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, UINT 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 : (float) 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     struct wined3d_device *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         UINT rt_height = device->fb.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(struct wined3d_device *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(struct wined3d_device *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     const 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->rt_mask > 1)
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_shader *shader = ins->ctx->shader;
1341     const struct wined3d_texture *texture;
1342     const char *tex_type;
1343     BOOL np2_fixup = FALSE;
1344     struct wined3d_device *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
1592     shader_arb_get_dst_param(ins, dst, dst_name);
1593     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1594
1595     /* The coissue flag changes the semantic of the cnd instruction in <= 1.3 shaders */
1596     if (shader_version <= WINED3D_SHADER_VERSION(1, 3) && ins->coissue)
1597     {
1598         shader_addline(buffer, "MOV%s %s, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name[1]);
1599     }
1600     else
1601     {
1602         struct wined3d_shader_src_param src0_copy = ins->src[0];
1603         char extra_neg;
1604
1605         /* src0 may have a negate srcmod set, so we can't blindly add "-" to the name */
1606         src0_copy.modifiers = negate_modifiers(src0_copy.modifiers, &extra_neg);
1607
1608         shader_arb_get_src_param(ins, &src0_copy, 0, src_name[0]);
1609         shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
1610         shader_addline(buffer, "ADD TA, %c%s, coefdiv.x;\n", extra_neg, src_name[0]);
1611         shader_addline(buffer, "CMP%s %s, TA, %s, %s;\n", shader_arb_get_modifier(ins),
1612                 dst_name, src_name[1], src_name[2]);
1613     }
1614 }
1615
1616 static void pshader_hw_cmp(const struct wined3d_shader_instruction *ins)
1617 {
1618     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1619     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1620     char dst_name[50];
1621     char src_name[3][50];
1622
1623     shader_arb_get_dst_param(ins, dst, dst_name);
1624
1625     /* Generate input register names (with modifiers) */
1626     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
1627     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1628     shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
1629
1630     shader_addline(buffer, "CMP%s %s, %s, %s, %s;\n", shader_arb_get_modifier(ins),
1631             dst_name, src_name[0], src_name[2], src_name[1]);
1632 }
1633
1634 /** Process the WINED3DSIO_DP2ADD instruction in ARB.
1635  * dst = dot2(src0, src1) + src2 */
1636 static void pshader_hw_dp2add(const struct wined3d_shader_instruction *ins)
1637 {
1638     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1639     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1640     char dst_name[50];
1641     char src_name[3][50];
1642     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
1643
1644     shader_arb_get_dst_param(ins, dst, dst_name);
1645     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
1646     shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
1647
1648     if(ctx->target_version >= NV3)
1649     {
1650         /* GL_NV_fragment_program2 has a 1:1 matching instruction */
1651         shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1652         shader_addline(buffer, "DP2A%s %s, %s, %s, %s;\n", shader_arb_get_modifier(ins),
1653                        dst_name, src_name[0], src_name[1], src_name[2]);
1654     }
1655     else if(ctx->target_version >= NV2)
1656     {
1657         /* dst.x = src2.?, src0.x, src1.x + src0.y * src1.y
1658          * dst.y = src2.?, src0.x, src1.z + src0.y * src1.w
1659          * dst.z = src2.?, src0.x, src1.x + src0.y * src1.y
1660          * dst.z = src2.?, src0.x, src1.z + src0.y * src1.w
1661          *
1662          * Make sure that src1.zw = src1.xy, then we get a classic dp2add
1663          *
1664          * .xyxy and other swizzles that we could get with this are not valid in
1665          * plain ARBfp, but luckily the NV extension grammar lifts this limitation.
1666          */
1667         struct wined3d_shader_src_param tmp_param = ins->src[1];
1668         DWORD swizzle = tmp_param.swizzle & 0xf; /* Selects .xy */
1669         tmp_param.swizzle = swizzle | (swizzle << 4); /* Creates .xyxy */
1670
1671         shader_arb_get_src_param(ins, &tmp_param, 1, src_name[1]);
1672
1673         shader_addline(buffer, "X2D%s %s, %s, %s, %s;\n", shader_arb_get_modifier(ins),
1674                        dst_name, src_name[2], src_name[0], src_name[1]);
1675     }
1676     else
1677     {
1678         shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1679         /* 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]
1680         * src_name[0] can be TA, but TA is a private temp for modifiers, so it is save to overwrite
1681         */
1682         shader_addline(buffer, "MOV TA, %s;\n", src_name[0]);
1683         shader_addline(buffer, "MOV TA.z, 0.0;\n");
1684         shader_addline(buffer, "DP3 TA, TA, %s;\n", src_name[1]);
1685         shader_addline(buffer, "ADD%s %s, TA, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name[2]);
1686     }
1687 }
1688
1689 /* Map the opcode 1-to-1 to the GL code */
1690 static void shader_hw_map2gl(const struct wined3d_shader_instruction *ins)
1691 {
1692     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1693     const char *instruction;
1694     char arguments[256], dst_str[50];
1695     unsigned int i;
1696     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1697
1698     switch (ins->handler_idx)
1699     {
1700         case WINED3DSIH_ABS: instruction = "ABS"; break;
1701         case WINED3DSIH_ADD: instruction = "ADD"; break;
1702         case WINED3DSIH_CRS: instruction = "XPD"; break;
1703         case WINED3DSIH_DP3: instruction = "DP3"; break;
1704         case WINED3DSIH_DP4: instruction = "DP4"; break;
1705         case WINED3DSIH_DST: instruction = "DST"; break;
1706         case WINED3DSIH_FRC: instruction = "FRC"; break;
1707         case WINED3DSIH_LIT: instruction = "LIT"; break;
1708         case WINED3DSIH_LRP: instruction = "LRP"; break;
1709         case WINED3DSIH_MAD: instruction = "MAD"; break;
1710         case WINED3DSIH_MAX: instruction = "MAX"; break;
1711         case WINED3DSIH_MIN: instruction = "MIN"; break;
1712         case WINED3DSIH_MOV: instruction = "MOV"; break;
1713         case WINED3DSIH_MUL: instruction = "MUL"; break;
1714         case WINED3DSIH_SGE: instruction = "SGE"; break;
1715         case WINED3DSIH_SLT: instruction = "SLT"; break;
1716         case WINED3DSIH_SUB: instruction = "SUB"; break;
1717         case WINED3DSIH_MOVA:instruction = "ARR"; break;
1718         case WINED3DSIH_DSX: instruction = "DDX"; break;
1719         default: instruction = "";
1720             FIXME("Unhandled opcode %#x\n", ins->handler_idx);
1721             break;
1722     }
1723
1724     /* Note that shader_arb_add_dst_param() adds spaces. */
1725     arguments[0] = '\0';
1726     shader_arb_get_dst_param(ins, dst, dst_str);
1727     for (i = 0; i < ins->src_count; ++i)
1728     {
1729         char operand[100];
1730         strcat(arguments, ", ");
1731         shader_arb_get_src_param(ins, &ins->src[i], i, operand);
1732         strcat(arguments, operand);
1733     }
1734     shader_addline(buffer, "%s%s %s%s;\n", instruction, shader_arb_get_modifier(ins), dst_str, arguments);
1735 }
1736
1737 static void shader_hw_nop(const struct wined3d_shader_instruction *ins)
1738 {
1739     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1740     shader_addline(buffer, "NOP;\n");
1741 }
1742
1743 static void shader_hw_mov(const struct wined3d_shader_instruction *ins)
1744 {
1745     const struct wined3d_shader *shader = ins->ctx->shader;
1746     const struct wined3d_shader_reg_maps *reg_maps = ins->ctx->reg_maps;
1747     BOOL pshader = shader_is_pshader_version(reg_maps->shader_version.type);
1748     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
1749     const char *zero = arb_get_helper_value(reg_maps->shader_version.type, ARB_ZERO);
1750     const char *one = arb_get_helper_value(reg_maps->shader_version.type, ARB_ONE);
1751     const char *two = arb_get_helper_value(reg_maps->shader_version.type, ARB_TWO);
1752
1753     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1754     char src0_param[256];
1755
1756     if (ins->handler_idx == WINED3DSIH_MOVA)
1757     {
1758         const struct arb_vshader_private *shader_data = shader->backend_data;
1759         char write_mask[6];
1760         const char *offset = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_VS_REL_OFFSET);
1761
1762         if(ctx->target_version >= NV2) {
1763             shader_hw_map2gl(ins);
1764             return;
1765         }
1766         shader_arb_get_src_param(ins, &ins->src[0], 0, src0_param);
1767         shader_arb_get_write_mask(ins, &ins->dst[0], write_mask);
1768
1769         /* This implements the mova formula used in GLSL. The first two instructions
1770          * prepare the sign() part. Note that it is fine to have my_sign(0.0) = 1.0
1771          * in this case:
1772          * mova A0.x, 0.0
1773          *
1774          * A0.x = arl(floor(abs(0.0) + 0.5) * 1.0) = floor(0.5) = 0.0 since arl does a floor
1775          *
1776          * The ARL is performed when A0 is used - the requested component is read from A0_SHADOW into
1777          * A0.x. We can use the overwritten component of A0_shadow as temporary storage for the sign.
1778          */
1779         shader_addline(buffer, "SGE A0_SHADOW%s, %s, %s;\n", write_mask, src0_param, zero);
1780         shader_addline(buffer, "MAD A0_SHADOW%s, A0_SHADOW, %s, -%s;\n", write_mask, two, one);
1781
1782         shader_addline(buffer, "ABS TA%s, %s;\n", write_mask, src0_param);
1783         shader_addline(buffer, "ADD TA%s, TA, rel_addr_const.x;\n", write_mask);
1784         shader_addline(buffer, "FLR TA%s, TA;\n", write_mask);
1785         if (shader_data->rel_offset)
1786         {
1787             shader_addline(buffer, "ADD TA%s, TA, %s;\n", write_mask, offset);
1788         }
1789         shader_addline(buffer, "MUL A0_SHADOW%s, TA, A0_SHADOW;\n", write_mask);
1790
1791         ((struct shader_arb_ctx_priv *)ins->ctx->backend_data)->addr_reg[0] = '\0';
1792     }
1793     else if (reg_maps->shader_version.major == 1
1794           && !shader_is_pshader_version(reg_maps->shader_version.type)
1795           && ins->dst[0].reg.type == WINED3DSPR_ADDR)
1796     {
1797         const struct arb_vshader_private *shader_data = shader->backend_data;
1798         src0_param[0] = '\0';
1799
1800         if (shader_data->rel_offset)
1801         {
1802             const char *offset = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_VS_REL_OFFSET);
1803             shader_arb_get_src_param(ins, &ins->src[0], 0, src0_param);
1804             shader_addline(buffer, "ADD TA.x, %s, %s;\n", src0_param, offset);
1805             shader_addline(buffer, "ARL A0.x, TA.x;\n");
1806         }
1807         else
1808         {
1809             /* Apple's ARB_vertex_program implementation does not accept an ARL source argument
1810              * with more than one component. Thus replicate the first source argument over all
1811              * 4 components. For example, .xyzw -> .x (or better: .xxxx), .zwxy -> .z, etc) */
1812             struct wined3d_shader_src_param tmp_src = ins->src[0];
1813             tmp_src.swizzle = (tmp_src.swizzle & 0x3) * 0x55;
1814             shader_arb_get_src_param(ins, &tmp_src, 0, src0_param);
1815             shader_addline(buffer, "ARL A0.x, %s;\n", src0_param);
1816         }
1817     }
1818     else if (ins->dst[0].reg.type == WINED3DSPR_COLOROUT && !ins->dst[0].reg.idx && pshader)
1819     {
1820         if (ctx->cur_ps_args->super.srgb_correction && shader->u.ps.color0_mov)
1821         {
1822             shader_addline(buffer, "#mov handled in srgb write code\n");
1823             return;
1824         }
1825         shader_hw_map2gl(ins);
1826     }
1827     else
1828     {
1829         shader_hw_map2gl(ins);
1830     }
1831 }
1832
1833 static void pshader_hw_texkill(const struct wined3d_shader_instruction *ins)
1834 {
1835     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1836     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1837     char reg_dest[40];
1838
1839     /* No swizzles are allowed in d3d's texkill. PS 1.x ignores the 4th component as documented,
1840      * but >= 2.0 honors it(undocumented, but tested by the d3d9 testsuit)
1841      */
1842     shader_arb_get_dst_param(ins, dst, reg_dest);
1843
1844     if (ins->ctx->reg_maps->shader_version.major >= 2)
1845     {
1846         const char *kilsrc = "TA";
1847         BOOL is_color;
1848
1849         shader_arb_get_register_name(ins, &dst->reg, reg_dest, &is_color);
1850         if(dst->write_mask == WINED3DSP_WRITEMASK_ALL)
1851         {
1852             kilsrc = reg_dest;
1853         }
1854         else
1855         {
1856             /* Sigh. KIL doesn't support swizzles/writemasks. KIL passes a writemask, but ".xy" for example
1857              * is not valid as a swizzle in ARB (needs ".xyyy"). Use SWZ to load the register properly, and set
1858              * masked out components to 0(won't kill)
1859              */
1860             char x = '0', y = '0', z = '0', w = '0';
1861             if(dst->write_mask & WINED3DSP_WRITEMASK_0) x = 'x';
1862             if(dst->write_mask & WINED3DSP_WRITEMASK_1) y = 'y';
1863             if(dst->write_mask & WINED3DSP_WRITEMASK_2) z = 'z';
1864             if(dst->write_mask & WINED3DSP_WRITEMASK_3) w = 'w';
1865             shader_addline(buffer, "SWZ TA, %s, %c, %c, %c, %c;\n", reg_dest, x, y, z, w);
1866         }
1867         shader_addline(buffer, "KIL %s;\n", kilsrc);
1868     } else {
1869         /* ARB fp doesn't like swizzles on the parameter of the KIL instruction. To mask the 4th component,
1870          * copy the register into our general purpose TMP variable, overwrite .w and pass TMP to KIL
1871          *
1872          * ps_1_3 shaders use the texcoord incarnation of the Tx register. ps_1_4 shaders can use the same,
1873          * or pass in any temporary register(in shader phase 2)
1874          */
1875         if(ins->ctx->reg_maps->shader_version.minor <= 3) {
1876             sprintf(reg_dest, "fragment.texcoord[%u]", dst->reg.idx);
1877         } else {
1878             shader_arb_get_dst_param(ins, dst, reg_dest);
1879         }
1880         shader_addline(buffer, "SWZ TA, %s, x, y, z, 1;\n", reg_dest);
1881         shader_addline(buffer, "KIL TA;\n");
1882     }
1883 }
1884
1885 static void pshader_hw_tex(const struct wined3d_shader_instruction *ins)
1886 {
1887     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
1888     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1889     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
1890             ins->ctx->reg_maps->shader_version.minor);
1891     struct wined3d_shader_src_param src;
1892
1893     char reg_dest[40];
1894     char reg_coord[40];
1895     DWORD reg_sampler_code;
1896     WORD myflags = 0;
1897
1898     /* All versions have a destination register */
1899     shader_arb_get_dst_param(ins, dst, reg_dest);
1900
1901     /* 1.0-1.4: Use destination register number as texture code.
1902        2.0+: Use provided sampler number as texure code. */
1903     if (shader_version < WINED3D_SHADER_VERSION(2,0))
1904         reg_sampler_code = dst->reg.idx;
1905     else
1906         reg_sampler_code = ins->src[1].reg.idx;
1907
1908     /* 1.0-1.3: Use the texcoord varying.
1909        1.4+: Use provided coordinate source register. */
1910     if (shader_version < WINED3D_SHADER_VERSION(1,4))
1911         sprintf(reg_coord, "fragment.texcoord[%u]", reg_sampler_code);
1912     else {
1913         /* TEX is the only instruction that can handle DW and DZ natively */
1914         src = ins->src[0];
1915         if(src.modifiers == WINED3DSPSM_DW) src.modifiers = WINED3DSPSM_NONE;
1916         if(src.modifiers == WINED3DSPSM_DZ) src.modifiers = WINED3DSPSM_NONE;
1917         shader_arb_get_src_param(ins, &src, 0, reg_coord);
1918     }
1919
1920     /* projection flag:
1921      * 1.1, 1.2, 1.3: Use WINED3DTSS_TEXTURETRANSFORMFLAGS
1922      * 1.4: Use WINED3DSPSM_DZ or WINED3DSPSM_DW on src[0]
1923      * 2.0+: Use WINED3DSI_TEXLD_PROJECT on the opcode
1924      */
1925     if (shader_version < WINED3D_SHADER_VERSION(1,4))
1926     {
1927         DWORD flags = 0;
1928         if (reg_sampler_code < MAX_TEXTURES)
1929             flags = priv->cur_ps_args->super.tex_transform >> reg_sampler_code * WINED3D_PSARGS_TEXTRANSFORM_SHIFT;
1930         if (flags & WINED3D_PSARGS_PROJECTED)
1931             myflags |= TEX_PROJ;
1932     }
1933     else if (shader_version < WINED3D_SHADER_VERSION(2,0))
1934     {
1935         DWORD src_mod = ins->src[0].modifiers;
1936         if (src_mod == WINED3DSPSM_DZ) {
1937             /* TXP cannot handle DZ natively, so move the z coordinate to .w. reg_coord is a read-only
1938              * varying register, so we need a temp reg
1939              */
1940             shader_addline(ins->ctx->buffer, "SWZ TA, %s, x, y, z, z;\n", reg_coord);
1941             strcpy(reg_coord, "TA");
1942             myflags |= TEX_PROJ;
1943         } else if(src_mod == WINED3DSPSM_DW) {
1944             myflags |= TEX_PROJ;
1945         }
1946     } else {
1947         if (ins->flags & WINED3DSI_TEXLD_PROJECT) myflags |= TEX_PROJ;
1948         if (ins->flags & WINED3DSI_TEXLD_BIAS) myflags |= TEX_BIAS;
1949     }
1950     shader_hw_sample(ins, reg_sampler_code, reg_dest, reg_coord, myflags, NULL, NULL);
1951 }
1952
1953 static void pshader_hw_texcoord(const struct wined3d_shader_instruction *ins)
1954 {
1955     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1956     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1957     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
1958             ins->ctx->reg_maps->shader_version.minor);
1959     char dst_str[50];
1960
1961     if (shader_version < WINED3D_SHADER_VERSION(1,4))
1962     {
1963         DWORD reg = dst->reg.idx;
1964
1965         shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
1966         shader_addline(buffer, "MOV_SAT %s, fragment.texcoord[%u];\n", dst_str, reg);
1967     } else {
1968         char reg_src[40];
1969
1970         shader_arb_get_src_param(ins, &ins->src[0], 0, reg_src);
1971         shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
1972         shader_addline(buffer, "MOV %s, %s;\n", dst_str, reg_src);
1973    }
1974 }
1975
1976 static void pshader_hw_texreg2ar(const struct wined3d_shader_instruction *ins)
1977 {
1978      struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1979      DWORD flags = 0;
1980
1981      DWORD reg1 = ins->dst[0].reg.idx;
1982      char dst_str[50];
1983      char src_str[50];
1984
1985      /* Note that texreg2ar treats Tx as a temporary register, not as a varying */
1986      shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
1987      shader_arb_get_src_param(ins, &ins->src[0], 0, src_str);
1988      /* Move .x first in case src_str is "TA" */
1989      shader_addline(buffer, "MOV TA.y, %s.x;\n", src_str);
1990      shader_addline(buffer, "MOV TA.x, %s.w;\n", src_str);
1991      if (reg1 < MAX_TEXTURES)
1992      {
1993          struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
1994          flags = priv->cur_ps_args->super.tex_transform >> reg1 * WINED3D_PSARGS_TEXTRANSFORM_SHIFT;
1995      }
1996      shader_hw_sample(ins, reg1, dst_str, "TA", flags & WINED3D_PSARGS_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
1997 }
1998
1999 static void pshader_hw_texreg2gb(const struct wined3d_shader_instruction *ins)
2000 {
2001      struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2002
2003      DWORD reg1 = ins->dst[0].reg.idx;
2004      char dst_str[50];
2005      char src_str[50];
2006
2007      /* Note that texreg2gb treats Tx as a temporary register, not as a varying */
2008      shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2009      shader_arb_get_src_param(ins, &ins->src[0], 0, src_str);
2010      shader_addline(buffer, "MOV TA.x, %s.y;\n", src_str);
2011      shader_addline(buffer, "MOV TA.y, %s.z;\n", src_str);
2012      shader_hw_sample(ins, reg1, dst_str, "TA", 0, NULL, NULL);
2013 }
2014
2015 static void pshader_hw_texreg2rgb(const struct wined3d_shader_instruction *ins)
2016 {
2017     DWORD reg1 = ins->dst[0].reg.idx;
2018     char dst_str[50];
2019     char src_str[50];
2020
2021     /* Note that texreg2rg 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_hw_sample(ins, reg1, dst_str, src_str, 0, NULL, NULL);
2025 }
2026
2027 static void pshader_hw_texbem(const struct wined3d_shader_instruction *ins)
2028 {
2029     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2030     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2031     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2032     char reg_coord[40], dst_reg[50], src_reg[50];
2033     DWORD reg_dest_code;
2034
2035     /* All versions have a destination register. The Tx where the texture coordinates come
2036      * from is the varying incarnation of the texture register
2037      */
2038     reg_dest_code = dst->reg.idx;
2039     shader_arb_get_dst_param(ins, &ins->dst[0], dst_reg);
2040     shader_arb_get_src_param(ins, &ins->src[0], 0, src_reg);
2041     sprintf(reg_coord, "fragment.texcoord[%u]", reg_dest_code);
2042
2043     /* Sampling the perturbation map in Tsrc was done already, including the signedness correction if needed
2044      * The Tx in which the perturbation map is stored is the tempreg incarnation of the texture register
2045      *
2046      * GL_NV_fragment_program_option could handle this in one instruction via X2D:
2047      * X2D TA.xy, fragment.texcoord, T%u, bumpenvmat%u.xzyw
2048      *
2049      * However, the NV extensions are never enabled for <= 2.0 shaders because of the performance penalty that
2050      * comes with it, and texbem is an 1.x only instruction. No 1.x instruction forces us to enable the NV
2051      * extension.
2052      */
2053     shader_addline(buffer, "SWZ TB, bumpenvmat%d, x, z, 0, 0;\n", reg_dest_code);
2054     shader_addline(buffer, "DP3 TA.x, TB, %s;\n", src_reg);
2055     shader_addline(buffer, "SWZ TB, bumpenvmat%d, y, w, 0, 0;\n", reg_dest_code);
2056     shader_addline(buffer, "DP3 TA.y, TB, %s;\n", src_reg);
2057
2058     /* with projective textures, texbem only divides the static texture coord, not the displacement,
2059      * so we can't let the GL handle this.
2060      */
2061     if ((priv->cur_ps_args->super.tex_transform >> reg_dest_code * WINED3D_PSARGS_TEXTRANSFORM_SHIFT)
2062             & WINED3D_PSARGS_PROJECTED)
2063     {
2064         shader_addline(buffer, "RCP TB.w, %s.w;\n", reg_coord);
2065         shader_addline(buffer, "MUL TB.xy, %s, TB.w;\n", reg_coord);
2066         shader_addline(buffer, "ADD TA.xy, TA, TB;\n");
2067     } else {
2068         shader_addline(buffer, "ADD TA.xy, TA, %s;\n", reg_coord);
2069     }
2070
2071     shader_hw_sample(ins, reg_dest_code, dst_reg, "TA", 0, NULL, NULL);
2072
2073     if (ins->handler_idx == WINED3DSIH_TEXBEML)
2074     {
2075         /* No src swizzles are allowed, so this is ok */
2076         shader_addline(buffer, "MAD TA, %s.z, luminance%d.x, luminance%d.y;\n",
2077                        src_reg, reg_dest_code, reg_dest_code);
2078         shader_addline(buffer, "MUL %s, %s, TA;\n", dst_reg, dst_reg);
2079     }
2080 }
2081
2082 static void pshader_hw_texm3x2pad(const struct wined3d_shader_instruction *ins)
2083 {
2084     DWORD reg = ins->dst[0].reg.idx;
2085     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2086     char src0_name[50], dst_name[50];
2087     BOOL is_color;
2088     struct wined3d_shader_register tmp_reg = ins->dst[0].reg;
2089
2090     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2091     /* The next instruction will be a texm3x2tex or texm3x2depth that writes to the uninitialized
2092      * T<reg+1> register. Use this register to store the calculated vector
2093      */
2094     tmp_reg.idx = reg + 1;
2095     shader_arb_get_register_name(ins, &tmp_reg, dst_name, &is_color);
2096     shader_addline(buffer, "DP3 %s.x, fragment.texcoord[%u], %s;\n", dst_name, reg, src0_name);
2097 }
2098
2099 static void pshader_hw_texm3x2tex(const struct wined3d_shader_instruction *ins)
2100 {
2101     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2102     DWORD flags;
2103     DWORD reg = ins->dst[0].reg.idx;
2104     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2105     char dst_str[50];
2106     char src0_name[50];
2107     char dst_reg[50];
2108     BOOL is_color;
2109
2110     /* We know that we're writing to the uninitialized T<reg> register, so use it for temporary storage */
2111     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_reg, &is_color);
2112
2113     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2114     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2115     shader_addline(buffer, "DP3 %s.y, fragment.texcoord[%u], %s;\n", dst_reg, reg, src0_name);
2116     flags = reg < MAX_TEXTURES ? priv->cur_ps_args->super.tex_transform >> reg * WINED3D_PSARGS_TEXTRANSFORM_SHIFT : 0;
2117     shader_hw_sample(ins, reg, dst_str, dst_reg, flags & WINED3DTTFF_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2118 }
2119
2120 static void pshader_hw_texm3x3pad(const struct wined3d_shader_instruction *ins)
2121 {
2122     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
2123     DWORD reg = ins->dst[0].reg.idx;
2124     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2125     char src0_name[50], dst_name[50];
2126     struct wined3d_shader_register tmp_reg = ins->dst[0].reg;
2127     BOOL is_color;
2128
2129     /* There are always 2 texm3x3pad instructions followed by one texm3x3[tex,vspec, ...] instruction, with
2130      * incrementing ins->dst[0].register_idx numbers. So the pad instruction already knows the final destination
2131      * register, and this register is uninitialized(otherwise the assembler complains that it is 'redeclared')
2132      */
2133     tmp_reg.idx = reg + 2 - tex_mx->current_row;
2134     shader_arb_get_register_name(ins, &tmp_reg, dst_name, &is_color);
2135
2136     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2137     shader_addline(buffer, "DP3 %s.%c, fragment.texcoord[%u], %s;\n",
2138                    dst_name, 'x' + tex_mx->current_row, reg, src0_name);
2139     tex_mx->texcoord_w[tex_mx->current_row++] = reg;
2140 }
2141
2142 static void pshader_hw_texm3x3tex(const struct wined3d_shader_instruction *ins)
2143 {
2144     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2145     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
2146     DWORD flags;
2147     DWORD reg = ins->dst[0].reg.idx;
2148     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2149     char dst_str[50];
2150     char src0_name[50], dst_name[50];
2151     BOOL is_color;
2152
2153     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2154     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2155     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_name, reg, src0_name);
2156
2157     /* Sample the texture using the calculated coordinates */
2158     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2159     flags = reg < MAX_TEXTURES ? priv->cur_ps_args->super.tex_transform >> reg * WINED3D_PSARGS_TEXTRANSFORM_SHIFT : 0;
2160     shader_hw_sample(ins, reg, dst_str, dst_name, flags & WINED3DTTFF_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2161     tex_mx->current_row = 0;
2162 }
2163
2164 static void pshader_hw_texm3x3vspec(const struct wined3d_shader_instruction *ins)
2165 {
2166     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2167     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
2168     DWORD flags;
2169     DWORD reg = ins->dst[0].reg.idx;
2170     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2171     char dst_str[50];
2172     char src0_name[50];
2173     char dst_reg[50];
2174     BOOL is_color;
2175
2176     /* Get the dst reg without writemask strings. We know this register is uninitialized, so we can use all
2177      * components for temporary data storage
2178      */
2179     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_reg, &is_color);
2180     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2181     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_reg, reg, src0_name);
2182
2183     /* Construct the eye-ray vector from w coordinates */
2184     shader_addline(buffer, "MOV TB.x, fragment.texcoord[%u].w;\n", tex_mx->texcoord_w[0]);
2185     shader_addline(buffer, "MOV TB.y, fragment.texcoord[%u].w;\n", tex_mx->texcoord_w[1]);
2186     shader_addline(buffer, "MOV TB.z, fragment.texcoord[%u].w;\n", reg);
2187
2188     /* Calculate reflection vector
2189      */
2190     shader_addline(buffer, "DP3 %s.w, %s, TB;\n", dst_reg, dst_reg);
2191     /* The .w is ignored when sampling, so I can use TB.w to calculate dot(N, N) */
2192     shader_addline(buffer, "DP3 TB.w, %s, %s;\n", dst_reg, dst_reg);
2193     shader_addline(buffer, "RCP TB.w, TB.w;\n");
2194     shader_addline(buffer, "MUL %s.w, %s.w, TB.w;\n", dst_reg, dst_reg);
2195     shader_addline(buffer, "MUL %s, %s.w, %s;\n", dst_reg, dst_reg, dst_reg);
2196     shader_addline(buffer, "MAD %s, coefmul.x, %s, -TB;\n", dst_reg, dst_reg);
2197
2198     /* Sample the texture using the calculated coordinates */
2199     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2200     flags = reg < MAX_TEXTURES ? priv->cur_ps_args->super.tex_transform >> reg * WINED3D_PSARGS_TEXTRANSFORM_SHIFT : 0;
2201     shader_hw_sample(ins, reg, dst_str, dst_reg, flags & WINED3DTTFF_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2202     tex_mx->current_row = 0;
2203 }
2204
2205 static void pshader_hw_texm3x3spec(const struct wined3d_shader_instruction *ins)
2206 {
2207     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2208     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
2209     DWORD flags;
2210     DWORD reg = ins->dst[0].reg.idx;
2211     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2212     char dst_str[50];
2213     char src0_name[50];
2214     char src1_name[50];
2215     char dst_reg[50];
2216     BOOL is_color;
2217
2218     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2219     shader_arb_get_src_param(ins, &ins->src[0], 1, src1_name);
2220     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_reg, &is_color);
2221     /* Note: dst_reg.xy is input here, generated by two texm3x3pad instructions */
2222     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_reg, reg, src0_name);
2223
2224     /* Calculate reflection vector.
2225      *
2226      *                   dot(N, E)
2227      * dst_reg.xyz = 2 * --------- * N - E
2228      *                   dot(N, N)
2229      *
2230      * Which normalizes the normal vector
2231      */
2232     shader_addline(buffer, "DP3 %s.w, %s, %s;\n", dst_reg, dst_reg, src1_name);
2233     shader_addline(buffer, "DP3 TC.w, %s, %s;\n", dst_reg, dst_reg);
2234     shader_addline(buffer, "RCP TC.w, TC.w;\n");
2235     shader_addline(buffer, "MUL %s.w, %s.w, TC.w;\n", dst_reg, dst_reg);
2236     shader_addline(buffer, "MUL %s, %s.w, %s;\n", dst_reg, dst_reg, dst_reg);
2237     shader_addline(buffer, "MAD %s, coefmul.x, %s, -%s;\n", dst_reg, dst_reg, src1_name);
2238
2239     /* Sample the texture using the calculated coordinates */
2240     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2241     flags = reg < MAX_TEXTURES ? priv->cur_ps_args->super.tex_transform >> reg * WINED3D_PSARGS_TEXTRANSFORM_SHIFT : 0;
2242     shader_hw_sample(ins, reg, dst_str, dst_reg, flags & WINED3DTTFF_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2243     tex_mx->current_row = 0;
2244 }
2245
2246 static void pshader_hw_texdepth(const struct wined3d_shader_instruction *ins)
2247 {
2248     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2249     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2250     char dst_name[50];
2251     const char *zero = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ZERO);
2252     const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
2253
2254     /* texdepth has an implicit destination, the fragment depth value. It's only parameter,
2255      * which is essentially an input, is the destination register because it is the first
2256      * parameter. According to the msdn, this must be register r5, but let's keep it more flexible
2257      * here(writemasks/swizzles are not valid on texdepth)
2258      */
2259     shader_arb_get_dst_param(ins, dst, dst_name);
2260
2261     /* According to the msdn, the source register(must be r5) is unusable after
2262      * the texdepth instruction, so we're free to modify it
2263      */
2264     shader_addline(buffer, "MIN %s.y, %s.y, %s;\n", dst_name, dst_name, one);
2265
2266     /* How to deal with the special case dst_name.g == 0? if r != 0, then
2267      * the r * (1 / 0) will give infinity, which is clamped to 1.0, the correct
2268      * result. But if r = 0.0, then 0 * inf = 0, which is incorrect.
