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 loop_control
  95 {
  96     unsigned int count;
  97     unsigned int start;
  98     int step;
  99 };
 100
 101 struct control_frame
 102 {
 103     struct                          list entry;
 104     enum
 105     {
 106         IF,
 107         IFC,
 108         LOOP,
 109         REP
 110     } type;
 111     BOOL                            muting;
 112     BOOL                            outer_loop;
 113     union
 114     {
 115         unsigned int                loop_no;
 116         unsigned int                ifc_no;
 117     };
 118     struct loop_control             loop_control;
 119     BOOL                            had_else;
 120 };
 121
 122 struct arb_ps_np2fixup_info
 123 {
 124     struct ps_np2fixup_info         super;
 125     /* For ARB we need a offset value:
 126      * With both GLSL and ARB mode the NP2 fixup information (the texture dimensions) are stored in a
 127      * consecutive way (GLSL uses a uniform array). Since ARB doesn't know the notion of a "standalone"
 128      * array we need an offset to the index inside the program local parameter array. */
 129     UINT                            offset;
 130 };
 131
 132 struct arb_ps_compile_args
 133 {
 134     struct ps_compile_args          super;
 135     DWORD                           bools; /* WORD is enough, use DWORD for alignment */
 136     unsigned char                   loop_ctrl[MAX_CONST_I][3];
 137 };
 138
 139 struct stb_const_desc
 140 {
 141     unsigned char           texunit;
 142     UINT                    const_num;
 143 };
 144
 145 struct arb_ps_compiled_shader
 146 {
 147     struct arb_ps_compile_args      args;
 148     struct arb_ps_np2fixup_info     np2fixup_info;
 149     struct stb_const_desc           bumpenvmatconst[MAX_TEXTURES];
 150     struct stb_const_desc           luminanceconst[MAX_TEXTURES];
 151     UINT                            int_consts[MAX_CONST_I];
 152     GLuint                          prgId;
 153     UINT                            ycorrection;
 154     unsigned char                   numbumpenvmatconsts;
 155     char                            num_int_consts;
 156 };
 157
 158 struct arb_vs_compile_args
 159 {
 160     struct vs_compile_args          super;
 161     union
 162     {
 163         struct
 164         {
 165             WORD                    bools;
 166             char                    clip_control[2];
 167         }                           boolclip;
 168         DWORD                       boolclip_compare;
 169     };
 170     DWORD                           ps_signature;
 171     union
 172     {
 173         unsigned char               vertex_samplers[4];
 174         DWORD                       vertex_samplers_compare;
 175     };
 176     unsigned char                   loop_ctrl[MAX_CONST_I][3];
 177 };
 178
 179 struct arb_vs_compiled_shader
 180 {
 181     struct arb_vs_compile_args      args;
 182     GLuint                          prgId;
 183     UINT                            int_consts[MAX_CONST_I];
 184     char                            num_int_consts;
 185     char                            need_color_unclamp;
 186     UINT                            pos_fixup;
 187 };
 188
 189 struct recorded_instruction
 190 {
 191     struct wined3d_shader_instruction ins;
 192     struct list entry;
 193 };
 194
 195 struct shader_arb_ctx_priv
 196 {
 197     char addr_reg[20];
 198     enum
 199     {
 200         /* plain GL_ARB_vertex_program or GL_ARB_fragment_program */
 201         ARB,
 202         /* GL_NV_vertex_progam2_option or GL_NV_fragment_program_option */
 203         NV2,
 204         /* GL_NV_vertex_program3 or GL_NV_fragment_program2 */
 205         NV3
 206     } target_version;
 207
 208     const struct arb_vs_compile_args    *cur_vs_args;
 209     const struct arb_ps_compile_args    *cur_ps_args;
 210     const struct arb_ps_compiled_shader *compiled_fprog;
 211     const struct arb_vs_compiled_shader *compiled_vprog;
 212     struct arb_ps_np2fixup_info         *cur_np2fixup_info;
 213     struct list                         control_frames;
 214     struct list                         record;
 215     BOOL                                recording;
 216     BOOL                                muted;
 217     unsigned int                        num_loops, loop_depth, num_ifcs;
 218     int                                 aL;
 219
 220     unsigned int                        vs_clipplanes;
 221     BOOL                                footer_written;
 222     BOOL                                in_main_func;
 223
 224     /* For 3.0 vertex shaders */
 225     const char                          *vs_output[MAX_REG_OUTPUT];
 226     /* For 2.x and earlier vertex shaders */
 227     const char                          *texcrd_output[8], *color_output[2], *fog_output;
 228
 229     /* 3.0 pshader input for compatibility with fixed function */
 230     const char                          *ps_input[MAX_REG_INPUT];
 231 };
 232
 233 struct ps_signature
 234 {
 235     struct wined3d_shader_signature_element *sig;
 236     DWORD                               idx;
 237     struct wine_rb_entry                entry;
 238 };
 239
 240 struct arb_pshader_private {
 241     struct arb_ps_compiled_shader   *gl_shaders;
 242     UINT                            num_gl_shaders, shader_array_size;
 243     BOOL                            has_signature_idx;
 244     DWORD                           input_signature_idx;
 245     DWORD                           clipplane_emulation;
 246     BOOL                            clamp_consts;
 247 };
 248
 249 struct arb_vshader_private {
 250     struct arb_vs_compiled_shader   *gl_shaders;
 251     UINT                            num_gl_shaders, shader_array_size;
 252 };
 253
 254 struct shader_arb_priv
 255 {
 256     GLuint                  current_vprogram_id;
 257     GLuint                  current_fprogram_id;
 258     const struct arb_ps_compiled_shader *compiled_fprog;
 259     const struct arb_vs_compiled_shader *compiled_vprog;
 260     GLuint                  depth_blt_vprogram_id;
 261     GLuint                  depth_blt_fprogram_id[tex_type_count];
 262     BOOL                    use_arbfp_fixed_func;
 263     struct wine_rb_tree     fragment_shaders;
 264     BOOL                    last_ps_const_clamped;
 265     BOOL                    last_vs_color_unclamp;
 266
 267     struct wine_rb_tree     signature_tree;
 268     DWORD ps_sig_number;
 269 };
 270
 271 /********************************************************
 272  * ARB_[vertex/fragment]_program helper functions follow
 273  ********************************************************/
 274
 275 /**
 276  * Loads floating point constants into the currently set ARB_vertex/fragment_program.
 277  * When constant_list == NULL, it will load all the constants.
