.y, %s.y;\n", dst_name, dst_name, dst_name);
2650         }
2651     }
2652 }
2653
2654 static void shader_hw_sgn(const struct wined3d_shader_instruction *ins)
2655 {
2656     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2657     char dst_name[50];
2658     char src_name[50];
2659     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
2660
2661     shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2662     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name);
2663
2664     /* SGN is only valid in vertex shaders */
2665     if(ctx->target_version >= NV2) {
2666         shader_addline(buffer, "SSG%s %s, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name);
2667         return;
2668     }
2669
2670     /* If SRC > 0.0, -SRC < SRC = TRUE, otherwise false.
2671      * if SRC < 0.0,  SRC < -SRC = TRUE. If neither is true, src = 0.0
2672      */
2673     if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE) {
2674         shader_addline(buffer, "SLT %s, -%s, %s;\n", dst_name, src_name, src_name);
2675     } else {
2676         /* src contains TA? Write to the dest first. This won't overwrite our destination.
2677          * Then use TA, and calculate the final result
2678          *
2679          * Not reading from TA? Store the first result in TA to avoid overwriting the
2680          * destination if src reg = dst reg
2681          */
2682         if(strstr(src_name, "TA"))
2683         {
2684             shader_addline(buffer, "SLT %s,  %s, -%s;\n", dst_name, src_name, src_name);
2685             shader_addline(buffer, "SLT TA, -%s, %s;\n", src_name, src_name);
2686             shader_addline(buffer, "ADD %s, %s, -TA;\n", dst_name, dst_name);
2687         }
2688         else
2689         {
2690             shader_addline(buffer, "SLT TA, -%s, %s;\n", src_name, src_name);
2691             shader_addline(buffer, "SLT %s,  %s, -%s;\n", dst_name, src_name, src_name);
2692             shader_addline(buffer, "ADD %s, TA, -%s;\n", dst_name, dst_name);
2693         }
2694     }
2695 }
2696
2697 static void shader_hw_dsy(const struct wined3d_shader_instruction *ins)
2698 {
2699     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2700     char src[50];
2701     char dst[50];
2702     char dst_name[50];
2703     BOOL is_color;
2704
2705     shader_arb_get_dst_param(ins, &ins->dst[0], dst);
2706     shader_arb_get_src_param(ins, &ins->src[0], 0, src);
2707     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2708
2709     shader_addline(buffer, "DDY %s, %s;\n", dst, src);
2710     shader_addline(buffer, "MUL%s %s, %s, ycorrection.y;\n", shader_arb_get_modifier(ins), dst, dst_name);
2711 }
2712
2713 static DWORD abs_modifier(DWORD mod, BOOL *need_abs)
2714 {
2715     *need_abs = FALSE;
2716
2717     switch(mod)
2718     {
2719         case WINED3DSPSM_NONE:      return WINED3DSPSM_ABS;
2720         case WINED3DSPSM_NEG:       return WINED3DSPSM_ABS;
2721         case WINED3DSPSM_BIAS:      *need_abs = TRUE; return WINED3DSPSM_BIAS;
2722         case WINED3DSPSM_BIASNEG:   *need_abs = TRUE; return WINED3DSPSM_BIASNEG;
2723         case WINED3DSPSM_SIGN:      *need_abs = TRUE; return WINED3DSPSM_SIGN;
2724         case WINED3DSPSM_SIGNNEG:   *need_abs = TRUE; return WINED3DSPSM_SIGNNEG;
2725         case WINED3DSPSM_COMP:      *need_abs = TRUE; return WINED3DSPSM_COMP;
2726         case WINED3DSPSM_X2:        *need_abs = TRUE; return WINED3DSPSM_X2;
2727         case WINED3DSPSM_X2NEG:     *need_abs = TRUE; return WINED3DSPSM_X2NEG;
2728         case WINED3DSPSM_DZ:        *need_abs = TRUE; return WINED3DSPSM_DZ;
2729         case WINED3DSPSM_DW:        *need_abs = TRUE; return WINED3DSPSM_DW;
2730         case WINED3DSPSM_ABS:       return WINED3DSPSM_ABS;
2731         case WINED3DSPSM_ABSNEG:    return WINED3DSPSM_ABS;
2732     }
2733     FIXME("Unknown modifier %u\n", mod);
2734     return mod;
2735 }
2736
2737 static void shader_hw_log(const struct wined3d_shader_instruction *ins)
2738 {
2739     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2740     char src0[50], dst[50];
2741     struct wined3d_shader_src_param src0_copy = ins->src[0];
2742     BOOL need_abs = FALSE;
2743     const char *instr;
2744
2745     switch(ins->handler_idx)
2746     {
2747         case WINED3DSIH_LOG:  instr = "LG2"; break;
2748         case WINED3DSIH_LOGP: instr = "LOG"; break;
2749         default:
2750             ERR("Unexpected instruction %d\n", ins->handler_idx);
2751             return;
2752     }
2753
2754     /* LOG and LOGP operate on the absolute value of the input */
2755     src0_copy.modifiers = abs_modifier(src0_copy.modifiers, &need_abs);
2756
2757     shader_arb_get_dst_param(ins, &ins->dst[0], dst);
2758     shader_arb_get_src_param(ins, &src0_copy, 0, src0);
2759
2760     if(need_abs)
2761     {
2762         shader_addline(buffer, "ABS TA, %s;\n", src0);
2763         shader_addline(buffer, "%s%s %s, TA;\n", instr, shader_arb_get_modifier(ins), dst);
2764     }
2765     else
2766     {
2767         shader_addline(buffer, "%s%s %s, %s;\n", instr, shader_arb_get_modifier(ins), dst, src0);
2768     }
2769 }
2770
2771 static void shader_hw_pow(const struct wined3d_shader_instruction *ins)
2772 {
2773     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2774     char src0[50], src1[50], dst[50];
2775     struct wined3d_shader_src_param src0_copy = ins->src[0];
2776     BOOL need_abs = FALSE;
2777     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2778     const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
2779
2780     /* POW operates on the absolute value of the input */
2781     src0_copy.modifiers = abs_modifier(src0_copy.modifiers, &need_abs);
2782
2783     shader_arb_get_dst_param(ins, &ins->dst[0], dst);
2784     shader_arb_get_src_param(ins, &src0_copy, 0, src0);
2785     shader_arb_get_src_param(ins, &ins->src[1], 1, src1);
2786
2787     if (need_abs)
2788         shader_addline(buffer, "ABS TA.x, %s;\n", src0);
2789     else
2790         shader_addline(buffer, "MOV TA.x, %s;\n", src0);
2791
2792     if (priv->target_version >= NV2)
2793     {
2794         shader_addline(buffer, "MOVC TA.y, %s;\n", src1);
2795         shader_addline(buffer, "POW%s %s, TA.x, TA.y;\n", shader_arb_get_modifier(ins), dst);
2796         shader_addline(buffer, "MOV %s (EQ.y), %s;\n", dst, one);
2797     }
2798     else
2799     {
2800         const char *zero = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ZERO);
2801         const char *flt_eps = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_EPS);
2802
2803         shader_addline(buffer, "ABS TA.y, %s;\n", src1);
2804         shader_addline(buffer, "SGE TA.y, -TA.y, %s;\n", zero);
2805         /* Possibly add flt_eps to avoid getting float special values */
2806         shader_addline(buffer, "MAD TA.z, TA.y, %s, %s;\n", flt_eps, src1);
2807         shader_addline(buffer, "POW%s TA.x, TA.x, TA.z;\n", shader_arb_get_modifier(ins));
2808         shader_addline(buffer, "MAD TA.x, -TA.x, TA.y, TA.x;\n");
2809         shader_addline(buffer, "MAD %s, TA.y, %s, TA.x;\n", dst, one);
2810     }
2811 }
2812
2813 static void shader_hw_loop(const struct wined3d_shader_instruction *ins)
2814 {
2815     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2816     char src_name[50];
2817     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2818
2819     /* src0 is aL */
2820     shader_arb_get_src_param(ins, &ins->src[1], 0, src_name);
2821
2822     if(vshader)
2823     {
2824         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2825         struct list *e = list_head(&priv->control_frames);
2826         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2827
2828         if(priv->loop_depth > 1) shader_addline(buffer, "PUSHA aL;\n");
2829         /* The constant loader makes sure to load -1 into iX.w */
2830         shader_addline(buffer, "ARLC aL, %s.xywz;\n", src_name);
2831         shader_addline(buffer, "BRA loop_%u_end (LE.x);\n", control_frame->no.loop);
2832         shader_addline(buffer, "loop_%u_start:\n", control_frame->no.loop);
2833     }
2834     else
2835     {
2836         shader_addline(buffer, "LOOP %s;\n", src_name);
2837     }
2838 }
2839
2840 static void shader_hw_rep(const struct wined3d_shader_instruction *ins)
2841 {
2842     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2843     char src_name[50];
2844     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2845
2846     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name);
2847
2848     /* The constant loader makes sure to load -1 into iX.w */
2849     if(vshader)
2850     {
2851         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2852         struct list *e = list_head(&priv->control_frames);
2853         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2854
2855         if(priv->loop_depth > 1) shader_addline(buffer, "PUSHA aL;\n");
2856
2857         shader_addline(buffer, "ARLC aL, %s.xywz;\n", src_name);
2858         shader_addline(buffer, "BRA loop_%u_end (LE.x);\n", control_frame->no.loop);
2859         shader_addline(buffer, "loop_%u_start:\n", control_frame->no.loop);
2860     }
2861     else
2862     {
2863         shader_addline(buffer, "REP %s;\n", src_name);
2864     }
2865 }
2866
2867 static void shader_hw_endloop(const struct wined3d_shader_instruction *ins)
2868 {
2869     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2870     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2871
2872     if(vshader)
2873     {
2874         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2875         struct list *e = list_head(&priv->control_frames);
2876         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2877
2878         shader_addline(buffer, "ARAC aL.xy, aL;\n");
2879         shader_addline(buffer, "BRA loop_%u_start (GT.x);\n", control_frame->no.loop);
2880         shader_addline(buffer, "loop_%u_end:\n", control_frame->no.loop);
2881
2882         if(priv->loop_depth > 1) shader_addline(buffer, "POPA aL;\n");
2883     }
2884     else
2885     {
2886         shader_addline(buffer, "ENDLOOP;\n");
2887     }
2888 }
2889
2890 static void shader_hw_endrep(const struct wined3d_shader_instruction *ins)
2891 {
2892     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2893     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2894
2895     if(vshader)
2896     {
2897         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2898         struct list *e = list_head(&priv->control_frames);
2899         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2900
2901         shader_addline(buffer, "ARAC aL.xy, aL;\n");
2902         shader_addline(buffer, "BRA loop_%u_start (GT.x);\n", control_frame->no.loop);
2903         shader_addline(buffer, "loop_%u_end:\n", control_frame->no.loop);
2904
2905         if(priv->loop_depth > 1) shader_addline(buffer, "POPA aL;\n");
2906     }
2907     else
2908     {
2909         shader_addline(buffer, "ENDREP;\n");
2910     }
2911 }
2912
2913 static const struct control_frame *find_last_loop(const struct shader_arb_ctx_priv *priv)
2914 {
2915     struct control_frame *control_frame;
2916
2917     LIST_FOR_EACH_ENTRY(control_frame, &priv->control_frames, struct control_frame, entry)
2918     {
2919         if(control_frame->type == LOOP || control_frame->type == REP) return control_frame;
2920     }
2921     ERR("Could not find loop for break\n");
2922     return NULL;
2923 }
2924
2925 static void shader_hw_break(const struct wined3d_shader_instruction *ins)
2926 {
2927     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2928     const struct control_frame *control_frame = find_last_loop(ins->ctx->backend_data);
2929     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2930
2931     if(vshader)
2932     {
2933         shader_addline(buffer, "BRA loop_%u_end;\n", control_frame->no.loop);
2934     }
2935     else
2936     {
2937         shader_addline(buffer, "BRK;\n");
2938     }
2939 }
2940
2941 static const char *get_compare(COMPARISON_TYPE flags)
2942 {
2943     switch (flags)
2944     {
2945         case COMPARISON_GT: return "GT";
2946         case COMPARISON_EQ: return "EQ";
2947         case COMPARISON_GE: return "GE";
2948         case COMPARISON_LT: return "LT";
2949         case COMPARISON_NE: return "NE";
2950         case COMPARISON_LE: return "LE";
2951         default:
2952             FIXME("Unrecognized comparison value: %u\n", flags);
2953             return "(\?\?)";
2954     }
2955 }
2956
2957 static COMPARISON_TYPE invert_compare(COMPARISON_TYPE flags)
2958 {
2959     switch (flags)
2960     {
2961         case COMPARISON_GT: return COMPARISON_LE;
2962         case COMPARISON_EQ: return COMPARISON_NE;
2963         case COMPARISON_GE: return COMPARISON_LT;
2964         case COMPARISON_LT: return COMPARISON_GE;
2965         case COMPARISON_NE: return COMPARISON_EQ;
2966         case COMPARISON_LE: return COMPARISON_GT;
2967         default:
2968             FIXME("Unrecognized comparison value: %u\n", flags);
2969             return -1;
2970     }
2971 }
2972
2973 static void shader_hw_breakc(const struct wined3d_shader_instruction *ins)
2974 {
2975     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2976     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2977     const struct control_frame *control_frame = find_last_loop(ins->ctx->backend_data);
2978     char src_name0[50];
2979     char src_name1[50];
2980     const char *comp = get_compare(ins->flags);
2981
2982     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name0);
2983     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name1);
2984
2985     if(vshader)
2986     {
2987         /* SUBC CC, src0, src1" works only in pixel shaders, so use TA to throw
2988          * away the subtraction result
2989          */
2990         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
2991         shader_addline(buffer, "BRA loop_%u_end (%s.x);\n", control_frame->no.loop, comp);
2992     }
2993     else
2994     {
2995         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
2996         shader_addline(buffer, "BRK (%s.x);\n", comp);
2997     }
2998 }
2999
3000 static void shader_hw_ifc(const struct wined3d_shader_instruction *ins)
3001 {
3002     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3003     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3004     struct list *e = list_head(&priv->control_frames);
3005     struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
3006     const char *comp;
3007     char src_name0[50];
3008     char src_name1[50];
3009     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
3010
3011     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name0);
3012     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name1);
3013
3014     if(vshader)
3015     {
3016         /* Invert the flag. We jump to the else label if the condition is NOT true */
3017         comp = get_compare(invert_compare(ins->flags));
3018         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
3019         shader_addline(buffer, "BRA ifc_%u_else (%s.x);\n", control_frame->no.ifc, comp);
3020     }
3021     else
3022     {
3023         comp = get_compare(ins->flags);
3024         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
3025         shader_addline(buffer, "IF %s.x;\n", comp);
3026     }
3027 }
3028
3029 static void shader_hw_else(const struct wined3d_shader_instruction *ins)
3030 {
3031     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3032     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3033     struct list *e = list_head(&priv->control_frames);
3034     struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
3035     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
3036
3037     if(vshader)
3038     {
3039         shader_addline(buffer, "BRA ifc_%u_endif;\n", control_frame->no.ifc);
3040         shader_addline(buffer, "ifc_%u_else:\n", control_frame->no.ifc);
3041         control_frame->had_else = TRUE;
3042     }
3043     else
3044     {
3045         shader_addline(buffer, "ELSE;\n");
3046     }
3047 }
3048
3049 static void shader_hw_endif(const struct wined3d_shader_instruction *ins)
3050 {
3051     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3052     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3053     struct list *e = list_head(&priv->control_frames);
3054     struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
3055     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
3056
3057     if(vshader)
3058     {
3059         if(control_frame->had_else)
3060         {
3061             shader_addline(buffer, "ifc_%u_endif:\n", control_frame->no.ifc);
3062         }
3063         else
3064         {
3065             shader_addline(buffer, "#No else branch. else is endif\n");
3066             shader_addline(buffer, "ifc_%u_else:\n", control_frame->no.ifc);
3067         }
3068     }
3069     else
3070     {
3071         shader_addline(buffer, "ENDIF;\n");
3072     }
3073 }
3074
3075 static void shader_hw_texldd(const struct wined3d_shader_instruction *ins)
3076 {
3077     DWORD sampler_idx = ins->src[1].reg.idx;
3078     char reg_dest[40];
3079     char reg_src[3][40];
3080     WORD flags = TEX_DERIV;
3081
3082     shader_arb_get_dst_param(ins, &ins->dst[0], reg_dest);
3083     shader_arb_get_src_param(ins, &ins->src[0], 0, reg_src[0]);
3084     shader_arb_get_src_param(ins, &ins->src[2], 1, reg_src[1]);
3085     shader_arb_get_src_param(ins, &ins->src[3], 2, reg_src[2]);
3086
3087     if (ins->flags & WINED3DSI_TEXLD_PROJECT) flags |= TEX_PROJ;
3088     if (ins->flags & WINED3DSI_TEXLD_BIAS) flags |= TEX_BIAS;
3089
3090     shader_hw_sample(ins, sampler_idx, reg_dest, reg_src[0], flags, reg_src[1], reg_src[2]);
3091 }
3092
3093 static void shader_hw_texldl(const struct wined3d_shader_instruction *ins)
3094 {
3095     DWORD sampler_idx = ins->src[1].reg.idx;
3096     char reg_dest[40];
3097     char reg_coord[40];
3098     WORD flags = TEX_LOD;
3099
3100     shader_arb_get_dst_param(ins, &ins->dst[0], reg_dest);
3101     shader_arb_get_src_param(ins, &ins->src[0], 0, reg_coord);
3102
3103     if (ins->flags & WINED3DSI_TEXLD_PROJECT) flags |= TEX_PROJ;
3104     if (ins->flags & WINED3DSI_TEXLD_BIAS) flags |= TEX_BIAS;
3105
3106     shader_hw_sample(ins, sampler_idx, reg_dest, reg_coord, flags, NULL, NULL);
3107 }
3108
3109 static void shader_hw_label(const struct wined3d_shader_instruction *ins)
3110 {
3111     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3112     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3113
3114     priv->in_main_func = FALSE;
3115     /* Call instructions activate the NV extensions, not labels and rets. If there is an uncalled
3116      * subroutine, don't generate a label that will make GL complain
3117      */
3118     if(priv->target_version == ARB) return;
3119
3120     shader_addline(buffer, "l%u:\n", ins->src[0].reg.idx);
3121 }
3122
3123 static void vshader_add_footer(struct shader_arb_ctx_priv *priv_ctx,
3124         const struct arb_vshader_private *shader_data, const struct arb_vs_compile_args *args,
3125         const struct wined3d_shader_reg_maps *reg_maps, const struct wined3d_gl_info *gl_info,
3126         struct wined3d_shader_buffer *buffer)
3127 {
3128     unsigned int i;
3129
3130     /* The D3DRS_FOGTABLEMODE render state defines if the shader-generated fog coord is used
3131      * or if the fragment depth is used. If the fragment depth is used(FOGTABLEMODE != NONE),
3132      * the fog frag coord is thrown away. If the fog frag coord is used, but not written by
3133      * the shader, it is set to 0.0(fully fogged, since start = 1.0, end = 0.0)
3134      */
3135     if (args->super.fog_src == VS_FOG_Z)
3136         shader_addline(buffer, "MOV result.fogcoord, TMP_OUT.z;\n");
3137     else if (!reg_maps->fog)
3138         /* posFixup.x is always 1.0, so we can safely use it */
3139         shader_addline(buffer, "ADD result.fogcoord, posFixup.x, -posFixup.x;\n");
3140
3141     /* Clipplanes are always stored without y inversion */
3142     if (use_nv_clip(gl_info) && priv_ctx->target_version >= NV2)
3143     {
3144         if (args->super.clip_enabled)
3145         {
3146             for (i = 0; i < priv_ctx->vs_clipplanes; i++)
3147             {
3148                 shader_addline(buffer, "DP4 result.clip[%u].x, TMP_OUT, state.clip[%u].plane;\n", i, i);
3149             }
3150         }
3151     }
3152     else if (args->clip.boolclip.clip_texcoord)
3153     {
3154         unsigned int cur_clip = 0;
3155         char component[4] = {'x', 'y', 'z', 'w'};
3156         const char *zero = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_ZERO);
3157
3158         for (i = 0; i < gl_info->limits.clipplanes; ++i)
3159         {
3160             if (args->clip.boolclip.clipplane_mask & (1 << i))
3161             {
3162                 shader_addline(buffer, "DP4 TA.%c, TMP_OUT, state.clip[%u].plane;\n",
3163                                component[cur_clip++], i);
3164             }
3165         }
3166         switch (cur_clip)
3167         {
3168             case 0:
3169                 shader_addline(buffer, "MOV TA, %s;\n", zero);
3170                 break;
3171             case 1:
3172                 shader_addline(buffer, "MOV TA.yzw, %s;\n", zero);
3173                 break;
3174             case 2:
3175                 shader_addline(buffer, "MOV TA.zw, %s;\n", zero);
3176                 break;
3177             case 3:
3178                 shader_addline(buffer, "MOV TA.w, %s;\n", zero);
3179                 break;
3180         }
3181         shader_addline(buffer, "MOV result.texcoord[%u], TA;\n",
3182                        args->clip.boolclip.clip_texcoord - 1);
3183     }
3184
3185     /* Write the final position.
3186      *
3187      * OpenGL coordinates specify the center of the pixel while d3d coords specify
3188      * the corner. The offsets are stored in z and w in posFixup. posFixup.y contains
3189      * 1.0 or -1.0 to turn the rendering upside down for offscreen rendering. PosFixup.x
3190      * contains 1.0 to allow a mad, but arb vs swizzles are too restricted for that.