2269      */
2270     shader_addline(buffer, "RCP %s.y, %s.y;\n", dst_name, dst_name);
2271     shader_addline(buffer, "MUL TA.x, %s.x, %s.y;\n", dst_name, dst_name);
2272     shader_addline(buffer, "MIN TA.x, TA.x, %s;\n", one);
2273     shader_addline(buffer, "MAX result.depth, TA.x, %s;\n", zero);
2274 }
2275
2276 /** Process the WINED3DSIO_TEXDP3TEX instruction in ARB:
2277  * Take a 3-component dot product of the TexCoord[dstreg] and src,
2278  * then perform a 1D texture lookup from stage dstregnum, place into dst. */
2279 static void pshader_hw_texdp3tex(const struct wined3d_shader_instruction *ins)
2280 {
2281     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2282     DWORD sampler_idx = ins->dst[0].reg.idx;
2283     char src0[50];
2284     char dst_str[50];
2285
2286     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2287     shader_addline(buffer, "MOV TB, 0.0;\n");
2288     shader_addline(buffer, "DP3 TB.x, fragment.texcoord[%u], %s;\n", sampler_idx, src0);
2289
2290     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2291     shader_hw_sample(ins, sampler_idx, dst_str, "TB", 0 /* Only one coord, can't be projected */, NULL, NULL);
2292 }
2293
2294 /** Process the WINED3DSIO_TEXDP3 instruction in ARB:
2295  * Take a 3-component dot product of the TexCoord[dstreg] and src. */
2296 static void pshader_hw_texdp3(const struct wined3d_shader_instruction *ins)
2297 {
2298     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2299     char src0[50];
2300     char dst_str[50];
2301     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2302
2303     /* Handle output register */
2304     shader_arb_get_dst_param(ins, dst, dst_str);
2305     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2306     shader_addline(buffer, "DP3 %s, fragment.texcoord[%u], %s;\n", dst_str, dst->reg.idx, src0);
2307 }
2308
2309 /** Process the WINED3DSIO_TEXM3X3 instruction in ARB
2310  * Perform the 3rd row of a 3x3 matrix multiply */
2311 static void pshader_hw_texm3x3(const struct wined3d_shader_instruction *ins)
2312 {
2313     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2314     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2315     char dst_str[50], dst_name[50];
2316     char src0[50];
2317     BOOL is_color;
2318
2319     shader_arb_get_dst_param(ins, dst, dst_str);
2320     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2321     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2322     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_name, dst->reg.idx, src0);
2323     shader_addline(buffer, "MOV %s, %s;\n", dst_str, dst_name);
2324 }
2325
2326 /** Process the WINED3DSIO_TEXM3X2DEPTH instruction in ARB:
2327  * Last row of a 3x2 matrix multiply, use the result to calculate the depth:
2328  * Calculate tmp0.y = TexCoord[dstreg] . src.xyz;  (tmp0.x has already been calculated)
2329  * depth = (tmp0.y == 0.0) ? 1.0 : tmp0.x / tmp0.y
2330  */
2331 static void pshader_hw_texm3x2depth(const struct wined3d_shader_instruction *ins)
2332 {
2333     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2334     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2335     char src0[50], dst_name[50];
2336     BOOL is_color;
2337     const char *zero = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ZERO);
2338     const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
2339
2340     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2341     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2342     shader_addline(buffer, "DP3 %s.y, fragment.texcoord[%u], %s;\n", dst_name, dst->reg.idx, src0);
2343
2344     /* How to deal with the special case dst_name.g == 0? if r != 0, then
2345      * the r * (1 / 0) will give infinity, which is clamped to 1.0, the correct
2346      * result. But if r = 0.0, then 0 * inf = 0, which is incorrect.
2347      */
2348     shader_addline(buffer, "RCP %s.y, %s.y;\n", dst_name, dst_name);
2349     shader_addline(buffer, "MUL %s.x, %s.x, %s.y;\n", dst_name, dst_name, dst_name);
2350     shader_addline(buffer, "MIN %s.x, %s.x, %s;\n", dst_name, dst_name, one);
2351     shader_addline(buffer, "MAX result.depth, %s.x, %s;\n", dst_name, zero);
2352 }
2353
2354 /** Handles transforming all WINED3DSIO_M?x? opcodes for
2355     Vertex/Pixel shaders to ARB_vertex_program codes */
2356 static void shader_hw_mnxn(const struct wined3d_shader_instruction *ins)
2357 {
2358     int i;
2359     int nComponents = 0;
2360     struct wined3d_shader_dst_param tmp_dst = {{0}};
2361     struct wined3d_shader_src_param tmp_src[2] = {{{0}}};
2362     struct wined3d_shader_instruction tmp_ins;
2363
2364     memset(&tmp_ins, 0, sizeof(tmp_ins));
2365
2366     /* Set constants for the temporary argument */
2367     tmp_ins.ctx = ins->ctx;
2368     tmp_ins.dst_count = 1;
2369     tmp_ins.dst = &tmp_dst;
2370     tmp_ins.src_count = 2;
2371     tmp_ins.src = tmp_src;
2372
2373     switch(ins->handler_idx)
2374     {
2375         case WINED3DSIH_M4x4:
2376             nComponents = 4;
2377             tmp_ins.handler_idx = WINED3DSIH_DP4;
2378             break;
2379         case WINED3DSIH_M4x3:
2380             nComponents = 3;
2381             tmp_ins.handler_idx = WINED3DSIH_DP4;
2382             break;
2383         case WINED3DSIH_M3x4:
2384             nComponents = 4;
2385             tmp_ins.handler_idx = WINED3DSIH_DP3;
2386             break;
2387         case WINED3DSIH_M3x3:
2388             nComponents = 3;
2389             tmp_ins.handler_idx = WINED3DSIH_DP3;
2390             break;
2391         case WINED3DSIH_M3x2:
2392             nComponents = 2;
2393             tmp_ins.handler_idx = WINED3DSIH_DP3;
2394             break;
2395         default:
2396             FIXME("Unhandled opcode %#x\n", ins->handler_idx);
2397             break;
2398     }
2399
2400     tmp_dst = ins->dst[0];
2401     tmp_src[0] = ins->src[0];
2402     tmp_src[1] = ins->src[1];
2403     for (i = 0; i < nComponents; i++) {
2404         tmp_dst.write_mask = WINED3DSP_WRITEMASK_0 << i;
2405         shader_hw_map2gl(&tmp_ins);
2406         ++tmp_src[1].reg.idx;
2407     }
2408 }
2409
2410 static void shader_hw_rcp(const struct wined3d_shader_instruction *ins)
2411 {
2412     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2413
2414     char dst[50];
2415     char src[50];
2416
2417     shader_arb_get_dst_param(ins, &ins->dst[0], dst); /* Destination */
2418     shader_arb_get_src_param(ins, &ins->src[0], 0, src);
2419     if (ins->src[0].swizzle == WINED3DSP_NOSWIZZLE)
2420     {
2421         /* Dx sdk says .x is used if no swizzle is given, but our test shows that
2422          * .w is used
2423          */
2424         strcat(src, ".w");
2425     }
2426
2427     shader_addline(buffer, "RCP%s %s, %s;\n", shader_arb_get_modifier(ins), dst, src);
2428 }
2429
2430 static void shader_hw_scalar_op(const struct wined3d_shader_instruction *ins)
2431 {
2432     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2433     const char *instruction;
2434
2435     char dst[50];
2436     char src[50];
2437
2438     switch(ins->handler_idx)
2439     {
2440         case WINED3DSIH_RSQ:  instruction = "RSQ"; break;
2441         case WINED3DSIH_RCP:  instruction = "RCP"; break;
2442         case WINED3DSIH_EXP:  instruction = "EX2"; break;
2443         case WINED3DSIH_EXPP: instruction = "EXP"; break;
2444         default: instruction = "";
2445             FIXME("Unhandled opcode %#x\n", ins->handler_idx);
2446             break;
2447     }
2448
2449     shader_arb_get_dst_param(ins, &ins->dst[0], dst); /* Destination */
2450     shader_arb_get_src_param(ins, &ins->src[0], 0, src);
2451     if (ins->src[0].swizzle == WINED3DSP_NOSWIZZLE)
2452     {
2453         /* Dx sdk says .x is used if no swizzle is given, but our test shows that
2454          * .w is used
2455          */
2456         strcat(src, ".w");
2457     }
2458
2459     shader_addline(buffer, "%s%s %s, %s;\n", instruction, shader_arb_get_modifier(ins), dst, src);
2460 }
2461
2462 static void shader_hw_nrm(const struct wined3d_shader_instruction *ins)
2463 {
2464     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2465     char dst_name[50];
2466     char src_name[50];
2467     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2468     BOOL pshader = shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type);
2469     const char *zero = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ZERO);
2470
2471     shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2472     shader_arb_get_src_param(ins, &ins->src[0], 1 /* Use TB */, src_name);
2473
2474     /* In D3D, NRM of a vector with length zero returns zero. Catch this situation, as
2475      * otherwise NRM or RSQ would return NaN */
2476     if(pshader && priv->target_version >= NV3)
2477     {
2478         /* GL_NV_fragment_program2's NRM needs protection against length zero vectors too
2479          *
2480          * TODO: Find out if DP3+NRM+MOV is really faster than DP3+RSQ+MUL
2481          */
2482         shader_addline(buffer, "DP3C TA, %s, %s;\n", src_name, src_name);
2483         shader_addline(buffer, "NRM%s %s, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name);
2484         shader_addline(buffer, "MOV %s (EQ), %s;\n", dst_name, zero);
2485     }
2486     else if(priv->target_version >= NV2)
2487     {
2488         shader_addline(buffer, "DP3C TA.x, %s, %s;\n", src_name, src_name);
2489         shader_addline(buffer, "RSQ TA.x (NE), TA.x;\n");
2490         shader_addline(buffer, "MUL%s %s, %s, TA.x;\n", shader_arb_get_modifier(ins), dst_name,
2491                        src_name);
2492     }
2493     else
2494     {
2495         const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
2496
2497         shader_addline(buffer, "DP3 TA.x, %s, %s;\n", src_name, src_name);
2498         /* Pass any non-zero value to RSQ if the input vector has a length of zero. The
2499          * RSQ result doesn't matter, as long as multiplying it by 0 returns 0.
2500          */
2501         shader_addline(buffer, "SGE TA.y, -TA.x, %s;\n", zero);
2502         shader_addline(buffer, "MAD TA.x, %s, TA.y, TA.x;\n", one);
2503
2504         shader_addline(buffer, "RSQ TA.x, TA.x;\n");
2505         /* dst.w = src[0].w * 1 / (src.x^2 + src.y^2 + src.z^2)^(1/2) according to msdn*/
2506         shader_addline(buffer, "MUL%s %s, %s, TA.x;\n", shader_arb_get_modifier(ins), dst_name,
2507                     src_name);
2508     }
2509 }
2510
2511 static void shader_hw_lrp(const struct wined3d_shader_instruction *ins)
2512 {
2513     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2514     char dst_name[50];
2515     char src_name[3][50];
2516
2517     /* ARB_fragment_program has a convenient LRP instruction */
2518     if(shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type)) {
2519         shader_hw_map2gl(ins);
2520         return;
2521     }
2522
2523     shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2524     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
2525     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
2526     shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
2527
2528     shader_addline(buffer, "SUB TA, %s, %s;\n", src_name[1], src_name[2]);
2529     shader_addline(buffer, "MAD%s %s, %s, TA, %s;\n", shader_arb_get_modifier(ins),
2530                    dst_name, src_name[0], src_name[2]);
2531 }
2532
2533 static void shader_hw_sincos(const struct wined3d_shader_instruction *ins)
2534 {
2535     /* This instruction exists in ARB, but the d3d instruction takes two extra parameters which
2536      * must contain fixed constants. So we need a separate function to filter those constants and
2537      * can't use map2gl
2538      */
2539     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2540     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2541     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2542     char dst_name[50];
2543     char src_name0[50], src_name1[50], src_name2[50];
2544     BOOL is_color;
2545
2546     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name0);
2547     if(shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type)) {
2548         shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2549         /* No modifiers are supported on SCS */
2550         shader_addline(buffer, "SCS %s, %s;\n", dst_name, src_name0);
2551
2552         if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE)
2553         {
2554             shader_arb_get_register_name(ins, &dst->reg, src_name0, &is_color);
2555             shader_addline(buffer, "MOV_SAT %s, %s;\n", dst_name, src_name0);
2556         }
2557     } else if(priv->target_version >= NV2) {
2558         shader_arb_get_register_name(ins, &dst->reg, dst_name, &is_color);
2559
2560         /* Sincos writemask must be .x, .y or .xy */
2561         if(dst->write_mask & WINED3DSP_WRITEMASK_0)
2562             shader_addline(buffer, "COS%s %s.x, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name0);
2563         if(dst->write_mask & WINED3DSP_WRITEMASK_1)
2564             shader_addline(buffer, "SIN%s %s.y, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name0);
2565     } else {
2566         /* Approximate sine and cosine with a taylor series, as per math textbook. The application passes 8
2567          * helper constants(D3DSINCOSCONST1 and D3DSINCOSCONST2) in src1 and src2.
2568          *
2569          * sin(x) = x - x^3/3! + x^5/5! - x^7/7! + ...
2570          * cos(x) = 1 - x^2/2! + x^4/4! - x^6/6! + ...
2571          *
2572          * The constants we get are:
2573          *
2574          *  +1   +1,     -1     -1     +1      +1      -1       -1
2575          *      ---- ,  ---- , ---- , ----- , ----- , ----- , ------
2576          *      1!*2    2!*4   3!*8   4!*16   5!*32   6!*64   7!*128
2577          *
2578          * If used with x^2, x^3, x^4 etc they calculate sin(x/2) and cos(x/2):
2579          *
2580          * (x/2)^2 = x^2 / 4
2581          * (x/2)^3 = x^3 / 8
2582          * (x/2)^4 = x^4 / 16
2583          * (x/2)^5 = x^5 / 32
2584          * etc
2585          *
2586          * To get the final result:
2587          * sin(x) = 2 * sin(x/2) * cos(x/2)
2588          * cos(x) = cos(x/2)^2 - sin(x/2)^2
2589          * (from sin(x+y) and cos(x+y) rules)
2590          *
2591          * As per MSDN, dst.z is undefined after the operation, and so is
2592          * dst.x and dst.y if they're masked out by the writemask. Ie
2593          * sincos dst.y, src1, c0, c1
2594          * returns the sine in dst.y. dst.x and dst.z are undefined, dst.w is not touched. The assembler
2595          * vsa.exe also stops with an error if the dest register is the same register as the source
2596          * register. This means we can use dest.xyz as temporary storage. The assembler vsa.exe output also
2597          * indicates that sincos consumes 8 instruction slots in vs_2_0(and, strangely, in vs_3_0).
2598          */
2599         shader_arb_get_src_param(ins, &ins->src[1], 1, src_name1);
2600         shader_arb_get_src_param(ins, &ins->src[2], 2, src_name2);
2601         shader_arb_get_register_name(ins, &dst->reg, dst_name, &is_color);
2602
2603         shader_addline(buffer, "MUL %s.x, %s, %s;\n", dst_name, src_name0, src_name0);  /* x ^ 2 */
2604         shader_addline(buffer, "MUL TA.y, %s.x, %s;\n", dst_name, src_name0);           /* x ^ 3 */
2605         shader_addline(buffer, "MUL %s.y, TA.y, %s;\n", dst_name, src_name0);           /* x ^ 4 */
2606         shader_addline(buffer, "MUL TA.z, %s.y, %s;\n", dst_name, src_name0);           /* x ^ 5 */
2607         shader_addline(buffer, "MUL %s.z, TA.z, %s;\n", dst_name, src_name0);           /* x ^ 6 */
2608         shader_addline(buffer, "MUL TA.w, %s.z, %s;\n", dst_name, src_name0);           /* x ^ 7 */
2609
2610         /* sin(x/2)
2611          *
2612          * Unfortunately we don't get the constants in a DP4-capable form. Is there a way to
2613          * properly merge that with MULs in the code above?
2614          * The swizzles .yz and xw however fit into the .yzxw swizzle added to ps_2_0. Maybe
2615          * we can merge the sine and cosine MAD rows to calculate them together.
2616          */
2617         shader_addline(buffer, "MUL TA.x, %s, %s.w;\n", src_name0, src_name2); /* x^1, +1/(1!*2) */
2618         shader_addline(buffer, "MAD TA.x, TA.y, %s.x, TA.x;\n", src_name2); /* -1/(3!*8) */
2619         shader_addline(buffer, "MAD TA.x, TA.z, %s.w, TA.x;\n", src_name1); /* +1/(5!*32) */
2620         shader_addline(buffer, "MAD TA.x, TA.w, %s.x, TA.x;\n", src_name1); /* -1/(7!*128) */
2621
2622         /* cos(x/2) */
2623         shader_addline(buffer, "MAD TA.y, %s.x, %s.y, %s.z;\n", dst_name, src_name2, src_name2); /* -1/(2!*4), +1.0 */
2624         shader_addline(buffer, "MAD TA.y, %s.y, %s.z, TA.y;\n", dst_name, src_name1); /* +1/(4!*16) */
2625         shader_addline(buffer, "MAD TA.y, %s.z, %s.y, TA.y;\n", dst_name, src_name1); /* -1/(6!*64) */
2626
2627         if(dst->write_mask & WINED3DSP_WRITEMASK_0) {
2628             /* cos x */
2629             shader_addline(buffer, "MUL TA.z, TA.y, TA.y;\n");
2630             shader_addline(buffer, "MAD %s.x, -TA.x, TA.x, TA.z;\n", dst_name);
2631         }
2632         if(dst->write_mask & WINED3DSP_WRITEMASK_1) {
2633             /* sin x */
2634             shader_addline(buffer, "MUL %s.y, TA.x, TA.y;\n", dst_name);
2635             shader_addline(buffer, "ADD %s.y, %s.y, %s.y;\n", dst_name, dst_name, dst_name);
2636         }
2637     }
2638 }
2639
2640 static void shader_hw_sgn(const struct wined3d_shader_instruction *ins)
2641 {
2642     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2643     char dst_name[50];
2644     char src_name[50];
2645     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
2646
2647     shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2648     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name);
2649
2650     /* SGN is only valid in vertex shaders */
2651     if(ctx->target_version >= NV2) {
2652         shader_addline(buffer, "SSG%s %s, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name);
2653         return;
2654     }
2655
2656     /* If SRC > 0.0, -SRC < SRC = TRUE, otherwise false.
2657      * if SRC < 0.0,  SRC < -SRC = TRUE. If neither is true, src = 0.0
2658      */
2659     if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE) {
2660         shader_addline(buffer, "SLT %s, -%s, %s;\n", dst_name, src_name, src_name);
2661     } else {
2662         /* src contains TA? Write to the dest first. This won't overwrite our destination.
2663          * Then use TA, and calculate the final result
2664          *
2665          * Not reading from TA? Store the first result in TA to avoid overwriting the
2666          * destination if src reg = dst reg
2667          */
2668         if(strstr(src_name, "TA"))
2669         {
2670             shader_addline(buffer, "SLT %s,  %s, -%s;\n", dst_name, src_name, src_name);
2671             shader_addline(buffer, "SLT TA, -%s, %s;\n", src_name, src_name);
2672             shader_addline(buffer, "ADD %s, %s, -TA;\n", dst_name, dst_name);
2673         }
2674         else
2675         {
2676             shader_addline(buffer, "SLT TA, -%s, %s;\n", src_name, src_name);
2677             shader_addline(buffer, "SLT %s,  %s, -%s;\n", dst_name, src_name, src_name);
2678             shader_addline(buffer, "ADD %s, TA, -%s;\n", dst_name, dst_name);
2679         }
2680     }
2681 }
2682
2683 static void shader_hw_dsy(const struct wined3d_shader_instruction *ins)
2684 {
2685     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2686     char src[50];
2687     char dst[50];
2688     char dst_name[50];
2689     BOOL is_color;
2690
2691     shader_arb_get_dst_param(ins, &ins->dst[0], dst);
2692     shader_arb_get_src_param(ins, &ins->src[0], 0, src);
2693     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2694
2695     shader_addline(buffer, "DDY %s, %s;\n", dst, src);
2696     shader_addline(buffer, "MUL%s %s, %s, ycorrection.y;\n", shader_arb_get_modifier(ins), dst, dst_name);
2697 }
2698
2699 static DWORD abs_modifier(DWORD mod, BOOL *need_abs)
2700 {
2701     *need_abs = FALSE;
2702
2703     switch(mod)
2704     {
2705         case WINED3DSPSM_NONE:      return WINED3DSPSM_ABS;
2706         case WINED3DSPSM_NEG:       return WINED3DSPSM_ABS;
2707         case WINED3DSPSM_BIAS:      *need_abs = TRUE; return WINED3DSPSM_BIAS;
2708         case WINED3DSPSM_BIASNEG:   *need_abs = TRUE; return WINED3DSPSM_BIASNEG;
2709         case WINED3DSPSM_SIGN:      *need_abs = TRUE; return WINED3DSPSM_SIGN;
2710         case WINED3DSPSM_SIGNNEG:   *need_abs = TRUE; return WINED3DSPSM_SIGNNEG;
2711         case WINED3DSPSM_COMP:      *need_abs = TRUE; return WINED3DSPSM_COMP;
2712         case WINED3DSPSM_X2:        *need_abs = TRUE; return WINED3DSPSM_X2;
2713         case WINED3DSPSM_X2NEG:     *need_abs = TRUE; return WINED3DSPSM_X2NEG;
2714         case WINED3DSPSM_DZ:        *need_abs = TRUE; return WINED3DSPSM_DZ;
2715         case WINED3DSPSM_DW:        *need_abs = TRUE; return WINED3DSPSM_DW;
2716         case WINED3DSPSM_ABS:       return WINED3DSPSM_ABS;
2717         case WINED3DSPSM_ABSNEG:    return WINED3DSPSM_ABS;
2718     }
2719     FIXME("Unknown modifier %u\n", mod);
2720     return mod;
2721 }
2722
2723 static void shader_hw_log(const struct wined3d_shader_instruction *ins)
2724 {
2725     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2726     char src0[50], dst[50];
2727     struct wined3d_shader_src_param src0_copy = ins->src[0];
2728     BOOL need_abs = FALSE;
2729     const char *instr;
2730
2731     switch(ins->handler_idx)
2732     {
2733         case WINED3DSIH_LOG:  instr = "LG2"; break;
2734         case WINED3DSIH_LOGP: instr = "LOG"; break;
2735         default:
2736             ERR("Unexpected instruction %d\n", ins->handler_idx);
2737             return;
2738     }
2739
2740     /* LOG and LOGP operate on the absolute value of the input */
2741     src0_copy.modifiers = abs_modifier(src0_copy.modifiers, &need_abs);
2742
2743     shader_arb_get_dst_param(ins, &ins->dst[0], dst);
2744     shader_arb_get_src_param(ins, &src0_copy, 0, src0);
2745
2746     if(need_abs)
2747     {
2748         shader_addline(buffer, "ABS TA, %s;\n", src0);
2749         shader_addline(buffer, "%s%s %s, TA;\n", instr, shader_arb_get_modifier(ins), dst);
2750     }
2751     else
2752     {
2753         shader_addline(buffer, "%s%s %s, %s;\n", instr, shader_arb_get_modifier(ins), dst, src0);
2754     }
2755 }
2756
2757 static void shader_hw_pow(const struct wined3d_shader_instruction *ins)
2758 {
2759     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2760     char src0[50], src1[50], dst[50];
2761     struct wined3d_shader_src_param src0_copy = ins->src[0];
2762     BOOL need_abs = FALSE;
2763     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2764     const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
2765
2766     /* POW operates on the absolute value of the input */
2767     src0_copy.modifiers = abs_modifier(src0_copy.modifiers, &need_abs);
2768
2769     shader_arb_get_dst_param(ins, &ins->dst[0], dst);
2770     shader_arb_get_src_param(ins, &src0_copy, 0, src0);
2771     shader_arb_get_src_param(ins, &ins->src[1], 1, src1);
2772
2773     if (need_abs)
2774         shader_addline(buffer, "ABS TA.x, %s;\n", src0);
2775     else
2776         shader_addline(buffer, "MOV TA.x, %s;\n", src0);
2777
2778     if (priv->target_version >= NV2)
2779     {
2780         shader_addline(buffer, "MOVC TA.y, %s;\n", src1);
2781         shader_addline(buffer, "POW%s %s, TA.x, TA.y;\n", shader_arb_get_modifier(ins), dst);
2782         shader_addline(buffer, "MOV %s (EQ.y), %s;\n", dst, one);
2783     }
2784     else
2785     {
2786         const char *zero = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ZERO);
2787         const char *flt_eps = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_EPS);
2788
2789         shader_addline(buffer, "ABS TA.y, %s;\n", src1);
2790         shader_addline(buffer, "SGE TA.y, -TA.y, %s;\n", zero);
2791         /* Possibly add flt_eps to avoid getting float special values */
2792         shader_addline(buffer, "MAD TA.z, TA.y, %s, %s;\n", flt_eps, src1);
2793         shader_addline(buffer, "POW%s TA.x, TA.x, TA.z;\n", shader_arb_get_modifier(ins));
2794         shader_addline(buffer, "MAD TA.x, -TA.x, TA.y, TA.x;\n");
2795         shader_addline(buffer, "MAD %s, TA.y, %s, TA.x;\n", dst, one);
2796     }
2797 }
2798
2799 static void shader_hw_loop(const struct wined3d_shader_instruction *ins)
2800 {
2801     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2802     char src_name[50];
2803     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2804
2805     /* src0 is aL */
2806     shader_arb_get_src_param(ins, &ins->src[1], 0, src_name);
2807
2808     if(vshader)
2809     {
2810         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2811         struct list *e = list_head(&priv->control_frames);
2812         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2813
2814         if(priv->loop_depth > 1) shader_addline(buffer, "PUSHA aL;\n");
2815         /* The constant loader makes sure to load -1 into iX.w */
2816         shader_addline(buffer, "ARLC aL, %s.xywz;\n", src_name);
2817         shader_addline(buffer, "BRA loop_%u_end (LE.x);\n", control_frame->no.loop);
2818         shader_addline(buffer, "loop_%u_start:\n", control_frame->no.loop);
2819     }
2820     else
2821     {
2822         shader_addline(buffer, "LOOP %s;\n", src_name);
2823     }
2824 }
2825
2826 static void shader_hw_rep(const struct wined3d_shader_instruction *ins)
2827 {
2828     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2829     char src_name[50];
2830     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2831
2832     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name);
2833
2834     /* The constant loader makes sure to load -1 into iX.w */
2835     if(vshader)
2836     {
2837         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2838         struct list *e = list_head(&priv->control_frames);
2839         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2840
2841         if(priv->loop_depth > 1) shader_addline(buffer, "PUSHA aL;\n");
2842
2843         shader_addline(buffer, "ARLC aL, %s.xywz;\n", src_name);
2844         shader_addline(buffer, "BRA loop_%u_end (LE.x);\n", control_frame->no.loop);
2845         shader_addline(buffer, "loop_%u_start:\n", control_frame->no.loop);
2846     }
2847     else
2848     {
2849         shader_addline(buffer, "REP %s;\n", src_name);
2850     }
2851 }
2852
2853 static void shader_hw_endloop(const struct wined3d_shader_instruction *ins)
2854 {
2855     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2856     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2857
2858     if(vshader)
2859     {
2860         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2861         struct list *e = list_head(&priv->control_frames);
2862         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2863
2864         shader_addline(buffer, "ARAC aL.xy, aL;\n");
2865         shader_addline(buffer, "BRA loop_%u_start (GT.x);\n", control_frame->no.loop);
2866         shader_addline(buffer, "loop_%u_end:\n", control_frame->no.loop);
2867
2868         if(priv->loop_depth > 1) shader_addline(buffer, "POPA aL;\n");
2869     }
2870     else
2871     {
2872         shader_addline(buffer, "ENDLOOP;\n");
2873     }
2874 }
2875
2876 static void shader_hw_endrep(const struct wined3d_shader_instruction *ins)
2877 {
2878     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2879     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2880
2881     if(vshader)
2882     {
2883         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2884         struct list *e = list_head(&priv->control_frames);
2885         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2886
2887         shader_addline(buffer, "ARAC aL.xy, aL;\n");
2888         shader_addline(buffer, "BRA loop_%u_start (GT.x);\n", control_frame->no.loop);
2889         shader_addline(buffer, "loop_%u_end:\n", control_frame->no.loop);
2890
2891         if(priv->loop_depth > 1) shader_addline(buffer, "POPA aL;\n");
2892     }
2893     else
2894     {
2895         shader_addline(buffer, "ENDREP;\n");
2896     }
2897 }
2898
2899 static const struct control_frame *find_last_loop(const struct shader_arb_ctx_priv *priv)
2900 {
2901     struct control_frame *control_frame;
2902
2903     LIST_FOR_EACH_ENTRY(control_frame, &priv->control_frames, struct control_frame, entry)
2904     {
2905         if(control_frame->type == LOOP || control_frame->type == REP) return control_frame;
2906     }
2907     ERR("Could not find loop for break\n");
2908     return NULL;
2909 }
2910
2911 static void shader_hw_break(const struct wined3d_shader_instruction *ins)
2912 {
2913     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2914     const struct control_frame *control_frame = find_last_loop(ins->ctx->backend_data);
2915     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2916
2917     if(vshader)
2918     {
2919         shader_addline(buffer, "BRA loop_%u_end;\n", control_frame->no.loop);
2920     }
2921     else
2922     {
2923         shader_addline(buffer, "BRK;\n");
2924     }
2925 }
2926
2927 static const char *get_compare(COMPARISON_TYPE flags)
2928 {
2929     switch (flags)
2930     {
2931         case COMPARISON_GT: return "GT";
2932         case COMPARISON_EQ: return "EQ";
2933         case COMPARISON_GE: return "GE";
2934         case COMPARISON_LT: return "LT";
2935         case COMPARISON_NE: return "NE";
2936         case COMPARISON_LE: return "LE";
2937         default:
2938             FIXME("Unrecognized comparison value: %u\n", flags);
2939             return "(\?\?)";
2940     }
2941 }
2942
2943 static COMPARISON_TYPE invert_compare(COMPARISON_TYPE flags)
2944 {
2945     switch (flags)
2946     {
2947         case COMPARISON_GT: return COMPARISON_LE;
2948         case COMPARISON_EQ: return COMPARISON_NE;
2949         case COMPARISON_GE: return COMPARISON_LT;
2950         case COMPARISON_LT: return COMPARISON_GE;
2951         case COMPARISON_NE: return COMPARISON_EQ;
2952         case COMPARISON_LE: return COMPARISON_GT;
2953         default:
2954             FIXME("Unrecognized comparison value: %u\n", flags);
2955             return -1;
2956     }
2957 }
2958
2959 static void shader_hw_breakc(const struct wined3d_shader_instruction *ins)
2960 {
2961     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2962     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2963     const struct control_frame *control_frame = find_last_loop(ins->ctx->backend_data);
2964     char src_name0[50];
2965     char src_name1[50];
2966     const char *comp = get_compare(ins->flags);
2967
2968     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name0);
2969     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name1);
2970
2971     if(vshader)
2972     {
2973         /* SUBC CC, src0, src1" works only in pixel shaders, so use TA to throw
2974          * away the subtraction result
2975          */
2976         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
2977         shader_addline(buffer, "BRA loop_%u_end (%s.x);\n", control_frame->no.loop, comp);
2978     }
2979     else
2980     {
2981         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
2982         shader_addline(buffer, "BRK (%s.x);\n", comp);
2983     }
2984 }
2985
2986 static void shader_hw_ifc(const struct wined3d_shader_instruction *ins)
2987 {
2988     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2989     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2990     struct list *e = list_head(&priv->control_frames);
2991     struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2992     const char *comp;
2993     char src_name0[50];
2994     char src_name1[50];
2995     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2996
2997     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name0);
2998     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name1);
2999
3000     if(vshader)
3001     {
3002         /* Invert the flag. We jump to the else label if the condition is NOT true */
3003         comp = get_compare(invert_compare(ins->flags));
3004         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
3005         shader_addline(buffer, "BRA ifc_%u_else (%s.x);\n", control_frame->no.ifc, comp);
3006     }
3007     else
3008     {
3009         comp = get_compare(ins->flags);
3010         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
3011         shader_addline(buffer, "IF %s.x;\n", comp);
3012     }
3013 }
3014
3015 static void shader_hw_else(const struct wined3d_shader_instruction *ins)
3016 {
3017     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3018     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3019     struct list *e = list_head(&priv->control_frames);
3020     struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
3021     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
3022
3023     if(vshader)
3024     {
3025         shader_addline(buffer, "BRA ifc_%u_endif;\n", control_frame->no.ifc);
3026         shader_addline(buffer, "ifc_%u_else:\n", control_frame->no.ifc);
3027         control_frame->had_else = TRUE;
3028     }
3029     else
3030     {
3031         shader_addline(buffer, "ELSE;\n");
3032     }
3033 }
3034
3035 static void shader_hw_endif(const struct wined3d_shader_instruction *ins)
3036 {
3037     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3038     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3039     struct list *e = list_head(&priv->control_frames);
3040     struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
3041     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
3042
3043     if(vshader)
3044     {
3045         if(control_frame->had_else)
3046         {
3047             shader_addline(buffer, "ifc_%u_endif:\n", control_frame->no.ifc);
3048         }
3049         else
3050         {
3051             shader_addline(buffer, "#No else branch. else is endif\n");
3052             shader_addline(buffer, "ifc_%u_else:\n", control_frame->no.ifc);
3053         }
3054     }
3055     else
3056     {
3057         shader_addline(buffer, "ENDIF;\n");
3058     }
3059 }
3060
3061 static void shader_hw_texldd(const struct wined3d_shader_instruction *ins)
3062 {
3063     DWORD sampler_idx = ins->src[1].reg.idx;
3064     char reg_dest[40];
3065     char reg_src[3][40];
3066     WORD flags = TEX_DERIV;
3067
3068     shader_arb_get_dst_param(ins, &ins->dst[0], reg_dest);
3069     shader_arb_get_src_param(ins, &ins->src[0], 0, reg_src[0]);
3070     shader_arb_get_src_param(ins, &ins->src[2], 1, reg_src[1]);
3071     shader_arb_get_src_param(ins, &ins->src[3], 2, reg_src[2]);
3072
3073     if (ins->flags & WINED3DSI_TEXLD_PROJECT) flags |= TEX_PROJ;
3074     if (ins->flags & WINED3DSI_TEXLD_BIAS) flags |= TEX_BIAS;
3075
3076     shader_hw_sample(ins, sampler_idx, reg_dest, reg_src[0], flags, reg_src[1], reg_src[2]);
3077 }
3078
3079 static void shader_hw_texldl(const struct wined3d_shader_instruction *ins)
3080 {
3081     DWORD sampler_idx = ins->src[1].reg.idx;
3082     char reg_dest[40];
3083     char reg_coord[40];
3084     WORD flags = TEX_LOD;
3085
3086     shader_arb_get_dst_param(ins, &ins->dst[0], reg_dest);
3087     shader_arb_get_src_param(ins, &ins->src[0], 0, reg_coord);
3088
3089     if (ins->flags & WINED3DSI_TEXLD_PROJECT) flags |= TEX_PROJ;
3090     if (ins->flags & WINED3DSI_TEXLD_BIAS) flags |= TEX_BIAS;
3091
3092     shader_hw_sample(ins, sampler_idx, reg_dest, reg_coord, flags, NULL, NULL);
3093 }
3094
3095 static void shader_hw_label(const struct wined3d_shader_instruction *ins)
3096 {
3097     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3098     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3099
3100     priv->in_main_func = FALSE;
3101     /* Call instructions activate the NV extensions, not labels and rets. If there is an uncalled
3102      * subroutine, don't generate a label that will make GL complain
3103      */
3104     if(priv->target_version == ARB) return;
3105
3106     shader_addline(buffer, "l%u:\n", ins->src[0].reg.idx);
3107 }
3108
3109 static void vshader_add_footer(struct shader_arb_ctx_priv *priv_ctx,
3110         const struct arb_vshader_private *shader_data, const struct arb_vs_compile_args *args,
3111         const struct wined3d_shader_reg_maps *reg_maps, const struct wined3d_gl_info *gl_info,
3112         struct wined3d_shader_buffer *buffer)
3113 {
3114     unsigned int i;
3115
3116     /* The D3DRS_FOGTABLEMODE render state defines if the shader-generated fog coord is used
3117      * or if the fragment depth is used. If the fragment depth is used(FOGTABLEMODE != NONE),
3118      * the fog frag coord is thrown away. If the fog frag coord is used, but not written by
3119      * the shader, it is set to 0.0(fully fogged, since start = 1.0, end = 0.0)
3120      */
3121     if (args->super.fog_src == VS_FOG_Z)
3122         shader_addline(buffer, "MOV result.fogcoord, TMP_OUT.z;\n");
3123     else if (!reg_maps->fog)
3124         /* posFixup.x is always 1.0, so we can safely use it */
3125         shader_addline(buffer, "ADD result.fogcoord, posFixup.x, -posFixup.x;\n");
3126
3127     /* Clipplanes are always stored without y inversion */
3128     if (use_nv_clip(gl_info) && priv_ctx->target_version >= NV2)
3129     {
3130         if (args->super.clip_enabled)
3131         {
3132             for (i = 0; i < priv_ctx->vs_clipplanes; i++)
3133             {
3134                 shader_addline(buffer, "DP4 result.clip[%u].x, TMP_OUT, state.clip[%u].plane;\n", i, i);
3135             }
3136         }
3137     }
3138     else if (args->clip.boolclip.clip_texcoord)
3139     {
3140         unsigned int cur_clip = 0;
3141         char component[4] = {'x', 'y', 'z', 'w'};
3142         const char *zero = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_ZERO);
3143
3144         for (i = 0; i < gl_info->limits.clipplanes; ++i)
3145         {
3146             if (args->clip.boolclip.clipplane_mask & (1 << i))
3147             {
3148                 shader_addline(buffer, "DP4 TA.%c, TMP_OUT, state.clip[%u].plane;\n",
3149                                component[cur_clip++], i);
3150             }
3151         }
3152         switch (cur_clip)
3153         {
3154             case 0:
3155                 shader_addline(buffer, "MOV TA, %s;\n", zero);
3156                 break;
3157             case 1:
3158                 shader_addline(buffer, "MOV TA.yzw, %s;\n", zero);
3159                 break;
3160             case 2:
3161                 shader_addline(buffer, "MOV TA.zw, %s;\n", zero);
3162                 break;
3163             case 3:
3164                 shader_addline(buffer, "MOV TA.w, %s;\n", zero);
3165                 break;
3166         }
3167         shader_addline(buffer, "MOV result.texcoord[%u], TA;\n",
3168                        args->clip.boolclip.clip_texcoord - 1);
3169     }
3170
3171     /* Write the final position.