 278  *
 279  * @target_type should be either GL_VERTEX_PROGRAM_ARB (for vertex shaders)
 280  *  or GL_FRAGMENT_PROGRAM_ARB (for pixel shaders)
 281  */
 282 /* GL locking is done by the caller */
 283 static unsigned int shader_arb_load_constantsF(IWineD3DBaseShaderImpl *This, const struct wined3d_gl_info *gl_info,
 284         GLuint target_type, unsigned int max_constants, const float *constants, char *dirty_consts)
 285 {
 286     local_constant* lconst;
 287     DWORD i, j;
 288     unsigned int ret;
 289
 290     if (TRACE_ON(d3d_shader)) {
 291         for(i = 0; i < max_constants; i++) {
 292             if(!dirty_consts[i]) continue;
 293             TRACE_(d3d_constants)("Loading constants %i: %f, %f, %f, %f\n", i,
 294                         constants[i * 4 + 0], constants[i * 4 + 1],
 295                         constants[i * 4 + 2], constants[i * 4 + 3]);
 296         }
 297     }
 298     /* In 1.X pixel shaders constants are implicitly clamped in the range [-1;1] */
 299     if (target_type == GL_FRAGMENT_PROGRAM_ARB && This->baseShader.reg_maps.shader_version.major == 1)
 300     {
 301         float lcl_const[4];
 302         for(i = 0; i < 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     } else {
 326         if(GL_SUPPORT(EXT_GPU_PROGRAM_PARAMETERS)) {
 327             /* TODO: Benchmark if we're better of with finding the dirty constants ourselves,
 328              * or just reloading *all* constants at once
 329              *
 330             GL_EXTCALL(glProgramEnvParameters4fvEXT(target_type, 0, max_constants, constants));
 331              */
 332             for(i = 0; i < max_constants; i++) {
 333                 if(!dirty_consts[i]) continue;
 334
 335                 /* Find the next block of dirty constants */
 336                 dirty_consts[i] = 0;
 337                 j = i;
 338                 for(i++; (i < max_constants) && dirty_consts[i]; i++) {
 339                     dirty_consts[i] = 0;
 340                 }
 341
 342                 GL_EXTCALL(glProgramEnvParameters4fvEXT(target_type, j, i - j, constants + (j * 4)));
 343             }
 344         } else {
 345             for(i = 0; i < max_constants; i++) {
 346                 if(dirty_consts[i]) {
 347                     dirty_consts[i] = 0;
 348                     GL_EXTCALL(glProgramEnvParameter4fvARB(target_type, i, constants + (i * 4)));
 349                 }
 350             }
 351         }
 352     }
 353     checkGLcall("glProgramEnvParameter4fvARB()");
 354
 355     /* Load immediate constants */
 356     if(This->baseShader.load_local_constsF) {
 357         if (TRACE_ON(d3d_shader)) {
 358             LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry) {
 359                 GLfloat* values = (GLfloat*)lconst->value;
 360                 TRACE_(d3d_constants)("Loading local constants %i: %f, %f, %f, %f\n", lconst->idx,
 361                         values[0], values[1], values[2], values[3]);
 362             }
 363         }
 364         /* Immediate constants are clamped for 1.X shaders at loading times */
 365         ret = 0;
 366         LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry) {
 367             dirty_consts[lconst->idx] = 1; /* Dirtify so the non-immediate constant overwrites it next time */
 368             ret = max(ret, lconst->idx + 1);
 369             GL_EXTCALL(glProgramEnvParameter4fvARB(target_type, lconst->idx, (GLfloat*)lconst->value));
 370         }
 371         checkGLcall("glProgramEnvParameter4fvARB()");
 372         return ret; /* The loaded immediate constants need reloading for the next shader */
 373     } else {
 374         return 0; /* No constants are dirty now */
 375     }
 376 }
 377
 378 /**
 379  * Loads the texture dimensions for NP2 fixup into the currently set ARB_[vertex/fragment]_programs.
 380  */
 381 static void shader_arb_load_np2fixup_constants(
 382     IWineD3DDevice* device,
 383     char usePixelShader,
 384     char useVertexShader) {
 385
 386     IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl *) device;
 387     const struct shader_arb_priv* const priv = (const struct shader_arb_priv *) deviceImpl->shader_priv;
 388     IWineD3DStateBlockImpl* stateBlock = deviceImpl->stateBlock;
 389     const struct wined3d_gl_info *gl_info = &deviceImpl->adapter->gl_info;
 390
 391     if (!usePixelShader) {
 392         /* NP2 texcoord fixup is (currently) only done for pixelshaders. */
 393         return;
 394     }
 395
 396     if (priv->compiled_fprog && priv->compiled_fprog->np2fixup_info.super.active) {
 397         const struct arb_ps_np2fixup_info* const fixup = &priv->compiled_fprog->np2fixup_info;
 398         UINT i;
 399         WORD active = fixup->super.active;
 400         GLfloat np2fixup_constants[4 * MAX_FRAGMENT_SAMPLERS];
 401
 402         for (i = 0; active; active >>= 1, ++i) {
 403             const unsigned char idx = fixup->super.idx[i];
 404             const IWineD3DTextureImpl* const tex = (const IWineD3DTextureImpl*) stateBlock->textures[i];
 405             GLfloat* tex_dim = &np2fixup_constants[(idx >> 1) * 4];
 406
 407             if (!(active & 1)) continue;
 408
 409             if (!tex) {
 410                 FIXME("Nonexistent texture is flagged for NP2 texcoord fixup\n");
 411                 continue;
 412             }
 413
 414             if (idx % 2) {
 415                 tex_dim[2] = tex->baseTexture.pow2Matrix[0]; tex_dim[3] = tex->baseTexture.pow2Matrix[5];
 416             } else {
 417                 tex_dim[0] = tex->baseTexture.pow2Matrix[0]; tex_dim[1] = tex->baseTexture.pow2Matrix[5];
 418             }
 419         }
 420
 421         for (i = 0; i < fixup->super.num_consts; ++i) {
 422             GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB,
 423                                                    fixup->offset + i, &np2fixup_constants[i * 4]));
 424         }
 425     }
 426 }
 427
 428 /* GL locking is done by the caller. */
 429 static inline void shader_arb_ps_local_constants(IWineD3DDeviceImpl* deviceImpl)
 430 {
 431     IWineD3DStateBlockImpl* stateBlock = deviceImpl->stateBlock;
 432     const struct wined3d_gl_info *gl_info = &deviceImpl->adapter->gl_info;
 433     unsigned char i;
 434     struct shader_arb_priv *priv = deviceImpl->shader_priv;
 435     const struct arb_ps_compiled_shader *gl_shader = priv->compiled_fprog;
 436
 437     for(i = 0; i < gl_shader->numbumpenvmatconsts; i++)
 438     {
 439         int texunit = gl_shader->bumpenvmatconst[i].texunit;
 440
 441         /* The state manager takes care that this function is always called if the bump env matrix changes */
 442         const float *data = (const float *)&stateBlock->textureState[texunit][WINED3DTSS_BUMPENVMAT00];
 443         GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, gl_shader->bumpenvmatconst[i].const_num, data));
 444
 445         if (gl_shader->luminanceconst[i].const_num != WINED3D_CONST_NUM_UNUSED)
 446         {
 447             /* WINED3DTSS_BUMPENVLSCALE and WINED3DTSS_BUMPENVLOFFSET are next to each other.
 448              * point gl to the scale, and load 4 floats. x = scale, y = offset, z and w are junk, we
 449              * don't care about them. The pointers are valid for sure because the stateblock is bigger.
 450              * (they're WINED3DTSS_TEXTURETRANSFORMFLAGS and WINED3DTSS_ADDRESSW, so most likely 0 or NaN
 451             */
 452             const float *scale = (const float *)&stateBlock->textureState[texunit][WINED3DTSS_BUMPENVLSCALE];
 453             GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, gl_shader->luminanceconst[i].const_num, scale));
 454         }
 455     }
 456     checkGLcall("Load bumpmap consts");
 457
 458     if(gl_shader->ycorrection != WINED3D_CONST_NUM_UNUSED)
 459     {
 460         /* ycorrection.x: Backbuffer height(onscreen) or 0(offscreen).
 461         * ycorrection.y: -1.0(onscreen), 1.0(offscreen)
 462         * ycorrection.z: 1.0
 463         * ycorrection.w: 0.0
 464         */
 465         float val[4];
 466         val[0] = deviceImpl->render_offscreen ? 0.0f : ((IWineD3DSurfaceImpl *) deviceImpl->render_targets[0])->currentDesc.Height;
 467         val[1] = deviceImpl->render_offscreen ? 1.0f : -1.0f;
 468         val[2] = 1.0f;
 469         val[3] = 0.0f;
 470         GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, gl_shader->ycorrection, val));
 471         checkGLcall("y correction loading");
 472     }
 473
 474     if(gl_shader->num_int_consts == 0) return;
 475
 476     for(i = 0; i < MAX_CONST_I; i++)
 477     {
 478         if(gl_shader->int_consts[i] != WINED3D_CONST_NUM_UNUSED)
 479         {
 480             float val[4];
 481             val[0] = stateBlock->pixelShaderConstantI[4 * i];
 482             val[1] = stateBlock->pixelShaderConstantI[4 * i + 1];
 483             val[2] = stateBlock->pixelShaderConstantI[4 * i + 2];
 484             val[3] = -1.0f;
 485
 486             GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, gl_shader->int_consts[i], val));
 487         }
 488     }
 489     checkGLcall("Load ps int consts");
 490 }
 491
 492 /* GL locking is done by the caller. */
 493 static inline void shader_arb_vs_local_constants(IWineD3DDeviceImpl* deviceImpl)
 494 {
 495     IWineD3DStateBlockImpl* stateBlock;
 496     const struct wined3d_gl_info *gl_info = &deviceImpl->adapter->gl_info;
 497     unsigned char i;
 498     struct shader_arb_priv *priv = deviceImpl->shader_priv;
 499     const struct arb_vs_compiled_shader *gl_shader = priv->compiled_vprog;
 500
 501     /* Upload the position fixup */
 502     GL_EXTCALL(glProgramLocalParameter4fvARB(GL_VERTEX_PROGRAM_ARB, gl_shader->pos_fixup, deviceImpl->posFixup));
 503
 504     if(gl_shader->num_int_consts == 0) return;
 505
 506     stateBlock = deviceImpl->stateBlock;
 507
 508     for(i = 0; i < MAX_CONST_I; i++)
 509     {
 510         if(gl_shader->int_consts[i] != WINED3D_CONST_NUM_UNUSED)
 511         {
 512             float val[4];
 513             val[0] = stateBlock->vertexShaderConstantI[4 * i];
 514             val[1] = stateBlock->vertexShaderConstantI[4 * i + 1];
 515             val[2] = stateBlock->vertexShaderConstantI[4 * i + 2];
 516             val[3] = -1.0f;
 517
 518             GL_EXTCALL(glProgramLocalParameter4fvARB(GL_VERTEX_PROGRAM_ARB, gl_shader->int_consts[i], val));
 519         }
 520     }
 521     checkGLcall("Load vs int consts");
 522 }
 523
 524 /**
 525  * Loads the app-supplied constants into the currently set ARB_[vertex/fragment]_programs.
