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 #define GLINFO_LOCATION      (*gl_info)
  43
  44 /* GL locking for state handlers is done by the caller. */
  45 static BOOL need_mova_const(IWineD3DBaseShader *shader, const struct wined3d_gl_info *gl_info)
  46 {
  47     IWineD3DBaseShaderImpl *This = (IWineD3DBaseShaderImpl *) shader;
  48     if(!This->baseShader.reg_maps.usesmova) return FALSE;
  49     return !GL_SUPPORT(NV_VERTEX_PROGRAM2_OPTION);
  50 }
  51
  52 /* Returns TRUE if result.clip from GL_NV_vertex_program2 should be used and FALSE otherwise */
  53 static inline BOOL use_nv_clip(const struct wined3d_gl_info *gl_info)
  54 {
  55     return GL_SUPPORT(NV_VERTEX_PROGRAM2_OPTION);
  56 }
  57
  58 static BOOL need_helper_const(const struct wined3d_gl_info *gl_info)
  59 {
  60     if (!GL_SUPPORT(NV_VERTEX_PROGRAM) /* Need to init colors. */
  61         || gl_info->quirks & WINED3D_QUIRK_ARB_VS_OFFSET_LIMIT /* Load the immval offset. */
  62         || gl_info->quirks & WINED3D_QUIRK_SET_TEXCOORD_W /* Have to init texcoords. */
  63         || (!use_nv_clip(gl_info)) /* Init the clip texcoord */)
  64     {
  65         return TRUE;
  66     }
  67     return FALSE;
  68 }
  69
  70 static unsigned int reserved_vs_const(IWineD3DBaseShader *shader, const struct wined3d_gl_info *gl_info)
  71 {
  72     unsigned int ret = 1;
  73     /* We use one PARAM for the pos fixup, and in some cases one to load
  74      * some immediate values into the shader
  75      */
  76     if(need_helper_const(gl_info)) ret++;
  77     if(need_mova_const(shader, gl_info)) ret++;
  78     return ret;
  79 }
  80
  81 static inline BOOL ffp_clip_emul(IWineD3DStateBlockImpl *stateblock)
  82 {
  83     return stateblock->lowest_disabled_stage < 7;
  84 }
  85
  86 /* Internally used shader constants. Applications can use constants 0 to GL_LIMITS(vshader_constantsF) - 1,
  87  * so upload them above that
  88  */
  89 #define ARB_SHADER_PRIVCONST_BASE (GL_LIMITS(vshader_constantsF) - 1)
  90 #define ARB_SHADER_PRIVCONST_POS ARB_SHADER_PRIVCONST_BASE + 0
  91
  92 /* ARB_program_shader private data */
  93
  94 struct control_frame
  95 {
  96     struct                          list entry;
  97     enum
  98     {
  99         IF,
 100         IFC,
 101         LOOP,
 102         REP
 103     } type;
 104     BOOL                            muting;
 105     BOOL                            outer_loop;
 106     union
 107     {
 108         unsigned int                loop_no;
 109         unsigned int                ifc_no;
 110     };
 111     struct wined3d_shader_loop_control loop_control;
 112     BOOL                            had_else;
 113 };
 114
 115 struct arb_ps_np2fixup_info
 116 {
 117     struct ps_np2fixup_info         super;
 118     /* For ARB we need a offset value:
 119      * With both GLSL and ARB mode the NP2 fixup information (the texture dimensions) are stored in a
 120      * consecutive way (GLSL uses a uniform array). Since ARB doesn't know the notion of a "standalone"
 121      * array we need an offset to the index inside the program local parameter array. */
 122     UINT                            offset;
 123 };
 124
 125 struct arb_ps_compile_args
 126 {
 127     struct ps_compile_args          super;
 128     WORD                            bools;
 129     WORD                            clip;  /* only a boolean, use a WORD for alignment */
 130     unsigned char                   loop_ctrl[MAX_CONST_I][3];
 131 };
 132
 133 struct stb_const_desc
 134 {
 135     unsigned char           texunit;
 136     UINT                    const_num;
 137 };
 138
 139 struct arb_ps_compiled_shader
 140 {
 141     struct arb_ps_compile_args      args;
 142     struct arb_ps_np2fixup_info     np2fixup_info;
 143     struct stb_const_desc           bumpenvmatconst[MAX_TEXTURES];
 144     struct stb_const_desc           luminanceconst[MAX_TEXTURES];
 145     UINT                            int_consts[MAX_CONST_I];
 146     GLuint                          prgId;
 147     UINT                            ycorrection;
 148     unsigned char                   numbumpenvmatconsts;
 149     char                            num_int_consts;
 150 };
 151
 152 struct arb_vs_compile_args
 153 {
 154     struct vs_compile_args          super;
 155     union
 156     {
 157         struct
 158         {
 159             WORD                    bools;
 160             char                    clip_control[2];
 161         }                           boolclip;
 162         DWORD                       boolclip_compare;
 163     };
 164     DWORD                           ps_signature;
 165     union
 166     {
 167         unsigned char               vertex_samplers[4];
 168         DWORD                       vertex_samplers_compare;
 169     };
 170     unsigned char                   loop_ctrl[MAX_CONST_I][3];
 171 };
 172
 173 struct arb_vs_compiled_shader
 174 {
 175     struct arb_vs_compile_args      args;
 176     GLuint                          prgId;
 177     UINT                            int_consts[MAX_CONST_I];
 178     char                            num_int_consts;
 179     char                            need_color_unclamp;
 180     UINT                            pos_fixup;
 181 };
 182
 183 struct recorded_instruction
 184 {
 185     struct wined3d_shader_instruction ins;
 186     struct list entry;
 187 };
 188
 189 struct shader_arb_ctx_priv
 190 {
 191     char addr_reg[20];
 192     enum
 193     {
 194         /* plain GL_ARB_vertex_program or GL_ARB_fragment_program */
 195         ARB,
 196         /* GL_NV_vertex_progam2_option or GL_NV_fragment_program_option */
 197         NV2,
 198         /* GL_NV_vertex_program3 or GL_NV_fragment_program2 */
 199         NV3
 200     } target_version;
 201
 202     const struct arb_vs_compile_args    *cur_vs_args;
 203     const struct arb_ps_compile_args    *cur_ps_args;
 204     const struct arb_ps_compiled_shader *compiled_fprog;
 205     const struct arb_vs_compiled_shader *compiled_vprog;
 206     struct arb_ps_np2fixup_info         *cur_np2fixup_info;
 207     struct list                         control_frames;
 208     struct list                         record;
 209     BOOL                                recording;
 210     BOOL                                muted;
 211     unsigned int                        num_loops, loop_depth, num_ifcs;
 212     int                                 aL;
 213
 214     unsigned int                        vs_clipplanes;
 215     BOOL                                footer_written;
 216     BOOL                                in_main_func;
 217
 218     /* For 3.0 vertex shaders */
 219     const char                          *vs_output[MAX_REG_OUTPUT];
 220     /* For 2.x and earlier vertex shaders */
 221     const char                          *texcrd_output[8], *color_output[2], *fog_output;
 222
 223     /* 3.0 pshader input for compatibility with fixed function */
 224     const char                          *ps_input[MAX_REG_INPUT];
 225 };
 226
 227 struct ps_signature
 228 {
 229     struct wined3d_shader_signature_element *sig;
 230     DWORD                               idx;
 231     struct wine_rb_entry                entry;
 232 };
 233
 234 struct arb_pshader_private {
 235     struct arb_ps_compiled_shader   *gl_shaders;
 236     UINT                            num_gl_shaders, shader_array_size;
 237     BOOL                            has_signature_idx;
 238     DWORD                           input_signature_idx;
 239     DWORD                           clipplane_emulation;
 240     BOOL                            clamp_consts;
 241 };
 242
 243 struct arb_vshader_private {
 244     struct arb_vs_compiled_shader   *gl_shaders;
 245     UINT                            num_gl_shaders, shader_array_size;
 246 };
 247
 248 struct shader_arb_priv
 249 {
 250     GLuint                  current_vprogram_id;
 251     GLuint                  current_fprogram_id;
 252     const struct arb_ps_compiled_shader *compiled_fprog;
 253     const struct arb_vs_compiled_shader *compiled_vprog;
 254     GLuint                  depth_blt_vprogram_id;
 255     GLuint                  depth_blt_fprogram_id[tex_type_count];
 256     BOOL                    use_arbfp_fixed_func;
 257     struct wine_rb_tree     fragment_shaders;
 258     BOOL                    last_ps_const_clamped;
 259     BOOL                    last_vs_color_unclamp;
 260
 261     struct wine_rb_tree     signature_tree;
 262     DWORD ps_sig_number;
 263 };
 264
 265 /********************************************************
 266  * ARB_[vertex/fragment]_program helper functions follow
 267  ********************************************************/
 268
 269 /* Loads floating point constants into the currently set ARB_vertex/fragment_program.
 270  * When constant_list == NULL, it will load all the constants.
 271  *
 272  * @target_type should be either GL_VERTEX_PROGRAM_ARB (for vertex shaders)
 273  *  or GL_FRAGMENT_PROGRAM_ARB (for pixel shaders)
 274  */
 275 /* GL locking is done by the caller */
 276 static unsigned int shader_arb_load_constantsF(IWineD3DBaseShaderImpl *This, const struct wined3d_gl_info *gl_info,
 277         GLuint target_type, unsigned int max_constants, const float *constants, char *dirty_consts)
 278 {
 279     local_constant* lconst;
 280     DWORD i, j;
 281     unsigned int ret;
 282
 283     if (TRACE_ON(d3d_constants))
 284     {
 285         for(i = 0; i < max_constants; i++) {
 286             if(!dirty_consts[i]) continue;
 287             TRACE_(d3d_constants)("Loading constants %i: %f, %f, %f, %f\n", i,
 288                         constants[i * 4 + 0], constants[i * 4 + 1],
 289                         constants[i * 4 + 2], constants[i * 4 + 3]);
 290         }
 291     }
 292
 293     i = 0;
 294
 295     /* In 1.X pixel shaders constants are implicitly clamped in the range [-1;1] */
 296     if (target_type == GL_FRAGMENT_PROGRAM_ARB && This->baseShader.reg_maps.shader_version.major == 1)
 297     {
 298         float lcl_const[4];
 299         /* ps 1.x supports only 8 constants, clamp only those. When switching between 1.x and higher
 300          * shaders, the first 8 constants are marked dirty for reload
 301          */
 302         for(; i < min(8, max_constants); i++) {
 303             if(!dirty_consts[i]) continue;
 304             dirty_consts[i] = 0;
 305
 306             j = 4 * i;
 307             if (constants[j + 0] > 1.0f) lcl_const[0] = 1.0f;
 308             else if (constants[j + 0] < -1.0f) lcl_const[0] = -1.0f;
 309             else lcl_const[0] = constants[j + 0];
 310
 311             if (constants[j + 1] > 1.0f) lcl_const[1] = 1.0f;
 312             else if (constants[j + 1] < -1.0f) lcl_const[1] = -1.0f;
 313             else lcl_const[1] = constants[j + 1];
 314
 315             if (constants[j + 2] > 1.0f) lcl_const[2] = 1.0f;
 316             else if (constants[j + 2] < -1.0f) lcl_const[2] = -1.0f;
 317             else lcl_const[2] = constants[j + 2];
 318
 319             if (constants[j + 3] > 1.0f) lcl_const[3] = 1.0f;
 320             else if (constants[j + 3] < -1.0f) lcl_const[3] = -1.0f;
 321             else lcl_const[3] = constants[j + 3];
 322
 323             GL_EXTCALL(glProgramEnvParameter4fvARB(target_type, i, lcl_const));
 324         }
 325
 326         /* If further constants are dirty, reload them without clamping.
 327          *
 328          * The alternative is not to touch them, but then we cannot reset the dirty constant count
 329          * to zero. That's bad for apps that only use PS 1.x shaders, because in that case the code
 330          * above would always re-check the first 8 constants since max_constant remains at the init
 331          * value
 332          */
 333     }
 334
 335     if(GL_SUPPORT(EXT_GPU_PROGRAM_PARAMETERS)) {
 336         /* TODO: Benchmark if we're better of with finding the dirty constants ourselves,
 337          * or just reloading *all* constants at once
 338          *
 339         GL_EXTCALL(glProgramEnvParameters4fvEXT(target_type, i, max_constants, constants + (i * 4)));
 340          */
 341         for(; i < max_constants; i++) {
 342             if(!dirty_consts[i]) continue;
 343
 344             /* Find the next block of dirty constants */
 345             dirty_consts[i] = 0;
 346             j = i;
 347             for(i++; (i < max_constants) && dirty_consts[i]; i++) {
 348                 dirty_consts[i] = 0;
 349             }
 350
 351             GL_EXTCALL(glProgramEnvParameters4fvEXT(target_type, j, i - j, constants + (j * 4)));
 352         }
 353     } else {
 354         for(; i < max_constants; i++) {
 355             if(dirty_consts[i]) {
 356                 dirty_consts[i] = 0;
 357                 GL_EXTCALL(glProgramEnvParameter4fvARB(target_type, i, constants + (i * 4)));
 358             }
 359         }
 360     }
 361     checkGLcall("glProgramEnvParameter4fvARB()");
 362
 363     /* Load immediate constants */
 364     if(This->baseShader.load_local_constsF) {
 365         if (TRACE_ON(d3d_shader)) {
 366             LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry) {
 367                 GLfloat* values = (GLfloat*)lconst->value;
 368                 TRACE_(d3d_constants)("Loading local constants %i: %f, %f, %f, %f\n", lconst->idx,
 369                         values[0], values[1], values[2], values[3]);
 370             }
 371         }
 372         /* Immediate constants are clamped for 1.X shaders at loading times */
 373         ret = 0;
 374         LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry) {
 375             dirty_consts[lconst->idx] = 1; /* Dirtify so the non-immediate constant overwrites it next time */
 376             ret = max(ret, lconst->idx + 1);
 377             GL_EXTCALL(glProgramEnvParameter4fvARB(target_type, lconst->idx, (GLfloat*)lconst->value));
 378         }
 379         checkGLcall("glProgramEnvParameter4fvARB()");
 380         return ret; /* The loaded immediate constants need reloading for the next shader */
 381     } else {
 382         return 0; /* No constants are dirty now */
 383     }
 384 }
 385
 386 /**
 387  * Loads the texture dimensions for NP2 fixup into the currently set ARB_[vertex/fragment]_programs.
 388  */
 389 static void shader_arb_load_np2fixup_constants(
 390     IWineD3DDevice* device,
 391     char usePixelShader,
 392     char useVertexShader) {
 393
 394     IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl *) device;
 395     const struct shader_arb_priv* const priv = (const struct shader_arb_priv *) deviceImpl->shader_priv;
 396     IWineD3DStateBlockImpl* stateBlock = deviceImpl->stateBlock;
 397     const struct wined3d_gl_info *gl_info = &deviceImpl->adapter->gl_info;
 398
 399     if (!usePixelShader) {
 400         /* NP2 texcoord fixup is (currently) only done for pixelshaders. */
 401         return;
 402     }
 403
 404     if (priv->compiled_fprog && priv->compiled_fprog->np2fixup_info.super.active) {
 405         const struct arb_ps_np2fixup_info* const fixup = &priv->compiled_fprog->np2fixup_info;
 406         UINT i;
 407         WORD active = fixup->super.active;
 408         GLfloat np2fixup_constants[4 * MAX_FRAGMENT_SAMPLERS];
 409
 410         for (i = 0; active; active >>= 1, ++i) {
 411             const unsigned char idx = fixup->super.idx[i];
 412             const IWineD3DTextureImpl* const tex = (const IWineD3DTextureImpl*) stateBlock->textures[i];
 413             GLfloat* tex_dim = &np2fixup_constants[(idx >> 1) * 4];
 414
 415             if (!(active & 1)) continue;
 416
 417             if (!tex) {
 418                 FIXME("Nonexistent texture is flagged for NP2 texcoord fixup\n");
 419                 continue;
 420             }
 421
 422             if (idx % 2) {
 423                 tex_dim[2] = tex->baseTexture.pow2Matrix[0]; tex_dim[3] = tex->baseTexture.pow2Matrix[5];
 424             } else {
 425                 tex_dim[0] = tex->baseTexture.pow2Matrix[0]; tex_dim[1] = tex->baseTexture.pow2Matrix[5];
 426             }
 427         }
 428
 429         for (i = 0; i < fixup->super.num_consts; ++i) {
 430             GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB,
 431                                                    fixup->offset + i, &np2fixup_constants[i * 4]));
 432         }
 433     }
 434 }
 435
 436 /* GL locking is done by the caller. */
 437 static inline void shader_arb_ps_local_constants(IWineD3DDeviceImpl* deviceImpl)
 438 {
 439     const struct wined3d_context *context = context_get_current();
 440     IWineD3DStateBlockImpl* stateBlock = deviceImpl->stateBlock;
 441     const struct wined3d_gl_info *gl_info = context->gl_info;
 442     unsigned char i;
 443     struct shader_arb_priv *priv = deviceImpl->shader_priv;
 444     const struct arb_ps_compiled_shader *gl_shader = priv->compiled_fprog;
 445
 446     for(i = 0; i < gl_shader->numbumpenvmatconsts; i++)
 447     {
 448         int texunit = gl_shader->bumpenvmatconst[i].texunit;
 449
 450         /* The state manager takes care that this function is always called if the bump env matrix changes */
 451         const float *data = (const float *)&stateBlock->textureState[texunit][WINED3DTSS_BUMPENVMAT00];
 452         GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, gl_shader->bumpenvmatconst[i].const_num, data));
 453
 454         if (gl_shader->luminanceconst[i].const_num != WINED3D_CONST_NUM_UNUSED)
 455         {
 456             /* WINED3DTSS_BUMPENVLSCALE and WINED3DTSS_BUMPENVLOFFSET are next to each other.
 457              * point gl to the scale, and load 4 floats. x = scale, y = offset, z and w are junk, we
 458              * don't care about them. The pointers are valid for sure because the stateblock is bigger.
 459              * (they're WINED3DTSS_TEXTURETRANSFORMFLAGS and WINED3DTSS_ADDRESSW, so most likely 0 or NaN
 460             */
 461             const float *scale = (const float *)&stateBlock->textureState[texunit][WINED3DTSS_BUMPENVLSCALE];
 462             GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, gl_shader->luminanceconst[i].const_num, scale));
 463         }
 464     }
 465     checkGLcall("Load bumpmap consts");
 466
 467     if(gl_shader->ycorrection != WINED3D_CONST_NUM_UNUSED)
 468     {
 469         /* ycorrection.x: Backbuffer height(onscreen) or 0(offscreen).
 470         * ycorrection.y: -1.0(onscreen), 1.0(offscreen)
 471         * ycorrection.z: 1.0
 472         * ycorrection.w: 0.0
 473         */
 474         float val[4];
 475         val[0] = context->render_offscreen ? 0.0f
 476                 : ((IWineD3DSurfaceImpl *) deviceImpl->render_targets[0])->currentDesc.Height;
 477         val[1] = context->render_offscreen ? 1.0f : -1.0f;
 478         val[2] = 1.0f;
 479         val[3] = 0.0f;
 480         GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, gl_shader->ycorrection, val));
 481         checkGLcall("y correction loading");
 482     }
 483
 484     if(gl_shader->num_int_consts == 0) return;
 485
 486     for(i = 0; i < MAX_CONST_I; i++)
 487     {
 488         if(gl_shader->int_consts[i] != WINED3D_CONST_NUM_UNUSED)
 489         {
 490             float val[4];
 491             val[0] = stateBlock->pixelShaderConstantI[4 * i];
 492             val[1] = stateBlock->pixelShaderConstantI[4 * i + 1];
 493             val[2] = stateBlock->pixelShaderConstantI[4 * i + 2];
 494             val[3] = -1.0f;
 495
 496             GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, gl_shader->int_consts[i], val));
 497         }
 498     }
 499     checkGLcall("Load ps int consts");
 500 }
 501
 502 /* GL locking is done by the caller. */
 503 static inline void shader_arb_vs_local_constants(IWineD3DDeviceImpl* deviceImpl)
 504 {
 505     IWineD3DStateBlockImpl* stateBlock;
 506     const struct wined3d_gl_info *gl_info = &deviceImpl->adapter->gl_info;
 507     unsigned char i;
 508     struct shader_arb_priv *priv = deviceImpl->shader_priv;
 509     const struct arb_vs_compiled_shader *gl_shader = priv->compiled_vprog;
 510
 511     /* Upload the position fixup */
 512     GL_EXTCALL(glProgramLocalParameter4fvARB(GL_VERTEX_PROGRAM_ARB, gl_shader->pos_fixup, deviceImpl->posFixup));
 513
 514     if(gl_shader->num_int_consts == 0) return;
 515
 516     stateBlock = deviceImpl->stateBlock;
 517
 518     for(i = 0; i < MAX_CONST_I; i++)
 519     {
 520         if(gl_shader->int_consts[i] != WINED3D_CONST_NUM_UNUSED)
 521         {
 522             float val[4];
 523             val[0] = stateBlock->vertexShaderConstantI[4 * i];
 524             val[1] = stateBlock->vertexShaderConstantI[4 * i + 1];
 525             val[2] = stateBlock->vertexShaderConstantI[4 * i + 2];
 526             val[3] = -1.0f;
 527
 528             GL_EXTCALL(glProgramLocalParameter4fvARB(GL_VERTEX_PROGRAM_ARB, gl_shader->int_consts[i], val));
 529         }
 530     }
 531     checkGLcall("Load vs int consts");
 532 }
 533
 534 /**
 535  * Loads the app-supplied constants into the currently set ARB_[vertex/fragment]_programs.
 536  *
 537  * We only support float constants in ARB at the moment, so don't
 538  * worry about the Integers or Booleans
 539  */
 540 /* GL locking is done by the caller (state handler) */
 541 static void shader_arb_load_constants(const struct wined3d_context *context, char usePixelShader, char useVertexShader)
 542 {
 543     IWineD3DDeviceImpl *device = ((IWineD3DSurfaceImpl *)context->surface)->resource.wineD3DDevice;
 544     IWineD3DStateBlockImpl* stateBlock = device->stateBlock;
 545     const struct wined3d_gl_info *gl_info = context->gl_info;
 546
 547     if (useVertexShader) {
 548         IWineD3DBaseShaderImpl* vshader = (IWineD3DBaseShaderImpl*) stateBlock->vertexShader;
 549
 550         /* Load DirectX 9 float constants for vertex shader */
 551         device->highest_dirty_vs_const = shader_arb_load_constantsF(vshader, gl_info, GL_VERTEX_PROGRAM_ARB,
 552                 device->highest_dirty_vs_const, stateBlock->vertexShaderConstantF, context->vshader_const_dirty);
 553         shader_arb_vs_local_constants(device);
 554     }
 555
 556     if (usePixelShader) {
 557         IWineD3DBaseShaderImpl* pshader = (IWineD3DBaseShaderImpl*) stateBlock->pixelShader;
 558
 559         /* Load DirectX 9 float constants for pixel shader */
 560         device->highest_dirty_ps_const = shader_arb_load_constantsF(pshader, gl_info, GL_FRAGMENT_PROGRAM_ARB,
 561                 device->highest_dirty_ps_const, stateBlock->pixelShaderConstantF, context->pshader_const_dirty);
 562         shader_arb_ps_local_constants(device);
 563     }
 564 }
 565
 566 static void shader_arb_update_float_vertex_constants(IWineD3DDevice *iface, UINT start, UINT count)
 567 {
 568     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
 569     struct wined3d_context *context = context_get_current();
 570
 571     /* We don't want shader constant dirtification to be an O(contexts), so just dirtify the active
 572      * context. On a context switch the old context will be fully dirtified */
 573     if (!context || ((IWineD3DSurfaceImpl *)context->surface)->resource.wineD3DDevice != This) return;
 574
 575     memset(context->vshader_const_dirty + start, 1, sizeof(*context->vshader_const_dirty) * count);
 576     This->highest_dirty_vs_const = max(This->highest_dirty_vs_const, start + count);
 577 }
 578
 579 static void shader_arb_update_float_pixel_constants(IWineD3DDevice *iface, UINT start, UINT count)
 580 {
 581     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
 582     struct wined3d_context *context = context_get_current();
 583
 584     /* We don't want shader constant dirtification to be an O(contexts), so just dirtify the active
 585      * context. On a context switch the old context will be fully dirtified */
 586     if (!context || ((IWineD3DSurfaceImpl *)context->surface)->resource.wineD3DDevice != This) return;
 587
 588     memset(context->pshader_const_dirty + start, 1, sizeof(*context->pshader_const_dirty) * count);
 589     This->highest_dirty_ps_const = max(This->highest_dirty_ps_const, start + count);
 590 }
 591
 592 static DWORD *local_const_mapping(IWineD3DBaseShaderImpl *This)
 593 {
 594     DWORD *ret;
 595     DWORD idx = 0;
 596     const local_constant *lconst;
 597
 598     if(This->baseShader.load_local_constsF || list_empty(&This->baseShader.constantsF)) return NULL;
 599
 600     ret = HeapAlloc(GetProcessHeap(), 0, sizeof(DWORD) * This->baseShader.limits.constant_float);
 601     if(!ret) {
 602         ERR("Out of memory\n");
 603         return NULL;
 604     }
 605
 606     LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry) {
 607         ret[lconst->idx] = idx++;
 608     }
 609     return ret;
 610 }
 611
 612 /* Generate the variable & register declarations for the ARB_vertex_program output target */
 613 static DWORD shader_generate_arb_declarations(IWineD3DBaseShader *iface, const shader_reg_maps *reg_maps,
 614         struct wined3d_shader_buffer *buffer, const struct wined3d_gl_info *gl_info, DWORD *lconst_map,
 615         DWORD *num_clipplanes, struct shader_arb_ctx_priv *ctx)
 616 {
 617     IWineD3DBaseShaderImpl* This = (IWineD3DBaseShaderImpl*) iface;
 618     DWORD i, next_local = 0;
 619     char pshader = shader_is_pshader_version(reg_maps->shader_version.type);
 620     unsigned max_constantsF;
 621     const local_constant *lconst;
 622     DWORD map;
 623
 624     /* In pixel shaders, all private constants are program local, we don't need anything
 625      * from program.env. Thus we can advertise the full set of constants in pixel shaders.
 626      * If we need a private constant the GL implementation will squeeze it in somewhere
 627      *
 628      * With vertex shaders we need the posFixup and on some GL implementations 4 helper
 629      * immediate values. The posFixup is loaded using program.env for now, so always
 630      * subtract one from the number of constants. If the shader uses indirect addressing,
 631      * account for the helper const too because we have to declare all availabke d3d constants
 632      * and don't know which are actually used.
 633      */
 634     if(pshader) {
 635         max_constantsF = GL_LIMITS(pshader_constantsF);
 636     } else {
 637         if(This->baseShader.reg_maps.usesrelconstF) {
 638             DWORD highest_constf = 0, clip_limit;
 639             max_constantsF = GL_LIMITS(vshader_constantsF) - reserved_vs_const(iface, gl_info);
 640             max_constantsF -= count_bits(This->baseShader.reg_maps.integer_constants);
 641
 642             for(i = 0; i < This->baseShader.limits.constant_float; i++)
 643             {
 644                 DWORD idx = i >> 5;
 645                 DWORD shift = i & 0x1f;
 646                 if(reg_maps->constf[idx] & (1 << shift)) highest_constf = i;
 647             }
 648
 649             clip_limit = GL_LIMITS(clipplanes);
 650             if(ctx->target_version == ARB) clip_limit = min(clip_limit, 4);
 651             *num_clipplanes = min(clip_limit, max_constantsF - highest_constf - 1);
 652             max_constantsF -= *num_clipplanes;
 653             if(*num_clipplanes < clip_limit)
 654             {
 655                 WARN("Only %u clipplanes out of %u enabled\n", *num_clipplanes, GL_LIMITS(clipplanes));
 656             }
 657         }
 658         else
 659         {
 660             if(ctx->target_version >= NV2) *num_clipplanes = GL_LIMITS(clipplanes);
 661             else *num_clipplanes = min(GL_LIMITS(clipplanes), 4);
 662             max_constantsF = GL_LIMITS(vshader_constantsF);
 663         }
 664     }
 665
 666     for (i = 0, map = reg_maps->temporary; map; map >>= 1, ++i)
 667     {
 668         if (map & 1) shader_addline(buffer, "TEMP R%u;\n", i);
 669     }
 670
 671     for (i = 0, map = reg_maps->address; map; map >>= 1, ++i)
 672     {
 673         if (map & 1) shader_addline(buffer, "ADDRESS A%u;\n", i);
 674     }
 675
 676     if (pshader && reg_maps->shader_version.major == 1 && reg_maps->shader_version.minor <= 3)
 677     {
 678         for (i = 0, map = reg_maps->texcoord; map; map >>= 1, ++i)
 679         {
 680             if (map & 1) shader_addline(buffer, "TEMP T%u;\n", i);
 681         }
 682     }
 683
 684     /* Load local constants using the program-local space,
 685      * this avoids reloading them each time the shader is used
 686      */
 687     if(lconst_map) {
 688         LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry) {
 689             shader_addline(buffer, "PARAM C%u = program.local[%u];\n", lconst->idx,
 690                            lconst_map[lconst->idx]);
 691             next_local = max(next_local, lconst_map[lconst->idx] + 1);
 692         }
 693     }
 694
 695     /* we use the array-based constants array if the local constants are marked for loading,
 696      * because then we use indirect addressing, or when the local constant list is empty,
 697      * because then we don't know if we're using indirect addressing or not. If we're hardcoding
 698      * local constants do not declare the loaded constants as an array because ARB compilers usually
 699      * do not optimize unused constants away
 700      */
 701     if(This->baseShader.reg_maps.usesrelconstF) {
 702         /* Need to PARAM the environment parameters (constants) so we can use relative addressing */
 703         shader_addline(buffer, "PARAM C[%d] = { program.env[0..%d] };\n",
 704                     max_constantsF, max_constantsF - 1);
 705     } else {
 706         for(i = 0; i < max_constantsF; i++) {
 707             DWORD idx, mask;
 708             idx = i >> 5;
 709             mask = 1 << (i & 0x1f);
 710             if(!shader_constant_is_local(This, i) && (This->baseShader.reg_maps.constf[idx] & mask)) {
 711                 shader_addline(buffer, "PARAM C%d = program.env[%d];\n",i, i);
 712             }
 713         }
 714     }
 715
 716     return next_local;
 717 }
 718
 719 static const char * const shift_tab[] = {
 720     "dummy",     /*  0 (none) */
 721     "coefmul.x", /*  1 (x2)   */
 722     "coefmul.y", /*  2 (x4)   */
 723     "coefmul.z", /*  3 (x8)   */
 724     "coefmul.w", /*  4 (x16)  */
 725     "dummy",     /*  5 (x32)  */
 726     "dummy",     /*  6 (x64)  */
 727     "dummy",     /*  7 (x128) */
 728     "dummy",     /*  8 (d256) */
 729     "dummy",     /*  9 (d128) */
 730     "dummy",     /* 10 (d64)  */
 731     "dummy",     /* 11 (d32)  */
 732     "coefdiv.w", /* 12 (d16)  */
 733     "coefdiv.z", /* 13 (d8)   */
 734     "coefdiv.y", /* 14 (d4)   */
 735     "coefdiv.x"  /* 15 (d2)   */
 736 };
 737
 738 static void shader_arb_get_write_mask(const struct wined3d_shader_instruction *ins,
 739         const struct wined3d_shader_dst_param *dst, char *write_mask)
 740 {
 741     char *ptr = write_mask;
 742
 743     if (dst->write_mask != WINED3DSP_WRITEMASK_ALL)
 744     {
 745         *ptr++ = '.';
 746         if (dst->write_mask & WINED3DSP_WRITEMASK_0) *ptr++ = 'x';
 747         if (dst->write_mask & WINED3DSP_WRITEMASK_1) *ptr++ = 'y';
 748         if (dst->write_mask & WINED3DSP_WRITEMASK_2) *ptr++ = 'z';
 749         if (dst->write_mask & WINED3DSP_WRITEMASK_3) *ptr++ = 'w';
 750     }
 751
 752     *ptr = '\0';
 753 }
 754
 755 static void shader_arb_get_swizzle(const struct wined3d_shader_src_param *param, BOOL fixup, char *swizzle_str)
 756 {
 757     /* For registers of type WINED3DDECLTYPE_D3DCOLOR, data is stored as "bgra",
 758      * but addressed as "rgba". To fix this we need to swap the register's x
 759      * and z components. */
 760     const char *swizzle_chars = fixup ? "zyxw" : "xyzw";
 761     char *ptr = swizzle_str;
 762
 763     /* swizzle bits fields: wwzzyyxx */
 764     DWORD swizzle = param->swizzle;
 765     DWORD swizzle_x = swizzle & 0x03;
 766     DWORD swizzle_y = (swizzle >> 2) & 0x03;
 767     DWORD swizzle_z = (swizzle >> 4) & 0x03;
 768     DWORD swizzle_w = (swizzle >> 6) & 0x03;
 769
 770     /* If the swizzle is the default swizzle (ie, "xyzw"), we don't need to
 771      * generate a swizzle string. Unless we need to our own swizzling. */
 772     if (swizzle != WINED3DSP_NOSWIZZLE || fixup)
 773     {
 774         *ptr++ = '.';
 775         if (swizzle_x == swizzle_y && swizzle_x == swizzle_z && swizzle_x == swizzle_w) {
 776             *ptr++ = swizzle_chars[swizzle_x];
 777         } else {
 778             *ptr++ = swizzle_chars[swizzle_x];
 779             *ptr++ = swizzle_chars[swizzle_y];
 780             *ptr++ = swizzle_chars[swizzle_z];
 781             *ptr++ = swizzle_chars[swizzle_w];
 782         }
 783     }
 784
 785     *ptr = '\0';
 786 }
 787
 788 static void shader_arb_request_a0(const struct wined3d_shader_instruction *ins, const char *src)
 789 {
 790     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
 791     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
 792
 793     if(strcmp(priv->addr_reg, src) == 0) return;
 794
 795     strcpy(priv->addr_reg, src);
 796     shader_addline(buffer, "ARL A0.x, %s;\n", src);
 797 }
 798
 799 static void shader_arb_get_src_param(const struct wined3d_shader_instruction *ins,
 800         const struct wined3d_shader_src_param *src, unsigned int tmpreg, char *outregstr);
 801
 802 static void shader_arb_get_register_name(const struct wined3d_shader_instruction *ins,
 803         const struct wined3d_shader_register *reg, char *register_name, BOOL *is_color)
 804 {
 805     /* oPos, oFog and oPts in D3D */
 806     static const char * const rastout_reg_names[] = {"TMP_OUT", "result.fogcoord", "result.pointsize"};
 807     IWineD3DBaseShaderImpl *This = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
 808     BOOL pshader = shader_is_pshader_version(This->baseShader.reg_maps.shader_version.type);
 809     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
 810
 811     *is_color = FALSE;
 812
 813     switch (reg->type)
 814     {
 815         case WINED3DSPR_TEMP:
 816             sprintf(register_name, "R%u", reg->idx);
 817             break;
 818
 819         case WINED3DSPR_INPUT:
 820             if (pshader)
 821             {
 822                 if(This->baseShader.reg_maps.shader_version.major < 3)
 823                 {
 824                     if (reg->idx == 0) strcpy(register_name, "fragment.color.primary");
 825                     else strcpy(register_name, "fragment.color.secondary");
 826                 }
 827                 else
 828                 {
 829                     if(reg->rel_addr)
 830                     {
 831                         char rel_reg[50];
 832                         shader_arb_get_src_param(ins, reg->rel_addr, 0, rel_reg);
 833
 834                         if(strcmp(rel_reg, "**aL_emul**") == 0)
 835                         {
 836                             DWORD idx = ctx->aL + reg->idx;
 837                             if(idx < MAX_REG_INPUT)
 838                             {
 839                                 strcpy(register_name, ctx->ps_input[idx]);
 840                             }
 841                             else
 842                             {
 843                                 ERR("Pixel shader input register out of bounds: %u\n", idx);
 844                                 sprintf(register_name, "out_of_bounds_%u", idx);
 845                             }
 846                         }
 847                         else if(This->baseShader.reg_maps.input_registers & 0x0300)
 848                         {
 849                             /* There are two ways basically:
 850                              *
 851                              * 1) Use the unrolling code that is used for loop emulation and unroll the loop.
 852                              *    That means trouble if the loop also contains a breakc or if the control values
 853                              *    aren't local constants.