3191      */
3192     shader_addline(buffer, "MUL TA, posFixup, TMP_OUT.w;\n");
3193     shader_addline(buffer, "ADD TMP_OUT.x, TMP_OUT.x, TA.z;\n");
3194     shader_addline(buffer, "MAD TMP_OUT.y, TMP_OUT.y, posFixup.y, TA.w;\n");
3195
3196     /* Z coord [0;1]->[-1;1] mapping, see comment in transform_projection in state.c
3197      * and the glsl equivalent
3198      */
3199     if (need_helper_const(shader_data, reg_maps, gl_info))
3200     {
3201         const char *two = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_TWO);
3202         shader_addline(buffer, "MAD TMP_OUT.z, TMP_OUT.z, %s, -TMP_OUT.w;\n", two);
3203     }
3204     else
3205     {
3206         shader_addline(buffer, "ADD TMP_OUT.z, TMP_OUT.z, TMP_OUT.z;\n");
3207         shader_addline(buffer, "ADD TMP_OUT.z, TMP_OUT.z, -TMP_OUT.w;\n");
3208     }
3209
3210     shader_addline(buffer, "MOV result.position, TMP_OUT;\n");
3211
3212     priv_ctx->footer_written = TRUE;
3213 }
3214
3215 static void shader_hw_ret(const struct wined3d_shader_instruction *ins)
3216 {
3217     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3218     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3219     struct wined3d_shader *shader = ins->ctx->shader;
3220     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
3221
3222     if(priv->target_version == ARB) return;
3223
3224     if(vshader)
3225     {
3226         if (priv->in_main_func) vshader_add_footer(priv, shader->backend_data,
3227                 priv->cur_vs_args, ins->ctx->reg_maps, ins->ctx->gl_info, buffer);
3228     }
3229
3230     shader_addline(buffer, "RET;\n");
3231 }
3232
3233 static void shader_hw_call(const struct wined3d_shader_instruction *ins)
3234 {
3235     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3236     shader_addline(buffer, "CAL l%u;\n", ins->src[0].reg.idx);
3237 }
3238
3239 /* GL locking is done by the caller */
3240 static GLuint create_arb_blt_vertex_program(const struct wined3d_gl_info *gl_info)
3241 {
3242     GLuint program_id = 0;
3243     GLint pos;
3244
3245     const char *blt_vprogram =
3246         "!!ARBvp1.0\n"
3247         "PARAM c[1] = { { 1, 0.5 } };\n"
3248         "MOV result.position, vertex.position;\n"
3249         "MOV result.color, c[0].x;\n"
3250         "MOV result.texcoord[0], vertex.texcoord[0];\n"
3251         "END\n";
3252
3253     GL_EXTCALL(glGenProgramsARB(1, &program_id));
3254     GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, program_id));
3255     GL_EXTCALL(glProgramStringARB(GL_VERTEX_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
3256             strlen(blt_vprogram), blt_vprogram));
3257     checkGLcall("glProgramStringARB()");
3258
3259     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
3260     if (pos != -1)
3261     {
3262         FIXME("Vertex program error at position %d: %s\n\n", pos,
3263             debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
3264         shader_arb_dump_program_source(blt_vprogram);
3265     }
3266     else
3267     {
3268         GLint native;
3269
3270         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
3271         checkGLcall("glGetProgramivARB()");
3272         if (!native) WARN("Program exceeds native resource limits.\n");
3273     }
3274
3275     return program_id;
3276 }
3277
3278 /* GL locking is done by the caller */
3279 static GLuint create_arb_blt_fragment_program(const struct wined3d_gl_info *gl_info,
3280         enum tex_types tex_type, BOOL masked)
3281 {
3282     GLuint program_id = 0;
3283     const char *fprogram;
3284     GLint pos;
3285
3286     static const char * const blt_fprograms_full[tex_type_count] =
3287     {
3288         /* tex_1d */
3289         NULL,
3290         /* tex_2d */
3291         "!!ARBfp1.0\n"
3292         "TEMP R0;\n"
3293         "TEX R0.x, fragment.texcoord[0], texture[0], 2D;\n"
3294         "MOV result.depth.z, R0.x;\n"
3295         "END\n",
3296         /* tex_3d */
3297         NULL,
3298         /* tex_cube */
3299         "!!ARBfp1.0\n"
3300         "TEMP R0;\n"
3301         "TEX R0.x, fragment.texcoord[0], texture[0], CUBE;\n"
3302         "MOV result.depth.z, R0.x;\n"
3303         "END\n",
3304         /* tex_rect */
3305         "!!ARBfp1.0\n"
3306         "TEMP R0;\n"
3307         "TEX R0.x, fragment.texcoord[0], texture[0], RECT;\n"
3308         "MOV result.depth.z, R0.x;\n"
3309         "END\n",
3310     };
3311
3312     static const char * const blt_fprograms_masked[tex_type_count] =
3313     {
3314         /* tex_1d */
3315         NULL,
3316         /* tex_2d */
3317         "!!ARBfp1.0\n"
3318         "PARAM mask = program.local[0];\n"
3319         "TEMP R0;\n"
3320         "SLT R0.xy, fragment.position, mask.zwzw;\n"
3321         "MUL R0.x, R0.x, R0.y;\n"
3322         "KIL -R0.x;\n"
3323         "TEX R0.x, fragment.texcoord[0], texture[0], 2D;\n"
3324         "MOV result.depth.z, R0.x;\n"
3325         "END\n",
3326         /* tex_3d */
3327         NULL,
3328         /* tex_cube */
3329         "!!ARBfp1.0\n"
3330         "PARAM mask = program.local[0];\n"
3331         "TEMP R0;\n"
3332         "SLT R0.xy, fragment.position, mask.zwzw;\n"
3333         "MUL R0.x, R0.x, R0.y;\n"
3334         "KIL -R0.x;\n"
3335         "TEX R0.x, fragment.texcoord[0], texture[0], CUBE;\n"
3336         "MOV result.depth.z, R0.x;\n"
3337         "END\n",
3338         /* tex_rect */
3339         "!!ARBfp1.0\n"
3340         "PARAM mask = program.local[0];\n"
3341         "TEMP R0;\n"
3342         "SLT R0.xy, fragment.position, mask.zwzw;\n"
3343         "MUL R0.x, R0.x, R0.y;\n"
3344         "KIL -R0.x;\n"
3345         "TEX R0.x, fragment.texcoord[0], texture[0], RECT;\n"
3346         "MOV result.depth.z, R0.x;\n"
3347         "END\n",
3348     };
3349
3350     fprogram = masked ? blt_fprograms_masked[tex_type] : blt_fprograms_full[tex_type];
3351     if (!fprogram)
3352     {
3353         FIXME("tex_type %#x not supported, falling back to tex_2d\n", tex_type);
3354         tex_type = tex_2d;
3355         fprogram = masked ? blt_fprograms_masked[tex_type] : blt_fprograms_full[tex_type];
3356     }
3357
3358     GL_EXTCALL(glGenProgramsARB(1, &program_id));
3359     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, program_id));
3360     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB, strlen(fprogram), fprogram));
3361     checkGLcall("glProgramStringARB()");
3362
3363     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
3364     if (pos != -1)
3365     {
3366         FIXME("Fragment program error at position %d: %s\n\n", pos,
3367             debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
3368         shader_arb_dump_program_source(fprogram);
3369     }
3370     else
3371     {
3372         GLint native;
3373
3374         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
3375         checkGLcall("glGetProgramivARB()");
3376         if (!native) WARN("Program exceeds native resource limits.\n");
3377     }
3378
3379     return program_id;
3380 }
3381
3382 static void arbfp_add_sRGB_correction(struct wined3d_shader_buffer *buffer, const char *fragcolor,
3383         const char *tmp1, const char *tmp2, const char *tmp3, const char *tmp4, BOOL condcode)
3384 {
3385     /* Perform sRGB write correction. See GLX_EXT_framebuffer_sRGB */
3386
3387     if(condcode)
3388     {
3389         /* Sigh. MOVC CC doesn't work, so use one of the temps as dummy dest */
3390         shader_addline(buffer, "SUBC %s, %s.x, srgb_consts1.y;\n", tmp1, fragcolor);
3391         /* Calculate the > 0.0031308 case */
3392         shader_addline(buffer, "POW %s.x (GE), %s.x, srgb_consts1.z;\n", fragcolor, fragcolor);
3393         shader_addline(buffer, "POW %s.y (GE), %s.y, srgb_consts1.z;\n", fragcolor, fragcolor);
3394         shader_addline(buffer, "POW %s.z (GE), %s.z, srgb_consts1.z;\n", fragcolor, fragcolor);
3395         shader_addline(buffer, "MUL %s.xyz (GE), %s, srgb_consts1.w;\n", fragcolor, fragcolor);
3396         shader_addline(buffer, "SUB %s.xyz (GE), %s, srgb_consts2.x;\n", fragcolor, fragcolor);
3397         /* Calculate the < case */
3398         shader_addline(buffer, "MUL %s.xyz (LT), srgb_consts1.x, %s;\n", fragcolor, fragcolor);
3399     }
3400     else
3401     {
3402         /* Calculate the > 0.0031308 case */
3403         shader_addline(buffer, "POW %s.x, %s.x, srgb_consts1.z;\n", tmp1, fragcolor);
3404         shader_addline(buffer, "POW %s.y, %s.y, srgb_consts1.z;\n", tmp1, fragcolor);
3405         shader_addline(buffer, "POW %s.z, %s.z, srgb_consts1.z;\n", tmp1, fragcolor);
3406         shader_addline(buffer, "MUL %s, %s, srgb_consts1.w;\n", tmp1, tmp1);
3407         shader_addline(buffer, "SUB %s, %s, srgb_consts2.x;\n", tmp1, tmp1);
3408         /* Calculate the < case */
3409         shader_addline(buffer, "MUL %s, srgb_consts1.x, %s;\n", tmp2, fragcolor);
3410         /* Get 1.0 / 0.0 masks for > 0.0031308 and < 0.0031308 */
3411         shader_addline(buffer, "SLT %s, srgb_consts1.y, %s;\n", tmp3, fragcolor);
3412         shader_addline(buffer, "SGE %s, srgb_consts1.y, %s;\n", tmp4, fragcolor);
3413         /* Store the components > 0.0031308 in the destination */
3414         shader_addline(buffer, "MUL %s.xyz, %s, %s;\n", fragcolor, tmp1, tmp3);
3415         /* Add the components that are < 0.0031308 */
3416         shader_addline(buffer, "MAD %s.xyz, %s, %s, %s;\n", fragcolor, tmp2, tmp4, fragcolor);
3417         /* Move everything into result.color at once. Nvidia hardware cannot handle partial
3418         * result.color writes(.rgb first, then .a), or handle overwriting already written
3419         * components. The assembler uses a temporary register in this case, which is usually
3420         * not allocated from one of our registers that were used earlier.
3421         */
3422     }
3423     /* [0.0;1.0] clamping. Not needed, this is done implicitly */
3424 }
3425
3426 static const DWORD *find_loop_control_values(const struct wined3d_shader *shader, DWORD idx)
3427 {
3428     const local_constant *constant;
3429
3430     LIST_FOR_EACH_ENTRY(constant, &shader->constantsI, local_constant, entry)
3431     {
3432         if (constant->idx == idx)
3433         {
3434             return constant->value;
3435         }
3436     }
3437     return NULL;
3438 }
3439
3440 static void init_ps_input(const struct wined3d_shader *shader,
3441         const struct arb_ps_compile_args *args, struct shader_arb_ctx_priv *priv)
3442 {
3443     static const char * const texcoords[8] =
3444     {
3445         "fragment.texcoord[0]", "fragment.texcoord[1]", "fragment.texcoord[2]", "fragment.texcoord[3]",
3446         "fragment.texcoord[4]", "fragment.texcoord[5]", "fragment.texcoord[6]", "fragment.texcoord[7]"
3447     };
3448     unsigned int i;
3449     const struct wined3d_shader_signature_element *sig = shader->input_signature;
3450     const char *semantic_name;
3451     DWORD semantic_idx;
3452
3453     switch(args->super.vp_mode)
3454     {
3455         case pretransformed:
3456         case fixedfunction:
3457             /* The pixelshader has to collect the varyings on its own. In any case properly load
3458              * color0 and color1. In the case of pretransformed vertices also load texcoords. Set
3459              * other attribs to 0.0.
3460              *
3461              * For fixedfunction this behavior is correct, according to the tests. For pretransformed
3462              * we'd either need a replacement shader that can load other attribs like BINORMAL, or
3463              * load the texcoord attrib pointers to match the pixel shader signature
3464              */
3465             for(i = 0; i < MAX_REG_INPUT; i++)
3466             {
3467                 semantic_name = sig[i].semantic_name;
3468                 semantic_idx = sig[i].semantic_idx;
3469                 if (!semantic_name) continue;
3470
3471                 if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_COLOR))
3472                 {
3473                     if (!semantic_idx) priv->ps_input[i] = "fragment.color.primary";
3474                     else if(semantic_idx == 1) priv->ps_input[i] = "fragment.color.secondary";
3475                     else priv->ps_input[i] = "0.0";
3476                 }
3477                 else if(args->super.vp_mode == fixedfunction)
3478                 {
3479                     priv->ps_input[i] = "0.0";
3480                 }
3481                 else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_TEXCOORD))
3482                 {
3483                     if(semantic_idx < 8) priv->ps_input[i] = texcoords[semantic_idx];
3484                     else priv->ps_input[i] = "0.0";
3485                 }
3486                 else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_FOG))
3487                 {
3488                     if (!semantic_idx) priv->ps_input[i] = "fragment.fogcoord";
3489                     else priv->ps_input[i] = "0.0";
3490                 }
3491                 else
3492                 {
3493                     priv->ps_input[i] = "0.0";
3494                 }
3495
3496                 TRACE("v%u, semantic %s%u is %s\n", i, semantic_name, semantic_idx, priv->ps_input[i]);
3497             }
3498             break;
3499
3500         case vertexshader:
3501             /* That one is easy. The vertex shaders provide v0-v7 in fragment.texcoord and v8 and v9 in
3502              * fragment.color
3503              */
3504             for(i = 0; i < 8; i++)
3505             {
3506                 priv->ps_input[i] = texcoords[i];
3507             }
3508             priv->ps_input[8] = "fragment.color.primary";
3509             priv->ps_input[9] = "fragment.color.secondary";
3510             break;
3511     }
3512 }
3513
3514 /* GL locking is done by the caller */
3515 static GLuint shader_arb_generate_pshader(struct wined3d_shader *shader,
3516         const struct wined3d_gl_info *gl_info, struct wined3d_shader_buffer *buffer,
3517         const struct arb_ps_compile_args *args, struct arb_ps_compiled_shader *compiled)
3518 {
3519     const struct wined3d_shader_reg_maps *reg_maps = &shader->reg_maps;
3520     const DWORD *function = shader->function;
3521     const local_constant *lconst;
3522     GLuint retval;
3523     char fragcolor[16];
3524     DWORD *lconst_map = local_const_mapping(shader), next_local, cur;
3525     struct shader_arb_ctx_priv priv_ctx;
3526     BOOL dcl_td = FALSE;
3527     BOOL want_nv_prog = FALSE;
3528     struct arb_pshader_private *shader_priv = shader->backend_data;
3529     GLint errPos;
3530     DWORD map;
3531
3532     char srgbtmp[4][4];
3533     unsigned int i, found = 0;
3534
3535     for (i = 0, map = reg_maps->temporary; map; map >>= 1, ++i)
3536     {
3537         if (!(map & 1)
3538                 || (shader->u.ps.color0_mov && i == shader->u.ps.color0_reg)
3539                 || (reg_maps->shader_version.major < 2 && !i))
3540             continue;
3541
3542         sprintf(srgbtmp[found], "R%u", i);
3543         ++found;
3544         if (found == 4) break;
3545     }
3546
3547     switch(found) {
3548         case 0:
3549             sprintf(srgbtmp[0], "TA");
3550             sprintf(srgbtmp[1], "TB");
3551             sprintf(srgbtmp[2], "TC");
3552             sprintf(srgbtmp[3], "TD");
3553             dcl_td = TRUE;
3554             break;
3555         case 1:
3556             sprintf(srgbtmp[1], "TA");
3557             sprintf(srgbtmp[2], "TB");
3558             sprintf(srgbtmp[3], "TC");
3559             break;
3560         case 2:
3561             sprintf(srgbtmp[2], "TA");
3562             sprintf(srgbtmp[3], "TB");
3563             break;
3564         case 3:
3565             sprintf(srgbtmp[3], "TA");
3566             break;
3567         case 4:
3568             break;
3569     }
3570
3571     /*  Create the hw ARB shader */
3572     memset(&priv_ctx, 0, sizeof(priv_ctx));
3573     priv_ctx.cur_ps_args = args;
3574     priv_ctx.compiled_fprog = compiled;
3575     priv_ctx.cur_np2fixup_info = &compiled->np2fixup_info;
3576     init_ps_input(shader, args, &priv_ctx);
3577     list_init(&priv_ctx.control_frames);
3578
3579     /* Avoid enabling NV_fragment_program* if we do not need it.
3580      *
3581      * Enabling GL_NV_fragment_program_option causes the driver to occupy a temporary register,
3582      * and it slows down the shader execution noticeably(about 5%). Usually our instruction emulation
3583      * is faster than what we gain from using higher native instructions. There are some things though
3584      * that cannot be emulated. In that case enable the extensions.
3585      * If the extension is enabled, instruction handlers that support both ways will use it.
3586      *
3587      * Testing shows no performance difference between OPTION NV_fragment_program2 and NV_fragment_program.
3588      * So enable the best we can get.
3589      */
3590     if(reg_maps->usesdsx || reg_maps->usesdsy || reg_maps->loop_depth > 0 || reg_maps->usestexldd ||
3591        reg_maps->usestexldl || reg_maps->usesfacing || reg_maps->usesifc || reg_maps->usescall)
3592     {
3593         want_nv_prog = TRUE;
3594     }
3595
3596     shader_addline(buffer, "!!ARBfp1.0\n");
3597     if (want_nv_prog && gl_info->supported[NV_FRAGMENT_PROGRAM2])
3598     {
3599         shader_addline(buffer, "OPTION NV_fragment_program2;\n");
3600         priv_ctx.target_version = NV3;
3601     }
3602     else if (want_nv_prog && gl_info->supported[NV_FRAGMENT_PROGRAM_OPTION])
3603     {
3604         shader_addline(buffer, "OPTION NV_fragment_program;\n");
3605         priv_ctx.target_version = NV2;
3606     } else {
3607         if(want_nv_prog)
3608         {
3609             /* This is an error - either we're advertising the wrong shader version, or aren't enforcing some
3610              * limits properly
3611              */
3612             ERR("The shader requires instructions that are not available in plain GL_ARB_fragment_program\n");
3613             ERR("Try GLSL\n");
3614         }
3615         priv_ctx.target_version = ARB;
3616     }
3617
3618     if (reg_maps->highest_render_target > 0)
3619     {
3620         shader_addline(buffer, "OPTION ARB_draw_buffers;\n");
3621     }
3622
3623     if (reg_maps->shader_version.major < 3)
3624     {
3625         switch(args->super.fog) {
3626             case FOG_OFF:
3627                 break;
3628             case FOG_LINEAR:
3629                 shader_addline(buffer, "OPTION ARB_fog_linear;\n");
3630                 break;
3631             case FOG_EXP:
3632                 shader_addline(buffer, "OPTION ARB_fog_exp;\n");
3633                 break;
3634             case FOG_EXP2:
3635                 shader_addline(buffer, "OPTION ARB_fog_exp2;\n");
3636                 break;
3637         }
3638     }
3639
3640     /* For now always declare the temps. At least the Nvidia assembler optimizes completely
3641      * unused temps away(but occupies them for the whole shader if they're used once). Always
3642      * declaring them avoids tricky bookkeeping work
3643      */
3644     shader_addline(buffer, "TEMP TA;\n");      /* Used for modifiers */
3645     shader_addline(buffer, "TEMP TB;\n");      /* Used for modifiers */
3646     shader_addline(buffer, "TEMP TC;\n");      /* Used for modifiers */
3647     if(dcl_td) shader_addline(buffer, "TEMP TD;\n"); /* Used for sRGB writing */
3648     shader_addline(buffer, "PARAM coefdiv = { 0.5, 0.25, 0.125, 0.0625 };\n");
3649     shader_addline(buffer, "PARAM coefmul = { 2, 4, 8, 16 };\n");
3650     shader_addline(buffer, "PARAM ps_helper_const = { 0.0, 1.0, %1.10f, 0.0 };\n", eps);
3651
3652     if (reg_maps->shader_version.major < 2)
3653     {
3654         strcpy(fragcolor, "R0");
3655     }
3656     else
3657     {
3658         if (args->super.srgb_correction)
3659         {
3660             if (shader->u.ps.color0_mov)
3661             {
3662                 sprintf(fragcolor, "R%u", shader->u.ps.color0_reg);
3663             }
3664             else
3665             {
3666                 shader_addline(buffer, "TEMP TMP_COLOR;\n");
3667                 strcpy(fragcolor, "TMP_COLOR");
3668             }
3669         } else {
3670             strcpy(fragcolor, "result.color");
3671         }
3672     }
3673
3674     if(args->super.srgb_correction) {
3675         shader_addline(buffer, "PARAM srgb_consts1 = {%f, %f, %f, %f};\n",
3676                        srgb_mul_low, srgb_cmp, srgb_pow, srgb_mul_high);
3677         shader_addline(buffer, "PARAM srgb_consts2 = {%f, %f, %f, %f};\n",
3678                        srgb_sub_high, 0.0, 0.0, 0.0);
3679     }
3680
3681     /* Base Declarations */
3682     next_local = shader_generate_arb_declarations(shader, reg_maps,
3683             buffer, gl_info, lconst_map, NULL, &priv_ctx);
3684
3685     for (i = 0, map = reg_maps->bumpmat; map; map >>= 1, ++i)
3686     {
3687         if (!(map & 1)) continue;
3688
3689         cur = compiled->numbumpenvmatconsts;
3690         compiled->bumpenvmatconst[cur].const_num = WINED3D_CONST_NUM_UNUSED;
3691         compiled->bumpenvmatconst[cur].texunit = i;
3692         compiled->luminanceconst[cur].const_num = WINED3D_CONST_NUM_UNUSED;
3693         compiled->luminanceconst[cur].texunit = i;
3694
3695         /* We can fit the constants into the constant limit for sure because texbem, texbeml, bem and beml are only supported
3696          * in 1.x shaders, and GL_ARB_fragment_program has a constant limit of 24 constants. So in the worst case we're loading
3697          * 8 shader constants, 8 bump matrices and 8 luminance parameters and are perfectly fine. (No NP2 fixup on bumpmapped
3698          * textures due to conditional NP2 restrictions)
3699          *
3700          * Use local constants to load the bump env parameters, not program.env. This avoids collisions with d3d constants of
3701          * shaders in newer shader models. Since the bump env parameters have to share their space with NP2 fixup constants,
3702          * their location is shader dependent anyway and they cannot be loaded globally.
3703          */
3704         compiled->bumpenvmatconst[cur].const_num = next_local++;
3705         shader_addline(buffer, "PARAM bumpenvmat%d = program.local[%d];\n",
3706                        i, compiled->bumpenvmatconst[cur].const_num);
3707         compiled->numbumpenvmatconsts = cur + 1;
3708
3709         if (!(reg_maps->luminanceparams & (1 << i))) continue;
3710
3711         compiled->luminanceconst[cur].const_num = next_local++;
3712         shader_addline(buffer, "PARAM luminance%d = program.local[%d];\n",
3713                        i, compiled->luminanceconst[cur].const_num);
3714     }
3715
3716     for(i = 0; i < MAX_CONST_I; i++)
3717     {
3718         compiled->int_consts[i] = WINED3D_CONST_NUM_UNUSED;
3719         if (reg_maps->integer_constants & (1 << i) && priv_ctx.target_version >= NV2)
3720         {
3721             const DWORD *control_values = find_loop_control_values(shader, i);
3722
3723             if(control_values)
3724             {
3725                 shader_addline(buffer, "PARAM I%u = {%u, %u, %u, -1};\n", i,
3726                                 control_values[0], control_values[1], control_values[2]);
3727             }
3728             else
3729             {
3730                 compiled->int_consts[i] = next_local;
3731                 compiled->num_int_consts++;
3732                 shader_addline(buffer, "PARAM I%u = program.local[%u];\n", i, next_local++);
3733             }
3734         }
3735     }
3736
3737     if(reg_maps->vpos || reg_maps->usesdsy)
3738     {
3739         compiled->ycorrection = next_local;
3740         shader_addline(buffer, "PARAM ycorrection = program.local[%u];\n", next_local++);
3741
3742         if(reg_maps->vpos)
3743         {
3744             shader_addline(buffer, "TEMP vpos;\n");
3745             /* ycorrection.x: Backbuffer height(onscreen) or 0(offscreen).
3746              * ycorrection.y: -1.0(onscreen), 1.0(offscreen)
3747              * ycorrection.z: 1.0
3748              * ycorrection.w: 0.0
3749              */
3750             shader_addline(buffer, "MAD vpos, fragment.position, ycorrection.zyww, ycorrection.wxww;\n");
3751             shader_addline(buffer, "FLR vpos.xy, vpos;\n");
3752         }
3753     }
3754     else
3755     {
3756         compiled->ycorrection = WINED3D_CONST_NUM_UNUSED;
3757     }
3758
3759     /* Load constants to fixup NP2 texcoords if there are still free constants left:
3760      * Constants (texture dimensions) for the NP2 fixup are loaded as local program parameters. This will consume
3761      * at most 8 (MAX_FRAGMENT_SAMPLERS / 2) parameters, which is highly unlikely, since the application had to
3762      * use 16 NP2 textures at the same time. In case that we run out of constants the fixup is simply not
3763      * applied / activated. This will probably result in wrong rendering of the texture, but will save us from
3764      * shader compilation errors and the subsequent errors when drawing with this shader. */
3765     if (priv_ctx.cur_ps_args->super.np2_fixup) {
3766
3767         struct arb_ps_np2fixup_info* const fixup = priv_ctx.cur_np2fixup_info;
3768         const WORD map = priv_ctx.cur_ps_args->super.np2_fixup;
3769         const UINT max_lconsts = gl_info->limits.arb_ps_local_constants;
3770
3771         fixup->offset = next_local;
3772         fixup->super.active = 0;
3773
3774         cur = 0;
3775         for (i = 0; i < MAX_FRAGMENT_SAMPLERS; ++i) {
3776             if (!(map & (1 << i))) continue;
3777
3778             if (fixup->offset + (cur >> 1) < max_lconsts) {
3779                 fixup->super.active |= (1 << i);
3780                 fixup->super.idx[i] = cur++;
3781             } else {
3782                 FIXME("No free constant found to load NP2 fixup data into shader. "
3783                       "Sampling from this texture will probably look wrong.\n");
3784                 break;
3785             }
3786         }
3787
3788         fixup->super.num_consts = (cur + 1) >> 1;
3789         if (fixup->super.num_consts) {
3790             shader_addline(buffer, "PARAM np2fixup[%u] = { program.env[%u..%u] };\n",
3791                            fixup->super.num_consts, fixup->offset, fixup->super.num_consts + fixup->offset - 1);
3792         }
3793
3794         next_local += fixup->super.num_consts;
3795     }
3796
3797     if (shader_priv->clipplane_emulation != ~0U && args->clip)
3798     {
3799         shader_addline(buffer, "KIL fragment.texcoord[%u];\n", shader_priv->clipplane_emulation);
3800     }
3801
3802     /* Base Shader Body */
3803     shader_generate_main(shader, buffer, reg_maps, function, &priv_ctx);
3804
3805     if(args->super.srgb_correction) {
3806         arbfp_add_sRGB_correction(buffer, fragcolor, srgbtmp[0], srgbtmp[1], srgbtmp[2], srgbtmp[3],
3807                                   priv_ctx.target_version >= NV2);
3808     }
3809
3810     if(strcmp(fragcolor, "result.color")) {
3811         shader_addline(buffer, "MOV result.color, %s;\n", fragcolor);
3812     }
3813     shader_addline(buffer, "END\n");
3814
3815     /* TODO: change to resource.glObjectHandle or something like that */
3816     GL_EXTCALL(glGenProgramsARB(1, &retval));
3817
3818     TRACE("Creating a hw pixel shader, prg=%d\n", retval);
3819     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, retval));
3820
3821     TRACE("Created hw pixel shader, prg=%d\n", retval);
3822     /* Create the program and check for errors */
3823     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
3824                buffer->bsize, buffer->buffer));
3825     checkGLcall("glProgramStringARB()");
3826
3827     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &errPos);
3828     if (errPos != -1)
3829     {
3830         FIXME("HW PixelShader Error at position %d: %s\n\n",
3831               errPos, debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
3832         shader_arb_dump_program_source(buffer->buffer);
3833         retval = 0;
3834     }
3835     else
3836     {
3837         GLint native;
3838
3839         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
3840         checkGLcall("glGetProgramivARB()");
3841         if (!native) WARN("Program exceeds native resource limits.\n");
3842     }
3843
3844     /* Load immediate constants */
3845     if (lconst_map)
3846     {
3847         LIST_FOR_EACH_ENTRY(lconst, &shader->constantsF, local_constant, entry)
3848         {
3849             const float *value = (const float *)lconst->value;
3850             GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, lconst_map[lconst->idx], value));
3851             checkGLcall("glProgramLocalParameter4fvARB");
3852         }
3853         HeapFree(GetProcessHeap(), 0, lconst_map);
3854     }
3855
3856     return retval;
3857 }
3858
3859 static int compare_sig(const struct wined3d_shader_signature_element *sig1, const struct wined3d_shader_signature_element *sig2)
3860 {
3861     unsigned int i;
3862     int ret;
3863
3864     for(i = 0; i < MAX_REG_INPUT; i++)
3865     {
3866         if (!sig1[i].semantic_name || !sig2[i].semantic_name)
3867         {
3868             /* Compare pointers, not contents. One string is NULL(element does not exist), the other one is not NULL */
3869             if(sig1[i].semantic_name != sig2[i].semantic_name) return sig1[i].semantic_name < sig2[i].semantic_name ? -1 : 1;
3870             continue;
3871         }
3872
3873         if ((ret = strcmp(sig1[i].semantic_name, sig2[i].semantic_name))) return ret;
3874         if(sig1[i].semantic_idx    != sig2[i].semantic_idx)    return sig1[i].semantic_idx    < sig2[i].semantic_idx    ? -1 : 1;
3875         if(sig1[i].sysval_semantic != sig2[i].sysval_semantic) return sig1[i].sysval_semantic < sig2[i].sysval_semantic ? -1 : 1;
3876         if(sig1[i].component_type  != sig2[i].component_type)  return sig1[i].sysval_semantic < sig2[i].component_type  ? -1 : 1;
3877         if(sig1[i].register_idx    != sig2[i].register_idx)    return sig1[i].register_idx    < sig2[i].register_idx    ? -1 : 1;
3878         if(sig1[i].mask            != sig2->mask)              return sig1[i].mask            < sig2[i].mask            ? -1 : 1;
3879     }
3880     return 0;
3881 }
3882
3883 static struct wined3d_shader_signature_element *clone_sig(const struct wined3d_shader_signature_element *sig)
3884 {
3885     struct wined3d_shader_signature_element *new;
3886     int i;
3887     char *name;
3888
3889     new = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*new) * MAX_REG_INPUT);
3890     for(i = 0; i < MAX_REG_INPUT; i++)
3891     {
3892         if (!sig[i].semantic_name) continue;
3893
3894         new[i] = sig[i];
3895         /* Clone the semantic string */
3896         name = HeapAlloc(GetProcessHeap(), 0, strlen(sig[i].semantic_name) + 1);
3897         strcpy(name, sig[i].semantic_name);
3898         new[i].semantic_name = name;
3899     }
3900     return new;
3901 }
3902
3903 static DWORD find_input_signature(struct shader_arb_priv *priv, const struct wined3d_shader_signature_element *sig)
3904 {
3905     struct wine_rb_entry *entry = wine_rb_get(&priv->signature_tree, sig);
3906     struct ps_signature *found_sig;
3907
3908     if (entry)
3909     {
3910         found_sig = WINE_RB_ENTRY_VALUE(entry, struct ps_signature, entry);
3911         TRACE("Found existing signature %u\n", found_sig->idx);
3912         return found_sig->idx;
3913     }
3914     found_sig = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*sig));
3915     found_sig->sig = clone_sig(sig);
3916     found_sig->idx = priv->ps_sig_number++;
3917     TRACE("New signature stored and assigned number %u\n", found_sig->idx);
3918     if(wine_rb_put(&priv->signature_tree, sig, &found_sig->entry) == -1)
3919     {
3920         ERR("Failed to insert program entry.\n");
3921     }
3922     return found_sig->idx;
3923 }
3924
3925 static void init_output_registers(struct wined3d_shader *shader, DWORD sig_num,
3926         struct shader_arb_ctx_priv *priv_ctx, struct arb_vs_compiled_shader *compiled)
3927 {
3928     unsigned int i, j;
3929     static const char * const texcoords[8] =
3930     {
3931         "result.texcoord[0]", "result.texcoord[1]", "result.texcoord[2]", "result.texcoord[3]",
3932         "result.texcoord[4]", "result.texcoord[5]", "result.texcoord[6]", "result.texcoord[7]"
3933     };
3934     IWineD3DDeviceImpl *device = shader->device;
3935     const struct wined3d_shader_signature_element *sig;
3936     const char *semantic_name;
3937     DWORD semantic_idx, reg_idx;
3938
3939     /* Write generic input varyings 0 to 7 to result.texcoord[], varying 8 to result.color.primary
3940      * and varying 9 to result.color.secondary
3941      */
3942     static const char * const decl_idx_to_string[MAX_REG_INPUT] =
3943     {
3944         "result.texcoord[0]", "result.texcoord[1]", "result.texcoord[2]", "result.texcoord[3]",
3945         "result.texcoord[4]", "result.texcoord[5]", "result.texcoord[6]", "result.texcoord[7]",
3946         "result.color.primary", "result.color.secondary"
3947     };
3948
3949     if(sig_num == ~0)
3950     {
3951         TRACE("Pixel shader uses builtin varyings\n");
3952         /* Map builtins to builtins */
3953         for(i = 0; i < 8; i++)
3954         {
3955             priv_ctx->texcrd_output[i] = texcoords[i];
3956         }
3957         priv_ctx->color_output[0] = "result.color.primary";
3958         priv_ctx->color_output[1] = "result.color.secondary";
3959         priv_ctx->fog_output = "result.fogcoord";
3960
3961         /* Map declared regs to builtins. Use "TA" to /dev/null unread output */
3962         for (i = 0; i < (sizeof(shader->output_signature) / sizeof(*shader->output_signature)); ++i)
3963         {
3964             semantic_name = shader->output_signature[i].semantic_name;
3965             if (!semantic_name) continue;
3966
3967             if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_POSITION))
3968             {
3969                 TRACE("o%u is TMP_OUT\n", i);
3970                 if (!shader->output_signature[i].semantic_idx) priv_ctx->vs_output[i] = "TMP_OUT";
3971                 else priv_ctx->vs_output[i] = "TA";
3972             }
3973             else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_PSIZE))
3974             {
3975                 TRACE("o%u is result.pointsize\n", i);
3976                 if (!shader->output_signature[i].semantic_idx) priv_ctx->vs_output[i] = "result.pointsize";
3977                 else priv_ctx->vs_output[i] = "TA";
3978             }
3979             else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_COLOR))
3980             {
3981                 TRACE("o%u is result.color.?, idx %u\n", i, shader->output_signature[i].semantic_idx);
3982                 if (!shader->output_signature[i].semantic_idx)
3983                     priv_ctx->vs_output[i] = "result.color.primary";
3984                 else if (shader->output_signature[i].semantic_idx == 1)
3985                     priv_ctx->vs_output[i] = "result.color.secondary";
3986                 else priv_ctx->vs_output[i] = "TA";
3987             }
3988             else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_TEXCOORD))
3989             {
3990                 TRACE("o%u is %s\n", i, texcoords[shader->output_signature[i].semantic_idx]);
3991                 if (shader->output_signature[i].semantic_idx >= 8) priv_ctx->vs_output[i] = "TA";
3992                 else priv_ctx->vs_output[i] = texcoords[shader->output_signature[i].semantic_idx];
3993             }
3994             else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_FOG))
3995             {
3996                 TRACE("o%u is result.fogcoord\n", i);
3997                 if (shader->output_signature[i].semantic_idx > 0) priv_ctx->vs_output[i] = "TA";
3998                 else priv_ctx->vs_output[i] = "result.fogcoord";
3999             }
4000             else
4001             {
4002                 priv_ctx->vs_output[i] = "TA";
4003             }
4004         }
4005         return;
4006     }
4007
4008     /* Instead of searching for the signature in the signature list, read the one from the current pixel shader.