3172      *
3173      * OpenGL coordinates specify the center of the pixel while d3d coords specify
3174      * the corner. The offsets are stored in z and w in posFixup. posFixup.y contains
3175      * 1.0 or -1.0 to turn the rendering upside down for offscreen rendering. PosFixup.x
3176      * contains 1.0 to allow a mad, but arb vs swizzles are too restricted for that.
3177      */
3178     shader_addline(buffer, "MUL TA, posFixup, TMP_OUT.w;\n");
3179     shader_addline(buffer, "ADD TMP_OUT.x, TMP_OUT.x, TA.z;\n");
3180     shader_addline(buffer, "MAD TMP_OUT.y, TMP_OUT.y, posFixup.y, TA.w;\n");
3181
3182     /* Z coord [0;1]->[-1;1] mapping, see comment in transform_projection in state.c
3183      * and the glsl equivalent
3184      */
3185     if (need_helper_const(shader_data, reg_maps, gl_info))
3186     {
3187         const char *two = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_TWO);
3188         shader_addline(buffer, "MAD TMP_OUT.z, TMP_OUT.z, %s, -TMP_OUT.w;\n", two);
3189     }
3190     else
3191     {
3192         shader_addline(buffer, "ADD TMP_OUT.z, TMP_OUT.z, TMP_OUT.z;\n");
3193         shader_addline(buffer, "ADD TMP_OUT.z, TMP_OUT.z, -TMP_OUT.w;\n");
3194     }
3195
3196     shader_addline(buffer, "MOV result.position, TMP_OUT;\n");
3197
3198     priv_ctx->footer_written = TRUE;
3199 }
3200
3201 static void shader_hw_ret(const struct wined3d_shader_instruction *ins)
3202 {
3203     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3204     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3205     const struct wined3d_shader *shader = ins->ctx->shader;
3206     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
3207
3208     if(priv->target_version == ARB) return;
3209
3210     if(vshader)
3211     {
3212         if (priv->in_main_func) vshader_add_footer(priv, shader->backend_data,
3213                 priv->cur_vs_args, ins->ctx->reg_maps, ins->ctx->gl_info, buffer);
3214     }
3215
3216     shader_addline(buffer, "RET;\n");
3217 }
3218
3219 static void shader_hw_call(const struct wined3d_shader_instruction *ins)
3220 {
3221     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3222     shader_addline(buffer, "CAL l%u;\n", ins->src[0].reg.idx);
3223 }
3224
3225 /* GL locking is done by the caller */
3226 static GLuint create_arb_blt_vertex_program(const struct wined3d_gl_info *gl_info)
3227 {
3228     GLuint program_id = 0;
3229     GLint pos;
3230
3231     const char *blt_vprogram =
3232         "!!ARBvp1.0\n"
3233         "PARAM c[1] = { { 1, 0.5 } };\n"
3234         "MOV result.position, vertex.position;\n"
3235         "MOV result.color, c[0].x;\n"
3236         "MOV result.texcoord[0], vertex.texcoord[0];\n"
3237         "END\n";
3238
3239     GL_EXTCALL(glGenProgramsARB(1, &program_id));
3240     GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, program_id));
3241     GL_EXTCALL(glProgramStringARB(GL_VERTEX_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
3242             strlen(blt_vprogram), blt_vprogram));
3243     checkGLcall("glProgramStringARB()");
3244
3245     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
3246     if (pos != -1)
3247     {
3248         FIXME("Vertex program error at position %d: %s\n\n", pos,
3249             debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
3250         shader_arb_dump_program_source(blt_vprogram);
3251     }
3252     else
3253     {
3254         GLint native;
3255
3256         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
3257         checkGLcall("glGetProgramivARB()");
3258         if (!native) WARN("Program exceeds native resource limits.\n");
3259     }
3260
3261     return program_id;
3262 }
3263
3264 /* GL locking is done by the caller */
3265 static GLuint create_arb_blt_fragment_program(const struct wined3d_gl_info *gl_info,
3266         enum tex_types tex_type, BOOL masked)
3267 {
3268     GLuint program_id = 0;
3269     const char *fprogram;
3270     GLint pos;
3271
3272     static const char * const blt_fprograms_full[tex_type_count] =
3273     {
3274         /* tex_1d */
3275         NULL,
3276         /* tex_2d */
3277         "!!ARBfp1.0\n"
3278         "TEMP R0;\n"
3279         "TEX R0.x, fragment.texcoord[0], texture[0], 2D;\n"
3280         "MOV result.depth.z, R0.x;\n"
3281         "END\n",
3282         /* tex_3d */
3283         NULL,
3284         /* tex_cube */
3285         "!!ARBfp1.0\n"
3286         "TEMP R0;\n"
3287         "TEX R0.x, fragment.texcoord[0], texture[0], CUBE;\n"
3288         "MOV result.depth.z, R0.x;\n"
3289         "END\n",
3290         /* tex_rect */
3291         "!!ARBfp1.0\n"
3292         "TEMP R0;\n"
3293         "TEX R0.x, fragment.texcoord[0], texture[0], RECT;\n"
3294         "MOV result.depth.z, R0.x;\n"
3295         "END\n",
3296     };
3297
3298     static const char * const blt_fprograms_masked[tex_type_count] =
3299     {
3300         /* tex_1d */
3301         NULL,
3302         /* tex_2d */
3303         "!!ARBfp1.0\n"
3304         "PARAM mask = program.local[0];\n"
3305         "TEMP R0;\n"
3306         "SLT R0.xy, fragment.position, mask.zwzw;\n"
3307         "MUL R0.x, R0.x, R0.y;\n"
3308         "KIL -R0.x;\n"
3309         "TEX R0.x, fragment.texcoord[0], texture[0], 2D;\n"
3310         "MOV result.depth.z, R0.x;\n"
3311         "END\n",
3312         /* tex_3d */
3313         NULL,
3314         /* tex_cube */
3315         "!!ARBfp1.0\n"
3316         "PARAM mask = program.local[0];\n"
3317         "TEMP R0;\n"
3318         "SLT R0.xy, fragment.position, mask.zwzw;\n"
3319         "MUL R0.x, R0.x, R0.y;\n"
3320         "KIL -R0.x;\n"
3321         "TEX R0.x, fragment.texcoord[0], texture[0], CUBE;\n"
3322         "MOV result.depth.z, R0.x;\n"
3323         "END\n",
3324         /* tex_rect */
3325         "!!ARBfp1.0\n"
3326         "PARAM mask = program.local[0];\n"
3327         "TEMP R0;\n"
3328         "SLT R0.xy, fragment.position, mask.zwzw;\n"
3329         "MUL R0.x, R0.x, R0.y;\n"
3330         "KIL -R0.x;\n"
3331         "TEX R0.x, fragment.texcoord[0], texture[0], RECT;\n"
3332         "MOV result.depth.z, R0.x;\n"
3333         "END\n",
3334     };
3335
3336     fprogram = masked ? blt_fprograms_masked[tex_type] : blt_fprograms_full[tex_type];
3337     if (!fprogram)
3338     {
3339         FIXME("tex_type %#x not supported, falling back to tex_2d\n", tex_type);
3340         tex_type = tex_2d;
3341         fprogram = masked ? blt_fprograms_masked[tex_type] : blt_fprograms_full[tex_type];
3342     }
3343
3344     GL_EXTCALL(glGenProgramsARB(1, &program_id));
3345     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, program_id));
3346     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB, strlen(fprogram), fprogram));
3347     checkGLcall("glProgramStringARB()");
3348
3349     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
3350     if (pos != -1)
3351     {
3352         FIXME("Fragment program error at position %d: %s\n\n", pos,
3353             debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
3354         shader_arb_dump_program_source(fprogram);
3355     }
3356     else
3357     {
3358         GLint native;
3359
3360         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
3361         checkGLcall("glGetProgramivARB()");
3362         if (!native) WARN("Program exceeds native resource limits.\n");
3363     }
3364
3365     return program_id;
3366 }
3367
3368 static void arbfp_add_sRGB_correction(struct wined3d_shader_buffer *buffer, const char *fragcolor,
3369         const char *tmp1, const char *tmp2, const char *tmp3, const char *tmp4, BOOL condcode)
3370 {
3371     /* Perform sRGB write correction. See GLX_EXT_framebuffer_sRGB */
3372
3373     if(condcode)
3374     {
3375         /* Sigh. MOVC CC doesn't work, so use one of the temps as dummy dest */
3376         shader_addline(buffer, "SUBC %s, %s.x, srgb_consts1.y;\n", tmp1, fragcolor);
3377         /* Calculate the > 0.0031308 case */
3378         shader_addline(buffer, "POW %s.x (GE), %s.x, srgb_consts1.z;\n", fragcolor, fragcolor);
3379         shader_addline(buffer, "POW %s.y (GE), %s.y, srgb_consts1.z;\n", fragcolor, fragcolor);
3380         shader_addline(buffer, "POW %s.z (GE), %s.z, srgb_consts1.z;\n", fragcolor, fragcolor);
3381         shader_addline(buffer, "MUL %s.xyz (GE), %s, srgb_consts1.w;\n", fragcolor, fragcolor);
3382         shader_addline(buffer, "SUB %s.xyz (GE), %s, srgb_consts2.x;\n", fragcolor, fragcolor);
3383         /* Calculate the < case */
3384         shader_addline(buffer, "MUL %s.xyz (LT), srgb_consts1.x, %s;\n", fragcolor, fragcolor);
3385     }
3386     else
3387     {
3388         /* Calculate the > 0.0031308 case */
3389         shader_addline(buffer, "POW %s.x, %s.x, srgb_consts1.z;\n", tmp1, fragcolor);
3390         shader_addline(buffer, "POW %s.y, %s.y, srgb_consts1.z;\n", tmp1, fragcolor);
3391         shader_addline(buffer, "POW %s.z, %s.z, srgb_consts1.z;\n", tmp1, fragcolor);
3392         shader_addline(buffer, "MUL %s, %s, srgb_consts1.w;\n", tmp1, tmp1);
3393         shader_addline(buffer, "SUB %s, %s, srgb_consts2.x;\n", tmp1, tmp1);
3394         /* Calculate the < case */
3395         shader_addline(buffer, "MUL %s, srgb_consts1.x, %s;\n", tmp2, fragcolor);
3396         /* Get 1.0 / 0.0 masks for > 0.0031308 and < 0.0031308 */
3397         shader_addline(buffer, "SLT %s, srgb_consts1.y, %s;\n", tmp3, fragcolor);
3398         shader_addline(buffer, "SGE %s, srgb_consts1.y, %s;\n", tmp4, fragcolor);
3399         /* Store the components > 0.0031308 in the destination */
3400         shader_addline(buffer, "MUL %s.xyz, %s, %s;\n", fragcolor, tmp1, tmp3);
3401         /* Add the components that are < 0.0031308 */
3402         shader_addline(buffer, "MAD %s.xyz, %s, %s, %s;\n", fragcolor, tmp2, tmp4, fragcolor);
3403         /* Move everything into result.color at once. Nvidia hardware cannot handle partial
3404         * result.color writes(.rgb first, then .a), or handle overwriting already written
3405         * components. The assembler uses a temporary register in this case, which is usually
3406         * not allocated from one of our registers that were used earlier.
3407         */
3408     }
3409     /* [0.0;1.0] clamping. Not needed, this is done implicitly */
3410 }
3411
3412 static const DWORD *find_loop_control_values(const struct wined3d_shader *shader, DWORD idx)
3413 {
3414     const local_constant *constant;
3415
3416     LIST_FOR_EACH_ENTRY(constant, &shader->constantsI, local_constant, entry)
3417     {
3418         if (constant->idx == idx)
3419         {
3420             return constant->value;
3421         }
3422     }
3423     return NULL;
3424 }
3425
3426 static void init_ps_input(const struct wined3d_shader *shader,
3427         const struct arb_ps_compile_args *args, struct shader_arb_ctx_priv *priv)
3428 {
3429     static const char * const texcoords[8] =
3430     {
3431         "fragment.texcoord[0]", "fragment.texcoord[1]", "fragment.texcoord[2]", "fragment.texcoord[3]",
3432         "fragment.texcoord[4]", "fragment.texcoord[5]", "fragment.texcoord[6]", "fragment.texcoord[7]"
3433     };
3434     unsigned int i;
3435     const struct wined3d_shader_signature_element *sig = shader->input_signature;
3436     const char *semantic_name;
3437     DWORD semantic_idx;
3438
3439     switch(args->super.vp_mode)
3440     {
3441         case pretransformed:
3442         case fixedfunction:
3443             /* The pixelshader has to collect the varyings on its own. In any case properly load
3444              * color0 and color1. In the case of pretransformed vertices also load texcoords. Set
3445              * other attribs to 0.0.
3446              *
3447              * For fixedfunction this behavior is correct, according to the tests. For pretransformed
3448              * we'd either need a replacement shader that can load other attribs like BINORMAL, or
3449              * load the texcoord attrib pointers to match the pixel shader signature
3450              */
3451             for(i = 0; i < MAX_REG_INPUT; i++)
3452             {
3453                 semantic_name = sig[i].semantic_name;
3454                 semantic_idx = sig[i].semantic_idx;
3455                 if (!semantic_name) continue;
3456
3457                 if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_COLOR))
3458                 {
3459                     if (!semantic_idx) priv->ps_input[i] = "fragment.color.primary";
3460                     else if(semantic_idx == 1) priv->ps_input[i] = "fragment.color.secondary";
3461                     else priv->ps_input[i] = "0.0";
3462                 }
3463                 else if(args->super.vp_mode == fixedfunction)
3464                 {
3465                     priv->ps_input[i] = "0.0";
3466                 }
3467                 else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_TEXCOORD))
3468                 {
3469                     if(semantic_idx < 8) priv->ps_input[i] = texcoords[semantic_idx];
3470                     else priv->ps_input[i] = "0.0";
3471                 }
3472                 else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_FOG))
3473                 {
3474                     if (!semantic_idx) priv->ps_input[i] = "fragment.fogcoord";
3475                     else priv->ps_input[i] = "0.0";
3476                 }
3477                 else
3478                 {
3479                     priv->ps_input[i] = "0.0";
3480                 }
3481
3482                 TRACE("v%u, semantic %s%u is %s\n", i, semantic_name, semantic_idx, priv->ps_input[i]);
3483             }
3484             break;
3485
3486         case vertexshader:
3487             /* That one is easy. The vertex shaders provide v0-v7 in fragment.texcoord and v8 and v9 in
3488              * fragment.color
3489              */
3490             for(i = 0; i < 8; i++)
3491             {
3492                 priv->ps_input[i] = texcoords[i];
3493             }
3494             priv->ps_input[8] = "fragment.color.primary";
3495             priv->ps_input[9] = "fragment.color.secondary";
3496             break;
3497     }
3498 }
3499
3500 /* GL locking is done by the caller */
3501 static GLuint shader_arb_generate_pshader(struct wined3d_shader *shader,
3502         const struct wined3d_gl_info *gl_info, struct wined3d_shader_buffer *buffer,
3503         const struct arb_ps_compile_args *args, struct arb_ps_compiled_shader *compiled)
3504 {
3505     const struct wined3d_shader_reg_maps *reg_maps = &shader->reg_maps;
3506     const DWORD *function = shader->function;
3507     const local_constant *lconst;
3508     GLuint retval;
3509     char fragcolor[16];
3510     DWORD *lconst_map = local_const_mapping(shader), next_local;
3511     struct shader_arb_ctx_priv priv_ctx;
3512     BOOL dcl_td = FALSE;
3513     BOOL want_nv_prog = FALSE;
3514     struct arb_pshader_private *shader_priv = shader->backend_data;
3515     GLint errPos;
3516     DWORD map;
3517
3518     char srgbtmp[4][4];
3519     unsigned int i, found = 0;
3520
3521     for (i = 0, map = reg_maps->temporary; map; map >>= 1, ++i)
3522     {
3523         if (!(map & 1)
3524                 || (shader->u.ps.color0_mov && i == shader->u.ps.color0_reg)
3525                 || (reg_maps->shader_version.major < 2 && !i))
3526             continue;
3527
3528         sprintf(srgbtmp[found], "R%u", i);
3529         ++found;
3530         if (found == 4) break;
3531     }
3532
3533     switch(found) {
3534         case 0:
3535             sprintf(srgbtmp[0], "TA");
3536             sprintf(srgbtmp[1], "TB");
3537             sprintf(srgbtmp[2], "TC");
3538             sprintf(srgbtmp[3], "TD");
3539             dcl_td = TRUE;
3540             break;
3541         case 1:
3542             sprintf(srgbtmp[1], "TA");
3543             sprintf(srgbtmp[2], "TB");
3544             sprintf(srgbtmp[3], "TC");
3545             break;
3546         case 2:
3547             sprintf(srgbtmp[2], "TA");
3548             sprintf(srgbtmp[3], "TB");
3549             break;
3550         case 3:
3551             sprintf(srgbtmp[3], "TA");
3552             break;
3553         case 4:
3554             break;
3555     }
3556
3557     /*  Create the hw ARB shader */
3558     memset(&priv_ctx, 0, sizeof(priv_ctx));
3559     priv_ctx.cur_ps_args = args;
3560     priv_ctx.compiled_fprog = compiled;
3561     priv_ctx.cur_np2fixup_info = &compiled->np2fixup_info;
3562     init_ps_input(shader, args, &priv_ctx);
3563     list_init(&priv_ctx.control_frames);
3564
3565     /* Avoid enabling NV_fragment_program* if we do not need it.
3566      *
3567      * Enabling GL_NV_fragment_program_option causes the driver to occupy a temporary register,
3568      * and it slows down the shader execution noticeably(about 5%). Usually our instruction emulation
3569      * is faster than what we gain from using higher native instructions. There are some things though
3570      * that cannot be emulated. In that case enable the extensions.
3571      * If the extension is enabled, instruction handlers that support both ways will use it.
3572      *
3573      * Testing shows no performance difference between OPTION NV_fragment_program2 and NV_fragment_program.
3574      * So enable the best we can get.
3575      */
3576     if(reg_maps->usesdsx || reg_maps->usesdsy || reg_maps->loop_depth > 0 || reg_maps->usestexldd ||
3577        reg_maps->usestexldl || reg_maps->usesfacing || reg_maps->usesifc || reg_maps->usescall)
3578     {
3579         want_nv_prog = TRUE;
3580     }
3581
3582     shader_addline(buffer, "!!ARBfp1.0\n");
3583     if (want_nv_prog && gl_info->supported[NV_FRAGMENT_PROGRAM2])
3584     {
3585         shader_addline(buffer, "OPTION NV_fragment_program2;\n");
3586         priv_ctx.target_version = NV3;
3587     }
3588     else if (want_nv_prog && gl_info->supported[NV_FRAGMENT_PROGRAM_OPTION])
3589     {
3590         shader_addline(buffer, "OPTION NV_fragment_program;\n");
3591         priv_ctx.target_version = NV2;
3592     } else {
3593         if(want_nv_prog)
3594         {
3595             /* This is an error - either we're advertising the wrong shader version, or aren't enforcing some
3596              * limits properly
3597              */
3598             ERR("The shader requires instructions that are not available in plain GL_ARB_fragment_program\n");
3599             ERR("Try GLSL\n");
3600         }
3601         priv_ctx.target_version = ARB;
3602     }
3603
3604     if (reg_maps->rt_mask > 1)
3605     {
3606         shader_addline(buffer, "OPTION ARB_draw_buffers;\n");
3607     }
3608
3609     if (reg_maps->shader_version.major < 3)
3610     {
3611         switch(args->super.fog) {
3612             case FOG_OFF:
3613                 break;
3614             case FOG_LINEAR:
3615                 shader_addline(buffer, "OPTION ARB_fog_linear;\n");
3616                 break;
3617             case FOG_EXP:
3618                 shader_addline(buffer, "OPTION ARB_fog_exp;\n");
3619                 break;
3620             case FOG_EXP2:
3621                 shader_addline(buffer, "OPTION ARB_fog_exp2;\n");
3622                 break;
3623         }
3624     }
3625
3626     /* For now always declare the temps. At least the Nvidia assembler optimizes completely
3627      * unused temps away(but occupies them for the whole shader if they're used once). Always
3628      * declaring them avoids tricky bookkeeping work
3629      */
3630     shader_addline(buffer, "TEMP TA;\n");      /* Used for modifiers */
3631     shader_addline(buffer, "TEMP TB;\n");      /* Used for modifiers */
3632     shader_addline(buffer, "TEMP TC;\n");      /* Used for modifiers */
3633     if(dcl_td) shader_addline(buffer, "TEMP TD;\n"); /* Used for sRGB writing */
3634     shader_addline(buffer, "PARAM coefdiv = { 0.5, 0.25, 0.125, 0.0625 };\n");
3635     shader_addline(buffer, "PARAM coefmul = { 2, 4, 8, 16 };\n");
3636     shader_addline(buffer, "PARAM ps_helper_const = { 0.0, 1.0, %1.10f, 0.0 };\n", eps);
3637
3638     if (reg_maps->shader_version.major < 2)
3639     {
3640         strcpy(fragcolor, "R0");
3641     }
3642     else
3643     {
3644         if (args->super.srgb_correction)
3645         {
3646             if (shader->u.ps.color0_mov)
3647             {
3648                 sprintf(fragcolor, "R%u", shader->u.ps.color0_reg);
3649             }
3650             else
3651             {
3652                 shader_addline(buffer, "TEMP TMP_COLOR;\n");
3653                 strcpy(fragcolor, "TMP_COLOR");
3654             }
3655         } else {
3656             strcpy(fragcolor, "result.color");
3657         }
3658     }
3659
3660     if(args->super.srgb_correction) {
3661         shader_addline(buffer, "PARAM srgb_consts1 = {%f, %f, %f, %f};\n",
3662                        srgb_mul_low, srgb_cmp, srgb_pow, srgb_mul_high);
3663         shader_addline(buffer, "PARAM srgb_consts2 = {%f, %f, %f, %f};\n",
3664                        srgb_sub_high, 0.0, 0.0, 0.0);
3665     }
3666
3667     /* Base Declarations */
3668     next_local = shader_generate_arb_declarations(shader, reg_maps,
3669             buffer, gl_info, lconst_map, NULL, &priv_ctx);
3670
3671     for (i = 0, map = reg_maps->bumpmat; map; map >>= 1, ++i)
3672     {
3673         unsigned char bump_const;
3674
3675         if (!(map & 1)) continue;
3676
3677         bump_const = compiled->numbumpenvmatconsts;
3678         compiled->bumpenvmatconst[bump_const].const_num = WINED3D_CONST_NUM_UNUSED;
3679         compiled->bumpenvmatconst[bump_const].texunit = i;
3680         compiled->luminanceconst[bump_const].const_num = WINED3D_CONST_NUM_UNUSED;
3681         compiled->luminanceconst[bump_const].texunit = i;
3682
3683         /* We can fit the constants into the constant limit for sure because texbem, texbeml, bem and beml are only supported
3684          * in 1.x shaders, and GL_ARB_fragment_program has a constant limit of 24 constants. So in the worst case we're loading
3685          * 8 shader constants, 8 bump matrices and 8 luminance parameters and are perfectly fine. (No NP2 fixup on bumpmapped
3686          * textures due to conditional NP2 restrictions)
3687          *
3688          * Use local constants to load the bump env parameters, not program.env. This avoids collisions with d3d constants of
3689          * shaders in newer shader models. Since the bump env parameters have to share their space with NP2 fixup constants,
3690          * their location is shader dependent anyway and they cannot be loaded globally.
3691          */
3692         compiled->bumpenvmatconst[bump_const].const_num = next_local++;
3693         shader_addline(buffer, "PARAM bumpenvmat%d = program.local[%d];\n",
3694                        i, compiled->bumpenvmatconst[bump_const].const_num);
3695         compiled->numbumpenvmatconsts = bump_const + 1;
3696
3697         if (!(reg_maps->luminanceparams & (1 << i))) continue;
3698
3699         compiled->luminanceconst[bump_const].const_num = next_local++;
3700         shader_addline(buffer, "PARAM luminance%d = program.local[%d];\n",
3701                        i, compiled->luminanceconst[bump_const].const_num);
3702     }
3703
3704     for(i = 0; i < MAX_CONST_I; i++)
3705     {
3706         compiled->int_consts[i] = WINED3D_CONST_NUM_UNUSED;
3707         if (reg_maps->integer_constants & (1 << i) && priv_ctx.target_version >= NV2)
3708         {
3709             const DWORD *control_values = find_loop_control_values(shader, i);
3710
3711             if(control_values)
3712             {
3713                 shader_addline(buffer, "PARAM I%u = {%u, %u, %u, -1};\n", i,
3714                                 control_values[0], control_values[1], control_values[2]);
3715             }
3716             else
3717             {
3718                 compiled->int_consts[i] = next_local;
3719                 compiled->num_int_consts++;
3720                 shader_addline(buffer, "PARAM I%u = program.local[%u];\n", i, next_local++);
3721             }
3722         }
3723     }
3724
3725     if(reg_maps->vpos || reg_maps->usesdsy)
3726     {
3727         compiled->ycorrection = next_local;
3728         shader_addline(buffer, "PARAM ycorrection = program.local[%u];\n", next_local++);
3729
3730         if(reg_maps->vpos)
3731         {
3732             shader_addline(buffer, "TEMP vpos;\n");
3733             /* ycorrection.x: Backbuffer height(onscreen) or 0(offscreen).