 854                              * 2) Generate an if block that checks if aL.y < 8, == 8 or == 9 and selects the
 855                              *    source dynamically. The trouble is that we cannot simply read aL.y because it
 856                              *    is an ADDRESS register. We could however push it, load .zw with a value and use
 857                              *    ADAC to load the condition code register and pop it again afterwards
 858                              */
 859                             FIXME("Relative input register addressing with more than 8 registers\n");
 860
 861                             /* This is better than nothing for now */
 862                             sprintf(register_name, "fragment.texcoord[%s + %u]", rel_reg, reg->idx);
 863                         }
 864                         else if(ctx->cur_ps_args->super.vp_mode != vertexshader)
 865                         {
 866                             /* This is problematic because we'd have to consult the ctx->ps_input strings
 867                              * for where to find the varying. Some may be "0.0", others can be texcoords or
 868                              * colors. This needs either a pipeline replacement to make the vertex shader feed
 869                              * proper varyings, or loop unrolling
 870                              *
 871                              * For now use the texcoords and hope for the best
 872                              */
 873                             FIXME("Non-vertex shader varying input with indirect addressing\n");
 874                             sprintf(register_name, "fragment.texcoord[%s + %u]", rel_reg, reg->idx);
 875                         }
 876                         else
 877                         {
 878                             /* D3D supports indirect addressing only with aL in loop registers. The loop instruction
 879                              * pulls GL_NV_fragment_program2 in
 880                              */
 881                             sprintf(register_name, "fragment.texcoord[%s + %u]", rel_reg, reg->idx);
 882                         }
 883                     }
 884                     else
 885                     {
 886                         if(reg->idx < MAX_REG_INPUT)
 887                         {
 888                             strcpy(register_name, ctx->ps_input[reg->idx]);
 889                         }
 890                         else
 891                         {
 892                             ERR("Pixel shader input register out of bounds: %u\n", reg->idx);
 893                             sprintf(register_name, "out_of_bounds_%u", reg->idx);
 894                         }
 895                     }
 896                 }
 897             }
 898             else
 899             {
 900                 if (ctx->cur_vs_args->super.swizzle_map & (1 << reg->idx)) *is_color = TRUE;
 901                 sprintf(register_name, "vertex.attrib[%u]", reg->idx);
 902             }
 903             break;
 904
 905         case WINED3DSPR_CONST:
 906             if (!pshader && reg->rel_addr)
 907             {
 908                 BOOL aL = FALSE;
 909                 char rel_reg[50];
 910                 UINT rel_offset = ((IWineD3DVertexShaderImpl *)This)->rel_offset;
 911                 if(This->baseShader.reg_maps.shader_version.major < 2) {
 912                     sprintf(rel_reg, "A0.x");
 913                 } else {
 914                     shader_arb_get_src_param(ins, reg->rel_addr, 0, rel_reg);
 915                     if(ctx->target_version == ARB) {
 916                         if(strcmp(rel_reg, "**aL_emul**") == 0) {
 917                             aL = TRUE;
 918                         } else {
 919                             shader_arb_request_a0(ins, rel_reg);
 920                             sprintf(rel_reg, "A0.x");
 921                         }
 922                     }
 923                 }
 924                 if(aL)
 925                     sprintf(register_name, "C[%u]", ctx->aL + reg->idx);
 926                 else if (reg->idx >= rel_offset)
 927                     sprintf(register_name, "C[%s + %u]", rel_reg, reg->idx - rel_offset);
 928                 else
 929                     sprintf(register_name, "C[%s - %u]", rel_reg, -reg->idx + rel_offset);
 930             }
 931             else
 932             {
 933                 if (This->baseShader.reg_maps.usesrelconstF)
 934                     sprintf(register_name, "C[%u]", reg->idx);
 935                 else
 936                     sprintf(register_name, "C%u", reg->idx);
 937             }
 938             break;
 939
 940         case WINED3DSPR_TEXTURE: /* case WINED3DSPR_ADDR: */
 941             if (pshader) {
 942                 if(This->baseShader.reg_maps.shader_version.major == 1 &&
 943                    This->baseShader.reg_maps.shader_version.minor <= 3) {
 944                     /* In ps <= 1.3, Tx is a temporary register as destination to all instructions,
 945                      * and as source to most instructions. For some instructions it is the texcoord
 946                      * input. Those instructions know about the special use
 947                      */
 948                     sprintf(register_name, "T%u", reg->idx);
 949                 } else {
 950                     /* in ps 1.4 and 2.x Tx is always a (read-only) varying */
 951                     sprintf(register_name, "fragment.texcoord[%u]", reg->idx);
 952                 }
 953             }
 954             else
 955             {
 956                 if(This->baseShader.reg_maps.shader_version.major == 1 || ctx->target_version >= NV2)
 957                 {
 958                     sprintf(register_name, "A%u", reg->idx);
 959                 }
 960                 else
 961                 {
 962                     sprintf(register_name, "A%u_SHADOW", reg->idx);
 963                 }
 964             }
 965             break;
 966
 967         case WINED3DSPR_COLOROUT:
 968             if(ctx->cur_ps_args->super.srgb_correction && reg->idx == 0)
 969             {
 970                 strcpy(register_name, "TMP_COLOR");
 971             }
 972             else
 973             {
 974                 if(ctx->cur_ps_args->super.srgb_correction) FIXME("sRGB correction on higher render targets\n");
 975                 if(This->baseShader.reg_maps.highest_render_target > 0)
 976                 {
 977                     sprintf(register_name, "result.color[%u]", reg->idx);
 978                 }
 979                 else
 980                 {
 981                     strcpy(register_name, "result.color");
 982                 }
 983             }
 984             break;
 985
 986         case WINED3DSPR_RASTOUT:
 987             if(reg->idx == 1) sprintf(register_name, "%s", ctx->fog_output);
 988             else sprintf(register_name, "%s", rastout_reg_names[reg->idx]);
 989             break;
 990
 991         case WINED3DSPR_DEPTHOUT:
 992             strcpy(register_name, "result.depth");
 993             break;
 994
 995         case WINED3DSPR_ATTROUT:
 996         /* case WINED3DSPR_OUTPUT: */
 997             if (pshader) sprintf(register_name, "oD[%u]", reg->idx);
 998             else strcpy(register_name, ctx->color_output[reg->idx]);
 999             break;
1000
1001         case WINED3DSPR_TEXCRDOUT:
1002             if (pshader)
1003             {
1004                 sprintf(register_name, "oT[%u]", reg->idx);
1005             }
1006             else
1007             {
1008                 if(This->baseShader.reg_maps.shader_version.major < 3)
1009                 {
1010                     strcpy(register_name, ctx->texcrd_output[reg->idx]);
1011                 }
1012                 else
1013                 {
1014                     strcpy(register_name, ctx->vs_output[reg->idx]);
1015                 }
1016             }
1017             break;
1018
1019         case WINED3DSPR_LOOP:
1020             if(ctx->target_version >= NV2)
1021             {
1022                 /* Pshader has an implicitly declared loop index counter A0.x that cannot be renamed */
1023                 if(pshader) sprintf(register_name, "A0.x");
1024                 else sprintf(register_name, "aL.y");
1025             }
1026             else
1027             {
1028                 /* Unfortunately this code cannot return the value of ctx->aL here. An immediate value
1029                  * would be valid, but if aL is used for indexing(its only use), there's likely an offset,
1030                  * thus the result would be something like C[15 + 30], which is not valid in the ARB program
1031                  * grammar. So return a marker for the emulated aL and intercept it in constant and varying
1032                  * indexing
1033                  */
1034                 sprintf(register_name, "**aL_emul**");
1035             }
1036
1037             break;
1038
1039         case WINED3DSPR_CONSTINT:
1040             sprintf(register_name, "I%u", reg->idx);
1041             break;
1042
1043         case WINED3DSPR_MISCTYPE:
1044             if(reg->idx == 0)
1045             {
1046                 sprintf(register_name, "vpos");
1047             }
1048             else if(reg->idx == 1)
1049             {
1050                 sprintf(register_name, "fragment.facing.x");
1051             }
1052             else
1053             {
1054                 FIXME("Unknown MISCTYPE register index %u\n", reg->idx);
1055             }
1056             break;
1057
1058         default:
1059             FIXME("Unhandled register type %#x[%u]\n", reg->type, reg->idx);
1060             sprintf(register_name, "unrecognized_register[%u]", reg->idx);
1061             break;
1062     }
1063 }
1064
1065 static void shader_arb_get_dst_param(const struct wined3d_shader_instruction *ins,
1066         const struct wined3d_shader_dst_param *wined3d_dst, char *str)
1067 {
1068     char register_name[255];
1069     char write_mask[6];
1070     BOOL is_color;
1071
1072     shader_arb_get_register_name(ins, &wined3d_dst->reg, register_name, &is_color);
1073     strcpy(str, register_name);
1074
1075     shader_arb_get_write_mask(ins, wined3d_dst, write_mask);
1076     strcat(str, write_mask);
1077 }
1078
1079 static const char *shader_arb_get_fixup_swizzle(enum fixup_channel_source channel_source)
1080 {
1081     switch(channel_source)
1082     {
1083         case CHANNEL_SOURCE_ZERO: return "0";
1084         case CHANNEL_SOURCE_ONE: return "1";
1085         case CHANNEL_SOURCE_X: return "x";
1086         case CHANNEL_SOURCE_Y: return "y";
1087         case CHANNEL_SOURCE_Z: return "z";
1088         case CHANNEL_SOURCE_W: return "w";
1089         default:
1090             FIXME("Unhandled channel source %#x\n", channel_source);
1091             return "undefined";
1092     }
1093 }
1094
1095 static void gen_color_correction(struct wined3d_shader_buffer *buffer, const char *reg,
1096         DWORD dst_mask, const char *one, const char *two, struct color_fixup_desc fixup)
1097 {
1098     DWORD mask;
1099
1100     if (is_yuv_fixup(fixup))
1101     {
1102         enum yuv_fixup yuv_fixup = get_yuv_fixup(fixup);
1103         FIXME("YUV fixup (%#x) not supported\n", yuv_fixup);
1104         return;
1105     }
1106
1107     mask = 0;
1108     if (fixup.x_source != CHANNEL_SOURCE_X) mask |= WINED3DSP_WRITEMASK_0;
1109     if (fixup.y_source != CHANNEL_SOURCE_Y) mask |= WINED3DSP_WRITEMASK_1;
1110     if (fixup.z_source != CHANNEL_SOURCE_Z) mask |= WINED3DSP_WRITEMASK_2;
1111     if (fixup.w_source != CHANNEL_SOURCE_W) mask |= WINED3DSP_WRITEMASK_3;
1112     mask &= dst_mask;
1113
1114     if (mask)
1115     {
1116         shader_addline(buffer, "SWZ %s, %s, %s, %s, %s, %s;\n", reg, reg,
1117                 shader_arb_get_fixup_swizzle(fixup.x_source), shader_arb_get_fixup_swizzle(fixup.y_source),
1118                 shader_arb_get_fixup_swizzle(fixup.z_source), shader_arb_get_fixup_swizzle(fixup.w_source));
1119     }
1120
1121     mask = 0;
1122     if (fixup.x_sign_fixup) mask |= WINED3DSP_WRITEMASK_0;
1123     if (fixup.y_sign_fixup) mask |= WINED3DSP_WRITEMASK_1;
1124     if (fixup.z_sign_fixup) mask |= WINED3DSP_WRITEMASK_2;
1125     if (fixup.w_sign_fixup) mask |= WINED3DSP_WRITEMASK_3;
1126     mask &= dst_mask;
1127
1128     if (mask)
1129     {
1130         char reg_mask[6];
1131         char *ptr = reg_mask;
1132
1133         if (mask != WINED3DSP_WRITEMASK_ALL)
1134         {
1135             *ptr++ = '.';
1136             if (mask & WINED3DSP_WRITEMASK_0) *ptr++ = 'x';
1137             if (mask & WINED3DSP_WRITEMASK_1) *ptr++ = 'y';
1138             if (mask & WINED3DSP_WRITEMASK_2) *ptr++ = 'z';
1139             if (mask & WINED3DSP_WRITEMASK_3) *ptr++ = 'w';
1140         }
1141         *ptr = '\0';
1142
1143         shader_addline(buffer, "MAD %s%s, %s, %s, -%s;\n", reg, reg_mask, reg, two, one);
1144     }
1145 }
1146
1147 static const char *shader_arb_get_modifier(const struct wined3d_shader_instruction *ins)
1148 {
1149     DWORD mod;
1150     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
1151     if (!ins->dst_count) return "";
1152
1153     mod = ins->dst[0].modifiers;
1154
1155     /* Silently ignore PARTIALPRECISION if its not supported */
1156     if(priv->target_version == ARB) mod &= ~WINED3DSPDM_PARTIALPRECISION;
1157
1158     if(mod & WINED3DSPDM_MSAMPCENTROID)
1159     {
1160         FIXME("Unhandled modifier WINED3DSPDM_MSAMPCENTROID\n");
1161         mod &= ~WINED3DSPDM_MSAMPCENTROID;
1162     }
1163
1164     switch(mod)
1165     {
1166         case WINED3DSPDM_SATURATE | WINED3DSPDM_PARTIALPRECISION:
1167             return "H_SAT";
1168
1169         case WINED3DSPDM_SATURATE:
1170             return "_SAT";
1171
1172         case WINED3DSPDM_PARTIALPRECISION:
1173             return "H";
1174
1175         case 0:
1176             return "";
1177
1178         default:
1179             FIXME("Unknown modifiers 0x%08x\n", mod);
1180             return "";
1181     }
1182 }
1183
1184 #define TEX_PROJ        0x1
1185 #define TEX_BIAS        0x2
1186 #define TEX_LOD         0x4
1187 #define TEX_DERIV       0x10
1188
1189 static void shader_hw_sample(const struct wined3d_shader_instruction *ins, DWORD sampler_idx,
1190         const char *dst_str, const char *coord_reg, WORD flags, const char *dsx, const char *dsy)
1191 {
1192     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1193     DWORD sampler_type = ins->ctx->reg_maps->sampler_type[sampler_idx];
1194     const char *tex_type;
1195     BOOL np2_fixup = FALSE;
1196     IWineD3DBaseShaderImpl *This = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
1197     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *) This->baseShader.device;
1198     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
1199     const char *mod;
1200     BOOL pshader = shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type);
1201
1202     /* D3D vertex shader sampler IDs are vertex samplers(0-3), not global d3d samplers */
1203     if(!pshader) sampler_idx += MAX_FRAGMENT_SAMPLERS;
1204
1205     switch(sampler_type) {
1206         case WINED3DSTT_1D:
1207             tex_type = "1D";
1208             break;
1209
1210         case WINED3DSTT_2D:
1211             if(device->stateBlock->textures[sampler_idx] &&
1212                IWineD3DBaseTexture_GetTextureDimensions(device->stateBlock->textures[sampler_idx]) == GL_TEXTURE_RECTANGLE_ARB) {
1213                 tex_type = "RECT";
1214             } else {
1215                 tex_type = "2D";
1216             }
1217             if (shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type))
1218             {
1219                 if (priv->cur_np2fixup_info->super.active & (1 << sampler_idx))
1220                 {
1221                     if (flags) FIXME("Only ordinary sampling from NP2 textures is supported.\n");
1222                     else np2_fixup = TRUE;
1223                 }
1224             }
1225             break;
1226
1227         case WINED3DSTT_VOLUME:
1228             tex_type = "3D";
1229             break;
1230
1231         case WINED3DSTT_CUBE:
1232             tex_type = "CUBE";
1233             break;
1234
1235         default:
1236             ERR("Unexpected texture type %d\n", sampler_type);
1237             tex_type = "";
1238     }
1239
1240     /* TEX, TXL, TXD and TXP do not support the "H" modifier,
1241      * so don't use shader_arb_get_modifier
1242      */
1243     if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE) mod = "_SAT";
1244     else mod = "";
1245
1246     /* Fragment samplers always have indentity mapping */
1247     if(sampler_idx >= MAX_FRAGMENT_SAMPLERS)
1248     {
1249         sampler_idx = priv->cur_vs_args->vertex_samplers[sampler_idx - MAX_FRAGMENT_SAMPLERS];
1250     }
1251
1252     if (flags & TEX_DERIV)
1253     {
1254         if(flags & TEX_PROJ) FIXME("Projected texture sampling with custom derivatives\n");
1255         if(flags & TEX_BIAS) FIXME("Biased texture sampling with custom derivatives\n");
1256         shader_addline(buffer, "TXD%s %s, %s, %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg,
1257                        dsx, dsy,sampler_idx, tex_type);
1258     }
1259     else if(flags & TEX_LOD)
1260     {
1261         if(flags & TEX_PROJ) FIXME("Projected texture sampling with explicit lod\n");
1262         if(flags & TEX_BIAS) FIXME("Biased texture sampling with explicit lod\n");
1263         shader_addline(buffer, "TXL%s %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg,
1264                        sampler_idx, tex_type);
1265     }
1266     else if (flags & TEX_BIAS)
1267     {
1268         /* Shouldn't be possible, but let's check for it */
1269         if(flags & TEX_PROJ) FIXME("Biased and Projected texture sampling\n");
1270         /* TXB takes the 4th component of the source vector automatically, as d3d. Nothing more to do */
1271         shader_addline(buffer, "TXB%s %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg, sampler_idx, tex_type);
1272     }
1273     else if (flags & TEX_PROJ)
1274     {
1275         shader_addline(buffer, "TXP%s %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg, sampler_idx, tex_type);
1276     }
1277     else
1278     {
1279         if (np2_fixup)
1280         {
1281             const unsigned char idx = priv->cur_np2fixup_info->super.idx[sampler_idx];
1282             shader_addline(buffer, "MUL TA, np2fixup[%u].%s, %s;\n", idx >> 1,
1283                            (idx % 2) ? "zwxy" : "xyzw", coord_reg);
1284
1285             shader_addline(buffer, "TEX%s %s, TA, texture[%u], %s;\n", mod, dst_str, sampler_idx, tex_type);
1286         }
1287         else
1288             shader_addline(buffer, "TEX%s %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg, sampler_idx, tex_type);
1289     }
1290
1291     if (pshader)
1292     {
1293         gen_color_correction(buffer, dst_str, ins->dst[0].write_mask,
1294                 "one", "coefmul.x", priv->cur_ps_args->super.color_fixup[sampler_idx]);
1295     }
1296 }
1297
1298 static void shader_arb_get_src_param(const struct wined3d_shader_instruction *ins,
1299         const struct wined3d_shader_src_param *src, unsigned int tmpreg, char *outregstr)
1300 {
1301     /* Generate a line that does the input modifier computation and return the input register to use */
1302     BOOL is_color = FALSE;
1303     char regstr[256];
1304     char swzstr[20];
1305     int insert_line;
1306     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1307     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
1308
1309     /* Assume a new line will be added */
1310     insert_line = 1;
1311
1312     /* Get register name */
1313     shader_arb_get_register_name(ins, &src->reg, regstr, &is_color);
1314     shader_arb_get_swizzle(src, is_color, swzstr);
1315
1316     switch (src->modifiers)
1317     {
1318     case WINED3DSPSM_NONE:
1319         sprintf(outregstr, "%s%s", regstr, swzstr);
1320         insert_line = 0;
1321         break;
1322     case WINED3DSPSM_NEG:
1323         sprintf(outregstr, "-%s%s", regstr, swzstr);
1324         insert_line = 0;
1325         break;
1326     case WINED3DSPSM_BIAS:
1327         shader_addline(buffer, "ADD T%c, %s, -coefdiv.x;\n", 'A' + tmpreg, regstr);
1328         break;
1329     case WINED3DSPSM_BIASNEG:
1330         shader_addline(buffer, "ADD T%c, -%s, coefdiv.x;\n", 'A' + tmpreg, regstr);
1331         break;
1332     case WINED3DSPSM_SIGN:
1333         shader_addline(buffer, "MAD T%c, %s, coefmul.x, -one.x;\n", 'A' + tmpreg, regstr);
1334         break;
1335     case WINED3DSPSM_SIGNNEG:
1336         shader_addline(buffer, "MAD T%c, %s, -coefmul.x, one.x;\n", 'A' + tmpreg, regstr);
1337         break;
1338     case WINED3DSPSM_COMP:
1339         shader_addline(buffer, "SUB T%c, one.x, %s;\n", 'A' + tmpreg, regstr);
1340         break;
1341     case WINED3DSPSM_X2:
1342         shader_addline(buffer, "ADD T%c, %s, %s;\n", 'A' + tmpreg, regstr, regstr);
1343         break;
1344     case WINED3DSPSM_X2NEG:
1345         shader_addline(buffer, "ADD T%c, -%s, -%s;\n", 'A' + tmpreg, regstr, regstr);
1346         break;
1347     case WINED3DSPSM_DZ:
1348         shader_addline(buffer, "RCP T%c, %s.z;\n", 'A' + tmpreg, regstr);
1349         shader_addline(buffer, "MUL T%c, %s, T%c;\n", 'A' + tmpreg, regstr, 'A' + tmpreg);
1350         break;
1351     case WINED3DSPSM_DW:
1352         shader_addline(buffer, "RCP T%c, %s.w;\n", 'A' + tmpreg, regstr);
1353         shader_addline(buffer, "MUL T%c, %s, T%c;\n", 'A' + tmpreg, regstr, 'A' + tmpreg);
1354         break;
1355     case WINED3DSPSM_ABS:
1356         if(ctx->target_version >= NV2) {
1357             sprintf(outregstr, "|%s%s|", regstr, swzstr);
1358             insert_line = 0;
1359         } else {
1360             shader_addline(buffer, "ABS T%c, %s;\n", 'A' + tmpreg, regstr);
1361         }
1362         break;
1363     case WINED3DSPSM_ABSNEG:
1364         if(ctx->target_version >= NV2) {
1365             sprintf(outregstr, "-|%s%s|", regstr, swzstr);
1366         } else {
1367             shader_addline(buffer, "ABS T%c, %s;\n", 'A' + tmpreg, regstr);
1368             sprintf(outregstr, "-T%c%s", 'A' + tmpreg, swzstr);
1369         }
1370         insert_line = 0;
1371         break;
1372     default:
1373         sprintf(outregstr, "%s%s", regstr, swzstr);
1374         insert_line = 0;
1375     }
1376
1377     /* Return modified or original register, with swizzle */
1378     if (insert_line)
1379         sprintf(outregstr, "T%c%s", 'A' + tmpreg, swzstr);
1380 }
1381
1382 static void pshader_hw_bem(const struct wined3d_shader_instruction *ins)
1383 {
1384     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1385     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1386     char dst_name[50];
1387     char src_name[2][50];
1388     DWORD sampler_code = dst->reg.idx;
1389
1390     shader_arb_get_dst_param(ins, dst, dst_name);
1391
1392     /* Sampling the perturbation map in Tsrc was done already, including the signedness correction if needed
1393      *
1394      * Keep in mind that src_name[1] can be "TB" and src_name[0] can be "TA" because modifiers like _x2 are valid
1395      * with bem. So delay loading the first parameter until after the perturbation calculation which needs two
1396      * temps is done.
1397      */
1398     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1399     shader_addline(buffer, "SWZ TA, bumpenvmat%d, x, z, 0, 0;\n", sampler_code);
1400     shader_addline(buffer, "DP3 TC.r, TA, %s;\n", src_name[1]);
1401     shader_addline(buffer, "SWZ TA, bumpenvmat%d, y, w, 0, 0;\n", sampler_code);
1402     shader_addline(buffer, "DP3 TC.g, TA, %s;\n", src_name[1]);
1403
1404     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
1405     shader_addline(buffer, "ADD %s, %s, TC;\n", dst_name, src_name[0]);
1406 }
1407
1408 static DWORD negate_modifiers(DWORD mod, char *extra_char)
1409 {
1410     *extra_char = ' ';
1411     switch(mod)
1412     {
1413         case WINED3DSPSM_NONE:      return WINED3DSPSM_NEG;
1414         case WINED3DSPSM_NEG:       return WINED3DSPSM_NONE;
1415         case WINED3DSPSM_BIAS:      return WINED3DSPSM_BIASNEG;
1416         case WINED3DSPSM_BIASNEG:   return WINED3DSPSM_BIAS;
1417         case WINED3DSPSM_SIGN:      return WINED3DSPSM_SIGNNEG;
1418         case WINED3DSPSM_SIGNNEG:   return WINED3DSPSM_SIGN;
1419         case WINED3DSPSM_COMP:      *extra_char = '-'; return WINED3DSPSM_COMP;
1420         case WINED3DSPSM_X2:        return WINED3DSPSM_X2NEG;
1421         case WINED3DSPSM_X2NEG:     return WINED3DSPSM_X2;
1422         case WINED3DSPSM_DZ:        *extra_char = '-'; return WINED3DSPSM_DZ;
1423         case WINED3DSPSM_DW:        *extra_char = '-'; return WINED3DSPSM_DW;
1424         case WINED3DSPSM_ABS:       return WINED3DSPSM_ABSNEG;
1425         case WINED3DSPSM_ABSNEG:    return WINED3DSPSM_ABS;
1426     }
1427     FIXME("Unknown modifier %u\n", mod);
1428     return mod;
1429 }
1430
1431 static void pshader_hw_cnd(const struct wined3d_shader_instruction *ins)
1432 {
1433     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1434     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1435     char dst_name[50];
1436     char src_name[3][50];
1437     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
1438             ins->ctx->reg_maps->shader_version.minor);
1439     BOOL is_color;
1440
1441     shader_arb_get_dst_param(ins, dst, dst_name);
1442     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1443
1444     /* The coissue flag changes the semantic of the cnd instruction in <= 1.3 shaders */
1445     if (shader_version <= WINED3D_SHADER_VERSION(1, 3) && ins->coissue)
1446     {
1447         shader_addline(buffer, "MOV%s %s, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name[1]);
1448     } else {
1449         struct wined3d_shader_src_param src0_copy = ins->src[0];
1450         char extra_neg;
1451
1452         /* src0 may have a negate srcmod set, so we can't blindly add "-" to the name */
1453         src0_copy.modifiers = negate_modifiers(src0_copy.modifiers, &extra_neg);
1454
1455         shader_arb_get_src_param(ins, &src0_copy, 0, src_name[0]);
1456         shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
1457         shader_addline(buffer, "ADD TA, %c%s, coefdiv.x;\n", extra_neg, src_name[0]);
1458         /* No modifiers supported on CMP */
1459         shader_addline(buffer, "CMP %s, TA, %s, %s;\n", dst_name, src_name[1], src_name[2]);
1460
1461         /* _SAT on CMP doesn't make much sense, but it is not a pure NOP */
1462         if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE)
1463         {
1464             shader_arb_get_register_name(ins, &dst->reg, src_name[0], &is_color);
1465             shader_addline(buffer, "MOV_SAT %s, %s;\n", dst_name, dst_name);
1466         }
1467     }
1468 }
1469
1470 static void pshader_hw_cmp(const struct wined3d_shader_instruction *ins)
1471 {
1472     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1473     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1474     char dst_name[50];
1475     char src_name[3][50];
1476     BOOL is_color;
1477
1478     shader_arb_get_dst_param(ins, dst, dst_name);
1479
1480     /* Generate input register names (with modifiers) */
1481     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
1482     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1483     shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
1484
1485     /* No modifiers are supported on CMP */
1486     shader_addline(buffer, "CMP %s, %s, %s, %s;\n", dst_name,
1487                    src_name[0], src_name[2], src_name[1]);
1488
1489     if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE)
1490     {
1491         shader_arb_get_register_name(ins, &dst->reg, src_name[0], &is_color);
1492         shader_addline(buffer, "MOV_SAT %s, %s;\n", dst_name, src_name[0]);
1493     }
1494 }
1495
1496 /** Process the WINED3DSIO_DP2ADD instruction in ARB.
1497  * dst = dot2(src0, src1) + src2 */
1498 static void pshader_hw_dp2add(const struct wined3d_shader_instruction *ins)
1499 {
1500     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1501     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1502     char dst_name[50];
1503     char src_name[3][50];
1504     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
1505
1506     shader_arb_get_dst_param(ins, dst, dst_name);
1507     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
1508     shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
1509
1510     if(ctx->target_version >= NV3)
1511     {
1512         /* GL_NV_fragment_program2 has a 1:1 matching instruction */
1513         shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1514         shader_addline(buffer, "DP2A%s %s, %s, %s, %s;\n", shader_arb_get_modifier(ins),
1515                        dst_name, src_name[0], src_name[1], src_name[2]);
1516     }
1517     else if(ctx->target_version >= NV2)
1518     {
1519         /* dst.x = src2.?, src0.x, src1.x + src0.y * src1.y
1520          * dst.y = src2.?, src0.x, src1.z + src0.y * src1.w
1521          * dst.z = src2.?, src0.x, src1.x + src0.y * src1.y
1522          * dst.z = src2.?, src0.x, src1.z + src0.y * src1.w
1523          *
1524          * Make sure that src1.zw = src1.xy, then we get a classic dp2add
1525          *
1526          * .xyxy and other swizzles that we could get with this are not valid in
1527          * plain ARBfp, but luckily the NV extension grammar lifts this limitation.
1528          */
1529         struct wined3d_shader_src_param tmp_param = ins->src[1];
1530         DWORD swizzle = tmp_param.swizzle & 0xf; /* Selects .xy */
1531         tmp_param.swizzle = swizzle | (swizzle << 4); /* Creates .xyxy */
1532
1533         shader_arb_get_src_param(ins, &tmp_param, 1, src_name[1]);
1534
1535         shader_addline(buffer, "X2D%s %s, %s, %s, %s;\n", shader_arb_get_modifier(ins),
1536                        dst_name, src_name[2], src_name[0], src_name[1]);
1537     }
1538     else
1539     {
1540         shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1541         /* 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]
1542         * src_name[0] can be TA, but TA is a private temp for modifiers, so it is save to overwrite
1543         */
1544         shader_addline(buffer, "MOV TA, %s;\n", src_name[0]);
1545         shader_addline(buffer, "MOV TA.z, 0.0;\n");
1546         shader_addline(buffer, "DP3 TA, TA, %s;\n", src_name[1]);
1547         shader_addline(buffer, "ADD%s %s, TA, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name[2]);
1548     }
1549 }
1550
1551 /* Map the opcode 1-to-1 to the GL code */
1552 static void shader_hw_map2gl(const struct wined3d_shader_instruction *ins)
1553 {
1554     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1555     const char *instruction;
1556     char arguments[256], dst_str[50];
1557     unsigned int i;
1558     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1559
1560     switch (ins->handler_idx)
1561     {
1562         case WINED3DSIH_ABS: instruction = "ABS"; break;
1563         case WINED3DSIH_ADD: instruction = "ADD"; break;
1564         case WINED3DSIH_CRS: instruction = "XPD"; break;
1565         case WINED3DSIH_DP3: instruction = "DP3"; break;
1566         case WINED3DSIH_DP4: instruction = "DP4"; break;
1567         case WINED3DSIH_DST: instruction = "DST"; break;
1568         case WINED3DSIH_FRC: instruction = "FRC"; break;
1569         case WINED3DSIH_LIT: instruction = "LIT"; break;
1570         case WINED3DSIH_LRP: instruction = "LRP"; break;
1571         case WINED3DSIH_MAD: instruction = "MAD"; break;
1572         case WINED3DSIH_MAX: instruction = "MAX"; break;
1573         case WINED3DSIH_MIN: instruction = "MIN"; break;
1574         case WINED3DSIH_MOV: instruction = "MOV"; break;
1575         case WINED3DSIH_MUL: instruction = "MUL"; break;
1576         case WINED3DSIH_SGE: instruction = "SGE"; break;
1577         case WINED3DSIH_SLT: instruction = "SLT"; break;
1578         case WINED3DSIH_SUB: instruction = "SUB"; break;
1579         case WINED3DSIH_MOVA:instruction = "ARR"; break;
1580         case WINED3DSIH_SGN: instruction = "SSG"; break;
1581         case WINED3DSIH_DSX: instruction = "DDX"; break;
1582         default: instruction = "";
1583             FIXME("Unhandled opcode %#x\n", ins->handler_idx);
1584             break;
1585     }
1586
1587     /* Note that shader_arb_add_dst_param() adds spaces. */
1588     arguments[0] = '\0';
1589     shader_arb_get_dst_param(ins, dst, dst_str);
1590     for (i = 0; i < ins->src_count; ++i)
1591     {
1592         char operand[100];
1593         strcat(arguments, ", ");
1594         shader_arb_get_src_param(ins, &ins->src[i], i, operand);
1595         strcat(arguments, operand);
1596     }
1597     shader_addline(buffer, "%s%s %s%s;\n", instruction, shader_arb_get_modifier(ins), dst_str, arguments);
1598 }
1599
1600 static void shader_hw_nop(const struct wined3d_shader_instruction *ins)
1601 {
1602     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1603     shader_addline(buffer, "NOP;\n");
1604 }
1605
1606 static void shader_hw_mov(const struct wined3d_shader_instruction *ins)
1607 {
1608     IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
1609     BOOL pshader = shader_is_pshader_version(shader->baseShader.reg_maps.shader_version.type);
1610     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
1611
1612     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1613     char src0_param[256];
1614
1615     if(ins->handler_idx == WINED3DSIH_MOVA) {
1616         char write_mask[6];
1617
1618         if(ctx->target_version >= NV2) {
1619             shader_hw_map2gl(ins);
1620             return;
1621         }
1622         shader_arb_get_src_param(ins, &ins->src[0], 0, src0_param);
1623         shader_arb_get_write_mask(ins, &ins->dst[0], write_mask);
1624
1625         /* This implements the mova formula used in GLSL. The first two instructions
1626          * prepare the sign() part. Note that it is fine to have my_sign(0.0) = 1.0
1627          * in this case:
1628          * mova A0.x, 0.0
1629          *
1630          * A0.x = arl(floor(abs(0.0) + 0.5) * 1.0) = floor(0.5) = 0.0 since arl does a floor
1631          *
1632          * The ARL is performed when A0 is used - the requested component is read from A0_SHADOW into
1633          * A0.x. We can use the overwritten component of A0_shadow as temporary storage for the sign.
1634          */
1635         shader_addline(buffer, "SGE A0_SHADOW%s, %s, mova_const.y;\n", write_mask, src0_param);
1636         shader_addline(buffer, "MAD A0_SHADOW%s, A0_SHADOW, mova_const.z, -mova_const.w;\n", write_mask);
1637
1638         shader_addline(buffer, "ABS TA%s, %s;\n", write_mask, src0_param);
1639         shader_addline(buffer, "ADD TA%s, TA, mova_const.x;\n", write_mask);
1640         shader_addline(buffer, "FLR TA%s, TA;\n", write_mask);
1641         if (((IWineD3DVertexShaderImpl *)shader)->rel_offset)
1642         {
1643             shader_addline(buffer, "ADD TA%s, TA, helper_const.z;\n", write_mask);
1644         }
1645         shader_addline(buffer, "MUL A0_SHADOW%s, TA, A0_SHADOW;\n", write_mask);
1646
1647         ((struct shader_arb_ctx_priv *)ins->ctx->backend_data)->addr_reg[0] = '\0';
1648     } else if (ins->ctx->reg_maps->shader_version.major == 1
1649           && !shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type)
1650           && ins->dst[0].reg.type == WINED3DSPR_ADDR)
1651     {
1652         src0_param[0] = '\0';
1653         if (((IWineD3DVertexShaderImpl *)shader)->rel_offset)
1654         {
1655             shader_arb_get_src_param(ins, &ins->src[0], 0, src0_param);
1656             shader_addline(buffer, "ADD TA.x, %s, helper_const.z;\n", src0_param);
1657             shader_addline(buffer, "ARL A0.x, TA.x;\n");
1658         }
1659         else
1660         {
1661             /* Apple's ARB_vertex_program implementation does not accept an ARL source argument
1662              * with more than one component. Thus replicate the first source argument over all
1663              * 4 components. For example, .xyzw -> .x (or better: .xxxx), .zwxy -> .z, etc) */
1664             struct wined3d_shader_src_param tmp_src = ins->src[0];
1665             tmp_src.swizzle = (tmp_src.swizzle & 0x3) * 0x55;
1666             shader_arb_get_src_param(ins, &tmp_src, 0, src0_param);
1667             shader_addline(buffer, "ARL A0.x, %s;\n", src0_param);
1668         }
1669     }
1670     else if(ins->dst[0].reg.type == WINED3DSPR_COLOROUT && ins->dst[0].reg.idx == 0 && pshader)
1671     {
1672         IWineD3DPixelShaderImpl *ps = (IWineD3DPixelShaderImpl *) shader;
1673         if(ctx->cur_ps_args->super.srgb_correction && ps->color0_mov)
1674         {
1675             shader_addline(buffer, "#mov handled in srgb write code\n");
1676             return;
1677         }
1678         shader_hw_map2gl(ins);
1679     }
1680     else
1681     {
1682         shader_hw_map2gl(ins);
1683     }
1684 }
1685
1686 static void pshader_hw_texkill(const struct wined3d_shader_instruction *ins)
1687 {
1688     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1689     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1690     char reg_dest[40];
1691
1692     /* No swizzles are allowed in d3d's texkill. PS 1.x ignores the 4th component as documented,
1693      * but >= 2.0 honors it(undocumented, but tested by the d3d9 testsuit)
1694      */
1695     shader_arb_get_dst_param(ins, dst, reg_dest);
1696
1697     if (ins->ctx->reg_maps->shader_version.major >= 2)
1698     {
1699         const char *kilsrc = "TA";
1700         BOOL is_color;
1701
1702         shader_arb_get_register_name(ins, &dst->reg, reg_dest, &is_color);
1703         if(dst->write_mask == WINED3DSP_WRITEMASK_ALL)
1704         {
1705             kilsrc = reg_dest;
1706         }
1707         else
1708         {
1709             /* Sigh. KIL doesn't support swizzles/writemasks. KIL passes a writemask, but ".xy" for example
1710              * is not valid as a swizzle in ARB (needs ".xyyy"). Use SWZ to load the register properly, and set
1711              * masked out components to 0(won't kill)
1712              */
1713             char x = '0', y = '0', z = '0', w = '0';
1714             if(dst->write_mask & WINED3DSP_WRITEMASK_0) x = 'x';
1715             if(dst->write_mask & WINED3DSP_WRITEMASK_1) y = 'y';
1716             if(dst->write_mask & WINED3DSP_WRITEMASK_2) z = 'z';
1717             if(dst->write_mask & WINED3DSP_WRITEMASK_3) w = 'w';
1718             shader_addline(buffer, "SWZ TA, %s, %c, %c, %c, %c;\n", reg_dest, x, y, z, w);
1719         }
1720         shader_addline(buffer, "KIL %s;\n", kilsrc);
1721     } else {
1722         /* ARB fp doesn't like swizzles on the parameter of the KIL instruction. To mask the 4th component,
1723          * copy the register into our general purpose TMP variable, overwrite .w and pass TMP to KIL
1724          *
1725          * ps_1_3 shaders use the texcoord incarnation of the Tx register. ps_1_4 shaders can use the same,
1726          * or pass in any temporary register(in shader phase 2)
1727          */
1728         if(ins->ctx->reg_maps->shader_version.minor <= 3) {
1729             sprintf(reg_dest, "fragment.texcoord[%u]", dst->reg.idx);
1730         } else {
1731             shader_arb_get_dst_param(ins, dst, reg_dest);
1732         }
1733         shader_addline(buffer, "SWZ TA, %s, x, y, z, 1;\n", reg_dest);
1734         shader_addline(buffer, "KIL TA;\n");
1735     }
1736 }
1737
1738 static void pshader_hw_tex(const struct wined3d_shader_instruction *ins)
1739 {
1740     IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)ins->ctx->shader;
1741     IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
1742     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1743     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
1744             ins->ctx->reg_maps->shader_version.minor);
1745     struct wined3d_shader_src_param src;
1746
1747     char reg_dest[40];
1748     char reg_coord[40];
1749     DWORD reg_sampler_code;
1750     DWORD myflags = 0;
1751
1752     /* All versions have a destination register */
1753     shader_arb_get_dst_param(ins, dst, reg_dest);
1754
1755     /* 1.0-1.4: Use destination register number as texture code.
1756        2.0+: Use provided sampler number as texure code. */
1757     if (shader_version < WINED3D_SHADER_VERSION(2,0))
1758         reg_sampler_code = dst->reg.idx;
1759     else
1760         reg_sampler_code = ins->src[1].reg.idx;
1761
1762     /* 1.0-1.3: Use the texcoord varying.