4009      * Its maybe not the shader where the signature came from, but it is the same signature and faster to find
4010      */
4011     sig = device->stateBlock->state.pixel_shader->input_signature;
4012     TRACE("Pixel shader uses declared varyings\n");
4013
4014     /* Map builtin to declared. /dev/null the results by default to the TA temp reg */
4015     for(i = 0; i < 8; i++)
4016     {
4017         priv_ctx->texcrd_output[i] = "TA";
4018     }
4019     priv_ctx->color_output[0] = "TA";
4020     priv_ctx->color_output[1] = "TA";
4021     priv_ctx->fog_output = "TA";
4022
4023     for(i = 0; i < MAX_REG_INPUT; i++)
4024     {
4025         semantic_name = sig[i].semantic_name;
4026         semantic_idx = sig[i].semantic_idx;
4027         reg_idx = sig[i].register_idx;
4028         if (!semantic_name) continue;
4029
4030         /* If a declared input register is not written by builtin arguments, don't write to it.
4031          * GL_NV_vertex_program makes sure the input defaults to 0.0, which is correct with D3D
4032          *
4033          * Don't care about POSITION and PSIZE here - this is a builtin vertex shader, position goes
4034          * to TMP_OUT in any case
4035          */
4036         if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_TEXCOORD))
4037         {
4038             if(semantic_idx < 8) priv_ctx->texcrd_output[semantic_idx] = decl_idx_to_string[reg_idx];
4039         }
4040         else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_COLOR))
4041         {
4042             if(semantic_idx < 2) priv_ctx->color_output[semantic_idx] = decl_idx_to_string[reg_idx];
4043         }
4044         else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_FOG))
4045         {
4046             if (!semantic_idx) priv_ctx->fog_output = decl_idx_to_string[reg_idx];
4047         }
4048         else
4049         {
4050             continue;
4051         }
4052
4053         if (!strcmp(decl_idx_to_string[reg_idx], "result.color.primary")
4054                 || !strcmp(decl_idx_to_string[reg_idx], "result.color.secondary"))
4055         {
4056             compiled->need_color_unclamp = TRUE;
4057         }
4058     }
4059
4060     /* Map declared to declared */
4061     for (i = 0; i < (sizeof(shader->output_signature) / sizeof(*shader->output_signature)); ++i)
4062     {
4063         /* Write unread output to TA to throw them away */
4064         priv_ctx->vs_output[i] = "TA";
4065         semantic_name = shader->output_signature[i].semantic_name;
4066         if (!semantic_name) continue;
4067
4068         if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_POSITION)
4069                 && !shader->output_signature[i].semantic_idx)
4070         {
4071             priv_ctx->vs_output[i] = "TMP_OUT";
4072             continue;
4073         }
4074         else if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_PSIZE)
4075                 && !shader->output_signature[i].semantic_idx)
4076         {
4077             priv_ctx->vs_output[i] = "result.pointsize";
4078             continue;
4079         }
4080
4081         for(j = 0; j < MAX_REG_INPUT; j++)
4082         {
4083             if (!sig[j].semantic_name) continue;
4084
4085             if (!strcmp(sig[j].semantic_name, semantic_name)
4086                     && sig[j].semantic_idx == shader->output_signature[i].semantic_idx)
4087             {
4088                 priv_ctx->vs_output[i] = decl_idx_to_string[sig[j].register_idx];
4089
4090                 if (!strcmp(priv_ctx->vs_output[i], "result.color.primary")
4091                         || !strcmp(priv_ctx->vs_output[i], "result.color.secondary"))
4092                 {
4093                     compiled->need_color_unclamp = TRUE;
4094                 }
4095             }
4096         }
4097     }
4098 }
4099
4100 /* GL locking is done by the caller */
4101 static GLuint shader_arb_generate_vshader(struct wined3d_shader *shader,
4102         const struct wined3d_gl_info *gl_info, struct wined3d_shader_buffer *buffer,
4103         const struct arb_vs_compile_args *args, struct arb_vs_compiled_shader *compiled)
4104 {
4105     const struct arb_vshader_private *shader_data = shader->backend_data;
4106     const struct wined3d_shader_reg_maps *reg_maps = &shader->reg_maps;
4107     const DWORD *function = shader->function;
4108     const local_constant *lconst;
4109     GLuint ret;
4110     DWORD next_local, *lconst_map = local_const_mapping(shader);
4111     struct shader_arb_ctx_priv priv_ctx;
4112     unsigned int i;
4113     GLint errPos;
4114
4115     memset(&priv_ctx, 0, sizeof(priv_ctx));
4116     priv_ctx.cur_vs_args = args;
4117     list_init(&priv_ctx.control_frames);
4118     init_output_registers(shader, args->ps_signature, &priv_ctx, compiled);
4119
4120     /*  Create the hw ARB shader */
4121     shader_addline(buffer, "!!ARBvp1.0\n");
4122
4123     /* Always enable the NV extension if available. Unlike fragment shaders, there is no
4124      * mesurable performance penalty, and we can always make use of it for clipplanes.
4125      */
4126     if (gl_info->supported[NV_VERTEX_PROGRAM3])
4127     {
4128         shader_addline(buffer, "OPTION NV_vertex_program3;\n");
4129         priv_ctx.target_version = NV3;
4130         shader_addline(buffer, "ADDRESS aL;\n");
4131     }
4132     else if (gl_info->supported[NV_VERTEX_PROGRAM2_OPTION])
4133     {
4134         shader_addline(buffer, "OPTION NV_vertex_program2;\n");
4135         priv_ctx.target_version = NV2;
4136         shader_addline(buffer, "ADDRESS aL;\n");
4137     } else {
4138         priv_ctx.target_version = ARB;
4139     }
4140
4141     shader_addline(buffer, "TEMP TMP_OUT;\n");
4142     if (need_helper_const(shader_data, reg_maps, gl_info))
4143     {
4144         shader_addline(buffer, "PARAM helper_const = { 0.0, 1.0, 2.0, %1.10f};\n", eps);
4145     }
4146     if (need_rel_addr_const(shader_data, reg_maps, gl_info))
4147     {
4148         shader_addline(buffer, "PARAM rel_addr_const = { 0.5, %d.0, 0.0, 0.0 };\n", shader_data->rel_offset);
4149         shader_addline(buffer, "TEMP A0_SHADOW;\n");
4150     }
4151
4152     shader_addline(buffer, "TEMP TA;\n");
4153     shader_addline(buffer, "TEMP TB;\n");
4154
4155     /* Base Declarations */
4156     next_local = shader_generate_arb_declarations(shader, reg_maps, buffer,
4157             gl_info, lconst_map, &priv_ctx.vs_clipplanes, &priv_ctx);
4158
4159     for(i = 0; i < MAX_CONST_I; i++)
4160     {
4161         compiled->int_consts[i] = WINED3D_CONST_NUM_UNUSED;
4162         if(reg_maps->integer_constants & (1 << i) && priv_ctx.target_version >= NV2)
4163         {
4164             const DWORD *control_values = find_loop_control_values(shader, i);
4165
4166             if(control_values)
4167             {
4168                 shader_addline(buffer, "PARAM I%u = {%u, %u, %u, -1};\n", i,
4169                                 control_values[0], control_values[1], control_values[2]);
4170             }
4171             else
4172             {
4173                 compiled->int_consts[i] = next_local;
4174                 compiled->num_int_consts++;
4175                 shader_addline(buffer, "PARAM I%u = program.local[%u];\n", i, next_local++);
4176             }
4177         }
4178     }
4179
4180     /* We need a constant to fixup the final position */
4181     shader_addline(buffer, "PARAM posFixup = program.local[%u];\n", next_local);
4182     compiled->pos_fixup = next_local++;
4183
4184     /* Initialize output parameters. GL_ARB_vertex_program does not require special initialization values
4185      * for output parameters. D3D in theory does not do that either, but some applications depend on a
4186      * proper initialization of the secondary color, and programs using the fixed function pipeline without
4187      * a replacement shader depend on the texcoord.w being set properly.
4188      *
4189      * GL_NV_vertex_program defines that all output values are initialized to {0.0, 0.0, 0.0, 1.0}. This
4190      * assertion is in effect even when using GL_ARB_vertex_program without any NV specific additions. So
4191      * skip this if NV_vertex_program is supported. Otherwise, initialize the secondary color. For the tex-
4192      * coords, we have a flag in the opengl caps. Many cards do not require the texcoord being set, and
4193      * this can eat a number of instructions, so skip it unless this cap is set as well
4194      */
4195     if (!gl_info->supported[NV_VERTEX_PROGRAM])
4196     {
4197         IWineD3DDeviceImpl *device = shader->device;
4198         const char *color_init = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_0001);
4199         shader_addline(buffer, "MOV result.color.secondary, %s;\n", color_init);
4200
4201         if (gl_info->quirks & WINED3D_QUIRK_SET_TEXCOORD_W && !device->frag_pipe->ffp_proj_control)
4202         {
4203             int i;
4204             const char *one = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_ONE);
4205             for(i = 0; i < min(8, MAX_REG_TEXCRD); i++)
4206             {
4207                 if (reg_maps->texcoord_mask[i] && reg_maps->texcoord_mask[i] != WINED3DSP_WRITEMASK_ALL)
4208                     shader_addline(buffer, "MOV result.texcoord[%u].w, %s\n", i, one);
4209             }
4210         }
4211     }
4212
4213     /* The shader starts with the main function */
4214     priv_ctx.in_main_func = TRUE;
4215     /* Base Shader Body */
4216     shader_generate_main(shader, buffer, reg_maps, function, &priv_ctx);
4217
4218     if (!priv_ctx.footer_written) vshader_add_footer(&priv_ctx,
4219             shader_data, args, reg_maps, gl_info, buffer);
4220
4221     shader_addline(buffer, "END\n");
4222
4223     /* TODO: change to resource.glObjectHandle or something like that */
4224     GL_EXTCALL(glGenProgramsARB(1, &ret));
4225
4226     TRACE("Creating a hw vertex shader, prg=%d\n", ret);
4227     GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, ret));
4228
4229     TRACE("Created hw vertex shader, prg=%d\n", ret);
4230     /* Create the program and check for errors */
4231     GL_EXTCALL(glProgramStringARB(GL_VERTEX_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
4232                buffer->bsize, buffer->buffer));
4233     checkGLcall("glProgramStringARB()");
4234
4235     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &errPos);
4236     if (errPos != -1)
4237     {
4238         FIXME("HW VertexShader Error at position %d: %s\n\n",
4239               errPos, debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
4240         shader_arb_dump_program_source(buffer->buffer);
4241         ret = -1;
4242     }
4243     else
4244     {
4245         GLint native;
4246
4247         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
4248         checkGLcall("glGetProgramivARB()");
4249         if (!native) WARN("Program exceeds native resource limits.\n");
4250
4251         /* Load immediate constants */
4252         if (lconst_map)
4253         {
4254             LIST_FOR_EACH_ENTRY(lconst, &shader->constantsF, local_constant, entry)
4255             {
4256                 const float *value = (const float *)lconst->value;
4257                 GL_EXTCALL(glProgramLocalParameter4fvARB(GL_VERTEX_PROGRAM_ARB, lconst_map[lconst->idx], value));
4258             }
4259         }
4260     }
4261     HeapFree(GetProcessHeap(), 0, lconst_map);
4262
4263     return ret;
4264 }
4265
4266 /* GL locking is done by the caller */
4267 static struct arb_ps_compiled_shader *find_arb_pshader(struct wined3d_shader *shader,
4268         const struct arb_ps_compile_args *args)
4269 {
4270     IWineD3DDeviceImpl *device = shader->device;
4271     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4272     UINT i;
4273     DWORD new_size;
4274     struct arb_ps_compiled_shader *new_array;
4275     struct wined3d_shader_buffer buffer;
4276     struct arb_pshader_private *shader_data;
4277     GLuint ret;
4278
4279     if (!shader->backend_data)
4280     {
4281         struct shader_arb_priv *priv = device->shader_priv;
4282
4283         shader->backend_data = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data));
4284         shader_data = shader->backend_data;
4285         shader_data->clamp_consts = shader->reg_maps.shader_version.major == 1;
4286
4287         if (shader->reg_maps.shader_version.major < 3)
4288             shader_data->input_signature_idx = ~0;
4289         else
4290             shader_data->input_signature_idx = find_input_signature(priv, shader->input_signature);
4291
4292         shader_data->has_signature_idx = TRUE;
4293         TRACE("Shader got assigned input signature index %u\n", shader_data->input_signature_idx);
4294
4295         if (!device->vs_clipping)
4296             shader_data->clipplane_emulation = shader_find_free_input_register(&shader->reg_maps,
4297                     gl_info->limits.texture_stages - 1);
4298         else
4299             shader_data->clipplane_emulation = ~0U;
4300     }
4301     shader_data = shader->backend_data;
4302
4303     /* Usually we have very few GL shaders for each d3d shader(just 1 or maybe 2),
4304      * so a linear search is more performant than a hashmap or a binary search
4305      * (cache coherency etc)
4306      */
4307     for (i = 0; i < shader_data->num_gl_shaders; ++i)
4308     {
4309         if (!memcmp(&shader_data->gl_shaders[i].args, args, sizeof(*args)))
4310             return &shader_data->gl_shaders[i];
4311     }
4312
4313     TRACE("No matching GL shader found, compiling a new shader\n");
4314     if(shader_data->shader_array_size == shader_data->num_gl_shaders) {
4315         if (shader_data->num_gl_shaders)
4316         {
4317             new_size = shader_data->shader_array_size + max(1, shader_data->shader_array_size / 2);
4318             new_array = HeapReAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, shader_data->gl_shaders,
4319                                     new_size * sizeof(*shader_data->gl_shaders));
4320         } else {
4321             new_array = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data->gl_shaders));
4322             new_size = 1;
4323         }
4324
4325         if(!new_array) {
4326             ERR("Out of memory\n");
4327             return 0;
4328         }
4329         shader_data->gl_shaders = new_array;
4330         shader_data->shader_array_size = new_size;
4331     }
4332
4333     shader_data->gl_shaders[shader_data->num_gl_shaders].args = *args;
4334
4335     pixelshader_update_samplers(&shader->reg_maps, device->stateBlock->state.textures);
4336
4337     if (!shader_buffer_init(&buffer))
4338     {
4339         ERR("Failed to initialize shader buffer.\n");
4340         return 0;
4341     }
4342
4343     ret = shader_arb_generate_pshader(shader, gl_info, &buffer, args,
4344             &shader_data->gl_shaders[shader_data->num_gl_shaders]);
4345     shader_buffer_free(&buffer);
4346     shader_data->gl_shaders[shader_data->num_gl_shaders].prgId = ret;
4347
4348     return &shader_data->gl_shaders[shader_data->num_gl_shaders++];
4349 }
4350
4351 static inline BOOL vs_args_equal(const struct arb_vs_compile_args *stored, const struct arb_vs_compile_args *new,
4352                                  const DWORD use_map, BOOL skip_int) {
4353     if((stored->super.swizzle_map & use_map) != new->super.swizzle_map) return FALSE;
4354     if(stored->super.clip_enabled != new->super.clip_enabled) return FALSE;
4355     if(stored->super.fog_src != new->super.fog_src) return FALSE;
4356     if(stored->clip.boolclip_compare != new->clip.boolclip_compare) return FALSE;
4357     if(stored->ps_signature != new->ps_signature) return FALSE;
4358     if(stored->vertex.samplers_compare != new->vertex.samplers_compare) return FALSE;
4359     if(skip_int) return TRUE;
4360
4361     return !memcmp(stored->loop_ctrl, new->loop_ctrl, sizeof(stored->loop_ctrl));
4362 }
4363
4364 static struct arb_vs_compiled_shader *find_arb_vshader(struct wined3d_shader *shader,
4365         const struct arb_vs_compile_args *args)
4366 {
4367     IWineD3DDeviceImpl *device = shader->device;
4368     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4369     DWORD use_map = device->strided_streams.use_map;
4370     UINT i;
4371     DWORD new_size;
4372     struct arb_vs_compiled_shader *new_array;
4373     struct wined3d_shader_buffer buffer;
4374     struct arb_vshader_private *shader_data;
4375     GLuint ret;
4376
4377     if (!shader->backend_data)
4378     {
4379         const struct wined3d_shader_reg_maps *reg_maps = &shader->reg_maps;
4380
4381         shader->backend_data = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data));
4382         shader_data = shader->backend_data;
4383
4384         if ((gl_info->quirks & WINED3D_QUIRK_ARB_VS_OFFSET_LIMIT)
4385                 && reg_maps->min_rel_offset <= reg_maps->max_rel_offset)
4386         {
4387             if (reg_maps->max_rel_offset - reg_maps->min_rel_offset > 127)
4388             {
4389                 FIXME("The difference between the minimum and maximum relative offset is > 127.\n");
4390                 FIXME("Which this OpenGL implementation does not support. Try using GLSL.\n");
4391                 FIXME("Min: %u, Max: %u.\n", reg_maps->min_rel_offset, reg_maps->max_rel_offset);
4392             }
4393             else if (reg_maps->max_rel_offset - reg_maps->min_rel_offset > 63)
4394                 shader_data->rel_offset = reg_maps->min_rel_offset + 63;
4395             else if (reg_maps->max_rel_offset > 63)
4396                 shader_data->rel_offset = reg_maps->min_rel_offset;
4397         }
4398     }
4399     shader_data = shader->backend_data;
4400
4401     /* Usually we have very few GL shaders for each d3d shader(just 1 or maybe 2),
4402      * so a linear search is more performant than a hashmap or a binary search
4403      * (cache coherency etc)
4404      */
4405     for(i = 0; i < shader_data->num_gl_shaders; i++) {
4406         if (vs_args_equal(&shader_data->gl_shaders[i].args, args,
4407                 use_map, gl_info->supported[NV_VERTEX_PROGRAM2_OPTION]))
4408         {
4409             return &shader_data->gl_shaders[i];
4410         }
4411     }
4412
4413     TRACE("No matching GL shader found, compiling a new shader\n");
4414
4415     if(shader_data->shader_array_size == shader_data->num_gl_shaders) {
4416         if (shader_data->num_gl_shaders)
4417         {
4418             new_size = shader_data->shader_array_size + max(1, shader_data->shader_array_size / 2);
4419             new_array = HeapReAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, shader_data->gl_shaders,
4420                                     new_size * sizeof(*shader_data->gl_shaders));
4421         } else {
4422             new_array = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data->gl_shaders));
4423             new_size = 1;
4424         }
4425
4426         if(!new_array) {
4427             ERR("Out of memory\n");
4428             return 0;
4429         }
4430         shader_data->gl_shaders = new_array;
4431         shader_data->shader_array_size = new_size;
4432     }
4433
4434     shader_data->gl_shaders[shader_data->num_gl_shaders].args = *args;
4435
4436     if (!shader_buffer_init(&buffer))
4437     {
4438         ERR("Failed to initialize shader buffer.\n");
4439         return 0;
4440     }
4441
4442     ret = shader_arb_generate_vshader(shader, gl_info, &buffer, args,
4443             &shader_data->gl_shaders[shader_data->num_gl_shaders]);
4444     shader_buffer_free(&buffer);
4445     shader_data->gl_shaders[shader_data->num_gl_shaders].prgId = ret;
4446
4447     return &shader_data->gl_shaders[shader_data->num_gl_shaders++];
4448 }
4449
4450 static void find_arb_ps_compile_args(const struct wined3d_state *state,
4451         const struct wined3d_shader *shader, struct arb_ps_compile_args *args)
4452 {
4453     IWineD3DDeviceImpl *device = shader->device;
4454     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4455     int i;
4456     WORD int_skip;
4457
4458     find_ps_compile_args(state, shader, &args->super);
4459
4460     /* This forces all local boolean constants to 1 to make them stateblock independent */
4461     args->bools = shader->reg_maps.local_bool_consts;
4462
4463     for(i = 0; i < MAX_CONST_B; i++)
4464     {
4465         if (state->ps_consts_b[i])
4466             args->bools |= ( 1 << i);
4467     }
4468
4469     /* Only enable the clip plane emulation KIL if at least one clipplane is enabled. The KIL instruction
4470      * is quite expensive because it forces the driver to disable early Z discards. It is cheaper to
4471      * duplicate the shader than have a no-op KIL instruction in every shader
4472      */
4473     if (!device->vs_clipping && use_vs(state)
4474             && state->render_states[WINED3DRS_CLIPPING]
4475             && state->render_states[WINED3DRS_CLIPPLANEENABLE])
4476         args->clip = 1;
4477     else
4478         args->clip = 0;
4479
4480     /* Skip if unused or local, or supported natively */
4481     int_skip = ~shader->reg_maps.integer_constants | shader->reg_maps.local_int_consts;
4482     if (int_skip == 0xffff || gl_info->supported[NV_FRAGMENT_PROGRAM_OPTION])
4483     {
4484         memset(&args->loop_ctrl, 0, sizeof(args->loop_ctrl));
4485         return;
4486     }
4487
4488     for(i = 0; i < MAX_CONST_I; i++)
4489     {
4490         if(int_skip & (1 << i))
4491         {
4492             args->loop_ctrl[i][0] = 0;
4493             args->loop_ctrl[i][1] = 0;
4494             args->loop_ctrl[i][2] = 0;
4495         }
4496         else
4497         {
4498             args->loop_ctrl[i][0] = state->ps_consts_i[i * 4];
4499             args->loop_ctrl[i][1] = state->ps_consts_i[i * 4 + 1];
4500             args->loop_ctrl[i][2] = state->ps_consts_i[i * 4 + 2];
4501         }
4502     }
4503 }
4504
4505 static void find_arb_vs_compile_args(const struct wined3d_state *state,
4506         const struct wined3d_shader *shader, struct arb_vs_compile_args *args)
4507 {
4508     IWineD3DDeviceImpl *device = shader->device;
4509     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4510     int i;
4511     WORD int_skip;
4512
4513     find_vs_compile_args(state, shader, &args->super);
4514
4515     args->clip.boolclip_compare = 0;
4516     if (use_ps(state))
4517     {
4518         const struct wined3d_shader *ps = state->pixel_shader;
4519         const struct arb_pshader_private *shader_priv = ps->backend_data;
4520         args->ps_signature = shader_priv->input_signature_idx;
4521
4522         args->clip.boolclip.clip_texcoord = shader_priv->clipplane_emulation + 1;
4523     }
4524     else
4525     {
4526         args->ps_signature = ~0;
4527         if (!device->vs_clipping)
4528         {
4529             args->clip.boolclip.clip_texcoord = ffp_clip_emul(state) ? gl_info->limits.texture_stages : 0;
4530         }
4531         /* Otherwise: Setting boolclip_compare set clip_texcoord to 0 */
4532     }
4533
4534     if (args->clip.boolclip.clip_texcoord)
4535     {
4536         if (state->render_states[WINED3DRS_CLIPPING])
4537             args->clip.boolclip.clipplane_mask = (unsigned char)state->render_states[WINED3DRS_CLIPPLANEENABLE];
4538         /* clipplane_mask was set to 0 by setting boolclip_compare to 0 */
4539     }
4540
4541     /* This forces all local boolean constants to 1 to make them stateblock independent */
4542     args->clip.boolclip.bools = shader->reg_maps.local_bool_consts;
4543     /* TODO: Figure out if it would be better to store bool constants as bitmasks in the stateblock */
4544     for(i = 0; i < MAX_CONST_B; i++)
4545     {
4546         if (state->vs_consts_b[i])
4547             args->clip.boolclip.bools |= ( 1 << i);
4548     }
4549
4550     args->vertex.samplers[0] = device->texUnitMap[MAX_FRAGMENT_SAMPLERS + 0];
4551     args->vertex.samplers[1] = device->texUnitMap[MAX_FRAGMENT_SAMPLERS + 1];
4552     args->vertex.samplers[2] = device->texUnitMap[MAX_FRAGMENT_SAMPLERS + 2];
4553     args->vertex.samplers[3] = 0;
4554
4555     /* Skip if unused or local */
4556     int_skip = ~shader->reg_maps.integer_constants | shader->reg_maps.local_int_consts;
4557     /* This is about flow control, not clipping. */
4558     if (int_skip == 0xffff || gl_info->supported[NV_VERTEX_PROGRAM2_OPTION])
4559     {
4560         memset(&args->loop_ctrl, 0, sizeof(args->loop_ctrl));
4561         return;
4562     }
4563
4564     for(i = 0; i < MAX_CONST_I; i++)
4565     {
4566         if(int_skip & (1 << i))
4567         {
4568             args->loop_ctrl[i][0] = 0;
4569             args->loop_ctrl[i][1] = 0;
4570             args->loop_ctrl[i][2] = 0;
4571         }
4572         else
4573         {
4574             args->loop_ctrl[i][0] = state->vs_consts_i[i * 4];
4575             args->loop_ctrl[i][1] = state->vs_consts_i[i * 4 + 1];
4576             args->loop_ctrl[i][2] = state->vs_consts_i[i * 4 + 2];
4577         }
4578     }
4579 }
4580
4581 /* GL locking is done by the caller */
4582 static void shader_arb_select(const struct wined3d_context *context, BOOL usePS, BOOL useVS)
4583 {
4584     IWineD3DDeviceImpl *This = context->swapchain->device;
4585     struct shader_arb_priv *priv = This->shader_priv;
4586     const struct wined3d_gl_info *gl_info = context->gl_info;
4587     const struct wined3d_state *state = &This->stateBlock->state;
4588     int i;
4589
4590     /* Deal with pixel shaders first so the vertex shader arg function has the input signature ready */
4591     if (usePS)
4592     {
4593         struct wined3d_shader *ps = state->pixel_shader;
4594         struct arb_ps_compile_args compile_args;
4595         struct arb_ps_compiled_shader *compiled;
4596