3734              * ycorrection.y: -1.0(onscreen), 1.0(offscreen)
3735              * ycorrection.z: 1.0
3736              * ycorrection.w: 0.0
3737              */
3738             shader_addline(buffer, "MAD vpos, fragment.position, ycorrection.zyww, ycorrection.wxww;\n");
3739             shader_addline(buffer, "FLR vpos.xy, vpos;\n");
3740         }
3741     }
3742     else
3743     {
3744         compiled->ycorrection = WINED3D_CONST_NUM_UNUSED;
3745     }
3746
3747     /* Load constants to fixup NP2 texcoords if there are still free constants left:
3748      * Constants (texture dimensions) for the NP2 fixup are loaded as local program parameters. This will consume
3749      * at most 8 (MAX_FRAGMENT_SAMPLERS / 2) parameters, which is highly unlikely, since the application had to
3750      * use 16 NP2 textures at the same time. In case that we run out of constants the fixup is simply not
3751      * applied / activated. This will probably result in wrong rendering of the texture, but will save us from
3752      * shader compilation errors and the subsequent errors when drawing with this shader. */
3753     if (priv_ctx.cur_ps_args->super.np2_fixup) {
3754         unsigned char cur_fixup_sampler = 0;
3755
3756         struct arb_ps_np2fixup_info* const fixup = priv_ctx.cur_np2fixup_info;
3757         const WORD map = priv_ctx.cur_ps_args->super.np2_fixup;
3758         const UINT max_lconsts = gl_info->limits.arb_ps_local_constants;
3759
3760         fixup->offset = next_local;
3761         fixup->super.active = 0;
3762
3763         for (i = 0; i < MAX_FRAGMENT_SAMPLERS; ++i) {
3764             if (!(map & (1 << i))) continue;
3765
3766             if (fixup->offset + (cur_fixup_sampler >> 1) < max_lconsts) {
3767                 fixup->super.active |= (1 << i);
3768                 fixup->super.idx[i] = cur_fixup_sampler++;
3769             } else {
3770                 FIXME("No free constant found to load NP2 fixup data into shader. "
3771                       "Sampling from this texture will probably look wrong.\n");
3772                 break;
3773             }
3774         }
3775
3776         fixup->super.num_consts = (cur_fixup_sampler + 1) >> 1;
3777         if (fixup->super.num_consts) {
3778             shader_addline(buffer, "PARAM np2fixup[%u] = { program.env[%u..%u] };\n",
3779                            fixup->super.num_consts, fixup->offset, fixup->super.num_consts + fixup->offset - 1);
3780         }
3781
3782         next_local += fixup->super.num_consts;
3783     }
3784
3785     if (shader_priv->clipplane_emulation != ~0U && args->clip)
3786     {
3787         shader_addline(buffer, "KIL fragment.texcoord[%u];\n", shader_priv->clipplane_emulation);
3788     }
3789
3790     /* Base Shader Body */
3791     shader_generate_main(shader, buffer, reg_maps, function, &priv_ctx);
3792
3793     if(args->super.srgb_correction) {
3794         arbfp_add_sRGB_correction(buffer, fragcolor, srgbtmp[0], srgbtmp[1], srgbtmp[2], srgbtmp[3],
3795                                   priv_ctx.target_version >= NV2);
3796     }
3797
3798     if(strcmp(fragcolor, "result.color")) {
3799         shader_addline(buffer, "MOV result.color, %s;\n", fragcolor);
3800     }
3801     shader_addline(buffer, "END\n");
3802
3803     /* TODO: change to resource.glObjectHandle or something like that */
3804     GL_EXTCALL(glGenProgramsARB(1, &retval));
3805
3806     TRACE("Creating a hw pixel shader, prg=%d\n", retval);
3807     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, retval));
3808
3809     TRACE("Created hw pixel shader, prg=%d\n", retval);
3810     /* Create the program and check for errors */
3811     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
3812                buffer->bsize, buffer->buffer));
3813     checkGLcall("glProgramStringARB()");
3814
3815     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &errPos);
3816     if (errPos != -1)
3817     {
3818         FIXME("HW PixelShader Error at position %d: %s\n\n",
3819               errPos, debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
3820         shader_arb_dump_program_source(buffer->buffer);
3821         retval = 0;
3822     }
3823     else
3824     {
3825         GLint native;
3826
3827         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
3828         checkGLcall("glGetProgramivARB()");
3829         if (!native) WARN("Program exceeds native resource limits.\n");
3830     }
3831
3832     /* Load immediate constants */
3833     if (lconst_map)
3834     {
3835         LIST_FOR_EACH_ENTRY(lconst, &shader->constantsF, local_constant, entry)
3836         {
3837             const float *value = (const float *)lconst->value;
3838             GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, lconst_map[lconst->idx], value));
3839             checkGLcall("glProgramLocalParameter4fvARB");
3840         }
3841         HeapFree(GetProcessHeap(), 0, lconst_map);
3842     }
3843
3844     return retval;
3845 }
3846
3847 static int compare_sig(const struct wined3d_shader_signature_element *sig1, const struct wined3d_shader_signature_element *sig2)
3848 {
3849     unsigned int i;
3850     int ret;
3851
3852     for(i = 0; i < MAX_REG_INPUT; i++)
3853     {
3854         if (!sig1[i].semantic_name || !sig2[i].semantic_name)
3855         {
3856             /* Compare pointers, not contents. One string is NULL(element does not exist), the other one is not NULL */
3857             if(sig1[i].semantic_name != sig2[i].semantic_name) return sig1[i].semantic_name < sig2[i].semantic_name ? -1 : 1;
3858             continue;
3859         }
3860
3861         if ((ret = strcmp(sig1[i].semantic_name, sig2[i].semantic_name))) return ret;
3862         if(sig1[i].semantic_idx    != sig2[i].semantic_idx)    return sig1[i].semantic_idx    < sig2[i].semantic_idx    ? -1 : 1;
3863         if(sig1[i].sysval_semantic != sig2[i].sysval_semantic) return sig1[i].sysval_semantic < sig2[i].sysval_semantic ? -1 : 1;
3864         if(sig1[i].component_type  != sig2[i].component_type)  return sig1[i].component_type  < sig2[i].component_type  ? -1 : 1;
3865         if(sig1[i].register_idx    != sig2[i].register_idx)    return sig1[i].register_idx    < sig2[i].register_idx    ? -1 : 1;
3866         if(sig1[i].mask            != sig2[i].mask)            return sig1[i].mask            < sig2[i].mask            ? -1 : 1;
3867     }
3868     return 0;
3869 }
3870
3871 static struct wined3d_shader_signature_element *clone_sig(const struct wined3d_shader_signature_element *sig)
3872 {
3873     struct wined3d_shader_signature_element *new;
3874     int i;
3875     char *name;
3876
3877     new = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*new) * MAX_REG_INPUT);
3878     for(i = 0; i < MAX_REG_INPUT; i++)
3879     {
3880         if (!sig[i].semantic_name) continue;
3881
3882         new[i] = sig[i];
3883         /* Clone the semantic string */
3884         name = HeapAlloc(GetProcessHeap(), 0, strlen(sig[i].semantic_name) + 1);
3885         strcpy(name, sig[i].semantic_name);
3886         new[i].semantic_name = name;
3887     }
3888     return new;
3889 }
3890
3891 static DWORD find_input_signature(struct shader_arb_priv *priv, const struct wined3d_shader_signature_element *sig)
3892 {
3893     struct wine_rb_entry *entry = wine_rb_get(&priv->signature_tree, sig);
3894     struct ps_signature *found_sig;
3895
3896     if (entry)
3897     {
3898         found_sig = WINE_RB_ENTRY_VALUE(entry, struct ps_signature, entry);
3899         TRACE("Found existing signature %u\n", found_sig->idx);
3900         return found_sig->idx;
3901     }
3902     found_sig = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*sig));
3903     found_sig->sig = clone_sig(sig);
3904     found_sig->idx = priv->ps_sig_number++;
3905     TRACE("New signature stored and assigned number %u\n", found_sig->idx);
3906     if(wine_rb_put(&priv->signature_tree, sig, &found_sig->entry) == -1)
3907     {
3908         ERR("Failed to insert program entry.\n");
3909     }
3910     return found_sig->idx;
3911 }
3912
3913 static void init_output_registers(struct wined3d_shader *shader, DWORD sig_num,
3914         struct shader_arb_ctx_priv *priv_ctx, struct arb_vs_compiled_shader *compiled)
3915 {
3916     unsigned int i, j;
3917     static const char * const texcoords[8] =
3918     {
3919         "result.texcoord[0]", "result.texcoord[1]", "result.texcoord[2]", "result.texcoord[3]",
3920         "result.texcoord[4]", "result.texcoord[5]", "result.texcoord[6]", "result.texcoord[7]"
3921     };
3922     struct wined3d_device *device = shader->device;
3923     const struct wined3d_shader_signature_element *sig;
3924     const char *semantic_name;
3925     DWORD semantic_idx, reg_idx;
3926
3927     /* Write generic input varyings 0 to 7 to result.texcoord[], varying 8 to result.color.primary
3928      * and varying 9 to result.color.secondary
3929      */
3930     static const char * const decl_idx_to_string[MAX_REG_INPUT] =
3931     {
3932         "result.texcoord[0]", "result.texcoord[1]", "result.texcoord[2]", "result.texcoord[3]",
3933         "result.texcoord[4]", "result.texcoord[5]", "result.texcoord[6]", "result.texcoord[7]",
3934         "result.color.primary", "result.color.secondary"
3935     };
3936
3937     if(sig_num == ~0)
3938     {
3939         TRACE("Pixel shader uses builtin varyings\n");
3940         /* Map builtins to builtins */
3941         for(i = 0; i < 8; i++)
3942         {
3943             priv_ctx->texcrd_output[i] = texcoords[i];
3944         }
3945         priv_ctx->color_output[0] = "result.color.primary";
3946         priv_ctx->color_output[1] = "result.color.secondary";
3947         priv_ctx->fog_output = "result.fogcoord";
3948
3949         /* Map declared regs to builtins. Use "TA" to /dev/null unread output */
3950         for (i = 0; i < (sizeof(shader->output_signature) / sizeof(*shader->output_signature)); ++i)
3951         {
3952             semantic_name = shader->output_signature[i].semantic_name;
3953             if (!semantic_name) continue;
3954
3955             if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_POSITION))
3956             {
3957                 TRACE("o%u is TMP_OUT\n", i);
3958                 if (!shader->output_signature[i].semantic_idx) priv_ctx->vs_output[i] = "TMP_OUT";
3959                 else priv_ctx->vs_output[i] = "TA";
3960             }
3961             else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_PSIZE))
3962             {
3963                 TRACE("o%u is result.pointsize\n", i);
3964                 if (!shader->output_signature[i].semantic_idx) priv_ctx->vs_output[i] = "result.pointsize";
3965                 else priv_ctx->vs_output[i] = "TA";
3966             }
3967             else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_COLOR))
3968             {
3969                 TRACE("o%u is result.color.?, idx %u\n", i, shader->output_signature[i].semantic_idx);
3970                 if (!shader->output_signature[i].semantic_idx)
3971                     priv_ctx->vs_output[i] = "result.color.primary";
3972                 else if (shader->output_signature[i].semantic_idx == 1)
3973                     priv_ctx->vs_output[i] = "result.color.secondary";
3974                 else priv_ctx->vs_output[i] = "TA";
3975             }
3976             else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_TEXCOORD))
3977             {
3978                 TRACE("o%u is %s\n", i, texcoords[shader->output_signature[i].semantic_idx]);
3979                 if (shader->output_signature[i].semantic_idx >= 8) priv_ctx->vs_output[i] = "TA";
3980                 else priv_ctx->vs_output[i] = texcoords[shader->output_signature[i].semantic_idx];
3981             }
3982             else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_FOG))
3983             {
3984                 TRACE("o%u is result.fogcoord\n", i);
3985                 if (shader->output_signature[i].semantic_idx > 0) priv_ctx->vs_output[i] = "TA";
3986                 else priv_ctx->vs_output[i] = "result.fogcoord";
3987             }
3988             else
3989             {
3990                 priv_ctx->vs_output[i] = "TA";
3991             }
3992         }
3993         return;
3994     }
3995
3996     /* Instead of searching for the signature in the signature list, read the one from the current pixel shader.
3997      * Its maybe not the shader where the signature came from, but it is the same signature and faster to find
3998      */
3999     sig = device->stateBlock->state.pixel_shader->input_signature;
4000     TRACE("Pixel shader uses declared varyings\n");
4001
4002     /* Map builtin to declared. /dev/null the results by default to the TA temp reg */
4003     for(i = 0; i < 8; i++)
4004     {
4005         priv_ctx->texcrd_output[i] = "TA";
4006     }
4007     priv_ctx->color_output[0] = "TA";
4008     priv_ctx->color_output[1] = "TA";
4009     priv_ctx->fog_output = "TA";
4010
4011     for(i = 0; i < MAX_REG_INPUT; i++)
4012     {
4013         semantic_name = sig[i].semantic_name;
4014         semantic_idx = sig[i].semantic_idx;
4015         reg_idx = sig[i].register_idx;
4016         if (!semantic_name) continue;
4017
4018         /* If a declared input register is not written by builtin arguments, don't write to it.
4019          * GL_NV_vertex_program makes sure the input defaults to 0.0, which is correct with D3D
4020          *
4021          * Don't care about POSITION and PSIZE here - this is a builtin vertex shader, position goes
4022          * to TMP_OUT in any case
4023          */
4024         if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_TEXCOORD))
4025         {
4026             if(semantic_idx < 8) priv_ctx->texcrd_output[semantic_idx] = decl_idx_to_string[reg_idx];
4027         }
4028         else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_COLOR))
4029         {
4030             if(semantic_idx < 2) priv_ctx->color_output[semantic_idx] = decl_idx_to_string[reg_idx];
4031         }
4032         else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_FOG))
4033         {
4034             if (!semantic_idx) priv_ctx->fog_output = decl_idx_to_string[reg_idx];
4035         }
4036         else
4037         {
4038             continue;
4039         }
4040
4041         if (!strcmp(decl_idx_to_string[reg_idx], "result.color.primary")
4042                 || !strcmp(decl_idx_to_string[reg_idx], "result.color.secondary"))
4043         {
4044             compiled->need_color_unclamp = TRUE;
4045         }
4046     }
4047
4048     /* Map declared to declared */
4049     for (i = 0; i < (sizeof(shader->output_signature) / sizeof(*shader->output_signature)); ++i)
4050     {
4051         /* Write unread output to TA to throw them away */
4052         priv_ctx->vs_output[i] = "TA";
4053         semantic_name = shader->output_signature[i].semantic_name;
4054         if (!semantic_name) continue;
4055
4056         if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_POSITION)
4057                 && !shader->output_signature[i].semantic_idx)
4058         {
4059             priv_ctx->vs_output[i] = "TMP_OUT";
4060             continue;
4061         }
4062         else if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_PSIZE)
4063                 && !shader->output_signature[i].semantic_idx)
4064         {
4065             priv_ctx->vs_output[i] = "result.pointsize";
4066             continue;
4067         }
4068
4069         for(j = 0; j < MAX_REG_INPUT; j++)
4070         {
4071             if (!sig[j].semantic_name) continue;
4072
4073             if (!strcmp(sig[j].semantic_name, semantic_name)
4074                     && sig[j].semantic_idx == shader->output_signature[i].semantic_idx)
4075             {
4076                 priv_ctx->vs_output[i] = decl_idx_to_string[sig[j].register_idx];
4077
4078                 if (!strcmp(priv_ctx->vs_output[i], "result.color.primary")
4079                         || !strcmp(priv_ctx->vs_output[i], "result.color.secondary"))
4080                 {
4081                     compiled->need_color_unclamp = TRUE;
4082                 }
4083             }
4084         }
4085     }
4086 }
4087
4088 /* GL locking is done by the caller */
4089 static GLuint shader_arb_generate_vshader(struct wined3d_shader *shader,
4090         const struct wined3d_gl_info *gl_info, struct wined3d_shader_buffer *buffer,
4091         const struct arb_vs_compile_args *args, struct arb_vs_compiled_shader *compiled)
4092 {
4093     const struct arb_vshader_private *shader_data = shader->backend_data;
4094     const struct wined3d_shader_reg_maps *reg_maps = &shader->reg_maps;
4095     const DWORD *function = shader->function;
4096     const local_constant *lconst;
4097     GLuint ret;
4098     DWORD next_local, *lconst_map = local_const_mapping(shader);
4099     struct shader_arb_ctx_priv priv_ctx;
4100     unsigned int i;
4101     GLint errPos;
4102
4103     memset(&priv_ctx, 0, sizeof(priv_ctx));
4104     priv_ctx.cur_vs_args = args;
4105     list_init(&priv_ctx.control_frames);
4106     init_output_registers(shader, args->ps_signature, &priv_ctx, compiled);
4107
4108     /*  Create the hw ARB shader */
4109     shader_addline(buffer, "!!ARBvp1.0\n");
4110
4111     /* Always enable the NV extension if available. Unlike fragment shaders, there is no
4112      * mesurable performance penalty, and we can always make use of it for clipplanes.
4113      */
4114     if (gl_info->supported[NV_VERTEX_PROGRAM3])
4115     {
4116         shader_addline(buffer, "OPTION NV_vertex_program3;\n");
4117         priv_ctx.target_version = NV3;
4118         shader_addline(buffer, "ADDRESS aL;\n");
4119     }
4120     else if (gl_info->supported[NV_VERTEX_PROGRAM2_OPTION])
4121     {
4122         shader_addline(buffer, "OPTION NV_vertex_program2;\n");
4123         priv_ctx.target_version = NV2;
4124         shader_addline(buffer, "ADDRESS aL;\n");
4125     } else {
4126         priv_ctx.target_version = ARB;
4127     }
4128
4129     shader_addline(buffer, "TEMP TMP_OUT;\n");
4130     if (need_helper_const(shader_data, reg_maps, gl_info))
4131     {
4132         shader_addline(buffer, "PARAM helper_const = { 0.0, 1.0, 2.0, %1.10f};\n", eps);
4133     }
4134     if (need_rel_addr_const(shader_data, reg_maps, gl_info))
4135     {
4136         shader_addline(buffer, "PARAM rel_addr_const = { 0.5, %d.0, 0.0, 0.0 };\n", shader_data->rel_offset);
4137         shader_addline(buffer, "TEMP A0_SHADOW;\n");
4138     }
4139
4140     shader_addline(buffer, "TEMP TA;\n");
4141     shader_addline(buffer, "TEMP TB;\n");
4142
4143     /* Base Declarations */
4144     next_local = shader_generate_arb_declarations(shader, reg_maps, buffer,
4145             gl_info, lconst_map, &priv_ctx.vs_clipplanes, &priv_ctx);
4146
4147     for(i = 0; i < MAX_CONST_I; i++)
4148     {
4149         compiled->int_consts[i] = WINED3D_CONST_NUM_UNUSED;
4150         if(reg_maps->integer_constants & (1 << i) && priv_ctx.target_version >= NV2)
4151         {
4152             const DWORD *control_values = find_loop_control_values(shader, i);
4153
4154             if(control_values)
4155             {
4156                 shader_addline(buffer, "PARAM I%u = {%u, %u, %u, -1};\n", i,
4157                                 control_values[0], control_values[1], control_values[2]);
4158             }
4159             else
4160             {
4161                 compiled->int_consts[i] = next_local;
4162                 compiled->num_int_consts++;
4163                 shader_addline(buffer, "PARAM I%u = program.local[%u];\n", i, next_local++);
4164             }
4165         }
4166     }
4167
4168     /* We need a constant to fixup the final position */
4169     shader_addline(buffer, "PARAM posFixup = program.local[%u];\n", next_local);
4170     compiled->pos_fixup = next_local++;
4171
4172     /* Initialize output parameters. GL_ARB_vertex_program does not require special initialization values
4173      * for output parameters. D3D in theory does not do that either, but some applications depend on a
4174      * proper initialization of the secondary color, and programs using the fixed function pipeline without
4175      * a replacement shader depend on the texcoord.w being set properly.
4176      *
4177      * GL_NV_vertex_program defines that all output values are initialized to {0.0, 0.0, 0.0, 1.0}. This
4178      * assertion is in effect even when using GL_ARB_vertex_program without any NV specific additions. So
4179      * skip this if NV_vertex_program is supported. Otherwise, initialize the secondary color. For the tex-
4180      * coords, we have a flag in the opengl caps. Many cards do not require the texcoord being set, and
4181      * this can eat a number of instructions, so skip it unless this cap is set as well
4182      */
4183     if (!gl_info->supported[NV_VERTEX_PROGRAM])
4184     {
4185         struct wined3d_device *device = shader->device;
4186         const char *color_init = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_0001);
4187         shader_addline(buffer, "MOV result.color.secondary, %s;\n", color_init);
4188
4189         if (gl_info->quirks & WINED3D_QUIRK_SET_TEXCOORD_W && !device->frag_pipe->ffp_proj_control)
4190         {
4191             int i;
4192             const char *one = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_ONE);
4193             for(i = 0; i < min(8, MAX_REG_TEXCRD); i++)
4194             {
4195                 if (reg_maps->texcoord_mask[i] && reg_maps->texcoord_mask[i] != WINED3DSP_WRITEMASK_ALL)
4196                     shader_addline(buffer, "MOV result.texcoord[%u].w, %s\n", i, one);
4197             }
4198         }
4199     }
4200
4201     /* The shader starts with the main function */
4202     priv_ctx.in_main_func = TRUE;
4203     /* Base Shader Body */
4204     shader_generate_main(shader, buffer, reg_maps, function, &priv_ctx);
4205
4206     if (!priv_ctx.footer_written) vshader_add_footer(&priv_ctx,
4207             shader_data, args, reg_maps, gl_info, buffer);
4208
4209     shader_addline(buffer, "END\n");
4210
4211     /* TODO: change to resource.glObjectHandle or something like that */
4212     GL_EXTCALL(glGenProgramsARB(1, &ret));
4213
4214     TRACE("Creating a hw vertex shader, prg=%d\n", ret);
4215     GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, ret));
4216
4217     TRACE("Created hw vertex shader, prg=%d\n", ret);
4218     /* Create the program and check for errors */
4219     GL_EXTCALL(glProgramStringARB(GL_VERTEX_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
4220                buffer->bsize, buffer->buffer));
4221     checkGLcall("glProgramStringARB()");
4222
4223     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &errPos);
4224     if (errPos != -1)
4225     {
4226         FIXME("HW VertexShader Error at position %d: %s\n\n",
4227               errPos, debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
4228         shader_arb_dump_program_source(buffer->buffer);
4229         ret = -1;
4230     }
4231     else
4232     {
4233         GLint native;
4234
4235         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
4236         checkGLcall("glGetProgramivARB()");
4237         if (!native) WARN("Program exceeds native resource limits.\n");
4238
4239         /* Load immediate constants */
4240         if (lconst_map)
4241         {
4242             LIST_FOR_EACH_ENTRY(lconst, &shader->constantsF, local_constant, entry)
4243             {
4244                 const float *value = (const float *)lconst->value;
4245                 GL_EXTCALL(glProgramLocalParameter4fvARB(GL_VERTEX_PROGRAM_ARB, lconst_map[lconst->idx], value));
4246             }
4247         }
4248     }
4249     HeapFree(GetProcessHeap(), 0, lconst_map);
4250
4251     return ret;
4252 }
4253
4254 /* GL locking is done by the caller */
4255 static struct arb_ps_compiled_shader *find_arb_pshader(struct wined3d_shader *shader,
4256         const struct arb_ps_compile_args *args)
4257 {
4258     struct wined3d_device *device = shader->device;
4259     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4260     UINT i;
4261     DWORD new_size;
4262     struct arb_ps_compiled_shader *new_array;
4263     struct wined3d_shader_buffer buffer;
4264     struct arb_pshader_private *shader_data;
4265     GLuint ret;
4266
4267     if (!shader->backend_data)
4268     {
4269         struct shader_arb_priv *priv = device->shader_priv;
4270
4271         shader->backend_data = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data));
4272         shader_data = shader->backend_data;
4273         shader_data->clamp_consts = shader->reg_maps.shader_version.major == 1;
4274
4275         if (shader->reg_maps.shader_version.major < 3)
4276             shader_data->input_signature_idx = ~0;
4277         else
4278             shader_data->input_signature_idx = find_input_signature(priv, shader->input_signature);
4279
4280         shader_data->has_signature_idx = TRUE;
4281         TRACE("Shader got assigned input signature index %u\n", shader_data->input_signature_idx);
4282
4283         if (!device->vs_clipping)
4284             shader_data->clipplane_emulation = shader_find_free_input_register(&shader->reg_maps,
4285                     gl_info->limits.texture_stages - 1);
4286         else
4287             shader_data->clipplane_emulation = ~0U;
4288     }
4289     shader_data = shader->backend_data;
4290
4291     /* Usually we have very few GL shaders for each d3d shader(just 1 or maybe 2),
4292      * so a linear search is more performant than a hashmap or a binary search
4293      * (cache coherency etc)
4294      */
4295     for (i = 0; i < shader_data->num_gl_shaders; ++i)
4296     {
4297         if (!memcmp(&shader_data->gl_shaders[i].args, args, sizeof(*args)))
4298             return &shader_data->gl_shaders[i];
4299     }
4300
4301     TRACE("No matching GL shader found, compiling a new shader\n");
4302     if(shader_data->shader_array_size == shader_data->num_gl_shaders) {
4303         if (shader_data->num_gl_shaders)
4304         {
4305             new_size = shader_data->shader_array_size + max(1, shader_data->shader_array_size / 2);
4306             new_array = HeapReAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, shader_data->gl_shaders,
4307                                     new_size * sizeof(*shader_data->gl_shaders));
4308         } else {
4309             new_array = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data->gl_shaders));
4310             new_size = 1;
4311         }
4312
4313         if(!new_array) {
4314             ERR("Out of memory\n");
4315             return 0;
4316         }
4317         shader_data->gl_shaders = new_array;
4318         shader_data->shader_array_size = new_size;
4319     }
4320
4321     shader_data->gl_shaders[shader_data->num_gl_shaders].args = *args;
4322
4323     pixelshader_update_samplers(&shader->reg_maps, device->stateBlock->state.textures);
4324
4325     if (!shader_buffer_init(&buffer))
4326     {
4327         ERR("Failed to initialize shader buffer.\n");
4328         return 0;
4329     }
4330
4331     ret = shader_arb_generate_pshader(shader, gl_info, &buffer, args,
4332             &shader_data->gl_shaders[shader_data->num_gl_shaders]);
4333     shader_buffer_free(&buffer);
4334     shader_data->gl_shaders[shader_data->num_gl_shaders].prgId = ret;
4335
4336     return &shader_data->gl_shaders[shader_data->num_gl_shaders++];
4337 }
4338
4339 static inline BOOL vs_args_equal(const struct arb_vs_compile_args *stored, const struct arb_vs_compile_args *new,
4340                                  const DWORD use_map, BOOL skip_int) {
4341     if((stored->super.swizzle_map & use_map) != new->super.swizzle_map) return FALSE;
4342     if(stored->super.clip_enabled != new->super.clip_enabled) return FALSE;
4343     if(stored->super.fog_src != new->super.fog_src) return FALSE;
4344     if(stored->clip.boolclip_compare != new->clip.boolclip_compare) return FALSE;
4345     if(stored->ps_signature != new->ps_signature) return FALSE;
4346     if(stored->vertex.samplers_compare != new->vertex.samplers_compare) return FALSE;
4347     if(skip_int) return TRUE;
4348
4349     return !memcmp(stored->loop_ctrl, new->loop_ctrl, sizeof(stored->loop_ctrl));
4350 }
4351
4352 static struct arb_vs_compiled_shader *find_arb_vshader(struct wined3d_shader *shader,
4353         const struct arb_vs_compile_args *args)
4354 {
4355     struct wined3d_device *device = shader->device;
4356     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4357     DWORD use_map = device->strided_streams.use_map;
4358     UINT i;
4359     DWORD new_size;
4360     struct arb_vs_compiled_shader *new_array;
4361     struct wined3d_shader_buffer buffer;
4362     struct arb_vshader_private *shader_data;
4363     GLuint ret;
4364
4365     if (!shader->backend_data)
4366     {
4367         const struct wined3d_shader_reg_maps *reg_maps = &shader->reg_maps;
4368
4369         shader->backend_data = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data));
4370         shader_data = shader->backend_data;
4371
4372         if ((gl_info->quirks & WINED3D_QUIRK_ARB_VS_OFFSET_LIMIT)
4373                 && reg_maps->min_rel_offset <= reg_maps->max_rel_offset)
4374         {
4375             if (reg_maps->max_rel_offset - reg_maps->min_rel_offset > 127)
4376             {
4377                 FIXME("The difference between the minimum and maximum relative offset is > 127.\n");
4378                 FIXME("Which this OpenGL implementation does not support. Try using GLSL.\n");
4379                 FIXME("Min: %u, Max: %u.\n", reg_maps->min_rel_offset, reg_maps->max_rel_offset);
4380             }
4381             else if (reg_maps->max_rel_offset - reg_maps->min_rel_offset > 63)
4382                 shader_data->rel_offset = reg_maps->min_rel_offset + 63;
4383             else if (reg_maps->max_rel_offset > 63)
4384                 shader_data->rel_offset = reg_maps->min_rel_offset;
4385         }
4386     }
4387     shader_data = shader->backend_data;
4388
4389     /* Usually we have very few GL shaders for each d3d shader(just 1 or maybe 2),
4390      * so a linear search is more performant than a hashmap or a binary search
4391      * (cache coherency etc)
4392      */
4393     for(i = 0; i < shader_data->num_gl_shaders; i++) {
4394         if (vs_args_equal(&shader_data->gl_shaders[i].args, args,
4395                 use_map, gl_info->supported[NV_VERTEX_PROGRAM2_OPTION]))
4396         {
4397             return &shader_data->gl_shaders[i];
4398         }
4399     }
4400
4401     TRACE("No matching GL shader found, compiling a new shader\n");
4402
4403     if(shader_data->shader_array_size == shader_data->num_gl_shaders) {
4404         if (shader_data->num_gl_shaders)
4405         {
4406             new_size = shader_data->shader_array_size + max(1, shader_data->shader_array_size / 2);
4407             new_array = HeapReAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, shader_data->gl_shaders,
4408                                     new_size * sizeof(*shader_data->gl_shaders));
4409         } else {
4410             new_array = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data->gl_shaders));
4411             new_size = 1;
4412         }
4413
4414         if(!new_array) {
4415             ERR("Out of memory\n");
4416             return 0;
4417         }
4418         shader_data->gl_shaders = new_array;
4419         shader_data->shader_array_size = new_size;
4420     }
4421
4422     shader_data->gl_shaders[shader_data->num_gl_shaders].args = *args;
4423
4424     if (!shader_buffer_init(&buffer))
4425     {
4426         ERR("Failed to initialize shader buffer.\n");
4427         return 0;
4428     }
4429
4430     ret = shader_arb_generate_vshader(shader, gl_info, &buffer, args,
4431             &shader_data->gl_shaders[shader_data->num_gl_shaders]);
4432     shader_buffer_free(&buffer);
4433     shader_data->gl_shaders[shader_data->num_gl_shaders].prgId = ret;
4434
4435     return &shader_data->gl_shaders[shader_data->num_gl_shaders++];
4436 }
4437
4438 static void find_arb_ps_compile_args(const struct wined3d_state *state,
4439         const struct wined3d_shader *shader, struct arb_ps_compile_args *args)
4440 {
4441     struct wined3d_device *device = shader->device;
4442     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4443     int i;
4444     WORD int_skip;
4445
4446     find_ps_compile_args(state, shader, &args->super);
4447
4448     /* This forces all local boolean constants to 1 to make them stateblock independent */
4449     args->bools = shader->reg_maps.local_bool_consts;
4450
4451     for(i = 0; i < MAX_CONST_B; i++)
4452     {
4453         if (state->ps_consts_b[i])
4454             args->bools |= ( 1 << i);
4455     }
4456
4457     /* Only enable the clip plane emulation KIL if at least one clipplane is enabled. The KIL instruction
4458      * is quite expensive because it forces the driver to disable early Z discards. It is cheaper to
4459      * duplicate the shader than have a no-op KIL instruction in every shader
4460      */
4461     if (!device->vs_clipping && use_vs(state)
4462             && state->render_states[WINED3DRS_CLIPPING]
4463             && state->render_states[WINED3DRS_CLIPPLANEENABLE])
4464         args->clip = 1;
4465     else
4466         args->clip = 0;
4467
4468     /* Skip if unused or local, or supported natively */
4469     int_skip = ~shader->reg_maps.integer_constants | shader->reg_maps.local_int_consts;
4470     if (int_skip == 0xffff || gl_info->supported[NV_FRAGMENT_PROGRAM_OPTION])
4471     {
4472         memset(&args->loop_ctrl, 0, sizeof(args->loop_ctrl));
4473         return;
4474     }
4475
4476     for(i = 0; i < MAX_CONST_I; i++)
4477     {
4478         if(int_skip & (1 << i))
4479         {
4480             args->loop_ctrl[i][0] = 0;
4481             args->loop_ctrl[i][1] = 0;
4482             args->loop_ctrl[i][2] = 0;
4483         }