1763        1.4+: Use provided coordinate source register. */
1764     if (shader_version < WINED3D_SHADER_VERSION(1,4))
1765         sprintf(reg_coord, "fragment.texcoord[%u]", reg_sampler_code);
1766     else {
1767         /* TEX is the only instruction that can handle DW and DZ natively */
1768         src = ins->src[0];
1769         if(src.modifiers == WINED3DSPSM_DW) src.modifiers = WINED3DSPSM_NONE;
1770         if(src.modifiers == WINED3DSPSM_DZ) src.modifiers = WINED3DSPSM_NONE;
1771         shader_arb_get_src_param(ins, &src, 0, reg_coord);
1772     }
1773
1774     /* projection flag:
1775      * 1.1, 1.2, 1.3: Use WINED3DTSS_TEXTURETRANSFORMFLAGS
1776      * 1.4: Use WINED3DSPSM_DZ or WINED3DSPSM_DW on src[0]
1777      * 2.0+: Use WINED3DSI_TEXLD_PROJECT on the opcode
1778      */
1779     if (shader_version < WINED3D_SHADER_VERSION(1,4))
1780     {
1781         DWORD flags = 0;
1782         if(reg_sampler_code < MAX_TEXTURES) {
1783             flags = deviceImpl->stateBlock->textureState[reg_sampler_code][WINED3DTSS_TEXTURETRANSFORMFLAGS];
1784         }
1785         if (flags & WINED3DTTFF_PROJECTED) {
1786             myflags |= TEX_PROJ;
1787         }
1788     }
1789     else if (shader_version < WINED3D_SHADER_VERSION(2,0))
1790     {
1791         DWORD src_mod = ins->src[0].modifiers;
1792         if (src_mod == WINED3DSPSM_DZ) {
1793             /* TXP cannot handle DZ natively, so move the z coordinate to .w. reg_coord is a read-only
1794              * varying register, so we need a temp reg
1795              */
1796             shader_addline(ins->ctx->buffer, "SWZ TA, %s, x, y, z, z;\n", reg_coord);
1797             strcpy(reg_coord, "TA");
1798             myflags |= TEX_PROJ;
1799         } else if(src_mod == WINED3DSPSM_DW) {
1800             myflags |= TEX_PROJ;
1801         }
1802     } else {
1803         if (ins->flags & WINED3DSI_TEXLD_PROJECT) myflags |= TEX_PROJ;
1804         if (ins->flags & WINED3DSI_TEXLD_BIAS) myflags |= TEX_BIAS;
1805     }
1806     shader_hw_sample(ins, reg_sampler_code, reg_dest, reg_coord, myflags, NULL, NULL);
1807 }
1808
1809 static void pshader_hw_texcoord(const struct wined3d_shader_instruction *ins)
1810 {
1811     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1812     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1813     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
1814             ins->ctx->reg_maps->shader_version.minor);
1815     char dst_str[50];
1816
1817     if (shader_version < WINED3D_SHADER_VERSION(1,4))
1818     {
1819         DWORD reg = dst->reg.idx;
1820
1821         shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
1822         shader_addline(buffer, "MOV_SAT %s, fragment.texcoord[%u];\n", dst_str, reg);
1823     } else {
1824         char reg_src[40];
1825
1826         shader_arb_get_src_param(ins, &ins->src[0], 0, reg_src);
1827         shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
1828         shader_addline(buffer, "MOV %s, %s;\n", dst_str, reg_src);
1829    }
1830 }
1831
1832 static void pshader_hw_texreg2ar(const struct wined3d_shader_instruction *ins)
1833 {
1834      struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1835      IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)ins->ctx->shader;
1836      IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
1837      DWORD flags;
1838
1839      DWORD reg1 = ins->dst[0].reg.idx;
1840      char dst_str[50];
1841      char src_str[50];
1842
1843      /* Note that texreg2ar treats Tx as a temporary register, not as a varying */
1844      shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
1845      shader_arb_get_src_param(ins, &ins->src[0], 0, src_str);
1846      /* Move .x first in case src_str is "TA" */
1847      shader_addline(buffer, "MOV TA.y, %s.x;\n", src_str);
1848      shader_addline(buffer, "MOV TA.x, %s.w;\n", src_str);
1849      flags = reg1 < MAX_TEXTURES ? deviceImpl->stateBlock->textureState[reg1][WINED3DTSS_TEXTURETRANSFORMFLAGS] : 0;
1850      shader_hw_sample(ins, reg1, dst_str, "TA", flags & WINED3DTTFF_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
1851 }
1852
1853 static void pshader_hw_texreg2gb(const struct wined3d_shader_instruction *ins)
1854 {
1855      struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1856
1857      DWORD reg1 = ins->dst[0].reg.idx;
1858      char dst_str[50];
1859      char src_str[50];
1860
1861      /* Note that texreg2gb treats Tx as a temporary register, not as a varying */
1862      shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
1863      shader_arb_get_src_param(ins, &ins->src[0], 0, src_str);
1864      shader_addline(buffer, "MOV TA.x, %s.y;\n", src_str);
1865      shader_addline(buffer, "MOV TA.y, %s.z;\n", src_str);
1866      shader_hw_sample(ins, reg1, dst_str, "TA", 0, NULL, NULL);
1867 }
1868
1869 static void pshader_hw_texreg2rgb(const struct wined3d_shader_instruction *ins)
1870 {
1871     DWORD reg1 = ins->dst[0].reg.idx;
1872     char dst_str[50];
1873     char src_str[50];
1874
1875     /* Note that texreg2rg treats Tx as a temporary register, not as a varying */
1876     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
1877     shader_arb_get_src_param(ins, &ins->src[0], 0, src_str);
1878     shader_hw_sample(ins, reg1, dst_str, src_str, 0, NULL, NULL);
1879 }
1880
1881 static void pshader_hw_texbem(const struct wined3d_shader_instruction *ins)
1882 {
1883     IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)ins->ctx->shader;
1884     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1885     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1886     char reg_coord[40], dst_reg[50], src_reg[50];
1887     DWORD reg_dest_code;
1888
1889     /* All versions have a destination register. The Tx where the texture coordinates come
1890      * from is the varying incarnation of the texture register
1891      */
1892     reg_dest_code = dst->reg.idx;
1893     shader_arb_get_dst_param(ins, &ins->dst[0], dst_reg);
1894     shader_arb_get_src_param(ins, &ins->src[0], 0, src_reg);
1895     sprintf(reg_coord, "fragment.texcoord[%u]", reg_dest_code);
1896
1897     /* Sampling the perturbation map in Tsrc was done already, including the signedness correction if needed
1898      * The Tx in which the perturbation map is stored is the tempreg incarnation of the texture register
1899      *
1900      * GL_NV_fragment_program_option could handle this in one instruction via X2D:
1901      * X2D TA.xy, fragment.texcoord, T%u, bumpenvmat%u.xzyw
1902      *
1903      * However, the NV extensions are never enabled for <= 2.0 shaders because of the performance penalty that
1904      * comes with it, and texbem is an 1.x only instruction. No 1.x instruction forces us to enable the NV
1905      * extension.
1906      */
1907     shader_addline(buffer, "SWZ TB, bumpenvmat%d, x, z, 0, 0;\n", reg_dest_code);
1908     shader_addline(buffer, "DP3 TA.x, TB, %s;\n", src_reg);
1909     shader_addline(buffer, "SWZ TB, bumpenvmat%d, y, w, 0, 0;\n", reg_dest_code);
1910     shader_addline(buffer, "DP3 TA.y, TB, %s;\n", src_reg);
1911
1912     /* with projective textures, texbem only divides the static texture coord, not the displacement,
1913      * so we can't let the GL handle this.
1914      */
1915     if (((IWineD3DDeviceImpl*) This->baseShader.device)->stateBlock->textureState[reg_dest_code][WINED3DTSS_TEXTURETRANSFORMFLAGS]
1916             & WINED3DTTFF_PROJECTED) {
1917         shader_addline(buffer, "RCP TB.w, %s.w;\n", reg_coord);
1918         shader_addline(buffer, "MUL TB.xy, %s, TB.w;\n", reg_coord);
1919         shader_addline(buffer, "ADD TA.xy, TA, TB;\n");
1920     } else {
1921         shader_addline(buffer, "ADD TA.xy, TA, %s;\n", reg_coord);
1922     }
1923
1924     shader_hw_sample(ins, reg_dest_code, dst_reg, "TA", 0, NULL, NULL);
1925
1926     if (ins->handler_idx == WINED3DSIH_TEXBEML)
1927     {
1928         /* No src swizzles are allowed, so this is ok */
1929         shader_addline(buffer, "MAD TA, %s.z, luminance%d.x, luminance%d.y;\n",
1930                        src_reg, reg_dest_code, reg_dest_code);
1931         shader_addline(buffer, "MUL %s, %s, TA;\n", dst_reg, dst_reg);
1932     }
1933 }
1934
1935 static void pshader_hw_texm3x2pad(const struct wined3d_shader_instruction *ins)
1936 {
1937     DWORD reg = ins->dst[0].reg.idx;
1938     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1939     char src0_name[50], dst_name[50];
1940     BOOL is_color;
1941     struct wined3d_shader_register tmp_reg = ins->dst[0].reg;
1942
1943     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
1944     /* The next instruction will be a texm3x2tex or texm3x2depth that writes to the uninitialized
1945      * T<reg+1> register. Use this register to store the calculated vector
1946      */
1947     tmp_reg.idx = reg + 1;
1948     shader_arb_get_register_name(ins, &tmp_reg, dst_name, &is_color);
1949     shader_addline(buffer, "DP3 %s.x, fragment.texcoord[%u], %s;\n", dst_name, reg, src0_name);
1950 }
1951
1952 static void pshader_hw_texm3x2tex(const struct wined3d_shader_instruction *ins)
1953 {
1954     IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)ins->ctx->shader;
1955     IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
1956     DWORD flags;
1957     DWORD reg = ins->dst[0].reg.idx;
1958     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1959     char dst_str[50];
1960     char src0_name[50];
1961     char dst_reg[50];
1962     BOOL is_color;
1963
1964     /* We know that we're writing to the uninitialized T<reg> register, so use it for temporary storage */
1965     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_reg, &is_color);
1966
1967     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
1968     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
1969     shader_addline(buffer, "DP3 %s.y, fragment.texcoord[%u], %s;\n", dst_reg, reg, src0_name);
1970     flags = reg < MAX_TEXTURES ? deviceImpl->stateBlock->textureState[reg][WINED3DTSS_TEXTURETRANSFORMFLAGS] : 0;
1971     shader_hw_sample(ins, reg, dst_str, dst_reg, flags & WINED3DTTFF_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
1972 }
1973
1974 static void pshader_hw_texm3x3pad(const struct wined3d_shader_instruction *ins)
1975 {
1976     IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)ins->ctx->shader;
1977     DWORD reg = ins->dst[0].reg.idx;
1978     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1979     SHADER_PARSE_STATE* current_state = &This->baseShader.parse_state;
1980     char src0_name[50], dst_name[50];
1981     struct wined3d_shader_register tmp_reg = ins->dst[0].reg;
1982     BOOL is_color;
1983
1984     /* There are always 2 texm3x3pad instructions followed by one texm3x3[tex,vspec, ...] instruction, with
1985      * incrementing ins->dst[0].register_idx numbers. So the pad instruction already knows the final destination
1986      * register, and this register is uninitialized(otherwise the assembler complains that it is 'redeclared')
1987      */
1988     tmp_reg.idx = reg + 2 - current_state->current_row;
1989     shader_arb_get_register_name(ins, &tmp_reg, dst_name, &is_color);
1990
1991     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
1992     shader_addline(buffer, "DP3 %s.%c, fragment.texcoord[%u], %s;\n",
1993                    dst_name, 'x' + current_state->current_row, reg, src0_name);
1994     current_state->texcoord_w[current_state->current_row++] = reg;
1995 }
1996
1997 static void pshader_hw_texm3x3tex(const struct wined3d_shader_instruction *ins)
1998 {
1999     IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)ins->ctx->shader;
2000     IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
2001     DWORD flags;
2002     DWORD reg = ins->dst[0].reg.idx;
2003     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2004     SHADER_PARSE_STATE* current_state = &This->baseShader.parse_state;
2005     char dst_str[50];
2006     char src0_name[50], dst_name[50];
2007     BOOL is_color;
2008
2009     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2010     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2011     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_name, reg, src0_name);
2012
2013     /* Sample the texture using the calculated coordinates */
2014     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2015     flags = reg < MAX_TEXTURES ? deviceImpl->stateBlock->textureState[reg][WINED3DTSS_TEXTURETRANSFORMFLAGS] : 0;
2016     shader_hw_sample(ins, reg, dst_str, dst_name, flags & WINED3DTTFF_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2017     current_state->current_row = 0;
2018 }
2019
2020 static void pshader_hw_texm3x3vspec(const struct wined3d_shader_instruction *ins)
2021 {
2022     IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)ins->ctx->shader;
2023     IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
2024     DWORD flags;
2025     DWORD reg = ins->dst[0].reg.idx;
2026     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2027     SHADER_PARSE_STATE* current_state = &This->baseShader.parse_state;
2028     char dst_str[50];
2029     char src0_name[50];
2030     char dst_reg[50];
2031     BOOL is_color;
2032
2033     /* Get the dst reg without writemask strings. We know this register is uninitialized, so we can use all
2034      * components for temporary data storage
2035      */
2036     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_reg, &is_color);
2037     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2038     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_reg, reg, src0_name);
2039
2040     /* Construct the eye-ray vector from w coordinates */
2041     shader_addline(buffer, "MOV TB.x, fragment.texcoord[%u].w;\n", current_state->texcoord_w[0]);
2042     shader_addline(buffer, "MOV TB.y, fragment.texcoord[%u].w;\n", current_state->texcoord_w[1]);
2043     shader_addline(buffer, "MOV TB.z, fragment.texcoord[%u].w;\n", reg);
2044
2045     /* Calculate reflection vector
2046      */
2047     shader_addline(buffer, "DP3 %s.w, %s, TB;\n", dst_reg, dst_reg);
2048     /* The .w is ignored when sampling, so I can use TB.w to calculate dot(N, N) */
2049     shader_addline(buffer, "DP3 TB.w, %s, %s;\n", dst_reg, dst_reg);
2050     shader_addline(buffer, "RCP TB.w, TB.w;\n");
2051     shader_addline(buffer, "MUL %s.w, %s.w, TB.w;\n", dst_reg, dst_reg);
2052     shader_addline(buffer, "MUL %s, %s.w, %s;\n", dst_reg, dst_reg, dst_reg);
2053     shader_addline(buffer, "MAD %s, coefmul.x, %s, -TB;\n", dst_reg, dst_reg);
2054
2055     /* Sample the texture using the calculated coordinates */
2056     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2057     flags = reg < MAX_TEXTURES ? deviceImpl->stateBlock->textureState[reg][WINED3DTSS_TEXTURETRANSFORMFLAGS] : 0;
2058     shader_hw_sample(ins, reg, dst_str, dst_reg, flags & WINED3DTTFF_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2059     current_state->current_row = 0;
2060 }
2061
2062 static void pshader_hw_texm3x3spec(const struct wined3d_shader_instruction *ins)
2063 {
2064     IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)ins->ctx->shader;
2065     IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
2066     DWORD flags;
2067     DWORD reg = ins->dst[0].reg.idx;
2068     SHADER_PARSE_STATE* current_state = &This->baseShader.parse_state;
2069     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2070     char dst_str[50];
2071     char src0_name[50];
2072     char src1_name[50];
2073     char dst_reg[50];
2074     BOOL is_color;
2075
2076     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2077     shader_arb_get_src_param(ins, &ins->src[0], 1, src1_name);
2078     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_reg, &is_color);
2079     /* Note: dst_reg.xy is input here, generated by two texm3x3pad instructions */
2080     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_reg, reg, src0_name);
2081
2082     /* Calculate reflection vector.
2083      *
2084      *                   dot(N, E)
2085      * dst_reg.xyz = 2 * --------- * N - E
2086      *                   dot(N, N)
2087      *
2088      * Which normalizes the normal vector
2089      */
2090     shader_addline(buffer, "DP3 %s.w, %s, %s;\n", dst_reg, dst_reg, src1_name);
2091     shader_addline(buffer, "DP3 TC.w, %s, %s;\n", dst_reg, dst_reg);
2092     shader_addline(buffer, "RCP TC.w, TC.w;\n");
2093     shader_addline(buffer, "MUL %s.w, %s.w, TC.w;\n", dst_reg, dst_reg);
2094     shader_addline(buffer, "MUL %s, %s.w, %s;\n", dst_reg, dst_reg, dst_reg);
2095     shader_addline(buffer, "MAD %s, coefmul.x, %s, -%s;\n", dst_reg, dst_reg, src1_name);
2096
2097     /* Sample the texture using the calculated coordinates */
2098     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2099     flags = reg < MAX_TEXTURES ? deviceImpl->stateBlock->textureState[reg][WINED3DTSS_TEXTURETRANSFORMFLAGS] : 0;
2100     shader_hw_sample(ins, reg, dst_str, dst_reg, flags & WINED3DTTFF_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2101     current_state->current_row = 0;
2102 }
2103
2104 static void pshader_hw_texdepth(const struct wined3d_shader_instruction *ins)
2105 {
2106     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2107     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2108     char dst_name[50];
2109
2110     /* texdepth has an implicit destination, the fragment depth value. It's only parameter,
2111      * which is essentially an input, is the destination register because it is the first
2112      * parameter. According to the msdn, this must be register r5, but let's keep it more flexible
2113      * here(writemasks/swizzles are not valid on texdepth)
2114      */
2115     shader_arb_get_dst_param(ins, dst, dst_name);
2116
2117     /* According to the msdn, the source register(must be r5) is unusable after
2118      * the texdepth instruction, so we're free to modify it
2119      */
2120     shader_addline(buffer, "MIN %s.y, %s.y, one.y;\n", dst_name, dst_name);
2121
2122     /* How to deal with the special case dst_name.g == 0? if r != 0, then
2123      * the r * (1 / 0) will give infinity, which is clamped to 1.0, the correct
2124      * result. But if r = 0.0, then 0 * inf = 0, which is incorrect.
2125      */
2126     shader_addline(buffer, "RCP %s.y, %s.y;\n", dst_name, dst_name);
2127     shader_addline(buffer, "MUL TA.x, %s.x, %s.y;\n", dst_name, dst_name);
2128     shader_addline(buffer, "MIN TA.x, TA.x, one.x;\n");
2129     shader_addline(buffer, "MAX result.depth, TA.x, 0.0;\n");
2130 }
2131
2132 /** Process the WINED3DSIO_TEXDP3TEX instruction in ARB:
2133  * Take a 3-component dot product of the TexCoord[dstreg] and src,
2134  * then perform a 1D texture lookup from stage dstregnum, place into dst. */
2135 static void pshader_hw_texdp3tex(const struct wined3d_shader_instruction *ins)
2136 {
2137     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2138     DWORD sampler_idx = ins->dst[0].reg.idx;
2139     char src0[50];
2140     char dst_str[50];
2141
2142     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2143     shader_addline(buffer, "MOV TB, 0.0;\n");
2144     shader_addline(buffer, "DP3 TB.x, fragment.texcoord[%u], %s;\n", sampler_idx, src0);
2145
2146     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2147     shader_hw_sample(ins, sampler_idx, dst_str, "TB", 0 /* Only one coord, can't be projected */, NULL, NULL);
2148 }
2149
2150 /** Process the WINED3DSIO_TEXDP3 instruction in ARB:
2151  * Take a 3-component dot product of the TexCoord[dstreg] and src. */
2152 static void pshader_hw_texdp3(const struct wined3d_shader_instruction *ins)
2153 {
2154     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2155     char src0[50];
2156     char dst_str[50];
2157     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2158
2159     /* Handle output register */
2160     shader_arb_get_dst_param(ins, dst, dst_str);
2161     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2162     shader_addline(buffer, "DP3 %s, fragment.texcoord[%u], %s;\n", dst_str, dst->reg.idx, src0);
2163 }
2164
2165 /** Process the WINED3DSIO_TEXM3X3 instruction in ARB
2166  * Perform the 3rd row of a 3x3 matrix multiply */
2167 static void pshader_hw_texm3x3(const struct wined3d_shader_instruction *ins)
2168 {
2169     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2170     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2171     char dst_str[50], dst_name[50];
2172     char src0[50];
2173     BOOL is_color;
2174
2175     shader_arb_get_dst_param(ins, dst, dst_str);
2176     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2177     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2178     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_name, dst->reg.idx, src0);
2179     shader_addline(buffer, "MOV %s, %s;\n", dst_str, dst_name);
2180 }
2181
2182 /** Process the WINED3DSIO_TEXM3X2DEPTH instruction in ARB:
2183  * Last row of a 3x2 matrix multiply, use the result to calculate the depth:
2184  * Calculate tmp0.y = TexCoord[dstreg] . src.xyz;  (tmp0.x has already been calculated)
2185  * depth = (tmp0.y == 0.0) ? 1.0 : tmp0.x / tmp0.y
2186  */
2187 static void pshader_hw_texm3x2depth(const struct wined3d_shader_instruction *ins)
2188 {
2189     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2190     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2191     char src0[50], dst_name[50];
2192     BOOL is_color;
2193
2194     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2195     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2196     shader_addline(buffer, "DP3 %s.y, fragment.texcoord[%u], %s;\n", dst_name, dst->reg.idx, src0);
2197
2198     /* How to deal with the special case dst_name.g == 0? if r != 0, then
2199      * the r * (1 / 0) will give infinity, which is clamped to 1.0, the correct
2200      * result. But if r = 0.0, then 0 * inf = 0, which is incorrect.
2201      */
2202     shader_addline(buffer, "RCP %s.y, %s.y;\n", dst_name, dst_name);
2203     shader_addline(buffer, "MUL %s.x, %s.x, %s.y;\n", dst_name, dst_name, dst_name);
2204     shader_addline(buffer, "MIN %s.x, %s.x, one.x;\n", dst_name, dst_name);
2205     shader_addline(buffer, "MAX result.depth, %s.x, 0.0;\n", dst_name);
2206 }
2207
2208 /** Handles transforming all WINED3DSIO_M?x? opcodes for
2209     Vertex/Pixel shaders to ARB_vertex_program codes */
2210 static void shader_hw_mnxn(const struct wined3d_shader_instruction *ins)
2211 {
2212     int i;
2213     int nComponents = 0;
2214     struct wined3d_shader_dst_param tmp_dst = {{0}};
2215     struct wined3d_shader_src_param tmp_src[2] = {{{0}}};
2216     struct wined3d_shader_instruction tmp_ins;
2217
2218     memset(&tmp_ins, 0, sizeof(tmp_ins));
2219
2220     /* Set constants for the temporary argument */
2221     tmp_ins.ctx = ins->ctx;
2222     tmp_ins.dst_count = 1;
2223     tmp_ins.dst = &tmp_dst;
2224     tmp_ins.src_count = 2;
2225     tmp_ins.src = tmp_src;
2226
2227     switch(ins->handler_idx)
2228     {
2229         case WINED3DSIH_M4x4:
2230             nComponents = 4;
2231             tmp_ins.handler_idx = WINED3DSIH_DP4;
2232             break;
2233         case WINED3DSIH_M4x3:
2234             nComponents = 3;
2235             tmp_ins.handler_idx = WINED3DSIH_DP4;
2236             break;
2237         case WINED3DSIH_M3x4:
2238             nComponents = 4;
2239             tmp_ins.handler_idx = WINED3DSIH_DP3;
2240             break;
2241         case WINED3DSIH_M3x3:
2242             nComponents = 3;
2243             tmp_ins.handler_idx = WINED3DSIH_DP3;
2244             break;
2245         case WINED3DSIH_M3x2:
2246             nComponents = 2;
2247             tmp_ins.handler_idx = WINED3DSIH_DP3;
2248             break;
2249         default:
2250             FIXME("Unhandled opcode %#x\n", ins->handler_idx);
2251             break;
2252     }
2253
2254     tmp_dst = ins->dst[0];
2255     tmp_src[0] = ins->src[0];
2256     tmp_src[1] = ins->src[1];
2257     for (i = 0; i < nComponents; i++) {
2258         tmp_dst.write_mask = WINED3DSP_WRITEMASK_0 << i;
2259         shader_hw_map2gl(&tmp_ins);
2260         ++tmp_src[1].reg.idx;
2261     }
2262 }
2263
2264 static void shader_hw_scalar_op(const struct wined3d_shader_instruction *ins)
2265 {
2266     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2267     const char *instruction;
2268
2269     char dst[50];
2270     char src[50];
2271
2272     switch(ins->handler_idx)
2273     {
2274         case WINED3DSIH_RSQ:  instruction = "RSQ"; break;
2275         case WINED3DSIH_RCP:  instruction = "RCP"; break;
2276         case WINED3DSIH_EXP:  instruction = "EX2"; break;
2277         case WINED3DSIH_EXPP: instruction = "EXP"; break;
2278         default: instruction = "";
2279             FIXME("Unhandled opcode %#x\n", ins->handler_idx);
2280             break;
2281     }
2282
2283     shader_arb_get_dst_param(ins, &ins->dst[0], dst); /* Destination */
2284     shader_arb_get_src_param(ins, &ins->src[0], 0, src);
2285     if (ins->src[0].swizzle == WINED3DSP_NOSWIZZLE)
2286     {
2287         /* Dx sdk says .x is used if no swizzle is given, but our test shows that
2288          * .w is used
2289          */
2290         strcat(src, ".w");
2291     }
2292
2293     shader_addline(buffer, "%s%s %s, %s;\n", instruction, shader_arb_get_modifier(ins), dst, src);
2294 }
2295
2296 static void shader_hw_nrm(const struct wined3d_shader_instruction *ins)
2297 {
2298     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2299     char dst_name[50];
2300     char src_name[50];
2301     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2302     BOOL pshader = shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type);
2303
2304     shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2305     shader_arb_get_src_param(ins, &ins->src[0], 1 /* Use TB */, src_name);
2306
2307     if(pshader && priv->target_version >= NV3)
2308     {
2309         shader_addline(buffer, "NRM%s %s, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name);
2310     }
2311     else
2312     {
2313         shader_addline(buffer, "DP3 TA, %s, %s;\n", src_name, src_name);
2314         shader_addline(buffer, "RSQ TA, TA.x;\n");
2315         /* dst.w = src[0].w * 1 / (src.x^2 + src.y^2 + src.z^2)^(1/2) according to msdn*/
2316         shader_addline(buffer, "MUL%s %s, %s, TA;\n", shader_arb_get_modifier(ins), dst_name,
2317                     src_name);
2318     }
2319 }
2320
2321 static void shader_hw_lrp(const struct wined3d_shader_instruction *ins)
2322 {
2323     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2324     char dst_name[50];
2325     char src_name[3][50];
2326
2327     /* ARB_fragment_program has a convenient LRP instruction */
2328     if(shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type)) {
2329         shader_hw_map2gl(ins);
2330         return;
2331     }
2332
2333     shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2334     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
2335     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
2336     shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
2337
2338     shader_addline(buffer, "SUB TA, %s, %s;\n", src_name[1], src_name[2]);
2339     shader_addline(buffer, "MAD%s %s, %s, TA, %s;\n", shader_arb_get_modifier(ins),
2340                    dst_name, src_name[0], src_name[2]);
2341 }
2342
2343 static void shader_hw_sincos(const struct wined3d_shader_instruction *ins)
2344 {
2345     /* This instruction exists in ARB, but the d3d instruction takes two extra parameters which
2346      * must contain fixed constants. So we need a separate function to filter those constants and
2347      * can't use map2gl
2348      */
2349     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2350     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2351     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2352     char dst_name[50];
2353     char src_name0[50], src_name1[50], src_name2[50];
2354     BOOL is_color;
2355
2356     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name0);
2357     if(shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type)) {
2358         shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2359         /* No modifiers are supported on SCS */
2360         shader_addline(buffer, "SCS %s, %s;\n", dst_name, src_name0);
2361
2362         if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE)
2363         {
2364             shader_arb_get_register_name(ins, &dst->reg, src_name0, &is_color);
2365             shader_addline(buffer, "MOV_SAT %s, %s;\n", dst_name, src_name0);
2366         }
2367     } else if(priv->target_version >= NV2) {
2368         shader_arb_get_register_name(ins, &dst->reg, dst_name, &is_color);
2369
2370         /* Sincos writemask must be .x, .y or .xy */
2371         if(dst->write_mask & WINED3DSP_WRITEMASK_0)
2372             shader_addline(buffer, "COS%s %s.x, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name0);
2373         if(dst->write_mask & WINED3DSP_WRITEMASK_1)
2374             shader_addline(buffer, "SIN%s %s.y, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name0);
2375     } else {
2376         /* Approximate sine and cosine with a taylor series, as per math textbook. The application passes 8
2377          * helper constants(D3DSINCOSCONST1 and D3DSINCOSCONST2) in src1 and src2.
2378          *
2379          * sin(x) = x - x^3/3! + x^5/5! - x^7/7! + ...
2380          * cos(x) = 1 - x^2/2! + x^4/4! - x^6/6! + ...
2381          *
2382          * The constants we get are:
2383          *
2384          *  +1   +1,     -1     -1     +1      +1      -1       -1
2385          *      ---- ,  ---- , ---- , ----- , ----- , ----- , ------
2386          *      1!*2    2!*4   3!*8   4!*16   5!*32   6!*64   7!*128
2387          *
2388          * If used with x^2, x^3, x^4 etc they calculate sin(x/2) and cos(x/2):
2389          *
2390          * (x/2)^2 = x^2 / 4
2391          * (x/2)^3 = x^3 / 8
2392          * (x/2)^4 = x^4 / 16
2393          * (x/2)^5 = x^5 / 32
2394          * etc
2395          *
2396          * To get the final result:
2397          * sin(x) = 2 * sin(x/2) * cos(x/2)
2398          * cos(x) = cos(x/2)^2 - sin(x/2)^2
2399          * (from sin(x+y) and cos(x+y) rules)
2400          *
2401          * As per MSDN, dst.z is undefined after the operation, and so is
2402          * dst.x and dst.y if they're masked out by the writemask. Ie
2403          * sincos dst.y, src1, c0, c1
2404          * returns the sine in dst.y. dst.x and dst.z are undefined, dst.w is not touched. The assembler
2405          * vsa.exe also stops with an error if the dest register is the same register as the source
2406          * register. This means we can use dest.xyz as temporary storage. The assembler vsa.exe output also
2407          * indicates that sincos consumes 8 instruction slots in vs_2_0(and, strangely, in vs_3_0).
2408          */
2409         shader_arb_get_src_param(ins, &ins->src[1], 1, src_name1);
2410         shader_arb_get_src_param(ins, &ins->src[2], 2, src_name2);
2411         shader_arb_get_register_name(ins, &dst->reg, dst_name, &is_color);
2412
2413         shader_addline(buffer, "MUL %s.x, %s, %s;\n", dst_name, src_name0, src_name0);  /* x ^ 2 */
2414         shader_addline(buffer, "MUL TA.y, %s.x, %s;\n", dst_name, src_name0);           /* x ^ 3 */
2415         shader_addline(buffer, "MUL %s.y, TA.y, %s;\n", dst_name, src_name0);           /* x ^ 4 */
2416         shader_addline(buffer, "MUL TA.z, %s.y, %s;\n", dst_name, src_name0);           /* x ^ 5 */
2417         shader_addline(buffer, "MUL %s.z, TA.z, %s;\n", dst_name, src_name0);           /* x ^ 6 */
2418         shader_addline(buffer, "MUL TA.w, %s.z, %s;\n", dst_name, src_name0);           /* x ^ 7 */
2419
2420         /* sin(x/2)
2421          *
2422          * Unfortunately we don't get the constants in a DP4-capable form. Is there a way to
2423          * properly merge that with MULs in the code above?
2424          * The swizzles .yz and xw however fit into the .yzxw swizzle added to ps_2_0. Maybe
2425          * we can merge the sine and cosine MAD rows to calculate them together.
2426          */
2427         shader_addline(buffer, "MUL TA.x, %s, %s.w;\n", src_name0, src_name2); /* x^1, +1/(1!*2) */
2428         shader_addline(buffer, "MAD TA.x, TA.y, %s.x, TA.x;\n", src_name2); /* -1/(3!*8) */
2429         shader_addline(buffer, "MAD TA.x, TA.z, %s.w, TA.x;\n", src_name1); /* +1/(5!*32) */
2430         shader_addline(buffer, "MAD TA.x, TA.w, %s.x, TA.x;\n", src_name1); /* -1/(7!*128) */
2431
2432         /* cos(x/2) */
2433         shader_addline(buffer, "MAD TA.y, %s.x, %s.y, %s.z;\n", dst_name, src_name2, src_name2); /* -1/(2!*4), +1.0 */
2434         shader_addline(buffer, "MAD TA.y, %s.y, %s.z, TA.y;\n", dst_name, src_name1); /* +1/(4!*16) */
2435         shader_addline(buffer, "MAD TA.y, %s.z, %s.y, TA.y;\n", dst_name, src_name1); /* -1/(6!*64) */
2436
2437         if(dst->write_mask & WINED3DSP_WRITEMASK_0) {
2438             /* cos x */
2439             shader_addline(buffer, "MUL TA.z, TA.y, TA.y;\n");
2440             shader_addline(buffer, "MAD %s.x, -TA.x, TA.x, TA.z;\n", dst_name);
2441         }
2442         if(dst->write_mask & WINED3DSP_WRITEMASK_1) {
2443             /* sin x */
2444             shader_addline(buffer, "MUL %s.y, TA.x, TA.y;\n", dst_name);
2445             shader_addline(buffer, "ADD %s.y, %s.y, %s.y;\n", dst_name, dst_name, dst_name);
2446         }
2447     }
2448 }
2449
2450 static void shader_hw_sgn(const struct wined3d_shader_instruction *ins)
2451 {
2452     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2453     char dst_name[50];
2454     char src_name[50];
2455     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
2456
2457     /* SGN is only valid in vertex shaders */
2458     if(ctx->target_version == NV2) {
2459         shader_hw_map2gl(ins);
2460         return;
2461     }
2462     shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2463     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name);
2464
2465     /* If SRC > 0.0, -SRC < SRC = TRUE, otherwise false.
2466      * if SRC < 0.0,  SRC < -SRC = TRUE. If neither is true, src = 0.0
2467      */
2468     if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE) {
2469         shader_addline(buffer, "SLT %s, -%s, %s;\n", dst_name, src_name, src_name);
2470     } else {
2471         /* src contains TA? Write to the dest first. This won't overwrite our destination.