4597         TRACE("Using pixel shader %p.\n", ps);
4598         find_arb_ps_compile_args(state, ps, &compile_args);
4599         compiled = find_arb_pshader(ps, &compile_args);
4600         priv->current_fprogram_id = compiled->prgId;
4601         priv->compiled_fprog = compiled;
4602
4603         /* Bind the fragment program */
4604         GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, priv->current_fprogram_id));
4605         checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, priv->current_fprogram_id);");
4606
4607         if(!priv->use_arbfp_fixed_func) {
4608             /* Enable OpenGL fragment programs */
4609             glEnable(GL_FRAGMENT_PROGRAM_ARB);
4610             checkGLcall("glEnable(GL_FRAGMENT_PROGRAM_ARB);");
4611         }
4612         TRACE("(%p) : Bound fragment program %u and enabled GL_FRAGMENT_PROGRAM_ARB\n", This, priv->current_fprogram_id);
4613
4614         /* Pixel Shader 1.x constants are clamped to [-1;1], Pixel Shader 2.0 constants are not. If switching between
4615          * a 1.x and newer shader, reload the first 8 constants
4616          */
4617         if (priv->last_ps_const_clamped != ((struct arb_pshader_private *)ps->backend_data)->clamp_consts)
4618         {
4619             priv->last_ps_const_clamped = ((struct arb_pshader_private *)ps->backend_data)->clamp_consts;
4620             This->highest_dirty_ps_const = max(This->highest_dirty_ps_const, 8);
4621             for(i = 0; i < 8; i++)
4622             {
4623                 context->pshader_const_dirty[i] = 1;
4624             }
4625             /* Also takes care of loading local constants */
4626             shader_arb_load_constants(context, TRUE, FALSE);
4627         }
4628         else
4629         {
4630             float rt_height = This->render_targets[0]->resource.height;
4631             shader_arb_ps_local_constants(compiled, context, state, rt_height);
4632         }
4633
4634         /* Force constant reloading for the NP2 fixup (see comment in shader_glsl_select for more info) */
4635         if (compiled->np2fixup_info.super.active)
4636             shader_arb_load_np2fixup_constants(priv, gl_info, state);
4637     }
4638     else if (gl_info->supported[ARB_FRAGMENT_PROGRAM] && !priv->use_arbfp_fixed_func)
4639     {
4640         /* Disable only if we're not using arbfp fixed function fragment processing. If this is used,
4641         * keep GL_FRAGMENT_PROGRAM_ARB enabled, and the fixed function pipeline will bind the fixed function
4642         * replacement shader
4643         */
4644         glDisable(GL_FRAGMENT_PROGRAM_ARB);
4645         checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
4646         priv->current_fprogram_id = 0;
4647     }
4648
4649     if (useVS)
4650     {
4651         struct wined3d_shader *vs = state->vertex_shader;
4652         struct arb_vs_compile_args compile_args;
4653         struct arb_vs_compiled_shader *compiled;
4654
4655         TRACE("Using vertex shader %p\n", vs);
4656         find_arb_vs_compile_args(state, vs, &compile_args);
4657         compiled = find_arb_vshader(vs, &compile_args);
4658         priv->current_vprogram_id = compiled->prgId;
4659         priv->compiled_vprog = compiled;
4660
4661         /* Bind the vertex program */
4662         GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->current_vprogram_id));
4663         checkGLcall("glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->current_vprogram_id);");
4664
4665         /* Enable OpenGL vertex programs */
4666         glEnable(GL_VERTEX_PROGRAM_ARB);
4667         checkGLcall("glEnable(GL_VERTEX_PROGRAM_ARB);");
4668         TRACE("(%p) : Bound vertex program %u and enabled GL_VERTEX_PROGRAM_ARB\n", This, priv->current_vprogram_id);
4669         shader_arb_vs_local_constants(compiled, context, state);
4670
4671         if(priv->last_vs_color_unclamp != compiled->need_color_unclamp) {
4672             priv->last_vs_color_unclamp = compiled->need_color_unclamp;
4673
4674             if (gl_info->supported[ARB_COLOR_BUFFER_FLOAT])
4675             {
4676                 GL_EXTCALL(glClampColorARB(GL_CLAMP_VERTEX_COLOR_ARB, !compiled->need_color_unclamp));
4677                 checkGLcall("glClampColorARB");
4678             } else {
4679                 FIXME("vertex color clamp needs to be changed, but extension not supported.\n");
4680             }
4681         }
4682     }
4683     else if (gl_info->supported[ARB_VERTEX_PROGRAM])
4684     {
4685         priv->current_vprogram_id = 0;
4686         glDisable(GL_VERTEX_PROGRAM_ARB);
4687         checkGLcall("glDisable(GL_VERTEX_PROGRAM_ARB)");
4688     }
4689 }
4690
4691 /* GL locking is done by the caller */
4692 static void shader_arb_select_depth_blt(void *shader_priv, const struct wined3d_gl_info *gl_info,
4693         enum tex_types tex_type, const SIZE *ds_mask_size)
4694 {
4695     const float mask[] = {0.0f, 0.0f, (float)ds_mask_size->cx, (float)ds_mask_size->cy};
4696     BOOL masked = ds_mask_size->cx && ds_mask_size->cy;
4697     struct shader_arb_priv *priv = shader_priv;
4698     GLuint *blt_fprogram;
4699
4700     if (!priv->depth_blt_vprogram_id) priv->depth_blt_vprogram_id = create_arb_blt_vertex_program(gl_info);
4701     GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->depth_blt_vprogram_id));
4702     glEnable(GL_VERTEX_PROGRAM_ARB);
4703
4704     blt_fprogram = masked ? &priv->depth_blt_fprogram_id_masked[tex_type] : &priv->depth_blt_fprogram_id_full[tex_type];
4705     if (!*blt_fprogram) *blt_fprogram = create_arb_blt_fragment_program(gl_info, tex_type, masked);
4706     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, *blt_fprogram));
4707     if (masked) GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, 0, mask));
4708     glEnable(GL_FRAGMENT_PROGRAM_ARB);
4709 }
4710
4711 /* GL locking is done by the caller */
4712 static void shader_arb_deselect_depth_blt(void *shader_priv, const struct wined3d_gl_info *gl_info)
4713 {
4714     struct shader_arb_priv *priv = shader_priv;
4715
4716     if (priv->current_vprogram_id) {
4717         GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->current_vprogram_id));
4718         checkGLcall("glBindProgramARB(GL_VERTEX_PROGRAM_ARB, vertexShader->prgId);");
4719
4720         TRACE("Bound vertex program %u and enabled GL_VERTEX_PROGRAM_ARB.\n", priv->current_vprogram_id);
4721     }
4722     else
4723     {
4724         glDisable(GL_VERTEX_PROGRAM_ARB);
4725         checkGLcall("glDisable(GL_VERTEX_PROGRAM_ARB)");
4726     }
4727
4728     if (priv->current_fprogram_id) {
4729         GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, priv->current_fprogram_id));
4730         checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, pixelShader->prgId);");
4731
4732         TRACE("Bound fragment program %u and enabled GL_FRAGMENT_PROGRAM_ARB.\n", priv->current_fprogram_id);
4733     }
4734     else if(!priv->use_arbfp_fixed_func)
4735     {
4736         glDisable(GL_FRAGMENT_PROGRAM_ARB);
4737         checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
4738     }
4739 }
4740
4741 static void shader_arb_destroy(struct wined3d_shader *shader)
4742 {
4743     IWineD3DDeviceImpl *device = shader->device;
4744     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4745
4746     if (shader_is_pshader_version(shader->reg_maps.shader_version.type))
4747     {
4748         struct arb_pshader_private *shader_data = shader->backend_data;
4749         UINT i;
4750
4751         if(!shader_data) return; /* This can happen if a shader was never compiled */
4752
4753         if (shader_data->num_gl_shaders)
4754         {
4755             struct wined3d_context *context = context_acquire(device, NULL);
4756
4757             ENTER_GL();
4758             for (i = 0; i < shader_data->num_gl_shaders; ++i)
4759             {
4760                 GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId));
4761                 checkGLcall("GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId))");
4762             }
4763             LEAVE_GL();
4764
4765             context_release(context);
4766         }
4767
4768         HeapFree(GetProcessHeap(), 0, shader_data->gl_shaders);
4769         HeapFree(GetProcessHeap(), 0, shader_data);
4770         shader->backend_data = NULL;
4771     }
4772     else
4773     {
4774         struct arb_vshader_private *shader_data = shader->backend_data;
4775         UINT i;
4776
4777         if(!shader_data) return; /* This can happen if a shader was never compiled */
4778
4779         if (shader_data->num_gl_shaders)
4780         {
4781             struct wined3d_context *context = context_acquire(device, NULL);
4782
4783             ENTER_GL();
4784             for (i = 0; i < shader_data->num_gl_shaders; ++i)
4785             {
4786                 GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId));
4787                 checkGLcall("GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId))");
4788             }
4789             LEAVE_GL();
4790
4791             context_release(context);
4792         }
4793
4794         HeapFree(GetProcessHeap(), 0, shader_data->gl_shaders);
4795         HeapFree(GetProcessHeap(), 0, shader_data);
4796         shader->backend_data = NULL;
4797     }
4798 }
4799
4800 static int sig_tree_compare(const void *key, const struct wine_rb_entry *entry)
4801 {
4802     struct ps_signature *e = WINE_RB_ENTRY_VALUE(entry, struct ps_signature, entry);
4803     return compare_sig(key, e->sig);
4804 }
4805
4806 static const struct wine_rb_functions sig_tree_functions =
4807 {
4808     wined3d_rb_alloc,
4809     wined3d_rb_realloc,
4810     wined3d_rb_free,
4811     sig_tree_compare
4812 };
4813
4814 static HRESULT shader_arb_alloc(IWineD3DDeviceImpl *device)
4815 {
4816     struct shader_arb_priv *priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*priv));
4817     if(wine_rb_init(&priv->signature_tree, &sig_tree_functions) == -1)
4818     {
4819         ERR("RB tree init failed\n");
4820         HeapFree(GetProcessHeap(), 0, priv);
4821         return E_OUTOFMEMORY;
4822     }
4823     device->shader_priv = priv;
4824     return WINED3D_OK;
4825 }
4826
4827 static void release_signature(struct wine_rb_entry *entry, void *context)
4828 {
4829     struct ps_signature *sig = WINE_RB_ENTRY_VALUE(entry, struct ps_signature, entry);
4830     int i;
4831     for(i = 0; i < MAX_REG_INPUT; i++)
4832     {
4833         HeapFree(GetProcessHeap(), 0, (char *) sig->sig[i].semantic_name);
4834     }
4835     HeapFree(GetProcessHeap(), 0, sig->sig);
4836     HeapFree(GetProcessHeap(), 0, sig);
4837 }
4838
4839 /* Context activation is done by the caller. */
4840 static void shader_arb_free(IWineD3DDeviceImpl *device)
4841 {
4842     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4843     struct shader_arb_priv *priv = device->shader_priv;
4844     int i;
4845
4846     ENTER_GL();
4847     if(priv->depth_blt_vprogram_id) {
4848         GL_EXTCALL(glDeleteProgramsARB(1, &priv->depth_blt_vprogram_id));
4849     }
4850     for (i = 0; i < tex_type_count; ++i)
4851     {
4852         if (priv->depth_blt_fprogram_id_full[i])
4853         {
4854             GL_EXTCALL(glDeleteProgramsARB(1, &priv->depth_blt_fprogram_id_full[i]));
4855         }
4856         if (priv->depth_blt_fprogram_id_masked[i])
4857         {
4858             GL_EXTCALL(glDeleteProgramsARB(1, &priv->depth_blt_fprogram_id_masked[i]));
4859         }
4860     }
4861     LEAVE_GL();
4862
4863     wine_rb_destroy(&priv->signature_tree, release_signature, NULL);
4864     HeapFree(GetProcessHeap(), 0, device->shader_priv);
4865 }
4866
4867 static BOOL shader_arb_dirty_const(void)
4868 {
4869     return TRUE;
4870 }
4871
4872 static void shader_arb_get_caps(const struct wined3d_gl_info *gl_info, struct shader_caps *caps)
4873 {
4874     if (gl_info->supported[ARB_VERTEX_PROGRAM])
4875     {
4876         DWORD vs_consts;
4877
4878         /* 96 is the minimum allowed value of MAX_PROGRAM_ENV_PARAMETERS_ARB
4879          * for vertex programs. If the native limit is less than that it's
4880          * not very useful, and e.g. Mesa swrast returns 0, probably to
4881          * indicate it's a software implementation. */
4882         if (gl_info->limits.arb_vs_native_constants < 96)
4883             vs_consts = gl_info->limits.arb_vs_float_constants;
4884         else
4885             vs_consts = min(gl_info->limits.arb_vs_float_constants, gl_info->limits.arb_vs_native_constants);
4886
4887         if (gl_info->supported[NV_VERTEX_PROGRAM3])
4888         {
4889             caps->VertexShaderVersion = WINED3DVS_VERSION(3,0);
4890             TRACE_(d3d_caps)("Hardware vertex shader version 3.0 enabled (NV_VERTEX_PROGRAM3)\n");
4891         }
4892         else if (vs_consts >= 256)
4893         {
4894             /* Shader Model 2.0 requires at least 256 vertex shader constants */
4895             caps->VertexShaderVersion = WINED3DVS_VERSION(2,0);
4896             TRACE_(d3d_caps)("Hardware vertex shader version 2.0 enabled (ARB_PROGRAM)\n");
4897         }
4898         else
4899         {
4900             caps->VertexShaderVersion = WINED3DVS_VERSION(1,1);
4901             TRACE_(d3d_caps)("Hardware vertex shader version 1.1 enabled (ARB_PROGRAM)\n");
4902         }
4903         caps->MaxVertexShaderConst = vs_consts;
4904     }
4905     else
4906     {
4907         caps->VertexShaderVersion = 0;
4908         caps->MaxVertexShaderConst = 0;
4909     }
4910
4911     if (gl_info->supported[ARB_FRAGMENT_PROGRAM])
4912     {
4913         DWORD ps_consts;
4914
4915         /* Similar as above for vertex programs, but the minimum for fragment
4916          * programs is 24. */
4917         if (gl_info->limits.arb_ps_native_constants < 24)
4918             ps_consts = gl_info->limits.arb_ps_float_constants;
4919         else
4920             ps_consts = min(gl_info->limits.arb_ps_float_constants, gl_info->limits.arb_ps_native_constants);
4921
4922         if (gl_info->supported[NV_FRAGMENT_PROGRAM2])
4923         {
4924             caps->PixelShaderVersion = WINED3DPS_VERSION(3,0);
4925             TRACE_(d3d_caps)("Hardware pixel shader version 3.0 enabled (NV_FRAGMENT_PROGRAM2)\n");
4926         }
4927         else if (ps_consts >= 32)
4928         {
4929             /* Shader Model 2.0 requires at least 32 pixel shader constants */
4930             caps->PixelShaderVersion = WINED3DPS_VERSION(2,0);
4931             TRACE_(d3d_caps)("Hardware pixel shader version 2.0 enabled (ARB_PROGRAM)\n");
4932         }
4933         else
4934         {
4935             caps->PixelShaderVersion = WINED3DPS_VERSION(1,4);
4936             TRACE_(d3d_caps)("Hardware pixel shader version 1.4 enabled (ARB_PROGRAM)\n");
4937         }
4938         caps->PixelShader1xMaxValue = 8.0f;
4939         caps->MaxPixelShaderConst = ps_consts;
4940     }
4941     else
4942     {
4943         caps->PixelShaderVersion = 0;
4944         caps->PixelShader1xMaxValue = 0.0f;
4945         caps->MaxPixelShaderConst = 0;
4946     }
4947
4948     caps->VSClipping = use_nv_clip(gl_info);
4949 }
4950
4951 static BOOL shader_arb_color_fixup_supported(struct color_fixup_desc fixup)
4952 {
4953     if (TRACE_ON(d3d_shader) && TRACE_ON(d3d))
4954     {
4955         TRACE("Checking support for color_fixup:\n");
4956         dump_color_fixup_desc(fixup);
4957     }
4958
4959     /* We support everything except complex conversions. */
4960     if (!is_complex_fixup(fixup))
4961     {
4962         TRACE("[OK]\n");
4963         return TRUE;
4964     }
4965
4966     TRACE("[FAILED]\n");
4967     return FALSE;
4968 }
4969
4970 static void shader_arb_add_instruction_modifiers(const struct wined3d_shader_instruction *ins) {
4971     DWORD shift;
4972     char write_mask[20], regstr[50];
4973     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
4974     BOOL is_color = FALSE;
4975     const struct wined3d_shader_dst_param *dst;
4976
4977     if (!ins->dst_count) return;
4978
4979     dst = &ins->dst[0];
4980     shift = dst->shift;
4981     if (!shift) return; /* Saturate alone is handled by the instructions */
4982
4983     shader_arb_get_write_mask(ins, dst, write_mask);
4984     shader_arb_get_register_name(ins, &dst->reg, regstr, &is_color);
4985
4986     /* Generate a line that does the output modifier computation
4987      * FIXME: _SAT vs shift? _SAT alone is already handled in the instructions, if this
4988      * maps problems in e.g. _d4_sat modify shader_arb_get_modifier
4989      */
4990     shader_addline(buffer, "MUL%s %s%s, %s, %s;\n", shader_arb_get_modifier(ins),
4991                    regstr, write_mask, regstr, shift_tab[shift]);
4992 }
4993
4994 static const SHADER_HANDLER shader_arb_instruction_handler_table[WINED3DSIH_TABLE_SIZE] =
4995 {
4996     /* WINED3DSIH_ABS           */ shader_hw_map2gl,
4997     /* WINED3DSIH_ADD           */ shader_hw_map2gl,
4998     /* WINED3DSIH_AND           */ NULL,
4999     /* WINED3DSIH_BEM           */ pshader_hw_bem,
5000     /* WINED3DSIH_BREAK         */ shader_hw_break,
5001     /* WINED3DSIH_BREAKC        */ shader_hw_breakc,
5002     /* WINED3DSIH_BREAKP        */ NULL,
5003     /* WINED3DSIH_CALL          */ shader_hw_call,
5004     /* WINED3DSIH_CALLNZ        */ NULL,
5005     /* WINED3DSIH_CMP           */ pshader_hw_cmp,
5006     /* WINED3DSIH_CND           */ pshader_hw_cnd,
5007     /* WINED3DSIH_CRS           */ shader_hw_map2gl,
5008     /* WINED3DSIH_CUT           */ NULL,
5009     /* WINED3DSIH_DCL           */ NULL,
5010     /* WINED3DSIH_DEF           */ NULL,
5011     /* WINED3DSIH_DEFB          */ NULL,
5012     /* WINED3DSIH_DEFI          */ NULL,
5013     /* WINED3DSIH_DIV           */ NULL,
5014     /* WINED3DSIH_DP2ADD        */ pshader_hw_dp2add,
5015     /* WINED3DSIH_DP3           */ shader_hw_map2gl,
5016     /* WINED3DSIH_DP4           */ shader_hw_map2gl,
5017     /* WINED3DSIH_DST           */ shader_hw_map2gl,
5018     /* WINED3DSIH_DSX           */ shader_hw_map2gl,
5019     /* WINED3DSIH_DSY           */ shader_hw_dsy,
5020     /* WINED3DSIH_ELSE          */ shader_hw_else,
5021     /* WINED3DSIH_EMIT          */ NULL,
5022     /* WINED3DSIH_ENDIF         */ shader_hw_endif,
5023     /* WINED3DSIH_ENDLOOP       */ shader_hw_endloop,
5024     /* WINED3DSIH_ENDREP        */ shader_hw_endrep,
5025     /* WINED3DSIH_EXP           */ shader_hw_scalar_op,
5026     /* WINED3DSIH_EXPP          */ shader_hw_scalar_op,
5027     /* WINED3DSIH_FRC           */ shader_hw_map2gl,
5028     /* WINED3DSIH_FTOI          */ NULL,
5029     /* WINED3DSIH_IADD          */ NULL,
5030     /* WINED3DSIH_IEQ           */ NULL,
5031     /* WINED3DSIH_IF            */ NULL /* Hardcoded into the shader */,
5032     /* WINED3DSIH_IFC           */ shader_hw_ifc,
5033     /* WINED3DSIH_IGE           */ NULL,
5034     /* WINED3DSIH_IMUL          */ NULL,
5035     /* WINED3DSIH_ITOF          */ NULL,
5036     /* WINED3DSIH_LABEL         */ shader_hw_label,
5037     /* WINED3DSIH_LD            */ NULL,
5038     /* WINED3DSIH_LIT           */ shader_hw_map2gl,
5039     /* WINED3DSIH_LOG           */ shader_hw_log,
5040     /* WINED3DSIH_LOGP          */ shader_hw_log,
5041     /* WINED3DSIH_LOOP          */ shader_hw_loop,
5042     /* WINED3DSIH_LRP           */ shader_hw_lrp,
5043     /* WINED3DSIH_LT            */ NULL,
5044     /* WINED3DSIH_M3x2          */ shader_hw_mnxn,
5045     /* WINED3DSIH_M3x3          */ shader_hw_mnxn,
5046     /* WINED3DSIH_M3x4          */ shader_hw_mnxn,
5047     /* WINED3DSIH_M4x3          */ shader_hw_mnxn,
5048     /* WINED3DSIH_M4x4          */ shader_hw_mnxn,
5049     /* WINED3DSIH_MAD           */ shader_hw_map2gl,
5050     /* WINED3DSIH_MAX           */ shader_hw_map2gl,
5051     /* WINED3DSIH_MIN           */ shader_hw_map2gl,
5052     /* WINED3DSIH_MOV           */ shader_hw_mov,
5053     /* WINED3DSIH_MOVA          */ shader_hw_mov,
5054     /* WINED3DSIH_MOVC          */ NULL,
5055     /* WINED3DSIH_MUL           */ shader_hw_map2gl,
5056     /* WINED3DSIH_NOP           */ shader_hw_nop,
5057     /* WINED3DSIH_NRM           */ shader_hw_nrm,
5058     /* WINED3DSIH_PHASE         */ NULL,
5059     /* WINED3DSIH_POW           */ shader_hw_pow,
5060     /* WINED3DSIH_RCP           */ shader_hw_rcp,
5061     /* WINED3DSIH_REP           */ shader_hw_rep,
5062     /* WINED3DSIH_RET           */ shader_hw_ret,
5063     /* WINED3DSIH_RSQ           */ shader_hw_scalar_op,
5064     /* WINED3DSIH_SAMPLE        */ NULL,
5065     /* WINED3DSIH_SAMPLE_GRAD   */ NULL,
5066     /* WINED3DSIH_SAMPLE_LOD    */ NULL,
5067     /* WINED3DSIH_SETP          */ NULL,
5068     /* WINED3DSIH_SGE           */ shader_hw_map2gl,
5069     /* WINED3DSIH_SGN           */ shader_hw_sgn,
5070     /* WINED3DSIH_SINCOS        */ shader_hw_sincos,
5071     /* WINED3DSIH_SLT           */ shader_hw_map2gl,
5072     /* WINED3DSIH_SQRT          */ NULL,
5073     /* WINED3DSIH_SUB           */ shader_hw_map2gl,
5074     /* WINED3DSIH_TEX           */ pshader_hw_tex,
5075     /* WINED3DSIH_TEXBEM        */ pshader_hw_texbem,
5076     /* WINED3DSIH_TEXBEML       */ pshader_hw_texbem,
5077     /* WINED3DSIH_TEXCOORD      */ pshader_hw_texcoord,
5078     /* WINED3DSIH_TEXDEPTH      */ pshader_hw_texdepth,
5079     /* WINED3DSIH_TEXDP3        */ pshader_hw_texdp3,
5080     /* WINED3DSIH_TEXDP3TEX     */ pshader_hw_texdp3tex,
5081     /* WINED3DSIH_TEXKILL       */ pshader_hw_texkill,
5082     /* WINED3DSIH_TEXLDD        */ shader_hw_texldd,
5083     /* WINED3DSIH_TEXLDL        */ shader_hw_texldl,
5084     /* WINED3DSIH_TEXM3x2DEPTH  */ pshader_hw_texm3x2depth,
5085     /* WINED3DSIH_TEXM3x2PAD    */ pshader_hw_texm3x2pad,
5086     /* WINED3DSIH_TEXM3x2TEX    */ pshader_hw_texm3x2tex,
5087     /* WINED3DSIH_TEXM3x3       */ pshader_hw_texm3x3,
5088     /* WINED3DSIH_TEXM3x3DIFF   */ NULL,
5089     /* WINED3DSIH_TEXM3x3PAD    */ pshader_hw_texm3x3pad,
5090     /* WINED3DSIH_TEXM3x3SPEC   */ pshader_hw_texm3x3spec,
5091     /* WINED3DSIH_TEXM3x3TEX    */ pshader_hw_texm3x3tex,
5092     /* WINED3DSIH_TEXM3x3VSPEC  */ pshader_hw_texm3x3vspec,
5093     /* WINED3DSIH_TEXREG2AR     */ pshader_hw_texreg2ar,
5094     /* WINED3DSIH_TEXREG2GB     */ pshader_hw_texreg2gb,
5095     /* WINED3DSIH_TEXREG2RGB    */ pshader_hw_texreg2rgb,
5096     /* WINED3DSIH_UTOF          */ NULL,
5097 };
5098
5099 static BOOL get_bool_const(const struct wined3d_shader_instruction *ins,
5100         const struct wined3d_shader *shader, DWORD idx)
5101 {
5102     const struct wined3d_shader_reg_maps *reg_maps = ins->ctx->reg_maps;
5103     BOOL vshader = shader_is_vshader_version(reg_maps->shader_version.type);
5104     WORD bools = 0;
5105     WORD flag = (1 << idx);
5106     const local_constant *constant;
5107     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
5108
5109     if (reg_maps->local_bool_consts & flag)
5110     {
5111         /* What good is a if(bool) with a hardcoded local constant? I don't know, but handle it */
5112         LIST_FOR_EACH_ENTRY(constant, &shader->constantsB, local_constant, entry)
5113         {
5114             if (constant->idx == idx)
5115             {
5116                 return constant->value[0];
5117             }
5118         }
5119         ERR("Local constant not found\n");
5120         return FALSE;
5121     }
5122     else
5123     {
5124         if(vshader) bools = priv->cur_vs_args->clip.boolclip.bools;
5125         else bools = priv->cur_ps_args->bools;
5126         return bools & flag;
5127     }
5128 }
5129
5130 static void get_loop_control_const(const struct wined3d_shader_instruction *ins,
5131         const struct wined3d_shader *shader, UINT idx, struct wined3d_shader_loop_control *loop_control)
5132 {
5133     const struct wined3d_shader_reg_maps *reg_maps = ins->ctx->reg_maps;
5134     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
5135
5136     /* Integer constants can either be a local constant, or they can be stored in the shader
5137      * type specific compile args. */
5138     if (reg_maps->local_int_consts & (1 << idx))
5139     {
5140         const local_constant *constant;
5141
5142         LIST_FOR_EACH_ENTRY(constant, &shader->constantsI, local_constant, entry)
5143         {
5144             if (constant->idx == idx)
5145             {
5146                 loop_control->count = constant->value[0];
5147                 loop_control->start = constant->value[1];
5148                 /* Step is signed. */
5149                 loop_control->step = (int)constant->value[2];
5150                 return;
5151             }
5152         }
5153         /* If this happens the flag was set incorrectly */
5154         ERR("Local constant not found\n");
5155         loop_control->count = 0;
5156         loop_control->start = 0;
5157         loop_control->step = 0;
5158         return;
5159     }
5160
5161     switch (reg_maps->shader_version.type)
5162     {
5163         case WINED3D_SHADER_TYPE_VERTEX:
5164             /* Count and aL start value are unsigned */
5165             loop_control->count = priv->cur_vs_args->loop_ctrl[idx][0];
5166             loop_control->start = priv->cur_vs_args->loop_ctrl[idx][1];
5167             /* Step is signed. */
5168             loop_control->step = ((char)priv->cur_vs_args->loop_ctrl[idx][2]);
5169             break;
5170
5171         case WINED3D_SHADER_TYPE_PIXEL:
5172             loop_control->count = priv->cur_ps_args->loop_ctrl[idx][0];
5173             loop_control->start = priv->cur_ps_args->loop_ctrl[idx][1];
5174             loop_control->step = ((char)priv->cur_ps_args->loop_ctrl[idx][2]);
5175             break;
5176
5177         default:
5178             FIXME("Unhandled shader type %#x.\n", reg_maps->shader_version.type);
5179             break;
5180     }
5181 }
5182