4484         else
4485         {
4486             args->loop_ctrl[i][0] = state->ps_consts_i[i * 4];
4487             args->loop_ctrl[i][1] = state->ps_consts_i[i * 4 + 1];
4488             args->loop_ctrl[i][2] = state->ps_consts_i[i * 4 + 2];
4489         }
4490     }
4491 }
4492
4493 static void find_arb_vs_compile_args(const struct wined3d_state *state,
4494         const struct wined3d_shader *shader, struct arb_vs_compile_args *args)
4495 {
4496     struct wined3d_device *device = shader->device;
4497     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4498     int i;
4499     WORD int_skip;
4500
4501     find_vs_compile_args(state, shader, &args->super);
4502
4503     args->clip.boolclip_compare = 0;
4504     if (use_ps(state))
4505     {
4506         const struct wined3d_shader *ps = state->pixel_shader;
4507         const struct arb_pshader_private *shader_priv = ps->backend_data;
4508         args->ps_signature = shader_priv->input_signature_idx;
4509
4510         args->clip.boolclip.clip_texcoord = shader_priv->clipplane_emulation + 1;
4511     }
4512     else
4513     {
4514         args->ps_signature = ~0;
4515         if (!device->vs_clipping)
4516         {
4517             args->clip.boolclip.clip_texcoord = ffp_clip_emul(state) ? gl_info->limits.texture_stages : 0;
4518         }
4519         /* Otherwise: Setting boolclip_compare set clip_texcoord to 0 */
4520     }
4521
4522     if (args->clip.boolclip.clip_texcoord)
4523     {
4524         if (state->render_states[WINED3DRS_CLIPPING])
4525             args->clip.boolclip.clipplane_mask = (unsigned char)state->render_states[WINED3DRS_CLIPPLANEENABLE];
4526         /* clipplane_mask was set to 0 by setting boolclip_compare to 0 */
4527     }
4528
4529     /* This forces all local boolean constants to 1 to make them stateblock independent */
4530     args->clip.boolclip.bools = shader->reg_maps.local_bool_consts;
4531     /* TODO: Figure out if it would be better to store bool constants as bitmasks in the stateblock */
4532     for(i = 0; i < MAX_CONST_B; i++)
4533     {
4534         if (state->vs_consts_b[i])
4535             args->clip.boolclip.bools |= ( 1 << i);
4536     }
4537
4538     args->vertex.samplers[0] = device->texUnitMap[MAX_FRAGMENT_SAMPLERS + 0];
4539     args->vertex.samplers[1] = device->texUnitMap[MAX_FRAGMENT_SAMPLERS + 1];
4540     args->vertex.samplers[2] = device->texUnitMap[MAX_FRAGMENT_SAMPLERS + 2];
4541     args->vertex.samplers[3] = 0;
4542
4543     /* Skip if unused or local */
4544     int_skip = ~shader->reg_maps.integer_constants | shader->reg_maps.local_int_consts;
4545     /* This is about flow control, not clipping. */
4546     if (int_skip == 0xffff || gl_info->supported[NV_VERTEX_PROGRAM2_OPTION])
4547     {
4548         memset(&args->loop_ctrl, 0, sizeof(args->loop_ctrl));
4549         return;
4550     }
4551
4552     for(i = 0; i < MAX_CONST_I; i++)
4553     {
4554         if(int_skip & (1 << i))
4555         {
4556             args->loop_ctrl[i][0] = 0;
4557             args->loop_ctrl[i][1] = 0;
4558             args->loop_ctrl[i][2] = 0;
4559         }
4560         else
4561         {
4562             args->loop_ctrl[i][0] = state->vs_consts_i[i * 4];
4563             args->loop_ctrl[i][1] = state->vs_consts_i[i * 4 + 1];
4564             args->loop_ctrl[i][2] = state->vs_consts_i[i * 4 + 2];
4565         }
4566     }
4567 }
4568
4569 /* GL locking is done by the caller */
4570 static void shader_arb_select(const struct wined3d_context *context, BOOL usePS, BOOL useVS)
4571 {
4572     struct wined3d_device *device = context->swapchain->device;
4573     struct shader_arb_priv *priv = device->shader_priv;
4574     const struct wined3d_gl_info *gl_info = context->gl_info;
4575     const struct wined3d_state *state = &device->stateBlock->state;
4576     int i;
4577
4578     /* Deal with pixel shaders first so the vertex shader arg function has the input signature ready */
4579     if (usePS)
4580     {
4581         struct wined3d_shader *ps = state->pixel_shader;
4582         struct arb_ps_compile_args compile_args;
4583         struct arb_ps_compiled_shader *compiled;
4584
4585         TRACE("Using pixel shader %p.\n", ps);
4586         find_arb_ps_compile_args(state, ps, &compile_args);
4587         compiled = find_arb_pshader(ps, &compile_args);
4588         priv->current_fprogram_id = compiled->prgId;
4589         priv->compiled_fprog = compiled;
4590
4591         /* Bind the fragment program */
4592         GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, priv->current_fprogram_id));
4593         checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, priv->current_fprogram_id);");
4594
4595         if(!priv->use_arbfp_fixed_func) {
4596             /* Enable OpenGL fragment programs */
4597             glEnable(GL_FRAGMENT_PROGRAM_ARB);
4598             checkGLcall("glEnable(GL_FRAGMENT_PROGRAM_ARB);");
4599         }
4600         TRACE("(%p) : Bound fragment program %u and enabled GL_FRAGMENT_PROGRAM_ARB\n",
4601                 device, priv->current_fprogram_id);
4602
4603         /* Pixel Shader 1.x constants are clamped to [-1;1], Pixel Shader 2.0 constants are not. If switching between
4604          * a 1.x and newer shader, reload the first 8 constants
4605          */
4606         if (priv->last_ps_const_clamped != ((struct arb_pshader_private *)ps->backend_data)->clamp_consts)
4607         {
4608             priv->last_ps_const_clamped = ((struct arb_pshader_private *)ps->backend_data)->clamp_consts;
4609             device->highest_dirty_ps_const = max(device->highest_dirty_ps_const, 8);
4610             for(i = 0; i < 8; i++)
4611             {
4612                 context->pshader_const_dirty[i] = 1;
4613             }
4614             /* Also takes care of loading local constants */
4615             shader_arb_load_constants(context, TRUE, FALSE);
4616         }
4617         else
4618         {
4619             UINT rt_height = device->fb.render_targets[0]->resource.height;
4620             shader_arb_ps_local_constants(compiled, context, state, rt_height);
4621         }
4622
4623         /* Force constant reloading for the NP2 fixup (see comment in shader_glsl_select for more info) */
4624         if (compiled->np2fixup_info.super.active)
4625             shader_arb_load_np2fixup_constants(priv, gl_info, state);
4626     }
4627     else if (gl_info->supported[ARB_FRAGMENT_PROGRAM] && !priv->use_arbfp_fixed_func)
4628     {
4629         /* Disable only if we're not using arbfp fixed function fragment processing. If this is used,
4630         * keep GL_FRAGMENT_PROGRAM_ARB enabled, and the fixed function pipeline will bind the fixed function
4631         * replacement shader
4632         */
4633         glDisable(GL_FRAGMENT_PROGRAM_ARB);
4634         checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
4635         priv->current_fprogram_id = 0;
4636     }
4637
4638     if (useVS)
4639     {
4640         struct wined3d_shader *vs = state->vertex_shader;
4641         struct arb_vs_compile_args compile_args;
4642         struct arb_vs_compiled_shader *compiled;
4643
4644         TRACE("Using vertex shader %p\n", vs);
4645         find_arb_vs_compile_args(state, vs, &compile_args);
4646         compiled = find_arb_vshader(vs, &compile_args);
4647         priv->current_vprogram_id = compiled->prgId;
4648         priv->compiled_vprog = compiled;
4649
4650         /* Bind the vertex program */
4651         GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->current_vprogram_id));
4652         checkGLcall("glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->current_vprogram_id);");
4653
4654         /* Enable OpenGL vertex programs */
4655         glEnable(GL_VERTEX_PROGRAM_ARB);
4656         checkGLcall("glEnable(GL_VERTEX_PROGRAM_ARB);");
4657         TRACE("(%p) : Bound vertex program %u and enabled GL_VERTEX_PROGRAM_ARB\n", device, priv->current_vprogram_id);
4658         shader_arb_vs_local_constants(compiled, context, state);
4659
4660         if(priv->last_vs_color_unclamp != compiled->need_color_unclamp) {
4661             priv->last_vs_color_unclamp = compiled->need_color_unclamp;
4662
4663             if (gl_info->supported[ARB_COLOR_BUFFER_FLOAT])
4664             {
4665                 GL_EXTCALL(glClampColorARB(GL_CLAMP_VERTEX_COLOR_ARB, !compiled->need_color_unclamp));
4666                 checkGLcall("glClampColorARB");
4667             } else {
4668                 FIXME("vertex color clamp needs to be changed, but extension not supported.\n");
4669             }
4670         }
4671     }
4672     else if (gl_info->supported[ARB_VERTEX_PROGRAM])
4673     {
4674         priv->current_vprogram_id = 0;
4675         glDisable(GL_VERTEX_PROGRAM_ARB);
4676         checkGLcall("glDisable(GL_VERTEX_PROGRAM_ARB)");
4677     }
4678 }
4679
4680 /* GL locking is done by the caller */
4681 static void shader_arb_select_depth_blt(void *shader_priv, const struct wined3d_gl_info *gl_info,
4682         enum tex_types tex_type, const SIZE *ds_mask_size)
4683 {
4684     const float mask[] = {0.0f, 0.0f, (float)ds_mask_size->cx, (float)ds_mask_size->cy};
4685     BOOL masked = ds_mask_size->cx && ds_mask_size->cy;
4686     struct shader_arb_priv *priv = shader_priv;
4687     GLuint *blt_fprogram;
4688
4689     if (!priv->depth_blt_vprogram_id) priv->depth_blt_vprogram_id = create_arb_blt_vertex_program(gl_info);
4690     GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->depth_blt_vprogram_id));
4691     glEnable(GL_VERTEX_PROGRAM_ARB);
4692
4693     blt_fprogram = masked ? &priv->depth_blt_fprogram_id_masked[tex_type] : &priv->depth_blt_fprogram_id_full[tex_type];
4694     if (!*blt_fprogram) *blt_fprogram = create_arb_blt_fragment_program(gl_info, tex_type, masked);
4695     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, *blt_fprogram));
4696     if (masked) GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, 0, mask));
4697     glEnable(GL_FRAGMENT_PROGRAM_ARB);
4698 }
4699
4700 /* GL locking is done by the caller */
4701 static void shader_arb_deselect_depth_blt(void *shader_priv, const struct wined3d_gl_info *gl_info)
4702 {
4703     struct shader_arb_priv *priv = shader_priv;
4704
4705     if (priv->current_vprogram_id) {
4706         GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->current_vprogram_id));
4707         checkGLcall("glBindProgramARB(GL_VERTEX_PROGRAM_ARB, vertexShader->prgId);");
4708
4709         TRACE("Bound vertex program %u and enabled GL_VERTEX_PROGRAM_ARB.\n", priv->current_vprogram_id);
4710     }
4711     else
4712     {
4713         glDisable(GL_VERTEX_PROGRAM_ARB);
4714         checkGLcall("glDisable(GL_VERTEX_PROGRAM_ARB)");
4715     }
4716
4717     if (priv->current_fprogram_id) {
4718         GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, priv->current_fprogram_id));
4719         checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, pixelShader->prgId);");
4720
4721         TRACE("Bound fragment program %u and enabled GL_FRAGMENT_PROGRAM_ARB.\n", priv->current_fprogram_id);
4722     }
4723     else if(!priv->use_arbfp_fixed_func)
4724     {
4725         glDisable(GL_FRAGMENT_PROGRAM_ARB);
4726         checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
4727     }
4728 }
4729
4730 static void shader_arb_destroy(struct wined3d_shader *shader)
4731 {
4732     struct wined3d_device *device = shader->device;
4733     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4734
4735     if (shader_is_pshader_version(shader->reg_maps.shader_version.type))
4736     {
4737         struct arb_pshader_private *shader_data = shader->backend_data;
4738         UINT i;
4739
4740         if(!shader_data) return; /* This can happen if a shader was never compiled */
4741
4742         if (shader_data->num_gl_shaders)
4743         {
4744             struct wined3d_context *context = context_acquire(device, NULL);
4745
4746             ENTER_GL();
4747             for (i = 0; i < shader_data->num_gl_shaders; ++i)
4748             {
4749                 GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId));
4750                 checkGLcall("GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId))");
4751             }
4752             LEAVE_GL();
4753
4754             context_release(context);
4755         }
4756
4757         HeapFree(GetProcessHeap(), 0, shader_data->gl_shaders);
4758         HeapFree(GetProcessHeap(), 0, shader_data);
4759         shader->backend_data = NULL;
4760     }
4761     else
4762     {
4763         struct arb_vshader_private *shader_data = shader->backend_data;
4764         UINT i;
4765
4766         if(!shader_data) return; /* This can happen if a shader was never compiled */
4767
4768         if (shader_data->num_gl_shaders)
4769         {
4770             struct wined3d_context *context = context_acquire(device, NULL);
4771
4772             ENTER_GL();
4773             for (i = 0; i < shader_data->num_gl_shaders; ++i)
4774             {
4775                 GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId));
4776                 checkGLcall("GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId))");
4777             }
4778             LEAVE_GL();
4779
4780             context_release(context);
4781         }
4782
4783         HeapFree(GetProcessHeap(), 0, shader_data->gl_shaders);
4784         HeapFree(GetProcessHeap(), 0, shader_data);
4785         shader->backend_data = NULL;
4786     }
4787 }
4788
4789 static int sig_tree_compare(const void *key, const struct wine_rb_entry *entry)
4790 {
4791     struct ps_signature *e = WINE_RB_ENTRY_VALUE(entry, struct ps_signature, entry);
4792     return compare_sig(key, e->sig);
4793 }
4794
4795 static const struct wine_rb_functions sig_tree_functions =
4796 {
4797     wined3d_rb_alloc,
4798     wined3d_rb_realloc,
4799     wined3d_rb_free,
4800     sig_tree_compare
4801 };
4802
4803 static HRESULT shader_arb_alloc(struct wined3d_device *device)
4804 {
4805     struct shader_arb_priv *priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*priv));
4806     if(wine_rb_init(&priv->signature_tree, &sig_tree_functions) == -1)
4807     {
4808         ERR("RB tree init failed\n");
4809         HeapFree(GetProcessHeap(), 0, priv);
4810         return E_OUTOFMEMORY;
4811     }
4812     device->shader_priv = priv;
4813     return WINED3D_OK;
4814 }
4815
4816 static void release_signature(struct wine_rb_entry *entry, void *context)
4817 {
4818     struct ps_signature *sig = WINE_RB_ENTRY_VALUE(entry, struct ps_signature, entry);
4819     int i;
4820     for(i = 0; i < MAX_REG_INPUT; i++)
4821     {
4822         HeapFree(GetProcessHeap(), 0, (char *) sig->sig[i].semantic_name);
4823     }
4824     HeapFree(GetProcessHeap(), 0, sig->sig);
4825     HeapFree(GetProcessHeap(), 0, sig);
4826 }
4827
4828 /* Context activation is done by the caller. */
4829 static void shader_arb_free(struct wined3d_device *device)
4830 {
4831     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4832     struct shader_arb_priv *priv = device->shader_priv;
4833     int i;
4834
4835     ENTER_GL();
4836     if(priv->depth_blt_vprogram_id) {
4837         GL_EXTCALL(glDeleteProgramsARB(1, &priv->depth_blt_vprogram_id));
4838     }
4839     for (i = 0; i < tex_type_count; ++i)
4840     {
4841         if (priv->depth_blt_fprogram_id_full[i])
4842         {
4843             GL_EXTCALL(glDeleteProgramsARB(1, &priv->depth_blt_fprogram_id_full[i]));
4844         }
4845         if (priv->depth_blt_fprogram_id_masked[i])
4846         {
4847             GL_EXTCALL(glDeleteProgramsARB(1, &priv->depth_blt_fprogram_id_masked[i]));
4848         }
4849     }
4850     LEAVE_GL();
4851
4852     wine_rb_destroy(&priv->signature_tree, release_signature, NULL);
4853     HeapFree(GetProcessHeap(), 0, device->shader_priv);
4854 }
4855
4856 static BOOL shader_arb_dirty_const(void)
4857 {
4858     return TRUE;
4859 }
4860
4861 static void shader_arb_get_caps(const struct wined3d_gl_info *gl_info, struct shader_caps *caps)
4862 {
4863     if (gl_info->supported[ARB_VERTEX_PROGRAM])
4864     {
4865         DWORD vs_consts;
4866
4867         /* 96 is the minimum allowed value of MAX_PROGRAM_ENV_PARAMETERS_ARB
4868          * for vertex programs. If the native limit is less than that it's
4869          * not very useful, and e.g. Mesa swrast returns 0, probably to
4870          * indicate it's a software implementation. */
4871         if (gl_info->limits.arb_vs_native_constants < 96)
4872             vs_consts = gl_info->limits.arb_vs_float_constants;
4873         else
4874             vs_consts = min(gl_info->limits.arb_vs_float_constants, gl_info->limits.arb_vs_native_constants);
4875
4876         if (gl_info->supported[NV_VERTEX_PROGRAM3])
4877         {
4878             caps->VertexShaderVersion = WINED3DVS_VERSION(3,0);
4879             TRACE_(d3d_caps)("Hardware vertex shader version 3.0 enabled (NV_VERTEX_PROGRAM3)\n");
4880         }
4881         else if (vs_consts >= 256)
4882         {
4883             /* Shader Model 2.0 requires at least 256 vertex shader constants */
4884             caps->VertexShaderVersion = WINED3DVS_VERSION(2,0);
4885             TRACE_(d3d_caps)("Hardware vertex shader version 2.0 enabled (ARB_PROGRAM)\n");
4886         }
4887         else
4888         {
4889             caps->VertexShaderVersion = WINED3DVS_VERSION(1,1);
4890             TRACE_(d3d_caps)("Hardware vertex shader version 1.1 enabled (ARB_PROGRAM)\n");
4891         }
4892         caps->MaxVertexShaderConst = vs_consts;
4893     }
4894     else
4895     {
4896         caps->VertexShaderVersion = 0;
4897         caps->MaxVertexShaderConst = 0;
4898     }
4899
4900     if (gl_info->supported[ARB_FRAGMENT_PROGRAM])
4901     {
4902         DWORD ps_consts;
4903
4904         /* Similar as above for vertex programs, but the minimum for fragment
4905          * programs is 24. */
4906         if (gl_info->limits.arb_ps_native_constants < 24)
4907             ps_consts = gl_info->limits.arb_ps_float_constants;
4908         else
4909             ps_consts = min(gl_info->limits.arb_ps_float_constants, gl_info->limits.arb_ps_native_constants);
4910
4911         if (gl_info->supported[NV_FRAGMENT_PROGRAM2])
4912         {
4913             caps->PixelShaderVersion = WINED3DPS_VERSION(3,0);
4914             TRACE_(d3d_caps)("Hardware pixel shader version 3.0 enabled (NV_FRAGMENT_PROGRAM2)\n");
4915         }
4916         else if (ps_consts >= 32)
4917         {
4918             /* Shader Model 2.0 requires at least 32 pixel shader constants */
4919             caps->PixelShaderVersion = WINED3DPS_VERSION(2,0);
4920             TRACE_(d3d_caps)("Hardware pixel shader version 2.0 enabled (ARB_PROGRAM)\n");
4921         }
4922         else
4923         {
4924             caps->PixelShaderVersion = WINED3DPS_VERSION(1,4);
4925             TRACE_(d3d_caps)("Hardware pixel shader version 1.4 enabled (ARB_PROGRAM)\n");
4926         }
4927         caps->PixelShader1xMaxValue = 8.0f;
4928         caps->MaxPixelShaderConst = ps_consts;
4929     }
4930     else
4931     {
4932         caps->PixelShaderVersion = 0;
4933         caps->PixelShader1xMaxValue = 0.0f;
4934         caps->MaxPixelShaderConst = 0;
4935     }
4936
4937     caps->VSClipping = use_nv_clip(gl_info);
4938 }
4939
4940 static BOOL shader_arb_color_fixup_supported(struct color_fixup_desc fixup)
4941 {
4942     if (TRACE_ON(d3d_shader) && TRACE_ON(d3d))
4943     {
4944         TRACE("Checking support for color_fixup:\n");
4945         dump_color_fixup_desc(fixup);
4946     }
4947
4948     /* We support everything except complex conversions. */
4949     if (!is_complex_fixup(fixup))
4950     {
4951         TRACE("[OK]\n");
4952         return TRUE;
4953     }
4954
4955     TRACE("[FAILED]\n");
4956     return FALSE;
4957 }
4958
4959 static void shader_arb_add_instruction_modifiers(const struct wined3d_shader_instruction *ins) {
4960     DWORD shift;
4961     char write_mask[20], regstr[50];
4962     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
4963     BOOL is_color = FALSE;
4964     const struct wined3d_shader_dst_param *dst;
4965
4966     if (!ins->dst_count) return;
4967
4968     dst = &ins->dst[0];
4969     shift = dst->shift;
4970     if (!shift) return; /* Saturate alone is handled by the instructions */
4971
4972     shader_arb_get_write_mask(ins, dst, write_mask);
4973     shader_arb_get_register_name(ins, &dst->reg, regstr, &is_color);
4974
4975     /* Generate a line that does the output modifier computation
4976      * FIXME: _SAT vs shift? _SAT alone is already handled in the instructions, if this
4977      * maps problems in e.g. _d4_sat modify shader_arb_get_modifier
4978      */
4979     shader_addline(buffer, "MUL%s %s%s, %s, %s;\n", shader_arb_get_modifier(ins),
4980                    regstr, write_mask, regstr, shift_tab[shift]);
4981 }
4982
4983 static const SHADER_HANDLER shader_arb_instruction_handler_table[WINED3DSIH_TABLE_SIZE] =
4984 {
4985     /* WINED3DSIH_ABS           */ shader_hw_map2gl,
4986     /* WINED3DSIH_ADD           */ shader_hw_map2gl,
4987     /* WINED3DSIH_AND           */ NULL,
4988     /* WINED3DSIH_BEM           */ pshader_hw_bem,
4989     /* WINED3DSIH_BREAK         */ shader_hw_break,
4990     /* WINED3DSIH_BREAKC        */ shader_hw_breakc,
4991     /* WINED3DSIH_BREAKP        */ NULL,
4992     /* WINED3DSIH_CALL          */ shader_hw_call,
4993     /* WINED3DSIH_CALLNZ        */ NULL,
4994     /* WINED3DSIH_CMP           */ pshader_hw_cmp,
4995     /* WINED3DSIH_CND           */ pshader_hw_cnd,
4996     /* WINED3DSIH_CRS           */ shader_hw_map2gl,
4997     /* WINED3DSIH_CUT           */ NULL,
4998     /* WINED3DSIH_DCL           */ NULL,
4999     /* WINED3DSIH_DEF           */ NULL,
5000     /* WINED3DSIH_DEFB          */ NULL,
5001     /* WINED3DSIH_DEFI          */ NULL,
5002     /* WINED3DSIH_DIV           */ NULL,
5003     /* WINED3DSIH_DP2ADD        */ pshader_hw_dp2add,
5004     /* WINED3DSIH_DP3           */ shader_hw_map2gl,
5005     /* WINED3DSIH_DP4           */ shader_hw_map2gl,
5006     /* WINED3DSIH_DST           */ shader_hw_map2gl,
5007     /* WINED3DSIH_DSX           */ shader_hw_map2gl,
5008     /* WINED3DSIH_DSY           */ shader_hw_dsy,
5009     /* WINED3DSIH_ELSE          */ shader_hw_else,
5010     /* WINED3DSIH_EMIT          */ NULL,
5011     /* WINED3DSIH_ENDIF         */ shader_hw_endif,
5012     /* WINED3DSIH_ENDLOOP       */ shader_hw_endloop,
5013     /* WINED3DSIH_ENDREP        */ shader_hw_endrep,
5014     /* WINED3DSIH_EXP           */ shader_hw_scalar_op,
5015     /* WINED3DSIH_EXPP          */ shader_hw_scalar_op,
5016     /* WINED3DSIH_FRC           */ shader_hw_map2gl,
5017     /* WINED3DSIH_FTOI          */ NULL,
5018     /* WINED3DSIH_IADD          */ NULL,
5019     /* WINED3DSIH_IEQ           */ NULL,
5020     /* WINED3DSIH_IF            */ NULL /* Hardcoded into the shader */,
5021     /* WINED3DSIH_IFC           */ shader_hw_ifc,
5022     /* WINED3DSIH_IGE           */ NULL,
5023     /* WINED3DSIH_IMUL          */ NULL,
5024     /* WINED3DSIH_ITOF          */ NULL,
5025     /* WINED3DSIH_LABEL         */ shader_hw_label,
5026     /* WINED3DSIH_LD            */ NULL,
5027     /* WINED3DSIH_LIT           */ shader_hw_map2gl,
5028     /* WINED3DSIH_LOG           */ shader_hw_log,
5029     /* WINED3DSIH_LOGP          */ shader_hw_log,
5030     /* WINED3DSIH_LOOP          */ shader_hw_loop,
5031     /* WINED3DSIH_LRP           */ shader_hw_lrp,
5032     /* WINED3DSIH_LT            */ NULL,
5033     /* WINED3DSIH_M3x2          */ shader_hw_mnxn,
5034     /* WINED3DSIH_M3x3          */ shader_hw_mnxn,
5035     /* WINED3DSIH_M3x4          */ shader_hw_mnxn,
5036     /* WINED3DSIH_M4x3          */ shader_hw_mnxn,
5037     /* WINED3DSIH_M4x4          */ shader_hw_mnxn,
5038     /* WINED3DSIH_MAD           */ shader_hw_map2gl,
5039     /* WINED3DSIH_MAX           */ shader_hw_map2gl,
5040     /* WINED3DSIH_MIN           */ shader_hw_map2gl,
5041     /* WINED3DSIH_MOV           */ shader_hw_mov,
5042     /* WINED3DSIH_MOVA          */ shader_hw_mov,
5043     /* WINED3DSIH_MOVC          */ NULL,
5044     /* WINED3DSIH_MUL           */ shader_hw_map2gl,
5045     /* WINED3DSIH_NOP           */ shader_hw_nop,
5046     /* WINED3DSIH_NRM           */ shader_hw_nrm,
5047     /* WINED3DSIH_PHASE         */ NULL,
5048     /* WINED3DSIH_POW           */ shader_hw_pow,
5049     /* WINED3DSIH_RCP           */ shader_hw_rcp,
5050     /* WINED3DSIH_REP           */ shader_hw_rep,
5051     /* WINED3DSIH_RET           */ shader_hw_ret,
5052     /* WINED3DSIH_RSQ           */ shader_hw_scalar_op,
5053     /* WINED3DSIH_SAMPLE        */ NULL,
5054     /* WINED3DSIH_SAMPLE_GRAD   */ NULL,
5055     /* WINED3DSIH_SAMPLE_LOD    */ NULL,
5056     /* WINED3DSIH_SETP          */ NULL,
5057     /* WINED3DSIH_SGE           */ shader_hw_map2gl,
5058     /* WINED3DSIH_SGN           */ shader_hw_sgn,
5059     /* WINED3DSIH_SINCOS        */ shader_hw_sincos,
5060     /* WINED3DSIH_SLT           */ shader_hw_map2gl,
5061     /* WINED3DSIH_SQRT          */ NULL,
5062     /* WINED3DSIH_SUB           */ shader_hw_map2gl,
5063     /* WINED3DSIH_TEX           */ pshader_hw_tex,
5064     /* WINED3DSIH_TEXBEM        */ pshader_hw_texbem,
5065     /* WINED3DSIH_TEXBEML       */ pshader_hw_texbem,
5066     /* WINED3DSIH_TEXCOORD      */ pshader_hw_texcoord,
5067     /* WINED3DSIH_TEXDEPTH      */ pshader_hw_texdepth,
5068     /* WINED3DSIH_TEXDP3        */ pshader_hw_texdp3,
5069     /* WINED3DSIH_TEXDP3TEX     */ pshader_hw_texdp3tex,
5070     /* WINED3DSIH_TEXKILL       */ pshader_hw_texkill,
5071     /* WINED3DSIH_TEXLDD        */ shader_hw_texldd,
5072     /* WINED3DSIH_TEXLDL        */ shader_hw_texldl,
5073     /* WINED3DSIH_TEXM3x2DEPTH  */ pshader_hw_texm3x2depth,
5074     /* WINED3DSIH_TEXM3x2PAD    */ pshader_hw_texm3x2pad,
5075     /* WINED3DSIH_TEXM3x2TEX    */ pshader_hw_texm3x2tex,
5076     /* WINED3DSIH_TEXM3x3       */ pshader_hw_texm3x3,
5077     /* WINED3DSIH_TEXM3x3DIFF   */ NULL,
5078     /* WINED3DSIH_TEXM3x3PAD    */ pshader_hw_texm3x3pad,
5079     /* WINED3DSIH_TEXM3x3SPEC   */ pshader_hw_texm3x3spec,
5080     /* WINED3DSIH_TEXM3x3TEX    */ pshader_hw_texm3x3tex,
5081     /* WINED3DSIH_TEXM3x3VSPEC  */ pshader_hw_texm3x3vspec,
5082     /* WINED3DSIH_TEXREG2AR     */ pshader_hw_texreg2ar,
5083     /* WINED3DSIH_TEXREG2GB     */ pshader_hw_texreg2gb,
5084     /* WINED3DSIH_TEXREG2RGB    */ pshader_hw_texreg2rgb,
5085     /* WINED3DSIH_UTOF          */ NULL,
5086 };
5087
5088 static BOOL get_bool_const(const struct wined3d_shader_instruction *ins,
5089         const struct wined3d_shader *shader, DWORD idx)
5090 {
5091     const struct wined3d_shader_reg_maps *reg_maps = ins->ctx->reg_maps;
5092     BOOL vshader = shader_is_vshader_version(reg_maps->shader_version.type);
5093     WORD bools = 0;
5094     WORD flag = (1 << idx);
5095     const local_constant *constant;
5096     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
5097
5098     if (reg_maps->local_bool_consts & flag)
5099     {
5100         /* What good is a if(bool) with a hardcoded local constant? I don't know, but handle it */
5101         LIST_FOR_EACH_ENTRY(constant, &shader->constantsB, local_constant, entry)
5102         {
5103             if (constant->idx == idx)
5104             {
5105                 return constant->value[0];
5106             }
5107         }
5108         ERR("Local constant not found\n");
5109         return FALSE;
5110     }
5111     else
5112     {
5113         if(vshader) bools = priv->cur_vs_args->clip.boolclip.bools;
5114         else bools = priv->cur_ps_args->bools;
5115         return bools & flag;
5116     }
5117 }
5118
5119 static void get_loop_control_const(const struct wined3d_shader_instruction *ins,
5120         const struct wined3d_shader *shader, UINT idx, struct wined3d_shader_loop_control *loop_control)
5121 {
5122     const struct wined3d_shader_reg_maps *reg_maps = ins->ctx->reg_maps;
5123     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
5124
5125     /* Integer constants can either be a local constant, or they can be stored in the shader
5126      * type specific compile args. */
5127     if (reg_maps->local_int_consts & (1 << idx))
5128     {
5129         const local_constant *constant;
5130
5131         LIST_FOR_EACH_ENTRY(constant, &shader->constantsI, local_constant, entry)
5132         {
5133             if (constant->idx == idx)
5134             {
5135                 loop_control->count = constant->value[0];
5136                 loop_control->start = constant->value[1];
5137                 /* Step is signed. */
5138                 loop_control->step = (int)constant->value[2];
5139                 return;
5140             }
5141         }
5142         /* If this happens the flag was set incorrectly */
5143         ERR("Local constant not found\n");
5144         loop_control->count = 0;
5145         loop_control->start = 0;
5146         loop_control->step = 0;
5147         return;
5148     }
5149
5150     switch (reg_maps->shader_version.type)
5151     {
5152         case WINED3D_SHADER_TYPE_VERTEX:
5153             /* Count and aL start value are unsigned */
5154             loop_control->count = priv->cur_vs_args->loop_ctrl[idx][0];
5155             loop_control->start = priv->cur_vs_args->loop_ctrl[idx][1];
5156             /* Step is signed. */
5157             loop_control->step = ((char)priv->cur_vs_args->loop_ctrl[idx][2]);
5158             break;
5159
5160         case WINED3D_SHADER_TYPE_PIXEL:
5161             loop_control->count = priv->cur_ps_args->loop_ctrl[idx][0];
5162             loop_control->start = priv->cur_ps_args->loop_ctrl[idx][1];
5163             loop_control->step = ((char)priv->cur_ps_args->loop_ctrl[idx][2]);
5164             break;
5165
5166         default:
5167             FIXME("Unhandled shader type %#x.\n", reg_maps->shader_version.type);
5168             break;
5169     }
5170 }
5171
5172 static void record_instruction(struct list *list, const struct wined3d_shader_instruction *ins)
5173 {
5174     unsigned int i;
5175     struct wined3d_shader_dst_param *dst_param = NULL;
5176     struct wined3d_shader_src_param *src_param = NULL, *rel_addr = NULL;
5177     struct recorded_instruction *rec = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*rec));
5178     if(!rec)
5179     {
5180         ERR("Out of memory\n");
5181         return;
5182     }
5183
5184     rec->ins = *ins;
5185     dst_param = HeapAlloc(GetProcessHeap(), 0, sizeof(*dst_param));
5186     if(!dst_param) goto free;
5187     *dst_param = *ins->dst;
5188     if(ins->dst->reg.rel_addr)
5189     {
5190         rel_addr = HeapAlloc(GetProcessHeap(), 0, sizeof(*dst_param->reg.rel_addr));
5191         if(!rel_addr) goto free;
5192         *rel_addr = *ins->dst->reg.rel_addr;
5193         dst_param->reg.rel_addr = rel_addr;
5194     }
5195     rec->ins.dst = dst_param;
5196
5197     src_param = HeapAlloc(GetProcessHeap(), 0, sizeof(*src_param) * ins->src_count);
5198     if(!src_param) goto free;
5199     for(i = 0; i < ins->src_count; i++)
5200     {
5201         src_param[i] = ins->src[i];
5202         if(ins->src[i].reg.rel_addr)
5203         {
5204             rel_addr = HeapAlloc(GetProcessHeap(), 0, sizeof(*rel_addr));
5205             if(!rel_addr) goto free;
5206             *rel_addr = *ins->src[i].reg.rel_addr;
5207             src_param[i].reg.rel_addr = rel_addr;
5208         }
5209     }
5210     rec->ins.src = src_param;
5211     list_add_tail(list, &rec->entry);
5212     return;
5213
5214 free:
5215     ERR("Out of memory\n");
5216     if(dst_param)
5217     {
5218         HeapFree(GetProcessHeap(), 0, (void *) dst_param->reg.rel_addr);
5219         HeapFree(GetProcessHeap(), 0, dst_param);
5220     }
5221     if(src_param)
5222     {
5223         for(i = 0; i < ins->src_count; i++)
5224         {
5225             HeapFree(GetProcessHeap(), 0, (void *) src_param[i].reg.rel_addr);
5226         }
5227         HeapFree(GetProcessHeap(), 0, src_param);
5228     }
5229     HeapFree(GetProcessHeap(), 0, rec);
5230 }
5231
5232 static void free_recorded_instruction(struct list *list)
5233 {
5234     struct recorded_instruction *rec_ins, *entry2;
5235     unsigned int i;
5236
5237     LIST_FOR_EACH_ENTRY_SAFE(rec_ins, entry2, list, struct recorded_instruction, entry)
5238     {
5239         list_remove(&rec_ins->entry);
5240         if(rec_ins->ins.dst)
5241         {
5242             HeapFree(GetProcessHeap(), 0, (void *) rec_ins->ins.dst->reg.rel_addr);
5243             HeapFree(GetProcessHeap(), 0, (void *) rec_ins->ins.dst);
5244         }
5245         if(rec_ins->ins.src)
5246         {
5247             for(i = 0; i < rec_ins->ins.src_count; i++)
5248             {
5249                 HeapFree(GetProcessHeap(), 0, (void *) rec_ins->ins.src[i].reg.rel_addr);
5250             }
5251             HeapFree(GetProcessHeap(), 0, (void *) rec_ins->ins.src);
5252         }
5253         HeapFree(GetProcessHeap(), 0, rec_ins);
5254     }
5255 }
5256
5257 static void shader_arb_handle_instruction(const struct wined3d_shader_instruction *ins) {
5258     SHADER_HANDLER hw_fct;
5259     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
5260     const struct wined3d_shader *shader = ins->ctx->shader;
5261     struct control_frame *control_frame;
5262     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
5263     BOOL bool_const;
5264
5265     if(ins->handler_idx == WINED3DSIH_LOOP || ins->handler_idx == WINED3DSIH_REP)
5266     {
5267         control_frame = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*control_frame));
5268         list_add_head(&priv->control_frames, &control_frame->entry);
5269
5270         if(ins->handler_idx == WINED3DSIH_LOOP) control_frame->type = LOOP;
5271         if(ins->handler_idx == WINED3DSIH_REP) control_frame->type = REP;
5272
5273         if(priv->target_version >= NV2)
5274         {
5275             control_frame->no.loop = priv->num_loops++;