2472          * Then use TA, and calculate the final result
2473          *
2474          * Not reading from TA? Store the first result in TA to avoid overwriting the
2475          * destination if src reg = dst reg
2476          */
2477         if(strstr(src_name, "TA"))
2478         {
2479             shader_addline(buffer, "SLT %s,  %s, -%s;\n", dst_name, src_name, src_name);
2480             shader_addline(buffer, "SLT TA, -%s, %s;\n", src_name, src_name);
2481             shader_addline(buffer, "ADD %s, %s, -TA;\n", dst_name, dst_name);
2482         }
2483         else
2484         {
2485             shader_addline(buffer, "SLT TA, -%s, %s;\n", src_name, src_name);
2486             shader_addline(buffer, "SLT %s,  %s, -%s;\n", dst_name, src_name, src_name);
2487             shader_addline(buffer, "ADD %s, TA, -%s;\n", dst_name, dst_name);
2488         }
2489     }
2490 }
2491
2492 static void shader_hw_dsy(const struct wined3d_shader_instruction *ins)
2493 {
2494     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2495     char src[50];
2496     char dst[50];
2497     char dst_name[50];
2498     BOOL is_color;
2499
2500     shader_arb_get_dst_param(ins, &ins->dst[0], dst);
2501     shader_arb_get_src_param(ins, &ins->src[0], 0, src);
2502     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2503
2504     shader_addline(buffer, "DDY %s, %s;\n", dst, src);
2505     shader_addline(buffer, "MUL%s %s, %s, ycorrection.y;\n", shader_arb_get_modifier(ins), dst, dst_name);
2506 }
2507
2508 static DWORD abs_modifier(DWORD mod, BOOL *need_abs)
2509 {
2510     *need_abs = FALSE;
2511
2512     switch(mod)
2513     {
2514         case WINED3DSPSM_NONE:      return WINED3DSPSM_ABS;
2515         case WINED3DSPSM_NEG:       return WINED3DSPSM_ABS;
2516         case WINED3DSPSM_BIAS:      *need_abs = TRUE; return WINED3DSPSM_BIAS;
2517         case WINED3DSPSM_BIASNEG:   *need_abs = TRUE; return WINED3DSPSM_BIASNEG;
2518         case WINED3DSPSM_SIGN:      *need_abs = TRUE; return WINED3DSPSM_SIGN;
2519         case WINED3DSPSM_SIGNNEG:   *need_abs = TRUE; return WINED3DSPSM_SIGNNEG;
2520         case WINED3DSPSM_COMP:      *need_abs = TRUE; return WINED3DSPSM_COMP;
2521         case WINED3DSPSM_X2:        *need_abs = TRUE; return WINED3DSPSM_X2;
2522         case WINED3DSPSM_X2NEG:     *need_abs = TRUE; return WINED3DSPSM_X2NEG;
2523         case WINED3DSPSM_DZ:        *need_abs = TRUE; return WINED3DSPSM_DZ;
2524         case WINED3DSPSM_DW:        *need_abs = TRUE; return WINED3DSPSM_DW;
2525         case WINED3DSPSM_ABS:       return WINED3DSPSM_ABS;
2526         case WINED3DSPSM_ABSNEG:    return WINED3DSPSM_ABS;
2527     }
2528     FIXME("Unknown modifier %u\n", mod);
2529     return mod;
2530 }
2531
2532 static void shader_hw_log_pow(const struct wined3d_shader_instruction *ins)
2533 {
2534     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2535     char src0[50], src1[50], dst[50];
2536     struct wined3d_shader_src_param src0_copy = ins->src[0];
2537     BOOL need_abs = FALSE;
2538     const char *instr;
2539     BOOL arg2 = FALSE;
2540
2541     switch(ins->handler_idx)
2542     {
2543         case WINED3DSIH_LOG:  instr = "LG2"; break;
2544         case WINED3DSIH_LOGP: instr = "LOG"; break;
2545         case WINED3DSIH_POW:  instr = "POW"; arg2 = TRUE; break;
2546         default:
2547             ERR("Unexpected instruction %d\n", ins->handler_idx);
2548             return;
2549     }
2550
2551     /* LOG, LOGP and POW operate on the absolute value of the input */
2552     src0_copy.modifiers = abs_modifier(src0_copy.modifiers, &need_abs);
2553
2554     shader_arb_get_dst_param(ins, &ins->dst[0], dst);
2555     shader_arb_get_src_param(ins, &src0_copy, 0, src0);
2556     if(arg2) shader_arb_get_src_param(ins, &ins->src[1], 1, src1);
2557
2558     if(need_abs)
2559     {
2560         shader_addline(buffer, "ABS TA, %s;\n", src0);
2561         if(arg2)
2562         {
2563             shader_addline(buffer, "%s%s %s, TA, %s;\n", instr, shader_arb_get_modifier(ins), dst, src1);
2564         }
2565         else
2566         {
2567             shader_addline(buffer, "%s%s %s, TA;\n", instr, shader_arb_get_modifier(ins), dst);
2568         }
2569     }
2570     else if(arg2)
2571     {
2572         shader_addline(buffer, "%s%s %s, %s, %s;\n", instr, shader_arb_get_modifier(ins), dst, src0, src1);
2573     }
2574     else
2575     {
2576         shader_addline(buffer, "%s%s %s, %s;\n", instr, shader_arb_get_modifier(ins), dst, src0);
2577     }
2578 }
2579
2580 static void shader_hw_loop(const struct wined3d_shader_instruction *ins)
2581 {
2582     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2583     char src_name[50];
2584     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2585
2586     /* src0 is aL */
2587     shader_arb_get_src_param(ins, &ins->src[1], 0, src_name);
2588
2589     if(vshader)
2590     {
2591         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2592         struct list *e = list_head(&priv->control_frames);
2593         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2594
2595         if(priv->loop_depth > 1) shader_addline(buffer, "PUSHA aL;\n");
2596         /* The constant loader makes sure to load -1 into iX.w */
2597         shader_addline(buffer, "ARLC aL, %s.xywz;\n", src_name);
2598         shader_addline(buffer, "BRA loop_%u_end (LE.x);\n", control_frame->loop_no);
2599         shader_addline(buffer, "loop_%u_start:\n", control_frame->loop_no);
2600     }
2601     else
2602     {
2603         shader_addline(buffer, "LOOP %s;\n", src_name);
2604     }
2605 }
2606
2607 static void shader_hw_rep(const struct wined3d_shader_instruction *ins)
2608 {
2609     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2610     char src_name[50];
2611     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2612
2613     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name);
2614
2615     /* The constant loader makes sure to load -1 into iX.w */
2616     if(vshader)
2617     {
2618         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2619         struct list *e = list_head(&priv->control_frames);
2620         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2621
2622         if(priv->loop_depth > 1) shader_addline(buffer, "PUSHA aL;\n");
2623
2624         shader_addline(buffer, "ARLC aL, %s.xywz;\n", src_name);
2625         shader_addline(buffer, "BRA loop_%u_end (LE.x);\n", control_frame->loop_no);
2626         shader_addline(buffer, "loop_%u_start:\n", control_frame->loop_no);
2627     }
2628     else
2629     {
2630         shader_addline(buffer, "REP %s;\n", src_name);
2631     }
2632 }
2633
2634 static void shader_hw_endloop(const struct wined3d_shader_instruction *ins)
2635 {
2636     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2637     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2638
2639     if(vshader)
2640     {
2641         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2642         struct list *e = list_head(&priv->control_frames);
2643         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2644
2645         shader_addline(buffer, "ARAC aL.xy, aL;\n");
2646         shader_addline(buffer, "BRA loop_%u_start (GT.x);\n", control_frame->loop_no);
2647         shader_addline(buffer, "loop_%u_end:\n", control_frame->loop_no);
2648
2649         if(priv->loop_depth > 1) shader_addline(buffer, "POPA aL;\n");
2650     }
2651     else
2652     {
2653         shader_addline(buffer, "ENDLOOP;\n");
2654     }
2655 }
2656
2657 static void shader_hw_endrep(const struct wined3d_shader_instruction *ins)
2658 {
2659     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2660     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2661
2662     if(vshader)
2663     {
2664         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2665         struct list *e = list_head(&priv->control_frames);
2666         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2667
2668         shader_addline(buffer, "ARAC aL.xy, aL;\n");
2669         shader_addline(buffer, "BRA loop_%u_start (GT.x);\n", control_frame->loop_no);
2670         shader_addline(buffer, "loop_%u_end:\n", control_frame->loop_no);
2671
2672         if(priv->loop_depth > 1) shader_addline(buffer, "POPA aL;\n");
2673     }
2674     else
2675     {
2676         shader_addline(buffer, "ENDREP;\n");
2677     }
2678 }
2679
2680 static const struct control_frame *find_last_loop(const struct shader_arb_ctx_priv *priv)
2681 {
2682     struct control_frame *control_frame;
2683
2684     LIST_FOR_EACH_ENTRY(control_frame, &priv->control_frames, struct control_frame, entry)
2685     {
2686         if(control_frame->type == LOOP || control_frame->type == REP) return control_frame;
2687     }
2688     ERR("Could not find loop for break\n");
2689     return NULL;
2690 }
2691
2692 static void shader_hw_break(const struct wined3d_shader_instruction *ins)
2693 {
2694     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2695     const struct control_frame *control_frame = find_last_loop(ins->ctx->backend_data);
2696     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2697
2698     if(vshader)
2699     {
2700         shader_addline(buffer, "BRA loop_%u_end;\n", control_frame->loop_no);
2701     }
2702     else
2703     {
2704         shader_addline(buffer, "BRK;\n");
2705     }
2706 }
2707
2708 static const char *get_compare(COMPARISON_TYPE flags)
2709 {
2710     switch (flags)
2711     {
2712         case COMPARISON_GT: return "GT";
2713         case COMPARISON_EQ: return "EQ";
2714         case COMPARISON_GE: return "GE";
2715         case COMPARISON_LT: return "LT";
2716         case COMPARISON_NE: return "NE";
2717         case COMPARISON_LE: return "LE";
2718         default:
2719             FIXME("Unrecognized comparison value: %u\n", flags);
2720             return "(\?\?)";
2721     }
2722 }
2723
2724 static COMPARISON_TYPE invert_compare(COMPARISON_TYPE flags)
2725 {
2726     switch (flags)
2727     {
2728         case COMPARISON_GT: return COMPARISON_LE;
2729         case COMPARISON_EQ: return COMPARISON_NE;
2730         case COMPARISON_GE: return COMPARISON_LT;
2731         case COMPARISON_LT: return COMPARISON_GE;
2732         case COMPARISON_NE: return COMPARISON_EQ;
2733         case COMPARISON_LE: return COMPARISON_GT;
2734         default:
2735             FIXME("Unrecognized comparison value: %u\n", flags);
2736             return -1;
2737     }
2738 }
2739
2740 static void shader_hw_breakc(const struct wined3d_shader_instruction *ins)
2741 {
2742     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2743     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2744     const struct control_frame *control_frame = find_last_loop(ins->ctx->backend_data);
2745     char src_name0[50];
2746     char src_name1[50];
2747     const char *comp = get_compare(ins->flags);
2748
2749     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name0);
2750     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name1);
2751
2752     if(vshader)
2753     {
2754         /* SUBC CC, src0, src1" works only in pixel shaders, so use TA to throw
2755          * away the subtraction result
2756          */
2757         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
2758         shader_addline(buffer, "BRA loop_%u_end (%s.x);\n", control_frame->loop_no, comp);
2759     }
2760     else
2761     {
2762         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
2763         shader_addline(buffer, "BRK (%s.x);\n", comp);
2764     }
2765 }
2766
2767 static void shader_hw_ifc(const struct wined3d_shader_instruction *ins)
2768 {
2769     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2770     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2771     struct list *e = list_head(&priv->control_frames);
2772     struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2773     const char *comp;
2774     char src_name0[50];
2775     char src_name1[50];
2776     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2777
2778     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name0);
2779     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name1);
2780
2781     if(vshader)
2782     {
2783         /* Invert the flag. We jump to the else label if the condition is NOT true */
2784         comp = get_compare(invert_compare(ins->flags));
2785         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
2786         shader_addline(buffer, "BRA ifc_%u_else (%s.x);\n", control_frame->ifc_no, comp);
2787     }
2788     else
2789     {
2790         comp = get_compare(ins->flags);
2791         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
2792         shader_addline(buffer, "IF %s.x;\n", comp);
2793     }
2794 }
2795
2796 static void shader_hw_else(const struct wined3d_shader_instruction *ins)
2797 {
2798     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2799     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2800     struct list *e = list_head(&priv->control_frames);
2801     struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2802     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2803
2804     if(vshader)
2805     {
2806         shader_addline(buffer, "BRA ifc_%u_endif;\n", control_frame->ifc_no);
2807         shader_addline(buffer, "ifc_%u_else:\n", control_frame->ifc_no);
2808         control_frame->had_else = TRUE;
2809     }
2810     else
2811     {
2812         shader_addline(buffer, "ELSE;\n");
2813     }
2814 }
2815
2816 static void shader_hw_endif(const struct wined3d_shader_instruction *ins)
2817 {
2818     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2819     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2820     struct list *e = list_head(&priv->control_frames);
2821     struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2822     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2823
2824     if(vshader)
2825     {
2826         if(control_frame->had_else)
2827         {
2828             shader_addline(buffer, "ifc_%u_endif:\n", control_frame->ifc_no);
2829         }
2830         else
2831         {
2832             shader_addline(buffer, "#No else branch. else is endif\n");
2833             shader_addline(buffer, "ifc_%u_else:\n", control_frame->ifc_no);
2834         }
2835     }
2836     else
2837     {
2838         shader_addline(buffer, "ENDIF;\n");
2839     }
2840 }
2841
2842 static void shader_hw_texldd(const struct wined3d_shader_instruction *ins)
2843 {
2844     DWORD sampler_idx = ins->src[1].reg.idx;
2845     char reg_dest[40];
2846     char reg_src[3][40];
2847     DWORD flags = TEX_DERIV;
2848
2849     shader_arb_get_dst_param(ins, &ins->dst[0], reg_dest);
2850     shader_arb_get_src_param(ins, &ins->src[0], 0, reg_src[0]);
2851     shader_arb_get_src_param(ins, &ins->src[2], 1, reg_src[1]);
2852     shader_arb_get_src_param(ins, &ins->src[3], 2, reg_src[2]);
2853
2854     if (ins->flags & WINED3DSI_TEXLD_PROJECT) flags |= TEX_PROJ;
2855     if (ins->flags & WINED3DSI_TEXLD_BIAS) flags |= TEX_BIAS;
2856
2857     shader_hw_sample(ins, sampler_idx, reg_dest, reg_src[0], flags, reg_src[1], reg_src[2]);
2858 }
2859
2860 static void shader_hw_texldl(const struct wined3d_shader_instruction *ins)
2861 {
2862     DWORD sampler_idx = ins->src[1].reg.idx;
2863     char reg_dest[40];
2864     char reg_coord[40];
2865     DWORD flags = TEX_LOD;
2866
2867     shader_arb_get_dst_param(ins, &ins->dst[0], reg_dest);
2868     shader_arb_get_src_param(ins, &ins->src[0], 0, reg_coord);
2869
2870     if (ins->flags & WINED3DSI_TEXLD_PROJECT) flags |= TEX_PROJ;
2871     if (ins->flags & WINED3DSI_TEXLD_BIAS) flags |= TEX_BIAS;
2872
2873     shader_hw_sample(ins, sampler_idx, reg_dest, reg_coord, flags, NULL, NULL);
2874 }
2875
2876 static void shader_hw_label(const struct wined3d_shader_instruction *ins)
2877 {
2878     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2879     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2880
2881     priv->in_main_func = FALSE;
2882     /* Call instructions activate the NV extensions, not labels and rets. If there is an uncalled
2883      * subroutine, don't generate a label that will make GL complain
2884      */
2885     if(priv->target_version == ARB) return;
2886
2887     shader_addline(buffer, "l%u:\n", ins->src[0].reg.idx);
2888 }
2889
2890 static void vshader_add_footer(IWineD3DVertexShaderImpl *This, struct wined3d_shader_buffer *buffer,
2891         const struct arb_vs_compile_args *args, struct shader_arb_ctx_priv *priv_ctx)
2892 {
2893     const shader_reg_maps *reg_maps = &This->baseShader.reg_maps;
2894     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *)This->baseShader.device;
2895     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
2896     unsigned int i;
2897
2898     /* The D3DRS_FOGTABLEMODE render state defines if the shader-generated fog coord is used
2899      * or if the fragment depth is used. If the fragment depth is used(FOGTABLEMODE != NONE),
2900      * the fog frag coord is thrown away. If the fog frag coord is used, but not written by
2901      * the shader, it is set to 0.0(fully fogged, since start = 1.0, end = 0.0)
2902      */
2903     if(args->super.fog_src == VS_FOG_Z) {
2904         shader_addline(buffer, "MOV result.fogcoord, TMP_OUT.z;\n");
2905     } else if (!reg_maps->fog) {
2906         /* posFixup.x is always 1.0, so we can savely use it */
2907         shader_addline(buffer, "ADD result.fogcoord, posFixup.x, -posFixup.x;\n");
2908     }
2909
2910     /* Write the final position.
2911      *
2912      * OpenGL coordinates specify the center of the pixel while d3d coords specify
2913      * the corner. The offsets are stored in z and w in posFixup. posFixup.y contains
2914      * 1.0 or -1.0 to turn the rendering upside down for offscreen rendering. PosFixup.x
2915      * contains 1.0 to allow a mad, but arb vs swizzles are too restricted for that.
2916      */
2917     shader_addline(buffer, "MUL TA, posFixup, TMP_OUT.w;\n");
2918     shader_addline(buffer, "ADD TMP_OUT.x, TMP_OUT.x, TA.z;\n");
2919     shader_addline(buffer, "MAD TMP_OUT.y, TMP_OUT.y, posFixup.y, TA.w;\n");
2920
2921     if(use_nv_clip(gl_info) && priv_ctx->target_version >= NV2)
2922     {
2923         for(i = 0; i < priv_ctx->vs_clipplanes; i++)
2924         {
2925             shader_addline(buffer, "DP4 result.clip[%u].x, TMP_OUT, state.clip[%u].plane;\n", i, i);
2926         }
2927     }
2928     else if(args->boolclip.clip_control[0])
2929     {
2930         unsigned int cur_clip = 0;
2931         char component[4] = {'x', 'y', 'z', 'w'};
2932
2933         for(i = 0; i < GL_LIMITS(clipplanes); i++)
2934         {
2935             if(args->boolclip.clip_control[1] & (1 << i))
2936             {
2937                 shader_addline(buffer, "DP4 TA.%c, TMP_OUT, state.clip[%u].plane;\n",
2938                                component[cur_clip++], i);
2939             }
2940         }
2941         switch(cur_clip)
2942         {
2943             case 0:
2944                 shader_addline(buffer, "MOV TA, -helper_const.w;\n");
2945                 break;
2946             case 1:
2947                 shader_addline(buffer, "MOV TA.yzw, -helper_const.w;\n");
2948                 break;
2949             case 2:
2950                 shader_addline(buffer, "MOV TA.zw, -helper_const.w;\n");
2951                 break;
2952             case 3:
2953                 shader_addline(buffer, "MOV TA.w, -helper_const.w;\n");
2954                 break;
2955         }
2956         shader_addline(buffer, "MOV result.texcoord[%u], TA;\n",
2957                        args->boolclip.clip_control[0] - 1);
2958     }
2959
2960     /* Z coord [0;1]->[-1;1] mapping, see comment in transform_projection in state.c
2961      * and the glsl equivalent
2962      */
2963     if(need_helper_const(gl_info)) {
2964         shader_addline(buffer, "MAD TMP_OUT.z, TMP_OUT.z, helper_const.x, -TMP_OUT.w;\n");
2965     } else {
2966         shader_addline(buffer, "ADD TMP_OUT.z, TMP_OUT.z, TMP_OUT.z;\n");
2967         shader_addline(buffer, "ADD TMP_OUT.z, TMP_OUT.z, -TMP_OUT.w;\n");
2968     }
2969
2970     shader_addline(buffer, "MOV result.position, TMP_OUT;\n");
2971
2972     priv_ctx->footer_written = TRUE;
2973 }
2974
2975 static void shader_hw_ret(const struct wined3d_shader_instruction *ins)
2976 {
2977     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2978     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2979     IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *) ins->ctx->shader;
2980     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2981
2982     if(priv->target_version == ARB) return;
2983
2984     if(vshader)
2985     {
2986         if(priv->in_main_func) vshader_add_footer((IWineD3DVertexShaderImpl *) shader, buffer, priv->cur_vs_args, priv);
2987     }
2988
2989     shader_addline(buffer, "RET;\n");
2990 }
2991
2992 static void shader_hw_call(const struct wined3d_shader_instruction *ins)
2993 {
2994     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2995     shader_addline(buffer, "CAL l%u;\n", ins->src[0].reg.idx);
2996 }
2997
2998 /* GL locking is done by the caller */
2999 static GLuint create_arb_blt_vertex_program(const struct wined3d_gl_info *gl_info)
3000 {
3001     GLuint program_id = 0;
3002     GLint pos;
3003
3004     const char *blt_vprogram =
3005         "!!ARBvp1.0\n"
3006         "PARAM c[1] = { { 1, 0.5 } };\n"
3007         "MOV result.position, vertex.position;\n"
3008         "MOV result.color, c[0].x;\n"
3009         "MOV result.texcoord[0], vertex.texcoord[0];\n"
3010         "END\n";
3011
3012     GL_EXTCALL(glGenProgramsARB(1, &program_id));
3013     GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, program_id));
3014     GL_EXTCALL(glProgramStringARB(GL_VERTEX_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
3015             strlen(blt_vprogram), blt_vprogram));
3016     checkGLcall("glProgramStringARB()");
3017
3018     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
3019     if (pos != -1)
3020     {
3021         FIXME("Vertex program error at position %d: %s\n", pos,
3022             debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
3023     }
3024     else
3025     {
3026         GLint native;
3027
3028         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
3029         checkGLcall("glGetProgramivARB()");
3030         if (!native) WARN("Program exceeds native resource limits.\n");
3031     }
3032
3033     return program_id;
3034 }
3035
3036 /* GL locking is done by the caller */
3037 static GLuint create_arb_blt_fragment_program(const struct wined3d_gl_info *gl_info, enum tex_types tex_type)
3038 {
3039     GLuint program_id = 0;
3040     GLint pos;
3041
3042     static const char * const blt_fprograms[tex_type_count] =
3043     {
3044         /* tex_1d */
3045         NULL,
3046         /* tex_2d */
3047         "!!ARBfp1.0\n"
3048         "TEMP R0;\n"
3049         "TEX R0.x, fragment.texcoord[0], texture[0], 2D;\n"
3050         "MOV result.depth.z, R0.x;\n"
3051         "END\n",
3052         /* tex_3d */
3053         NULL,
3054         /* tex_cube */
3055         "!!ARBfp1.0\n"
3056         "TEMP R0;\n"
3057         "TEX R0.x, fragment.texcoord[0], texture[0], CUBE;\n"
3058         "MOV result.depth.z, R0.x;\n"
3059         "END\n",
3060         /* tex_rect */
3061         "!!ARBfp1.0\n"
3062         "TEMP R0;\n"
3063         "TEX R0.x, fragment.texcoord[0], texture[0], RECT;\n"
3064         "MOV result.depth.z, R0.x;\n"
3065         "END\n",
3066     };
3067
3068     if (!blt_fprograms[tex_type])
3069     {
3070         FIXME("tex_type %#x not supported\n", tex_type);
3071         tex_type = tex_2d;
3072     }
3073
3074     GL_EXTCALL(glGenProgramsARB(1, &program_id));
3075     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, program_id));
3076     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
3077             strlen(blt_fprograms[tex_type]), blt_fprograms[tex_type]));
3078     checkGLcall("glProgramStringARB()");
3079
3080     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
3081     if (pos != -1)
3082     {
3083         FIXME("Fragment program error at position %d: %s\n", pos,
3084             debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
3085     }
3086     else
3087     {
3088         GLint native;
3089
3090         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
3091         checkGLcall("glGetProgramivARB()");
3092         if (!native) WARN("Program exceeds native resource limits.\n");
3093     }
3094
3095     return program_id;
3096 }
3097
3098 static void arbfp_add_sRGB_correction(struct wined3d_shader_buffer *buffer, const char *fragcolor,
3099         const char *tmp1, const char *tmp2, const char *tmp3, const char *tmp4, BOOL condcode)
3100 {
3101     /* Perform sRGB write correction. See GLX_EXT_framebuffer_sRGB */
3102
3103     if(condcode)
3104     {
3105         /* Sigh. MOVC CC doesn't work, so use one of the temps as dummy dest */
3106         shader_addline(buffer, "SUBC %s, %s.x, srgb_consts1.y;\n", tmp1, fragcolor);
3107         /* Calculate the > 0.0031308 case */
3108         shader_addline(buffer, "POW %s.x (GE), %s.x, srgb_consts1.z;\n", fragcolor, fragcolor);
3109         shader_addline(buffer, "POW %s.y (GE), %s.y, srgb_consts1.z;\n", fragcolor, fragcolor);
3110         shader_addline(buffer, "POW %s.z (GE), %s.z, srgb_consts1.z;\n", fragcolor, fragcolor);
3111         shader_addline(buffer, "MUL %s.xyz (GE), %s, srgb_consts1.w;\n", fragcolor, fragcolor);
3112         shader_addline(buffer, "SUB %s.xyz (GE), %s, srgb_consts2.x;\n", fragcolor, fragcolor);
3113         /* Calculate the < case */
3114         shader_addline(buffer, "MUL %s.xyz (LT), srgb_consts1.x, %s;\n", fragcolor, fragcolor);
3115     }
3116     else
3117     {
3118         /* Calculate the > 0.0031308 case */
3119         shader_addline(buffer, "POW %s.x, %s.x, srgb_consts1.z;\n", tmp1, fragcolor);
3120         shader_addline(buffer, "POW %s.y, %s.y, srgb_consts1.z;\n", tmp1, fragcolor);
3121         shader_addline(buffer, "POW %s.z, %s.z, srgb_consts1.z;\n", tmp1, fragcolor);
3122         shader_addline(buffer, "MUL %s, %s, srgb_consts1.w;\n", tmp1, tmp1);
3123         shader_addline(buffer, "SUB %s, %s, srgb_consts2.x;\n", tmp1, tmp1);
3124         /* Calculate the < case */
3125         shader_addline(buffer, "MUL %s, srgb_consts1.x, %s;\n", tmp2, fragcolor);
3126         /* Get 1.0 / 0.0 masks for > 0.0031308 and < 0.0031308 */
3127         shader_addline(buffer, "SLT %s, srgb_consts1.y, %s;\n", tmp3, fragcolor);
3128         shader_addline(buffer, "SGE %s, srgb_consts1.y, %s;\n", tmp4, fragcolor);
3129         /* Store the components > 0.0031308 in the destination */
3130         shader_addline(buffer, "MUL %s.xyz, %s, %s;\n", fragcolor, tmp1, tmp3);
3131         /* Add the components that are < 0.0031308 */
3132         shader_addline(buffer, "MAD %s.xyz, %s, %s, %s;\n", fragcolor, tmp2, tmp4, fragcolor);
3133         /* Move everything into result.color at once. Nvidia hardware cannot handle partial
3134         * result.color writes(.rgb first, then .a), or handle overwriting already written
3135         * components. The assembler uses a temporary register in this case, which is usually
3136         * not allocated from one of our registers that were used earlier.
3137         */
3138     }
3139     shader_addline(buffer, "MOV result.color, %s;\n", fragcolor);
3140     /* [0.0;1.0] clamping. Not needed, this is done implicitly */
3141 }
3142
3143 static const DWORD *find_loop_control_values(IWineD3DBaseShaderImpl *This, DWORD idx)
3144 {
3145     const local_constant *constant;
3146
3147     LIST_FOR_EACH_ENTRY(constant, &This->baseShader.constantsI, local_constant, entry)
3148     {
3149         if (constant->idx == idx)
3150         {
3151             return constant->value;
3152         }
3153     }
3154     return NULL;
3155 }
3156
3157 static void init_ps_input(const IWineD3DPixelShaderImpl *This, const struct arb_ps_compile_args *args,
3158                           struct shader_arb_ctx_priv *priv)
3159 {
3160     const char *texcoords[8] =
3161     {
3162         "fragment.texcoord[0]", "fragment.texcoord[1]", "fragment.texcoord[2]", "fragment.texcoord[3]",
3163         "fragment.texcoord[4]", "fragment.texcoord[5]", "fragment.texcoord[6]", "fragment.texcoord[7]"
3164     };
3165     unsigned int i;
3166     const struct wined3d_shader_signature_element *sig = This->input_signature;
3167     const char *semantic_name;
3168     DWORD semantic_idx;
3169
3170     switch(args->super.vp_mode)
3171     {
3172         case pretransformed:
3173         case fixedfunction:
3174             /* The pixelshader has to collect the varyings on its own. In any case properly load
3175              * color0 and color1. In the case of pretransformed vertices also load texcoords. Set
3176              * other attribs to 0.0.
3177              *
3178              * For fixedfunction this behavior is correct, according to the tests. For pretransformed
3179              * we'd either need a replacement shader that can load other attribs like BINORMAL, or
3180              * load the texcoord attrib pointers to match the pixel shader signature
3181              */
3182             for(i = 0; i < MAX_REG_INPUT; i++)
3183             {
3184                 semantic_name = sig[i].semantic_name;
3185                 semantic_idx = sig[i].semantic_idx;
3186                 if(semantic_name == NULL) continue;
3187
3188                 if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_COLOR))
3189                 {
3190                     if(semantic_idx == 0) priv->ps_input[i] = "fragment.color.primary";
3191                     else if(semantic_idx == 1) priv->ps_input[i] = "fragment.color.secondary";
3192                     else priv->ps_input[i] = "0.0";
3193                 }
3194                 else if(args->super.vp_mode == fixedfunction)
3195                 {
3196                     priv->ps_input[i] = "0.0";
3197                 }
3198                 else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_TEXCOORD))
3199                 {
3200                     if(semantic_idx < 8) priv->ps_input[i] = texcoords[semantic_idx];
3201                     else priv->ps_input[i] = "0.0";
3202                 }
3203                 else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_FOG))
3204                 {
3205                     if(semantic_idx == 0) priv->ps_input[i] = "fragment.fogcoord";
3206                     else priv->ps_input[i] = "0.0";
3207                 }
3208                 else
3209                 {
3210                     priv->ps_input[i] = "0.0";
3211                 }
3212
3213                 TRACE("v%u, semantic %s%u is %s\n", i, semantic_name, semantic_idx, priv->ps_input[i]);
3214             }
3215             break;
3216
3217         case vertexshader:
3218             /* That one is easy. The vertex shaders provide v0-v7 in fragment.texcoord and v8 and v9 in
3219              * fragment.color
3220              */
3221             for(i = 0; i < 8; i++)
3222             {
3223                 priv->ps_input[i] = texcoords[i];
3224             }
3225             priv->ps_input[8] = "fragment.color.primary";
3226             priv->ps_input[9] = "fragment.color.secondary";
3227             break;
3228     }
3229 }
3230
3231 /* GL locking is done by the caller */
3232 static GLuint shader_arb_generate_pshader(IWineD3DPixelShaderImpl *This, struct wined3d_shader_buffer *buffer,
3233         const struct arb_ps_compile_args *args, struct arb_ps_compiled_shader *compiled)
3234 {
3235     const shader_reg_maps* reg_maps = &This->baseShader.reg_maps;
3236     CONST DWORD *function = This->baseShader.function;
3237     const struct wined3d_gl_info *gl_info = &((IWineD3DDeviceImpl *)This->baseShader.device)->adapter->gl_info;
3238     const local_constant *lconst;
3239     GLuint retval;
3240     char fragcolor[16];
3241     DWORD *lconst_map = local_const_mapping((IWineD3DBaseShaderImpl *) This), next_local, cur;
3242     struct shader_arb_ctx_priv priv_ctx;
3243     BOOL dcl_tmp = args->super.srgb_correction, dcl_td = FALSE;
3244     BOOL want_nv_prog = FALSE;
3245     struct arb_pshader_private *shader_priv = This->backend_priv;
3246     GLint errPos;
3247     DWORD map;
3248
3249     char srgbtmp[4][4];
3250     unsigned int i, found = 0;
3251
3252     for (i = 0, map = reg_maps->temporary; map; map >>= 1, ++i)
3253     {
3254         if (!(map & 1)
3255                 || (This->color0_mov && i == This->color0_reg)
3256                 || (reg_maps->shader_version.major < 2 && i == 0))
3257             continue;
3258
3259         sprintf(srgbtmp[found], "R%u", i);
3260         ++found;
3261         if (found == 4) break;
3262     }
3263
3264     switch(found) {
3265         case 4: dcl_tmp = FALSE; break;
3266         case 0:
3267             sprintf(srgbtmp[0], "TA");
3268             sprintf(srgbtmp[1], "TB");
3269             sprintf(srgbtmp[2], "TC");
3270             sprintf(srgbtmp[3], "TD");
3271             dcl_td = TRUE;
3272             break;
3273         case 1:
3274             sprintf(srgbtmp[1], "TA");
3275             sprintf(srgbtmp[2], "TB");
3276             sprintf(srgbtmp[3], "TC");
3277             break;
3278         case 2:
3279             sprintf(srgbtmp[2], "TA");
3280             sprintf(srgbtmp[3], "TB");
3281             break;
3282         case 3:
3283             sprintf(srgbtmp[3], "TA");
3284             break;
3285     }
3286
3287     /*  Create the hw ARB shader */
3288     memset(&priv_ctx, 0, sizeof(priv_ctx));
3289     priv_ctx.cur_ps_args = args;
3290     priv_ctx.compiled_fprog = compiled;
3291     priv_ctx.cur_np2fixup_info = &compiled->np2fixup_info;
3292     init_ps_input(This, args, &priv_ctx);
3293     list_init(&priv_ctx.control_frames);
3294
3295     /* Avoid enabling NV_fragment_program* if we do not need it.
3296      *
3297      * Enabling GL_NV_fragment_program_option causes the driver to occupy a temporary register,
3298      * and it slows down the shader execution noticeably(about 5%). Usually our instruction emulation
3299      * is faster than what we gain from using higher native instructions. There are some things though
3300      * that cannot be emulated. In that case enable the extensions.
3301      * If the extension is enabled, instruction handlers that support both ways will use it.
3302      *
3303      * Testing shows no performance difference between OPTION NV_fragment_program2 and NV_fragment_program.
3304      * So enable the best we can get.
3305      */
3306     if(reg_maps->usesdsx || reg_maps->usesdsy || reg_maps->loop_depth > 0 || reg_maps->usestexldd ||
3307        reg_maps->usestexldl || reg_maps->usesfacing || reg_maps->usesifc || reg_maps->usescall)
3308     {
3309         want_nv_prog = TRUE;
3310     }
3311
3312     shader_addline(buffer, "!!ARBfp1.0\n");
3313     if(want_nv_prog && GL_SUPPORT(NV_FRAGMENT_PROGRAM2)) {
3314         shader_addline(buffer, "OPTION NV_fragment_program2;\n");
3315         priv_ctx.target_version = NV3;
3316     } else if(want_nv_prog && GL_SUPPORT(NV_FRAGMENT_PROGRAM_OPTION)) {
3317         shader_addline(buffer, "OPTION NV_fragment_program;\n");
3318         priv_ctx.target_version = NV2;
3319     } else {
3320         if(want_nv_prog)
3321         {
3322             /* This is an error - either we're advertising the wrong shader version, or aren't enforcing some
3323              * limits properly
3324              */
3325             ERR("The shader requires instructions that are not available in plain GL_ARB_fragment_program\n");
3326             ERR("Try GLSL\n");
3327         }
3328         priv_ctx.target_version = ARB;
3329     }
3330
3331     if(This->baseShader.reg_maps.highest_render_target > 0)
3332     {
3333         shader_addline(buffer, "OPTION ARB_draw_buffers;\n");
3334     }
3335
3336     if (reg_maps->shader_version.major < 3)
3337     {
3338         switch(args->super.fog) {
3339             case FOG_OFF:
3340                 break;
3341             case FOG_LINEAR:
3342                 shader_addline(buffer, "OPTION ARB_fog_linear;\n");
3343                 break;
3344             case FOG_EXP:
3345                 shader_addline(buffer, "OPTION ARB_fog_exp;\n");
3346                 break;
3347             case FOG_EXP2:
3348                 shader_addline(buffer, "OPTION ARB_fog_exp2;\n");
3349                 break;
3350         }
3351     }
3352
3353     /* For now always declare the temps. At least the Nvidia assembler optimizes completely
3354      * unused temps away(but occupies them for the whole shader if they're used once). Always
3355      * declaring them avoids tricky bookkeeping work
3356      */
3357     shader_addline(buffer, "TEMP TA;\n");      /* Used for modifiers */
3358     shader_addline(buffer, "TEMP TB;\n");      /* Used for modifiers */
3359     shader_addline(buffer, "TEMP TC;\n");      /* Used for modifiers */
3360     if(dcl_td) shader_addline(buffer, "TEMP TD;\n"); /* Used for sRGB writing */
3361     shader_addline(buffer, "PARAM coefdiv = { 0.5, 0.25, 0.125, 0.0625 };\n");
3362     shader_addline(buffer, "PARAM coefmul = { 2, 4, 8, 16 };\n");
3363     shader_addline(buffer, "PARAM one = { 1.0, 1.0, 1.0, 1.0 };\n");
3364
3365     if (reg_maps->shader_version.major < 2)
3366     {
3367         strcpy(fragcolor, "R0");
3368     } else {
3369         if(args->super.srgb_correction) {
3370             if(This->color0_mov) {
3371                 sprintf(fragcolor, "R%u", This->color0_reg);
3372             } else {
3373                 shader_addline(buffer, "TEMP TMP_COLOR;\n");
3374                 strcpy(fragcolor, "TMP_COLOR");
3375             }
3376         } else {
3377             strcpy(fragcolor, "result.color");
3378         }
3379     }
3380
3381     if(args->super.srgb_correction) {
3382         shader_addline(buffer, "PARAM srgb_consts1 = {%f, %f, %f, %f};\n",
3383                        srgb_mul_low, srgb_cmp, srgb_pow, srgb_mul_high);
3384         shader_addline(buffer, "PARAM srgb_consts2 = {%f, %f, %f, %f};\n",
3385                        srgb_sub_high, 0.0, 0.0, 0.0);
3386     }
3387
3388     /* Base Declarations */
3389     next_local = shader_generate_arb_declarations( (IWineD3DBaseShader*) This, reg_maps, buffer, &GLINFO_LOCATION,
3390             lconst_map, NULL, &priv_ctx);
3391
3392     for (i = 0, map = reg_maps->bumpmat; map; map >>= 1, ++i)
3393     {
3394         if (!(map & 1)) continue;
3395
3396         cur = compiled->numbumpenvmatconsts;
3397         compiled->bumpenvmatconst[cur].const_num = WINED3D_CONST_NUM_UNUSED;
3398         compiled->bumpenvmatconst[cur].texunit = i;
3399         compiled->luminanceconst[cur].const_num = WINED3D_CONST_NUM_UNUSED;
3400         compiled->luminanceconst[cur].texunit = i;
3401
3402         /* We can fit the constants into the constant limit for sure because texbem, texbeml, bem and beml are only supported
3403          * in 1.x shaders, and GL_ARB_fragment_program has a constant limit of 24 constants. So in the worst case we're loading
3404          * 8 shader constants, 8 bump matrices and 8 luminance parameters and are perfectly fine. (No NP2 fixup on bumpmapped
3405          * textures due to conditional NP2 restrictions)
3406          *
3407          * Use local constants to load the bump env parameters, not program.env. This avoids collisions with d3d constants of
3408          * shaders in newer shader models. Since the bump env parameters have to share their space with NP2 fixup constants,
3409          * their location is shader dependent anyway and they cannot be loaded globally.
3410          */
3411         compiled->bumpenvmatconst[cur].const_num = next_local++;
3412         shader_addline(buffer, "PARAM bumpenvmat%d = program.local[%d];\n",
3413                        i, compiled->bumpenvmatconst[cur].const_num);
3414         compiled->numbumpenvmatconsts = cur + 1;
3415
3416         if (!(reg_maps->luminanceparams & (1 << i))) continue;
3417
3418         compiled->luminanceconst[cur].const_num = next_local++;
3419         shader_addline(buffer, "PARAM luminance%d = program.local[%d];\n",
3420                        i, compiled->luminanceconst[cur].const_num);
3421     }
3422
3423     for(i = 0; i < MAX_CONST_I; i++)
3424     {
3425         compiled->int_consts[i] = WINED3D_CONST_NUM_UNUSED;
3426         if (reg_maps->integer_constants & (1 << i) && priv_ctx.target_version >= NV2)
3427         {
3428             const DWORD *control_values = find_loop_control_values((IWineD3DBaseShaderImpl *) This, i);
3429
3430             if(control_values)
3431             {
3432                 shader_addline(buffer, "PARAM I%u = {%u, %u, %u, -1};\n", i,
3433                                 control_values[0], control_values[1], control_values[2]);
3434             }
3435             else
3436             {
3437                 compiled->int_consts[i] = next_local;
3438                 compiled->num_int_consts++;
3439                 shader_addline(buffer, "PARAM I%u = program.local[%u];\n", i, next_local++);
3440             }
3441         }
3442     }
3443
3444     if(reg_maps->vpos || reg_maps->usesdsy)
3445     {
3446         compiled->ycorrection = next_local;
3447         shader_addline(buffer, "PARAM ycorrection = program.local[%u];\n", next_local++);
3448
3449         if(reg_maps->vpos)
3450         {
3451             shader_addline(buffer, "TEMP vpos;\n");
3452             /* ycorrection.x: Backbuffer height(onscreen) or 0(offscreen).