 526  *
 527  * We only support float constants in ARB at the moment, so don't
 528  * worry about the Integers or Booleans
 529  */
 530 /* GL locking is done by the caller (state handler) */
 531 static void shader_arb_load_constants(
 532     IWineD3DDevice* device,
 533     char usePixelShader,
 534     char useVertexShader) {
 535
 536     IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) device;
 537     IWineD3DStateBlockImpl* stateBlock = deviceImpl->stateBlock;
 538     const struct wined3d_context *context = context_get_current();
 539     const struct wined3d_gl_info *gl_info = context->gl_info;
 540
 541     if (useVertexShader) {
 542         IWineD3DBaseShaderImpl* vshader = (IWineD3DBaseShaderImpl*) stateBlock->vertexShader;
 543
 544         /* Load DirectX 9 float constants for vertex shader */
 545         deviceImpl->highest_dirty_vs_const = shader_arb_load_constantsF(vshader, gl_info, GL_VERTEX_PROGRAM_ARB,
 546                 deviceImpl->highest_dirty_vs_const, stateBlock->vertexShaderConstantF, context->vshader_const_dirty);
 547
 548         shader_arb_vs_local_constants(deviceImpl);
 549     }
 550
 551     if (usePixelShader) {
 552         IWineD3DBaseShaderImpl* pshader = (IWineD3DBaseShaderImpl*) stateBlock->pixelShader;
 553
 554         /* Load DirectX 9 float constants for pixel shader */
 555         deviceImpl->highest_dirty_ps_const = shader_arb_load_constantsF(pshader, gl_info, GL_FRAGMENT_PROGRAM_ARB,
 556                 deviceImpl->highest_dirty_ps_const, stateBlock->pixelShaderConstantF, context->pshader_const_dirty);
 557         shader_arb_ps_local_constants(deviceImpl);
 558     }
 559 }
 560
 561 static void shader_arb_update_float_vertex_constants(IWineD3DDevice *iface, UINT start, UINT count)
 562 {
 563     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
 564     struct wined3d_context *context = context_get_current();
 565
 566     /* We don't want shader constant dirtification to be an O(contexts), so just dirtify the active
 567      * context. On a context switch the old context will be fully dirtified */
 568     if (!context || ((IWineD3DSurfaceImpl *)context->surface)->resource.wineD3DDevice != This) return;
 569
 570     memset(context->vshader_const_dirty + start, 1, sizeof(*context->vshader_const_dirty) * count);
 571     This->highest_dirty_vs_const = max(This->highest_dirty_vs_const, start + count);
 572 }
 573
 574 static void shader_arb_update_float_pixel_constants(IWineD3DDevice *iface, UINT start, UINT count)
 575 {
 576     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
 577     struct wined3d_context *context = context_get_current();
 578
 579     /* We don't want shader constant dirtification to be an O(contexts), so just dirtify the active
 580      * context. On a context switch the old context will be fully dirtified */
 581     if (!context || ((IWineD3DSurfaceImpl *)context->surface)->resource.wineD3DDevice != This) return;
 582
 583     memset(context->pshader_const_dirty + start, 1, sizeof(*context->pshader_const_dirty) * count);
 584     This->highest_dirty_ps_const = max(This->highest_dirty_ps_const, start + count);
 585 }
 586
 587 static DWORD *local_const_mapping(IWineD3DBaseShaderImpl *This)
 588 {
 589     DWORD *ret;
 590     DWORD idx = 0;
 591     const local_constant *lconst;
 592
 593     if(This->baseShader.load_local_constsF || list_empty(&This->baseShader.constantsF)) return NULL;
 594
 595     ret = HeapAlloc(GetProcessHeap(), 0, sizeof(DWORD) * This->baseShader.limits.constant_float);
 596     if(!ret) {
 597         ERR("Out of memory\n");
 598         return NULL;
 599     }
 600
 601     LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry) {
 602         ret[lconst->idx] = idx++;
 603     }
 604     return ret;
 605 }
 606
 607 /* Generate the variable & register declarations for the ARB_vertex_program output target */
 608 static DWORD shader_generate_arb_declarations(IWineD3DBaseShader *iface, const shader_reg_maps *reg_maps,
 609         struct wined3d_shader_buffer *buffer, const struct wined3d_gl_info *gl_info, DWORD *lconst_map,
 610         DWORD *num_clipplanes, struct shader_arb_ctx_priv *ctx)
 611 {
 612     IWineD3DBaseShaderImpl* This = (IWineD3DBaseShaderImpl*) iface;
 613     DWORD i, next_local = 0;
 614     char pshader = shader_is_pshader_version(reg_maps->shader_version.type);
 615     unsigned max_constantsF;
 616     const local_constant *lconst;
 617
 618     /* In pixel shaders, all private constants are program local, we don't need anything
 619      * from program.env. Thus we can advertise the full set of constants in pixel shaders.
 620      * If we need a private constant the GL implementation will squeeze it in somewhere
 621      *
 622      * With vertex shaders we need the posFixup and on some GL implementations 4 helper
 623      * immediate values. The posFixup is loaded using program.env for now, so always
 624      * subtract one from the number of constants. If the shader uses indirect addressing,
 625      * account for the helper const too because we have to declare all availabke d3d constants
 626      * and don't know which are actually used.