5276             priv->loop_depth++;
5277         }
5278         else
5279         {
5280             /* Don't bother recording when we're in a not used if branch */
5281             if(priv->muted)
5282             {
5283                 return;
5284             }
5285
5286             if(!priv->recording)
5287             {
5288                 list_init(&priv->record);
5289                 priv->recording = TRUE;
5290                 control_frame->outer_loop = TRUE;
5291                 get_loop_control_const(ins, shader, ins->src[0].reg.idx, &control_frame->loop_control);
5292                 return; /* Instruction is handled */
5293             }
5294             /* Record this loop in the outer loop's recording */
5295         }
5296     }
5297     else if(ins->handler_idx == WINED3DSIH_ENDLOOP || ins->handler_idx == WINED3DSIH_ENDREP)
5298     {
5299         if(priv->target_version >= NV2)
5300         {
5301             /* Nothing to do. The control frame is popped after the HW instr handler */
5302         }
5303         else
5304         {
5305             struct list *e = list_head(&priv->control_frames);
5306             control_frame = LIST_ENTRY(e, struct control_frame, entry);
5307             list_remove(&control_frame->entry);
5308
5309             if(control_frame->outer_loop)
5310             {
5311                 unsigned int iteration;
5312                 int aL = 0;
5313                 struct list copy;
5314
5315                 /* Turn off recording before playback */
5316                 priv->recording = FALSE;
5317
5318                 /* Move the recorded instructions to a separate list and get them out of the private data
5319                  * structure. If there are nested loops, the shader_arb_handle_instruction below will
5320                  * be recorded again, thus priv->record might be overwritten
5321                  */
5322                 list_init(&copy);
5323                 list_move_tail(&copy, &priv->record);
5324                 list_init(&priv->record);
5325
5326                 if(ins->handler_idx == WINED3DSIH_ENDLOOP)
5327                 {
5328                     shader_addline(buffer, "#unrolling loop: %u iterations, aL=%u, inc %d\n",
5329                                    control_frame->loop_control.count, control_frame->loop_control.start,
5330                                    control_frame->loop_control.step);
5331                     aL = control_frame->loop_control.start;
5332                 }
5333                 else
5334                 {
5335                     shader_addline(buffer, "#unrolling rep: %u iterations\n", control_frame->loop_control.count);
5336                 }
5337
5338                 for (iteration = 0; iteration < control_frame->loop_control.count; ++iteration)
5339                 {
5340                     struct recorded_instruction *rec_ins;
5341                     if(ins->handler_idx == WINED3DSIH_ENDLOOP)
5342                     {
5343                         priv->aL = aL;
5344                         shader_addline(buffer, "#Iteration %u, aL=%d\n", iteration, aL);
5345                     }
5346                     else
5347                     {
5348                         shader_addline(buffer, "#Iteration %u\n", iteration);
5349                     }
5350
5351                     LIST_FOR_EACH_ENTRY(rec_ins, &copy, struct recorded_instruction, entry)
5352                     {
5353                         shader_arb_handle_instruction(&rec_ins->ins);
5354                     }
5355
5356                     if(ins->handler_idx == WINED3DSIH_ENDLOOP)
5357                     {
5358                         aL += control_frame->loop_control.step;
5359                     }
5360                 }
5361                 shader_addline(buffer, "#end loop/rep\n");
5362
5363                 free_recorded_instruction(&copy);
5364                 HeapFree(GetProcessHeap(), 0, control_frame);
5365                 return; /* Instruction is handled */
5366             }
5367             else
5368             {
5369                 /* This is a nested loop. Proceed to the normal recording function */
5370                 HeapFree(GetProcessHeap(), 0, control_frame);
5371             }
5372         }
5373     }
5374
5375     if(priv->recording)
5376     {
5377         record_instruction(&priv->record, ins);
5378         return;
5379     }
5380
5381     /* boolean if */
5382     if(ins->handler_idx == WINED3DSIH_IF)
5383     {
5384         control_frame = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*control_frame));
5385         list_add_head(&priv->control_frames, &control_frame->entry);
5386         control_frame->type = IF;
5387
5388         bool_const = get_bool_const(ins, shader, ins->src[0].reg.idx);
5389         if(ins->src[0].modifiers == WINED3DSPSM_NOT) bool_const = !bool_const;
5390         if (!priv->muted && !bool_const)
5391         {
5392             shader_addline(buffer, "#if(FALSE){\n");
5393             priv->muted = TRUE;
5394             control_frame->muting = TRUE;
5395         }
5396         else shader_addline(buffer, "#if(TRUE) {\n");
5397
5398         return; /* Instruction is handled */
5399     }
5400     else if(ins->handler_idx == WINED3DSIH_IFC)
5401     {
5402         /* IF(bool) and if_cond(a, b) use the same ELSE and ENDIF tokens */
5403         control_frame = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*control_frame));
5404         control_frame->type = IFC;
5405         control_frame->no.ifc = priv->num_ifcs++;
5406         list_add_head(&priv->control_frames, &control_frame->entry);
5407     }
5408     else if(ins->handler_idx == WINED3DSIH_ELSE)
5409     {
5410         struct list *e = list_head(&priv->control_frames);
5411         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5412
5413         if(control_frame->type == IF)
5414         {
5415             shader_addline(buffer, "#} else {\n");
5416             if(!priv->muted && !control_frame->muting)
5417             {
5418                 priv->muted = TRUE;
5419                 control_frame->muting = TRUE;
5420             }
5421             else if(control_frame->muting) priv->muted = FALSE;
5422             return; /* Instruction is handled. */
5423         }
5424         /* In case of an ifc, generate a HW shader instruction */
5425     }
5426     else if(ins->handler_idx == WINED3DSIH_ENDIF)
5427     {
5428         struct list *e = list_head(&priv->control_frames);
5429         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5430
5431         if(control_frame->type == IF)
5432         {
5433             shader_addline(buffer, "#} endif\n");
5434             if(control_frame->muting) priv->muted = FALSE;
5435             list_remove(&control_frame->entry);
5436             HeapFree(GetProcessHeap(), 0, control_frame);
5437             return; /* Instruction is handled */
5438         }
5439     }
5440
5441     if(priv->muted) return;
5442
5443     /* Select handler */
5444     hw_fct = shader_arb_instruction_handler_table[ins->handler_idx];
5445
5446     /* Unhandled opcode */
5447     if (!hw_fct)
5448     {
5449         FIXME("Backend can't handle opcode %#x\n", ins->handler_idx);
5450         return;
5451     }
5452     hw_fct(ins);
5453
5454     if(ins->handler_idx == WINED3DSIH_ENDLOOP || ins->handler_idx == WINED3DSIH_ENDREP)
5455     {
5456         struct list *e = list_head(&priv->control_frames);
5457         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5458         list_remove(&control_frame->entry);
5459         HeapFree(GetProcessHeap(), 0, control_frame);
5460         priv->loop_depth--;
5461     }
5462     else if(ins->handler_idx == WINED3DSIH_ENDIF)
5463     {
5464         /* Non-ifc ENDIFs don't reach that place because of the return in the if block above */
5465         struct list *e = list_head(&priv->control_frames);
5466         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5467         list_remove(&control_frame->entry);
5468         HeapFree(GetProcessHeap(), 0, control_frame);
5469     }
5470
5471
5472     shader_arb_add_instruction_modifiers(ins);
5473 }
5474
5475 const struct wined3d_shader_backend_ops arb_program_shader_backend =
5476 {
5477     shader_arb_handle_instruction,
5478     shader_arb_select,
5479     shader_arb_select_depth_blt,
5480     shader_arb_deselect_depth_blt,
5481     shader_arb_update_float_vertex_constants,
5482     shader_arb_update_float_pixel_constants,
5483     shader_arb_load_constants,
5484     shader_arb_load_np2fixup_constants,
5485     shader_arb_destroy,
5486     shader_arb_alloc,
5487     shader_arb_free,
5488     shader_arb_dirty_const,
5489     shader_arb_get_caps,
5490     shader_arb_color_fixup_supported,
5491 };
5492
5493 /* ARB_fragment_program fixed function pipeline replacement definitions */
5494 #define ARB_FFP_CONST_TFACTOR           0
5495 #define ARB_FFP_CONST_SPECULAR_ENABLE   ((ARB_FFP_CONST_TFACTOR) + 1)
5496 #define ARB_FFP_CONST_CONSTANT(i)       ((ARB_FFP_CONST_SPECULAR_ENABLE) + 1 + i)
5497 #define ARB_FFP_CONST_BUMPMAT(i)        ((ARB_FFP_CONST_CONSTANT(7)) + 1 + i)
5498 #define ARB_FFP_CONST_LUMINANCE(i)      ((ARB_FFP_CONST_BUMPMAT(7)) + 1 + i)
5499
5500 struct arbfp_ffp_desc
5501 {
5502     struct ffp_frag_desc parent;
5503     GLuint shader;
5504     unsigned int num_textures_used;
5505 };
5506
5507 /* Context activation and GL locking are done by the caller. */
5508 static void arbfp_enable(BOOL enable)
5509 {
5510     if(enable) {
5511         glEnable(GL_FRAGMENT_PROGRAM_ARB);
5512         checkGLcall("glEnable(GL_FRAGMENT_PROGRAM_ARB)");
5513     } else {
5514         glDisable(GL_FRAGMENT_PROGRAM_ARB);
5515         checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
5516     }
5517 }
5518
5519 static HRESULT arbfp_alloc(struct wined3d_device *device)
5520 {
5521     struct shader_arb_priv *priv;
5522     /* Share private data between the shader backend and the pipeline replacement, if both
5523      * are the arb implementation. This is needed to figure out whether ARBfp should be disabled
5524      * if no pixel shader is bound or not
5525      */
5526     if (device->shader_backend == &arb_program_shader_backend)
5527     {
5528         device->fragment_priv = device->shader_priv;
5529     }
5530     else
5531     {
5532         device->fragment_priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(struct shader_arb_priv));
5533         if (!device->fragment_priv) return E_OUTOFMEMORY;
5534     }
5535     priv = device->fragment_priv;
5536     if (wine_rb_init(&priv->fragment_shaders, &wined3d_ffp_frag_program_rb_functions) == -1)
5537     {
5538         ERR("Failed to initialize rbtree.\n");
5539         HeapFree(GetProcessHeap(), 0, device->fragment_priv);
5540         return E_OUTOFMEMORY;
5541     }
5542     priv->use_arbfp_fixed_func = TRUE;
5543     return WINED3D_OK;
5544 }
5545
5546 /* Context activation is done by the caller. */
5547 static void arbfp_free_ffpshader(struct wine_rb_entry *entry, void *context)
5548 {
5549     const struct wined3d_gl_info *gl_info = context;
5550     struct arbfp_ffp_desc *entry_arb = WINE_RB_ENTRY_VALUE(entry, struct arbfp_ffp_desc, parent.entry);
5551
5552     ENTER_GL();
5553     GL_EXTCALL(glDeleteProgramsARB(1, &entry_arb->shader));
5554     checkGLcall("glDeleteProgramsARB(1, &entry_arb->shader)");
5555     HeapFree(GetProcessHeap(), 0, entry_arb);
5556     LEAVE_GL();
5557 }
5558
5559 /* Context activation is done by the caller. */
5560 static void arbfp_free(struct wined3d_device *device)
5561 {
5562     struct shader_arb_priv *priv = device->fragment_priv;
5563
5564     wine_rb_destroy(&priv->fragment_shaders, arbfp_free_ffpshader, &device->adapter->gl_info);
5565     priv->use_arbfp_fixed_func = FALSE;
5566
5567     if (device->shader_backend != &arb_program_shader_backend)
5568     {
5569         HeapFree(GetProcessHeap(), 0, device->fragment_priv);
5570     }
5571 }
5572
5573 static void arbfp_get_caps(const struct wined3d_gl_info *gl_info, struct fragment_caps *caps)
5574 {
5575     caps->PrimitiveMiscCaps = WINED3DPMISCCAPS_TSSARGTEMP;
5576     caps->TextureOpCaps =  WINED3DTEXOPCAPS_DISABLE                     |
5577                            WINED3DTEXOPCAPS_SELECTARG1                  |
5578                            WINED3DTEXOPCAPS_SELECTARG2                  |
5579                            WINED3DTEXOPCAPS_MODULATE4X                  |
5580                            WINED3DTEXOPCAPS_MODULATE2X                  |
5581                            WINED3DTEXOPCAPS_MODULATE                    |
5582                            WINED3DTEXOPCAPS_ADDSIGNED2X                 |
5583                            WINED3DTEXOPCAPS_ADDSIGNED                   |
5584                            WINED3DTEXOPCAPS_ADD                         |
5585                            WINED3DTEXOPCAPS_SUBTRACT                    |
5586                            WINED3DTEXOPCAPS_ADDSMOOTH                   |
5587                            WINED3DTEXOPCAPS_BLENDCURRENTALPHA           |
5588                            WINED3DTEXOPCAPS_BLENDFACTORALPHA            |
5589                            WINED3DTEXOPCAPS_BLENDTEXTUREALPHA           |
5590                            WINED3DTEXOPCAPS_BLENDDIFFUSEALPHA           |
5591                            WINED3DTEXOPCAPS_BLENDTEXTUREALPHAPM         |
5592                            WINED3DTEXOPCAPS_MODULATEALPHA_ADDCOLOR      |
5593                            WINED3DTEXOPCAPS_MODULATECOLOR_ADDALPHA      |
5594                            WINED3DTEXOPCAPS_MODULATEINVCOLOR_ADDALPHA   |
5595                            WINED3DTEXOPCAPS_MODULATEINVALPHA_ADDCOLOR   |
5596                            WINED3DTEXOPCAPS_DOTPRODUCT3                 |
5597                            WINED3DTEXOPCAPS_MULTIPLYADD                 |
5598                            WINED3DTEXOPCAPS_LERP                        |
5599                            WINED3DTEXOPCAPS_BUMPENVMAP                  |
5600                            WINED3DTEXOPCAPS_BUMPENVMAPLUMINANCE;
5601
5602     /* TODO: Implement WINED3DTEXOPCAPS_PREMODULATE */
5603
5604     caps->MaxTextureBlendStages   = 8;
5605     caps->MaxSimultaneousTextures = min(gl_info->limits.fragment_samplers, 8);
5606 }
5607
5608 static void state_texfactor_arbfp(DWORD state_id,
5609         struct wined3d_stateblock *stateblock, struct wined3d_context *context)
5610 {
5611     const struct wined3d_gl_info *gl_info = context->gl_info;
5612     const struct wined3d_state *state = &stateblock->state;
5613     struct wined3d_device *device = stateblock->device;
5614     float col[4];
5615
5616     /* Don't load the parameter if we're using an arbfp pixel shader, otherwise we'll overwrite
5617      * application provided constants
5618      */
5619     if (device->shader_backend == &arb_program_shader_backend)
5620     {
5621         if (use_ps(state)) return;
5622
5623         context->pshader_const_dirty[ARB_FFP_CONST_TFACTOR] = 1;
5624         device->highest_dirty_ps_const = max(device->highest_dirty_ps_const, ARB_FFP_CONST_TFACTOR + 1);
5625     }
5626
5627     D3DCOLORTOGLFLOAT4(state->render_states[WINED3DRS_TEXTUREFACTOR], col);
5628     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_TFACTOR, col));
5629     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_TFACTOR, col)");
5630
5631 }
5632
5633 static void state_arb_specularenable(DWORD state_id,
5634         struct wined3d_stateblock *stateblock, struct wined3d_context *context)
5635 {
5636     const struct wined3d_gl_info *gl_info = context->gl_info;
5637     const struct wined3d_state *state = &stateblock->state;
5638     struct wined3d_device *device = stateblock->device;
5639     float col[4];
5640
5641     /* Don't load the parameter if we're using an arbfp pixel shader, otherwise we'll overwrite
5642      * application provided constants
5643      */
5644     if (device->shader_backend == &arb_program_shader_backend)
5645     {
5646         if (use_ps(state)) return;
5647
5648         context->pshader_const_dirty[ARB_FFP_CONST_SPECULAR_ENABLE] = 1;
5649         device->highest_dirty_ps_const = max(device->highest_dirty_ps_const, ARB_FFP_CONST_SPECULAR_ENABLE + 1);
5650     }
5651
5652     if (state->render_states[WINED3DRS_SPECULARENABLE])
5653     {
5654         /* The specular color has no alpha */
5655         col[0] = 1.0f; col[1] = 1.0f;
5656         col[2] = 1.0f; col[3] = 0.0f;
5657     } else {
5658         col[0] = 0.0f; col[1] = 0.0f;
5659         col[2] = 0.0f; col[3] = 0.0f;
5660     }
5661     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_SPECULAR_ENABLE, col));
5662     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_SPECULAR_ENABLE, col)");
5663 }
5664
5665 static void set_bumpmat_arbfp(DWORD state_id, struct wined3d_stateblock *stateblock, struct wined3d_context *context)
5666 {
5667     DWORD stage = (state_id - STATE_TEXTURESTAGE(0, 0)) / (WINED3D_HIGHEST_TEXTURE_STATE + 1);
5668     const struct wined3d_gl_info *gl_info = context->gl_info;
5669     const struct wined3d_state *state = &stateblock->state;
5670     struct wined3d_device *device = stateblock->device;
5671     float mat[2][2];
5672
5673     if (use_ps(state))
5674     {
5675         if (stage && (state->pixel_shader->reg_maps.bumpmat & (1 << stage)))
5676         {
5677             /* The pixel shader has to know the bump env matrix. Do a constants update if it isn't scheduled
5678              * anyway
5679              */
5680             if (!isStateDirty(context, STATE_PIXELSHADERCONSTANT))
5681                 stateblock_apply_state(STATE_PIXELSHADERCONSTANT, stateblock, context);
5682         }
5683
5684         if(device->shader_backend == &arb_program_shader_backend) {
5685             /* Exit now, don't set the bumpmat below, otherwise we may overwrite pixel shader constants */
5686             return;
5687         }
5688     } else if(device->shader_backend == &arb_program_shader_backend) {
5689         context->pshader_const_dirty[ARB_FFP_CONST_BUMPMAT(stage)] = 1;
5690         device->highest_dirty_ps_const = max(device->highest_dirty_ps_const, ARB_FFP_CONST_BUMPMAT(stage) + 1);
5691     }
5692
5693     mat[0][0] = *((float *)&state->texture_states[stage][WINED3DTSS_BUMPENVMAT00]);
5694     mat[0][1] = *((float *)&state->texture_states[stage][WINED3DTSS_BUMPENVMAT01]);
5695     mat[1][0] = *((float *)&state->texture_states[stage][WINED3DTSS_BUMPENVMAT10]);
5696     mat[1][1] = *((float *)&state->texture_states[stage][WINED3DTSS_BUMPENVMAT11]);
5697
5698     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_BUMPMAT(stage), &mat[0][0]));
5699     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_BUMPMAT(stage), &mat[0][0])");
5700 }
5701
5702 static void tex_bumpenvlum_arbfp(DWORD state_id,
5703         struct wined3d_stateblock *stateblock, struct wined3d_context *context)
5704 {
5705     DWORD stage = (state_id - STATE_TEXTURESTAGE(0, 0)) / (WINED3D_HIGHEST_TEXTURE_STATE + 1);
5706     const struct wined3d_gl_info *gl_info = context->gl_info;
5707     const struct wined3d_state *state = &stateblock->state;
5708     struct wined3d_device *device = stateblock->device;
5709     float param[4];
5710
5711     if (use_ps(state))
5712     {
5713         if (stage && (state->pixel_shader->reg_maps.luminanceparams & (1 << stage)))
5714         {
5715             /* The pixel shader has to know the luminance offset. Do a constants update if it
5716              * isn't scheduled anyway
5717              */
5718             if (!isStateDirty(context, STATE_PIXELSHADERCONSTANT))
5719                 stateblock_apply_state(STATE_PIXELSHADERCONSTANT, stateblock, context);
5720         }
5721
5722         if(device->shader_backend == &arb_program_shader_backend) {
5723             /* Exit now, don't set the bumpmat below, otherwise we may overwrite pixel shader constants */
5724             return;
5725         }
5726     } else if(device->shader_backend == &arb_program_shader_backend) {
5727         context->pshader_const_dirty[ARB_FFP_CONST_LUMINANCE(stage)] = 1;
5728         device->highest_dirty_ps_const = max(device->highest_dirty_ps_const, ARB_FFP_CONST_LUMINANCE(stage) + 1);
5729     }
5730
5731     param[0] = *((float *)&state->texture_states[stage][WINED3DTSS_BUMPENVLSCALE]);
5732     param[1] = *((float *)&state->texture_states[stage][WINED3DTSS_BUMPENVLOFFSET]);
5733     param[2] = 0.0f;
5734     param[3] = 0.0f;
5735
5736     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_LUMINANCE(stage), param));
5737     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_LUMINANCE(stage), param)");
5738 }
5739
5740 static const char *get_argreg(struct wined3d_shader_buffer *buffer, DWORD argnum, unsigned int stage, DWORD arg)
5741 {
5742     const char *ret;
5743
5744     if(arg == ARG_UNUSED) return "unused"; /* This is the marker for unused registers */
5745
5746     switch(arg & WINED3DTA_SELECTMASK) {
5747         case WINED3DTA_DIFFUSE:
5748             ret = "fragment.color.primary"; break;
5749
5750         case WINED3DTA_CURRENT:
5751             if (!stage) ret = "fragment.color.primary";
5752             else ret = "ret";
5753             break;
5754
5755         case WINED3DTA_TEXTURE:
5756             switch(stage) {
5757                 case 0: ret = "tex0"; break;
5758                 case 1: ret = "tex1"; break;
5759                 case 2: ret = "tex2"; break;
5760                 case 3: ret = "tex3"; break;
5761                 case 4: ret = "tex4"; break;
5762                 case 5: ret = "tex5"; break;
5763                 case 6: ret = "tex6"; break;
5764                 case 7: ret = "tex7"; break;
5765                 default: ret = "unknown texture";
5766             }
5767             break;
5768
5769         case WINED3DTA_TFACTOR:
5770             ret = "tfactor"; break;
5771
5772         case WINED3DTA_SPECULAR:
5773             ret = "fragment.color.secondary"; break;
5774
5775         case WINED3DTA_TEMP:
5776             ret = "tempreg"; break;
5777
5778         case WINED3DTA_CONSTANT:
5779             FIXME("Implement perstage constants\n");
5780             switch(stage) {
5781                 case 0: ret = "const0"; break;
5782                 case 1: ret = "const1"; break;
5783                 case 2: ret = "const2"; break;
5784                 case 3: ret = "const3"; break;
5785                 case 4: ret = "const4"; break;
5786                 case 5: ret = "const5"; break;
5787                 case 6: ret = "const6"; break;
5788                 case 7: ret = "const7"; break;
5789                 default: ret = "unknown constant";
5790             }
5791             break;
5792
5793         default:
5794             return "unknown";
5795     }
5796
5797     if(arg & WINED3DTA_COMPLEMENT) {
5798         shader_addline(buffer, "SUB arg%u, const.x, %s;\n", argnum, ret);
5799         if(argnum == 0) ret = "arg0";
5800         if(argnum == 1) ret = "arg1";
5801         if(argnum == 2) ret = "arg2";
5802     }
5803     if(arg & WINED3DTA_ALPHAREPLICATE) {
5804         shader_addline(buffer, "MOV arg%u, %s.w;\n", argnum, ret);
5805         if(argnum == 0) ret = "arg0";
5806         if(argnum == 1) ret = "arg1";
5807         if(argnum == 2) ret = "arg2";
5808     }
5809     return ret;
5810 }
5811
5812 static void gen_ffp_instr(struct wined3d_shader_buffer *buffer, unsigned int stage, BOOL color,
5813         BOOL alpha, DWORD dst, DWORD op, DWORD dw_arg0, DWORD dw_arg1, DWORD dw_arg2)
5814 {
5815     const char *dstmask, *dstreg, *arg0, *arg1, *arg2;
5816     unsigned int mul = 1;
5817     BOOL mul_final_dest = FALSE;
5818
5819     if(color && alpha) dstmask = "";
5820     else if(color) dstmask = ".xyz";
5821     else dstmask = ".w";
5822
5823     if(dst == tempreg) dstreg = "tempreg";
5824     else dstreg = "ret";
5825
5826     arg0 = get_argreg(buffer, 0, stage, dw_arg0);
5827     arg1 = get_argreg(buffer, 1, stage, dw_arg1);
5828     arg2 = get_argreg(buffer, 2, stage, dw_arg2);
5829
5830     switch(op) {
5831         case WINED3DTOP_DISABLE:
5832             if (!stage) shader_addline(buffer, "MOV %s%s, fragment.color.primary;\n", dstreg, dstmask);
5833             break;
5834
5835         case WINED3DTOP_SELECTARG2:
5836             arg1 = arg2;
5837             /* FALLTHROUGH */
5838         case WINED3DTOP_SELECTARG1:
5839             shader_addline(buffer, "MOV %s%s, %s;\n", dstreg, dstmask, arg1);
5840             break;
5841
5842         case WINED3DTOP_MODULATE4X:
5843             mul = 2;
5844             /* FALLTHROUGH */
5845         case WINED3DTOP_MODULATE2X:
5846             mul *= 2;
5847             if (!strcmp(dstreg, "result.color"))
5848             {
5849                 dstreg = "ret";
5850                 mul_final_dest = TRUE;
5851             }
5852             /* FALLTHROUGH */
5853         case WINED3DTOP_MODULATE:
5854             shader_addline(buffer, "MUL %s%s, %s, %s;\n", dstreg, dstmask, arg1, arg2);
5855             break;
5856
5857         case WINED3DTOP_ADDSIGNED2X:
5858             mul = 2;
5859             if (!strcmp(dstreg, "result.color"))
5860             {
5861                 dstreg = "ret";
5862                 mul_final_dest = TRUE;
5863             }
5864             /* FALLTHROUGH */
5865         case WINED3DTOP_ADDSIGNED:
5866             shader_addline(buffer, "SUB arg2, %s, const.w;\n", arg2);
5867             arg2 = "arg2";
5868             /* FALLTHROUGH */
5869         case WINED3DTOP_ADD:
5870             shader_addline(buffer, "ADD_SAT %s%s, %s, %s;\n", dstreg, dstmask, arg1, arg2);
5871             break;
5872
5873         case WINED3DTOP_SUBTRACT:
5874             shader_addline(buffer, "SUB_SAT %s%s, %s, %s;\n", dstreg, dstmask, arg1, arg2);
5875             break;
5876
5877         case WINED3DTOP_ADDSMOOTH:
5878             shader_addline(buffer, "SUB arg1, const.x, %s;\n", arg1);
5879             shader_addline(buffer, "MAD_SAT %s%s, arg1, %s, %s;\n", dstreg, dstmask, arg2, arg1);
5880             break;
5881
5882         case WINED3DTOP_BLENDCURRENTALPHA:
5883             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_CURRENT);
5884             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
5885             break;
5886         case WINED3DTOP_BLENDFACTORALPHA:
5887             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_TFACTOR);
5888             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
5889             break;
5890         case WINED3DTOP_BLENDTEXTUREALPHA:
5891             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_TEXTURE);
5892             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
5893             break;
5894         case WINED3DTOP_BLENDDIFFUSEALPHA:
5895             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_DIFFUSE);
5896             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
5897             break;
5898
5899         case WINED3DTOP_BLENDTEXTUREALPHAPM:
5900             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_TEXTURE);
5901             shader_addline(buffer, "SUB arg0.w, const.x, %s.w;\n", arg0);
5902             shader_addline(buffer, "MAD_SAT %s%s, %s, arg0.w, %s;\n", dstreg, dstmask, arg2, arg1);
5903             break;
5904
5905         /* D3DTOP_PREMODULATE ???? */
5906
5907         case WINED3DTOP_MODULATEINVALPHA_ADDCOLOR:
5908             shader_addline(buffer, "SUB arg0.w, const.x, %s;\n", arg1);
5909             shader_addline(buffer, "MAD_SAT %s%s, arg0.w, %s, %s;\n", dstreg, dstmask, arg2, arg1);
5910             break;
5911         case WINED3DTOP_MODULATEALPHA_ADDCOLOR:
5912             shader_addline(buffer, "MAD_SAT %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg1, arg2, arg1);
5913             break;
5914         case WINED3DTOP_MODULATEINVCOLOR_ADDALPHA:
5915             shader_addline(buffer, "SUB arg0, const.x, %s;\n", arg1);
5916             shader_addline(buffer, "MAD_SAT %s%s, arg0, %s, %s.w;\n", dstreg, dstmask, arg2, arg1);
5917             break;
5918         case WINED3DTOP_MODULATECOLOR_ADDALPHA:
5919             shader_addline(buffer, "MAD_SAT %s%s, %s, %s, %s.w;\n", dstreg, dstmask, arg1, arg2, arg1);
5920             break;
5921
5922         case WINED3DTOP_DOTPRODUCT3:
5923             mul = 4;
5924             if (!strcmp(dstreg, "result.color"))
5925             {
5926                 dstreg = "ret";
5927                 mul_final_dest = TRUE;
5928             }
5929             shader_addline(buffer, "SUB arg1, %s, const.w;\n", arg1);
5930             shader_addline(buffer, "SUB arg2, %s, const.w;\n", arg2);
5931             shader_addline(buffer, "DP3_SAT %s%s, arg1, arg2;\n", dstreg, dstmask);
5932             break;
5933
5934         case WINED3DTOP_MULTIPLYADD:
5935             shader_addline(buffer, "MAD_SAT %s%s, %s, %s, %s;\n", dstreg, dstmask, arg1, arg2, arg0);
5936             break;
5937
5938         case WINED3DTOP_LERP:
5939             /* The msdn is not quite right here */
5940             shader_addline(buffer, "LRP %s%s, %s, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
5941             break;
5942
5943         case WINED3DTOP_BUMPENVMAP:
5944         case WINED3DTOP_BUMPENVMAPLUMINANCE:
5945             /* Those are handled in the first pass of the shader(generation pass 1 and 2) already */
5946             break;
5947
5948         default:
5949             FIXME("Unhandled texture op %08x\n", op);
5950     }
5951
5952     if(mul == 2) {
5953         shader_addline(buffer, "MUL_SAT %s%s, %s, const.y;\n", mul_final_dest ? "result.color" : dstreg, dstmask, dstreg);
5954     } else if(mul == 4) {
5955         shader_addline(buffer, "MUL_SAT %s%s, %s, const.z;\n", mul_final_dest ? "result.color" : dstreg, dstmask, dstreg);
5956     }
5957 }
5958
5959 static GLuint gen_arbfp_ffp_shader(const struct ffp_frag_settings *settings, struct wined3d_stateblock *stateblock)
5960 {
5961     const struct wined3d_gl_info *gl_info = &stateblock->device->adapter->gl_info;
5962     unsigned int stage;
5963     struct wined3d_shader_buffer buffer;
5964     BOOL tex_read[MAX_TEXTURES] = {FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE};
5965     BOOL bump_used[MAX_TEXTURES] = {FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE};
5966     BOOL luminance_used[MAX_TEXTURES] = {FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE};
5967     const char *textype;
5968     const char *instr, *sat;
5969     char colorcor_dst[8];
5970     GLuint ret;
5971     DWORD arg0, arg1, arg2;
5972     BOOL tempreg_used = FALSE, tfactor_used = FALSE;
5973     BOOL op_equal;
5974     const char *final_combiner_src = "ret";
5975     GLint pos;
5976
5977     /* Find out which textures are read */
5978     for(stage = 0; stage < MAX_TEXTURES; stage++) {
5979         if(settings->op[stage].cop == WINED3DTOP_DISABLE) break;
5980         arg0 = settings->op[stage].carg0 & WINED3DTA_SELECTMASK;
5981         arg1 = settings->op[stage].carg1 & WINED3DTA_SELECTMASK;
5982         arg2 = settings->op[stage].carg2 & WINED3DTA_SELECTMASK;
5983         if(arg0 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
5984         if(arg1 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
5985         if(arg2 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
5986
5987         if(settings->op[stage].cop == WINED3DTOP_BLENDTEXTUREALPHA) tex_read[stage] = TRUE;
5988         if(settings->op[stage].cop == WINED3DTOP_BLENDTEXTUREALPHAPM) tex_read[stage] = TRUE;
5989         if(settings->op[stage].cop == WINED3DTOP_BUMPENVMAP) {
5990             bump_used[stage] = TRUE;
5991             tex_read[stage] = TRUE;
5992         }
5993         if(settings->op[stage].cop == WINED3DTOP_BUMPENVMAPLUMINANCE) {
5994             bump_used[stage] = TRUE;
5995             tex_read[stage] = TRUE;
5996             luminance_used[stage] = TRUE;
5997         } else if(settings->op[stage].cop == WINED3DTOP_BLENDFACTORALPHA) {
5998             tfactor_used = TRUE;
5999         }
6000
6001         if(arg0 == WINED3DTA_TFACTOR || arg1 == WINED3DTA_TFACTOR || arg2 == WINED3DTA_TFACTOR) {
6002             tfactor_used = TRUE;
6003         }
6004
6005         if(settings->op[stage].dst == tempreg) tempreg_used = TRUE;
6006         if(arg0 == WINED3DTA_TEMP || arg1 == WINED3DTA_TEMP || arg2 == WINED3DTA_TEMP) {
6007             tempreg_used = TRUE;
6008         }
6009
6010         if(settings->op[stage].aop == WINED3DTOP_DISABLE) continue;
6011         arg0 = settings->op[stage].aarg0 & WINED3DTA_SELECTMASK;
6012         arg1 = settings->op[stage].aarg1 & WINED3DTA_SELECTMASK;
6013         arg2 = settings->op[stage].aarg2 & WINED3DTA_SELECTMASK;
6014         if(arg0 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
6015         if(arg1 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
6016         if(arg2 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
6017
6018         if(arg0 == WINED3DTA_TEMP || arg1 == WINED3DTA_TEMP || arg2 == WINED3DTA_TEMP) {
6019             tempreg_used = TRUE;
6020         }
6021         if(arg0 == WINED3DTA_TFACTOR || arg1 == WINED3DTA_TFACTOR || arg2 == WINED3DTA_TFACTOR) {
6022             tfactor_used = TRUE;
6023         }
6024     }
6025
6026     /* Shader header */
6027     if (!shader_buffer_init(&buffer))
6028     {
6029         ERR("Failed to initialize shader buffer.\n");
6030         return 0;
6031     }
6032
6033     shader_addline(&buffer, "!!ARBfp1.0\n");
6034
6035     switch(settings->fog) {
6036         case FOG_OFF:                                                         break;
6037         case FOG_LINEAR: shader_addline(&buffer, "OPTION ARB_fog_linear;\n"); break;
6038         case FOG_EXP:    shader_addline(&buffer, "OPTION ARB_fog_exp;\n");    break;
6039         case FOG_EXP2:   shader_addline(&buffer, "OPTION ARB_fog_exp2;\n");   break;
6040         default: FIXME("Unexpected fog setting %d\n", settings->fog);
6041     }
6042
6043     shader_addline(&buffer, "PARAM const = {1, 2, 4, 0.5};\n");