3453              * ycorrection.y: -1.0(onscreen), 1.0(offscreen)
3454              * ycorrection.z: 1.0
3455              * ycorrection.w: 0.0
3456              */
3457             shader_addline(buffer, "MAD vpos, fragment.position, ycorrection.zyww, ycorrection.wxww;\n");
3458             shader_addline(buffer, "FLR vpos.xy, vpos;\n");
3459         }
3460     }
3461     else
3462     {
3463         compiled->ycorrection = WINED3D_CONST_NUM_UNUSED;
3464     }
3465
3466     /* Load constants to fixup NP2 texcoords if there are still free constants left:
3467      * Constants (texture dimensions) for the NP2 fixup are loaded as local program parameters. This will consume
3468      * at most 8 (MAX_FRAGMENT_SAMPLERS / 2) parameters, which is highly unlikely, since the application had to
3469      * use 16 NP2 textures at the same time. In case that we run out of constants the fixup is simply not
3470      * applied / activated. This will probably result in wrong rendering of the texture, but will save us from
3471      * shader compilation errors and the subsequent errors when drawing with this shader. */
3472     if (priv_ctx.cur_ps_args->super.np2_fixup) {
3473
3474         struct arb_ps_np2fixup_info* const fixup = priv_ctx.cur_np2fixup_info;
3475         const WORD map = priv_ctx.cur_ps_args->super.np2_fixup;
3476         const UINT max_lconsts = gl_info->ps_arb_max_local_constants;
3477
3478         fixup->offset = next_local;
3479         fixup->super.active = 0;
3480
3481         cur = 0;
3482         for (i = 0; i < MAX_FRAGMENT_SAMPLERS; ++i) {
3483             if (!(map & (1 << i))) continue;
3484
3485             if (fixup->offset + (cur >> 1) < max_lconsts) {
3486                 fixup->super.active |= (1 << i);
3487                 fixup->super.idx[i] = cur++;
3488             } else {
3489                 FIXME("No free constant found to load NP2 fixup data into shader. "
3490                       "Sampling from this texture will probably look wrong.\n");
3491                 break;
3492             }
3493         }
3494
3495         fixup->super.num_consts = (cur + 1) >> 1;
3496         if (fixup->super.num_consts) {
3497             shader_addline(buffer, "PARAM np2fixup[%u] = { program.env[%u..%u] };\n",
3498                            fixup->super.num_consts, fixup->offset, fixup->super.num_consts + fixup->offset - 1);
3499         }
3500
3501         next_local += fixup->super.num_consts;
3502     }
3503
3504     if (shader_priv->clipplane_emulation != ~0U && args->clip)
3505     {
3506         shader_addline(buffer, "KIL fragment.texcoord[%u];\n", shader_priv->clipplane_emulation);
3507     }
3508
3509     /* Base Shader Body */
3510     shader_generate_main((IWineD3DBaseShader *)This, buffer, reg_maps, function, &priv_ctx);
3511
3512     if(args->super.srgb_correction) {
3513         arbfp_add_sRGB_correction(buffer, fragcolor, srgbtmp[0], srgbtmp[1], srgbtmp[2], srgbtmp[3],
3514                                   priv_ctx.target_version >= NV2);
3515     } else if(reg_maps->shader_version.major < 2) {
3516         shader_addline(buffer, "MOV result.color, %s;\n", fragcolor);
3517     }
3518     shader_addline(buffer, "END\n");
3519
3520     /* TODO: change to resource.glObjectHandle or something like that */
3521     GL_EXTCALL(glGenProgramsARB(1, &retval));
3522
3523     TRACE("Creating a hw pixel shader, prg=%d\n", retval);
3524     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, retval));
3525
3526     TRACE("Created hw pixel shader, prg=%d\n", retval);
3527     /* Create the program and check for errors */
3528     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
3529                buffer->bsize, buffer->buffer));
3530     checkGLcall("glProgramStringARB()");
3531
3532     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &errPos);
3533     if (errPos != -1)
3534     {
3535         FIXME("HW PixelShader Error at position %d: %s\n",
3536               errPos, debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
3537         retval = 0;
3538     }
3539     else
3540     {
3541         GLint native;
3542
3543         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
3544         checkGLcall("glGetProgramivARB()");
3545         if (!native) WARN("Program exceeds native resource limits.\n");
3546     }
3547
3548     /* Load immediate constants */
3549     if(lconst_map) {
3550         LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry) {
3551             const float *value = (const float *)lconst->value;
3552             GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, lconst_map[lconst->idx], value));
3553             checkGLcall("glProgramLocalParameter4fvARB");
3554         }
3555         HeapFree(GetProcessHeap(), 0, lconst_map);
3556     }
3557
3558     return retval;
3559 }
3560
3561 static int compare_sig(const struct wined3d_shader_signature_element *sig1, const struct wined3d_shader_signature_element *sig2)
3562 {
3563     unsigned int i;
3564     int ret;
3565
3566     for(i = 0; i < MAX_REG_INPUT; i++)
3567     {
3568         if(sig1[i].semantic_name == NULL || sig2[i].semantic_name == NULL)
3569         {
3570             /* Compare pointers, not contents. One string is NULL(element does not exist), the other one is not NULL */
3571             if(sig1[i].semantic_name != sig2[i].semantic_name) return sig1[i].semantic_name < sig2[i].semantic_name ? -1 : 1;
3572             continue;
3573         }
3574
3575         ret = strcmp(sig1[i].semantic_name, sig2[i].semantic_name);
3576         if(ret != 0) return ret;
3577         if(sig1[i].semantic_idx    != sig2[i].semantic_idx)    return sig1[i].semantic_idx    < sig2[i].semantic_idx    ? -1 : 1;
3578         if(sig1[i].sysval_semantic != sig2[i].sysval_semantic) return sig1[i].sysval_semantic < sig2[i].sysval_semantic ? -1 : 1;
3579         if(sig1[i].component_type  != sig2[i].component_type)  return sig1[i].sysval_semantic < sig2[i].component_type  ? -1 : 1;
3580         if(sig1[i].register_idx    != sig2[i].register_idx)    return sig1[i].register_idx    < sig2[i].register_idx    ? -1 : 1;
3581         if(sig1[i].mask            != sig2->mask)              return sig1[i].mask            < sig2[i].mask            ? -1 : 1;
3582     }
3583     return 0;
3584 }
3585
3586 static struct wined3d_shader_signature_element *clone_sig(const struct wined3d_shader_signature_element *sig)
3587 {
3588     struct wined3d_shader_signature_element *new;
3589     int i;
3590     char *name;
3591
3592     new = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*new) * MAX_REG_INPUT);
3593     for(i = 0; i < MAX_REG_INPUT; i++)
3594     {
3595         if(sig[i].semantic_name == NULL)
3596         {
3597             continue;
3598         }
3599
3600         new[i] = sig[i];
3601         /* Clone the semantic string */
3602         name = HeapAlloc(GetProcessHeap(), 0, strlen(sig[i].semantic_name) + 1);
3603         strcpy(name, sig[i].semantic_name);
3604         new[i].semantic_name = name;
3605     }
3606     return new;
3607 }
3608
3609 static DWORD find_input_signature(struct shader_arb_priv *priv, const struct wined3d_shader_signature_element *sig)
3610 {
3611     struct wine_rb_entry *entry = wine_rb_get(&priv->signature_tree, sig);
3612     struct ps_signature *found_sig;
3613
3614     if(entry != NULL)
3615     {
3616         found_sig = WINE_RB_ENTRY_VALUE(entry, struct ps_signature, entry);
3617         TRACE("Found existing signature %u\n", found_sig->idx);
3618         return found_sig->idx;
3619     }
3620     found_sig = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*sig));
3621     found_sig->sig = clone_sig(sig);
3622     found_sig->idx = priv->ps_sig_number++;
3623     TRACE("New signature stored and assigned number %u\n", found_sig->idx);
3624     if(wine_rb_put(&priv->signature_tree, sig, &found_sig->entry) == -1)
3625     {
3626         ERR("Failed to insert program entry.\n");
3627     }
3628     return found_sig->idx;
3629 }
3630
3631 static void init_output_registers(IWineD3DVertexShaderImpl *shader, DWORD sig_num, struct shader_arb_ctx_priv *priv_ctx,
3632                                   struct arb_vs_compiled_shader *compiled)
3633 {
3634     unsigned int i, j;
3635     static const char *texcoords[8] =
3636     {
3637         "result.texcoord[0]", "result.texcoord[1]", "result.texcoord[2]", "result.texcoord[3]",
3638         "result.texcoord[4]", "result.texcoord[5]", "result.texcoord[6]", "result.texcoord[7]"
3639     };
3640     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *) shader->baseShader.device;
3641     const struct wined3d_shader_signature_element *sig;
3642     const char *semantic_name;
3643     DWORD semantic_idx, reg_idx;
3644
3645     /* Write generic input varyings 0 to 7 to result.texcoord[], varying 8 to result.color.primary
3646      * and varying 9 to result.color.secondary
3647      */
3648     const char *decl_idx_to_string[MAX_REG_INPUT] =
3649     {
3650         texcoords[0], texcoords[1], texcoords[2], texcoords[3],
3651         texcoords[4], texcoords[5], texcoords[6], texcoords[7],
3652         "result.color.primary", "result.color.secondary"
3653     };
3654
3655     if(sig_num == ~0)
3656     {
3657         TRACE("Pixel shader uses builtin varyings\n");
3658         /* Map builtins to builtins */
3659         for(i = 0; i < 8; i++)
3660         {
3661             priv_ctx->texcrd_output[i] = texcoords[i];
3662         }
3663         priv_ctx->color_output[0] = "result.color.primary";
3664         priv_ctx->color_output[1] = "result.color.secondary";
3665         priv_ctx->fog_output = "result.fogcoord";
3666
3667         /* Map declared regs to builtins. Use "TA" to /dev/null unread output */
3668         for(i = 0; i < (sizeof(shader->output_signature) / sizeof(*shader->output_signature)); i++)
3669         {
3670             semantic_name = shader->output_signature[i].semantic_name;
3671             if(semantic_name == NULL) continue;
3672
3673             if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_POSITION))
3674             {
3675                 TRACE("o%u is TMP_OUT\n", i);
3676                 if(shader->output_signature[i].semantic_idx == 0) priv_ctx->vs_output[i] = "TMP_OUT";
3677                 else priv_ctx->vs_output[i] = "TA";
3678             }
3679             else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_PSIZE))
3680             {
3681                 TRACE("o%u is result.pointsize\n", i);
3682                 if(shader->output_signature[i].semantic_idx == 0) priv_ctx->vs_output[i] = "result.pointsize";
3683                 else priv_ctx->vs_output[i] = "TA";
3684             }
3685             else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_COLOR))
3686             {
3687                 TRACE("o%u is result.color.?, idx %u\n", i, shader->output_signature[i].semantic_idx);
3688                 if(shader->output_signature[i].semantic_idx == 0) priv_ctx->vs_output[i] = "result.color.primary";
3689                 else if(shader->output_signature[i].semantic_idx == 1) priv_ctx->vs_output[i] = "result.color.secondary";
3690                 else priv_ctx->vs_output[i] = "TA";
3691             }
3692             else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_TEXCOORD))
3693             {
3694                 TRACE("o%u is %s\n", i, texcoords[shader->output_signature[i].semantic_idx]);
3695                 if(shader->output_signature[i].semantic_idx >= 8) priv_ctx->vs_output[i] = "TA";
3696                 else priv_ctx->vs_output[i] = texcoords[shader->output_signature[i].semantic_idx];
3697             }
3698             else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_FOG))
3699             {
3700                 TRACE("o%u is result.fogcoord\n", i);
3701                 if(shader->output_signature[i].semantic_idx > 0) priv_ctx->vs_output[i] = "TA";
3702                 else priv_ctx->vs_output[i] = "result.fogcoord";
3703             }
3704             else
3705             {
3706                 priv_ctx->vs_output[i] = "TA";
3707             }
3708         }
3709         return;
3710     }
3711
3712     /* Instead of searching for the signature in the signature list, read the one from the current pixel shader.
3713      * Its maybe not the shader where the signature came from, but it is the same signature and faster to find
3714      */
3715     sig = ((IWineD3DPixelShaderImpl *)device->stateBlock->pixelShader)->input_signature;
3716     TRACE("Pixel shader uses declared varyings\n");
3717
3718     /* Map builtin to declared. /dev/null the results by default to the TA temp reg */
3719     for(i = 0; i < 8; i++)
3720     {
3721         priv_ctx->texcrd_output[i] = "TA";
3722     }
3723     priv_ctx->color_output[0] = "TA";
3724     priv_ctx->color_output[1] = "TA";
3725     priv_ctx->fog_output = "TA";
3726
3727     for(i = 0; i < MAX_REG_INPUT; i++)
3728     {
3729         semantic_name = sig[i].semantic_name;
3730         semantic_idx = sig[i].semantic_idx;
3731         reg_idx = sig[i].register_idx;
3732         if(semantic_name == NULL) continue;
3733
3734         /* If a declared input register is not written by builtin arguments, don't write to it.
3735          * GL_NV_vertex_program makes sure the input defaults to 0.0, which is correct with D3D
3736          *
3737          * Don't care about POSITION and PSIZE here - this is a builtin vertex shader, position goes
3738          * to TMP_OUT in any case
3739          */
3740         if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_TEXCOORD))
3741         {
3742             if(semantic_idx < 8) priv_ctx->texcrd_output[semantic_idx] = decl_idx_to_string[reg_idx];
3743         }
3744         else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_COLOR))
3745         {
3746             if(semantic_idx < 2) priv_ctx->color_output[semantic_idx] = decl_idx_to_string[reg_idx];
3747         }
3748         else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_FOG))
3749         {
3750             if(semantic_idx == 0) priv_ctx->fog_output = decl_idx_to_string[reg_idx];
3751         }
3752         else
3753         {
3754             continue;
3755         }
3756
3757         if(strcmp(decl_idx_to_string[reg_idx], "result.color.primary") == 0 ||
3758            strcmp(decl_idx_to_string[reg_idx], "result.color.secondary") == 0)
3759         {
3760             compiled->need_color_unclamp = TRUE;
3761         }
3762     }
3763
3764     /* Map declared to declared */
3765     for(i = 0; i < (sizeof(shader->output_signature) / sizeof(*shader->output_signature)); i++)
3766     {
3767         /* Write unread output to TA to throw them away */
3768         priv_ctx->vs_output[i] = "TA";
3769         semantic_name = shader->output_signature[i].semantic_name;
3770         if(semantic_name == NULL)
3771         {
3772             continue;
3773         }
3774
3775         if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_POSITION) &&
3776            shader->output_signature[i].semantic_idx == 0)
3777         {
3778             priv_ctx->vs_output[i] = "TMP_OUT";
3779             continue;
3780         }
3781         else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_PSIZE) &&
3782            shader->output_signature[i].semantic_idx == 0)
3783         {
3784             priv_ctx->vs_output[i] = "result.pointsize";
3785             continue;
3786         }
3787
3788         for(j = 0; j < MAX_REG_INPUT; j++)
3789         {
3790             if(sig[j].semantic_name == NULL)
3791             {
3792                 continue;
3793             }
3794
3795             if(strcmp(sig[j].semantic_name, semantic_name) == 0 &&
3796                sig[j].semantic_idx == shader->output_signature[i].semantic_idx)
3797             {
3798                 priv_ctx->vs_output[i] = decl_idx_to_string[sig[j].register_idx];
3799
3800                 if(strcmp(priv_ctx->vs_output[i], "result.color.primary") == 0 ||
3801                    strcmp(priv_ctx->vs_output[i], "result.color.secondary") == 0)
3802                 {
3803                     compiled->need_color_unclamp = TRUE;
3804                 }
3805             }
3806         }
3807     }
3808 }
3809
3810 /* GL locking is done by the caller */
3811 static GLuint shader_arb_generate_vshader(IWineD3DVertexShaderImpl *This, struct wined3d_shader_buffer *buffer,
3812         const struct arb_vs_compile_args *args, struct arb_vs_compiled_shader *compiled)
3813 {
3814     const shader_reg_maps *reg_maps = &This->baseShader.reg_maps;
3815     CONST DWORD *function = This->baseShader.function;
3816     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *)This->baseShader.device;
3817     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
3818     const local_constant *lconst;
3819     GLuint ret;
3820     DWORD next_local, *lconst_map = local_const_mapping((IWineD3DBaseShaderImpl *) This);
3821     struct shader_arb_ctx_priv priv_ctx;
3822     unsigned int i;
3823     GLint errPos;
3824
3825     memset(&priv_ctx, 0, sizeof(priv_ctx));
3826     priv_ctx.cur_vs_args = args;
3827     list_init(&priv_ctx.control_frames);
3828     init_output_registers(This, args->ps_signature, &priv_ctx, compiled);
3829
3830     /*  Create the hw ARB shader */
3831     shader_addline(buffer, "!!ARBvp1.0\n");
3832
3833     /* Always enable the NV extension if available. Unlike fragment shaders, there is no
3834      * mesurable performance penalty, and we can always make use of it for clipplanes.
3835      */
3836     if(GL_SUPPORT(NV_VERTEX_PROGRAM3)) {
3837         shader_addline(buffer, "OPTION NV_vertex_program3;\n");
3838         priv_ctx.target_version = NV3;
3839         shader_addline(buffer, "ADDRESS aL;\n");
3840     } else if(GL_SUPPORT(NV_VERTEX_PROGRAM2_OPTION)) {
3841         shader_addline(buffer, "OPTION NV_vertex_program2;\n");
3842         priv_ctx.target_version = NV2;
3843         shader_addline(buffer, "ADDRESS aL;\n");
3844     } else {
3845         priv_ctx.target_version = ARB;
3846     }
3847
3848     shader_addline(buffer, "TEMP TMP_OUT;\n");
3849     if(need_helper_const(gl_info)) {
3850         shader_addline(buffer, "PARAM helper_const = { 2.0, -1.0, %d.0, 0.0 };\n", This->rel_offset);
3851     }
3852     if(need_mova_const((IWineD3DBaseShader *) This, gl_info)) {
3853         shader_addline(buffer, "PARAM mova_const = { 0.5, 0.0, 2.0, 1.0 };\n");
3854         shader_addline(buffer, "TEMP A0_SHADOW;\n");
3855     }
3856
3857     shader_addline(buffer, "TEMP TA;\n");
3858
3859     /* Base Declarations */
3860     next_local = shader_generate_arb_declarations( (IWineD3DBaseShader*) This, reg_maps, buffer, &GLINFO_LOCATION,
3861             lconst_map, &priv_ctx.vs_clipplanes, &priv_ctx);
3862
3863     for(i = 0; i < MAX_CONST_I; i++)
3864     {
3865         compiled->int_consts[i] = WINED3D_CONST_NUM_UNUSED;
3866         if(reg_maps->integer_constants & (1 << i) && priv_ctx.target_version >= NV2)
3867         {
3868             const DWORD *control_values = find_loop_control_values((IWineD3DBaseShaderImpl *) This, i);
3869
3870             if(control_values)
3871             {
3872                 shader_addline(buffer, "PARAM I%u = {%u, %u, %u, -1};\n", i,
3873                                 control_values[0], control_values[1], control_values[2]);
3874             }
3875             else
3876             {
3877                 compiled->int_consts[i] = next_local;
3878                 compiled->num_int_consts++;
3879                 shader_addline(buffer, "PARAM I%u = program.local[%u];\n", i, next_local++);
3880             }
3881         }
3882     }
3883
3884     /* We need a constant to fixup the final position */
3885     shader_addline(buffer, "PARAM posFixup = program.local[%u];\n", next_local);
3886     compiled->pos_fixup = next_local++;
3887
3888     /* Initialize output parameters. GL_ARB_vertex_program does not require special initialization values
3889      * for output parameters. D3D in theory does not do that either, but some applications depend on a
3890      * proper initialization of the secondary color, and programs using the fixed function pipeline without
3891      * a replacement shader depend on the texcoord.w being set properly.
3892      *
3893      * GL_NV_vertex_program defines that all output values are initialized to {0.0, 0.0, 0.0, 1.0}. This
3894      * assertion is in effect even when using GL_ARB_vertex_program without any NV specific additions. So
3895      * skip this if NV_vertex_program is supported. Otherwise, initialize the secondary color. For the tex-
3896      * coords, we have a flag in the opengl caps. Many cards do not require the texcoord being set, and
3897      * this can eat a number of instructions, so skip it unless this cap is set as well
3898      */
3899     if(!GL_SUPPORT(NV_VERTEX_PROGRAM)) {
3900         shader_addline(buffer, "MOV result.color.secondary, -helper_const.wwwy;\n");
3901
3902         if ((GLINFO_LOCATION).quirks & WINED3D_QUIRK_SET_TEXCOORD_W && !device->frag_pipe->ffp_proj_control)
3903         {
3904             int i;
3905             for(i = 0; i < min(8, MAX_REG_TEXCRD); i++) {
3906                 if(This->baseShader.reg_maps.texcoord_mask[i] != 0 &&
3907                 This->baseShader.reg_maps.texcoord_mask[i] != WINED3DSP_WRITEMASK_ALL) {
3908                     shader_addline(buffer, "MOV result.texcoord[%u].w, -helper_const.y;\n", i);
3909                 }
3910             }
3911         }
3912     }
3913
3914     /* The shader starts with the main function */
3915     priv_ctx.in_main_func = TRUE;
3916     /* Base Shader Body */
3917     shader_generate_main((IWineD3DBaseShader *)This, buffer, reg_maps, function, &priv_ctx);
3918
3919     if(!priv_ctx.footer_written) vshader_add_footer(This, buffer, args, &priv_ctx);
3920
3921     shader_addline(buffer, "END\n");
3922
3923     /* TODO: change to resource.glObjectHandle or something like that */
3924     GL_EXTCALL(glGenProgramsARB(1, &ret));
3925
3926     TRACE("Creating a hw vertex shader, prg=%d\n", ret);
3927     GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, ret));
3928
3929     TRACE("Created hw vertex shader, prg=%d\n", ret);
3930     /* Create the program and check for errors */
3931     GL_EXTCALL(glProgramStringARB(GL_VERTEX_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
3932                buffer->bsize, buffer->buffer));
3933     checkGLcall("glProgramStringARB()");
3934
3935     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &errPos);
3936     if (errPos != -1)
3937     {
3938         FIXME("HW VertexShader Error at position %d: %s\n",
3939               errPos, debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
3940         ret = -1;
3941     }
3942     else
3943     {
3944         GLint native;
3945
3946         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
3947         checkGLcall("glGetProgramivARB()");
3948         if (!native) WARN("Program exceeds native resource limits.\n");
3949
3950         /* Load immediate constants */
3951         if(lconst_map) {
3952             LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry) {
3953                 const float *value = (const float *)lconst->value;
3954                 GL_EXTCALL(glProgramLocalParameter4fvARB(GL_VERTEX_PROGRAM_ARB, lconst_map[lconst->idx], value));
3955             }
3956         }
3957     }
3958     HeapFree(GetProcessHeap(), 0, lconst_map);
3959
3960     return ret;
3961 }
3962
3963 /* GL locking is done by the caller */
3964 static struct arb_ps_compiled_shader *find_arb_pshader(IWineD3DPixelShaderImpl *shader, const struct arb_ps_compile_args *args)
3965 {
3966     UINT i;
3967     DWORD new_size;
3968     struct arb_ps_compiled_shader *new_array;
3969     struct wined3d_shader_buffer buffer;
3970     struct arb_pshader_private *shader_data;
3971     GLuint ret;
3972
3973     if(!shader->backend_priv) {
3974         IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *) shader->baseShader.device;
3975         const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
3976         struct shader_arb_priv *priv = device->shader_priv;
3977
3978         shader->backend_priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data));
3979         shader_data = shader->backend_priv;
3980         shader_data->clamp_consts = shader->baseShader.reg_maps.shader_version.major == 1;
3981
3982         if(shader->baseShader.reg_maps.shader_version.major < 3) shader_data->input_signature_idx = ~0;
3983         else shader_data->input_signature_idx = find_input_signature(priv, shader->input_signature);
3984
3985         shader_data->has_signature_idx = TRUE;
3986         TRACE("Shader got assigned input signature index %u\n", shader_data->input_signature_idx);
3987
3988         if (!device->vs_clipping)
3989             shader_data->clipplane_emulation = shader_find_free_input_register(&shader->baseShader.reg_maps,
3990                     GL_LIMITS(texture_stages) - 1);
3991         else
3992             shader_data->clipplane_emulation = ~0U;
3993     }
3994     shader_data = shader->backend_priv;
3995
3996     /* Usually we have very few GL shaders for each d3d shader(just 1 or maybe 2),
3997      * so a linear search is more performant than a hashmap or a binary search
3998      * (cache coherency etc)
3999      */
4000     for(i = 0; i < shader_data->num_gl_shaders; i++) {
4001         if(memcmp(&shader_data->gl_shaders[i].args, args, sizeof(*args)) == 0) {
4002             return &shader_data->gl_shaders[i];
4003         }
4004     }
4005
4006     TRACE("No matching GL shader found, compiling a new shader\n");
4007     if(shader_data->shader_array_size == shader_data->num_gl_shaders) {
4008         if (shader_data->num_gl_shaders)
4009         {
4010             new_size = shader_data->shader_array_size + max(1, shader_data->shader_array_size / 2);
4011             new_array = HeapReAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, shader_data->gl_shaders,
4012                                     new_size * sizeof(*shader_data->gl_shaders));
4013         } else {
4014             new_array = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data->gl_shaders));
4015             new_size = 1;
4016         }
4017
4018         if(!new_array) {
4019             ERR("Out of memory\n");
4020             return 0;
4021         }
4022         shader_data->gl_shaders = new_array;
4023         shader_data->shader_array_size = new_size;
4024     }
4025
4026     shader_data->gl_shaders[shader_data->num_gl_shaders].args = *args;
4027
4028     pixelshader_update_samplers(&shader->baseShader.reg_maps,
4029             ((IWineD3DDeviceImpl *)shader->baseShader.device)->stateBlock->textures);
4030
4031     if (!shader_buffer_init(&buffer))
4032     {
4033         ERR("Failed to initialize shader buffer.\n");
4034         return 0;
4035     }
4036
4037     ret = shader_arb_generate_pshader(shader, &buffer, args,
4038                                       &shader_data->gl_shaders[shader_data->num_gl_shaders]);
4039     shader_buffer_free(&buffer);
4040     shader_data->gl_shaders[shader_data->num_gl_shaders].prgId = ret;
4041
4042     return &shader_data->gl_shaders[shader_data->num_gl_shaders++];
4043 }
4044
4045 static inline BOOL vs_args_equal(const struct arb_vs_compile_args *stored, const struct arb_vs_compile_args *new,
4046                                  const DWORD use_map, BOOL skip_int) {
4047     if((stored->super.swizzle_map & use_map) != new->super.swizzle_map) return FALSE;
4048     if(stored->super.fog_src != new->super.fog_src) return FALSE;
4049     if(stored->boolclip_compare != new->boolclip_compare) return FALSE;
4050     if(stored->ps_signature != new->ps_signature) return FALSE;
4051     if(stored->vertex_samplers_compare != new->vertex_samplers_compare) return FALSE;
4052     if(skip_int) return TRUE;
4053
4054     return memcmp(stored->loop_ctrl, new->loop_ctrl, sizeof(stored->loop_ctrl)) == 0;
4055 }
4056
4057 static struct arb_vs_compiled_shader *find_arb_vshader(IWineD3DVertexShaderImpl *shader, const struct arb_vs_compile_args *args)
4058 {
4059     UINT i;
4060     DWORD new_size;
4061     struct arb_vs_compiled_shader *new_array;
4062     DWORD use_map = ((IWineD3DDeviceImpl *)shader->baseShader.device)->strided_streams.use_map;
4063     struct wined3d_shader_buffer buffer;
4064     struct arb_vshader_private *shader_data;
4065     GLuint ret;
4066     const struct wined3d_gl_info *gl_info = &((IWineD3DDeviceImpl *)shader->baseShader.device)->adapter->gl_info;
4067
4068     if(!shader->backend_priv) {
4069         shader->backend_priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data));
4070     }
4071     shader_data = shader->backend_priv;
4072
4073     /* Usually we have very few GL shaders for each d3d shader(just 1 or maybe 2),
4074      * so a linear search is more performant than a hashmap or a binary search
4075      * (cache coherency etc)
4076      */
4077     for(i = 0; i < shader_data->num_gl_shaders; i++) {
4078         if(vs_args_equal(&shader_data->gl_shaders[i].args, args, use_map, GL_SUPPORT(NV_VERTEX_PROGRAM2_OPTION))) {
4079             return &shader_data->gl_shaders[i];
4080         }
4081     }
4082
4083     TRACE("No matching GL shader found, compiling a new shader\n");
4084
4085     if(shader_data->shader_array_size == shader_data->num_gl_shaders) {
4086         if (shader_data->num_gl_shaders)
4087         {
4088             new_size = shader_data->shader_array_size + max(1, shader_data->shader_array_size / 2);
4089             new_array = HeapReAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, shader_data->gl_shaders,
4090                                     new_size * sizeof(*shader_data->gl_shaders));
4091         } else {
4092             new_array = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data->gl_shaders));
4093             new_size = 1;
4094         }
4095
4096         if(!new_array) {
4097             ERR("Out of memory\n");
4098             return 0;
4099         }
4100         shader_data->gl_shaders = new_array;
4101         shader_data->shader_array_size = new_size;
4102     }
4103
4104     shader_data->gl_shaders[shader_data->num_gl_shaders].args = *args;
4105
4106     if (!shader_buffer_init(&buffer))
4107     {
4108         ERR("Failed to initialize shader buffer.\n");
4109         return 0;
4110     }
4111
4112     ret = shader_arb_generate_vshader(shader, &buffer, args,
4113             &shader_data->gl_shaders[shader_data->num_gl_shaders]);
4114     shader_buffer_free(&buffer);
4115     shader_data->gl_shaders[shader_data->num_gl_shaders].prgId = ret;
4116
4117     return &shader_data->gl_shaders[shader_data->num_gl_shaders++];
4118 }
4119
4120 static inline void find_arb_ps_compile_args(IWineD3DPixelShaderImpl *shader, IWineD3DStateBlockImpl *stateblock,
4121         struct arb_ps_compile_args *args)
4122 {
4123     int i;
4124     WORD int_skip;
4125     const struct wined3d_gl_info *gl_info = &((IWineD3DDeviceImpl *)shader->baseShader.device)->adapter->gl_info;
4126     find_ps_compile_args(shader, stateblock, &args->super);
4127
4128     /* This forces all local boolean constants to 1 to make them stateblock independent */
4129     args->bools = shader->baseShader.reg_maps.local_bool_consts;
4130
4131     for(i = 0; i < MAX_CONST_B; i++)
4132     {
4133         if(stateblock->pixelShaderConstantB[i]) args->bools |= ( 1 << i);
4134     }
4135
4136     /* Only enable the clip plane emulation KIL if at least one clipplane is enabled. The KIL instruction
4137      * is quite expensive because it forces the driver to disable early Z discards. It is cheaper to
4138      * duplicate the shader than have a no-op KIL instruction in every shader
4139      */
4140     if((!((IWineD3DDeviceImpl *) shader->baseShader.device)->vs_clipping) && use_vs(stateblock) &&
4141        stateblock->renderState[WINED3DRS_CLIPPING] && stateblock->renderState[WINED3DRS_CLIPPLANEENABLE])
4142     {
4143         args->clip = 1;
4144     }
4145     else
4146     {
4147         args->clip = 0;
4148     }
4149
4150     /* Skip if unused or local, or supported natively */
4151     int_skip = ~shader->baseShader.reg_maps.integer_constants | shader->baseShader.reg_maps.local_int_consts;
4152     if(int_skip == 0xffff || GL_SUPPORT(NV_FRAGMENT_PROGRAM_OPTION))
4153     {
4154         memset(&args->loop_ctrl, 0, sizeof(args->loop_ctrl));
4155         return;
4156     }
4157
4158     for(i = 0; i < MAX_CONST_I; i++)
4159     {
4160         if(int_skip & (1 << i))
4161         {
4162             args->loop_ctrl[i][0] = 0;
4163             args->loop_ctrl[i][1] = 0;
4164             args->loop_ctrl[i][2] = 0;
4165         }
4166         else
4167         {
4168             args->loop_ctrl[i][0] = stateblock->pixelShaderConstantI[i * 4];
4169             args->loop_ctrl[i][1] = stateblock->pixelShaderConstantI[i * 4 + 1];
4170             args->loop_ctrl[i][2] = stateblock->pixelShaderConstantI[i * 4 + 2];
4171         }
4172     }
4173 }
4174
4175 static inline void find_arb_vs_compile_args(IWineD3DVertexShaderImpl *shader, IWineD3DStateBlockImpl *stateblock,
4176         struct arb_vs_compile_args *args)
4177 {
4178     int i;
4179     WORD int_skip;
4180     IWineD3DDeviceImpl *dev = (IWineD3DDeviceImpl *)shader->baseShader.device;
4181     const struct wined3d_gl_info *gl_info = &dev->adapter->gl_info;
4182     find_vs_compile_args(shader, stateblock, &args->super);
4183
4184     args->boolclip_compare = 0;
4185     if(use_ps(stateblock))
4186     {
4187         IWineD3DPixelShaderImpl *ps = (IWineD3DPixelShaderImpl *) stateblock->pixelShader;
4188         struct arb_pshader_private *shader_priv = ps->backend_priv;
4189         args->ps_signature = shader_priv->input_signature_idx;
4190
4191         args->boolclip.clip_control[0] = shader_priv->clipplane_emulation + 1;
4192     }
4193     else
4194     {
4195         args->ps_signature = ~0;
4196         if(!dev->vs_clipping)
4197         {
4198             args->boolclip.clip_control[0] = ffp_clip_emul(stateblock) ? GL_LIMITS(texture_stages) : 0;
4199         }
4200         /* Otherwise: Setting boolclip_compare set clip_control[0] to 0 */
4201     }
4202
4203     if(args->boolclip.clip_control[0])
4204     {
4205         if(stateblock->renderState[WINED3DRS_CLIPPING])
4206         {
4207             args->boolclip.clip_control[1] = stateblock->renderState[WINED3DRS_CLIPPLANEENABLE];
4208         }
4209         /* clip_control[1] was set to 0 by setting boolclip_compare to 0 */
4210     }
4211
4212     /* This forces all local boolean constants to 1 to make them stateblock independent */
4213     args->boolclip.bools = shader->baseShader.reg_maps.local_bool_consts;
4214     /* TODO: Figure out if it would be better to store bool constants as bitmasks in the stateblock */
4215     for(i = 0; i < MAX_CONST_B; i++)
4216     {
4217         if(stateblock->vertexShaderConstantB[i]) args->boolclip.bools |= ( 1 << i);
4218     }
4219
4220     args->vertex_samplers[0] = dev->texUnitMap[MAX_FRAGMENT_SAMPLERS + 0];
4221     args->vertex_samplers[1] = dev->texUnitMap[MAX_FRAGMENT_SAMPLERS + 1];
4222     args->vertex_samplers[2] = dev->texUnitMap[MAX_FRAGMENT_SAMPLERS + 2];
4223     args->vertex_samplers[3] = 0;
4224
4225     /* Skip if unused or local */
4226     int_skip = ~shader->baseShader.reg_maps.integer_constants | shader->baseShader.reg_maps.local_int_consts;
4227     if(int_skip == 0xffff || GL_SUPPORT(NV_VERTEX_PROGRAM2_OPTION)) /* This is about flow control, not clipping */
4228     {
4229         memset(&args->loop_ctrl, 0, sizeof(args->loop_ctrl));
4230         return;
4231     }
4232
4233     for(i = 0; i < MAX_CONST_I; i++)
4234     {
4235         if(int_skip & (1 << i))
4236         {
4237             args->loop_ctrl[i][0] = 0;
4238             args->loop_ctrl[i][1] = 0;
4239             args->loop_ctrl[i][2] = 0;
4240         }
4241         else
4242         {
4243             args->loop_ctrl[i][0] = stateblock->vertexShaderConstantI[i * 4];
4244             args->loop_ctrl[i][1] = stateblock->vertexShaderConstantI[i * 4 + 1];
4245             args->loop_ctrl[i][2] = stateblock->vertexShaderConstantI[i * 4 + 2];
4246         }
4247     }
4248 }
4249
4250 /* GL locking is done by the caller */
4251 static void shader_arb_select(const struct wined3d_context *context, BOOL usePS, BOOL useVS)
4252 {
4253     IWineD3DDeviceImpl *This = ((IWineD3DSurfaceImpl *)context->surface)->resource.wineD3DDevice;
4254     struct shader_arb_priv *priv = This->shader_priv;
4255     const struct wined3d_gl_info *gl_info = context->gl_info;
4256     int i;
4257
4258     /* Deal with pixel shaders first so the vertex shader arg function has the input signature ready */
4259     if (usePS) {
4260         struct arb_ps_compile_args compile_args;
4261         struct arb_ps_compiled_shader *compiled;
4262         IWineD3DPixelShaderImpl *ps = (IWineD3DPixelShaderImpl *) This->stateBlock->pixelShader;
4263
4264         TRACE("Using pixel shader %p\n", This->stateBlock->pixelShader);
4265         find_arb_ps_compile_args(ps, This->stateBlock, &compile_args);
4266         compiled = find_arb_pshader(ps, &compile_args);
4267         priv->current_fprogram_id = compiled->prgId;
4268         priv->compiled_fprog = compiled;
4269
4270         /* Bind the fragment program */
4271         GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, priv->current_fprogram_id));
4272         checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, priv->current_fprogram_id);");