5183 static void record_instruction(struct list *list, const struct wined3d_shader_instruction *ins)
5184 {
5185     unsigned int i;
5186     struct wined3d_shader_dst_param *dst_param = NULL;
5187     struct wined3d_shader_src_param *src_param = NULL, *rel_addr = NULL;
5188     struct recorded_instruction *rec = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*rec));
5189     if(!rec)
5190     {
5191         ERR("Out of memory\n");
5192         return;
5193     }
5194
5195     rec->ins = *ins;
5196     dst_param = HeapAlloc(GetProcessHeap(), 0, sizeof(*dst_param));
5197     if(!dst_param) goto free;
5198     *dst_param = *ins->dst;
5199     if(ins->dst->reg.rel_addr)
5200     {
5201         rel_addr = HeapAlloc(GetProcessHeap(), 0, sizeof(*dst_param->reg.rel_addr));
5202         if(!rel_addr) goto free;
5203         *rel_addr = *ins->dst->reg.rel_addr;
5204         dst_param->reg.rel_addr = rel_addr;
5205     }
5206     rec->ins.dst = dst_param;
5207
5208     src_param = HeapAlloc(GetProcessHeap(), 0, sizeof(*src_param) * ins->src_count);
5209     if(!src_param) goto free;
5210     for(i = 0; i < ins->src_count; i++)
5211     {
5212         src_param[i] = ins->src[i];
5213         if(ins->src[i].reg.rel_addr)
5214         {
5215             rel_addr = HeapAlloc(GetProcessHeap(), 0, sizeof(*rel_addr));
5216             if(!rel_addr) goto free;
5217             *rel_addr = *ins->src[i].reg.rel_addr;
5218             src_param[i].reg.rel_addr = rel_addr;
5219         }
5220     }
5221     rec->ins.src = src_param;
5222     list_add_tail(list, &rec->entry);
5223     return;
5224
5225 free:
5226     ERR("Out of memory\n");
5227     if(dst_param)
5228     {
5229         HeapFree(GetProcessHeap(), 0, (void *) dst_param->reg.rel_addr);
5230         HeapFree(GetProcessHeap(), 0, dst_param);
5231     }
5232     if(src_param)
5233     {
5234         for(i = 0; i < ins->src_count; i++)
5235         {
5236             HeapFree(GetProcessHeap(), 0, (void *) src_param[i].reg.rel_addr);
5237         }
5238         HeapFree(GetProcessHeap(), 0, src_param);
5239     }
5240     HeapFree(GetProcessHeap(), 0, rec);
5241 }
5242
5243 static void free_recorded_instruction(struct list *list)
5244 {
5245     struct recorded_instruction *rec_ins, *entry2;
5246     unsigned int i;
5247
5248     LIST_FOR_EACH_ENTRY_SAFE(rec_ins, entry2, list, struct recorded_instruction, entry)
5249     {
5250         list_remove(&rec_ins->entry);
5251         if(rec_ins->ins.dst)
5252         {
5253             HeapFree(GetProcessHeap(), 0, (void *) rec_ins->ins.dst->reg.rel_addr);
5254             HeapFree(GetProcessHeap(), 0, (void *) rec_ins->ins.dst);
5255         }
5256         if(rec_ins->ins.src)
5257         {
5258             for(i = 0; i < rec_ins->ins.src_count; i++)
5259             {
5260                 HeapFree(GetProcessHeap(), 0, (void *) rec_ins->ins.src[i].reg.rel_addr);
5261             }
5262             HeapFree(GetProcessHeap(), 0, (void *) rec_ins->ins.src);
5263         }
5264         HeapFree(GetProcessHeap(), 0, rec_ins);
5265     }
5266 }
5267
5268 static void shader_arb_handle_instruction(const struct wined3d_shader_instruction *ins) {
5269     SHADER_HANDLER hw_fct;
5270     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
5271     struct wined3d_shader *shader = ins->ctx->shader;
5272     struct control_frame *control_frame;
5273     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
5274     BOOL bool_const;
5275
5276     if(ins->handler_idx == WINED3DSIH_LOOP || ins->handler_idx == WINED3DSIH_REP)
5277     {
5278         control_frame = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*control_frame));
5279         list_add_head(&priv->control_frames, &control_frame->entry);
5280
5281         if(ins->handler_idx == WINED3DSIH_LOOP) control_frame->type = LOOP;
5282         if(ins->handler_idx == WINED3DSIH_REP) control_frame->type = REP;
5283
5284         if(priv->target_version >= NV2)
5285         {
5286             control_frame->no.loop = priv->num_loops++;
5287             priv->loop_depth++;
5288         }
5289         else
5290         {
5291             /* Don't bother recording when we're in a not used if branch */
5292             if(priv->muted)
5293             {
5294                 return;
5295             }
5296
5297             if(!priv->recording)
5298             {
5299                 list_init(&priv->record);
5300                 priv->recording = TRUE;
5301                 control_frame->outer_loop = TRUE;
5302                 get_loop_control_const(ins, shader, ins->src[0].reg.idx, &control_frame->loop_control);
5303                 return; /* Instruction is handled */
5304             }
5305             /* Record this loop in the outer loop's recording */
5306         }
5307     }
5308     else if(ins->handler_idx == WINED3DSIH_ENDLOOP || ins->handler_idx == WINED3DSIH_ENDREP)
5309     {
5310         if(priv->target_version >= NV2)
5311         {
5312             /* Nothing to do. The control frame is popped after the HW instr handler */
5313         }
5314         else
5315         {
5316             struct list *e = list_head(&priv->control_frames);
5317             control_frame = LIST_ENTRY(e, struct control_frame, entry);
5318             list_remove(&control_frame->entry);
5319
5320             if(control_frame->outer_loop)
5321             {
5322                 unsigned int iteration;
5323                 int aL = 0;
5324                 struct list copy;
5325
5326                 /* Turn off recording before playback */
5327                 priv->recording = FALSE;
5328
5329                 /* Move the recorded instructions to a separate list and get them out of the private data
5330                  * structure. If there are nested loops, the shader_arb_handle_instruction below will
5331                  * be recorded again, thus priv->record might be overwritten
5332                  */
5333                 list_init(&copy);
5334                 list_move_tail(&copy, &priv->record);
5335                 list_init(&priv->record);
5336
5337                 if(ins->handler_idx == WINED3DSIH_ENDLOOP)
5338                 {
5339                     shader_addline(buffer, "#unrolling loop: %u iterations, aL=%u, inc %d\n",
5340                                    control_frame->loop_control.count, control_frame->loop_control.start,
5341                                    control_frame->loop_control.step);
5342                     aL = control_frame->loop_control.start;
5343                 }
5344                 else
5345                 {
5346                     shader_addline(buffer, "#unrolling rep: %u iterations\n", control_frame->loop_control.count);
5347                 }
5348
5349                 for (iteration = 0; iteration < control_frame->loop_control.count; ++iteration)
5350                 {
5351                     struct recorded_instruction *rec_ins;
5352                     if(ins->handler_idx == WINED3DSIH_ENDLOOP)
5353                     {
5354                         priv->aL = aL;
5355                         shader_addline(buffer, "#Iteration %u, aL=%d\n", iteration, aL);
5356                     }
5357                     else
5358                     {
5359                         shader_addline(buffer, "#Iteration %u\n", iteration);
5360                     }
5361
5362                     LIST_FOR_EACH_ENTRY(rec_ins, &copy, struct recorded_instruction, entry)
5363                     {
5364                         shader_arb_handle_instruction(&rec_ins->ins);
5365                     }
5366
5367                     if(ins->handler_idx == WINED3DSIH_ENDLOOP)
5368                     {
5369                         aL += control_frame->loop_control.step;
5370                     }
5371                 }
5372                 shader_addline(buffer, "#end loop/rep\n");
5373
5374                 free_recorded_instruction(&copy);
5375                 HeapFree(GetProcessHeap(), 0, control_frame);
5376                 return; /* Instruction is handled */
5377             }
5378             else
5379             {
5380                 /* This is a nested loop. Proceed to the normal recording function */
5381                 HeapFree(GetProcessHeap(), 0, control_frame);
5382             }
5383         }
5384     }
5385
5386     if(priv->recording)
5387     {
5388         record_instruction(&priv->record, ins);
5389         return;
5390     }
5391
5392     /* boolean if */
5393     if(ins->handler_idx == WINED3DSIH_IF)
5394     {
5395         control_frame = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*control_frame));
5396         list_add_head(&priv->control_frames, &control_frame->entry);
5397         control_frame->type = IF;
5398
5399         bool_const = get_bool_const(ins, shader, ins->src[0].reg.idx);
5400         if(ins->src[0].modifiers == WINED3DSPSM_NOT) bool_const = !bool_const;
5401         if (!priv->muted && !bool_const)
5402         {
5403             shader_addline(buffer, "#if(FALSE){\n");
5404             priv->muted = TRUE;
5405             control_frame->muting = TRUE;
5406         }
5407         else shader_addline(buffer, "#if(TRUE) {\n");
5408
5409         return; /* Instruction is handled */
5410     }
5411     else if(ins->handler_idx == WINED3DSIH_IFC)
5412     {
5413         /* IF(bool) and if_cond(a, b) use the same ELSE and ENDIF tokens */
5414         control_frame = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*control_frame));
5415         control_frame->type = IFC;
5416         control_frame->no.ifc = priv->num_ifcs++;
5417         list_add_head(&priv->control_frames, &control_frame->entry);
5418     }
5419     else if(ins->handler_idx == WINED3DSIH_ELSE)
5420     {
5421         struct list *e = list_head(&priv->control_frames);
5422         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5423
5424         if(control_frame->type == IF)
5425         {
5426             shader_addline(buffer, "#} else {\n");
5427             if(!priv->muted && !control_frame->muting)
5428             {
5429                 priv->muted = TRUE;
5430                 control_frame->muting = TRUE;
5431             }
5432             else if(control_frame->muting) priv->muted = FALSE;
5433             return; /* Instruction is handled. */
5434         }
5435         /* In case of an ifc, generate a HW shader instruction */
5436     }
5437     else if(ins->handler_idx == WINED3DSIH_ENDIF)
5438     {
5439         struct list *e = list_head(&priv->control_frames);
5440         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5441
5442         if(control_frame->type == IF)
5443         {
5444             shader_addline(buffer, "#} endif\n");
5445             if(control_frame->muting) priv->muted = FALSE;
5446             list_remove(&control_frame->entry);
5447             HeapFree(GetProcessHeap(), 0, control_frame);
5448             return; /* Instruction is handled */
5449         }
5450     }
5451
5452     if(priv->muted) return;
5453
5454     /* Select handler */
5455     hw_fct = shader_arb_instruction_handler_table[ins->handler_idx];
5456
5457     /* Unhandled opcode */
5458     if (!hw_fct)
5459     {
5460         FIXME("Backend can't handle opcode %#x\n", ins->handler_idx);
5461         return;
5462     }
5463     hw_fct(ins);
5464
5465     if(ins->handler_idx == WINED3DSIH_ENDLOOP || ins->handler_idx == WINED3DSIH_ENDREP)
5466     {
5467         struct list *e = list_head(&priv->control_frames);
5468         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5469         list_remove(&control_frame->entry);
5470         HeapFree(GetProcessHeap(), 0, control_frame);
5471         priv->loop_depth--;
5472     }
5473     else if(ins->handler_idx == WINED3DSIH_ENDIF)
5474     {
5475         /* Non-ifc ENDIFs don't reach that place because of the return in the if block above */
5476         struct list *e = list_head(&priv->control_frames);
5477         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5478         list_remove(&control_frame->entry);
5479         HeapFree(GetProcessHeap(), 0, control_frame);
5480     }
5481
5482
5483     shader_arb_add_instruction_modifiers(ins);
5484 }
5485
5486 const struct wined3d_shader_backend_ops arb_program_shader_backend =
5487 {
5488     shader_arb_handle_instruction,
5489     shader_arb_select,
5490     shader_arb_select_depth_blt,
5491     shader_arb_deselect_depth_blt,
5492     shader_arb_update_float_vertex_constants,
5493     shader_arb_update_float_pixel_constants,
5494     shader_arb_load_constants,
5495     shader_arb_load_np2fixup_constants,
5496     shader_arb_destroy,
5497     shader_arb_alloc,
5498     shader_arb_free,
5499     shader_arb_dirty_const,
5500     shader_arb_get_caps,
5501     shader_arb_color_fixup_supported,
5502 };
5503
5504 /* ARB_fragment_program fixed function pipeline replacement definitions */
5505 #define ARB_FFP_CONST_TFACTOR           0
5506 #define ARB_FFP_CONST_SPECULAR_ENABLE   ((ARB_FFP_CONST_TFACTOR) + 1)
5507 #define ARB_FFP_CONST_CONSTANT(i)       ((ARB_FFP_CONST_SPECULAR_ENABLE) + 1 + i)
5508 #define ARB_FFP_CONST_BUMPMAT(i)        ((ARB_FFP_CONST_CONSTANT(7)) + 1 + i)
5509 #define ARB_FFP_CONST_LUMINANCE(i)      ((ARB_FFP_CONST_BUMPMAT(7)) + 1 + i)
5510
5511 struct arbfp_ffp_desc
5512 {
5513     struct ffp_frag_desc parent;
5514     GLuint shader;
5515     unsigned int num_textures_used;
5516 };
5517
5518 /* Context activation is done by the caller. */
5519 static void arbfp_enable(BOOL enable)
5520 {
5521     ENTER_GL();
5522     if(enable) {
5523         glEnable(GL_FRAGMENT_PROGRAM_ARB);
5524         checkGLcall("glEnable(GL_FRAGMENT_PROGRAM_ARB)");
5525     } else {
5526         glDisable(GL_FRAGMENT_PROGRAM_ARB);
5527         checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
5528     }
5529     LEAVE_GL();
5530 }
5531
5532 static HRESULT arbfp_alloc(IWineD3DDeviceImpl *device)
5533 {
5534     struct shader_arb_priv *priv;
5535     /* Share private data between the shader backend and the pipeline replacement, if both
5536      * are the arb implementation. This is needed to figure out whether ARBfp should be disabled
5537      * if no pixel shader is bound or not
5538      */
5539     if (device->shader_backend == &arb_program_shader_backend)
5540     {
5541         device->fragment_priv = device->shader_priv;
5542     }
5543     else
5544     {
5545         device->fragment_priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(struct shader_arb_priv));
5546         if (!device->fragment_priv) return E_OUTOFMEMORY;
5547     }
5548     priv = device->fragment_priv;
5549     if (wine_rb_init(&priv->fragment_shaders, &wined3d_ffp_frag_program_rb_functions) == -1)
5550     {
5551         ERR("Failed to initialize rbtree.\n");
5552         HeapFree(GetProcessHeap(), 0, device->fragment_priv);
5553         return E_OUTOFMEMORY;
5554     }
5555     priv->use_arbfp_fixed_func = TRUE;
5556     return WINED3D_OK;
5557 }
5558
5559 /* Context activation is done by the caller. */
5560 static void arbfp_free_ffpshader(struct wine_rb_entry *entry, void *context)
5561 {
5562     const struct wined3d_gl_info *gl_info = context;
5563     struct arbfp_ffp_desc *entry_arb = WINE_RB_ENTRY_VALUE(entry, struct arbfp_ffp_desc, parent.entry);
5564
5565     ENTER_GL();
5566     GL_EXTCALL(glDeleteProgramsARB(1, &entry_arb->shader));
5567     checkGLcall("glDeleteProgramsARB(1, &entry_arb->shader)");
5568     HeapFree(GetProcessHeap(), 0, entry_arb);
5569     LEAVE_GL();
5570 }
5571
5572 /* Context activation is done by the caller. */
5573 static void arbfp_free(IWineD3DDeviceImpl *device)
5574 {
5575     struct shader_arb_priv *priv = device->fragment_priv;
5576
5577     wine_rb_destroy(&priv->fragment_shaders, arbfp_free_ffpshader, &device->adapter->gl_info);
5578     priv->use_arbfp_fixed_func = FALSE;
5579
5580     if (device->shader_backend != &arb_program_shader_backend)
5581     {
5582         HeapFree(GetProcessHeap(), 0, device->fragment_priv);
5583     }
5584 }
5585
5586 static void arbfp_get_caps(const struct wined3d_gl_info *gl_info, struct fragment_caps *caps)
5587 {
5588     caps->PrimitiveMiscCaps = WINED3DPMISCCAPS_TSSARGTEMP;
5589     caps->TextureOpCaps =  WINED3DTEXOPCAPS_DISABLE                     |
5590                            WINED3DTEXOPCAPS_SELECTARG1                  |
5591                            WINED3DTEXOPCAPS_SELECTARG2                  |
5592                            WINED3DTEXOPCAPS_MODULATE4X                  |
5593                            WINED3DTEXOPCAPS_MODULATE2X                  |
5594                            WINED3DTEXOPCAPS_MODULATE                    |
5595                            WINED3DTEXOPCAPS_ADDSIGNED2X                 |
5596                            WINED3DTEXOPCAPS_ADDSIGNED                   |
5597                            WINED3DTEXOPCAPS_ADD                         |
5598                            WINED3DTEXOPCAPS_SUBTRACT                    |
5599                            WINED3DTEXOPCAPS_ADDSMOOTH                   |
5600                            WINED3DTEXOPCAPS_BLENDCURRENTALPHA           |
5601                            WINED3DTEXOPCAPS_BLENDFACTORALPHA            |
5602                            WINED3DTEXOPCAPS_BLENDTEXTUREALPHA           |
5603                            WINED3DTEXOPCAPS_BLENDDIFFUSEALPHA           |
5604                            WINED3DTEXOPCAPS_BLENDTEXTUREALPHAPM         |
5605                            WINED3DTEXOPCAPS_MODULATEALPHA_ADDCOLOR      |
5606                            WINED3DTEXOPCAPS_MODULATECOLOR_ADDALPHA      |
5607                            WINED3DTEXOPCAPS_MODULATEINVCOLOR_ADDALPHA   |
5608                            WINED3DTEXOPCAPS_MODULATEINVALPHA_ADDCOLOR   |
5609                            WINED3DTEXOPCAPS_DOTPRODUCT3                 |
5610                            WINED3DTEXOPCAPS_MULTIPLYADD                 |
5611                            WINED3DTEXOPCAPS_LERP                        |
5612                            WINED3DTEXOPCAPS_BUMPENVMAP                  |
5613                            WINED3DTEXOPCAPS_BUMPENVMAPLUMINANCE;
5614
5615     /* TODO: Implement WINED3DTEXOPCAPS_PREMODULATE */
5616
5617     caps->MaxTextureBlendStages   = 8;
5618     caps->MaxSimultaneousTextures = min(gl_info->limits.fragment_samplers, 8);
5619 }
5620
5621 static void state_texfactor_arbfp(DWORD state_id,
5622         struct wined3d_stateblock *stateblock, struct wined3d_context *context)
5623 {
5624     const struct wined3d_gl_info *gl_info = context->gl_info;
5625     const struct wined3d_state *state = &stateblock->state;
5626     IWineD3DDeviceImpl *device = stateblock->device;
5627     float col[4];
5628
5629     /* Don't load the parameter if we're using an arbfp pixel shader, otherwise we'll overwrite
5630      * application provided constants
5631      */
5632     if (device->shader_backend == &arb_program_shader_backend)
5633     {
5634         if (use_ps(state)) return;
5635
5636         context->pshader_const_dirty[ARB_FFP_CONST_TFACTOR] = 1;
5637         device->highest_dirty_ps_const = max(device->highest_dirty_ps_const, ARB_FFP_CONST_TFACTOR + 1);
5638     }
5639
5640     D3DCOLORTOGLFLOAT4(state->render_states[WINED3DRS_TEXTUREFACTOR], col);
5641     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_TFACTOR, col));
5642     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_TFACTOR, col)");
5643
5644 }
5645
5646 static void state_arb_specularenable(DWORD state_id,
5647         struct wined3d_stateblock *stateblock, struct wined3d_context *context)
5648 {
5649     const struct wined3d_gl_info *gl_info = context->gl_info;
5650     const struct wined3d_state *state = &stateblock->state;
5651     IWineD3DDeviceImpl *device = stateblock->device;
5652     float col[4];
5653
5654     /* Don't load the parameter if we're using an arbfp pixel shader, otherwise we'll overwrite
5655      * application provided constants
5656      */
5657     if (device->shader_backend == &arb_program_shader_backend)
5658     {
5659         if (use_ps(state)) return;
5660
5661         context->pshader_const_dirty[ARB_FFP_CONST_SPECULAR_ENABLE] = 1;
5662         device->highest_dirty_ps_const = max(device->highest_dirty_ps_const, ARB_FFP_CONST_SPECULAR_ENABLE + 1);
5663     }
5664
5665     if (state->render_states[WINED3DRS_SPECULARENABLE])
5666     {
5667         /* The specular color has no alpha */
5668         col[0] = 1.0f; col[1] = 1.0f;
5669         col[2] = 1.0f; col[3] = 0.0f;
5670     } else {
5671         col[0] = 0.0f; col[1] = 0.0f;
5672         col[2] = 0.0f; col[3] = 0.0f;
5673     }
5674     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_SPECULAR_ENABLE, col));
5675     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_SPECULAR_ENABLE, col)");
5676 }
5677
5678 static void set_bumpmat_arbfp(DWORD state_id, struct wined3d_stateblock *stateblock, struct wined3d_context *context)
5679 {
5680     DWORD stage = (state_id - STATE_TEXTURESTAGE(0, 0)) / (WINED3D_HIGHEST_TEXTURE_STATE + 1);
5681     const struct wined3d_gl_info *gl_info = context->gl_info;
5682     const struct wined3d_state *state = &stateblock->state;
5683     IWineD3DDeviceImpl *device = stateblock->device;
5684     float mat[2][2];
5685
5686     if (use_ps(state))
5687     {
5688         if (stage && (state->pixel_shader->reg_maps.bumpmat & (1 << stage)))
5689         {
5690             /* The pixel shader has to know the bump env matrix. Do a constants update if it isn't scheduled
5691              * anyway
5692              */
5693             if (!isStateDirty(context, STATE_PIXELSHADERCONSTANT))
5694                 stateblock_apply_state(STATE_PIXELSHADERCONSTANT, stateblock, context);
5695         }
5696
5697         if(device->shader_backend == &arb_program_shader_backend) {
5698             /* Exit now, don't set the bumpmat below, otherwise we may overwrite pixel shader constants */
5699             return;
5700         }
5701     } else if(device->shader_backend == &arb_program_shader_backend) {
5702         context->pshader_const_dirty[ARB_FFP_CONST_BUMPMAT(stage)] = 1;
5703         device->highest_dirty_ps_const = max(device->highest_dirty_ps_const, ARB_FFP_CONST_BUMPMAT(stage) + 1);
5704     }
5705
5706     mat[0][0] = *((float *)&state->texture_states[stage][WINED3DTSS_BUMPENVMAT00]);
5707     mat[0][1] = *((float *)&state->texture_states[stage][WINED3DTSS_BUMPENVMAT01]);
5708     mat[1][0] = *((float *)&state->texture_states[stage][WINED3DTSS_BUMPENVMAT10]);
5709     mat[1][1] = *((float *)&state->texture_states[stage][WINED3DTSS_BUMPENVMAT11]);
5710
5711     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_BUMPMAT(stage), &mat[0][0]));
5712     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_BUMPMAT(stage), &mat[0][0])");
5713 }
5714
5715 static void tex_bumpenvlum_arbfp(DWORD state_id,
5716         struct wined3d_stateblock *stateblock, struct wined3d_context *context)
5717 {
5718     DWORD stage = (state_id - STATE_TEXTURESTAGE(0, 0)) / (WINED3D_HIGHEST_TEXTURE_STATE + 1);
5719     const struct wined3d_gl_info *gl_info = context->gl_info;
5720     const struct wined3d_state *state = &stateblock->state;
5721     IWineD3DDeviceImpl *device = stateblock->device;
5722     float param[4];
5723
5724     if (use_ps(state))
5725     {
5726         if (stage && (state->pixel_shader->reg_maps.luminanceparams & (1 << stage)))
5727         {
5728             /* The pixel shader has to know the luminance offset. Do a constants update if it
5729              * isn't scheduled anyway
5730              */
5731             if (!isStateDirty(context, STATE_PIXELSHADERCONSTANT))
5732                 stateblock_apply_state(STATE_PIXELSHADERCONSTANT, stateblock, context);
5733         }
5734
5735         if(device->shader_backend == &arb_program_shader_backend) {
5736             /* Exit now, don't set the bumpmat below, otherwise we may overwrite pixel shader constants */
5737             return;
5738         }
5739     } else if(device->shader_backend == &arb_program_shader_backend) {
5740         context->pshader_const_dirty[ARB_FFP_CONST_LUMINANCE(stage)] = 1;
5741         device->highest_dirty_ps_const = max(device->highest_dirty_ps_const, ARB_FFP_CONST_LUMINANCE(stage) + 1);
5742     }
5743
5744     param[0] = *((float *)&state->texture_states[stage][WINED3DTSS_BUMPENVLSCALE]);
5745     param[1] = *((float *)&state->texture_states[stage][WINED3DTSS_BUMPENVLOFFSET]);
5746     param[2] = 0.0f;
5747     param[3] = 0.0f;
5748
5749     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_LUMINANCE(stage), param));
5750     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_LUMINANCE(stage), param)");
5751 }
5752
5753 static const char *get_argreg(struct wined3d_shader_buffer *buffer, DWORD argnum, unsigned int stage, DWORD arg)
5754 {
5755     const char *ret;
5756
5757     if(arg == ARG_UNUSED) return "unused"; /* This is the marker for unused registers */
5758
5759     switch(arg & WINED3DTA_SELECTMASK) {
5760         case WINED3DTA_DIFFUSE:
5761             ret = "fragment.color.primary"; break;
5762
5763         case WINED3DTA_CURRENT:
5764             if (!stage) ret = "fragment.color.primary";
5765             else ret = "ret";
5766             break;
5767
5768         case WINED3DTA_TEXTURE:
5769             switch(stage) {
5770                 case 0: ret = "tex0"; break;
5771                 case 1: ret = "tex1"; break;
5772                 case 2: ret = "tex2"; break;
5773                 case 3: ret = "tex3"; break;
5774                 case 4: ret = "tex4"; break;
5775                 case 5: ret = "tex5"; break;
5776                 case 6: ret = "tex6"; break;
5777                 case 7: ret = "tex7"; break;
5778                 default: ret = "unknown texture";