 627      */
 628     if(pshader) {
 629         max_constantsF = GL_LIMITS(pshader_constantsF);
 630     } else {
 631         if(This->baseShader.reg_maps.usesrelconstF) {
 632             DWORD highest_constf = 0, clip_limit;
 633             max_constantsF = GL_LIMITS(vshader_constantsF) - reserved_vs_const(iface, gl_info);
 634             max_constantsF -= count_bits(This->baseShader.reg_maps.integer_constants);
 635
 636             for(i = 0; i < This->baseShader.limits.constant_float; i++)
 637             {
 638                 DWORD idx = i >> 5;
 639                 DWORD shift = i & 0x1f;
 640                 if(reg_maps->constf[idx] & (1 << shift)) highest_constf = i;
 641             }
 642
 643             clip_limit = GL_LIMITS(clipplanes);
 644             if(ctx->target_version == ARB) clip_limit = min(clip_limit, 4);
 645             *num_clipplanes = min(clip_limit, max_constantsF - highest_constf - 1);
 646             max_constantsF -= *num_clipplanes;
 647             if(*num_clipplanes < clip_limit)
 648             {
 649                 WARN("Only %u clipplanes out of %u enabled\n", *num_clipplanes, GL_LIMITS(clipplanes));
 650             }
 651         }
 652         else
 653         {
 654             if(ctx->target_version >= NV2) *num_clipplanes = GL_LIMITS(clipplanes);
 655             else *num_clipplanes = min(GL_LIMITS(clipplanes), 4);
 656             max_constantsF = GL_LIMITS(vshader_constantsF);
 657         }
 658     }
 659
 660     for(i = 0; i < This->baseShader.limits.temporary; i++) {
 661         if (reg_maps->temporary[i])
 662             shader_addline(buffer, "TEMP R%u;\n", i);
 663     }
 664
 665     for (i = 0; i < This->baseShader.limits.address; i++) {
 666         if (reg_maps->address[i])
 667             shader_addline(buffer, "ADDRESS A%d;\n", i);
 668     }
 669
 670     if(pshader && reg_maps->shader_version.major == 1 && reg_maps->shader_version.minor <= 3) {
 671         for(i = 0; i < This->baseShader.limits.texcoord; i++) {
 672             if (reg_maps->texcoord[i] && pshader)
 673                 shader_addline(buffer,"TEMP T%u;\n", i);
 674         }
 675     }
 676
 677     /* Load local constants using the program-local space,
 678      * this avoids reloading them each time the shader is used
 679      */
 680     if(lconst_map) {
 681         LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry) {
 682             shader_addline(buffer, "PARAM C%u = program.local[%u];\n", lconst->idx,
 683                            lconst_map[lconst->idx]);
 684             next_local = max(next_local, lconst_map[lconst->idx] + 1);
 685         }
 686     }
 687
 688     /* we use the array-based constants array if the local constants are marked for loading,
 689      * because then we use indirect addressing, or when the local constant list is empty,
 690      * because then we don't know if we're using indirect addressing or not. If we're hardcoding
 691      * local constants do not declare the loaded constants as an array because ARB compilers usually
 692      * do not optimize unused constants away
 693      */
 694     if(This->baseShader.reg_maps.usesrelconstF) {
 695         /* Need to PARAM the environment parameters (constants) so we can use relative addressing */
 696         shader_addline(buffer, "PARAM C[%d] = { program.env[0..%d] };\n",
 697                     max_constantsF, max_constantsF - 1);
 698     } else {
 699         for(i = 0; i < max_constantsF; i++) {
 700             DWORD idx, mask;
 701             idx = i >> 5;
 702             mask = 1 << (i & 0x1f);
 703             if(!shader_constant_is_local(This, i) && (This->baseShader.reg_maps.constf[idx] & mask)) {
 704                 shader_addline(buffer, "PARAM C%d = program.env[%d];\n",i, i);
 705             }
 706         }
 707     }
 708
 709     return next_local;
 710 }
 711
 712 static const char * const shift_tab[] = {
 713     "dummy",     /*  0 (none) */
 714     "coefmul.x", /*  1 (x2)   */
 715     "coefmul.y", /*  2 (x4)   */
 716     "coefmul.z", /*  3 (x8)   */
 717     "coefmul.w", /*  4 (x16)  */
 718     "dummy",     /*  5 (x32)  */
 719     "dummy",     /*  6 (x64)  */
 720     "dummy",     /*  7 (x128) */
 721     "dummy",     /*  8 (d256) */
 722     "dummy",     /*  9 (d128) */
 723     "dummy",     /* 10 (d64)  */
 724     "dummy",     /* 11 (d32)  */
 725     "coefdiv.w", /* 12 (d16)  */
 726     "coefdiv.z", /* 13 (d8)   */
 727     "coefdiv.y", /* 14 (d4)   */
 728     "coefdiv.x"  /* 15 (d2)   */
 729 };
 730
 731 static void shader_arb_get_write_mask(const struct wined3d_shader_instruction *ins,
 732         const struct wined3d_shader_dst_param *dst, char *write_mask)
 733 {
 734     char *ptr = write_mask;
 735
 736     if (dst->write_mask != WINED3DSP_WRITEMASK_ALL)
 737     {
 738         *ptr++ = '.';
 739         if (dst->write_mask & WINED3DSP_WRITEMASK_0) *ptr++ = 'x';
 740         if (dst->write_mask & WINED3DSP_WRITEMASK_1) *ptr++ = 'y';
 741         if (dst->write_mask & WINED3DSP_WRITEMASK_2) *ptr++ = 'z';
 742         if (dst->write_mask & WINED3DSP_WRITEMASK_3) *ptr++ = 'w';
 743     }
 744
 745     *ptr = '\0';
 746 }
 747
 748 static void shader_arb_get_swizzle(const struct wined3d_shader_src_param *param, BOOL fixup, char *swizzle_str)
 749 {
 750     /* For registers of type WINED3DDECLTYPE_D3DCOLOR, data is stored as "bgra",
 751      * but addressed as "rgba". To fix this we need to swap the register's x
 752      * and z components. */
 753     const char *swizzle_chars = fixup ? "zyxw" : "xyzw";
 754     char *ptr = swizzle_str;
 755
 756     /* swizzle bits fields: wwzzyyxx */
 757     DWORD swizzle = param->swizzle;
 758     DWORD swizzle_x = swizzle & 0x03;
 759     DWORD swizzle_y = (swizzle >> 2) & 0x03;
 760     DWORD swizzle_z = (swizzle >> 4) & 0x03;
 761     DWORD swizzle_w = (swizzle >> 6) & 0x03;
 762
 763     /* If the swizzle is the default swizzle (ie, "xyzw"), we don't need to
 764      * generate a swizzle string. Unless we need to our own swizzling. */
 765     if (swizzle != WINED3DSP_NOSWIZZLE || fixup)
 766     {
 767         *ptr++ = '.';
 768         if (swizzle_x == swizzle_y && swizzle_x == swizzle_z && swizzle_x == swizzle_w) {
 769             *ptr++ = swizzle_chars[swizzle_x];
 770         } else {
 771             *ptr++ = swizzle_chars[swizzle_x];
 772             *ptr++ = swizzle_chars[swizzle_y];
 773             *ptr++ = swizzle_chars[swizzle_z];
 774             *ptr++ = swizzle_chars[swizzle_w];
 775         }
 776     }
 777
 778     *ptr = '\0';
 779 }
 780
 781 static void shader_arb_request_a0(const struct wined3d_shader_instruction *ins, const char *src)
 782 {
 783     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
 784     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
 785
 786     if(strcmp(priv->addr_reg, src) == 0) return;
 787
 788     strcpy(priv->addr_reg, src);
 789     shader_addline(buffer, "ARL A0.x, %s;\n", src);
 790 }
 791
 792 static void shader_arb_get_src_param(const struct wined3d_shader_instruction *ins,
 793         const struct wined3d_shader_src_param *src, unsigned int tmpreg, char *outregstr);
 794
 795 static void shader_arb_get_register_name(const struct wined3d_shader_instruction *ins,
 796         const struct wined3d_shader_register *reg, char *register_name, BOOL *is_color)
 797 {
 798     /* oPos, oFog and oPts in D3D */
 799     static const char * const rastout_reg_names[] = {"TMP_OUT", "result.fogcoord", "result.pointsize"};
 800     IWineD3DBaseShaderImpl *This = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
 801     BOOL pshader = shader_is_pshader_version(This->baseShader.reg_maps.shader_version.type);
 802     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
 803
 804     *is_color = FALSE;
 805
 806     switch (reg->type)
 807     {
 808         case WINED3DSPR_TEMP:
 809             sprintf(register_name, "R%u", reg->idx);
 810             break;
 811
 812         case WINED3DSPR_INPUT:
 813             if (pshader)
 814             {
 815                 if(This->baseShader.reg_maps.shader_version.major < 3)
 816                 {
 817                     if (reg->idx == 0) strcpy(register_name, "fragment.color.primary");
 818                     else strcpy(register_name, "fragment.color.secondary");
 819                 }
 820                 else
 821                 {
 822                     if(reg->rel_addr)
 823                     {
 824                         char rel_reg[50];
 825                         shader_arb_get_src_param(ins, reg->rel_addr, 0, rel_reg);
 826
 827                         if(strcmp(rel_reg, "**aL_emul**") == 0)
 828                         {
 829                             DWORD idx = ctx->aL + reg->idx;
 830                             if(idx < MAX_REG_INPUT)
 831                             {
 832                                 strcpy(register_name, ctx->ps_input[idx]);
 833                             }
 834                             else
 835                             {
 836                                 ERR("Pixel shader input register out of bounds: %u\n", idx);
 837                                 sprintf(register_name, "out_of_bounds_%u", idx);
 838                             }
 839                         }
 840                         else if(This->baseShader.reg_maps.input_registers & 0x0300)
 841                         {
 842                             /* There are two ways basically:
 843                              *
 844                              * 1) Use the unrolling code that is used for loop emulation and unroll the loop.