4273
4274         if(!priv->use_arbfp_fixed_func) {
4275             /* Enable OpenGL fragment programs */
4276             glEnable(GL_FRAGMENT_PROGRAM_ARB);
4277             checkGLcall("glEnable(GL_FRAGMENT_PROGRAM_ARB);");
4278         }
4279         TRACE("(%p) : Bound fragment program %u and enabled GL_FRAGMENT_PROGRAM_ARB\n", This, priv->current_fprogram_id);
4280
4281         /* Pixel Shader 1.x constants are clamped to [-1;1], Pixel Shader 2.0 constants are not. If switching between
4282          * a 1.x and newer shader, reload the first 8 constants
4283          */
4284         if(priv->last_ps_const_clamped != ((struct arb_pshader_private *) ps->backend_priv)->clamp_consts)
4285         {
4286             priv->last_ps_const_clamped = ((struct arb_pshader_private *) ps->backend_priv)->clamp_consts;
4287             This->highest_dirty_ps_const = max(This->highest_dirty_ps_const, 8);
4288             for(i = 0; i < 8; i++)
4289             {
4290                 context->pshader_const_dirty[i] = 1;
4291             }
4292             /* Also takes care of loading local constants */
4293             shader_arb_load_constants(context, TRUE, FALSE);
4294         }
4295         else
4296         {
4297             shader_arb_ps_local_constants(This);
4298         }
4299
4300         /* Force constant reloading for the NP2 fixup (see comment in shader_glsl_select for more info) */
4301         if (compiled->np2fixup_info.super.active)
4302             shader_arb_load_np2fixup_constants((IWineD3DDevice *)This, usePS, useVS);
4303     } else if(GL_SUPPORT(ARB_FRAGMENT_PROGRAM) && !priv->use_arbfp_fixed_func) {
4304         /* Disable only if we're not using arbfp fixed function fragment processing. If this is used,
4305         * keep GL_FRAGMENT_PROGRAM_ARB enabled, and the fixed function pipeline will bind the fixed function
4306         * replacement shader
4307         */
4308         glDisable(GL_FRAGMENT_PROGRAM_ARB);
4309         checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
4310         priv->current_fprogram_id = 0;
4311     }
4312
4313     if (useVS) {
4314         struct arb_vs_compile_args compile_args;
4315         struct arb_vs_compiled_shader *compiled;
4316         IWineD3DVertexShaderImpl *vs = (IWineD3DVertexShaderImpl *) This->stateBlock->vertexShader;
4317
4318         TRACE("Using vertex shader %p\n", This->stateBlock->vertexShader);
4319         find_arb_vs_compile_args(vs, This->stateBlock, &compile_args);
4320         compiled = find_arb_vshader(vs, &compile_args);
4321         priv->current_vprogram_id = compiled->prgId;
4322         priv->compiled_vprog = compiled;
4323
4324         /* Bind the vertex program */
4325         GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->current_vprogram_id));
4326         checkGLcall("glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->current_vprogram_id);");
4327
4328         /* Enable OpenGL vertex programs */
4329         glEnable(GL_VERTEX_PROGRAM_ARB);
4330         checkGLcall("glEnable(GL_VERTEX_PROGRAM_ARB);");
4331         TRACE("(%p) : Bound vertex program %u and enabled GL_VERTEX_PROGRAM_ARB\n", This, priv->current_vprogram_id);
4332         shader_arb_vs_local_constants(This);
4333
4334         if(priv->last_vs_color_unclamp != compiled->need_color_unclamp) {
4335             priv->last_vs_color_unclamp = compiled->need_color_unclamp;
4336
4337             if (GL_SUPPORT(ARB_COLOR_BUFFER_FLOAT)) {
4338                 GL_EXTCALL(glClampColorARB(GL_CLAMP_VERTEX_COLOR_ARB, !compiled->need_color_unclamp));
4339                 checkGLcall("glClampColorARB");
4340             } else {
4341                 FIXME("vertex color clamp needs to be changed, but extension not supported.\n");
4342             }
4343         }
4344     } else if(GL_SUPPORT(ARB_VERTEX_PROGRAM)) {
4345         priv->current_vprogram_id = 0;
4346         glDisable(GL_VERTEX_PROGRAM_ARB);
4347         checkGLcall("glDisable(GL_VERTEX_PROGRAM_ARB)");
4348     }
4349 }
4350
4351 /* GL locking is done by the caller */
4352 static void shader_arb_select_depth_blt(IWineD3DDevice *iface, enum tex_types tex_type) {
4353     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
4354     struct shader_arb_priv *priv = This->shader_priv;
4355     GLuint *blt_fprogram = &priv->depth_blt_fprogram_id[tex_type];
4356     const struct wined3d_gl_info *gl_info = &This->adapter->gl_info;
4357
4358     if (!priv->depth_blt_vprogram_id) priv->depth_blt_vprogram_id = create_arb_blt_vertex_program(gl_info);
4359     GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->depth_blt_vprogram_id));
4360     glEnable(GL_VERTEX_PROGRAM_ARB);
4361
4362     if (!*blt_fprogram) *blt_fprogram = create_arb_blt_fragment_program(gl_info, tex_type);
4363     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, *blt_fprogram));
4364     glEnable(GL_FRAGMENT_PROGRAM_ARB);
4365 }
4366
4367 /* GL locking is done by the caller */
4368 static void shader_arb_deselect_depth_blt(IWineD3DDevice *iface) {
4369     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
4370     struct shader_arb_priv *priv = This->shader_priv;
4371     const struct wined3d_gl_info *gl_info = &This->adapter->gl_info;
4372
4373     if (priv->current_vprogram_id) {
4374         GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->current_vprogram_id));
4375         checkGLcall("glBindProgramARB(GL_VERTEX_PROGRAM_ARB, vertexShader->prgId);");
4376
4377         TRACE("(%p) : Bound vertex program %u and enabled GL_VERTEX_PROGRAM_ARB\n", This, priv->current_vprogram_id);
4378     } else {
4379         glDisable(GL_VERTEX_PROGRAM_ARB);
4380         checkGLcall("glDisable(GL_VERTEX_PROGRAM_ARB)");
4381     }
4382
4383     if (priv->current_fprogram_id) {
4384         GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, priv->current_fprogram_id));
4385         checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, pixelShader->prgId);");
4386
4387         TRACE("(%p) : Bound fragment program %u and enabled GL_FRAGMENT_PROGRAM_ARB\n", This, priv->current_fprogram_id);
4388     } else if(!priv->use_arbfp_fixed_func) {
4389         glDisable(GL_FRAGMENT_PROGRAM_ARB);
4390         checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
4391     }
4392 }
4393
4394 static void shader_arb_destroy(IWineD3DBaseShader *iface) {
4395     IWineD3DBaseShaderImpl *baseShader = (IWineD3DBaseShaderImpl *) iface;
4396     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *)baseShader->baseShader.device;
4397     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4398
4399     if (shader_is_pshader_version(baseShader->baseShader.reg_maps.shader_version.type))
4400     {
4401         IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *) iface;
4402         struct arb_pshader_private *shader_data = This->backend_priv;
4403         UINT i;
4404
4405         if(!shader_data) return; /* This can happen if a shader was never compiled */
4406         ENTER_GL();
4407
4408         if(shader_data->num_gl_shaders) ActivateContext(device, NULL, CTXUSAGE_RESOURCELOAD);
4409
4410         for(i = 0; i < shader_data->num_gl_shaders; i++) {
4411             GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId));
4412             checkGLcall("GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId))");
4413         }
4414         LEAVE_GL();
4415         HeapFree(GetProcessHeap(), 0, shader_data->gl_shaders);
4416         HeapFree(GetProcessHeap(), 0, shader_data);
4417         This->backend_priv = NULL;
4418     } else {
4419         IWineD3DVertexShaderImpl *This = (IWineD3DVertexShaderImpl *) iface;
4420         struct arb_vshader_private *shader_data = This->backend_priv;
4421         UINT i;
4422
4423         if(!shader_data) return; /* This can happen if a shader was never compiled */
4424         ENTER_GL();
4425
4426         if(shader_data->num_gl_shaders) ActivateContext(device, NULL, CTXUSAGE_RESOURCELOAD);
4427
4428         for(i = 0; i < shader_data->num_gl_shaders; i++) {
4429             GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId));
4430             checkGLcall("GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId))");
4431         }
4432         LEAVE_GL();
4433         HeapFree(GetProcessHeap(), 0, shader_data->gl_shaders);
4434         HeapFree(GetProcessHeap(), 0, shader_data);
4435         This->backend_priv = NULL;
4436     }
4437 }
4438
4439 static int sig_tree_compare(const void *key, const struct wine_rb_entry *entry)
4440 {
4441     struct ps_signature *e = WINE_RB_ENTRY_VALUE(entry, struct ps_signature, entry);
4442     return compare_sig(key, e->sig);
4443 }
4444
4445 static const struct wine_rb_functions sig_tree_functions =
4446 {
4447     wined3d_rb_alloc,
4448     wined3d_rb_realloc,
4449     wined3d_rb_free,
4450     sig_tree_compare
4451 };
4452
4453 static HRESULT shader_arb_alloc(IWineD3DDevice *iface) {
4454     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
4455     struct shader_arb_priv *priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*priv));
4456     if(wine_rb_init(&priv->signature_tree, &sig_tree_functions) == -1)
4457     {
4458         ERR("RB tree init failed\n");
4459         HeapFree(GetProcessHeap(), 0, priv);
4460         return E_OUTOFMEMORY;
4461     }
4462     This->shader_priv = priv;
4463     return WINED3D_OK;
4464 }
4465
4466 static void release_signature(struct wine_rb_entry *entry, void *context)
4467 {
4468     struct ps_signature *sig = WINE_RB_ENTRY_VALUE(entry, struct ps_signature, entry);
4469     int i;
4470     for(i = 0; i < MAX_REG_INPUT; i++)
4471     {
4472         HeapFree(GetProcessHeap(), 0, (char *) sig->sig[i].semantic_name);
4473     }
4474     HeapFree(GetProcessHeap(), 0, sig->sig);
4475     HeapFree(GetProcessHeap(), 0, sig);
4476 }
4477
4478 /* Context activation is done by the caller. */
4479 static void shader_arb_free(IWineD3DDevice *iface) {
4480     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
4481     const struct wined3d_gl_info *gl_info = &This->adapter->gl_info;
4482     struct shader_arb_priv *priv = This->shader_priv;
4483     int i;
4484
4485     ENTER_GL();
4486     if(priv->depth_blt_vprogram_id) {
4487         GL_EXTCALL(glDeleteProgramsARB(1, &priv->depth_blt_vprogram_id));
4488     }
4489     for (i = 0; i < tex_type_count; ++i) {
4490         if (priv->depth_blt_fprogram_id[i]) {
4491             GL_EXTCALL(glDeleteProgramsARB(1, &priv->depth_blt_fprogram_id[i]));
4492         }
4493     }
4494     LEAVE_GL();
4495
4496     wine_rb_destroy(&priv->signature_tree, release_signature, NULL);
4497     HeapFree(GetProcessHeap(), 0, This->shader_priv);
4498 }
4499
4500 static BOOL shader_arb_dirty_const(IWineD3DDevice *iface) {
4501     return TRUE;
4502 }
4503
4504 static void shader_arb_get_caps(WINED3DDEVTYPE devtype, const struct wined3d_gl_info *gl_info,
4505         struct shader_caps *pCaps)
4506 {
4507     /* We don't have an ARB fixed function pipeline yet, so let the none backend set its caps,
4508      * then overwrite the shader specific ones
4509      */
4510     none_shader_backend.shader_get_caps(devtype, gl_info, pCaps);
4511
4512     if(GL_SUPPORT(ARB_VERTEX_PROGRAM)) {
4513         if(GL_SUPPORT(NV_VERTEX_PROGRAM3))
4514         {
4515             pCaps->VertexShaderVersion = WINED3DVS_VERSION(3,0);
4516             TRACE_(d3d_caps)("Hardware vertex shader version 3.0 enabled (NV_VERTEX_PROGRAM3)\n");
4517         }
4518         else if(GL_LIMITS(vshader_constantsF) >= 256)
4519         {
4520             /* Shader Model 2.0 requires at least 256 vertex shader constants */
4521             pCaps->VertexShaderVersion = WINED3DVS_VERSION(2,0);
4522             TRACE_(d3d_caps)("Hardware vertex shader version 2.0 enabled (ARB_PROGRAM)\n");
4523         }
4524         else
4525         {
4526             pCaps->VertexShaderVersion = WINED3DVS_VERSION(1,1);
4527             TRACE_(d3d_caps)("Hardware vertex shader version 1.1 enabled (ARB_PROGRAM)\n");
4528         }
4529         pCaps->MaxVertexShaderConst = GL_LIMITS(vshader_constantsF);
4530     }
4531
4532     if(GL_SUPPORT(ARB_FRAGMENT_PROGRAM)) {
4533         if(GL_SUPPORT(NV_FRAGMENT_PROGRAM2))
4534         {
4535             pCaps->PixelShaderVersion    = WINED3DPS_VERSION(3,0);
4536             TRACE_(d3d_caps)("Hardware pixel shader version 3.0 enabled (NV_FRAGMENT_PROGRAM2)\n");
4537         }
4538         else if(GL_LIMITS(pshader_constantsF) >= 32)
4539         {
4540             /* Shader Model 2.0 requires at least 32 pixel shader constants */
4541             pCaps->PixelShaderVersion    = WINED3DPS_VERSION(2,0);
4542             TRACE_(d3d_caps)("Hardware pixel shader version 2.0 enabled (ARB_PROGRAM)\n");
4543         }
4544         else
4545         {
4546             pCaps->PixelShaderVersion    = WINED3DPS_VERSION(1,4);
4547             TRACE_(d3d_caps)("Hardware pixel shader version 1.4 enabled (ARB_PROGRAM)\n");
4548         }
4549         pCaps->PixelShader1xMaxValue = 8.0f;
4550         pCaps->MaxPixelShaderConst = GL_LIMITS(pshader_constantsF);
4551     }
4552
4553     pCaps->VSClipping = use_nv_clip(gl_info);
4554 }
4555
4556 static BOOL shader_arb_color_fixup_supported(struct color_fixup_desc fixup)
4557 {
4558     if (TRACE_ON(d3d_shader) && TRACE_ON(d3d))
4559     {
4560         TRACE("Checking support for color_fixup:\n");
4561         dump_color_fixup_desc(fixup);
4562     }
4563
4564     /* We support everything except YUV conversions. */
4565     if (!is_yuv_fixup(fixup))
4566     {
4567         TRACE("[OK]\n");
4568         return TRUE;
4569     }
4570
4571     TRACE("[FAILED]\n");
4572     return FALSE;
4573 }
4574
4575 static void shader_arb_add_instruction_modifiers(const struct wined3d_shader_instruction *ins) {
4576     DWORD shift;
4577     char write_mask[20], regstr[50];
4578     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
4579     BOOL is_color = FALSE;
4580     const struct wined3d_shader_dst_param *dst;
4581
4582     if (!ins->dst_count) return;
4583
4584     dst = &ins->dst[0];
4585     shift = dst->shift;
4586     if(shift == 0) return; /* Saturate alone is handled by the instructions */
4587
4588     shader_arb_get_write_mask(ins, dst, write_mask);
4589     shader_arb_get_register_name(ins, &dst->reg, regstr, &is_color);
4590
4591     /* Generate a line that does the output modifier computation
4592      * FIXME: _SAT vs shift? _SAT alone is already handled in the instructions, if this
4593      * maps problems in e.g. _d4_sat modify shader_arb_get_modifier
4594      */
4595     shader_addline(buffer, "MUL%s %s%s, %s, %s;\n", shader_arb_get_modifier(ins),
4596                    regstr, write_mask, regstr, shift_tab[shift]);
4597 }
4598
4599 static const SHADER_HANDLER shader_arb_instruction_handler_table[WINED3DSIH_TABLE_SIZE] =
4600 {
4601     /* WINED3DSIH_ABS           */ shader_hw_map2gl,
4602     /* WINED3DSIH_ADD           */ shader_hw_map2gl,
4603     /* WINED3DSIH_BEM           */ pshader_hw_bem,
4604     /* WINED3DSIH_BREAK         */ shader_hw_break,
4605     /* WINED3DSIH_BREAKC        */ shader_hw_breakc,
4606     /* WINED3DSIH_BREAKP        */ NULL,
4607     /* WINED3DSIH_CALL          */ shader_hw_call,
4608     /* WINED3DSIH_CALLNZ        */ NULL,
4609     /* WINED3DSIH_CMP           */ pshader_hw_cmp,
4610     /* WINED3DSIH_CND           */ pshader_hw_cnd,
4611     /* WINED3DSIH_CRS           */ shader_hw_map2gl,
4612     /* WINED3DSIH_DCL           */ NULL,
4613     /* WINED3DSIH_DEF           */ NULL,
4614     /* WINED3DSIH_DEFB          */ NULL,
4615     /* WINED3DSIH_DEFI          */ NULL,
4616     /* WINED3DSIH_DP2ADD        */ pshader_hw_dp2add,
4617     /* WINED3DSIH_DP3           */ shader_hw_map2gl,
4618     /* WINED3DSIH_DP4           */ shader_hw_map2gl,
4619     /* WINED3DSIH_DST           */ shader_hw_map2gl,
4620     /* WINED3DSIH_DSX           */ shader_hw_map2gl,
4621     /* WINED3DSIH_DSY           */ shader_hw_dsy,
4622     /* WINED3DSIH_ELSE          */ shader_hw_else,
4623     /* WINED3DSIH_ENDIF         */ shader_hw_endif,
4624     /* WINED3DSIH_ENDLOOP       */ shader_hw_endloop,
4625     /* WINED3DSIH_ENDREP        */ shader_hw_endrep,
4626     /* WINED3DSIH_EXP           */ shader_hw_scalar_op,
4627     /* WINED3DSIH_EXPP          */ shader_hw_scalar_op,
4628     /* WINED3DSIH_FRC           */ shader_hw_map2gl,
4629     /* WINED3DSIH_IF            */ NULL /* Hardcoded into the shader */,
4630     /* WINED3DSIH_IFC           */ shader_hw_ifc,
4631     /* WINED3DSIH_LABEL         */ shader_hw_label,
4632     /* WINED3DSIH_LIT           */ shader_hw_map2gl,
4633     /* WINED3DSIH_LOG           */ shader_hw_log_pow,
4634     /* WINED3DSIH_LOGP          */ shader_hw_log_pow,
4635     /* WINED3DSIH_LOOP          */ shader_hw_loop,
4636     /* WINED3DSIH_LRP           */ shader_hw_lrp,
4637     /* WINED3DSIH_M3x2          */ shader_hw_mnxn,
4638     /* WINED3DSIH_M3x3          */ shader_hw_mnxn,
4639     /* WINED3DSIH_M3x4          */ shader_hw_mnxn,
4640     /* WINED3DSIH_M4x3          */ shader_hw_mnxn,
4641     /* WINED3DSIH_M4x4          */ shader_hw_mnxn,
4642     /* WINED3DSIH_MAD           */ shader_hw_map2gl,
4643     /* WINED3DSIH_MAX           */ shader_hw_map2gl,
4644     /* WINED3DSIH_MIN           */ shader_hw_map2gl,
4645     /* WINED3DSIH_MOV           */ shader_hw_mov,
4646     /* WINED3DSIH_MOVA          */ shader_hw_mov,
4647     /* WINED3DSIH_MUL           */ shader_hw_map2gl,
4648     /* WINED3DSIH_NOP           */ shader_hw_nop,
4649     /* WINED3DSIH_NRM           */ shader_hw_nrm,
4650     /* WINED3DSIH_PHASE         */ NULL,
4651     /* WINED3DSIH_POW           */ shader_hw_log_pow,
4652     /* WINED3DSIH_RCP           */ shader_hw_scalar_op,
4653     /* WINED3DSIH_REP           */ shader_hw_rep,
4654     /* WINED3DSIH_RET           */ shader_hw_ret,
4655     /* WINED3DSIH_RSQ           */ shader_hw_scalar_op,
4656     /* WINED3DSIH_SETP          */ NULL,
4657     /* WINED3DSIH_SGE           */ shader_hw_map2gl,
4658     /* WINED3DSIH_SGN           */ shader_hw_sgn,
4659     /* WINED3DSIH_SINCOS        */ shader_hw_sincos,
4660     /* WINED3DSIH_SLT           */ shader_hw_map2gl,
4661     /* WINED3DSIH_SUB           */ shader_hw_map2gl,
4662     /* WINED3DSIH_TEX           */ pshader_hw_tex,
4663     /* WINED3DSIH_TEXBEM        */ pshader_hw_texbem,
4664     /* WINED3DSIH_TEXBEML       */ pshader_hw_texbem,
4665     /* WINED3DSIH_TEXCOORD      */ pshader_hw_texcoord,
4666     /* WINED3DSIH_TEXDEPTH      */ pshader_hw_texdepth,
4667     /* WINED3DSIH_TEXDP3        */ pshader_hw_texdp3,
4668     /* WINED3DSIH_TEXDP3TEX     */ pshader_hw_texdp3tex,
4669     /* WINED3DSIH_TEXKILL       */ pshader_hw_texkill,
4670     /* WINED3DSIH_TEXLDD        */ shader_hw_texldd,
4671     /* WINED3DSIH_TEXLDL        */ shader_hw_texldl,
4672     /* WINED3DSIH_TEXM3x2DEPTH  */ pshader_hw_texm3x2depth,
4673     /* WINED3DSIH_TEXM3x2PAD    */ pshader_hw_texm3x2pad,
4674     /* WINED3DSIH_TEXM3x2TEX    */ pshader_hw_texm3x2tex,
4675     /* WINED3DSIH_TEXM3x3       */ pshader_hw_texm3x3,
4676     /* WINED3DSIH_TEXM3x3DIFF   */ NULL,
4677     /* WINED3DSIH_TEXM3x3PAD    */ pshader_hw_texm3x3pad,
4678     /* WINED3DSIH_TEXM3x3SPEC   */ pshader_hw_texm3x3spec,
4679     /* WINED3DSIH_TEXM3x3TEX    */ pshader_hw_texm3x3tex,
4680     /* WINED3DSIH_TEXM3x3VSPEC  */ pshader_hw_texm3x3vspec,
4681     /* WINED3DSIH_TEXREG2AR     */ pshader_hw_texreg2ar,
4682     /* WINED3DSIH_TEXREG2GB     */ pshader_hw_texreg2gb,
4683     /* WINED3DSIH_TEXREG2RGB    */ pshader_hw_texreg2rgb,
4684 };
4685
4686 static inline BOOL get_bool_const(const struct wined3d_shader_instruction *ins, IWineD3DBaseShaderImpl *This, DWORD idx)
4687 {
4688     BOOL vshader = shader_is_vshader_version(This->baseShader.reg_maps.shader_version.type);
4689     WORD bools = 0;
4690     WORD flag = (1 << idx);
4691     const local_constant *constant;
4692     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
4693
4694     if(This->baseShader.reg_maps.local_bool_consts & flag)
4695     {
4696         /* What good is a if(bool) with a hardcoded local constant? I don't know, but handle it */
4697         LIST_FOR_EACH_ENTRY(constant, &This->baseShader.constantsB, local_constant, entry)
4698         {
4699             if (constant->idx == idx)
4700             {
4701                 return constant->value[0];
4702             }
4703         }
4704         ERR("Local constant not found\n");
4705         return FALSE;
4706     }
4707     else
4708     {
4709         if(vshader) bools = priv->cur_vs_args->boolclip.bools;
4710         else bools = priv->cur_ps_args->bools;
4711         return bools & flag;
4712     }
4713 }
4714
4715 static void get_loop_control_const(const struct wined3d_shader_instruction *ins,
4716         IWineD3DBaseShaderImpl *This, UINT idx, struct wined3d_shader_loop_control *loop_control)
4717 {
4718     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
4719
4720     /* Integer constants can either be a local constant, or they can be stored in the shader
4721      * type specific compile args. */
4722     if (This->baseShader.reg_maps.local_int_consts & (1 << idx))
4723     {
4724         const local_constant *constant;
4725
4726         LIST_FOR_EACH_ENTRY(constant, &This->baseShader.constantsI, local_constant, entry)
4727         {
4728             if (constant->idx == idx)
4729             {
4730                 loop_control->count = constant->value[0];
4731                 loop_control->start = constant->value[1];
4732                 /* Step is signed. */
4733                 loop_control->step = (int)constant->value[2];
4734                 return;
4735             }
4736         }
4737         /* If this happens the flag was set incorrectly */
4738         ERR("Local constant not found\n");
4739         loop_control->count = 0;
4740         loop_control->start = 0;
4741         loop_control->step = 0;
4742         return;
4743     }
4744
4745     switch (This->baseShader.reg_maps.shader_version.type)
4746     {
4747         case WINED3D_SHADER_TYPE_VERTEX:
4748             /* Count and aL start value are unsigned */
4749             loop_control->count = priv->cur_vs_args->loop_ctrl[idx][0];
4750             loop_control->start = priv->cur_vs_args->loop_ctrl[idx][1];
4751             /* Step is signed. */
4752             loop_control->step = ((char)priv->cur_vs_args->loop_ctrl[idx][2]);
4753             break;
4754
4755         case WINED3D_SHADER_TYPE_PIXEL:
4756             loop_control->count = priv->cur_ps_args->loop_ctrl[idx][0];
4757             loop_control->start = priv->cur_ps_args->loop_ctrl[idx][1];
4758             loop_control->step = ((char)priv->cur_ps_args->loop_ctrl[idx][2]);
4759             break;
4760
4761         default:
4762             FIXME("Unhandled shader type %#x.\n", This->baseShader.reg_maps.shader_version.type);
4763             break;
4764     }
4765 }
4766
4767 static void record_instruction(struct list *list, const struct wined3d_shader_instruction *ins)
4768 {
4769     unsigned int i;
4770     struct wined3d_shader_dst_param *dst_param = NULL;
4771     struct wined3d_shader_src_param *src_param = NULL, *rel_addr = NULL;
4772     struct recorded_instruction *rec = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*rec));
4773     if(!rec)
4774     {
4775         ERR("Out of memory\n");
4776         return;
4777     }
4778
4779     rec->ins = *ins;
4780     dst_param = HeapAlloc(GetProcessHeap(), 0, sizeof(*dst_param));
4781     if(!dst_param) goto free;
4782     *dst_param = *ins->dst;
4783     if(ins->dst->reg.rel_addr)
4784     {
4785         rel_addr = HeapAlloc(GetProcessHeap(), 0, sizeof(*dst_param->reg.rel_addr));
4786         if(!rel_addr) goto free;
4787         *rel_addr = *ins->dst->reg.rel_addr;
4788         dst_param->reg.rel_addr = rel_addr;
4789     }
4790     rec->ins.dst = dst_param;
4791
4792     src_param = HeapAlloc(GetProcessHeap(), 0, sizeof(*src_param) * ins->src_count);
4793     if(!src_param) goto free;
4794     for(i = 0; i < ins->src_count; i++)
4795     {
4796         src_param[i] = ins->src[i];
4797         if(ins->src[i].reg.rel_addr)
4798         {
4799             rel_addr = HeapAlloc(GetProcessHeap(), 0, sizeof(*rel_addr));
4800             if(!rel_addr) goto free;
4801             *rel_addr = *ins->src[i].reg.rel_addr;
4802             src_param[i].reg.rel_addr = rel_addr;
4803         }
4804     }
4805     rec->ins.src = src_param;
4806     list_add_tail(list, &rec->entry);
4807     return;
4808
4809 free:
4810     ERR("Out of memory\n");
4811     if(dst_param)
4812     {
4813         HeapFree(GetProcessHeap(), 0, (void *) dst_param->reg.rel_addr);
4814         HeapFree(GetProcessHeap(), 0, dst_param);
4815     }
4816     if(src_param)
4817     {
4818         for(i = 0; i < ins->src_count; i++)
4819         {
4820             HeapFree(GetProcessHeap(), 0, (void *) src_param[i].reg.rel_addr);
4821         }
4822         HeapFree(GetProcessHeap(), 0, src_param);
4823     }
4824     HeapFree(GetProcessHeap(), 0, rec);
4825 }
4826
4827 static void free_recorded_instruction(struct list *list)
4828 {
4829     struct recorded_instruction *rec_ins, *entry2;
4830     unsigned int i;
4831
4832     LIST_FOR_EACH_ENTRY_SAFE(rec_ins, entry2, list, struct recorded_instruction, entry)
4833     {
4834         list_remove(&rec_ins->entry);
4835         if(rec_ins->ins.dst)
4836         {
4837             HeapFree(GetProcessHeap(), 0, (void *) rec_ins->ins.dst->reg.rel_addr);
4838             HeapFree(GetProcessHeap(), 0, (void *) rec_ins->ins.dst);
4839         }
4840         if(rec_ins->ins.src)
4841         {
4842             for(i = 0; i < rec_ins->ins.src_count; i++)
4843             {
4844                 HeapFree(GetProcessHeap(), 0, (void *) rec_ins->ins.src[i].reg.rel_addr);
4845             }
4846             HeapFree(GetProcessHeap(), 0, (void *) rec_ins->ins.src);
4847         }
4848         HeapFree(GetProcessHeap(), 0, rec_ins);
4849     }
4850 }
4851
4852 static void shader_arb_handle_instruction(const struct wined3d_shader_instruction *ins) {
4853     SHADER_HANDLER hw_fct;
4854     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
4855     IWineD3DBaseShaderImpl *This = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
4856     struct control_frame *control_frame;
4857     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
4858     BOOL bool_const;
4859
4860     if(ins->handler_idx == WINED3DSIH_LOOP || ins->handler_idx == WINED3DSIH_REP)
4861     {
4862         control_frame = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*control_frame));
4863         list_add_head(&priv->control_frames, &control_frame->entry);
4864
4865         if(ins->handler_idx == WINED3DSIH_LOOP) control_frame->type = LOOP;
4866         if(ins->handler_idx == WINED3DSIH_REP) control_frame->type = REP;
4867
4868         if(priv->target_version >= NV2)
4869         {
4870             control_frame->loop_no = priv->num_loops++;
4871             priv->loop_depth++;
4872         }
4873         else
4874         {
4875             /* Don't bother recording when we're in a not used if branch */
4876             if(priv->muted)
4877             {
4878                 return;
4879             }
4880
4881             if(!priv->recording)
4882             {
4883                 list_init(&priv->record);
4884                 priv->recording = TRUE;
4885                 control_frame->outer_loop = TRUE;
4886                 get_loop_control_const(ins, This, ins->src[0].reg.idx, &control_frame->loop_control);
4887                 return; /* Instruction is handled */
4888             }
4889             /* Record this loop in the outer loop's recording */
4890         }
4891     }
4892     else if(ins->handler_idx == WINED3DSIH_ENDLOOP || ins->handler_idx == WINED3DSIH_ENDREP)
4893     {
4894         if(priv->target_version >= NV2)
4895         {
4896             /* Nothing to do. The control frame is popped after the HW instr handler */
4897         }
4898         else
4899         {
4900             struct list *e = list_head(&priv->control_frames);
4901             control_frame = LIST_ENTRY(e, struct control_frame, entry);
4902             list_remove(&control_frame->entry);
4903
4904             if(control_frame->outer_loop)
4905             {
4906                 int iteration, aL = 0;
4907                 struct list copy;
4908
4909                 /* Turn off recording before playback */
4910                 priv->recording = FALSE;
4911
4912                 /* Move the recorded instructions to a separate list and get them out of the private data
4913                  * structure. If there are nested loops, the shader_arb_handle_instruction below will
4914                  * be recorded again, thus priv->record might be overwritten
4915                  */
4916                 list_init(&copy);
4917                 list_move_tail(&copy, &priv->record);
4918                 list_init(&priv->record);
4919
4920                 if(ins->handler_idx == WINED3DSIH_ENDLOOP)
4921                 {
4922                     shader_addline(buffer, "#unrolling loop: %u iterations, aL=%u, inc %d\n",
4923                                    control_frame->loop_control.count, control_frame->loop_control.start,
4924                                    control_frame->loop_control.step);
4925                     aL = control_frame->loop_control.start;
4926                 }
4927                 else
4928                 {
4929                     shader_addline(buffer, "#unrolling rep: %u iterations\n", control_frame->loop_control.count);
4930                 }
4931
4932                 for (iteration = 0; iteration < control_frame->loop_control.count; ++iteration)
4933                 {
4934                     struct recorded_instruction *rec_ins;
4935                     if(ins->handler_idx == WINED3DSIH_ENDLOOP)
4936                     {
4937                         priv->aL = aL;
4938                         shader_addline(buffer, "#Iteration %d, aL=%d\n", iteration, aL);
4939                     }
4940                     else
4941                     {
4942                         shader_addline(buffer, "#Iteration %d\n", iteration);
4943                     }
4944
4945                     LIST_FOR_EACH_ENTRY(rec_ins, &copy, struct recorded_instruction, entry)
4946                     {
4947                         shader_arb_handle_instruction(&rec_ins->ins);
4948                     }
4949
4950                     if(ins->handler_idx == WINED3DSIH_ENDLOOP)
4951                     {
4952                         aL += control_frame->loop_control.step;
4953                     }
4954                 }
4955                 shader_addline(buffer, "#end loop/rep\n");
4956
4957                 free_recorded_instruction(&copy);
4958                 HeapFree(GetProcessHeap(), 0, control_frame);
4959                 return; /* Instruction is handled */
4960             }
4961             else
4962             {
4963                 /* This is a nested loop. Proceed to the normal recording function */
4964                 HeapFree(GetProcessHeap(), 0, control_frame);
4965             }
4966         }
4967     }
4968
4969     if(priv->recording)
4970     {
4971         record_instruction(&priv->record, ins);
4972         return;
4973     }
4974
4975     /* boolean if */
4976     if(ins->handler_idx == WINED3DSIH_IF)
4977     {
4978         control_frame = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*control_frame));
4979         list_add_head(&priv->control_frames, &control_frame->entry);
4980         control_frame->type = IF;
4981
4982         bool_const = get_bool_const(ins, This, ins->src[0].reg.idx);
4983         if(ins->src[0].modifiers == WINED3DSPSM_NOT) bool_const = !bool_const;
4984         if(!priv->muted && bool_const == FALSE)
4985         {
4986             shader_addline(buffer, "#if(FALSE){\n");
4987             priv->muted = TRUE;
4988             control_frame->muting = TRUE;
4989         }
4990         else shader_addline(buffer, "#if(TRUE) {\n");
4991
4992         return; /* Instruction is handled */
4993     }
4994     else if(ins->handler_idx == WINED3DSIH_IFC)
4995     {
4996         /* IF(bool) and if_cond(a, b) use the same ELSE and ENDIF tokens */
4997         control_frame = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*control_frame));
4998         control_frame->type = IFC;
4999         control_frame->ifc_no = priv->num_ifcs++;
5000         list_add_head(&priv->control_frames, &control_frame->entry);
5001     }
5002     else if(ins->handler_idx == WINED3DSIH_ELSE)
5003     {
5004         struct list *e = list_head(&priv->control_frames);
5005         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5006
5007         if(control_frame->type == IF)
5008         {
5009             shader_addline(buffer, "#} else {\n");
5010             if(!priv->muted && !control_frame->muting)
5011             {
5012                 priv->muted = TRUE;
5013                 control_frame->muting = TRUE;
5014             }
5015             else if(control_frame->muting) priv->muted = FALSE;
5016             return; /* Instruction is handled. */
5017         }
5018         /* In case of an ifc, generate a HW shader instruction */
5019     }
5020     else if(ins->handler_idx == WINED3DSIH_ENDIF)
5021     {
5022         struct list *e = list_head(&priv->control_frames);
5023         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5024
5025         if(control_frame->type == IF)
5026         {
5027             shader_addline(buffer, "#} endif\n");
5028             if(control_frame->muting) priv->muted = FALSE;
5029             list_remove(&control_frame->entry);
5030             HeapFree(GetProcessHeap(), 0, control_frame);
5031             return; /* Instruction is handled */
5032         }
5033     }
5034
5035     if(priv->muted) return;
5036
5037     /* Select handler */
5038     hw_fct = shader_arb_instruction_handler_table[ins->handler_idx];
5039
5040     /* Unhandled opcode */
5041     if (!hw_fct)
5042     {
5043         FIXME("Backend can't handle opcode %#x\n", ins->handler_idx);
5044         return;
5045     }
5046     hw_fct(ins);
5047
5048     if(ins->handler_idx == WINED3DSIH_ENDLOOP || ins->handler_idx == WINED3DSIH_ENDREP)
5049     {
5050         struct list *e = list_head(&priv->control_frames);
5051         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5052         list_remove(&control_frame->entry);
5053         HeapFree(GetProcessHeap(), 0, control_frame);
5054         priv->loop_depth--;
5055     }
5056     else if(ins->handler_idx == WINED3DSIH_ENDIF)
5057     {
5058         /* Non-ifc ENDIFs don't reach that place because of the return in the if block above */
5059         struct list *e = list_head(&priv->control_frames);
5060         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5061         list_remove(&control_frame->entry);
5062         HeapFree(GetProcessHeap(), 0, control_frame);
5063     }
5064
5065
5066     shader_arb_add_instruction_modifiers(ins);
5067 }
5068
5069 const shader_backend_t arb_program_shader_backend = {
5070     shader_arb_handle_instruction,
5071     shader_arb_select,
5072     shader_arb_select_depth_blt,
5073     shader_arb_deselect_depth_blt,
5074     shader_arb_update_float_vertex_constants,
5075     shader_arb_update_float_pixel_constants,
5076     shader_arb_load_constants,
5077     shader_arb_load_np2fixup_constants,
5078     shader_arb_destroy,
5079     shader_arb_alloc,
5080     shader_arb_free,
5081     shader_arb_dirty_const,
5082     shader_arb_get_caps,
5083     shader_arb_color_fixup_supported,
5084 };
5085
5086 /* ARB_fragment_program fixed function pipeline replacement definitions */
5087 #define ARB_FFP_CONST_TFACTOR           0
5088 #define ARB_FFP_CONST_SPECULAR_ENABLE   ((ARB_FFP_CONST_TFACTOR) + 1)
5089 #define ARB_FFP_CONST_CONSTANT(i)       ((ARB_FFP_CONST_SPECULAR_ENABLE) + 1 + i)
5090 #define ARB_FFP_CONST_BUMPMAT(i)        ((ARB_FFP_CONST_CONSTANT(7)) + 1 + i)
5091 #define ARB_FFP_CONST_LUMINANCE(i)      ((ARB_FFP_CONST_BUMPMAT(7)) + 1 + i)
5092
5093 struct arbfp_ffp_desc
5094 {
5095     struct ffp_frag_desc parent;
5096     GLuint shader;
5097     unsigned int num_textures_used;
5098 };
5099
5100 /* Context activation is done by the caller. */
5101 static void arbfp_enable(IWineD3DDevice *iface, BOOL enable) {
5102     ENTER_GL();
5103     if(enable) {
5104         glEnable(GL_FRAGMENT_PROGRAM_ARB);
5105         checkGLcall("glEnable(GL_FRAGMENT_PROGRAM_ARB)");
5106     } else {
5107         glDisable(GL_FRAGMENT_PROGRAM_ARB);
5108         checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
5109     }
5110     LEAVE_GL();
5111 }
5112
5113 static HRESULT arbfp_alloc(IWineD3DDevice *iface) {
5114     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *) iface;
5115     struct shader_arb_priv *priv;
5116     /* Share private data between the shader backend and the pipeline replacement, if both
5117      * are the arb implementation. This is needed to figure out whether ARBfp should be disabled
5118      * if no pixel shader is bound or not
5119      */
5120     if(This->shader_backend == &arb_program_shader_backend) {
5121         This->fragment_priv = This->shader_priv;
5122     } else {
5123         This->fragment_priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(struct shader_arb_priv));
5124         if(!This->fragment_priv) return E_OUTOFMEMORY;