5779             }
5780             break;
5781
5782         case WINED3DTA_TFACTOR:
5783             ret = "tfactor"; break;
5784
5785         case WINED3DTA_SPECULAR:
5786             ret = "fragment.color.secondary"; break;
5787
5788         case WINED3DTA_TEMP:
5789             ret = "tempreg"; break;
5790
5791         case WINED3DTA_CONSTANT:
5792             FIXME("Implement perstage constants\n");
5793             switch(stage) {
5794                 case 0: ret = "const0"; break;
5795                 case 1: ret = "const1"; break;
5796                 case 2: ret = "const2"; break;
5797                 case 3: ret = "const3"; break;
5798                 case 4: ret = "const4"; break;
5799                 case 5: ret = "const5"; break;
5800                 case 6: ret = "const6"; break;
5801                 case 7: ret = "const7"; break;
5802                 default: ret = "unknown constant";
5803             }
5804             break;
5805
5806         default:
5807             return "unknown";
5808     }
5809
5810     if(arg & WINED3DTA_COMPLEMENT) {
5811         shader_addline(buffer, "SUB arg%u, const.x, %s;\n", argnum, ret);
5812         if(argnum == 0) ret = "arg0";
5813         if(argnum == 1) ret = "arg1";
5814         if(argnum == 2) ret = "arg2";
5815     }
5816     if(arg & WINED3DTA_ALPHAREPLICATE) {
5817         shader_addline(buffer, "MOV arg%u, %s.w;\n", argnum, ret);
5818         if(argnum == 0) ret = "arg0";
5819         if(argnum == 1) ret = "arg1";
5820         if(argnum == 2) ret = "arg2";
5821     }
5822     return ret;
5823 }
5824
5825 static void gen_ffp_instr(struct wined3d_shader_buffer *buffer, unsigned int stage, BOOL color,
5826         BOOL alpha, DWORD dst, DWORD op, DWORD dw_arg0, DWORD dw_arg1, DWORD dw_arg2)
5827 {
5828     const char *dstmask, *dstreg, *arg0, *arg1, *arg2;
5829     unsigned int mul = 1;
5830     BOOL mul_final_dest = FALSE;
5831
5832     if(color && alpha) dstmask = "";
5833     else if(color) dstmask = ".xyz";
5834     else dstmask = ".w";
5835
5836     if(dst == tempreg) dstreg = "tempreg";
5837     else dstreg = "ret";
5838
5839     arg0 = get_argreg(buffer, 0, stage, dw_arg0);
5840     arg1 = get_argreg(buffer, 1, stage, dw_arg1);
5841     arg2 = get_argreg(buffer, 2, stage, dw_arg2);
5842
5843     switch(op) {
5844         case WINED3DTOP_DISABLE:
5845             if (!stage) shader_addline(buffer, "MOV %s%s, fragment.color.primary;\n", dstreg, dstmask);
5846             break;
5847
5848         case WINED3DTOP_SELECTARG2:
5849             arg1 = arg2;
5850             /* FALLTHROUGH */
5851         case WINED3DTOP_SELECTARG1:
5852             shader_addline(buffer, "MOV %s%s, %s;\n", dstreg, dstmask, arg1);
5853             break;
5854
5855         case WINED3DTOP_MODULATE4X:
5856             mul = 2;
5857             /* FALLTHROUGH */
5858         case WINED3DTOP_MODULATE2X:
5859             mul *= 2;
5860             if (!strcmp(dstreg, "result.color"))
5861             {
5862                 dstreg = "ret";
5863                 mul_final_dest = TRUE;
5864             }
5865             /* FALLTHROUGH */
5866         case WINED3DTOP_MODULATE:
5867             shader_addline(buffer, "MUL %s%s, %s, %s;\n", dstreg, dstmask, arg1, arg2);
5868             break;
5869
5870         case WINED3DTOP_ADDSIGNED2X:
5871             mul = 2;
5872             if (!strcmp(dstreg, "result.color"))
5873             {
5874                 dstreg = "ret";
5875                 mul_final_dest = TRUE;
5876             }
5877             /* FALLTHROUGH */
5878         case WINED3DTOP_ADDSIGNED:
5879             shader_addline(buffer, "SUB arg2, %s, const.w;\n", arg2);
5880             arg2 = "arg2";
5881             /* FALLTHROUGH */
5882         case WINED3DTOP_ADD:
5883             shader_addline(buffer, "ADD_SAT %s%s, %s, %s;\n", dstreg, dstmask, arg1, arg2);
5884             break;
5885
5886         case WINED3DTOP_SUBTRACT:
5887             shader_addline(buffer, "SUB_SAT %s%s, %s, %s;\n", dstreg, dstmask, arg1, arg2);
5888             break;
5889
5890         case WINED3DTOP_ADDSMOOTH:
5891             shader_addline(buffer, "SUB arg1, const.x, %s;\n", arg1);
5892             shader_addline(buffer, "MAD_SAT %s%s, arg1, %s, %s;\n", dstreg, dstmask, arg2, arg1);
5893             break;
5894
5895         case WINED3DTOP_BLENDCURRENTALPHA:
5896             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_CURRENT);
5897             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
5898             break;
5899         case WINED3DTOP_BLENDFACTORALPHA:
5900             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_TFACTOR);
5901             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
5902             break;
5903         case WINED3DTOP_BLENDTEXTUREALPHA:
5904             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_TEXTURE);
5905             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
5906             break;
5907         case WINED3DTOP_BLENDDIFFUSEALPHA:
5908             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_DIFFUSE);
5909             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
5910             break;
5911
5912         case WINED3DTOP_BLENDTEXTUREALPHAPM:
5913             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_TEXTURE);
5914             shader_addline(buffer, "SUB arg0.w, const.x, %s.w;\n", arg0);
5915             shader_addline(buffer, "MAD_SAT %s%s, %s, arg0.w, %s;\n", dstreg, dstmask, arg2, arg1);
5916             break;
5917
5918         /* D3DTOP_PREMODULATE ???? */
5919
5920         case WINED3DTOP_MODULATEINVALPHA_ADDCOLOR:
5921             shader_addline(buffer, "SUB arg0.w, const.x, %s;\n", arg1);
5922             shader_addline(buffer, "MAD_SAT %s%s, arg0.w, %s, %s;\n", dstreg, dstmask, arg2, arg1);
5923             break;
5924         case WINED3DTOP_MODULATEALPHA_ADDCOLOR:
5925             shader_addline(buffer, "MAD_SAT %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg1, arg2, arg1);
5926             break;
5927         case WINED3DTOP_MODULATEINVCOLOR_ADDALPHA:
5928             shader_addline(buffer, "SUB arg0, const.x, %s;\n", arg1);
5929             shader_addline(buffer, "MAD_SAT %s%s, arg0, %s, %s.w;\n", dstreg, dstmask, arg2, arg1);
5930             break;
5931         case WINED3DTOP_MODULATECOLOR_ADDALPHA:
5932             shader_addline(buffer, "MAD_SAT %s%s, %s, %s, %s.w;\n", dstreg, dstmask, arg1, arg2, arg1);
5933             break;
5934
5935         case WINED3DTOP_DOTPRODUCT3:
5936             mul = 4;
5937             if (!strcmp(dstreg, "result.color"))
5938             {
5939                 dstreg = "ret";
5940                 mul_final_dest = TRUE;
5941             }
5942             shader_addline(buffer, "SUB arg1, %s, const.w;\n", arg1);
5943             shader_addline(buffer, "SUB arg2, %s, const.w;\n", arg2);
5944             shader_addline(buffer, "DP3_SAT %s%s, arg1, arg2;\n", dstreg, dstmask);
5945             break;
5946
5947         case WINED3DTOP_MULTIPLYADD:
5948             shader_addline(buffer, "MAD_SAT %s%s, %s, %s, %s;\n", dstreg, dstmask, arg1, arg2, arg0);
5949             break;
5950
5951         case WINED3DTOP_LERP:
5952             /* The msdn is not quite right here */
5953             shader_addline(buffer, "LRP %s%s, %s, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
5954             break;
5955
5956         case WINED3DTOP_BUMPENVMAP:
5957         case WINED3DTOP_BUMPENVMAPLUMINANCE:
5958             /* Those are handled in the first pass of the shader(generation pass 1 and 2) already */
5959             break;
5960
5961         default:
5962             FIXME("Unhandled texture op %08x\n", op);
5963     }
5964
5965     if(mul == 2) {
5966         shader_addline(buffer, "MUL_SAT %s%s, %s, const.y;\n", mul_final_dest ? "result.color" : dstreg, dstmask, dstreg);
5967     } else if(mul == 4) {
5968         shader_addline(buffer, "MUL_SAT %s%s, %s, const.z;\n", mul_final_dest ? "result.color" : dstreg, dstmask, dstreg);
5969     }
5970 }
5971
5972 static GLuint gen_arbfp_ffp_shader(const struct ffp_frag_settings *settings, struct wined3d_stateblock *stateblock)
5973 {
5974     const struct wined3d_gl_info *gl_info = &stateblock->device->adapter->gl_info;
5975     unsigned int stage;
5976     struct wined3d_shader_buffer buffer;
5977     BOOL tex_read[MAX_TEXTURES] = {FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE};
5978     BOOL bump_used[MAX_TEXTURES] = {FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE};
5979     BOOL luminance_used[MAX_TEXTURES] = {FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE};
5980     const char *textype;
5981     const char *instr, *sat;
5982     char colorcor_dst[8];
5983     GLuint ret;
5984     DWORD arg0, arg1, arg2;
5985     BOOL tempreg_used = FALSE, tfactor_used = FALSE;
5986     BOOL op_equal;
5987     const char *final_combiner_src = "ret";
5988     GLint pos;
5989
5990     /* Find out which textures are read */
5991     for(stage = 0; stage < MAX_TEXTURES; stage++) {
5992         if(settings->op[stage].cop == WINED3DTOP_DISABLE) break;
5993         arg0 = settings->op[stage].carg0 & WINED3DTA_SELECTMASK;
5994         arg1 = settings->op[stage].carg1 & WINED3DTA_SELECTMASK;
5995         arg2 = settings->op[stage].carg2 & WINED3DTA_SELECTMASK;
5996         if(arg0 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
5997         if(arg1 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
5998         if(arg2 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
5999
6000         if(settings->op[stage].cop == WINED3DTOP_BLENDTEXTUREALPHA) tex_read[stage] = TRUE;
6001         if(settings->op[stage].cop == WINED3DTOP_BLENDTEXTUREALPHAPM) tex_read[stage] = TRUE;
6002         if(settings->op[stage].cop == WINED3DTOP_BUMPENVMAP) {
6003             bump_used[stage] = TRUE;
6004             tex_read[stage] = TRUE;
6005         }
6006         if(settings->op[stage].cop == WINED3DTOP_BUMPENVMAPLUMINANCE) {
6007             bump_used[stage] = TRUE;
6008             tex_read[stage] = TRUE;
6009             luminance_used[stage] = TRUE;
6010         } else if(settings->op[stage].cop == WINED3DTOP_BLENDFACTORALPHA) {
6011             tfactor_used = TRUE;
6012         }
6013
6014         if(arg0 == WINED3DTA_TFACTOR || arg1 == WINED3DTA_TFACTOR || arg2 == WINED3DTA_TFACTOR) {
6015             tfactor_used = TRUE;
6016         }
6017
6018         if(settings->op[stage].dst == tempreg) tempreg_used = TRUE;
6019         if(arg0 == WINED3DTA_TEMP || arg1 == WINED3DTA_TEMP || arg2 == WINED3DTA_TEMP) {
6020             tempreg_used = TRUE;
6021         }
6022
6023         if(settings->op[stage].aop == WINED3DTOP_DISABLE) continue;
6024         arg0 = settings->op[stage].aarg0 & WINED3DTA_SELECTMASK;
6025         arg1 = settings->op[stage].aarg1 & WINED3DTA_SELECTMASK;
6026         arg2 = settings->op[stage].aarg2 & WINED3DTA_SELECTMASK;
6027         if(arg0 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
6028         if(arg1 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
6029         if(arg2 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
6030
6031         if(arg0 == WINED3DTA_TEMP || arg1 == WINED3DTA_TEMP || arg2 == WINED3DTA_TEMP) {
6032             tempreg_used = TRUE;
6033         }
6034         if(arg0 == WINED3DTA_TFACTOR || arg1 == WINED3DTA_TFACTOR || arg2 == WINED3DTA_TFACTOR) {
6035             tfactor_used = TRUE;
6036         }
6037     }
6038
6039     /* Shader header */
6040     if (!shader_buffer_init(&buffer))
6041     {
6042         ERR("Failed to initialize shader buffer.\n");
6043         return 0;
6044     }
6045
6046     shader_addline(&buffer, "!!ARBfp1.0\n");
6047
6048     switch(settings->fog) {
6049         case FOG_OFF:                                                         break;
6050         case FOG_LINEAR: shader_addline(&buffer, "OPTION ARB_fog_linear;\n"); break;
6051         case FOG_EXP:    shader_addline(&buffer, "OPTION ARB_fog_exp;\n");    break;
6052         case FOG_EXP2:   shader_addline(&buffer, "OPTION ARB_fog_exp2;\n");   break;
6053         default: FIXME("Unexpected fog setting %d\n", settings->fog);
6054     }
6055
6056     shader_addline(&buffer, "PARAM const = {1, 2, 4, 0.5};\n");
6057     shader_addline(&buffer, "TEMP TMP;\n");
6058     shader_addline(&buffer, "TEMP ret;\n");
6059     if(tempreg_used || settings->sRGB_write) shader_addline(&buffer, "TEMP tempreg;\n");
6060     shader_addline(&buffer, "TEMP arg0;\n");
6061     shader_addline(&buffer, "TEMP arg1;\n");
6062     shader_addline(&buffer, "TEMP arg2;\n");
6063     for(stage = 0; stage < MAX_TEXTURES; stage++) {
6064         if(!tex_read[stage]) continue;
6065         shader_addline(&buffer, "TEMP tex%u;\n", stage);
6066         if(!bump_used[stage]) continue;
6067         shader_addline(&buffer, "PARAM bumpmat%u = program.env[%u];\n", stage, ARB_FFP_CONST_BUMPMAT(stage));
6068         if(!luminance_used[stage]) continue;
6069         shader_addline(&buffer, "PARAM luminance%u = program.env[%u];\n", stage, ARB_FFP_CONST_LUMINANCE(stage));
6070     }
6071     if(tfactor_used) {
6072         shader_addline(&buffer, "PARAM tfactor = program.env[%u];\n", ARB_FFP_CONST_TFACTOR);
6073     }
6074         shader_addline(&buffer, "PARAM specular_enable = program.env[%u];\n", ARB_FFP_CONST_SPECULAR_ENABLE);
6075
6076     if(settings->sRGB_write) {
6077         shader_addline(&buffer, "PARAM srgb_consts1 = {%f, %f, %f, %f};\n",
6078                        srgb_mul_low, srgb_cmp, srgb_pow, srgb_mul_high);
6079         shader_addline(&buffer, "PARAM srgb_consts2 = {%f, %f, %f, %f};\n",
6080                        srgb_sub_high, 0.0, 0.0, 0.0);
6081     }
6082
6083     if (ffp_clip_emul(&stateblock->state) && settings->emul_clipplanes)
6084         shader_addline(&buffer, "KIL fragment.texcoord[7];\n");
6085
6086     /* Generate texture sampling instructions) */
6087     for(stage = 0; stage < MAX_TEXTURES && settings->op[stage].cop != WINED3DTOP_DISABLE; stage++) {
6088         if(!tex_read[stage]) continue;
6089
6090         switch(settings->op[stage].tex_type) {
6091             case tex_1d:                    textype = "1D";     break;
6092             case tex_2d:                    textype = "2D";     break;
6093             case tex_3d:                    textype = "3D";     break;
6094             case tex_cube:                  textype = "CUBE";   break;
6095             case tex_rect:                  textype = "RECT";   break;
6096             default: textype = "unexpected_textype";   break;
6097         }
6098
6099         if(settings->op[stage].cop == WINED3DTOP_BUMPENVMAP ||
6100            settings->op[stage].cop == WINED3DTOP_BUMPENVMAPLUMINANCE) {
6101             sat = "";
6102         } else {
6103             sat = "_SAT";
6104         }
6105
6106         if(settings->op[stage].projected == proj_none) {
6107             instr = "TEX";
6108         } else if(settings->op[stage].projected == proj_count4 ||
6109                   settings->op[stage].projected == proj_count3) {
6110             instr = "TXP";
6111         } else {
6112             FIXME("Unexpected projection mode %d\n", settings->op[stage].projected);
6113             instr = "TXP";
6114         }
6115
6116         if(stage > 0 &&
6117            (settings->op[stage - 1].cop == WINED3DTOP_BUMPENVMAP ||
6118             settings->op[stage - 1].cop == WINED3DTOP_BUMPENVMAPLUMINANCE)) {
6119             shader_addline(&buffer, "SWZ arg1, bumpmat%u, x, z, 0, 0;\n", stage - 1);
6120             shader_addline(&buffer, "DP3 ret.x, arg1, tex%u;\n", stage - 1);
6121             shader_addline(&buffer, "SWZ arg1, bumpmat%u, y, w, 0, 0;\n", stage - 1);
6122             shader_addline(&buffer, "DP3 ret.y, arg1, tex%u;\n", stage - 1);
6123
6124             /* with projective textures, texbem only divides the static texture coord, not the displacement,
6125              * so multiply the displacement with the dividing parameter before passing it to TXP
6126              */
6127             if (settings->op[stage].projected != proj_none) {
6128                 if(settings->op[stage].projected == proj_count4) {
6129                     shader_addline(&buffer, "MOV ret.w, fragment.texcoord[%u].w;\n", stage);
6130                     shader_addline(&buffer, "MUL ret.xyz, ret, fragment.texcoord[%u].w, fragment.texcoord[%u];\n", stage, stage);
6131                 } else {
6132                     shader_addline(&buffer, "MOV ret.w, fragment.texcoord[%u].z;\n", stage);
6133                     shader_addline(&buffer, "MAD ret.xyz, ret, fragment.texcoord[%u].z, fragment.texcoord[%u];\n", stage, stage);
6134                 }
6135             } else {
6136                 shader_addline(&buffer, "ADD ret, ret, fragment.texcoord[%u];\n", stage);
6137             }
6138
6139             shader_addline(&buffer, "%s%s tex%u, ret, texture[%u], %s;\n",
6140                            instr, sat, stage, stage, textype);
6141             if(settings->op[stage - 1].cop == WINED3DTOP_BUMPENVMAPLUMINANCE) {
6142                 shader_addline(&buffer, "MAD_SAT ret.x, tex%u.z, luminance%u.x, luminance%u.y;\n",
6143                                stage - 1, stage - 1, stage - 1);
6144                 shader_addline(&buffer, "MUL tex%u, tex%u, ret.x;\n", stage, stage);
6145             }
6146         } else if(settings->op[stage].projected == proj_count3) {
6147             shader_addline(&buffer, "MOV ret, fragment.texcoord[%u];\n", stage);
6148             shader_addline(&buffer, "MOV ret.w, ret.z;\n");
6149             shader_addline(&buffer, "%s%s tex%u, ret, texture[%u], %s;\n",
6150                             instr, sat, stage, stage, textype);
6151         } else {
6152             shader_addline(&buffer, "%s%s tex%u, fragment.texcoord[%u], texture[%u], %s;\n",
6153                             instr, sat, stage, stage, stage, textype);
6154         }
6155
6156         sprintf(colorcor_dst, "tex%u", stage);
6157         gen_color_correction(&buffer, colorcor_dst, WINED3DSP_WRITEMASK_ALL, "const.x", "const.y",
6158                 settings->op[stage].color_fixup);
6159     }
6160
6161     /* Generate the main shader */
6162     for (stage = 0; stage < MAX_TEXTURES; ++stage)
6163     {
6164         if (settings->op[stage].cop == WINED3DTOP_DISABLE)
6165         {
6166             if (!stage) final_combiner_src = "fragment.color.primary";
6167             break;
6168         }
6169
6170         if(settings->op[stage].cop == WINED3DTOP_SELECTARG1 &&
6171            settings->op[stage].aop == WINED3DTOP_SELECTARG1) {
6172             op_equal = settings->op[stage].carg1 == settings->op[stage].aarg1;
6173         } else if(settings->op[stage].cop == WINED3DTOP_SELECTARG1 &&
6174                   settings->op[stage].aop == WINED3DTOP_SELECTARG2) {
6175             op_equal = settings->op[stage].carg1 == settings->op[stage].aarg2;
6176         } else if(settings->op[stage].cop == WINED3DTOP_SELECTARG2 &&
6177                   settings->op[stage].aop == WINED3DTOP_SELECTARG1) {
6178             op_equal = settings->op[stage].carg2 == settings->op[stage].aarg1;
6179         } else if(settings->op[stage].cop == WINED3DTOP_SELECTARG2 &&
6180                   settings->op[stage].aop == WINED3DTOP_SELECTARG2) {
6181             op_equal = settings->op[stage].carg2 == settings->op[stage].aarg2;
6182         } else {
6183             op_equal = settings->op[stage].aop   == settings->op[stage].cop &&
6184                        settings->op[stage].carg0 == settings->op[stage].aarg0 &&
6185                        settings->op[stage].carg1 == settings->op[stage].aarg1 &&
6186                        settings->op[stage].carg2 == settings->op[stage].aarg2;
6187         }
6188
6189         if(settings->op[stage].aop == WINED3DTOP_DISABLE) {
6190             gen_ffp_instr(&buffer, stage, TRUE, FALSE, settings->op[stage].dst,
6191                           settings->op[stage].cop, settings->op[stage].carg0,
6192                           settings->op[stage].carg1, settings->op[stage].carg2);
6193             if (!stage)
6194                 shader_addline(&buffer, "MOV ret.w, fragment.color.primary.w;\n");
6195         }
6196         else if (op_equal)
6197         {
6198             gen_ffp_instr(&buffer, stage, TRUE, TRUE, settings->op[stage].dst,
6199                           settings->op[stage].cop, settings->op[stage].carg0,
6200                           settings->op[stage].carg1, settings->op[stage].carg2);
6201         } else {
6202             gen_ffp_instr(&buffer, stage, TRUE, FALSE, settings->op[stage].dst,
6203                           settings->op[stage].cop, settings->op[stage].carg0,
6204                           settings->op[stage].carg1, settings->op[stage].carg2);
6205             gen_ffp_instr(&buffer, stage, FALSE, TRUE, settings->op[stage].dst,
6206                           settings->op[stage].aop, settings->op[stage].aarg0,
6207                           settings->op[stage].aarg1, settings->op[stage].aarg2);
6208         }
6209     }
6210
6211     if(settings->sRGB_write) {
6212         shader_addline(&buffer, "MAD ret, fragment.color.secondary, specular_enable, %s;\n", final_combiner_src);
6213         arbfp_add_sRGB_correction(&buffer, "ret", "arg0", "arg1", "arg2", "tempreg", FALSE);
6214         shader_addline(&buffer, "MOV result.color, ret;\n");
6215     } else {
6216         shader_addline(&buffer, "MAD result.color, fragment.color.secondary, specular_enable, %s;\n", final_combiner_src);
6217     }
6218
6219     /* Footer */
6220     shader_addline(&buffer, "END\n");
6221
6222     /* Generate the shader */
6223     GL_EXTCALL(glGenProgramsARB(1, &ret));
6224     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, ret));
6225     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
6226             strlen(buffer.buffer), buffer.buffer));
6227     checkGLcall("glProgramStringARB()");
6228
6229     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
6230     if (pos != -1)
6231     {
6232         FIXME("Fragment program error at position %d: %s\n\n", pos,
6233               debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
6234         shader_arb_dump_program_source(buffer.buffer);
6235     }
6236     else
6237     {
6238         GLint native;
6239
6240         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
6241         checkGLcall("glGetProgramivARB()");
6242         if (!native) WARN("Program exceeds native resource limits.\n");
6243     }
6244
6245     shader_buffer_free(&buffer);
6246     return ret;
6247 }
6248
6249 static void fragment_prog_arbfp(DWORD state_id, struct wined3d_stateblock *stateblock, struct wined3d_context *context)
6250 {
6251     const struct wined3d_gl_info *gl_info = context->gl_info;
6252     const struct wined3d_state *state = &stateblock->state;
6253     IWineD3DDeviceImpl *device = stateblock->device;
6254     struct shader_arb_priv *priv = device->fragment_priv;
6255     BOOL use_vshader = use_vs(state);
6256     BOOL use_pshader = use_ps(state);
6257     struct ffp_frag_settings settings;
6258     const struct arbfp_ffp_desc *desc;
6259     unsigned int i;
6260
6261     TRACE("state_id %#x, stateblock %p, context %p\n", state_id, stateblock, context);
6262
6263     if(isStateDirty(context, STATE_RENDER(WINED3DRS_FOGENABLE))) {
6264         if(!use_pshader && device->shader_backend == &arb_program_shader_backend && context->last_was_pshader) {
6265             /* Reload fixed function constants since they collide with the pixel shader constants */
6266             for(i = 0; i < MAX_TEXTURES; i++) {
6267                 set_bumpmat_arbfp(STATE_TEXTURESTAGE(i, WINED3DTSS_BUMPENVMAT00), stateblock, context);
6268             }
6269             state_texfactor_arbfp(STATE_RENDER(WINED3DRS_TEXTUREFACTOR), stateblock, context);
6270             state_arb_specularenable(STATE_RENDER(WINED3DRS_SPECULARENABLE), stateblock, context);
6271         } else if(use_pshader && !isStateDirty(context, device->StateTable[STATE_VSHADER].representative)) {
6272             device->shader_backend->shader_select(context, use_pshader, use_vshader);
6273         }
6274         return;
6275     }
6276
6277     if(!use_pshader) {
6278         /* Find or create a shader implementing the fixed function pipeline settings, then activate it */
6279         gen_ffp_frag_op(stateblock, &settings, FALSE);
6280         desc = (const struct arbfp_ffp_desc *)find_ffp_frag_shader(&priv->fragment_shaders, &settings);
6281         if(!desc) {
6282             struct arbfp_ffp_desc *new_desc = HeapAlloc(GetProcessHeap(), 0, sizeof(*new_desc));
6283             if (!new_desc)
6284             {
6285                 ERR("Out of memory\n");
6286                 return;
6287             }
6288             new_desc->num_textures_used = 0;
6289             for (i = 0; i < gl_info->limits.texture_stages; ++i)
6290             {
6291                 if(settings.op[i].cop == WINED3DTOP_DISABLE) break;
6292                 new_desc->num_textures_used = i;
6293             }
6294
6295             memcpy(&new_desc->parent.settings, &settings, sizeof(settings));
6296             new_desc->shader = gen_arbfp_ffp_shader(&settings, stateblock);
6297             add_ffp_frag_shader(&priv->fragment_shaders, &new_desc->parent);
6298             TRACE("Allocated fixed function replacement shader descriptor %p\n", new_desc);
6299             desc = new_desc;
6300         }
6301
6302         /* Now activate the replacement program. GL_FRAGMENT_PROGRAM_ARB is already active(however, note the
6303          * comment above the shader_select call below). If e.g. GLSL is active, the shader_select call will
6304          * deactivate it.