 845                              *    That means trouble if the loop also contains a breakc or if the control values
 846                              *    aren't local constants.
 847                              * 2) Generate an if block that checks if aL.y < 8, == 8 or == 9 and selects the
 848                              *    source dynamically. The trouble is that we cannot simply read aL.y because it
 849                              *    is an ADDRESS register. We could however push it, load .zw with a value and use
 850                              *    ADAC to load the condition code register and pop it again afterwards
 851                              */
 852                             FIXME("Relative input register addressing with more than 8 registers\n");
 853
 854                             /* This is better than nothing for now */
 855                             sprintf(register_name, "fragment.texcoord[%s + %u]", rel_reg, reg->idx);
 856                         }
 857                         else if(ctx->cur_ps_args->super.vp_mode != vertexshader)
 858                         {
 859                             /* This is problematic because we'd have to consult the ctx->ps_input strings
 860                              * for where to find the varying. Some may be "0.0", others can be texcoords or
 861                              * colors. This needs either a pipeline replacement to make the vertex shader feed
 862                              * proper varyings, or loop unrolling
 863                              *
 864                              * For now use the texcoords and hope for the best
 865                              */
 866                             FIXME("Non-vertex shader varying input with indirect addressing\n");
 867                             sprintf(register_name, "fragment.texcoord[%s + %u]", rel_reg, reg->idx);
 868                         }
 869                         else
 870                         {
 871                             /* D3D supports indirect addressing only with aL in loop registers. The loop instruction
 872                              * pulls GL_NV_fragment_program2 in
 873                              */
 874                             sprintf(register_name, "fragment.texcoord[%s + %u]", rel_reg, reg->idx);
 875                         }
 876                     }
 877                     else
 878                     {
 879                         if(reg->idx < MAX_REG_INPUT)
 880                         {
 881                             strcpy(register_name, ctx->ps_input[reg->idx]);
 882                         }
 883                         else
 884                         {
 885                             ERR("Pixel shader input register out of bounds: %u\n", reg->idx);
 886                             sprintf(register_name, "out_of_bounds_%u", reg->idx);
 887                         }
 888                     }
 889                 }
 890             }
 891             else
 892             {
 893                 if (ctx->cur_vs_args->super.swizzle_map & (1 << reg->idx)) *is_color = TRUE;
 894                 sprintf(register_name, "vertex.attrib[%u]", reg->idx);
 895             }
 896             break;
 897
 898         case WINED3DSPR_CONST:
 899             if (!pshader && reg->rel_addr)
 900             {
 901                 BOOL aL = FALSE;
 902                 char rel_reg[50];
 903                 UINT rel_offset = ((IWineD3DVertexShaderImpl *)This)->rel_offset;
 904                 if(This->baseShader.reg_maps.shader_version.major < 2) {
 905                     sprintf(rel_reg, "A0.x");
 906                 } else {
 907                     shader_arb_get_src_param(ins, reg->rel_addr, 0, rel_reg);
 908                     if(ctx->target_version == ARB) {
 909                         if(strcmp(rel_reg, "**aL_emul**") == 0) {
 910                             aL = TRUE;
 911                         } else {
 912                             shader_arb_request_a0(ins, rel_reg);
 913                             sprintf(rel_reg, "A0.x");
 914                         }
 915                     }
 916                 }
 917                 if(aL)
 918                     sprintf(register_name, "C[%u]", ctx->aL + reg->idx);
 919                 else if (reg->idx >= rel_offset)
 920                     sprintf(register_name, "C[%s + %u]", rel_reg, reg->idx - rel_offset);
 921                 else
 922                     sprintf(register_name, "C[%s - %u]", rel_reg, -reg->idx + rel_offset);
 923             }
 924             else
 925             {
 926                 if (This->baseShader.reg_maps.usesrelconstF)
 927                     sprintf(register_name, "C[%u]", reg->idx);
 928                 else
 929                     sprintf(register_name, "C%u", reg->idx);
 930             }
 931             break;
 932
 933         case WINED3DSPR_TEXTURE: /* case WINED3DSPR_ADDR: */
 934             if (pshader) {
 935                 if(This->baseShader.reg_maps.shader_version.major == 1 &&
 936                    This->baseShader.reg_maps.shader_version.minor <= 3) {
 937                     /* In ps <= 1.3, Tx is a temporary register as destination to all instructions,
 938                      * and as source to most instructions. For some instructions it is the texcoord
 939                      * input. Those instructions know about the special use
 940                      */
 941                     sprintf(register_name, "T%u", reg->idx);
 942                 } else {
 943                     /* in ps 1.4 and 2.x Tx is always a (read-only) varying */
 944                     sprintf(register_name, "fragment.texcoord[%u]", reg->idx);
 945                 }
 946             }
 947             else
 948             {
 949                 if(This->baseShader.reg_maps.shader_version.major == 1 || ctx->target_version >= NV2)
 950                 {
 951                     sprintf(register_name, "A%u", reg->idx);
 952                 }
 953                 else
 954                 {
 955                     sprintf(register_name, "A%u_SHADOW", reg->idx);
 956                 }
 957             }
 958             break;
 959
 960         case WINED3DSPR_COLOROUT:
 961             if(ctx->cur_ps_args->super.srgb_correction && reg->idx == 0)
 962             {
 963                 strcpy(register_name, "TMP_COLOR");
 964             }
 965             else
 966             {
 967                 if(ctx->cur_ps_args->super.srgb_correction) FIXME("sRGB correction on higher render targets\n");
 968                 if(This->baseShader.reg_maps.highest_render_target > 0)
 969                 {
 970                     sprintf(register_name, "result.color[%u]", reg->idx);
 971                 }
 972                 else
 973                 {
 974                     strcpy(register_name, "result.color");
 975                 }
 976             }
 977             break;
 978
 979         case WINED3DSPR_RASTOUT:
 980             if(reg->idx == 1) sprintf(register_name, "%s", ctx->fog_output);
 981             else sprintf(register_name, "%s", rastout_reg_names[reg->idx]);
 982             break;
 983
 984         case WINED3DSPR_DEPTHOUT:
 985             strcpy(register_name, "result.depth");
 986             break;
 987
 988         case WINED3DSPR_ATTROUT:
 989         /* case WINED3DSPR_OUTPUT: */
 990             if (pshader) sprintf(register_name, "oD[%u]", reg->idx);
 991             else strcpy(register_name, ctx->color_output[reg->idx]);
 992             break;
 993
 994         case WINED3DSPR_TEXCRDOUT:
 995             if (pshader)
 996             {
 997                 sprintf(register_name, "oT[%u]", reg->idx);
 998             }
 999             else
1000             {
1001                 if(This->baseShader.reg_maps.shader_version.major < 3)
1002                 {