5125     }
5126     priv = This->fragment_priv;
5127     if (wine_rb_init(&priv->fragment_shaders, &wined3d_ffp_frag_program_rb_functions) == -1)
5128     {
5129         ERR("Failed to initialize rbtree.\n");
5130         HeapFree(GetProcessHeap(), 0, This->fragment_priv);
5131         return E_OUTOFMEMORY;
5132     }
5133     priv->use_arbfp_fixed_func = TRUE;
5134     return WINED3D_OK;
5135 }
5136
5137 /* Context activation is done by the caller. */
5138 static void arbfp_free_ffpshader(struct wine_rb_entry *entry, void *context)
5139 {
5140     const struct wined3d_gl_info *gl_info = context;
5141     struct arbfp_ffp_desc *entry_arb = WINE_RB_ENTRY_VALUE(entry, struct arbfp_ffp_desc, parent.entry);
5142
5143     ENTER_GL();
5144     GL_EXTCALL(glDeleteProgramsARB(1, &entry_arb->shader));
5145     checkGLcall("glDeleteProgramsARB(1, &entry_arb->shader)");
5146     HeapFree(GetProcessHeap(), 0, entry_arb);
5147     LEAVE_GL();
5148 }
5149
5150 /* Context activation is done by the caller. */
5151 static void arbfp_free(IWineD3DDevice *iface) {
5152     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *) iface;
5153     struct shader_arb_priv *priv = This->fragment_priv;
5154
5155     wine_rb_destroy(&priv->fragment_shaders, arbfp_free_ffpshader, &This->adapter->gl_info);
5156     priv->use_arbfp_fixed_func = FALSE;
5157
5158     if(This->shader_backend != &arb_program_shader_backend) {
5159         HeapFree(GetProcessHeap(), 0, This->fragment_priv);
5160     }
5161 }
5162
5163 static void arbfp_get_caps(WINED3DDEVTYPE devtype, const struct wined3d_gl_info *gl_info, struct fragment_caps *caps)
5164 {
5165     caps->TextureOpCaps =  WINED3DTEXOPCAPS_DISABLE                     |
5166                            WINED3DTEXOPCAPS_SELECTARG1                  |
5167                            WINED3DTEXOPCAPS_SELECTARG2                  |
5168                            WINED3DTEXOPCAPS_MODULATE4X                  |
5169                            WINED3DTEXOPCAPS_MODULATE2X                  |
5170                            WINED3DTEXOPCAPS_MODULATE                    |
5171                            WINED3DTEXOPCAPS_ADDSIGNED2X                 |
5172                            WINED3DTEXOPCAPS_ADDSIGNED                   |
5173                            WINED3DTEXOPCAPS_ADD                         |
5174                            WINED3DTEXOPCAPS_SUBTRACT                    |
5175                            WINED3DTEXOPCAPS_ADDSMOOTH                   |
5176                            WINED3DTEXOPCAPS_BLENDCURRENTALPHA           |
5177                            WINED3DTEXOPCAPS_BLENDFACTORALPHA            |
5178                            WINED3DTEXOPCAPS_BLENDTEXTUREALPHA           |
5179                            WINED3DTEXOPCAPS_BLENDDIFFUSEALPHA           |
5180                            WINED3DTEXOPCAPS_BLENDTEXTUREALPHAPM         |
5181                            WINED3DTEXOPCAPS_MODULATEALPHA_ADDCOLOR      |
5182                            WINED3DTEXOPCAPS_MODULATECOLOR_ADDALPHA      |
5183                            WINED3DTEXOPCAPS_MODULATEINVCOLOR_ADDALPHA   |
5184                            WINED3DTEXOPCAPS_MODULATEINVALPHA_ADDCOLOR   |
5185                            WINED3DTEXOPCAPS_DOTPRODUCT3                 |
5186                            WINED3DTEXOPCAPS_MULTIPLYADD                 |
5187                            WINED3DTEXOPCAPS_LERP                        |
5188                            WINED3DTEXOPCAPS_BUMPENVMAP                  |
5189                            WINED3DTEXOPCAPS_BUMPENVMAPLUMINANCE;
5190
5191     /* TODO: Implement WINED3DTEXOPCAPS_PREMODULATE */
5192
5193     caps->MaxTextureBlendStages   = 8;
5194     caps->MaxSimultaneousTextures = min(GL_LIMITS(fragment_samplers), 8);
5195
5196     caps->PrimitiveMiscCaps |= WINED3DPMISCCAPS_TSSARGTEMP;
5197 }
5198 #undef GLINFO_LOCATION
5199
5200 #define GLINFO_LOCATION stateblock->wineD3DDevice->adapter->gl_info
5201 static void state_texfactor_arbfp(DWORD state, IWineD3DStateBlockImpl *stateblock, struct wined3d_context *context)
5202 {
5203     float col[4];
5204     IWineD3DDeviceImpl *device = stateblock->wineD3DDevice;
5205
5206     /* Don't load the parameter if we're using an arbfp pixel shader, otherwise we'll overwrite
5207      * application provided constants
5208      */
5209     if(device->shader_backend == &arb_program_shader_backend) {
5210         if (use_ps(stateblock)) return;
5211
5212         device = stateblock->wineD3DDevice;
5213         context->pshader_const_dirty[ARB_FFP_CONST_TFACTOR] = 1;
5214         device->highest_dirty_ps_const = max(device->highest_dirty_ps_const, ARB_FFP_CONST_TFACTOR + 1);
5215     }
5216
5217     D3DCOLORTOGLFLOAT4(stateblock->renderState[WINED3DRS_TEXTUREFACTOR], col);
5218     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_TFACTOR, col));
5219     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_TFACTOR, col)");
5220
5221 }
5222
5223 static void state_arb_specularenable(DWORD state, IWineD3DStateBlockImpl *stateblock, struct wined3d_context *context)
5224 {
5225     float col[4];
5226     IWineD3DDeviceImpl *device = stateblock->wineD3DDevice;
5227
5228     /* Don't load the parameter if we're using an arbfp pixel shader, otherwise we'll overwrite
5229      * application provided constants
5230      */
5231     if(device->shader_backend == &arb_program_shader_backend) {
5232         if (use_ps(stateblock)) return;
5233
5234         device = stateblock->wineD3DDevice;
5235         context->pshader_const_dirty[ARB_FFP_CONST_SPECULAR_ENABLE] = 1;
5236         device->highest_dirty_ps_const = max(device->highest_dirty_ps_const, ARB_FFP_CONST_SPECULAR_ENABLE + 1);
5237     }
5238
5239     if(stateblock->renderState[WINED3DRS_SPECULARENABLE]) {
5240         /* The specular color has no alpha */
5241         col[0] = 1.0f; col[1] = 1.0f;
5242         col[2] = 1.0f; col[3] = 0.0f;
5243     } else {
5244         col[0] = 0.0f; col[1] = 0.0f;
5245         col[2] = 0.0f; col[3] = 0.0f;
5246     }
5247     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_SPECULAR_ENABLE, col));
5248     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_SPECULAR_ENABLE, col)");
5249 }
5250
5251 static void set_bumpmat_arbfp(DWORD state, IWineD3DStateBlockImpl *stateblock, struct wined3d_context *context)
5252 {
5253     DWORD stage = (state - STATE_TEXTURESTAGE(0, 0)) / (WINED3D_HIGHEST_TEXTURE_STATE + 1);
5254     IWineD3DDeviceImpl *device = stateblock->wineD3DDevice;
5255     float mat[2][2];
5256
5257     if (use_ps(stateblock))
5258     {
5259         if (stage != 0
5260                 && (((IWineD3DPixelShaderImpl *)stateblock->pixelShader)->baseShader.reg_maps.bumpmat & (1 << stage)))
5261         {
5262             /* The pixel shader has to know the bump env matrix. Do a constants update if it isn't scheduled
5263              * anyway
5264              */
5265             if(!isStateDirty(context, STATE_PIXELSHADERCONSTANT)) {
5266                 device->StateTable[STATE_PIXELSHADERCONSTANT].apply(STATE_PIXELSHADERCONSTANT, stateblock, context);
5267             }
5268         }
5269
5270         if(device->shader_backend == &arb_program_shader_backend) {
5271             /* Exit now, don't set the bumpmat below, otherwise we may overwrite pixel shader constants */
5272             return;
5273         }
5274     } else if(device->shader_backend == &arb_program_shader_backend) {
5275         context->pshader_const_dirty[ARB_FFP_CONST_BUMPMAT(stage)] = 1;
5276         device->highest_dirty_ps_const = max(device->highest_dirty_ps_const, ARB_FFP_CONST_BUMPMAT(stage) + 1);
5277     }
5278
5279     mat[0][0] = *((float *) &stateblock->textureState[stage][WINED3DTSS_BUMPENVMAT00]);
5280     mat[0][1] = *((float *) &stateblock->textureState[stage][WINED3DTSS_BUMPENVMAT01]);
5281     mat[1][0] = *((float *) &stateblock->textureState[stage][WINED3DTSS_BUMPENVMAT10]);
5282     mat[1][1] = *((float *) &stateblock->textureState[stage][WINED3DTSS_BUMPENVMAT11]);
5283
5284     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_BUMPMAT(stage), &mat[0][0]));
5285     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_BUMPMAT(stage), &mat[0][0])");
5286 }
5287
5288 static void tex_bumpenvlum_arbfp(DWORD state, IWineD3DStateBlockImpl *stateblock, struct wined3d_context *context)
5289 {
5290     DWORD stage = (state - STATE_TEXTURESTAGE(0, 0)) / (WINED3D_HIGHEST_TEXTURE_STATE + 1);
5291     IWineD3DDeviceImpl *device = stateblock->wineD3DDevice;
5292     float param[4];
5293
5294     if (use_ps(stateblock))
5295     {
5296         if (stage != 0
5297                 && (((IWineD3DPixelShaderImpl *)stateblock->pixelShader)->baseShader.reg_maps.luminanceparams & (1 << stage)))
5298         {
5299             /* The pixel shader has to know the luminance offset. Do a constants update if it
5300              * isn't scheduled anyway
5301              */
5302             if(!isStateDirty(context, STATE_PIXELSHADERCONSTANT)) {
5303                 device->StateTable[STATE_PIXELSHADERCONSTANT].apply(STATE_PIXELSHADERCONSTANT, stateblock, context);
5304             }
5305         }
5306
5307         if(device->shader_backend == &arb_program_shader_backend) {
5308             /* Exit now, don't set the bumpmat below, otherwise we may overwrite pixel shader constants */
5309             return;
5310         }
5311     } else if(device->shader_backend == &arb_program_shader_backend) {
5312         context->pshader_const_dirty[ARB_FFP_CONST_LUMINANCE(stage)] = 1;
5313         device->highest_dirty_ps_const = max(device->highest_dirty_ps_const, ARB_FFP_CONST_LUMINANCE(stage) + 1);
5314     }
5315
5316     param[0] = *((float *) &stateblock->textureState[stage][WINED3DTSS_BUMPENVLSCALE]);
5317     param[1] = *((float *) &stateblock->textureState[stage][WINED3DTSS_BUMPENVLOFFSET]);
5318     param[2] = 0.0f;
5319     param[3] = 0.0f;
5320
5321     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_LUMINANCE(stage), param));
5322     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_LUMINANCE(stage), param)");
5323 }
5324
5325 static const char *get_argreg(struct wined3d_shader_buffer *buffer, DWORD argnum, unsigned int stage, DWORD arg)
5326 {
5327     const char *ret;
5328
5329     if(arg == ARG_UNUSED) return "unused"; /* This is the marker for unused registers */
5330
5331     switch(arg & WINED3DTA_SELECTMASK) {
5332         case WINED3DTA_DIFFUSE:
5333             ret = "fragment.color.primary"; break;
5334
5335         case WINED3DTA_CURRENT:
5336             if(stage == 0) ret = "fragment.color.primary";
5337             else ret = "ret";
5338             break;
5339
5340         case WINED3DTA_TEXTURE:
5341             switch(stage) {
5342                 case 0: ret = "tex0"; break;
5343                 case 1: ret = "tex1"; break;
5344                 case 2: ret = "tex2"; break;
5345                 case 3: ret = "tex3"; break;
5346                 case 4: ret = "tex4"; break;
5347                 case 5: ret = "tex5"; break;
5348                 case 6: ret = "tex6"; break;
5349                 case 7: ret = "tex7"; break;
5350                 default: ret = "unknown texture";
5351             }
5352             break;
5353
5354         case WINED3DTA_TFACTOR:
5355             ret = "tfactor"; break;
5356
5357         case WINED3DTA_SPECULAR:
5358             ret = "fragment.color.secondary"; break;
5359
5360         case WINED3DTA_TEMP:
5361             ret = "tempreg"; break;
5362
5363         case WINED3DTA_CONSTANT:
5364             FIXME("Implement perstage constants\n");
5365             switch(stage) {
5366                 case 0: ret = "const0"; break;
5367                 case 1: ret = "const1"; break;
5368                 case 2: ret = "const2"; break;
5369                 case 3: ret = "const3"; break;
5370                 case 4: ret = "const4"; break;
5371                 case 5: ret = "const5"; break;
5372                 case 6: ret = "const6"; break;
5373                 case 7: ret = "const7"; break;
5374                 default: ret = "unknown constant";
5375             }
5376             break;
5377
5378         default:
5379             return "unknown";
5380     }
5381
5382     if(arg & WINED3DTA_COMPLEMENT) {
5383         shader_addline(buffer, "SUB arg%u, const.x, %s;\n", argnum, ret);
5384         if(argnum == 0) ret = "arg0";
5385         if(argnum == 1) ret = "arg1";
5386         if(argnum == 2) ret = "arg2";
5387     }
5388     if(arg & WINED3DTA_ALPHAREPLICATE) {
5389         shader_addline(buffer, "MOV arg%u, %s.w;\n", argnum, ret);
5390         if(argnum == 0) ret = "arg0";
5391         if(argnum == 1) ret = "arg1";
5392         if(argnum == 2) ret = "arg2";
5393     }
5394     return ret;
5395 }
5396
5397 static void gen_ffp_instr(struct wined3d_shader_buffer *buffer, unsigned int stage, BOOL color,
5398         BOOL alpha, DWORD dst, DWORD op, DWORD dw_arg0, DWORD dw_arg1, DWORD dw_arg2)
5399 {
5400     const char *dstmask, *dstreg, *arg0, *arg1, *arg2;
5401     unsigned int mul = 1;
5402     BOOL mul_final_dest = FALSE;
5403
5404     if(color && alpha) dstmask = "";
5405     else if(color) dstmask = ".xyz";
5406     else dstmask = ".w";
5407
5408     if(dst == tempreg) dstreg = "tempreg";
5409     else dstreg = "ret";
5410
5411     arg0 = get_argreg(buffer, 0, stage, dw_arg0);
5412     arg1 = get_argreg(buffer, 1, stage, dw_arg1);
5413     arg2 = get_argreg(buffer, 2, stage, dw_arg2);
5414
5415     switch(op) {
5416         case WINED3DTOP_DISABLE:
5417             if(stage == 0) shader_addline(buffer, "MOV %s%s, fragment.color.primary;\n", dstreg, dstmask);
5418             break;
5419
5420         case WINED3DTOP_SELECTARG2:
5421             arg1 = arg2;
5422         case WINED3DTOP_SELECTARG1:
5423             shader_addline(buffer, "MOV %s%s, %s;\n", dstreg, dstmask, arg1);
5424             break;
5425
5426         case WINED3DTOP_MODULATE4X:
5427             mul = 2;
5428         case WINED3DTOP_MODULATE2X:
5429             mul *= 2;
5430             if(strcmp(dstreg, "result.color") == 0) {
5431                 dstreg = "ret";
5432                 mul_final_dest = TRUE;
5433             }
5434         case WINED3DTOP_MODULATE:
5435             shader_addline(buffer, "MUL %s%s, %s, %s;\n", dstreg, dstmask, arg1, arg2);
5436             break;
5437
5438         case WINED3DTOP_ADDSIGNED2X:
5439             mul = 2;
5440             if(strcmp(dstreg, "result.color") == 0) {
5441                 dstreg = "ret";
5442                 mul_final_dest = TRUE;
5443             }
5444         case WINED3DTOP_ADDSIGNED:
5445             shader_addline(buffer, "SUB arg2, %s, const.w;\n", arg2);
5446             arg2 = "arg2";
5447         case WINED3DTOP_ADD:
5448             shader_addline(buffer, "ADD_SAT %s%s, %s, %s;\n", dstreg, dstmask, arg1, arg2);
5449             break;
5450
5451         case WINED3DTOP_SUBTRACT:
5452             shader_addline(buffer, "SUB_SAT %s%s, %s, %s;\n", dstreg, dstmask, arg1, arg2);
5453             break;
5454
5455         case WINED3DTOP_ADDSMOOTH:
5456             shader_addline(buffer, "SUB arg1, const.x, %s;\n", arg1);
5457             shader_addline(buffer, "MAD_SAT %s%s, arg1, %s, %s;\n", dstreg, dstmask, arg2, arg1);
5458             break;
5459
5460         case WINED3DTOP_BLENDCURRENTALPHA:
5461             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_CURRENT);
5462             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
5463             break;
5464         case WINED3DTOP_BLENDFACTORALPHA:
5465             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_TFACTOR);
5466             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
5467             break;
5468         case WINED3DTOP_BLENDTEXTUREALPHA:
5469             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_TEXTURE);
5470             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
5471             break;
5472         case WINED3DTOP_BLENDDIFFUSEALPHA:
5473             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_DIFFUSE);
5474             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
5475             break;
5476
5477         case WINED3DTOP_BLENDTEXTUREALPHAPM:
5478             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_TEXTURE);
5479             shader_addline(buffer, "SUB arg0.w, const.x, %s.w;\n", arg0);
5480             shader_addline(buffer, "MAD_SAT %s%s, %s, arg0.w, %s;\n", dstreg, dstmask, arg2, arg1);
5481             break;
5482
5483         /* D3DTOP_PREMODULATE ???? */
5484
5485         case WINED3DTOP_MODULATEINVALPHA_ADDCOLOR:
5486             shader_addline(buffer, "SUB arg0.w, const.x, %s;\n", arg1);
5487             shader_addline(buffer, "MAD_SAT %s%s, arg0.w, %s, %s;\n", dstreg, dstmask, arg2, arg1);
5488             break;
5489         case WINED3DTOP_MODULATEALPHA_ADDCOLOR:
5490             shader_addline(buffer, "MAD_SAT %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg1, arg2, arg1);
5491             break;
5492         case WINED3DTOP_MODULATEINVCOLOR_ADDALPHA:
5493             shader_addline(buffer, "SUB arg0, const.x, %s;\n", arg1);
5494             shader_addline(buffer, "MAD_SAT %s%s, arg0, %s, %s.w;\n", dstreg, dstmask, arg2, arg1);
5495             break;
5496         case WINED3DTOP_MODULATECOLOR_ADDALPHA:
5497             shader_addline(buffer, "MAD_SAT %s%s, %s, %s, %s.w;\n", dstreg, dstmask, arg1, arg2, arg1);
5498             break;
5499
5500         case WINED3DTOP_DOTPRODUCT3:
5501             mul = 4;
5502             if(strcmp(dstreg, "result.color") == 0) {
5503                 dstreg = "ret";
5504                 mul_final_dest = TRUE;
5505             }
5506             shader_addline(buffer, "SUB arg1, %s, const.w;\n", arg1);
5507             shader_addline(buffer, "SUB arg2, %s, const.w;\n", arg2);
5508             shader_addline(buffer, "DP3_SAT %s%s, arg1, arg2;\n", dstreg, dstmask);
5509             break;
5510
5511         case WINED3DTOP_MULTIPLYADD:
5512             shader_addline(buffer, "MAD_SAT %s%s, %s, %s, %s;\n", dstreg, dstmask, arg1, arg2, arg0);
5513             break;
5514
5515         case WINED3DTOP_LERP:
5516             /* The msdn is not quite right here */
5517             shader_addline(buffer, "LRP %s%s, %s, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
5518             break;
5519
5520         case WINED3DTOP_BUMPENVMAP:
5521         case WINED3DTOP_BUMPENVMAPLUMINANCE:
5522             /* Those are handled in the first pass of the shader(generation pass 1 and 2) already */
5523             break;
5524
5525         default:
5526             FIXME("Unhandled texture op %08x\n", op);
5527     }
5528
5529     if(mul == 2) {
5530         shader_addline(buffer, "MUL_SAT %s%s, %s, const.y;\n", mul_final_dest ? "result.color" : dstreg, dstmask, dstreg);
5531     } else if(mul == 4) {
5532         shader_addline(buffer, "MUL_SAT %s%s, %s, const.z;\n", mul_final_dest ? "result.color" : dstreg, dstmask, dstreg);
5533     }
5534 }
5535
5536 /* The stateblock is passed for GLINFO_LOCATION */
5537 static GLuint gen_arbfp_ffp_shader(const struct ffp_frag_settings *settings, IWineD3DStateBlockImpl *stateblock)
5538 {
5539     unsigned int stage;
5540     struct wined3d_shader_buffer buffer;
5541     BOOL tex_read[MAX_TEXTURES] = {FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE};
5542     BOOL bump_used[MAX_TEXTURES] = {FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE};
5543     BOOL luminance_used[MAX_TEXTURES] = {FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE};
5544     const char *textype;
5545     const char *instr, *sat;
5546     char colorcor_dst[8];
5547     GLuint ret;
5548     DWORD arg0, arg1, arg2;
5549     BOOL tempreg_used = FALSE, tfactor_used = FALSE;
5550     BOOL op_equal;
5551     const char *final_combiner_src = "ret";
5552     GLint pos;
5553
5554     /* Find out which textures are read */
5555     for(stage = 0; stage < MAX_TEXTURES; stage++) {
5556         if(settings->op[stage].cop == WINED3DTOP_DISABLE) break;
5557         arg0 = settings->op[stage].carg0 & WINED3DTA_SELECTMASK;
5558         arg1 = settings->op[stage].carg1 & WINED3DTA_SELECTMASK;
5559         arg2 = settings->op[stage].carg2 & WINED3DTA_SELECTMASK;
5560         if(arg0 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
5561         if(arg1 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
5562         if(arg2 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
5563
5564         if(settings->op[stage].cop == WINED3DTOP_BLENDTEXTUREALPHA) tex_read[stage] = TRUE;
5565         if(settings->op[stage].cop == WINED3DTOP_BLENDTEXTUREALPHAPM) tex_read[stage] = TRUE;
5566         if(settings->op[stage].cop == WINED3DTOP_BUMPENVMAP) {
5567             bump_used[stage] = TRUE;
5568             tex_read[stage] = TRUE;
5569         }
5570         if(settings->op[stage].cop == WINED3DTOP_BUMPENVMAPLUMINANCE) {
5571             bump_used[stage] = TRUE;
5572             tex_read[stage] = TRUE;
5573             luminance_used[stage] = TRUE;
5574         } else if(settings->op[stage].cop == WINED3DTOP_BLENDFACTORALPHA) {
5575             tfactor_used = TRUE;
5576         }
5577
5578         if(arg0 == WINED3DTA_TFACTOR || arg1 == WINED3DTA_TFACTOR || arg2 == WINED3DTA_TFACTOR) {
5579             tfactor_used = TRUE;
5580         }
5581
5582         if(settings->op[stage].dst == tempreg) tempreg_used = TRUE;
5583         if(arg0 == WINED3DTA_TEMP || arg1 == WINED3DTA_TEMP || arg2 == WINED3DTA_TEMP) {
5584             tempreg_used = TRUE;
5585         }
5586
5587         if(settings->op[stage].aop == WINED3DTOP_DISABLE) continue;
5588         arg0 = settings->op[stage].aarg0 & WINED3DTA_SELECTMASK;
5589         arg1 = settings->op[stage].aarg1 & WINED3DTA_SELECTMASK;
5590         arg2 = settings->op[stage].aarg2 & WINED3DTA_SELECTMASK;
5591         if(arg0 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
5592         if(arg1 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
5593         if(arg2 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
5594
5595         if(arg0 == WINED3DTA_TEMP || arg1 == WINED3DTA_TEMP || arg2 == WINED3DTA_TEMP) {
5596             tempreg_used = TRUE;
5597         }
5598         if(arg0 == WINED3DTA_TFACTOR || arg1 == WINED3DTA_TFACTOR || arg2 == WINED3DTA_TFACTOR) {
5599             tfactor_used = TRUE;
5600         }
5601     }
5602
5603     /* Shader header */
5604     if (!shader_buffer_init(&buffer))
5605     {
5606         ERR("Failed to initialize shader buffer.\n");
5607         return 0;
5608     }
5609
5610     shader_addline(&buffer, "!!ARBfp1.0\n");
5611
5612     switch(settings->fog) {
5613         case FOG_OFF:                                                         break;
5614         case FOG_LINEAR: shader_addline(&buffer, "OPTION ARB_fog_linear;\n"); break;
5615         case FOG_EXP:    shader_addline(&buffer, "OPTION ARB_fog_exp;\n");    break;
5616         case FOG_EXP2:   shader_addline(&buffer, "OPTION ARB_fog_exp2;\n");   break;
5617         default: FIXME("Unexpected fog setting %d\n", settings->fog);
5618     }
5619
5620     shader_addline(&buffer, "PARAM const = {1, 2, 4, 0.5};\n");
5621     shader_addline(&buffer, "TEMP TMP;\n");
5622     shader_addline(&buffer, "TEMP ret;\n");
5623     if(tempreg_used || settings->sRGB_write) shader_addline(&buffer, "TEMP tempreg;\n");
5624     shader_addline(&buffer, "TEMP arg0;\n");
5625     shader_addline(&buffer, "TEMP arg1;\n");
5626     shader_addline(&buffer, "TEMP arg2;\n");
5627     for(stage = 0; stage < MAX_TEXTURES; stage++) {
5628         if(!tex_read[stage]) continue;
5629         shader_addline(&buffer, "TEMP tex%u;\n", stage);
5630         if(!bump_used[stage]) continue;
5631         shader_addline(&buffer, "PARAM bumpmat%u = program.env[%u];\n", stage, ARB_FFP_CONST_BUMPMAT(stage));
5632         if(!luminance_used[stage]) continue;
5633         shader_addline(&buffer, "PARAM luminance%u = program.env[%u];\n", stage, ARB_FFP_CONST_LUMINANCE(stage));
5634     }
5635     if(tfactor_used) {
5636         shader_addline(&buffer, "PARAM tfactor = program.env[%u];\n", ARB_FFP_CONST_TFACTOR);
5637     }
5638         shader_addline(&buffer, "PARAM specular_enable = program.env[%u];\n", ARB_FFP_CONST_SPECULAR_ENABLE);
5639
5640     if(settings->sRGB_write) {
5641         shader_addline(&buffer, "PARAM srgb_consts1 = {%f, %f, %f, %f};\n",
5642                        srgb_mul_low, srgb_cmp, srgb_pow, srgb_mul_high);
5643         shader_addline(&buffer, "PARAM srgb_consts2 = {%f, %f, %f, %f};\n",
5644                        srgb_sub_high, 0.0, 0.0, 0.0);
5645     }
5646
5647     if(ffp_clip_emul(stateblock) && settings->emul_clipplanes) shader_addline(&buffer, "KIL fragment.texcoord[7];\n");
5648
5649     /* Generate texture sampling instructions) */
5650     for(stage = 0; stage < MAX_TEXTURES && settings->op[stage].cop != WINED3DTOP_DISABLE; stage++) {
5651         if(!tex_read[stage]) continue;
5652
5653         switch(settings->op[stage].tex_type) {
5654             case tex_1d:                    textype = "1D";     break;
5655             case tex_2d:                    textype = "2D";     break;
5656             case tex_3d:                    textype = "3D";     break;
5657             case tex_cube:                  textype = "CUBE";   break;
5658             case tex_rect:                  textype = "RECT";   break;
5659             default: textype = "unexpected_textype";   break;
5660         }
5661
5662         if(settings->op[stage].cop == WINED3DTOP_BUMPENVMAP ||
5663            settings->op[stage].cop == WINED3DTOP_BUMPENVMAPLUMINANCE) {
5664             sat = "";
5665         } else {
5666             sat = "_SAT";
5667         }
5668
5669         if(settings->op[stage].projected == proj_none) {
5670             instr = "TEX";
5671         } else if(settings->op[stage].projected == proj_count4 ||
5672                   settings->op[stage].projected == proj_count3) {
5673             instr = "TXP";
5674         } else {
5675             FIXME("Unexpected projection mode %d\n", settings->op[stage].projected);
5676             instr = "TXP";
5677         }
5678
5679         if(stage > 0 &&
5680            (settings->op[stage - 1].cop == WINED3DTOP_BUMPENVMAP ||
5681             settings->op[stage - 1].cop == WINED3DTOP_BUMPENVMAPLUMINANCE)) {
5682             shader_addline(&buffer, "SWZ arg1, bumpmat%u, x, z, 0, 0;\n", stage - 1);
5683             shader_addline(&buffer, "DP3 ret.x, arg1, tex%u;\n", stage - 1);
5684             shader_addline(&buffer, "SWZ arg1, bumpmat%u, y, w, 0, 0;\n", stage - 1);
5685             shader_addline(&buffer, "DP3 ret.y, arg1, tex%u;\n", stage - 1);
5686
5687             /* with projective textures, texbem only divides the static texture coord, not the displacement,
5688              * so multiply the displacement with the dividing parameter before passing it to TXP
5689              */
5690             if (settings->op[stage].projected != proj_none) {
5691                 if(settings->op[stage].projected == proj_count4) {
5692                     shader_addline(&buffer, "MOV ret.w, fragment.texcoord[%u].w;\n", stage);
5693                     shader_addline(&buffer, "MUL ret.xyz, ret, fragment.texcoord[%u].w, fragment.texcoord[%u];\n", stage, stage);
5694                 } else {
5695                     shader_addline(&buffer, "MOV ret.w, fragment.texcoord[%u].z;\n", stage);
5696                     shader_addline(&buffer, "MAD ret.xyz, ret, fragment.texcoord[%u].z, fragment.texcoord[%u];\n", stage, stage);
5697                 }
5698             } else {
5699                 shader_addline(&buffer, "ADD ret, ret, fragment.texcoord[%u];\n", stage);
5700             }
5701
5702             shader_addline(&buffer, "%s%s tex%u, ret, texture[%u], %s;\n",
5703                            instr, sat, stage, stage, textype);
5704             if(settings->op[stage - 1].cop == WINED3DTOP_BUMPENVMAPLUMINANCE) {
5705                 shader_addline(&buffer, "MAD_SAT ret.x, tex%u.z, luminance%u.x, luminance%u.y;\n",
5706                                stage - 1, stage - 1, stage - 1);
5707                 shader_addline(&buffer, "MUL tex%u, tex%u, ret.x;\n", stage, stage);
5708             }
5709         } else if(settings->op[stage].projected == proj_count3) {
5710             shader_addline(&buffer, "MOV ret, fragment.texcoord[%u];\n", stage);
5711             shader_addline(&buffer, "MOV ret.w, ret.z;\n");
5712             shader_addline(&buffer, "%s%s tex%u, ret, texture[%u], %s;\n",
5713                             instr, sat, stage, stage, textype);
5714         } else {
5715             shader_addline(&buffer, "%s%s tex%u, fragment.texcoord[%u], texture[%u], %s;\n",
5716                             instr, sat, stage, stage, stage, textype);
5717         }
5718
5719         sprintf(colorcor_dst, "tex%u", stage);
5720         gen_color_correction(&buffer, colorcor_dst, WINED3DSP_WRITEMASK_ALL, "const.x", "const.y",
5721                 settings->op[stage].color_fixup);
5722     }
5723
5724     /* Generate the main shader */
5725     for(stage = 0; stage < MAX_TEXTURES; stage++) {
5726         if(settings->op[stage].cop == WINED3DTOP_DISABLE) {
5727             if(stage == 0) {
5728                 final_combiner_src = "fragment.color.primary";
5729             }
5730             break;
5731         }
5732
5733         if(settings->op[stage].cop == WINED3DTOP_SELECTARG1 &&
5734            settings->op[stage].aop == WINED3DTOP_SELECTARG1) {
5735             op_equal = settings->op[stage].carg1 == settings->op[stage].aarg1;
5736         } else if(settings->op[stage].cop == WINED3DTOP_SELECTARG1 &&
5737                   settings->op[stage].aop == WINED3DTOP_SELECTARG2) {
5738             op_equal = settings->op[stage].carg1 == settings->op[stage].aarg2;
5739         } else if(settings->op[stage].cop == WINED3DTOP_SELECTARG2 &&
5740                   settings->op[stage].aop == WINED3DTOP_SELECTARG1) {
5741             op_equal = settings->op[stage].carg2 == settings->op[stage].aarg1;
5742         } else if(settings->op[stage].cop == WINED3DTOP_SELECTARG2 &&
5743                   settings->op[stage].aop == WINED3DTOP_SELECTARG2) {
5744             op_equal = settings->op[stage].carg2 == settings->op[stage].aarg2;
5745         } else {
5746             op_equal = settings->op[stage].aop   == settings->op[stage].cop &&
5747                        settings->op[stage].carg0 == settings->op[stage].aarg0 &&
5748                        settings->op[stage].carg1 == settings->op[stage].aarg1 &&
5749                        settings->op[stage].carg2 == settings->op[stage].aarg2;
5750         }
5751
5752         if(settings->op[stage].aop == WINED3DTOP_DISABLE) {
5753             gen_ffp_instr(&buffer, stage, TRUE, FALSE, settings->op[stage].dst,
5754                           settings->op[stage].cop, settings->op[stage].carg0,
5755                           settings->op[stage].carg1, settings->op[stage].carg2);
5756             if(stage == 0) {
5757                 shader_addline(&buffer, "MOV ret.w, fragment.color.primary.w;\n");
5758             }
5759         } else if(op_equal) {
5760             gen_ffp_instr(&buffer, stage, TRUE, TRUE, settings->op[stage].dst,
5761                           settings->op[stage].cop, settings->op[stage].carg0,
5762                           settings->op[stage].carg1, settings->op[stage].carg2);
5763         } else {
5764             gen_ffp_instr(&buffer, stage, TRUE, FALSE, settings->op[stage].dst,
5765                           settings->op[stage].cop, settings->op[stage].carg0,
5766                           settings->op[stage].carg1, settings->op[stage].carg2);
5767             gen_ffp_instr(&buffer, stage, FALSE, TRUE, settings->op[stage].dst,
5768                           settings->op[stage].aop, settings->op[stage].aarg0,
5769                           settings->op[stage].aarg1, settings->op[stage].aarg2);
5770         }
5771     }
5772
5773     if(settings->sRGB_write) {
5774         shader_addline(&buffer, "MAD ret, fragment.color.secondary, specular_enable, %s;\n", final_combiner_src);
5775         arbfp_add_sRGB_correction(&buffer, "ret", "arg0", "arg1", "arg2", "tempreg", FALSE);
5776     } else {
5777         shader_addline(&buffer, "MAD result.color, fragment.color.secondary, specular_enable, %s;\n", final_combiner_src);
5778     }
5779
5780     /* Footer */
5781     shader_addline(&buffer, "END\n");
5782
5783     /* Generate the shader */
5784     GL_EXTCALL(glGenProgramsARB(1, &ret));
5785     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, ret));
5786     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
5787             strlen(buffer.buffer), buffer.buffer));
5788     checkGLcall("glProgramStringARB()");
5789
5790     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
5791     if (pos != -1)
5792     {
5793         FIXME("Fragment program error at position %d: %s\n", pos,
5794               debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
5795     }
5796     else
5797     {
5798         GLint native;
5799
5800         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
5801         checkGLcall("glGetProgramivARB()");
5802         if (!native) WARN("Program exceeds native resource limits.\n");
5803     }
5804
5805     shader_buffer_free(&buffer);
5806     return ret;
5807 }
5808
5809 static void fragment_prog_arbfp(DWORD state, IWineD3DStateBlockImpl *stateblock, struct wined3d_context *context)
5810 {
5811     IWineD3DDeviceImpl *device = stateblock->wineD3DDevice;
5812     struct shader_arb_priv *priv = device->fragment_priv;
5813     BOOL use_pshader = use_ps(stateblock);
5814     BOOL use_vshader = use_vs(stateblock);
5815     struct ffp_frag_settings settings;
5816     const struct arbfp_ffp_desc *desc;
5817     unsigned int i;
5818
5819     TRACE("state %#x, stateblock %p, context %p\n", state, stateblock, context);
5820
5821     if(isStateDirty(context, STATE_RENDER(WINED3DRS_FOGENABLE))) {
5822         if(!use_pshader && device->shader_backend == &arb_program_shader_backend && context->last_was_pshader) {
5823             /* Reload fixed function constants since they collide with the pixel shader constants */
5824             for(i = 0; i < MAX_TEXTURES; i++) {
5825                 set_bumpmat_arbfp(STATE_TEXTURESTAGE(i, WINED3DTSS_BUMPENVMAT00), stateblock, context);
5826             }
5827             state_texfactor_arbfp(STATE_RENDER(WINED3DRS_TEXTUREFACTOR), stateblock, context);
5828             state_arb_specularenable(STATE_RENDER(WINED3DRS_SPECULARENABLE), stateblock, context);
5829         } else if(use_pshader && !isStateDirty(context, device->StateTable[STATE_VSHADER].representative)) {
5830             device->shader_backend->shader_select(context, use_pshader, use_vshader);
5831         }
5832         return;
5833     }
5834
5835     if(!use_pshader) {
5836         /* Find or create a shader implementing the fixed function pipeline settings, then activate it */
5837         gen_ffp_frag_op(stateblock, &settings, FALSE);
5838         desc = (const struct arbfp_ffp_desc *)find_ffp_frag_shader(&priv->fragment_shaders, &settings);
5839         if(!desc) {
5840             struct arbfp_ffp_desc *new_desc = HeapAlloc(GetProcessHeap(), 0, sizeof(*new_desc));
5841             if (!new_desc)
5842             {
5843                 ERR("Out of memory\n");
5844                 return;
5845             }
5846             new_desc->num_textures_used = 0;
5847             for(i = 0; i < GL_LIMITS(texture_stages); i++) {
5848                 if(settings.op[i].cop == WINED3DTOP_DISABLE) break;
5849                 new_desc->num_textures_used = i;
5850             }
5851
5852             memcpy(&new_desc->parent.settings, &settings, sizeof(settings));
5853             new_desc->shader = gen_arbfp_ffp_shader(&settings, stateblock);
5854             add_ffp_frag_shader(&priv->fragment_shaders, &new_desc->parent);
5855             TRACE("Allocated fixed function replacement shader descriptor %p\n", new_desc);
5856             desc = new_desc;
5857         }
5858
5859         /* Now activate the replacement program. GL_FRAGMENT_PROGRAM_ARB is already active(however, note the
5860          * comment above the shader_select call below). If e.g. GLSL is active, the shader_select call will
5861          * deactivate it.
5862          */
5863         GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, desc->shader));
5864         checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, desc->shader)");
5865         priv->current_fprogram_id = desc->shader;
5866
5867         if(device->shader_backend == &arb_program_shader_backend && context->last_was_pshader) {
5868             /* Reload fixed function constants since they collide with the pixel shader constants */
5869             for(i = 0; i < MAX_TEXTURES; i++) {
5870                 set_bumpmat_arbfp(STATE_TEXTURESTAGE(i, WINED3DTSS_BUMPENVMAT00), stateblock, context);
5871             }
5872             state_texfactor_arbfp(STATE_RENDER(WINED3DRS_TEXTUREFACTOR), stateblock, context);
5873             state_arb_specularenable(STATE_RENDER(WINED3DRS_SPECULARENABLE), stateblock, context);
5874         }
5875         context->last_was_pshader = FALSE;
5876     } else {
5877         context->last_was_pshader = TRUE;
5878     }
5879
5880     /* Finally, select the shader. If a pixel shader is used, it will be set and enabled by the shader backend.