6305          */
6306         GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, desc->shader));
6307         checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, desc->shader)");
6308         priv->current_fprogram_id = desc->shader;
6309
6310         if(device->shader_backend == &arb_program_shader_backend && context->last_was_pshader) {
6311             /* Reload fixed function constants since they collide with the pixel shader constants */
6312             for(i = 0; i < MAX_TEXTURES; i++) {
6313                 set_bumpmat_arbfp(STATE_TEXTURESTAGE(i, WINED3DTSS_BUMPENVMAT00), stateblock, context);
6314             }
6315             state_texfactor_arbfp(STATE_RENDER(WINED3DRS_TEXTUREFACTOR), stateblock, context);
6316             state_arb_specularenable(STATE_RENDER(WINED3DRS_SPECULARENABLE), stateblock, context);
6317         }
6318         context->last_was_pshader = FALSE;
6319     } else {
6320         context->last_was_pshader = TRUE;
6321     }
6322
6323     /* Finally, select the shader. If a pixel shader is used, it will be set and enabled by the shader backend.
6324      * If this shader backend is arbfp(most likely), then it will simply overwrite the last fixed function replace-
6325      * ment shader. If the shader backend is not ARB, it currently is important that the opengl implementation
6326      * type overwrites GL_ARB_fragment_program. This is currently the case with GLSL. If we really want to use
6327      * atifs or nvrc pixel shaders with arb fragment programs we'd have to disable GL_FRAGMENT_PROGRAM_ARB here
6328      *
6329      * Don't call shader_select if the vertex shader is dirty, because it will be called later on by the vertex
6330      * shader handler
6331      */
6332     if(!isStateDirty(context, device->StateTable[STATE_VSHADER].representative)) {
6333         device->shader_backend->shader_select(context, use_pshader, use_vshader);
6334
6335         if (!isStateDirty(context, STATE_VERTEXSHADERCONSTANT) && (use_vshader || use_pshader))
6336             stateblock_apply_state(STATE_VERTEXSHADERCONSTANT, stateblock, context);
6337     }
6338     if (use_pshader) stateblock_apply_state(STATE_PIXELSHADERCONSTANT, stateblock, context);
6339 }
6340
6341 /* We can't link the fog states to the fragment state directly since the vertex pipeline links them
6342  * to FOGENABLE. A different linking in different pipeline parts can't be expressed in the combined
6343  * state table, so we need to handle that with a forwarding function. The other invisible side effect
6344  * is that changing the fog start and fog end(which links to FOGENABLE in vertex) results in the
6345  * fragment_prog_arbfp function being called because FOGENABLE is dirty, which calls this function here
6346  */
6347 static void state_arbfp_fog(DWORD state_id, struct wined3d_stateblock *stateblock, struct wined3d_context *context)
6348 {
6349     const struct wined3d_state *state = &stateblock->state;
6350     enum fogsource new_source;
6351
6352     TRACE("state_id %#x, stateblock %p, context %p\n", state_id, stateblock, context);
6353
6354     if(!isStateDirty(context, STATE_PIXELSHADER)) {
6355         fragment_prog_arbfp(state_id, stateblock, context);
6356     }
6357
6358     if (!state->render_states[WINED3DRS_FOGENABLE]) return;
6359
6360     if (state->render_states[WINED3DRS_FOGTABLEMODE] == WINED3DFOG_NONE)
6361     {
6362         if (use_vs(state))
6363         {
6364             new_source = FOGSOURCE_VS;
6365         }
6366         else
6367         {
6368             if (state->render_states[WINED3DRS_FOGVERTEXMODE] == WINED3DFOG_NONE || context->last_was_rhw)
6369                 new_source = FOGSOURCE_COORD;
6370             else
6371                 new_source = FOGSOURCE_FFP;
6372         }
6373     } else {
6374         new_source = FOGSOURCE_FFP;
6375     }
6376     if(new_source != context->fog_source) {
6377         context->fog_source = new_source;
6378         state_fogstartend(STATE_RENDER(WINED3DRS_FOGSTART), stateblock, context);
6379     }
6380 }
6381
6382 static void textransform(DWORD state, struct wined3d_stateblock *stateblock, struct wined3d_context *context)
6383 {
6384     if (!isStateDirty(context, STATE_PIXELSHADER))
6385         fragment_prog_arbfp(state, stateblock, context);
6386 }
6387
6388 static const struct StateEntryTemplate arbfp_fragmentstate_template[] = {
6389     {STATE_RENDER(WINED3DRS_TEXTUREFACTOR),               { STATE_RENDER(WINED3DRS_TEXTUREFACTOR),              state_texfactor_arbfp   }, WINED3D_GL_EXT_NONE             },
6390     {STATE_TEXTURESTAGE(0, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6391     {STATE_TEXTURESTAGE(0, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6392     {STATE_TEXTURESTAGE(0, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6393     {STATE_TEXTURESTAGE(0, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6394     {STATE_TEXTURESTAGE(0, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6395     {STATE_TEXTURESTAGE(0, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6396     {STATE_TEXTURESTAGE(0, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6397     {STATE_TEXTURESTAGE(0, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6398     {STATE_TEXTURESTAGE(0, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6399     {STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6400     {STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6401     {STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6402     {STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6403     {STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6404     {STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVLSCALE),    NULL                    }, WINED3D_GL_EXT_NONE             },
6405     {STATE_TEXTURESTAGE(1, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6406     {STATE_TEXTURESTAGE(1, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6407     {STATE_TEXTURESTAGE(1, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6408     {STATE_TEXTURESTAGE(1, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6409     {STATE_TEXTURESTAGE(1, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6410     {STATE_TEXTURESTAGE(1, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6411     {STATE_TEXTURESTAGE(1, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6412     {STATE_TEXTURESTAGE(1, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6413     {STATE_TEXTURESTAGE(1, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6414     {STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6415     {STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6416     {STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6417     {STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6418     {STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6419     {STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVLSCALE),    NULL                    }, WINED3D_GL_EXT_NONE             },
6420     {STATE_TEXTURESTAGE(2, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6421     {STATE_TEXTURESTAGE(2, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6422     {STATE_TEXTURESTAGE(2, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6423     {STATE_TEXTURESTAGE(2, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6424     {STATE_TEXTURESTAGE(2, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6425     {STATE_TEXTURESTAGE(2, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6426     {STATE_TEXTURESTAGE(2, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6427     {STATE_TEXTURESTAGE(2, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6428     {STATE_TEXTURESTAGE(2, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6429     {STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6430     {STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6431     {STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6432     {STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6433     {STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6434     {STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVLSCALE),    NULL                    }, WINED3D_GL_EXT_NONE             },
6435     {STATE_TEXTURESTAGE(3, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6436     {STATE_TEXTURESTAGE(3, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6437     {STATE_TEXTURESTAGE(3, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6438     {STATE_TEXTURESTAGE(3, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6439     {STATE_TEXTURESTAGE(3, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6440     {STATE_TEXTURESTAGE(3, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6441     {STATE_TEXTURESTAGE(3, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6442     {STATE_TEXTURESTAGE(3, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6443     {STATE_TEXTURESTAGE(3, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6444     {STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6445     {STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6446     {STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6447     {STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6448     {STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6449     {STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVLSCALE),    NULL                    }, WINED3D_GL_EXT_NONE             },
6450     {STATE_TEXTURESTAGE(4, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6451     {STATE_TEXTURESTAGE(4, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6452     {STATE_TEXTURESTAGE(4, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6453     {STATE_TEXTURESTAGE(4, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6454     {STATE_TEXTURESTAGE(4, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6455     {STATE_TEXTURESTAGE(4, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6456     {STATE_TEXTURESTAGE(4, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6457     {STATE_TEXTURESTAGE(4, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6458     {STATE_TEXTURESTAGE(4, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6459     {STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6460     {STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6461     {STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6462     {STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6463     {STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6464     {STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVLSCALE),    NULL                    }, WINED3D_GL_EXT_NONE             },
6465     {STATE_TEXTURESTAGE(5, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6466     {STATE_TEXTURESTAGE(5, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6467     {STATE_TEXTURESTAGE(5, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6468     {STATE_TEXTURESTAGE(5, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6469     {STATE_TEXTURESTAGE(5, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6470     {STATE_TEXTURESTAGE(5, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6471     {STATE_TEXTURESTAGE(5, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6472     {STATE_TEXTURESTAGE(5, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6473     {STATE_TEXTURESTAGE(5, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6474     {STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6475     {STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6476     {STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6477     {STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6478     {STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6479     {STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVLSCALE),    NULL                    }, WINED3D_GL_EXT_NONE             },
6480     {STATE_TEXTURESTAGE(6, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6481     {STATE_TEXTURESTAGE(6, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6482     {STATE_TEXTURESTAGE(6, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6483     {STATE_TEXTURESTAGE(6, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6484     {STATE_TEXTURESTAGE(6, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6485     {STATE_TEXTURESTAGE(6, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6486     {STATE_TEXTURESTAGE(6, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6487     {STATE_TEXTURESTAGE(6, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6488     {STATE_TEXTURESTAGE(6, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6489     {STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6490     {STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6491     {STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6492     {STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6493     {STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6494     {STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVLSCALE),    NULL                    }, WINED3D_GL_EXT_NONE             },
6495     {STATE_TEXTURESTAGE(7, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6496     {STATE_TEXTURESTAGE(7, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6497     {STATE_TEXTURESTAGE(7, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6498     {STATE_TEXTURESTAGE(7, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6499     {STATE_TEXTURESTAGE(7, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6500     {STATE_TEXTURESTAGE(7, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6501     {STATE_TEXTURESTAGE(7, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6502     {STATE_TEXTURESTAGE(7, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6503     {STATE_TEXTURESTAGE(7, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6504     {STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6505     {STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6506     {STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6507     {STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6508     {STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6509     {STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVLSCALE),    NULL                    }, WINED3D_GL_EXT_NONE             },
6510     {STATE_PIXELSHADER,                                   { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6511     {STATE_RENDER(WINED3DRS_FOGENABLE),                   { STATE_RENDER(WINED3DRS_FOGENABLE),                  state_arbfp_fog         }, WINED3D_GL_EXT_NONE             },
6512     {STATE_RENDER(WINED3DRS_FOGTABLEMODE),                { STATE_RENDER(WINED3DRS_FOGENABLE),                  NULL                    }, WINED3D_GL_EXT_NONE             },
6513     {STATE_RENDER(WINED3DRS_FOGVERTEXMODE),               { STATE_RENDER(WINED3DRS_FOGENABLE),                  NULL                    }, WINED3D_GL_EXT_NONE             },
6514     {STATE_RENDER(WINED3DRS_FOGSTART),                    { STATE_RENDER(WINED3DRS_FOGSTART),                   state_fogstartend       }, WINED3D_GL_EXT_NONE             },
6515     {STATE_RENDER(WINED3DRS_FOGEND),                      { STATE_RENDER(WINED3DRS_FOGSTART),                   NULL                    }, WINED3D_GL_EXT_NONE             },
6516     {STATE_RENDER(WINED3DRS_SRGBWRITEENABLE),             { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6517     {STATE_RENDER(WINED3DRS_FOGCOLOR),                    { STATE_RENDER(WINED3DRS_FOGCOLOR),                   state_fogcolor          }, WINED3D_GL_EXT_NONE             },
6518     {STATE_RENDER(WINED3DRS_FOGDENSITY),                  { STATE_RENDER(WINED3DRS_FOGDENSITY),                 state_fogdensity        }, WINED3D_GL_EXT_NONE             },
6519     {STATE_TEXTURESTAGE(0,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(0, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6520     {STATE_TEXTURESTAGE(1,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(1, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6521     {STATE_TEXTURESTAGE(2,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(2, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6522     {STATE_TEXTURESTAGE(3,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(3, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6523     {STATE_TEXTURESTAGE(4,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(4, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6524     {STATE_TEXTURESTAGE(5,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(5, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6525     {STATE_TEXTURESTAGE(6,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(6, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6526     {STATE_TEXTURESTAGE(7,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(7, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6527     {STATE_RENDER(WINED3DRS_SPECULARENABLE),              { STATE_RENDER(WINED3DRS_SPECULARENABLE),             state_arb_specularenable}, WINED3D_GL_EXT_NONE             },
6528     {0 /* Terminate */,                                   { 0,                                                  0                       }, WINED3D_GL_EXT_NONE             },
6529 };
6530
6531 const struct fragment_pipeline arbfp_fragment_pipeline = {
6532     arbfp_enable,
6533     arbfp_get_caps,
6534     arbfp_alloc,
6535     arbfp_free,
6536     shader_arb_color_fixup_supported,
6537     arbfp_fragmentstate_template,
6538     TRUE /* We can disable projected textures */
6539 };
6540
6541 struct arbfp_blit_priv {
6542     GLenum yuy2_rect_shader, yuy2_2d_shader;
6543     GLenum uyvy_rect_shader, uyvy_2d_shader;
6544     GLenum yv12_rect_shader, yv12_2d_shader;
6545     GLenum p8_rect_shader, p8_2d_shader;
6546     GLuint palette_texture;
6547 };
6548
6549 static HRESULT arbfp_blit_alloc(IWineD3DDeviceImpl *device)
6550 {
6551     device->blit_priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(struct arbfp_blit_priv));
6552     if(!device->blit_priv) {
6553         ERR("Out of memory\n");
6554         return E_OUTOFMEMORY;
6555     }
6556     return WINED3D_OK;
6557 }
6558
6559 /* Context activation is done by the caller. */
6560 static void arbfp_blit_free(IWineD3DDeviceImpl *device)
6561 {
6562     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
6563     struct arbfp_blit_priv *priv = device->blit_priv;
6564
6565     ENTER_GL();
6566     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yuy2_rect_shader));
6567     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yuy2_2d_shader));
6568     GL_EXTCALL(glDeleteProgramsARB(1, &priv->uyvy_rect_shader));
6569     GL_EXTCALL(glDeleteProgramsARB(1, &priv->uyvy_2d_shader));
6570     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yv12_rect_shader));
6571     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yv12_2d_shader));
6572     GL_EXTCALL(glDeleteProgramsARB(1, &priv->p8_rect_shader));
6573     GL_EXTCALL(glDeleteProgramsARB(1, &priv->p8_2d_shader));
6574     checkGLcall("Delete yuv and p8 programs");
6575
6576     if(priv->palette_texture) glDeleteTextures(1, &priv->palette_texture);
6577     LEAVE_GL();
6578
6579     HeapFree(GetProcessHeap(), 0, device->blit_priv);
6580     device->blit_priv = NULL;
6581 }
6582
6583 static BOOL gen_planar_yuv_read(struct wined3d_shader_buffer *buffer, enum complex_fixup fixup,
6584         GLenum textype, char *luminance)
6585 {
6586     char chroma;
6587     const char *tex, *texinstr;
6588
6589     if (fixup == COMPLEX_FIXUP_UYVY) {
6590         chroma = 'x';
6591         *luminance = 'w';
6592     } else {
6593         chroma = 'w';
6594         *luminance = 'x';
6595     }
6596     switch(textype) {
6597         case GL_TEXTURE_2D:             tex = "2D";     texinstr = "TXP"; break;
6598         case GL_TEXTURE_RECTANGLE_ARB:  tex = "RECT";   texinstr = "TEX"; break;
6599         default:
6600             /* This is more tricky than just replacing the texture type - we have to navigate
6601              * properly in the texture to find the correct chroma values
6602              */
6603             FIXME("Implement yuv correction for non-2d, non-rect textures\n");
6604             return FALSE;
6605     }
6606
6607     /* First we have to read the chroma values. This means we need at least two pixels(no filtering),
6608      * or 4 pixels(with filtering). To get the unmodified chromas, we have to rid ourselves of the
6609      * filtering when we sample the texture.
6610      *
6611      * These are the rules for reading the chroma:
6612      *
6613      * Even pixel: Cr
6614      * Even pixel: U
6615      * Odd pixel: V
6616      *
6617      * So we have to get the sampling x position in non-normalized coordinates in integers
6618      */
6619     if(textype != GL_TEXTURE_RECTANGLE_ARB) {
6620         shader_addline(buffer, "MUL texcrd.xy, fragment.texcoord[0], size.x;\n");
6621         shader_addline(buffer, "MOV texcrd.w, size.x;\n");
6622     } else {
6623         shader_addline(buffer, "MOV texcrd, fragment.texcoord[0];\n");
6624     }
6625     /* We must not allow filtering between pixel x and x+1, this would mix U and V
6626      * Vertical filtering is ok. However, bear in mind that the pixel center is at
6627      * 0.5, so add 0.5.
6628      */
6629     shader_addline(buffer, "FLR texcrd.x, texcrd.x;\n");
6630     shader_addline(buffer, "ADD texcrd.x, texcrd.x, coef.y;\n");
6631
6632     /* Divide the x coordinate by 0.5 and get the fraction. This gives 0.25 and 0.75 for the
6633      * even and odd pixels respectively
6634      */
6635     shader_addline(buffer, "MUL texcrd2, texcrd, coef.y;\n");
6636     shader_addline(buffer, "FRC texcrd2, texcrd2;\n");
6637
6638     /* Sample Pixel 1 */
6639     shader_addline(buffer, "%s luminance, texcrd, texture[0], %s;\n", texinstr, tex);
6640
6641     /* Put the value into either of the chroma values */
6642     shader_addline(buffer, "SGE temp.x, texcrd2.x, coef.y;\n");
6643     shader_addline(buffer, "MUL chroma.x, luminance.%c, temp.x;\n", chroma);
6644     shader_addline(buffer, "SLT temp.x, texcrd2.x, coef.y;\n");
6645     shader_addline(buffer, "MUL chroma.y, luminance.%c, temp.x;\n", chroma);
6646
6647     /* Sample pixel 2. If we read an even pixel(SLT above returned 1), sample
6648      * the pixel right to the current one. Otherwise, sample the left pixel.
6649      * Bias and scale the SLT result to -1;1 and add it to the texcrd.x.