1003                     strcpy(register_name, ctx->texcrd_output[reg->idx]);
1004                 }
1005                 else
1006                 {
1007                     strcpy(register_name, ctx->vs_output[reg->idx]);
1008                 }
1009             }
1010             break;
1011
1012         case WINED3DSPR_LOOP:
1013             if(ctx->target_version >= NV2)
1014             {
1015                 /* Pshader has an implicitly declared loop index counter A0.x that cannot be renamed */
1016                 if(pshader) sprintf(register_name, "A0.x");
1017                 else sprintf(register_name, "aL.y");
1018             }
1019             else
1020             {
1021                 /* Unfortunately this code cannot return the value of ctx->aL here. An immediate value
1022                  * would be valid, but if aL is used for indexing(its only use), there's likely an offset,
1023                  * thus the result would be something like C[15 + 30], which is not valid in the ARB program
1024                  * grammar. So return a marker for the emulated aL and intercept it in constant and varying
1025                  * indexing
1026                  */
1027                 sprintf(register_name, "**aL_emul**");
1028             }
1029
1030             break;
1031
1032         case WINED3DSPR_CONSTINT:
1033             sprintf(register_name, "I%u", reg->idx);
1034             break;
1035
1036         case WINED3DSPR_MISCTYPE:
1037             if(reg->idx == 0)
1038             {
1039                 sprintf(register_name, "vpos");
1040             }
1041             else if(reg->idx == 1)
1042             {
1043                 sprintf(register_name, "fragment.facing.x");
1044             }
1045             else
1046             {
1047                 FIXME("Unknown MISCTYPE register index %u\n", reg->idx);
1048             }
1049             break;
1050
1051         default:
1052             FIXME("Unhandled register type %#x[%u]\n", reg->type, reg->idx);
1053             sprintf(register_name, "unrecognized_register[%u]", reg->idx);
1054             break;
1055     }
1056 }
1057
1058 static void shader_arb_get_dst_param(const struct wined3d_shader_instruction *ins,
1059         const struct wined3d_shader_dst_param *wined3d_dst, char *str)
1060 {
1061     char register_name[255];
1062     char write_mask[6];
1063     BOOL is_color;
1064
1065     shader_arb_get_register_name(ins, &wined3d_dst->reg, register_name, &is_color);
1066     strcpy(str, register_name);
1067
1068     shader_arb_get_write_mask(ins, wined3d_dst, write_mask);
1069     strcat(str, write_mask);
1070 }
1071
1072 static const char *shader_arb_get_fixup_swizzle(enum fixup_channel_source channel_source)
1073 {
1074     switch(channel_source)
1075     {
1076         case CHANNEL_SOURCE_ZERO: return "0";
1077         case CHANNEL_SOURCE_ONE: return "1";
1078         case CHANNEL_SOURCE_X: return "x";
1079         case CHANNEL_SOURCE_Y: return "y";
1080         case CHANNEL_SOURCE_Z: return "z";
1081         case CHANNEL_SOURCE_W: return "w";
1082         default:
1083             FIXME("Unhandled channel source %#x\n", channel_source);
1084             return "undefined";
1085     }
1086 }
1087
1088 static void gen_color_correction(struct wined3d_shader_buffer *buffer, const char *reg,
1089         DWORD dst_mask, const char *one, const char *two, struct color_fixup_desc fixup)
1090 {
1091     DWORD mask;
1092
1093     if (is_yuv_fixup(fixup))
1094     {
1095         enum yuv_fixup yuv_fixup = get_yuv_fixup(fixup);
1096         FIXME("YUV fixup (%#x) not supported\n", yuv_fixup);
1097         return;
1098     }
1099
1100     mask = 0;
1101     if (fixup.x_source != CHANNEL_SOURCE_X) mask |= WINED3DSP_WRITEMASK_0;
1102     if (fixup.y_source != CHANNEL_SOURCE_Y) mask |= WINED3DSP_WRITEMASK_1;
1103     if (fixup.z_source != CHANNEL_SOURCE_Z) mask |= WINED3DSP_WRITEMASK_2;
1104     if (fixup.w_source != CHANNEL_SOURCE_W) mask |= WINED3DSP_WRITEMASK_3;
1105     mask &= dst_mask;
1106
1107     if (mask)
1108     {
1109         shader_addline(buffer, "SWZ %s, %s, %s, %s, %s, %s;\n", reg, reg,
1110                 shader_arb_get_fixup_swizzle(fixup.x_source), shader_arb_get_fixup_swizzle(fixup.y_source),
1111                 shader_arb_get_fixup_swizzle(fixup.z_source), shader_arb_get_fixup_swizzle(fixup.w_source));
1112     }
1113
1114     mask = 0;
1115     if (fixup.x_sign_fixup) mask |= WINED3DSP_WRITEMASK_0;
1116     if (fixup.y_sign_fixup) mask |= WINED3DSP_WRITEMASK_1;
1117     if (fixup.z_sign_fixup) mask |= WINED3DSP_WRITEMASK_2;
1118     if (fixup.w_sign_fixup) mask |= WINED3DSP_WRITEMASK_3;
1119     mask &= dst_mask;
1120
1121     if (mask)
1122     {
1123         char reg_mask[6];
1124         char *ptr = reg_mask;
1125
1126         if (mask != WINED3DSP_WRITEMASK_ALL)
1127         {
1128             *ptr++ = '.';
1129             if (mask & WINED3DSP_WRITEMASK_0) *ptr++ = 'x';
1130             if (mask & WINED3DSP_WRITEMASK_1) *ptr++ = 'y';
1131             if (mask & WINED3DSP_WRITEMASK_2) *ptr++ = 'z';
1132             if (mask & WINED3DSP_WRITEMASK_3) *ptr++ = 'w';
1133         }
1134         *ptr = '\0';
1135
1136         shader_addline(buffer, "MAD %s%s, %s, %s, -%s;\n", reg, reg_mask, reg, two, one);
1137     }
1138 }
1139
1140 static const char *shader_arb_get_modifier(const struct wined3d_shader_instruction *ins)
1141 {
1142     DWORD mod;
1143     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
1144     if (!ins->dst_count) return "";
1145
1146     mod = ins->dst[0].modifiers;
1147
1148     /* Silently ignore PARTIALPRECISION if its not supported */
1149     if(priv->target_version == ARB) mod &= ~WINED3DSPDM_PARTIALPRECISION;
1150
1151     if(mod & WINED3DSPDM_MSAMPCENTROID)
1152     {
1153         FIXME("Unhandled modifier WINED3DSPDM_MSAMPCENTROID\n");
1154         mod &= ~WINED3DSPDM_MSAMPCENTROID;
1155     }
1156
1157     switch(mod)
1158     {
1159         case WINED3DSPDM_SATURATE | WINED3DSPDM_PARTIALPRECISION:
1160             return "H_SAT";
1161
1162         case WINED3DSPDM_SATURATE:
1163             return "_SAT";
1164
1165         case WINED3DSPDM_PARTIALPRECISION:
1166             return "H";
1167
1168         case 0:
1169             return "";
1170
1171         default:
1172             FIXME("Unknown modifiers 0x%08x\n", mod);
1173             return "";
1174     }
1175 }
1176
1177 #define TEX_PROJ        0x1
1178 #define TEX_BIAS        0x2
1179 #define TEX_LOD         0x4
1180 #define TEX_DERIV       0x10
1181
1182 static void shader_hw_sample(const struct wined3d_shader_instruction *ins, DWORD sampler_idx,
1183         const char *dst_str, const char *coord_reg, WORD flags, const char *dsx, const char *dsy)
1184 {
1185     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1186     DWORD sampler_type = ins->ctx->reg_maps->sampler_type[sampler_idx];
1187     const char *tex_type;
1188     BOOL np2_fixup = FALSE;
1189     IWineD3DBaseShaderImpl *This = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
1190     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *) This->baseShader.device;
1191     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
1192     const char *mod;
1193     BOOL pshader = shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type);
1194
1195     /* D3D vertex shader sampler IDs are vertex samplers(0-3), not global d3d samplers */
1196     if(!pshader) sampler_idx += MAX_FRAGMENT_SAMPLERS;
1197
1198     switch(sampler_type) {
1199         case WINED3DSTT_1D:
1200             tex_type = "1D";
1201             break;
1202
1203         case WINED3DSTT_2D:
1204             if(device->stateBlock->textures[sampler_idx] &&
1205                IWineD3DBaseTexture_GetTextureDimensions(device->stateBlock->textures[sampler_idx]) == GL_TEXTURE_RECTANGLE_ARB) {
1206                 tex_type = "RECT";