5881      * If this shader backend is arbfp(most likely), then it will simply overwrite the last fixed function replace-
5882      * ment shader. If the shader backend is not ARB, it currently is important that the opengl implementation
5883      * type overwrites GL_ARB_fragment_program. This is currently the case with GLSL. If we really want to use
5884      * atifs or nvrc pixel shaders with arb fragment programs we'd have to disable GL_FRAGMENT_PROGRAM_ARB here
5885      *
5886      * Don't call shader_select if the vertex shader is dirty, because it will be called later on by the vertex
5887      * shader handler
5888      */
5889     if(!isStateDirty(context, device->StateTable[STATE_VSHADER].representative)) {
5890         device->shader_backend->shader_select(context, use_pshader, use_vshader);
5891
5892         if (!isStateDirty(context, STATE_VERTEXSHADERCONSTANT) && (use_vshader || use_pshader)) {
5893             device->StateTable[STATE_VERTEXSHADERCONSTANT].apply(STATE_VERTEXSHADERCONSTANT, stateblock, context);
5894         }
5895     }
5896     if(use_pshader) {
5897         device->StateTable[STATE_PIXELSHADERCONSTANT].apply(STATE_PIXELSHADERCONSTANT, stateblock, context);
5898     }
5899 }
5900
5901 /* We can't link the fog states to the fragment state directly since the vertex pipeline links them
5902  * to FOGENABLE. A different linking in different pipeline parts can't be expressed in the combined
5903  * state table, so we need to handle that with a forwarding function. The other invisible side effect
5904  * is that changing the fog start and fog end(which links to FOGENABLE in vertex) results in the
5905  * fragment_prog_arbfp function being called because FOGENABLE is dirty, which calls this function here
5906  */
5907 static void state_arbfp_fog(DWORD state, IWineD3DStateBlockImpl *stateblock, struct wined3d_context *context)
5908 {
5909     enum fogsource new_source;
5910
5911     TRACE("state %#x, stateblock %p, context %p\n", state, stateblock, context);
5912
5913     if(!isStateDirty(context, STATE_PIXELSHADER)) {
5914         fragment_prog_arbfp(state, stateblock, context);
5915     }
5916
5917     if(!stateblock->renderState[WINED3DRS_FOGENABLE]) return;
5918
5919     if(stateblock->renderState[WINED3DRS_FOGTABLEMODE] == WINED3DFOG_NONE) {
5920         if(use_vs(stateblock)) {
5921             new_source = FOGSOURCE_VS;
5922         } else {
5923             if(stateblock->renderState[WINED3DRS_FOGVERTEXMODE] == WINED3DFOG_NONE || context->last_was_rhw) {
5924                 new_source = FOGSOURCE_COORD;
5925             } else {
5926                 new_source = FOGSOURCE_FFP;
5927             }
5928         }
5929     } else {
5930         new_source = FOGSOURCE_FFP;
5931     }
5932     if(new_source != context->fog_source) {
5933         context->fog_source = new_source;
5934         state_fogstartend(STATE_RENDER(WINED3DRS_FOGSTART), stateblock, context);
5935     }
5936 }
5937
5938 static void textransform(DWORD state, IWineD3DStateBlockImpl *stateblock, struct wined3d_context *context)
5939 {
5940     if(!isStateDirty(context, STATE_PIXELSHADER)) {
5941         fragment_prog_arbfp(state, stateblock, context);
5942     }
5943 }
5944
5945 #undef GLINFO_LOCATION
5946
5947 static const struct StateEntryTemplate arbfp_fragmentstate_template[] = {
5948     {STATE_RENDER(WINED3DRS_TEXTUREFACTOR),               { STATE_RENDER(WINED3DRS_TEXTUREFACTOR),              state_texfactor_arbfp   }, WINED3D_GL_EXT_NONE             },
5949     {STATE_TEXTURESTAGE(0, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5950     {STATE_TEXTURESTAGE(0, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5951     {STATE_TEXTURESTAGE(0, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5952     {STATE_TEXTURESTAGE(0, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5953     {STATE_TEXTURESTAGE(0, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5954     {STATE_TEXTURESTAGE(0, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5955     {STATE_TEXTURESTAGE(0, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5956     {STATE_TEXTURESTAGE(0, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5957     {STATE_TEXTURESTAGE(0, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5958     {STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5959     {STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5960     {STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5961     {STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5962     {STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
5963     {STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
5964     {STATE_TEXTURESTAGE(1, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5965     {STATE_TEXTURESTAGE(1, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5966     {STATE_TEXTURESTAGE(1, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5967     {STATE_TEXTURESTAGE(1, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5968     {STATE_TEXTURESTAGE(1, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5969     {STATE_TEXTURESTAGE(1, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5970     {STATE_TEXTURESTAGE(1, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5971     {STATE_TEXTURESTAGE(1, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5972     {STATE_TEXTURESTAGE(1, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5973     {STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5974     {STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5975     {STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5976     {STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5977     {STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
5978     {STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
5979     {STATE_TEXTURESTAGE(2, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5980     {STATE_TEXTURESTAGE(2, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5981     {STATE_TEXTURESTAGE(2, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5982     {STATE_TEXTURESTAGE(2, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5983     {STATE_TEXTURESTAGE(2, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5984     {STATE_TEXTURESTAGE(2, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5985     {STATE_TEXTURESTAGE(2, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5986     {STATE_TEXTURESTAGE(2, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5987     {STATE_TEXTURESTAGE(2, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5988     {STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5989     {STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5990     {STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5991     {STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5992     {STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
5993     {STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
5994     {STATE_TEXTURESTAGE(3, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5995     {STATE_TEXTURESTAGE(3, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5996     {STATE_TEXTURESTAGE(3, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5997     {STATE_TEXTURESTAGE(3, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5998     {STATE_TEXTURESTAGE(3, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5999     {STATE_TEXTURESTAGE(3, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6000     {STATE_TEXTURESTAGE(3, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6001     {STATE_TEXTURESTAGE(3, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6002     {STATE_TEXTURESTAGE(3, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6003     {STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6004     {STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6005     {STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6006     {STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6007     {STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6008     {STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6009     {STATE_TEXTURESTAGE(4, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6010     {STATE_TEXTURESTAGE(4, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6011     {STATE_TEXTURESTAGE(4, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6012     {STATE_TEXTURESTAGE(4, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6013     {STATE_TEXTURESTAGE(4, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6014     {STATE_TEXTURESTAGE(4, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6015     {STATE_TEXTURESTAGE(4, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6016     {STATE_TEXTURESTAGE(4, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6017     {STATE_TEXTURESTAGE(4, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6018     {STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6019     {STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6020     {STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6021     {STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6022     {STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6023     {STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6024     {STATE_TEXTURESTAGE(5, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6025     {STATE_TEXTURESTAGE(5, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6026     {STATE_TEXTURESTAGE(5, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6027     {STATE_TEXTURESTAGE(5, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6028     {STATE_TEXTURESTAGE(5, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6029     {STATE_TEXTURESTAGE(5, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6030     {STATE_TEXTURESTAGE(5, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6031     {STATE_TEXTURESTAGE(5, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6032     {STATE_TEXTURESTAGE(5, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6033     {STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6034     {STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6035     {STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6036     {STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6037     {STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6038     {STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6039     {STATE_TEXTURESTAGE(6, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6040     {STATE_TEXTURESTAGE(6, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6041     {STATE_TEXTURESTAGE(6, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6042     {STATE_TEXTURESTAGE(6, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6043     {STATE_TEXTURESTAGE(6, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6044     {STATE_TEXTURESTAGE(6, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6045     {STATE_TEXTURESTAGE(6, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6046     {STATE_TEXTURESTAGE(6, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6047     {STATE_TEXTURESTAGE(6, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6048     {STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6049     {STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6050     {STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6051     {STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6052     {STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6053     {STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6054     {STATE_TEXTURESTAGE(7, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6055     {STATE_TEXTURESTAGE(7, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6056     {STATE_TEXTURESTAGE(7, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6057     {STATE_TEXTURESTAGE(7, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6058     {STATE_TEXTURESTAGE(7, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6059     {STATE_TEXTURESTAGE(7, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6060     {STATE_TEXTURESTAGE(7, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6061     {STATE_TEXTURESTAGE(7, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6062     {STATE_TEXTURESTAGE(7, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6063     {STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6064     {STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6065     {STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6066     {STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6067     {STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6068     {STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6069     {STATE_SAMPLER(0),                                    { STATE_SAMPLER(0),                                   sampler_texdim          }, WINED3D_GL_EXT_NONE             },
6070     {STATE_SAMPLER(1),                                    { STATE_SAMPLER(1),                                   sampler_texdim          }, WINED3D_GL_EXT_NONE             },
6071     {STATE_SAMPLER(2),                                    { STATE_SAMPLER(2),                                   sampler_texdim          }, WINED3D_GL_EXT_NONE             },
6072     {STATE_SAMPLER(3),                                    { STATE_SAMPLER(3),                                   sampler_texdim          }, WINED3D_GL_EXT_NONE             },
6073     {STATE_SAMPLER(4),                                    { STATE_SAMPLER(4),                                   sampler_texdim          }, WINED3D_GL_EXT_NONE             },
6074     {STATE_SAMPLER(5),                                    { STATE_SAMPLER(5),                                   sampler_texdim          }, WINED3D_GL_EXT_NONE             },
6075     {STATE_SAMPLER(6),                                    { STATE_SAMPLER(6),                                   sampler_texdim          }, WINED3D_GL_EXT_NONE             },
6076     {STATE_SAMPLER(7),                                    { STATE_SAMPLER(7),                                   sampler_texdim          }, WINED3D_GL_EXT_NONE             },
6077     {STATE_PIXELSHADER,                                   { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6078     {STATE_RENDER(WINED3DRS_FOGENABLE),                   { STATE_RENDER(WINED3DRS_FOGENABLE),                  state_arbfp_fog         }, WINED3D_GL_EXT_NONE             },
6079     {STATE_RENDER(WINED3DRS_FOGTABLEMODE),                { STATE_RENDER(WINED3DRS_FOGENABLE),                  state_arbfp_fog         }, WINED3D_GL_EXT_NONE             },
6080     {STATE_RENDER(WINED3DRS_FOGVERTEXMODE),               { STATE_RENDER(WINED3DRS_FOGENABLE),                  state_arbfp_fog         }, WINED3D_GL_EXT_NONE             },
6081     {STATE_RENDER(WINED3DRS_FOGSTART),                    { STATE_RENDER(WINED3DRS_FOGSTART),                   state_fogstartend       }, WINED3D_GL_EXT_NONE             },
6082     {STATE_RENDER(WINED3DRS_FOGEND),                      { STATE_RENDER(WINED3DRS_FOGSTART),                   state_fogstartend       }, WINED3D_GL_EXT_NONE             },
6083     {STATE_RENDER(WINED3DRS_SRGBWRITEENABLE),             { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6084     {STATE_RENDER(WINED3DRS_FOGCOLOR),                    { STATE_RENDER(WINED3DRS_FOGCOLOR),                   state_fogcolor          }, WINED3D_GL_EXT_NONE             },
6085     {STATE_RENDER(WINED3DRS_FOGDENSITY),                  { STATE_RENDER(WINED3DRS_FOGDENSITY),                 state_fogdensity        }, WINED3D_GL_EXT_NONE             },
6086     {STATE_TEXTURESTAGE(0,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(0, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6087     {STATE_TEXTURESTAGE(1,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(1, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6088     {STATE_TEXTURESTAGE(2,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(2, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6089     {STATE_TEXTURESTAGE(3,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(3, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6090     {STATE_TEXTURESTAGE(4,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(4, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6091     {STATE_TEXTURESTAGE(5,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(5, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6092     {STATE_TEXTURESTAGE(6,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(6, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6093     {STATE_TEXTURESTAGE(7,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(7, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6094     {STATE_RENDER(WINED3DRS_SPECULARENABLE),              { STATE_RENDER(WINED3DRS_SPECULARENABLE),             state_arb_specularenable}, WINED3D_GL_EXT_NONE             },
6095     {0 /* Terminate */,                                   { 0,                                                  0                       }, WINED3D_GL_EXT_NONE             },
6096 };
6097
6098 const struct fragment_pipeline arbfp_fragment_pipeline = {
6099     arbfp_enable,
6100     arbfp_get_caps,
6101     arbfp_alloc,
6102     arbfp_free,
6103     shader_arb_color_fixup_supported,
6104     arbfp_fragmentstate_template,
6105     TRUE /* We can disable projected textures */
6106 };
6107
6108 #define GLINFO_LOCATION device->adapter->gl_info
6109
6110 struct arbfp_blit_priv {
6111     GLenum yuy2_rect_shader, yuy2_2d_shader;
6112     GLenum uyvy_rect_shader, uyvy_2d_shader;
6113     GLenum yv12_rect_shader, yv12_2d_shader;
6114 };
6115
6116 static HRESULT arbfp_blit_alloc(IWineD3DDevice *iface) {
6117     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *) iface;
6118     device->blit_priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(struct arbfp_blit_priv));
6119     if(!device->blit_priv) {
6120         ERR("Out of memory\n");
6121         return E_OUTOFMEMORY;
6122     }
6123     return WINED3D_OK;
6124 }
6125
6126 /* Context activation is done by the caller. */
6127 static void arbfp_blit_free(IWineD3DDevice *iface) {
6128     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *) iface;
6129     struct arbfp_blit_priv *priv = device->blit_priv;
6130
6131     ENTER_GL();
6132     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yuy2_rect_shader));
6133     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yuy2_2d_shader));
6134     GL_EXTCALL(glDeleteProgramsARB(1, &priv->uyvy_rect_shader));
6135     GL_EXTCALL(glDeleteProgramsARB(1, &priv->uyvy_2d_shader));
6136     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yv12_rect_shader));
6137     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yv12_2d_shader));
6138     checkGLcall("Delete yuv programs");
6139     LEAVE_GL();
6140 }
6141
6142 static BOOL gen_planar_yuv_read(struct wined3d_shader_buffer *buffer, enum yuv_fixup yuv_fixup,
6143         GLenum textype, char *luminance)
6144 {
6145     char chroma;
6146     const char *tex, *texinstr;
6147
6148     if (yuv_fixup == YUV_FIXUP_UYVY) {
6149         chroma = 'x';
6150         *luminance = 'w';
6151     } else {
6152         chroma = 'w';
6153         *luminance = 'x';
6154     }
6155     switch(textype) {
6156         case GL_TEXTURE_2D:             tex = "2D";     texinstr = "TXP"; break;
6157         case GL_TEXTURE_RECTANGLE_ARB:  tex = "RECT";   texinstr = "TEX"; break;
6158         default:
6159             /* This is more tricky than just replacing the texture type - we have to navigate
6160              * properly in the texture to find the correct chroma values
6161              */
6162             FIXME("Implement yuv correction for non-2d, non-rect textures\n");
6163             return FALSE;
6164     }
6165
6166     /* First we have to read the chroma values. This means we need at least two pixels(no filtering),
6167      * or 4 pixels(with filtering). To get the unmodified chromas, we have to rid ourselves of the
6168      * filtering when we sample the texture.
6169      *
6170      * These are the rules for reading the chroma:
6171      *
6172      * Even pixel: Cr
6173      * Even pixel: U
6174      * Odd pixel: V
6175      *
6176      * So we have to get the sampling x position in non-normalized coordinates in integers
6177      */
6178     if(textype != GL_TEXTURE_RECTANGLE_ARB) {
6179         shader_addline(buffer, "MUL texcrd.xy, fragment.texcoord[0], size.x;\n");
6180         shader_addline(buffer, "MOV texcrd.w, size.x;\n");
6181     } else {
6182         shader_addline(buffer, "MOV texcrd, fragment.texcoord[0];\n");
6183     }
6184     /* We must not allow filtering between pixel x and x+1, this would mix U and V
6185      * Vertical filtering is ok. However, bear in mind that the pixel center is at
6186      * 0.5, so add 0.5.
6187      */
6188     shader_addline(buffer, "FLR texcrd.x, texcrd.x;\n");
6189     shader_addline(buffer, "ADD texcrd.x, texcrd.x, coef.y;\n");
6190
6191     /* Divide the x coordinate by 0.5 and get the fraction. This gives 0.25 and 0.75 for the
6192      * even and odd pixels respectively
6193      */
6194     shader_addline(buffer, "MUL texcrd2, texcrd, coef.y;\n");
6195     shader_addline(buffer, "FRC texcrd2, texcrd2;\n");
6196
6197     /* Sample Pixel 1 */
6198     shader_addline(buffer, "%s luminance, texcrd, texture[0], %s;\n", texinstr, tex);
6199
6200     /* Put the value into either of the chroma values */
6201     shader_addline(buffer, "SGE temp.x, texcrd2.x, coef.y;\n");
6202     shader_addline(buffer, "MUL chroma.x, luminance.%c, temp.x;\n", chroma);
6203     shader_addline(buffer, "SLT temp.x, texcrd2.x, coef.y;\n");
6204     shader_addline(buffer, "MUL chroma.y, luminance.%c, temp.x;\n", chroma);
6205
6206     /* Sample pixel 2. If we read an even pixel(SLT above returned 1), sample
6207      * the pixel right to the current one. Otherwise, sample the left pixel.
6208      * Bias and scale the SLT result to -1;1 and add it to the texcrd.x.
6209      */
6210     shader_addline(buffer, "MAD temp.x, temp.x, coef.z, -coef.x;\n");
6211     shader_addline(buffer, "ADD texcrd.x, texcrd, temp.x;\n");
6212     shader_addline(buffer, "%s luminance, texcrd, texture[0], %s;\n", texinstr, tex);
6213
6214     /* Put the value into the other chroma */
6215     shader_addline(buffer, "SGE temp.x, texcrd2.x, coef.y;\n");
6216     shader_addline(buffer, "MAD chroma.y, luminance.%c, temp.x, chroma.y;\n", chroma);
6217     shader_addline(buffer, "SLT temp.x, texcrd2.x, coef.y;\n");
6218     shader_addline(buffer, "MAD chroma.x, luminance.%c, temp.x, chroma.x;\n", chroma);
6219
6220     /* TODO: If filtering is enabled, sample a 2nd pair of pixels left or right of
6221      * the current one and lerp the two U and V values
6222      */
6223
6224     /* This gives the correctly filtered luminance value */
6225     shader_addline(buffer, "TEX luminance, fragment.texcoord[0], texture[0], %s;\n", tex);
6226
6227     return TRUE;
6228 }
6229
6230 static BOOL gen_yv12_read(struct wined3d_shader_buffer *buffer, GLenum textype, char *luminance)
6231 {
6232     const char *tex;
6233
6234     switch(textype) {
6235         case GL_TEXTURE_2D:             tex = "2D";     break;
6236         case GL_TEXTURE_RECTANGLE_ARB:  tex = "RECT";   break;
6237         default:
6238             FIXME("Implement yv12 correction for non-2d, non-rect textures\n");
6239             return FALSE;
6240     }
6241
6242     /* YV12 surfaces contain a WxH sized luminance plane, followed by a (W/2)x(H/2)
6243      * V and a (W/2)x(H/2) U plane, each with 8 bit per pixel. So the effective
6244      * bitdepth is 12 bits per pixel. Since the U and V planes have only half the
6245      * pitch of the luminance plane, the packing into the gl texture is a bit
6246      * unfortunate. If the whole texture is interpreted as luminance data it looks
6247      * approximately like this:
6248      *
6249      *        +----------------------------------+----
6250      *        |                                  |
6251      *        |                                  |
6252      *        |                                  |
6253      *        |                                  |
6254      *        |                                  |   2
6255      *        |            LUMINANCE             |   -
6256      *        |                                  |   3
6257      *        |                                  |
6258      *        |                                  |
6259      *        |                                  |
6260      *        |                                  |
6261      *        +----------------+-----------------+----
6262      *        |                |                 |
6263      *        |  U even rows   |  U odd rows     |
6264      *        |                |                 |   1
6265      *        +----------------+------------------   -
6266      *        |                |                 |   3
6267      *        |  V even rows   |  V odd rows     |
6268      *        |                |                 |
6269      *        +----------------+-----------------+----
6270      *        |                |                 |
6271      *        |     0.5        |       0.5       |
6272      *
6273      * So it appears as if there are 4 chroma images, but in fact the odd rows
6274      * in the chroma images are in the same row as the even ones. So its is
6275      * kinda tricky to read
6276      *
6277      * When reading from rectangle textures, keep in mind that the input y coordinates
6278      * go from 0 to d3d_height, whereas the opengl texture height is 1.5 * d3d_height
6279      */
6280     shader_addline(buffer, "PARAM yv12_coef = {%f, %f, %f, %f};\n",
6281             2.0f / 3.0f, 1.0f / 6.0f, (2.0f / 3.0f) + (1.0f / 6.0f), 1.0f / 3.0f);
6282
6283     shader_addline(buffer, "MOV texcrd, fragment.texcoord[0];\n");
6284     /* the chroma planes have only half the width */
6285     shader_addline(buffer, "MUL texcrd.x, texcrd.x, coef.y;\n");
6286
6287     /* The first value is between 2/3 and 5/6th of the texture's height, so scale+bias
6288      * the coordinate. Also read the right side of the image when reading odd lines
6289      *
6290      * Don't forget to clamp the y values in into the range, otherwise we'll get filtering
6291      * bleeding
6292      */
6293     if(textype == GL_TEXTURE_2D) {
6294
6295         shader_addline(buffer, "RCP chroma.w, size.y;\n");
6296
6297         shader_addline(buffer, "MUL texcrd2.y, texcrd.y, size.y;\n");
6298
6299         shader_addline(buffer, "FLR texcrd2.y, texcrd2.y;\n");
6300         shader_addline(buffer, "MAD texcrd.y, texcrd.y, yv12_coef.y, yv12_coef.x;\n");
6301
6302         /* Read odd lines from the right side(add size * 0.5 to the x coordinate */
6303         shader_addline(buffer, "ADD texcrd2.x, texcrd2.y, yv12_coef.y;\n"); /* To avoid 0.5 == 0.5 comparisons */
6304         shader_addline(buffer, "FRC texcrd2.x, texcrd2.x;\n");
6305         shader_addline(buffer, "SGE texcrd2.x, texcrd2.x, coef.y;\n");
6306         shader_addline(buffer, "MAD texcrd.x, texcrd2.x, coef.y, texcrd.x;\n");
6307
6308         /* clamp, keep the half pixel origin in mind */
6309         shader_addline(buffer, "MAD temp.y, coef.y, chroma.w, yv12_coef.x;\n");
6310         shader_addline(buffer, "MAX texcrd.y, temp.y, texcrd.y;\n");
6311         shader_addline(buffer, "MAD temp.y, -coef.y, chroma.w, yv12_coef.z;\n");
6312         shader_addline(buffer, "MIN texcrd.y, temp.y, texcrd.y;\n");
6313     } else {
6314         /* Read from [size - size+size/4] */
6315         shader_addline(buffer, "FLR texcrd.y, texcrd.y;\n");
6316         shader_addline(buffer, "MAD texcrd.y, texcrd.y, coef.w, size.y;\n");
6317
6318         /* Read odd lines from the right side(add size * 0.5 to the x coordinate */
6319         shader_addline(buffer, "ADD texcrd2.x, texcrd.y, yv12_coef.y;\n"); /* To avoid 0.5 == 0.5 comparisons */
6320         shader_addline(buffer, "FRC texcrd2.x, texcrd2.x;\n");
6321         shader_addline(buffer, "SGE texcrd2.x, texcrd2.x, coef.y;\n");
6322         shader_addline(buffer, "MUL texcrd2.x, texcrd2.x, size.x;\n");
6323         shader_addline(buffer, "MAD texcrd.x, texcrd2.x, coef.y, texcrd.x;\n");
6324
6325         /* Make sure to read exactly from the pixel center */
6326         shader_addline(buffer, "FLR texcrd.y, texcrd.y;\n");
6327         shader_addline(buffer, "ADD texcrd.y, texcrd.y, coef.y;\n");
6328
6329         /* Clamp */
6330         shader_addline(buffer, "MAD temp.y, size.y, coef.w, size.y;\n");
6331         shader_addline(buffer, "ADD temp.y, temp.y, -coef.y;\n");
6332         shader_addline(buffer, "MIN texcrd.y, temp.y, texcrd.y;\n");
6333         shader_addline(buffer, "ADD temp.y, size.y, -coef.y;\n");
6334         shader_addline(buffer, "MAX texcrd.y, temp.y, texcrd.y;\n");
6335     }
6336     /* Read the texture, put the result into the output register */
6337     shader_addline(buffer, "TEX temp, texcrd, texture[0], %s;\n", tex);
6338     shader_addline(buffer, "MOV chroma.x, temp.w;\n");
6339
6340     /* The other chroma value is 1/6th of the texture lower, from 5/6th to 6/6th
6341      * No need to clamp because we're just reusing the already clamped value from above
6342      */
6343     if(textype == GL_TEXTURE_2D) {
6344         shader_addline(buffer, "ADD texcrd.y, texcrd.y, yv12_coef.y;\n");
6345     } else {
6346         shader_addline(buffer, "MAD texcrd.y, size.y, coef.w, texcrd.y;\n");
6347     }
6348     shader_addline(buffer, "TEX temp, texcrd, texture[0], %s;\n", tex);
6349     shader_addline(buffer, "MOV chroma.y, temp.w;\n");
6350
6351     /* Sample the luminance value. It is in the top 2/3rd of the texture, so scale the y coordinate.
6352      * Clamp the y coordinate to prevent the chroma values from bleeding into the sampled luminance
6353      * values due to filtering
6354      */
6355     shader_addline(buffer, "MOV texcrd, fragment.texcoord[0];\n");
6356     if(textype == GL_TEXTURE_2D) {
6357         /* Multiply the y coordinate by 2/3 and clamp it */
6358         shader_addline(buffer, "MUL texcrd.y, texcrd.y, yv12_coef.x;\n");
6359         shader_addline(buffer, "MAD temp.y, -coef.y, chroma.w, yv12_coef.x;\n");
6360         shader_addline(buffer, "MIN texcrd.y, temp.y, texcrd.y;\n");
6361         shader_addline(buffer, "TEX luminance, texcrd, texture[0], %s;\n", tex);
6362     } else {
6363         /* Reading from texture_rectangles is pretty straightforward, just use the unmodified
6364          * texture coordinate. It is still a good idea to clamp it though, since the opengl texture
6365          * is bigger
6366          */
6367         shader_addline(buffer, "ADD temp.x, size.y, -coef.y;\n");
6368         shader_addline(buffer, "MIN texcrd.y, texcrd.y, size.x;\n");
6369         shader_addline(buffer, "TEX luminance, texcrd, texture[0], %s;\n", tex);
6370     }
6371     *luminance = 'a';
6372
6373     return TRUE;
6374 }
6375
6376 /* Context activation is done by the caller. */
6377 static GLuint gen_yuv_shader(IWineD3DDeviceImpl *device, enum yuv_fixup yuv_fixup, GLenum textype)
6378 {
6379     GLenum shader;
6380     struct wined3d_shader_buffer buffer;
6381     char luminance_component;
6382     struct arbfp_blit_priv *priv = device->blit_priv;
6383     GLint pos;
6384
6385     /* Shader header */
6386     if (!shader_buffer_init(&buffer))
6387     {
6388         ERR("Failed to initialize shader buffer.\n");
6389         return 0;
6390     }
6391
6392     ENTER_GL();
6393     GL_EXTCALL(glGenProgramsARB(1, &shader));
6394     checkGLcall("GL_EXTCALL(glGenProgramsARB(1, &shader))");
6395     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader));
6396     checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader)");
6397     LEAVE_GL();
6398     if(!shader) {
6399         shader_buffer_free(&buffer);
6400         return 0;
6401     }
6402
6403     /* The YUY2 and UYVY formats contain two pixels packed into a 32 bit macropixel,
6404      * giving effectively 16 bit per pixel. The color consists of a luminance(Y) and
6405      * two chroma(U and V) values. Each macropixel has two luminance values, one for
6406      * each single pixel it contains, and one U and one V value shared between both
6407      * pixels.
6408      *
6409      * The data is loaded into an A8L8 texture. With YUY2, the luminance component
6410      * contains the luminance and alpha the chroma. With UYVY it is vice versa. Thus
6411      * take the format into account when generating the read swizzles
6412      *
6413      * Reading the Y value is straightforward - just sample the texture. The hardware
6414      * takes care of filtering in the horizontal and vertical direction.
6415      *
6416      * Reading the U and V values is harder. We have to avoid filtering horizontally,
6417      * because that would mix the U and V values of one pixel or two adjacent pixels.
6418      * Thus floor the texture coordinate and add 0.5 to get an unfiltered read,
6419      * regardless of the filtering setting. Vertical filtering works automatically
6420      * though - the U and V values of two rows are mixed nicely.
6421      *
6422      * Appart of avoiding filtering issues, the code has to know which value it just
6423      * read, and where it can find the other one. To determine this, it checks if
6424      * it sampled an even or odd pixel, and shifts the 2nd read accordingly.
6425      *
6426      * Handling horizontal filtering of U and V values requires reading a 2nd pair
6427      * of pixels, extracting U and V and mixing them. This is not implemented yet.
6428      *
6429      * An alternative implementation idea is to load the texture as A8R8G8B8 texture,
6430      * with width / 2. This way one read gives all 3 values, finding U and V is easy
6431      * in an unfiltered situation. Finding the luminance on the other hand requires
6432      * finding out if it is an odd or even pixel. The real drawback of this approach
6433      * is filtering. This would have to be emulated completely in the shader, reading
6434      * up two 2 packed pixels in up to 2 rows and interpolating both horizontally and
6435      * vertically. Beyond that it would require adjustments to the texture handling
6436      * code to deal with the width scaling
6437      */
6438     shader_addline(&buffer, "!!ARBfp1.0\n");
6439     shader_addline(&buffer, "TEMP luminance;\n");
6440     shader_addline(&buffer, "TEMP temp;\n");
6441     shader_addline(&buffer, "TEMP chroma;\n");
6442     shader_addline(&buffer, "TEMP texcrd;\n");
6443     shader_addline(&buffer, "TEMP texcrd2;\n");
6444     shader_addline(&buffer, "PARAM coef = {1.0, 0.5, 2.0, 0.25};\n");
6445     shader_addline(&buffer, "PARAM yuv_coef = {1.403, 0.344, 0.714, 1.770};\n");
6446     shader_addline(&buffer, "PARAM size = program.local[0];\n");
6447
6448     switch (yuv_fixup)
6449     {
6450         case YUV_FIXUP_UYVY:
6451         case YUV_FIXUP_YUY2:
6452             if (!gen_planar_yuv_read(&buffer, yuv_fixup, textype, &luminance_component))
6453             {
6454                 shader_buffer_free(&buffer);
6455                 return 0;
6456             }
6457             break;
6458
6459         case YUV_FIXUP_YV12:
6460             if (!gen_yv12_read(&buffer, textype, &luminance_component))
6461             {
6462                 shader_buffer_free(&buffer);
6463                 return 0;
6464             }
6465             break;
6466
6467         default:
6468             FIXME("Unsupported YUV fixup %#x\n", yuv_fixup);
6469             shader_buffer_free(&buffer);
6470             return 0;
6471     }
6472
6473     /* Calculate the final result. Formula is taken from
6474      * http://www.fourcc.org/fccyvrgb.php. Note that the chroma
6475      * ranges from -0.5 to 0.5
6476      */
6477     shader_addline(&buffer, "SUB chroma.xy, chroma, coef.y;\n");
6478
6479     shader_addline(&buffer, "MAD result.color.x, chroma.x, yuv_coef.x, luminance.%c;\n", luminance_component);
6480     shader_addline(&buffer, "MAD temp.x, -chroma.y, yuv_coef.y, luminance.%c;\n", luminance_component);
6481     shader_addline(&buffer, "MAD result.color.y, -chroma.x, yuv_coef.z, temp.x;\n");
6482     shader_addline(&buffer, "MAD result.color.z, chroma.y, yuv_coef.w, luminance.%c;\n", luminance_component);
6483     shader_addline(&buffer, "END\n");
6484
6485     ENTER_GL();
6486     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
6487             strlen(buffer.buffer), buffer.buffer));
6488     checkGLcall("glProgramStringARB()");
6489
6490     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
6491     if (pos != -1)
6492     {
6493         FIXME("Fragment program error at position %d: %s\n", pos,
6494               debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
6495     }
6496     else
6497     {
6498         GLint native;
6499
6500         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
6501         checkGLcall("glGetProgramivARB()");
6502         if (!native) WARN("Program exceeds native resource limits.\n");
6503     }
6504
6505     shader_buffer_free(&buffer);
6506     LEAVE_GL();
6507
6508     switch (yuv_fixup)
6509     {
6510         case YUV_FIXUP_YUY2:
6511             if (textype == GL_TEXTURE_RECTANGLE_ARB) priv->yuy2_rect_shader = shader;
6512             else priv->yuy2_2d_shader = shader;
6513             break;
6514
6515         case YUV_FIXUP_UYVY:
6516             if (textype == GL_TEXTURE_RECTANGLE_ARB) priv->uyvy_rect_shader = shader;
6517             else priv->uyvy_2d_shader = shader;
6518             break;
6519
6520         case YUV_FIXUP_YV12:
6521             if (textype == GL_TEXTURE_RECTANGLE_ARB) priv->yv12_rect_shader = shader;
6522             else priv->yv12_2d_shader = shader;
6523             break;
6524     }
6525
6526     return shader;
6527 }
6528
6529 /* Context activation is done by the caller. */
6530 static HRESULT arbfp_blit_set(IWineD3DDevice *iface, const struct GlPixelFormatDesc *format_desc,
6531         GLenum textype, UINT width, UINT height)
6532 {
6533     GLenum shader;
6534     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *) iface;
6535     float size[4] = {width, height, 1, 1};
6536     struct arbfp_blit_priv *priv = device->blit_priv;
6537     enum yuv_fixup yuv_fixup;
6538
6539     if (!is_yuv_fixup(format_desc->color_fixup))
6540     {
6541         TRACE("Fixup:\n");
6542         dump_color_fixup_desc(format_desc->color_fixup);
6543         /* Don't bother setting up a shader for unconverted formats */
6544         ENTER_GL();
6545         glEnable(textype);
6546         checkGLcall("glEnable(textype)");
6547         LEAVE_GL();
6548         return WINED3D_OK;
6549     }
6550
6551     yuv_fixup = get_yuv_fixup(format_desc->color_fixup);
6552
6553     switch(yuv_fixup)
6554     {
6555         case YUV_FIXUP_YUY2:
6556             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->yuy2_rect_shader : priv->yuy2_2d_shader;
6557             break;
6558
6559         case YUV_FIXUP_UYVY:
6560             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->uyvy_rect_shader : priv->uyvy_2d_shader;
6561             break;
6562
6563         case YUV_FIXUP_YV12:
6564             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->yv12_rect_shader : priv->yv12_2d_shader;
6565             break;
6566
6567         default:
6568             FIXME("Unsupported YUV fixup %#x, not setting a shader\n", yuv_fixup);
6569             ENTER_GL();
6570             glEnable(textype);
6571             checkGLcall("glEnable(textype)");
6572             LEAVE_GL();
6573             return E_NOTIMPL;
6574     }
6575
6576     if (!shader) shader = gen_yuv_shader(device, yuv_fixup, textype);
6577
6578     ENTER_GL();
6579     glEnable(GL_FRAGMENT_PROGRAM_ARB);
6580     checkGLcall("glEnable(GL_FRAGMENT_PROGRAM_ARB)");
6581     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader));
6582     checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader)");
6583     GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, 0, size));
6584     checkGLcall("glProgramLocalParameter4fvARB");
6585     LEAVE_GL();
6586
6587     return WINED3D_OK;
6588 }
6589
6590 /* Context activation is done by the caller. */
6591 static void arbfp_blit_unset(IWineD3DDevice *iface) {
6592     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *) iface;
6593
6594     ENTER_GL();
6595     glDisable(GL_FRAGMENT_PROGRAM_ARB);
6596     checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
6597     glDisable(GL_TEXTURE_2D);
6598     checkGLcall("glDisable(GL_TEXTURE_2D)");
6599     if(GL_SUPPORT(ARB_TEXTURE_CUBE_MAP)) {
6600         glDisable(GL_TEXTURE_CUBE_MAP_ARB);
6601         checkGLcall("glDisable(GL_TEXTURE_CUBE_MAP_ARB)");
6602     }
6603     if(GL_SUPPORT(ARB_TEXTURE_RECTANGLE)) {
6604         glDisable(GL_TEXTURE_RECTANGLE_ARB);
6605         checkGLcall("glDisable(GL_TEXTURE_RECTANGLE_ARB)");
6606     }
6607     LEAVE_GL();
6608 }
6609
6610 static BOOL arbfp_blit_color_fixup_supported(struct color_fixup_desc fixup)
6611 {
6612     enum yuv_fixup yuv_fixup;
6613
6614     if (TRACE_ON(d3d_shader) && TRACE_ON(d3d))
6615     {
6616         TRACE("Checking support for fixup:\n");
6617         dump_color_fixup_desc(fixup);
6618     }
6619
6620     if (is_identity_fixup(fixup))
6621     {
6622         TRACE("[OK]\n");
6623         return TRUE;
6624     }
6625
6626     /* We only support YUV conversions. */
6627     if (!is_yuv_fixup(fixup))
6628     {
6629         TRACE("[FAILED]\n");
6630         return FALSE;
6631     }
6632
6633     yuv_fixup = get_yuv_fixup(fixup);
6634     switch(yuv_fixup)
6635     {
6636         case YUV_FIXUP_YUY2:
6637         case YUV_FIXUP_UYVY:
6638         case YUV_FIXUP_YV12:
6639             TRACE("[OK]\n");
6640             return TRUE;
6641
6642         default:
6643             FIXME("Unsupported YUV fixup %#x\n", yuv_fixup);
6644             TRACE("[FAILED]\n");
6645             return FALSE;
6646     }
6647 }
6648
6649 const struct blit_shader arbfp_blit = {
6650     arbfp_blit_alloc,
6651     arbfp_blit_free,
6652     arbfp_blit_set,
6653     arbfp_blit_unset,
6654     arbfp_blit_color_fixup_supported,
6655 };
6656
6657 #undef GLINFO_LOCATION