6650      */
6651     shader_addline(buffer, "MAD temp.x, temp.x, coef.z, -coef.x;\n");
6652     shader_addline(buffer, "ADD texcrd.x, texcrd, temp.x;\n");
6653     shader_addline(buffer, "%s luminance, texcrd, texture[0], %s;\n", texinstr, tex);
6654
6655     /* Put the value into the other chroma */
6656     shader_addline(buffer, "SGE temp.x, texcrd2.x, coef.y;\n");
6657     shader_addline(buffer, "MAD chroma.y, luminance.%c, temp.x, chroma.y;\n", chroma);
6658     shader_addline(buffer, "SLT temp.x, texcrd2.x, coef.y;\n");
6659     shader_addline(buffer, "MAD chroma.x, luminance.%c, temp.x, chroma.x;\n", chroma);
6660
6661     /* TODO: If filtering is enabled, sample a 2nd pair of pixels left or right of
6662      * the current one and lerp the two U and V values
6663      */
6664
6665     /* This gives the correctly filtered luminance value */
6666     shader_addline(buffer, "TEX luminance, fragment.texcoord[0], texture[0], %s;\n", tex);
6667
6668     return TRUE;
6669 }
6670
6671 static BOOL gen_yv12_read(struct wined3d_shader_buffer *buffer, GLenum textype, char *luminance)
6672 {
6673     const char *tex;
6674
6675     switch(textype) {
6676         case GL_TEXTURE_2D:             tex = "2D";     break;
6677         case GL_TEXTURE_RECTANGLE_ARB:  tex = "RECT";   break;
6678         default:
6679             FIXME("Implement yv12 correction for non-2d, non-rect textures\n");
6680             return FALSE;
6681     }
6682
6683     /* YV12 surfaces contain a WxH sized luminance plane, followed by a (W/2)x(H/2)
6684      * V and a (W/2)x(H/2) U plane, each with 8 bit per pixel. So the effective
6685      * bitdepth is 12 bits per pixel. Since the U and V planes have only half the
6686      * pitch of the luminance plane, the packing into the gl texture is a bit
6687      * unfortunate. If the whole texture is interpreted as luminance data it looks
6688      * approximately like this:
6689      *
6690      *        +----------------------------------+----
6691      *        |                                  |
6692      *        |                                  |
6693      *        |                                  |
6694      *        |                                  |
6695      *        |                                  |   2
6696      *        |            LUMINANCE             |   -
6697      *        |                                  |   3
6698      *        |                                  |
6699      *        |                                  |
6700      *        |                                  |
6701      *        |                                  |
6702      *        +----------------+-----------------+----
6703      *        |                |                 |
6704      *        |  U even rows   |  U odd rows     |
6705      *        |                |                 |   1
6706      *        +----------------+------------------   -
6707      *        |                |                 |   3
6708      *        |  V even rows   |  V odd rows     |
6709      *        |                |                 |
6710      *        +----------------+-----------------+----
6711      *        |                |                 |
6712      *        |     0.5        |       0.5       |
6713      *
6714      * So it appears as if there are 4 chroma images, but in fact the odd rows
6715      * in the chroma images are in the same row as the even ones. So its is
6716      * kinda tricky to read
6717      *
6718      * When reading from rectangle textures, keep in mind that the input y coordinates
6719      * go from 0 to d3d_height, whereas the opengl texture height is 1.5 * d3d_height
6720      */
6721     shader_addline(buffer, "PARAM yv12_coef = {%f, %f, %f, %f};\n",
6722             2.0f / 3.0f, 1.0f / 6.0f, (2.0f / 3.0f) + (1.0f / 6.0f), 1.0f / 3.0f);
6723
6724     shader_addline(buffer, "MOV texcrd, fragment.texcoord[0];\n");
6725     /* the chroma planes have only half the width */
6726     shader_addline(buffer, "MUL texcrd.x, texcrd.x, coef.y;\n");
6727
6728     /* The first value is between 2/3 and 5/6th of the texture's height, so scale+bias
6729      * the coordinate. Also read the right side of the image when reading odd lines
6730      *
6731      * Don't forget to clamp the y values in into the range, otherwise we'll get filtering
6732      * bleeding
6733      */
6734     if(textype == GL_TEXTURE_2D) {
6735
6736         shader_addline(buffer, "RCP chroma.w, size.y;\n");
6737
6738         shader_addline(buffer, "MUL texcrd2.y, texcrd.y, size.y;\n");
6739
6740         shader_addline(buffer, "FLR texcrd2.y, texcrd2.y;\n");
6741         shader_addline(buffer, "MAD texcrd.y, texcrd.y, yv12_coef.y, yv12_coef.x;\n");
6742
6743         /* Read odd lines from the right side(add size * 0.5 to the x coordinate */
6744         shader_addline(buffer, "ADD texcrd2.x, texcrd2.y, yv12_coef.y;\n"); /* To avoid 0.5 == 0.5 comparisons */
6745         shader_addline(buffer, "FRC texcrd2.x, texcrd2.x;\n");
6746         shader_addline(buffer, "SGE texcrd2.x, texcrd2.x, coef.y;\n");
6747         shader_addline(buffer, "MAD texcrd.x, texcrd2.x, coef.y, texcrd.x;\n");
6748
6749         /* clamp, keep the half pixel origin in mind */
6750         shader_addline(buffer, "MAD temp.y, coef.y, chroma.w, yv12_coef.x;\n");
6751         shader_addline(buffer, "MAX texcrd.y, temp.y, texcrd.y;\n");
6752         shader_addline(buffer, "MAD temp.y, -coef.y, chroma.w, yv12_coef.z;\n");
6753         shader_addline(buffer, "MIN texcrd.y, temp.y, texcrd.y;\n");
6754     } else {
6755         /* Read from [size - size+size/4] */
6756         shader_addline(buffer, "FLR texcrd.y, texcrd.y;\n");
6757         shader_addline(buffer, "MAD texcrd.y, texcrd.y, coef.w, size.y;\n");
6758
6759         /* Read odd lines from the right side(add size * 0.5 to the x coordinate */
6760         shader_addline(buffer, "ADD texcrd2.x, texcrd.y, yv12_coef.y;\n"); /* To avoid 0.5 == 0.5 comparisons */
6761         shader_addline(buffer, "FRC texcrd2.x, texcrd2.x;\n");
6762         shader_addline(buffer, "SGE texcrd2.x, texcrd2.x, coef.y;\n");
6763         shader_addline(buffer, "MUL texcrd2.x, texcrd2.x, size.x;\n");
6764         shader_addline(buffer, "MAD texcrd.x, texcrd2.x, coef.y, texcrd.x;\n");
6765
6766         /* Make sure to read exactly from the pixel center */
6767         shader_addline(buffer, "FLR texcrd.y, texcrd.y;\n");
6768         shader_addline(buffer, "ADD texcrd.y, texcrd.y, coef.y;\n");
6769
6770         /* Clamp */
6771         shader_addline(buffer, "MAD temp.y, size.y, coef.w, size.y;\n");
6772         shader_addline(buffer, "ADD temp.y, temp.y, -coef.y;\n");
6773         shader_addline(buffer, "MIN texcrd.y, temp.y, texcrd.y;\n");
6774         shader_addline(buffer, "ADD temp.y, size.y, -coef.y;\n");
6775         shader_addline(buffer, "MAX texcrd.y, temp.y, texcrd.y;\n");
6776     }
6777     /* Read the texture, put the result into the output register */
6778     shader_addline(buffer, "TEX temp, texcrd, texture[0], %s;\n", tex);
6779     shader_addline(buffer, "MOV chroma.x, temp.w;\n");
6780
6781     /* The other chroma value is 1/6th of the texture lower, from 5/6th to 6/6th
6782      * No need to clamp because we're just reusing the already clamped value from above
6783      */
6784     if(textype == GL_TEXTURE_2D) {
6785         shader_addline(buffer, "ADD texcrd.y, texcrd.y, yv12_coef.y;\n");
6786     } else {
6787         shader_addline(buffer, "MAD texcrd.y, size.y, coef.w, texcrd.y;\n");
6788     }
6789     shader_addline(buffer, "TEX temp, texcrd, texture[0], %s;\n", tex);
6790     shader_addline(buffer, "MOV chroma.y, temp.w;\n");
6791
6792     /* Sample the luminance value. It is in the top 2/3rd of the texture, so scale the y coordinate.
6793      * Clamp the y coordinate to prevent the chroma values from bleeding into the sampled luminance
6794      * values due to filtering
6795      */
6796     shader_addline(buffer, "MOV texcrd, fragment.texcoord[0];\n");
6797     if(textype == GL_TEXTURE_2D) {
6798         /* Multiply the y coordinate by 2/3 and clamp it */
6799         shader_addline(buffer, "MUL texcrd.y, texcrd.y, yv12_coef.x;\n");
6800         shader_addline(buffer, "MAD temp.y, -coef.y, chroma.w, yv12_coef.x;\n");
6801         shader_addline(buffer, "MIN texcrd.y, temp.y, texcrd.y;\n");
6802         shader_addline(buffer, "TEX luminance, texcrd, texture[0], %s;\n", tex);
6803     } else {
6804         /* Reading from texture_rectangles is pretty straightforward, just use the unmodified
6805          * texture coordinate. It is still a good idea to clamp it though, since the opengl texture
6806          * is bigger
6807          */
6808         shader_addline(buffer, "ADD temp.x, size.y, -coef.y;\n");
6809         shader_addline(buffer, "MIN texcrd.y, texcrd.y, size.x;\n");
6810         shader_addline(buffer, "TEX luminance, texcrd, texture[0], %s;\n", tex);
6811     }
6812     *luminance = 'a';
6813
6814     return TRUE;
6815 }
6816
6817 static GLuint gen_p8_shader(struct arbfp_blit_priv *priv,
6818         const struct wined3d_gl_info *gl_info, GLenum textype)
6819 {
6820     GLenum shader;
6821     struct wined3d_shader_buffer buffer;
6822     GLint pos;
6823
6824     /* Shader header */
6825     if (!shader_buffer_init(&buffer))
6826     {
6827         ERR("Failed to initialize shader buffer.\n");
6828         return 0;
6829     }
6830
6831     ENTER_GL();
6832     GL_EXTCALL(glGenProgramsARB(1, &shader));
6833     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader));
6834     LEAVE_GL();
6835     if(!shader) {
6836         shader_buffer_free(&buffer);
6837         return 0;
6838     }
6839
6840     shader_addline(&buffer, "!!ARBfp1.0\n");
6841     shader_addline(&buffer, "TEMP index;\n");
6842
6843     /* { 255/256, 0.5/255*255/256, 0, 0 } */
6844     shader_addline(&buffer, "PARAM constants = { 0.996, 0.00195, 0, 0 };\n");
6845
6846     /* The alpha-component contains the palette index */
6847     if(textype == GL_TEXTURE_RECTANGLE_ARB)
6848         shader_addline(&buffer, "TXP index, fragment.texcoord[0], texture[0], RECT;\n");
6849     else
6850         shader_addline(&buffer, "TEX index, fragment.texcoord[0], texture[0], 2D;\n");
6851
6852     /* Scale the index by 255/256 and add a bias of '0.5' in order to sample in the middle */
6853     shader_addline(&buffer, "MAD index.a, index.a, constants.x, constants.y;\n");
6854
6855     /* Use the alpha-component as an index in the palette to get the final color */
6856     shader_addline(&buffer, "TEX result.color, index.a, texture[1], 1D;\n");
6857     shader_addline(&buffer, "END\n");
6858
6859     ENTER_GL();
6860     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
6861             strlen(buffer.buffer), buffer.buffer));
6862     checkGLcall("glProgramStringARB()");
6863
6864     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
6865     if (pos != -1)
6866     {
6867         FIXME("Fragment program error at position %d: %s\n\n", pos,
6868               debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
6869         shader_arb_dump_program_source(buffer.buffer);
6870     }
6871
6872     if (textype == GL_TEXTURE_RECTANGLE_ARB)
6873         priv->p8_rect_shader = shader;
6874     else
6875         priv->p8_2d_shader = shader;
6876
6877     shader_buffer_free(&buffer);
6878     LEAVE_GL();
6879
6880     return shader;
6881 }
6882
6883 /* Context activation is done by the caller. */
6884 static void upload_palette(struct wined3d_surface *surface)
6885 {
6886     BYTE table[256][4];
6887     IWineD3DDeviceImpl *device = surface->resource.device;
6888     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
6889     struct arbfp_blit_priv *priv = device->blit_priv;
6890     BOOL colorkey = (surface->CKeyFlags & WINEDDSD_CKSRCBLT) ? TRUE : FALSE;
6891
6892     d3dfmt_p8_init_palette(surface, table, colorkey);
6893
6894     ENTER_GL();
6895     if (!priv->palette_texture)
6896         glGenTextures(1, &priv->palette_texture);
6897
6898     GL_EXTCALL(glActiveTextureARB(GL_TEXTURE1));
6899     glBindTexture(GL_TEXTURE_1D, priv->palette_texture);
6900
6901     glTexEnvi(GL_TEXTURE_ENV, GL_TEXTURE_ENV_MODE, GL_REPLACE);
6902
6903     glTexParameteri(GL_TEXTURE_1D, GL_TEXTURE_MIN_FILTER, GL_NEAREST);
6904     /* Make sure we have discrete color levels. */
6905     glTexParameteri(GL_TEXTURE_1D, GL_TEXTURE_MAG_FILTER, GL_NEAREST);
6906     glTexParameteri(GL_TEXTURE_1D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
6907     /* Upload the palette */
6908     /* TODO: avoid unneeed uploads in the future by adding some SFLAG_PALETTE_DIRTY mechanism */
6909     glTexImage1D(GL_TEXTURE_1D, 0, GL_RGBA, 256, 0, GL_RGBA, GL_UNSIGNED_BYTE, table);
6910
6911     /* Switch back to unit 0 in which the 2D texture will be stored. */
6912     GL_EXTCALL(glActiveTextureARB(GL_TEXTURE0));
6913     LEAVE_GL();
6914 }
6915
6916 /* Context activation is done by the caller. */
6917 static GLuint gen_yuv_shader(struct arbfp_blit_priv *priv, const struct wined3d_gl_info *gl_info,
6918         enum complex_fixup yuv_fixup, GLenum textype)
6919 {
6920     GLenum shader;
6921     struct wined3d_shader_buffer buffer;
6922     char luminance_component;
6923     GLint pos;
6924
6925     /* Shader header */
6926     if (!shader_buffer_init(&buffer))
6927     {
6928         ERR("Failed to initialize shader buffer.\n");
6929         return 0;
6930     }
6931
6932     ENTER_GL();
6933     GL_EXTCALL(glGenProgramsARB(1, &shader));
6934     checkGLcall("GL_EXTCALL(glGenProgramsARB(1, &shader))");
6935     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader));
6936     checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader)");
6937     LEAVE_GL();
6938     if(!shader) {
6939         shader_buffer_free(&buffer);
6940         return 0;
6941     }
6942
6943     /* The YUY2 and UYVY formats contain two pixels packed into a 32 bit macropixel,
6944      * giving effectively 16 bit per pixel. The color consists of a luminance(Y) and
6945      * two chroma(U and V) values. Each macropixel has two luminance values, one for
6946      * each single pixel it contains, and one U and one V value shared between both
6947      * pixels.
6948      *
6949      * The data is loaded into an A8L8 texture. With YUY2, the luminance component
6950      * contains the luminance and alpha the chroma. With UYVY it is vice versa. Thus
6951      * take the format into account when generating the read swizzles
6952      *
6953      * Reading the Y value is straightforward - just sample the texture. The hardware
6954      * takes care of filtering in the horizontal and vertical direction.
6955      *
6956      * Reading the U and V values is harder. We have to avoid filtering horizontally,
6957      * because that would mix the U and V values of one pixel or two adjacent pixels.
6958      * Thus floor the texture coordinate and add 0.5 to get an unfiltered read,
6959      * regardless of the filtering setting. Vertical filtering works automatically
6960      * though - the U and V values of two rows are mixed nicely.
6961      *
6962      * Appart of avoiding filtering issues, the code has to know which value it just
6963      * read, and where it can find the other one. To determine this, it checks if
6964      * it sampled an even or odd pixel, and shifts the 2nd read accordingly.
6965      *
6966      * Handling horizontal filtering of U and V values requires reading a 2nd pair
6967      * of pixels, extracting U and V and mixing them. This is not implemented yet.
6968      *
6969      * An alternative implementation idea is to load the texture as A8R8G8B8 texture,
6970      * with width / 2. This way one read gives all 3 values, finding U and V is easy
6971      * in an unfiltered situation. Finding the luminance on the other hand requires
6972      * finding out if it is an odd or even pixel. The real drawback of this approach
6973      * is filtering. This would have to be emulated completely in the shader, reading
6974      * up two 2 packed pixels in up to 2 rows and interpolating both horizontally and
6975      * vertically. Beyond that it would require adjustments to the texture handling
6976      * code to deal with the width scaling
6977      */
6978     shader_addline(&buffer, "!!ARBfp1.0\n");
6979     shader_addline(&buffer, "TEMP luminance;\n");
6980     shader_addline(&buffer, "TEMP temp;\n");
6981     shader_addline(&buffer, "TEMP chroma;\n");
6982     shader_addline(&buffer, "TEMP texcrd;\n");
6983     shader_addline(&buffer, "TEMP texcrd2;\n");
6984     shader_addline(&buffer, "PARAM coef = {1.0, 0.5, 2.0, 0.25};\n");
6985     shader_addline(&buffer, "PARAM yuv_coef = {1.403, 0.344, 0.714, 1.770};\n");
6986     shader_addline(&buffer, "PARAM size = program.local[0];\n");
6987
6988     switch (yuv_fixup)
6989     {
6990         case COMPLEX_FIXUP_UYVY:
6991         case COMPLEX_FIXUP_YUY2:
6992             if (!gen_planar_yuv_read(&buffer, yuv_fixup, textype, &luminance_component))
6993             {
6994                 shader_buffer_free(&buffer);
6995                 return 0;
6996             }
6997             break;
6998
6999         case COMPLEX_FIXUP_YV12:
7000             if (!gen_yv12_read(&buffer, textype, &luminance_component))
7001             {
7002                 shader_buffer_free(&buffer);
7003                 return 0;
7004             }
7005             break;
7006
7007         default:
7008             FIXME("Unsupported YUV fixup %#x\n", yuv_fixup);
7009             shader_buffer_free(&buffer);
7010             return 0;
7011     }
7012
7013     /* Calculate the final result. Formula is taken from
7014      * http://www.fourcc.org/fccyvrgb.php. Note that the chroma
7015      * ranges from -0.5 to 0.5
7016      */
7017     shader_addline(&buffer, "SUB chroma.xy, chroma, coef.y;\n");
7018
7019     shader_addline(&buffer, "MAD result.color.x, chroma.x, yuv_coef.x, luminance.%c;\n", luminance_component);
7020     shader_addline(&buffer, "MAD temp.x, -chroma.y, yuv_coef.y, luminance.%c;\n", luminance_component);
7021     shader_addline(&buffer, "MAD result.color.y, -chroma.x, yuv_coef.z, temp.x;\n");
7022     shader_addline(&buffer, "MAD result.color.z, chroma.y, yuv_coef.w, luminance.%c;\n", luminance_component);
7023     shader_addline(&buffer, "END\n");
7024
7025     ENTER_GL();
7026     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
7027             strlen(buffer.buffer), buffer.buffer));
7028     checkGLcall("glProgramStringARB()");
7029
7030     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
7031     if (pos != -1)
7032     {
7033         FIXME("Fragment program error at position %d: %s\n\n", pos,
7034               debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
7035         shader_arb_dump_program_source(buffer.buffer);
7036     }
7037     else
7038     {
7039         GLint native;
7040
7041         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
7042         checkGLcall("glGetProgramivARB()");
7043         if (!native) WARN("Program exceeds native resource limits.\n");
7044     }
7045
7046     shader_buffer_free(&buffer);
7047     LEAVE_GL();
7048
7049     switch (yuv_fixup)
7050     {
7051         case COMPLEX_FIXUP_YUY2:
7052             if (textype == GL_TEXTURE_RECTANGLE_ARB) priv->yuy2_rect_shader = shader;
7053             else priv->yuy2_2d_shader = shader;
7054             break;
7055
7056         case COMPLEX_FIXUP_UYVY:
7057             if (textype == GL_TEXTURE_RECTANGLE_ARB) priv->uyvy_rect_shader = shader;
7058             else priv->uyvy_2d_shader = shader;
7059             break;
7060
7061         case COMPLEX_FIXUP_YV12:
7062             if (textype == GL_TEXTURE_RECTANGLE_ARB) priv->yv12_rect_shader = shader;
7063             else priv->yv12_2d_shader = shader;
7064             break;
7065         default:
7066             ERR("Unsupported complex fixup: %d\n", yuv_fixup);
7067     }
7068
7069     return shader;
7070 }
7071
7072 /* Context activation is done by the caller. */
7073 static HRESULT arbfp_blit_set(void *blit_priv, const struct wined3d_gl_info *gl_info, struct wined3d_surface *surface)
7074 {
7075     GLenum shader;
7076     float size[4] = {(float) surface->pow2Width, (float) surface->pow2Height, 1.0f, 1.0f};
7077     struct arbfp_blit_priv *priv = blit_priv;
7078     enum complex_fixup fixup;
7079     GLenum textype = surface->texture_target;
7080
7081     if (!is_complex_fixup(surface->resource.format->color_fixup))
7082     {
7083         TRACE("Fixup:\n");
7084         dump_color_fixup_desc(surface->resource.format->color_fixup);
7085         /* Don't bother setting up a shader for unconverted formats */
7086         ENTER_GL();
7087         glEnable(textype);
7088         checkGLcall("glEnable(textype)");
7089         LEAVE_GL();
7090         return WINED3D_OK;
7091     }
7092
7093     fixup = get_complex_fixup(surface->resource.format->color_fixup);
7094
7095     switch(fixup)
7096     {
7097         case COMPLEX_FIXUP_YUY2:
7098             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->yuy2_rect_shader : priv->yuy2_2d_shader;
7099             break;
7100
7101         case COMPLEX_FIXUP_UYVY:
7102             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->uyvy_rect_shader : priv->uyvy_2d_shader;
7103             break;
7104
7105         case COMPLEX_FIXUP_YV12:
7106             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->yv12_rect_shader : priv->yv12_2d_shader;
7107             break;
7108
7109         case COMPLEX_FIXUP_P8:
7110             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->p8_rect_shader : priv->p8_2d_shader;
7111             if (!shader) shader = gen_p8_shader(priv, gl_info, textype);
7112
7113             upload_palette(surface);
7114             break;
7115
7116         default:
7117             FIXME("Unsupported complex fixup %#x, not setting a shader\n", fixup);
7118             ENTER_GL();
7119             glEnable(textype);
7120             checkGLcall("glEnable(textype)");
7121             LEAVE_GL();
7122             return E_NOTIMPL;
7123     }
7124
7125     if (!shader) shader = gen_yuv_shader(priv, gl_info, fixup, textype);
7126
7127     ENTER_GL();
7128     glEnable(GL_FRAGMENT_PROGRAM_ARB);
7129     checkGLcall("glEnable(GL_FRAGMENT_PROGRAM_ARB)");
7130     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader));
7131     checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader)");
7132     GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, 0, size));
7133     checkGLcall("glProgramLocalParameter4fvARB");
7134     LEAVE_GL();
7135
7136     return WINED3D_OK;
7137 }
7138
7139 /* Context activation is done by the caller. */
7140 static void arbfp_blit_unset(const struct wined3d_gl_info *gl_info)
7141 {
7142     ENTER_GL();
7143     glDisable(GL_FRAGMENT_PROGRAM_ARB);
7144     checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
7145     glDisable(GL_TEXTURE_2D);
7146     checkGLcall("glDisable(GL_TEXTURE_2D)");
7147     if (gl_info->supported[ARB_TEXTURE_CUBE_MAP])
7148     {
7149         glDisable(GL_TEXTURE_CUBE_MAP_ARB);
7150         checkGLcall("glDisable(GL_TEXTURE_CUBE_MAP_ARB)");
7151     }
7152     if (gl_info->supported[ARB_TEXTURE_RECTANGLE])
7153     {
7154         glDisable(GL_TEXTURE_RECTANGLE_ARB);
7155         checkGLcall("glDisable(GL_TEXTURE_RECTANGLE_ARB)");
7156     }
7157     LEAVE_GL();
7158 }
7159
7160 static BOOL arbfp_blit_supported(const struct wined3d_gl_info *gl_info, enum wined3d_blit_op blit_op,
7161         const RECT *src_rect, DWORD src_usage, WINED3DPOOL src_pool, const struct wined3d_format *src_format,
7162         const RECT *dst_rect, DWORD dst_usage, WINED3DPOOL dst_pool, const struct wined3d_format *dst_format)
7163 {
7164     enum complex_fixup src_fixup;
7165
7166     if (!gl_info->supported[ARB_FRAGMENT_PROGRAM])
7167         return FALSE;
7168
7169     if (blit_op != WINED3D_BLIT_OP_COLOR_BLIT)
7170     {
7171         TRACE("Unsupported blit_op=%d\n", blit_op);
7172         return FALSE;
7173     }
7174
7175     src_fixup = get_complex_fixup(src_format->color_fixup);
7176     if (TRACE_ON(d3d_shader) && TRACE_ON(d3d))
7177     {
7178         TRACE("Checking support for fixup:\n");
7179         dump_color_fixup_desc(src_format->color_fixup);
7180     }
7181
7182     if (!is_identity_fixup(dst_format->color_fixup))
7183     {
7184         TRACE("Destination fixups are not supported\n");
7185         return FALSE;
7186     }
7187
7188     if (is_identity_fixup(src_format->color_fixup))
7189     {
7190         TRACE("[OK]\n");
7191         return TRUE;
7192     }
7193
7194      /* We only support YUV conversions. */
7195     if (!is_complex_fixup(src_format->color_fixup))
7196     {
7197         TRACE("[FAILED]\n");
7198         return FALSE;
7199     }
7200
7201     switch(src_fixup)
7202     {
7203         case COMPLEX_FIXUP_YUY2:
7204         case COMPLEX_FIXUP_UYVY:
7205         case COMPLEX_FIXUP_YV12:
7206         case COMPLEX_FIXUP_P8:
7207             TRACE("[OK]\n");
7208             return TRUE;
7209
7210         default:
7211             FIXME("Unsupported YUV fixup %#x\n", src_fixup);
7212             TRACE("[FAILED]\n");
7213             return FALSE;
7214     }
7215 }
7216
7217 HRESULT arbfp_blit_surface(IWineD3DDeviceImpl *device, struct wined3d_surface *src_surface, const RECT *src_rect,
7218         struct wined3d_surface *dst_surface, const RECT *dst_rect_in, enum wined3d_blit_op blit_op, DWORD Filter)
7219 {
7220     struct wined3d_context *context;
7221     RECT dst_rect = *dst_rect_in;
7222
7223     /* Now load the surface */
7224     surface_internal_preload(src_surface, SRGB_RGB);
7225
7226     /* Activate the destination context, set it up for blitting */
7227     context = context_acquire(device, dst_surface);
7228     context_apply_blit_state(context, device);
7229
7230     if (!surface_is_offscreen(dst_surface))
7231         surface_translate_drawable_coords(dst_surface, context->win_handle, &dst_rect);
7232
7233     arbfp_blit_set(device->blit_priv, context->gl_info, src_surface);
7234
7235     ENTER_GL();
7236
7237     /* Draw a textured quad */
7238     draw_textured_quad(src_surface, src_rect, &dst_rect, Filter);
7239
7240     LEAVE_GL();
7241
7242     /* Leave the opengl state valid for blitting */
7243     arbfp_blit_unset(context->gl_info);
7244
7245     if (wined3d_settings.strict_draw_ordering
7246             || (dst_surface->container.type == WINED3D_CONTAINER_SWAPCHAIN
7247             && (dst_surface->container.u.swapchain->front_buffer == dst_surface)))
7248         wglFlush(); /* Flush to ensure ordering across contexts. */
7249
7250     context_release(context);
7251
7252     surface_modify_location(dst_surface, SFLAG_INDRAWABLE, TRUE);
7253     return WINED3D_OK;
7254 }
7255
7256 /* Do not call while under the GL lock. */
7257 static HRESULT arbfp_blit_color_fill(IWineD3DDeviceImpl *device, struct wined3d_surface *dst_surface,
7258         const RECT *dst_rect, const WINED3DCOLORVALUE *color)
7259 {
7260     FIXME("Color filling not implemented by arbfp_blit\n");
7261     return WINED3DERR_INVALIDCALL;
7262 }
7263
7264 /* Do not call while under the GL lock. */
7265 static HRESULT arbfp_blit_depth_fill(IWineD3DDeviceImpl *device,
7266         struct wined3d_surface *surface, const RECT *rect, float depth)
7267 {
7268     FIXME("Depth filling not implemented by arbfp_blit.\n");
7269     return WINED3DERR_INVALIDCALL;
7270 }
7271
7272 const struct blit_shader arbfp_blit = {
7273     arbfp_blit_alloc,
7274     arbfp_blit_free,
7275     arbfp_blit_set,
7276     arbfp_blit_unset,
7277     arbfp_blit_supported,
7278     arbfp_blit_color_fill,
7279     arbfp_blit_depth_fill,
7280 };