1207             } else {
1208                 tex_type = "2D";
1209             }
1210             if (shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type))
1211             {
1212                 if (priv->cur_np2fixup_info->super.active & (1 << sampler_idx))
1213                 {
1214                     if (flags) FIXME("Only ordinary sampling from NP2 textures is supported.\n");
1215                     else np2_fixup = TRUE;
1216                 }
1217             }
1218             break;
1219
1220         case WINED3DSTT_VOLUME:
1221             tex_type = "3D";
1222             break;
1223
1224         case WINED3DSTT_CUBE:
1225             tex_type = "CUBE";
1226             break;
1227
1228         default:
1229             ERR("Unexpected texture type %d\n", sampler_type);
1230             tex_type = "";
1231     }
1232
1233     /* TEX, TXL, TXD and TXP do not support the "H" modifier,
1234      * so don't use shader_arb_get_modifier
1235      */
1236     if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE) mod = "_SAT";
1237     else mod = "";
1238
1239     /* Fragment samplers always have indentity mapping */
1240     if(sampler_idx >= MAX_FRAGMENT_SAMPLERS)
1241     {
1242         sampler_idx = priv->cur_vs_args->vertex_samplers[sampler_idx - MAX_FRAGMENT_SAMPLERS];
1243     }
1244
1245     if (flags & TEX_DERIV)
1246     {
1247         if(flags & TEX_PROJ) FIXME("Projected texture sampling with custom derivates\n");
1248         if(flags & TEX_BIAS) FIXME("Biased texture sampling with custom derivates\n");
1249         shader_addline(buffer, "TXD%s %s, %s, %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg,
1250                        dsx, dsy,sampler_idx, tex_type);
1251     }
1252     else if(flags & TEX_LOD)
1253     {
1254         if(flags & TEX_PROJ) FIXME("Projected texture sampling with explicit lod\n");
1255         if(flags & TEX_BIAS) FIXME("Biased texture sampling with explicit lod\n");
1256         shader_addline(buffer, "TXL%s %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg,
1257                        sampler_idx, tex_type);
1258     }
1259     else if (flags & TEX_BIAS)
1260     {
1261         /* Shouldn't be possible, but let's check for it */
1262         if(flags & TEX_PROJ) FIXME("Biased and Projected texture sampling\n");
1263         /* TXB takes the 4th component of the source vector automatically, as d3d. Nothing more to do */
1264         shader_addline(buffer, "TXB%s %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg, sampler_idx, tex_type);
1265     }
1266     else if (flags & TEX_PROJ)
1267     {
1268         shader_addline(buffer, "TXP%s %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg, sampler_idx, tex_type);
1269     }
1270     else
1271     {
1272         if (np2_fixup)
1273         {
1274             const unsigned char idx = priv->cur_np2fixup_info->super.idx[sampler_idx];
1275             shader_addline(buffer, "MUL TA, np2fixup[%u].%s, %s;\n", idx >> 1,
1276                            (idx % 2) ? "zwxy" : "xyzw", coord_reg);
1277
1278             shader_addline(buffer, "TEX%s %s, TA, texture[%u], %s;\n", mod, dst_str, sampler_idx, tex_type);
1279         }
1280         else
1281             shader_addline(buffer, "TEX%s %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg, sampler_idx, tex_type);
1282     }
1283
1284     if (pshader)
1285     {
1286         gen_color_correction(buffer, dst_str, ins->dst[0].write_mask,
1287                 "one", "coefmul.x", priv->cur_ps_args->super.color_fixup[sampler_idx]);
1288     }
1289 }
1290
1291 static void shader_arb_get_src_param(const struct wined3d_shader_instruction *ins,
1292         const struct wined3d_shader_src_param *src, unsigned int tmpreg, char *outregstr)
1293 {
1294     /* Generate a line that does the input modifier computation and return the input register to use */
1295     BOOL is_color = FALSE;
1296     char regstr[256];
1297     char swzstr[20];
1298     int insert_line;
1299     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1300     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
1301
1302     /* Assume a new line will be added */
1303     insert_line = 1;
1304
1305     /* Get register name */
1306     shader_arb_get_register_name(ins, &src->reg, regstr, &is_color);
1307     shader_arb_get_swizzle(src, is_color, swzstr);
1308
1309     switch (src->modifiers)
1310     {
1311     case WINED3DSPSM_NONE:
1312         sprintf(outregstr, "%s%s", regstr, swzstr);
1313         insert_line = 0;
1314         break;
1315     case WINED3DSPSM_NEG:
1316         sprintf(outregstr, "-%s%s", regstr, swzstr);
1317         insert_line = 0;
1318         break;
1319     case WINED3DSPSM_BIAS:
1320         shader_addline(buffer, "ADD T%c, %s, -coefdiv.x;\n", 'A' + tmpreg, regstr);
1321         break;
1322     case WINED3DSPSM_BIASNEG:
1323         shader_addline(buffer, "ADD T%c, -%s, coefdiv.x;\n", 'A' + tmpreg, regstr);
1324         break;
1325     case WINED3DSPSM_SIGN:
1326         shader_addline(buffer, "MAD T%c, %s, coefmul.x, -one.x;\n", 'A' + tmpreg, regstr);
1327         break;
1328     case WINED3DSPSM_SIGNNEG:
1329         shader_addline(buffer, "MAD T%c, %s, -coefmul.x, one.x;\n", 'A' + tmpreg, regstr);
1330         break;
1331     case WINED3DSPSM_COMP:
1332         shader_addline(buffer, "SUB T%c, one.x, %s;\n", 'A' + tmpreg, regstr);
1333         break;
1334     case WINED3DSPSM_X2:
1335         shader_addline(buffer, "ADD T%c, %s, %s;\n", 'A' + tmpreg, regstr, regstr);
1336         break;
1337     case WINED3DSPSM_X2NEG:
1338         shader_addline(buffer, "ADD T%c, -%s, -%s;\n", 'A' + tmpreg, regstr, regstr);
1339         break;
1340     case WINED3DSPSM_DZ:
1341         shader_addline(buffer, "RCP T%c, %s.z;\n", 'A' + tmpreg, regstr);
1342         shader_addline(buffer, "MUL T%c, %s, T%c;\n", 'A' + tmpreg, regstr, 'A' + tmpreg);
1343         break;
1344     case WINED3DSPSM_DW:
1345         shader_addline(buffer, "RCP T%c, %s.w;\n", 'A' + tmpreg, regstr);
1346         shader_addline(buffer, "MUL T%c, %s, T%c;\n", 'A' + tmpreg, regstr, 'A' + tmpreg);
1347         break;
1348     case WINED3DSPSM_ABS:
1349         if(ctx->target_version >= NV2) {
1350             sprintf(outregstr, "|%s%s|", regstr, swzstr);
1351             insert_line = 0;
1352         } else {
1353             shader_addline(buffer, "ABS T%c, %s;\n", 'A' + tmpreg, regstr);
1354         }
1355         break;
1356     case WINED3DSPSM_ABSNEG:
1357         if(ctx->target_version >= NV2) {
1358             sprintf(outregstr, "-|%s%s|", regstr, swzstr);
1359         } else {
1360             shader_addline(buffer, "ABS T%c, %s;\n", 'A' + tmpreg, regstr);
1361             sprintf(outregstr, "-T%c%s", 'A' + tmpreg, swzstr);
1362         }
1363         insert_line = 0;
1364         break;
1365     default:
1366         sprintf(outregstr, "%s%s", regstr, swzstr);
1367         insert_line = 0;
1368     }
1369
1370     /* Return modified or original register, with swizzle */
1371     if (insert_line)
1372         sprintf(outregstr, "T%c%s", 'A' + tmpreg, swzstr);
1373 }
1374
1375 static void pshader_hw_bem(const struct wined3d_shader_instruction *ins)
1376 {
1377     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1378     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1379     char dst_name[50];
1380     char src_name[2][50];
1381     DWORD sampler_code = dst->reg.idx;
1382
1383     shader_arb_get_dst_param(ins, dst, dst_name);
1384
1385     /* Sampling the perturbation map in Tsrc was done already, including the signedness correction if needed
1386      *
1387      * Keep in mind that src_name[1] can be "TB" and src_name[0] can be "TA" because modifiers like _x2 are valid
1388      * with bem. So delay loading the first parameter until after the perturbation calculation which needs two
1389      * temps is done.