6044     shader_addline(&buffer, "TEMP TMP;\n");
6045     shader_addline(&buffer, "TEMP ret;\n");
6046     if(tempreg_used || settings->sRGB_write) shader_addline(&buffer, "TEMP tempreg;\n");
6047     shader_addline(&buffer, "TEMP arg0;\n");
6048     shader_addline(&buffer, "TEMP arg1;\n");
6049     shader_addline(&buffer, "TEMP arg2;\n");
6050     for(stage = 0; stage < MAX_TEXTURES; stage++) {
6051         if(!tex_read[stage]) continue;
6052         shader_addline(&buffer, "TEMP tex%u;\n", stage);
6053         if(!bump_used[stage]) continue;
6054         shader_addline(&buffer, "PARAM bumpmat%u = program.env[%u];\n", stage, ARB_FFP_CONST_BUMPMAT(stage));
6055         if(!luminance_used[stage]) continue;
6056         shader_addline(&buffer, "PARAM luminance%u = program.env[%u];\n", stage, ARB_FFP_CONST_LUMINANCE(stage));
6057     }
6058     if(tfactor_used) {
6059         shader_addline(&buffer, "PARAM tfactor = program.env[%u];\n", ARB_FFP_CONST_TFACTOR);
6060     }
6061         shader_addline(&buffer, "PARAM specular_enable = program.env[%u];\n", ARB_FFP_CONST_SPECULAR_ENABLE);
6062
6063     if(settings->sRGB_write) {
6064         shader_addline(&buffer, "PARAM srgb_consts1 = {%f, %f, %f, %f};\n",
6065                        srgb_mul_low, srgb_cmp, srgb_pow, srgb_mul_high);
6066         shader_addline(&buffer, "PARAM srgb_consts2 = {%f, %f, %f, %f};\n",
6067                        srgb_sub_high, 0.0, 0.0, 0.0);
6068     }
6069
6070     if (ffp_clip_emul(&stateblock->state) && settings->emul_clipplanes)
6071         shader_addline(&buffer, "KIL fragment.texcoord[7];\n");
6072
6073     /* Generate texture sampling instructions) */
6074     for(stage = 0; stage < MAX_TEXTURES && settings->op[stage].cop != WINED3DTOP_DISABLE; stage++) {
6075         if(!tex_read[stage]) continue;
6076
6077         switch(settings->op[stage].tex_type) {
6078             case tex_1d:                    textype = "1D";     break;
6079             case tex_2d:                    textype = "2D";     break;
6080             case tex_3d:                    textype = "3D";     break;
6081             case tex_cube:                  textype = "CUBE";   break;
6082             case tex_rect:                  textype = "RECT";   break;
6083             default: textype = "unexpected_textype";   break;
6084         }
6085
6086         if(settings->op[stage].cop == WINED3DTOP_BUMPENVMAP ||
6087            settings->op[stage].cop == WINED3DTOP_BUMPENVMAPLUMINANCE) {
6088             sat = "";
6089         } else {
6090             sat = "_SAT";
6091         }
6092
6093         if(settings->op[stage].projected == proj_none) {
6094             instr = "TEX";
6095         } else if(settings->op[stage].projected == proj_count4 ||
6096                   settings->op[stage].projected == proj_count3) {
6097             instr = "TXP";
6098         } else {
6099             FIXME("Unexpected projection mode %d\n", settings->op[stage].projected);
6100             instr = "TXP";
6101         }
6102
6103         if(stage > 0 &&
6104            (settings->op[stage - 1].cop == WINED3DTOP_BUMPENVMAP ||
6105             settings->op[stage - 1].cop == WINED3DTOP_BUMPENVMAPLUMINANCE)) {
6106             shader_addline(&buffer, "SWZ arg1, bumpmat%u, x, z, 0, 0;\n", stage - 1);
6107             shader_addline(&buffer, "DP3 ret.x, arg1, tex%u;\n", stage - 1);
6108             shader_addline(&buffer, "SWZ arg1, bumpmat%u, y, w, 0, 0;\n", stage - 1);
6109             shader_addline(&buffer, "DP3 ret.y, arg1, tex%u;\n", stage - 1);
6110
6111             /* with projective textures, texbem only divides the static texture coord, not the displacement,
6112              * so multiply the displacement with the dividing parameter before passing it to TXP
6113              */
6114             if (settings->op[stage].projected != proj_none) {
6115                 if(settings->op[stage].projected == proj_count4) {
6116                     shader_addline(&buffer, "MOV ret.w, fragment.texcoord[%u].w;\n", stage);
6117                     shader_addline(&buffer, "MUL ret.xyz, ret, fragment.texcoord[%u].w, fragment.texcoord[%u];\n", stage, stage);
6118                 } else {
6119                     shader_addline(&buffer, "MOV ret.w, fragment.texcoord[%u].z;\n", stage);
6120                     shader_addline(&buffer, "MAD ret.xyz, ret, fragment.texcoord[%u].z, fragment.texcoord[%u];\n", stage, stage);
6121                 }
6122             } else {
6123                 shader_addline(&buffer, "ADD ret, ret, fragment.texcoord[%u];\n", stage);
6124             }
6125
6126             shader_addline(&buffer, "%s%s tex%u, ret, texture[%u], %s;\n",
6127                            instr, sat, stage, stage, textype);
6128             if(settings->op[stage - 1].cop == WINED3DTOP_BUMPENVMAPLUMINANCE) {
6129                 shader_addline(&buffer, "MAD_SAT ret.x, tex%u.z, luminance%u.x, luminance%u.y;\n",
6130                                stage - 1, stage - 1, stage - 1);
6131                 shader_addline(&buffer, "MUL tex%u, tex%u, ret.x;\n", stage, stage);
6132             }
6133         } else if(settings->op[stage].projected == proj_count3) {
6134             shader_addline(&buffer, "MOV ret, fragment.texcoord[%u];\n", stage);
6135             shader_addline(&buffer, "MOV ret.w, ret.z;\n");
6136             shader_addline(&buffer, "%s%s tex%u, ret, texture[%u], %s;\n",
6137                             instr, sat, stage, stage, textype);
6138         } else {
6139             shader_addline(&buffer, "%s%s tex%u, fragment.texcoord[%u], texture[%u], %s;\n",
6140                             instr, sat, stage, stage, stage, textype);
6141         }
6142
6143         sprintf(colorcor_dst, "tex%u", stage);
6144         gen_color_correction(&buffer, colorcor_dst, WINED3DSP_WRITEMASK_ALL, "const.x", "const.y",
6145                 settings->op[stage].color_fixup);
6146     }
6147
6148     /* Generate the main shader */
6149     for (stage = 0; stage < MAX_TEXTURES; ++stage)
6150     {
6151         if (settings->op[stage].cop == WINED3DTOP_DISABLE)
6152         {
6153             if (!stage) final_combiner_src = "fragment.color.primary";
6154             break;
6155         }
6156
6157         if(settings->op[stage].cop == WINED3DTOP_SELECTARG1 &&
6158            settings->op[stage].aop == WINED3DTOP_SELECTARG1) {
6159             op_equal = settings->op[stage].carg1 == settings->op[stage].aarg1;
6160         } else if(settings->op[stage].cop == WINED3DTOP_SELECTARG1 &&
6161                   settings->op[stage].aop == WINED3DTOP_SELECTARG2) {
6162             op_equal = settings->op[stage].carg1 == settings->op[stage].aarg2;
6163         } else if(settings->op[stage].cop == WINED3DTOP_SELECTARG2 &&
6164                   settings->op[stage].aop == WINED3DTOP_SELECTARG1) {
6165             op_equal = settings->op[stage].carg2 == settings->op[stage].aarg1;
6166         } else if(settings->op[stage].cop == WINED3DTOP_SELECTARG2 &&
6167                   settings->op[stage].aop == WINED3DTOP_SELECTARG2) {
6168             op_equal = settings->op[stage].carg2 == settings->op[stage].aarg2;
6169         } else {
6170             op_equal = settings->op[stage].aop   == settings->op[stage].cop &&
6171                        settings->op[stage].carg0 == settings->op[stage].aarg0 &&
6172                        settings->op[stage].carg1 == settings->op[stage].aarg1 &&
6173                        settings->op[stage].carg2 == settings->op[stage].aarg2;
6174         }
6175
6176         if(settings->op[stage].aop == WINED3DTOP_DISABLE) {
6177             gen_ffp_instr(&buffer, stage, TRUE, FALSE, settings->op[stage].dst,
6178                           settings->op[stage].cop, settings->op[stage].carg0,
6179                           settings->op[stage].carg1, settings->op[stage].carg2);
6180             if (!stage)
6181                 shader_addline(&buffer, "MOV ret.w, fragment.color.primary.w;\n");
6182         }
6183         else if (op_equal)
6184         {
6185             gen_ffp_instr(&buffer, stage, TRUE, TRUE, settings->op[stage].dst,
6186                           settings->op[stage].cop, settings->op[stage].carg0,
6187                           settings->op[stage].carg1, settings->op[stage].carg2);
6188         } else {
6189             gen_ffp_instr(&buffer, stage, TRUE, FALSE, settings->op[stage].dst,
6190                           settings->op[stage].cop, settings->op[stage].carg0,
6191                           settings->op[stage].carg1, settings->op[stage].carg2);
6192             gen_ffp_instr(&buffer, stage, FALSE, TRUE, settings->op[stage].dst,
6193                           settings->op[stage].aop, settings->op[stage].aarg0,
6194                           settings->op[stage].aarg1, settings->op[stage].aarg2);
6195         }
6196     }
6197
6198     if(settings->sRGB_write) {
6199         shader_addline(&buffer, "MAD ret, fragment.color.secondary, specular_enable, %s;\n", final_combiner_src);
6200         arbfp_add_sRGB_correction(&buffer, "ret", "arg0", "arg1", "arg2", "tempreg", FALSE);
6201         shader_addline(&buffer, "MOV result.color, ret;\n");
6202     } else {
6203         shader_addline(&buffer, "MAD result.color, fragment.color.secondary, specular_enable, %s;\n", final_combiner_src);
6204     }
6205
6206     /* Footer */
6207     shader_addline(&buffer, "END\n");
6208
6209     /* Generate the shader */
6210     GL_EXTCALL(glGenProgramsARB(1, &ret));
6211     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, ret));
6212     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
6213             strlen(buffer.buffer), buffer.buffer));
6214     checkGLcall("glProgramStringARB()");
6215
6216     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
6217     if (pos != -1)
6218     {
6219         FIXME("Fragment program error at position %d: %s\n\n", pos,
6220               debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
6221         shader_arb_dump_program_source(buffer.buffer);
6222     }
6223     else
6224     {
6225         GLint native;
6226
6227         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
6228         checkGLcall("glGetProgramivARB()");
6229         if (!native) WARN("Program exceeds native resource limits.\n");
6230     }
6231
6232     shader_buffer_free(&buffer);
6233     return ret;
6234 }
6235
6236 static void fragment_prog_arbfp(DWORD state_id, struct wined3d_stateblock *stateblock, struct wined3d_context *context)
6237 {
6238     const struct wined3d_gl_info *gl_info = context->gl_info;
6239     const struct wined3d_state *state = &stateblock->state;
6240     struct wined3d_device *device = stateblock->device;
6241     struct shader_arb_priv *priv = device->fragment_priv;
6242     BOOL use_vshader = use_vs(state);
6243     BOOL use_pshader = use_ps(state);
6244     struct ffp_frag_settings settings;
6245     const struct arbfp_ffp_desc *desc;
6246     unsigned int i;
6247
6248     TRACE("state_id %#x, stateblock %p, context %p\n", state_id, stateblock, context);
6249
6250     if(isStateDirty(context, STATE_RENDER(WINED3DRS_FOGENABLE))) {
6251         if(!use_pshader && device->shader_backend == &arb_program_shader_backend && context->last_was_pshader) {
6252             /* Reload fixed function constants since they collide with the pixel shader constants */
6253             for(i = 0; i < MAX_TEXTURES; i++) {
6254                 set_bumpmat_arbfp(STATE_TEXTURESTAGE(i, WINED3DTSS_BUMPENVMAT00), stateblock, context);
6255             }
6256             state_texfactor_arbfp(STATE_RENDER(WINED3DRS_TEXTUREFACTOR), stateblock, context);
6257             state_arb_specularenable(STATE_RENDER(WINED3DRS_SPECULARENABLE), stateblock, context);
6258         } else if(use_pshader && !isStateDirty(context, device->StateTable[STATE_VSHADER].representative)) {
6259             device->shader_backend->shader_select(context, use_pshader, use_vshader);
6260         }
6261         return;
6262     }
6263
6264     if(!use_pshader) {
6265         /* Find or create a shader implementing the fixed function pipeline settings, then activate it */
6266         gen_ffp_frag_op(stateblock, &settings, FALSE);
6267         desc = (const struct arbfp_ffp_desc *)find_ffp_frag_shader(&priv->fragment_shaders, &settings);
6268         if(!desc) {
6269             struct arbfp_ffp_desc *new_desc = HeapAlloc(GetProcessHeap(), 0, sizeof(*new_desc));
6270             if (!new_desc)
6271             {
6272                 ERR("Out of memory\n");
6273                 return;
6274             }
6275             new_desc->num_textures_used = 0;
6276             for (i = 0; i < gl_info->limits.texture_stages; ++i)
6277             {
6278                 if(settings.op[i].cop == WINED3DTOP_DISABLE) break;
6279                 new_desc->num_textures_used = i;
6280             }
6281
6282             memcpy(&new_desc->parent.settings, &settings, sizeof(settings));
6283             new_desc->shader = gen_arbfp_ffp_shader(&settings, stateblock);
6284             add_ffp_frag_shader(&priv->fragment_shaders, &new_desc->parent);
6285             TRACE("Allocated fixed function replacement shader descriptor %p\n", new_desc);
6286             desc = new_desc;
6287         }
6288
6289         /* Now activate the replacement program. GL_FRAGMENT_PROGRAM_ARB is already active(however, note the
6290          * comment above the shader_select call below). If e.g. GLSL is active, the shader_select call will
6291          * deactivate it.
6292          */
6293         GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, desc->shader));
6294         checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, desc->shader)");
6295         priv->current_fprogram_id = desc->shader;
6296
6297         if(device->shader_backend == &arb_program_shader_backend && context->last_was_pshader) {
6298             /* Reload fixed function constants since they collide with the pixel shader constants */
6299             for(i = 0; i < MAX_TEXTURES; i++) {
6300                 set_bumpmat_arbfp(STATE_TEXTURESTAGE(i, WINED3DTSS_BUMPENVMAT00), stateblock, context);
6301             }
6302             state_texfactor_arbfp(STATE_RENDER(WINED3DRS_TEXTUREFACTOR), stateblock, context);
6303             state_arb_specularenable(STATE_RENDER(WINED3DRS_SPECULARENABLE), stateblock, context);
6304         }
6305         context->last_was_pshader = FALSE;
6306     } else {
6307         context->last_was_pshader = TRUE;
6308     }
6309
6310     /* Finally, select the shader. If a pixel shader is used, it will be set and enabled by the shader backend.
6311      * If this shader backend is arbfp(most likely), then it will simply overwrite the last fixed function replace-
6312      * ment shader. If the shader backend is not ARB, it currently is important that the opengl implementation
6313      * type overwrites GL_ARB_fragment_program. This is currently the case with GLSL. If we really want to use
6314      * atifs or nvrc pixel shaders with arb fragment programs we'd have to disable GL_FRAGMENT_PROGRAM_ARB here
6315      *
6316      * Don't call shader_select if the vertex shader is dirty, because it will be called later on by the vertex
6317      * shader handler
6318      */
6319     if(!isStateDirty(context, device->StateTable[STATE_VSHADER].representative)) {
6320         device->shader_backend->shader_select(context, use_pshader, use_vshader);
6321
6322         if (!isStateDirty(context, STATE_VERTEXSHADERCONSTANT) && (use_vshader || use_pshader))
6323             stateblock_apply_state(STATE_VERTEXSHADERCONSTANT, stateblock, context);
6324     }
6325     if (use_pshader) stateblock_apply_state(STATE_PIXELSHADERCONSTANT, stateblock, context);
6326 }
6327
6328 /* We can't link the fog states to the fragment state directly since the vertex pipeline links them
6329  * to FOGENABLE. A different linking in different pipeline parts can't be expressed in the combined
6330  * state table, so we need to handle that with a forwarding function. The other invisible side effect
6331  * is that changing the fog start and fog end(which links to FOGENABLE in vertex) results in the
6332  * fragment_prog_arbfp function being called because FOGENABLE is dirty, which calls this function here
6333  */
6334 static void state_arbfp_fog(DWORD state_id, struct wined3d_stateblock *stateblock, struct wined3d_context *context)
6335 {
6336     const struct wined3d_state *state = &stateblock->state;
6337     enum fogsource new_source;
6338
6339     TRACE("state_id %#x, stateblock %p, context %p\n", state_id, stateblock, context);
6340
6341     if(!isStateDirty(context, STATE_PIXELSHADER)) {
6342         fragment_prog_arbfp(state_id, stateblock, context);
6343     }
6344
6345     if (!state->render_states[WINED3DRS_FOGENABLE]) return;
6346
6347     if (state->render_states[WINED3DRS_FOGTABLEMODE] == WINED3DFOG_NONE)
6348     {
6349         if (use_vs(state))
6350         {
6351             new_source = FOGSOURCE_VS;
6352         }
6353         else
6354         {
6355             if (state->render_states[WINED3DRS_FOGVERTEXMODE] == WINED3DFOG_NONE || context->last_was_rhw)
6356                 new_source = FOGSOURCE_COORD;
6357             else
6358                 new_source = FOGSOURCE_FFP;
6359         }
6360     } else {
6361         new_source = FOGSOURCE_FFP;
6362     }
6363     if(new_source != context->fog_source) {
6364         context->fog_source = new_source;
6365         state_fogstartend(STATE_RENDER(WINED3DRS_FOGSTART), stateblock, context);
6366     }
6367 }
6368
6369 static void textransform(DWORD state, struct wined3d_stateblock *stateblock, struct wined3d_context *context)
6370 {
6371     if (!isStateDirty(context, STATE_PIXELSHADER))
6372         fragment_prog_arbfp(state, stateblock, context);
6373 }
6374
6375 static const struct StateEntryTemplate arbfp_fragmentstate_template[] = {
6376     {STATE_RENDER(WINED3DRS_TEXTUREFACTOR),               { STATE_RENDER(WINED3DRS_TEXTUREFACTOR),              state_texfactor_arbfp   }, WINED3D_GL_EXT_NONE             },
6377     {STATE_TEXTURESTAGE(0, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6378     {STATE_TEXTURESTAGE(0, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6379     {STATE_TEXTURESTAGE(0, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6380     {STATE_TEXTURESTAGE(0, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6381     {STATE_TEXTURESTAGE(0, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6382     {STATE_TEXTURESTAGE(0, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6383     {STATE_TEXTURESTAGE(0, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6384     {STATE_TEXTURESTAGE(0, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6385     {STATE_TEXTURESTAGE(0, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6386     {STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6387     {STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6388     {STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6389     {STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6390     {STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6391     {STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVLSCALE),    NULL                    }, WINED3D_GL_EXT_NONE             },
6392     {STATE_TEXTURESTAGE(1, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6393     {STATE_TEXTURESTAGE(1, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6394     {STATE_TEXTURESTAGE(1, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6395     {STATE_TEXTURESTAGE(1, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6396     {STATE_TEXTURESTAGE(1, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6397     {STATE_TEXTURESTAGE(1, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6398     {STATE_TEXTURESTAGE(1, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6399     {STATE_TEXTURESTAGE(1, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6400     {STATE_TEXTURESTAGE(1, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6401     {STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6402     {STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6403     {STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6404     {STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6405     {STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6406     {STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVLSCALE),    NULL                    }, WINED3D_GL_EXT_NONE             },
6407     {STATE_TEXTURESTAGE(2, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6408     {STATE_TEXTURESTAGE(2, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6409     {STATE_TEXTURESTAGE(2, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6410     {STATE_TEXTURESTAGE(2, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6411     {STATE_TEXTURESTAGE(2, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6412     {STATE_TEXTURESTAGE(2, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6413     {STATE_TEXTURESTAGE(2, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6414     {STATE_TEXTURESTAGE(2, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6415     {STATE_TEXTURESTAGE(2, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6416     {STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6417     {STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6418     {STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6419     {STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6420     {STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6421     {STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVLSCALE),    NULL                    }, WINED3D_GL_EXT_NONE             },
6422     {STATE_TEXTURESTAGE(3, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6423     {STATE_TEXTURESTAGE(3, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6424     {STATE_TEXTURESTAGE(3, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6425     {STATE_TEXTURESTAGE(3, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6426     {STATE_TEXTURESTAGE(3, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6427     {STATE_TEXTURESTAGE(3, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6428     {STATE_TEXTURESTAGE(3, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6429     {STATE_TEXTURESTAGE(3, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6430     {STATE_TEXTURESTAGE(3, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6431     {STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6432     {STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6433     {STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6434     {STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6435     {STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6436     {STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVLSCALE),    NULL                    }, WINED3D_GL_EXT_NONE             },
6437     {STATE_TEXTURESTAGE(4, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6438     {STATE_TEXTURESTAGE(4, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6439     {STATE_TEXTURESTAGE(4, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6440     {STATE_TEXTURESTAGE(4, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6441     {STATE_TEXTURESTAGE(4, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6442     {STATE_TEXTURESTAGE(4, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6443     {STATE_TEXTURESTAGE(4, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6444     {STATE_TEXTURESTAGE(4, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6445     {STATE_TEXTURESTAGE(4, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6446     {STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6447     {STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6448     {STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6449     {STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6450     {STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6451     {STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVLSCALE),    NULL                    }, WINED3D_GL_EXT_NONE             },
6452     {STATE_TEXTURESTAGE(5, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6453     {STATE_TEXTURESTAGE(5, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6454     {STATE_TEXTURESTAGE(5, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6455     {STATE_TEXTURESTAGE(5, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6456     {STATE_TEXTURESTAGE(5, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6457     {STATE_TEXTURESTAGE(5, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6458     {STATE_TEXTURESTAGE(5, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6459     {STATE_TEXTURESTAGE(5, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6460     {STATE_TEXTURESTAGE(5, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6461     {STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6462     {STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6463     {STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6464     {STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6465     {STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6466     {STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVLSCALE),    NULL                    }, WINED3D_GL_EXT_NONE             },
6467     {STATE_TEXTURESTAGE(6, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6468     {STATE_TEXTURESTAGE(6, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6469     {STATE_TEXTURESTAGE(6, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6470     {STATE_TEXTURESTAGE(6, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6471     {STATE_TEXTURESTAGE(6, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6472     {STATE_TEXTURESTAGE(6, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6473     {STATE_TEXTURESTAGE(6, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6474     {STATE_TEXTURESTAGE(6, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6475     {STATE_TEXTURESTAGE(6, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6476     {STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6477     {STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6478     {STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6479     {STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6480     {STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6481     {STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVLSCALE),    NULL                    }, WINED3D_GL_EXT_NONE             },
6482     {STATE_TEXTURESTAGE(7, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6483     {STATE_TEXTURESTAGE(7, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6484     {STATE_TEXTURESTAGE(7, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6485     {STATE_TEXTURESTAGE(7, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6486     {STATE_TEXTURESTAGE(7, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6487     {STATE_TEXTURESTAGE(7, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6488     {STATE_TEXTURESTAGE(7, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6489     {STATE_TEXTURESTAGE(7, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6490     {STATE_TEXTURESTAGE(7, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6491     {STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6492     {STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6493     {STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6494     {STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6495     {STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6496     {STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVLSCALE),    NULL                    }, WINED3D_GL_EXT_NONE             },
6497     {STATE_PIXELSHADER,                                   { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6498     {STATE_RENDER(WINED3DRS_FOGENABLE),                   { STATE_RENDER(WINED3DRS_FOGENABLE),                  state_arbfp_fog         }, WINED3D_GL_EXT_NONE             },
6499     {STATE_RENDER(WINED3DRS_FOGTABLEMODE),                { STATE_RENDER(WINED3DRS_FOGENABLE),                  NULL                    }, WINED3D_GL_EXT_NONE             },
6500     {STATE_RENDER(WINED3DRS_FOGVERTEXMODE),               { STATE_RENDER(WINED3DRS_FOGENABLE),                  NULL                    }, WINED3D_GL_EXT_NONE             },
6501     {STATE_RENDER(WINED3DRS_FOGSTART),                    { STATE_RENDER(WINED3DRS_FOGSTART),                   state_fogstartend       }, WINED3D_GL_EXT_NONE             },
6502     {STATE_RENDER(WINED3DRS_FOGEND),                      { STATE_RENDER(WINED3DRS_FOGSTART),                   NULL                    }, WINED3D_GL_EXT_NONE             },
6503     {STATE_RENDER(WINED3DRS_SRGBWRITEENABLE),             { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6504     {STATE_RENDER(WINED3DRS_FOGCOLOR),                    { STATE_RENDER(WINED3DRS_FOGCOLOR),                   state_fogcolor          }, WINED3D_GL_EXT_NONE             },
6505     {STATE_RENDER(WINED3DRS_FOGDENSITY),                  { STATE_RENDER(WINED3DRS_FOGDENSITY),                 state_fogdensity        }, WINED3D_GL_EXT_NONE             },
6506     {STATE_TEXTURESTAGE(0,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(0, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6507     {STATE_TEXTURESTAGE(1,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(1, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6508     {STATE_TEXTURESTAGE(2,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(2, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6509     {STATE_TEXTURESTAGE(3,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(3, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6510     {STATE_TEXTURESTAGE(4,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(4, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6511     {STATE_TEXTURESTAGE(5,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(5, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6512     {STATE_TEXTURESTAGE(6,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(6, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6513     {STATE_TEXTURESTAGE(7,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(7, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6514     {STATE_RENDER(WINED3DRS_SPECULARENABLE),              { STATE_RENDER(WINED3DRS_SPECULARENABLE),             state_arb_specularenable}, WINED3D_GL_EXT_NONE             },
6515     {0 /* Terminate */,                                   { 0,                                                  0                       }, WINED3D_GL_EXT_NONE             },
6516 };
6517
6518 const struct fragment_pipeline arbfp_fragment_pipeline = {
6519     arbfp_enable,
6520     arbfp_get_caps,
6521     arbfp_alloc,
6522     arbfp_free,
6523     shader_arb_color_fixup_supported,
6524     arbfp_fragmentstate_template,
6525     TRUE /* We can disable projected textures */
6526 };
6527
6528 struct arbfp_blit_priv {
6529     GLenum yuy2_rect_shader, yuy2_2d_shader;
6530     GLenum uyvy_rect_shader, uyvy_2d_shader;
6531     GLenum yv12_rect_shader, yv12_2d_shader;
6532     GLenum p8_rect_shader, p8_2d_shader;
6533     GLuint palette_texture;
6534 };
6535
6536 static HRESULT arbfp_blit_alloc(struct wined3d_device *device)
6537 {
6538     device->blit_priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(struct arbfp_blit_priv));
6539     if(!device->blit_priv) {
6540         ERR("Out of memory\n");
6541         return E_OUTOFMEMORY;
6542     }
6543     return WINED3D_OK;
6544 }
6545
6546 /* Context activation is done by the caller. */
6547 static void arbfp_blit_free(struct wined3d_device *device)
6548 {
6549     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
6550     struct arbfp_blit_priv *priv = device->blit_priv;
6551
6552     ENTER_GL();
6553     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yuy2_rect_shader));
6554     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yuy2_2d_shader));
6555     GL_EXTCALL(glDeleteProgramsARB(1, &priv->uyvy_rect_shader));
6556     GL_EXTCALL(glDeleteProgramsARB(1, &priv->uyvy_2d_shader));
6557     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yv12_rect_shader));
6558     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yv12_2d_shader));
6559     GL_EXTCALL(glDeleteProgramsARB(1, &priv->p8_rect_shader));
6560     GL_EXTCALL(glDeleteProgramsARB(1, &priv->p8_2d_shader));
6561     checkGLcall("Delete yuv and p8 programs");
6562
6563     if(priv->palette_texture) glDeleteTextures(1, &priv->palette_texture);
6564     LEAVE_GL();
6565
6566     HeapFree(GetProcessHeap(), 0, device->blit_priv);
6567     device->blit_priv = NULL;
6568 }
6569
6570 static BOOL gen_planar_yuv_read(struct wined3d_shader_buffer *buffer, enum complex_fixup fixup,
6571         GLenum textype, char *luminance)
6572 {
6573     char chroma;
6574     const char *tex, *texinstr;
6575
6576     if (fixup == COMPLEX_FIXUP_UYVY) {
6577         chroma = 'x';
6578         *luminance = 'w';
6579     } else {
6580         chroma = 'w';
6581         *luminance = 'x';
6582     }
6583     switch(textype) {
6584         case GL_TEXTURE_2D:             tex = "2D";     texinstr = "TXP"; break;
6585         case GL_TEXTURE_RECTANGLE_ARB:  tex = "RECT";   texinstr = "TEX"; break;
6586         default:
6587             /* This is more tricky than just replacing the texture type - we have to navigate
6588              * properly in the texture to find the correct chroma values
6589              */
6590             FIXME("Implement yuv correction for non-2d, non-rect textures\n");
6591             return FALSE;
6592     }
6593
6594     /* First we have to read the chroma values. This means we need at least two pixels(no filtering),
6595      * or 4 pixels(with filtering). To get the unmodified chromas, we have to rid ourselves of the
6596      * filtering when we sample the texture.
6597      *
6598      * These are the rules for reading the chroma:
6599      *
6600      * Even pixel: Cr
6601      * Even pixel: U
6602      * Odd pixel: V
6603      *
6604      * So we have to get the sampling x position in non-normalized coordinates in integers
6605      */
6606     if(textype != GL_TEXTURE_RECTANGLE_ARB) {
6607         shader_addline(buffer, "MUL texcrd.xy, fragment.texcoord[0], size.x;\n");
6608         shader_addline(buffer, "MOV texcrd.w, size.x;\n");
6609     } else {
6610         shader_addline(buffer, "MOV texcrd, fragment.texcoord[0];\n");
6611     }
6612     /* We must not allow filtering between pixel x and x+1, this would mix U and V
6613      * Vertical filtering is ok. However, bear in mind that the pixel center is at
6614      * 0.5, so add 0.5.
6615      */
6616     shader_addline(buffer, "FLR texcrd.x, texcrd.x;\n");
6617     shader_addline(buffer, "ADD texcrd.x, texcrd.x, coef.y;\n");
6618
6619     /* Divide the x coordinate by 0.5 and get the fraction. This gives 0.25 and 0.75 for the
6620      * even and odd pixels respectively
6621      */
6622     shader_addline(buffer, "MUL texcrd2, texcrd, coef.y;\n");
6623     shader_addline(buffer, "FRC texcrd2, texcrd2;\n");
6624
6625     /* Sample Pixel 1 */
6626     shader_addline(buffer, "%s luminance, texcrd, texture[0], %s;\n", texinstr, tex);
6627
6628     /* Put the value into either of the chroma values */
6629     shader_addline(buffer, "SGE temp.x, texcrd2.x, coef.y;\n");
6630     shader_addline(buffer, "MUL chroma.x, luminance.%c, temp.x;\n", chroma);
6631     shader_addline(buffer, "SLT temp.x, texcrd2.x, coef.y;\n");
6632     shader_addline(buffer, "MUL chroma.y, luminance.%c, temp.x;\n", chroma);
6633
6634     /* Sample pixel 2. If we read an even pixel(SLT above returned 1), sample
6635      * the pixel right to the current one. Otherwise, sample the left pixel.
6636      * Bias and scale the SLT result to -1;1 and add it to the texcrd.x.
6637      */
6638     shader_addline(buffer, "MAD temp.x, temp.x, coef.z, -coef.x;\n");
6639     shader_addline(buffer, "ADD texcrd.x, texcrd, temp.x;\n");
6640     shader_addline(buffer, "%s luminance, texcrd, texture[0], %s;\n", texinstr, tex);
6641
6642     /* Put the value into the other chroma */
6643     shader_addline(buffer, "SGE temp.x, texcrd2.x, coef.y;\n");
6644     shader_addline(buffer, "MAD chroma.y, luminance.%c, temp.x, chroma.y;\n", chroma);
6645     shader_addline(buffer, "SLT temp.x, texcrd2.x, coef.y;\n");
6646     shader_addline(buffer, "MAD chroma.x, luminance.%c, temp.x, chroma.x;\n", chroma);
6647
6648     /* TODO: If filtering is enabled, sample a 2nd pair of pixels left or right of
6649      * the current one and lerp the two U and V values
6650      */
6651
6652     /* This gives the correctly filtered luminance value */
6653     shader_addline(buffer, "TEX luminance, fragment.texcoord[0], texture[0], %s;\n", tex);
6654
6655     return TRUE;
6656 }
6657
6658 static BOOL gen_yv12_read(struct wined3d_shader_buffer *buffer, GLenum textype, char *luminance)
6659 {
6660     const char *tex;
6661
6662     switch(textype) {
6663         case GL_TEXTURE_2D:             tex = "2D";     break;
6664         case GL_TEXTURE_RECTANGLE_ARB:  tex = "RECT";   break;
6665         default:
6666             FIXME("Implement yv12 correction for non-2d, non-rect textures\n");
6667             return FALSE;
6668     }
6669
6670     /* YV12 surfaces contain a WxH sized luminance plane, followed by a (W/2)x(H/2)
6671      * V and a (W/2)x(H/2) U plane, each with 8 bit per pixel. So the effective
6672      * bitdepth is 12 bits per pixel. Since the U and V planes have only half the
6673      * pitch of the luminance plane, the packing into the gl texture is a bit
6674      * unfortunate. If the whole texture is interpreted as luminance data it looks
6675      * approximately like this:
6676      *
6677      *        +----------------------------------+----
6678      *        |                                  |
6679      *        |                                  |
6680      *        |                                  |
6681      *        |                                  |
6682      *        |                                  |   2
6683      *        |            LUMINANCE             |   -
6684      *        |                                  |   3
6685      *        |                                  |
6686      *        |                                  |
6687      *        |                                  |
6688      *        |                                  |
6689      *        +----------------+-----------------+----
6690      *        |                |                 |
6691      *        |  U even rows   |  U odd rows     |
6692      *        |                |                 |   1
6693      *        +----------------+------------------   -
6694      *        |                |                 |   3
6695      *        |  V even rows   |  V odd rows     |
6696      *        |                |                 |
6697      *        +----------------+-----------------+----
6698      *        |                |                 |
6699      *        |     0.5        |       0.5       |
6700      *
6701      * So it appears as if there are 4 chroma images, but in fact the odd rows
6702      * in the chroma images are in the same row as the even ones. So its is
6703      * kinda tricky to read
6704      *
6705      * When reading from rectangle textures, keep in mind that the input y coordinates
6706      * go from 0 to d3d_height, whereas the opengl texture height is 1.5 * d3d_height
6707      */
6708     shader_addline(buffer, "PARAM yv12_coef = {%f, %f, %f, %f};\n",
6709             2.0f / 3.0f, 1.0f / 6.0f, (2.0f / 3.0f) + (1.0f / 6.0f), 1.0f / 3.0f);
6710
6711     shader_addline(buffer, "MOV texcrd, fragment.texcoord[0];\n");
6712     /* the chroma planes have only half the width */
6713     shader_addline(buffer, "MUL texcrd.x, texcrd.x, coef.y;\n");
6714
6715     /* The first value is between 2/3 and 5/6th of the texture's height, so scale+bias
6716      * the coordinate. Also read the right side of the image when reading odd lines
6717      *
6718      * Don't forget to clamp the y values in into the range, otherwise we'll get filtering
6719      * bleeding
6720      */
6721     if(textype == GL_TEXTURE_2D) {
6722
6723         shader_addline(buffer, "RCP chroma.w, size.y;\n");
6724
6725         shader_addline(buffer, "MUL texcrd2.y, texcrd.y, size.y;\n");
6726
6727         shader_addline(buffer, "FLR texcrd2.y, texcrd2.y;\n");
6728         shader_addline(buffer, "MAD texcrd.y, texcrd.y, yv12_coef.y, yv12_coef.x;\n");
6729
6730         /* Read odd lines from the right side(add size * 0.5 to the x coordinate */
6731         shader_addline(buffer, "ADD texcrd2.x, texcrd2.y, yv12_coef.y;\n"); /* To avoid 0.5 == 0.5 comparisons */
6732         shader_addline(buffer, "FRC texcrd2.x, texcrd2.x;\n");
6733         shader_addline(buffer, "SGE texcrd2.x, texcrd2.x, coef.y;\n");
6734         shader_addline(buffer, "MAD texcrd.x, texcrd2.x, coef.y, texcrd.x;\n");
6735
6736         /* clamp, keep the half pixel origin in mind */
6737         shader_addline(buffer, "MAD temp.y, coef.y, chroma.w, yv12_coef.x;\n");
6738         shader_addline(buffer, "MAX texcrd.y, temp.y, texcrd.y;\n");
6739         shader_addline(buffer, "MAD temp.y, -coef.y, chroma.w, yv12_coef.z;\n");
6740         shader_addline(buffer, "MIN texcrd.y, temp.y, texcrd.y;\n");
6741     } else {
6742         /* Read from [size - size+size/4] */
6743         shader_addline(buffer, "FLR texcrd.y, texcrd.y;\n");
6744         shader_addline(buffer, "MAD texcrd.y, texcrd.y, coef.w, size.y;\n");
6745
6746         /* Read odd lines from the right side(add size * 0.5 to the x coordinate */
6747         shader_addline(buffer, "ADD texcrd2.x, texcrd.y, yv12_coef.y;\n"); /* To avoid 0.5 == 0.5 comparisons */
6748         shader_addline(buffer, "FRC texcrd2.x, texcrd2.x;\n");
6749         shader_addline(buffer, "SGE texcrd2.x, texcrd2.x, coef.y;\n");
6750         shader_addline(buffer, "MUL texcrd2.x, texcrd2.x, size.x;\n");
6751         shader_addline(buffer, "MAD texcrd.x, texcrd2.x, coef.y, texcrd.x;\n");
6752
6753         /* Make sure to read exactly from the pixel center */
6754         shader_addline(buffer, "FLR texcrd.y, texcrd.y;\n");
6755         shader_addline(buffer, "ADD texcrd.y, texcrd.y, coef.y;\n");
6756
6757         /* Clamp */
6758         shader_addline(buffer, "MAD temp.y, size.y, coef.w, size.y;\n");
6759         shader_addline(buffer, "ADD temp.y, temp.y, -coef.y;\n");
6760         shader_addline(buffer, "MIN texcrd.y, temp.y, texcrd.y;\n");
6761         shader_addline(buffer, "ADD temp.y, size.y, -coef.y;\n");
6762         shader_addline(buffer, "MAX texcrd.y, temp.y, texcrd.y;\n");
6763     }
6764     /* Read the texture, put the result into the output register */
6765     shader_addline(buffer, "TEX temp, texcrd, texture[0], %s;\n", tex);
6766     shader_addline(buffer, "MOV chroma.x, temp.w;\n");
6767
6768     /* The other chroma value is 1/6th of the texture lower, from 5/6th to 6/6th
6769      * No need to clamp because we're just reusing the already clamped value from above
6770      */
6771     if(textype == GL_TEXTURE_2D) {
6772         shader_addline(buffer, "ADD texcrd.y, texcrd.y, yv12_coef.y;\n");
6773     } else {
6774         shader_addline(buffer, "MAD texcrd.y, size.y, coef.w, texcrd.y;\n");
6775     }
6776     shader_addline(buffer, "TEX temp, texcrd, texture[0], %s;\n", tex);
6777     shader_addline(buffer, "MOV chroma.y, temp.w;\n");
6778
6779     /* Sample the luminance value. It is in the top 2/3rd of the texture, so scale the y coordinate.
6780      * Clamp the y coordinate to prevent the chroma values from bleeding into the sampled luminance
6781      * values due to filtering
6782      */
6783     shader_addline(buffer, "MOV texcrd, fragment.texcoord[0];\n");
6784     if(textype == GL_TEXTURE_2D) {
6785         /* Multiply the y coordinate by 2/3 and clamp it */
6786         shader_addline(buffer, "MUL texcrd.y, texcrd.y, yv12_coef.x;\n");
6787         shader_addline(buffer, "MAD temp.y, -coef.y, chroma.w, yv12_coef.x;\n");
6788         shader_addline(buffer, "MIN texcrd.y, temp.y, texcrd.y;\n");
6789         shader_addline(buffer, "TEX luminance, texcrd, texture[0], %s;\n", tex);
6790     } else {
6791         /* Reading from texture_rectangles is pretty straightforward, just use the unmodified
6792          * texture coordinate. It is still a good idea to clamp it though, since the opengl texture
6793          * is bigger
6794          */
6795         shader_addline(buffer, "ADD temp.x, size.y, -coef.y;\n");
6796         shader_addline(buffer, "MIN texcrd.y, texcrd.y, size.x;\n");
6797         shader_addline(buffer, "TEX luminance, texcrd, texture[0], %s;\n", tex);
6798     }
6799     *luminance = 'a';
6800
6801     return TRUE;
6802 }
6803
6804 static GLuint gen_p8_shader(struct arbfp_blit_priv *priv,
6805         const struct wined3d_gl_info *gl_info, GLenum textype)
6806 {
6807     GLenum shader;
6808     struct wined3d_shader_buffer buffer;
6809     GLint pos;
6810
6811     /* Shader header */
6812     if (!shader_buffer_init(&buffer))
6813     {
6814         ERR("Failed to initialize shader buffer.\n");
6815         return 0;
6816     }
6817
6818     ENTER_GL();
6819     GL_EXTCALL(glGenProgramsARB(1, &shader));
6820     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader));
6821     LEAVE_GL();
6822     if(!shader) {
6823         shader_buffer_free(&buffer);
6824         return 0;
6825     }
6826
6827     shader_addline(&buffer, "!!ARBfp1.0\n");
6828     shader_addline(&buffer, "TEMP index;\n");
6829
6830     /* { 255/256, 0.5/255*255/256, 0, 0 } */
6831     shader_addline(&buffer, "PARAM constants = { 0.996, 0.00195, 0, 0 };\n");
6832
6833     /* The alpha-component contains the palette index */
6834     if(textype == GL_TEXTURE_RECTANGLE_ARB)
6835         shader_addline(&buffer, "TXP index, fragment.texcoord[0], texture[0], RECT;\n");
6836     else
6837         shader_addline(&buffer, "TEX index, fragment.texcoord[0], texture[0], 2D;\n");
6838
6839     /* Scale the index by 255/256 and add a bias of '0.5' in order to sample in the middle */
6840     shader_addline(&buffer, "MAD index.a, index.a, constants.x, constants.y;\n");
6841
6842     /* Use the alpha-component as an index in the palette to get the final color */
6843     shader_addline(&buffer, "TEX result.color, index.a, texture[1], 1D;\n");
6844     shader_addline(&buffer, "END\n");
6845
6846     ENTER_GL();
6847     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
6848             strlen(buffer.buffer), buffer.buffer));
6849     checkGLcall("glProgramStringARB()");
6850
6851     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
6852     if (pos != -1)
6853     {
6854         FIXME("Fragment program error at position %d: %s\n\n", pos,
6855               debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
6856         shader_arb_dump_program_source(buffer.buffer);
6857     }
6858
6859     if (textype == GL_TEXTURE_RECTANGLE_ARB)
6860         priv->p8_rect_shader = shader;
6861     else
6862         priv->p8_2d_shader = shader;
6863
6864     shader_buffer_free(&buffer);
6865     LEAVE_GL();
6866
6867     return shader;
6868 }
6869
6870 /* Context activation is done by the caller. */
6871 static void upload_palette(struct wined3d_surface *surface)
6872 {
6873     BYTE table[256][4];
6874     struct wined3d_device *device = surface->resource.device;
6875     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
6876     struct arbfp_blit_priv *priv = device->blit_priv;
6877     BOOL colorkey = (surface->CKeyFlags & WINEDDSD_CKSRCBLT) ? TRUE : FALSE;
6878
6879     d3dfmt_p8_init_palette(surface, table, colorkey);
6880
6881     ENTER_GL();
6882     if (!priv->palette_texture)
6883         glGenTextures(1, &priv->palette_texture);
6884
6885     GL_EXTCALL(glActiveTextureARB(GL_TEXTURE1));
6886     glBindTexture(GL_TEXTURE_1D, priv->palette_texture);
6887
6888     glTexEnvi(GL_TEXTURE_ENV, GL_TEXTURE_ENV_MODE, GL_REPLACE);
6889
6890     glTexParameteri(GL_TEXTURE_1D, GL_TEXTURE_MIN_FILTER, GL_NEAREST);
6891     /* Make sure we have discrete color levels. */
6892     glTexParameteri(GL_TEXTURE_1D, GL_TEXTURE_MAG_FILTER, GL_NEAREST);
6893     glTexParameteri(GL_TEXTURE_1D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
6894     /* Upload the palette */
6895     /* TODO: avoid unneeed uploads in the future by adding some SFLAG_PALETTE_DIRTY mechanism */
6896     glTexImage1D(GL_TEXTURE_1D, 0, GL_RGBA, 256, 0, GL_RGBA, GL_UNSIGNED_BYTE, table);
6897
6898     /* Switch back to unit 0 in which the 2D texture will be stored. */
6899     GL_EXTCALL(glActiveTextureARB(GL_TEXTURE0));
6900     LEAVE_GL();
6901 }
6902
6903 /* Context activation is done by the caller. */
6904 static GLuint gen_yuv_shader(struct arbfp_blit_priv *priv, const struct wined3d_gl_info *gl_info,
6905         enum complex_fixup yuv_fixup, GLenum textype)
6906 {
6907     GLenum shader;
6908     struct wined3d_shader_buffer buffer;
6909     char luminance_component;
6910     GLint pos;
6911
6912     /* Shader header */
6913     if (!shader_buffer_init(&buffer))
6914     {
6915         ERR("Failed to initialize shader buffer.\n");
6916         return 0;
6917     }
6918
6919     ENTER_GL();
6920     GL_EXTCALL(glGenProgramsARB(1, &shader));
6921     checkGLcall("GL_EXTCALL(glGenProgramsARB(1, &shader))");
6922     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader));
6923     checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader)");
6924     LEAVE_GL();
6925     if(!shader) {
6926         shader_buffer_free(&buffer);
6927         return 0;
6928     }
6929
6930     /* The YUY2 and UYVY formats contain two pixels packed into a 32 bit macropixel,
6931      * giving effectively 16 bit per pixel. The color consists of a luminance(Y) and
6932      * two chroma(U and V) values. Each macropixel has two luminance values, one for
6933      * each single pixel it contains, and one U and one V value shared between both
6934      * pixels.
6935      *
6936      * The data is loaded into an A8L8 texture. With YUY2, the luminance component
6937      * contains the luminance and alpha the chroma. With UYVY it is vice versa. Thus
6938      * take the format into account when generating the read swizzles
6939      *
6940      * Reading the Y value is straightforward - just sample the texture. The hardware
6941      * takes care of filtering in the horizontal and vertical direction.
6942      *
6943      * Reading the U and V values is harder. We have to avoid filtering horizontally,
6944      * because that would mix the U and V values of one pixel or two adjacent pixels.
6945      * Thus floor the texture coordinate and add 0.5 to get an unfiltered read,
6946      * regardless of the filtering setting. Vertical filtering works automatically
6947      * though - the U and V values of two rows are mixed nicely.
6948      *
6949      * Appart of avoiding filtering issues, the code has to know which value it just
6950      * read, and where it can find the other one. To determine this, it checks if
6951      * it sampled an even or odd pixel, and shifts the 2nd read accordingly.
6952      *
6953      * Handling horizontal filtering of U and V values requires reading a 2nd pair
6954      * of pixels, extracting U and V and mixing them. This is not implemented yet.
6955      *
6956      * An alternative implementation idea is to load the texture as A8R8G8B8 texture,
6957      * with width / 2. This way one read gives all 3 values, finding U and V is easy
6958      * in an unfiltered situation. Finding the luminance on the other hand requires
6959      * finding out if it is an odd or even pixel. The real drawback of this approach
6960      * is filtering. This would have to be emulated completely in the shader, reading
6961      * up two 2 packed pixels in up to 2 rows and interpolating both horizontally and
6962      * vertically. Beyond that it would require adjustments to the texture handling
6963      * code to deal with the width scaling
6964      */
6965     shader_addline(&buffer, "!!ARBfp1.0\n");
6966     shader_addline(&buffer, "TEMP luminance;\n");
6967     shader_addline(&buffer, "TEMP temp;\n");
6968     shader_addline(&buffer, "TEMP chroma;\n");
6969     shader_addline(&buffer, "TEMP texcrd;\n");
6970     shader_addline(&buffer, "TEMP texcrd2;\n");
6971     shader_addline(&buffer, "PARAM coef = {1.0, 0.5, 2.0, 0.25};\n");
6972     shader_addline(&buffer, "PARAM yuv_coef = {1.403, 0.344, 0.714, 1.770};\n");
6973     shader_addline(&buffer, "PARAM size = program.local[0];\n");
6974
6975     switch (yuv_fixup)
6976     {
6977         case COMPLEX_FIXUP_UYVY:
6978         case COMPLEX_FIXUP_YUY2:
6979             if (!gen_planar_yuv_read(&buffer, yuv_fixup, textype, &luminance_component))
6980             {
6981                 shader_buffer_free(&buffer);
6982                 return 0;
6983             }
6984             break;
6985
6986         case COMPLEX_FIXUP_YV12:
6987             if (!gen_yv12_read(&buffer, textype, &luminance_component))
6988             {
6989                 shader_buffer_free(&buffer);
6990                 return 0;
6991             }
6992             break;
6993
6994         default:
6995             FIXME("Unsupported YUV fixup %#x\n", yuv_fixup);
6996             shader_buffer_free(&buffer);
6997             return 0;
6998     }
6999
7000     /* Calculate the final result. Formula is taken from
7001      * http://www.fourcc.org/fccyvrgb.php. Note that the chroma
7002      * ranges from -0.5 to 0.5
7003      */
7004     shader_addline(&buffer, "SUB chroma.xy, chroma, coef.y;\n");
7005
7006     shader_addline(&buffer, "MAD result.color.x, chroma.x, yuv_coef.x, luminance.%c;\n", luminance_component);
7007     shader_addline(&buffer, "MAD temp.x, -chroma.y, yuv_coef.y, luminance.%c;\n", luminance_component);
7008     shader_addline(&buffer, "MAD result.color.y, -chroma.x, yuv_coef.z, temp.x;\n");
7009     shader_addline(&buffer, "MAD result.color.z, chroma.y, yuv_coef.w, luminance.%c;\n", luminance_component);
7010     shader_addline(&buffer, "END\n");
7011
7012     ENTER_GL();
7013     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
7014             strlen(buffer.buffer), buffer.buffer));
7015     checkGLcall("glProgramStringARB()");
7016
7017     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
7018     if (pos != -1)
7019     {
7020         FIXME("Fragment program error at position %d: %s\n\n", pos,
7021               debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
7022         shader_arb_dump_program_source(buffer.buffer);
7023     }
7024     else
7025     {
7026         GLint native;
7027
7028         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
7029         checkGLcall("glGetProgramivARB()");
7030         if (!native) WARN("Program exceeds native resource limits.\n");
7031     }
7032
7033     shader_buffer_free(&buffer);
7034     LEAVE_GL();
7035
7036     switch (yuv_fixup)
7037     {
7038         case COMPLEX_FIXUP_YUY2:
7039             if (textype == GL_TEXTURE_RECTANGLE_ARB) priv->yuy2_rect_shader = shader;
7040             else priv->yuy2_2d_shader = shader;
7041             break;
7042
7043         case COMPLEX_FIXUP_UYVY:
7044             if (textype == GL_TEXTURE_RECTANGLE_ARB) priv->uyvy_rect_shader = shader;
7045             else priv->uyvy_2d_shader = shader;
7046             break;
7047
7048         case COMPLEX_FIXUP_YV12:
7049             if (textype == GL_TEXTURE_RECTANGLE_ARB) priv->yv12_rect_shader = shader;
7050             else priv->yv12_2d_shader = shader;
7051             break;
7052         default:
7053             ERR("Unsupported complex fixup: %d\n", yuv_fixup);
7054     }
7055
7056     return shader;
7057 }
7058
7059 /* Context activation is done by the caller. */
7060 static HRESULT arbfp_blit_set(void *blit_priv, const struct wined3d_gl_info *gl_info, struct wined3d_surface *surface)
7061 {
7062     GLenum shader;
7063     float size[4] = {(float) surface->pow2Width, (float) surface->pow2Height, 1.0f, 1.0f};
7064     struct arbfp_blit_priv *priv = blit_priv;
7065     enum complex_fixup fixup;
7066     GLenum textype = surface->texture_target;
7067
7068     if (!is_complex_fixup(surface->resource.format->color_fixup))
7069     {
7070         TRACE("Fixup:\n");
7071         dump_color_fixup_desc(surface->resource.format->color_fixup);
7072         /* Don't bother setting up a shader for unconverted formats */
7073         ENTER_GL();
7074         glEnable(textype);
7075         checkGLcall("glEnable(textype)");
7076         LEAVE_GL();
7077         return WINED3D_OK;
7078     }
7079
7080     fixup = get_complex_fixup(surface->resource.format->color_fixup);
7081
7082     switch(fixup)
7083     {
7084         case COMPLEX_FIXUP_YUY2:
7085             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->yuy2_rect_shader : priv->yuy2_2d_shader;
7086             break;
7087
7088         case COMPLEX_FIXUP_UYVY:
7089             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->uyvy_rect_shader : priv->uyvy_2d_shader;
7090             break;
7091
7092         case COMPLEX_FIXUP_YV12:
7093             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->yv12_rect_shader : priv->yv12_2d_shader;
7094             break;
7095
7096         case COMPLEX_FIXUP_P8:
7097             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->p8_rect_shader : priv->p8_2d_shader;
7098             if (!shader) shader = gen_p8_shader(priv, gl_info, textype);
7099
7100             upload_palette(surface);
7101             break;
7102
7103         default:
7104             FIXME("Unsupported complex fixup %#x, not setting a shader\n", fixup);
7105             ENTER_GL();
7106             glEnable(textype);
7107             checkGLcall("glEnable(textype)");
7108             LEAVE_GL();
7109             return E_NOTIMPL;
7110     }
7111
7112     if (!shader) shader = gen_yuv_shader(priv, gl_info, fixup, textype);
7113
7114     ENTER_GL();
7115     glEnable(GL_FRAGMENT_PROGRAM_ARB);
7116     checkGLcall("glEnable(GL_FRAGMENT_PROGRAM_ARB)");
7117     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader));
7118     checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader)");
7119     GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, 0, size));
7120     checkGLcall("glProgramLocalParameter4fvARB");
7121     LEAVE_GL();
7122
7123     return WINED3D_OK;
7124 }
7125
7126 /* Context activation is done by the caller. */
7127 static void arbfp_blit_unset(const struct wined3d_gl_info *gl_info)
7128 {
7129     ENTER_GL();
7130     glDisable(GL_FRAGMENT_PROGRAM_ARB);
7131     checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
7132     glDisable(GL_TEXTURE_2D);
7133     checkGLcall("glDisable(GL_TEXTURE_2D)");
7134     if (gl_info->supported[ARB_TEXTURE_CUBE_MAP])
7135     {
7136         glDisable(GL_TEXTURE_CUBE_MAP_ARB);
7137         checkGLcall("glDisable(GL_TEXTURE_CUBE_MAP_ARB)");
7138     }
7139     if (gl_info->supported[ARB_TEXTURE_RECTANGLE])
7140     {
7141         glDisable(GL_TEXTURE_RECTANGLE_ARB);
7142         checkGLcall("glDisable(GL_TEXTURE_RECTANGLE_ARB)");
7143     }
7144     LEAVE_GL();
7145 }
7146
7147 static BOOL arbfp_blit_supported(const struct wined3d_gl_info *gl_info, enum wined3d_blit_op blit_op,
7148         const RECT *src_rect, DWORD src_usage, WINED3DPOOL src_pool, const struct wined3d_format *src_format,
7149         const RECT *dst_rect, DWORD dst_usage, WINED3DPOOL dst_pool, const struct wined3d_format *dst_format)
7150 {
7151     enum complex_fixup src_fixup;
7152
7153     if (!gl_info->supported[ARB_FRAGMENT_PROGRAM])
7154         return FALSE;
7155
7156     if (blit_op != WINED3D_BLIT_OP_COLOR_BLIT)
7157     {
7158         TRACE("Unsupported blit_op=%d\n", blit_op);
7159         return FALSE;
7160     }
7161
7162     src_fixup = get_complex_fixup(src_format->color_fixup);
7163     if (TRACE_ON(d3d_shader) && TRACE_ON(d3d))
7164     {
7165         TRACE("Checking support for fixup:\n");
7166         dump_color_fixup_desc(src_format->color_fixup);
7167     }
7168
7169     if (!is_identity_fixup(dst_format->color_fixup))
7170     {
7171         TRACE("Destination fixups are not supported\n");
7172         return FALSE;
7173     }
7174
7175     if (is_identity_fixup(src_format->color_fixup))
7176     {
7177         TRACE("[OK]\n");
7178         return TRUE;
7179     }
7180
7181      /* We only support YUV conversions. */
7182     if (!is_complex_fixup(src_format->color_fixup))
7183     {
7184         TRACE("[FAILED]\n");
7185         return FALSE;
7186     }
7187
7188     switch(src_fixup)
7189     {
7190         case COMPLEX_FIXUP_YUY2:
7191         case COMPLEX_FIXUP_UYVY:
7192         case COMPLEX_FIXUP_YV12:
7193         case COMPLEX_FIXUP_P8:
7194             TRACE("[OK]\n");
7195             return TRUE;
7196
7197         default:
7198             FIXME("Unsupported YUV fixup %#x\n", src_fixup);
7199             TRACE("[FAILED]\n");
7200             return FALSE;
7201     }
7202 }
7203
7204 HRESULT arbfp_blit_surface(struct wined3d_device *device, struct wined3d_surface *src_surface, const RECT *src_rect,
7205         struct wined3d_surface *dst_surface, const RECT *dst_rect_in, enum wined3d_blit_op blit_op, DWORD Filter)
7206 {
7207     struct wined3d_context *context;
7208     RECT dst_rect = *dst_rect_in;
7209
7210     /* Now load the surface */
7211     surface_internal_preload(src_surface, SRGB_RGB);
7212
7213     /* Activate the destination context, set it up for blitting */
7214     context = context_acquire(device, dst_surface);
7215     context_apply_blit_state(context, device);
7216
7217     if (!surface_is_offscreen(dst_surface))
7218         surface_translate_drawable_coords(dst_surface, context->win_handle, &dst_rect);
7219
7220     arbfp_blit_set(device->blit_priv, context->gl_info, src_surface);
7221
7222     ENTER_GL();
7223
7224     /* Draw a textured quad */
7225     draw_textured_quad(src_surface, src_rect, &dst_rect, Filter);
7226
7227     LEAVE_GL();
7228
7229     /* Leave the opengl state valid for blitting */
7230     arbfp_blit_unset(context->gl_info);
7231
7232     if (wined3d_settings.strict_draw_ordering
7233             || (dst_surface->container.type == WINED3D_CONTAINER_SWAPCHAIN
7234             && (dst_surface->container.u.swapchain->front_buffer == dst_surface)))
7235         wglFlush(); /* Flush to ensure ordering across contexts. */
7236
7237     context_release(context);
7238
7239     surface_modify_location(dst_surface, SFLAG_INDRAWABLE, TRUE);
7240     return WINED3D_OK;
7241 }
7242
7243 /* Do not call while under the GL lock. */
7244 static HRESULT arbfp_blit_color_fill(struct wined3d_device *device, struct wined3d_surface *dst_surface,
7245         const RECT *dst_rect, const WINED3DCOLORVALUE *color)
7246 {
7247     FIXME("Color filling not implemented by arbfp_blit\n");
7248     return WINED3DERR_INVALIDCALL;
7249 }
7250
7251 /* Do not call while under the GL lock. */
7252 static HRESULT arbfp_blit_depth_fill(struct wined3d_device *device,
7253         struct wined3d_surface *surface, const RECT *rect, float depth)
7254 {
7255     FIXME("Depth filling not implemented by arbfp_blit.\n");
7256     return WINED3DERR_INVALIDCALL;
7257 }
7258
7259 const struct blit_shader arbfp_blit = {
7260     arbfp_blit_alloc,
7261     arbfp_blit_free,
7262     arbfp_blit_set,
7263     arbfp_blit_unset,
7264     arbfp_blit_supported,
7265     arbfp_blit_color_fill,
7266     arbfp_blit_depth_fill,
7267 };