1390      */
1391     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1392     shader_addline(buffer, "SWZ TA, bumpenvmat%d, x, z, 0, 0;\n", sampler_code);
1393     shader_addline(buffer, "DP3 TC.r, TA, %s;\n", src_name[1]);
1394     shader_addline(buffer, "SWZ TA, bumpenvmat%d, y, w, 0, 0;\n", sampler_code);
1395     shader_addline(buffer, "DP3 TC.g, TA, %s;\n", src_name[1]);
1396
1397     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
1398     shader_addline(buffer, "ADD %s, %s, TC;\n", dst_name, src_name[0]);
1399 }
1400
1401 static DWORD negate_modifiers(DWORD mod, char *extra_char)
1402 {
1403     *extra_char = ' ';
1404     switch(mod)
1405     {
1406         case WINED3DSPSM_NONE:      return WINED3DSPSM_NEG;
1407         case WINED3DSPSM_NEG:       return WINED3DSPSM_NONE;
1408         case WINED3DSPSM_BIAS:      return WINED3DSPSM_BIASNEG;
1409         case WINED3DSPSM_BIASNEG:   return WINED3DSPSM_BIAS;
1410         case WINED3DSPSM_SIGN:      return WINED3DSPSM_SIGNNEG;
1411         case WINED3DSPSM_SIGNNEG:   return WINED3DSPSM_SIGN;
1412         case WINED3DSPSM_COMP:      *extra_char = '-'; return WINED3DSPSM_COMP;
1413         case WINED3DSPSM_X2:        return WINED3DSPSM_X2NEG;
1414         case WINED3DSPSM_X2NEG:     return WINED3DSPSM_X2;
1415         case WINED3DSPSM_DZ:        *extra_char = '-'; return WINED3DSPSM_DZ;
1416         case WINED3DSPSM_DW:        *extra_char = '-'; return WINED3DSPSM_DW;
1417         case WINED3DSPSM_ABS:       return WINED3DSPSM_ABSNEG;
1418         case WINED3DSPSM_ABSNEG:    return WINED3DSPSM_ABS;
1419     }
1420     FIXME("Unknown modifier %u\n", mod);
1421     return mod;
1422 }
1423
1424 static void pshader_hw_cnd(const struct wined3d_shader_instruction *ins)
1425 {
1426     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1427     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1428     char dst_name[50];
1429     char src_name[3][50];
1430     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
1431             ins->ctx->reg_maps->shader_version.minor);
1432     BOOL is_color;
1433
1434     shader_arb_get_dst_param(ins, dst, dst_name);
1435     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1436
1437     /* The coissue flag changes the semantic of the cnd instruction in <= 1.3 shaders */
1438     if (shader_version <= WINED3D_SHADER_VERSION(1, 3) && ins->coissue)
1439     {
1440         shader_addline(buffer, "MOV%s %s, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name[1]);
1441     } else {
1442         struct wined3d_shader_src_param src0_copy = ins->src[0];
1443         char extra_neg;
1444
1445         /* src0 may have a negate srcmod set, so we can't blindly add "-" to the name */
1446         src0_copy.modifiers = negate_modifiers(src0_copy.modifiers, &extra_neg);
1447
1448         shader_arb_get_src_param(ins, &src0_copy, 0, src_name[0]);
1449         shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
1450         shader_addline(buffer, "ADD TA, %c%s, coefdiv.x;\n", extra_neg, src_name[0]);
1451         /* No modifiers supported on CMP */
1452         shader_addline(buffer, "CMP %s, TA, %s, %s;\n", dst_name, src_name[1], src_name[2]);
1453
1454         /* _SAT on CMP doesn't make much sense, but it is not a pure NOP */
1455         if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE)
1456         {
1457             shader_arb_get_register_name(ins, &dst->reg, src_name[0], &is_color);
1458             shader_addline(buffer, "MOV_SAT %s, %s;\n", dst_name, dst_name);
1459         }
1460     }
1461 }
1462
1463 static void pshader_hw_cmp(const struct wined3d_shader_instruction *ins)
1464 {
1465     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1466     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1467     char dst_name[50];
1468     char src_name[3][50];
1469     BOOL is_color;
1470
1471     shader_arb_get_dst_param(ins, dst, dst_name);
1472
1473     /* Generate input register names (with modifiers) */
1474     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
1475     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1476     shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
1477
1478     /* No modifiers are supported on CMP */
1479     shader_addline(buffer, "CMP %s, %s, %s, %s;\n", dst_name,
1480                    src_name[0], src_name[2], src_name[1]);
1481
1482     if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE)
1483     {
1484         shader_arb_get_register_name(ins, &dst->reg, src_name[0], &is_color);
1485         shader_addline(buffer, "MOV_SAT %s, %s;\n", dst_name, src_name[0]);
1486     }
1487 }
1488
1489 /** Process the WINED3DSIO_DP2ADD instruction in ARB.
1490  * dst = dot2(src0, src1) + src2 */
1491 static void pshader_hw_dp2add(const struct wined3d_shader_instruction *ins)
1492 {
1493     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1494     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1495     char dst_name[50];
1496     char src_name[3][50];
1497     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
1498
1499     shader_arb_get_dst_param(ins, dst, dst_name);
1500     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
1501     shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
1502
1503     if(ctx->target_version >= NV3)
1504     {
1505         /* GL_NV_fragment_program2 has a 1:1 matching instruction */
1506         shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1507         shader_addline(buffer, "DP2A%s %s, %s, %s, %s;\n", shader_arb_get_modifier(ins),
1508                        dst_name, src_name[0], src_name[1], src_name[2]);
1509     }
1510     else if(ctx->target_version >= NV2)
1511     {
1512         /* dst.x = src2.?, src0.x, src1.x + src0.y * src1.y
1513          * dst.y = src2.?, src0.x, src1.z + src0.y * src1.w
1514          * dst.z = src2.?, src0.x, src1.x + src0.y * src1.y
1515          * dst.z = src2.?, src0.x, src1.z + src0.y * src1.w
1516          *
1517          * Make sure that src1.zw = src1.xy, then we get a classic dp2add
1518          *
1519          * .xyxy and other swizzles that we could get with this are not valid in
1520          * plain ARBfp, but luckily the NV extension grammar lifts this limitation.
1521          */
1522         struct wined3d_shader_src_param tmp_param = ins->src[1];
1523         DWORD swizzle = tmp_param.swizzle & 0xf; /* Selects .xy */
1524         tmp_param.swizzle = swizzle | (swizzle << 4); /* Creates .xyxy */
1525
1526         shader_arb_get_src_param(ins, &tmp_param, 1, src_name[1]);
1527
1528         shader_addline(buffer, "X2D%s %s, %s, %s, %s;\n", shader_arb_get_modifier(ins),
1529                        dst_name, src_name[2], src_name[0], src_name[1]);
1530     }
1531     else
1532     {
1533         shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1534         /* 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]
1535         * src_name[0] can be TA, but TA is a private temp for modifiers, so it is save to overwrite
1536         */
1537         shader_addline(buffer, "MOV TA, %s;\n", src_name[0]);
1538         shader_addline(buffer, "MOV TA.z, 0.0;\n");
1539         shader_addline(buffer, "DP3 TA, TA, %s;\n", src_name[1]);
1540         shader_addline(buffer, "ADD%s %s, TA, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name[2]);
1541     }
1542 }
1543
1544 /* Map the opcode 1-to-1 to the GL code */
1545 static void shader_hw_map2gl(const struct wined3d_shader_instruction *ins)
1546 {
1547     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1548     const char *instruction;
1549     char arguments[256], dst_str[50];
1550     unsigned int i;
1551     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1552
1553     switch (ins->handler_idx)
1554     {
1555         case WINED3DSIH_ABS: instruction = "ABS"; break;
1556         case WINED3DSIH_ADD: instruction = "ADD"; break;
1557         case WINED3DSIH_CRS: instruction = "XPD"; break;
1558         case WINED3DSIH_DP3: instruction = "DP3"; break;
1559         case WINED3DSIH_DP4: instruction = "DP4"; break;
1560         case WINED3DSIH_DST: instruction = "DST"; break;
1561         case WINED3DSIH_FRC: instruction = "FRC"; break;
1562         case WINED3DSIH_LIT: instruction = "LIT"; break;
1563         case WINED3DSIH_LRP: instruction = "LRP"; break;
1564         case WINED3DSIH_MAD: instruction = "MAD"; break;
1565         case WINED3DSIH_MAX: instruction = "MAX"; break;
1566         case WINED3DSIH_MIN: instruction = "MIN"; break;
1567         case WINED3DSIH_MOV: instruction = "MOV"; break;
1568         case WINED3DSIH_MUL: instruction = "MUL"; break;
1569         case WINED3DSIH_SGE: instruction = "SGE"; break;
1570         case WINED3DSIH_SLT: instruction = "SLT"; break;
1571         case WINED3DSIH_SUB: instruction = "SUB"; break;
1572         case WINED3DSIH_MOVA:instruction = "ARR"; break;
1573         case WINED3DSIH_SGN: instruction = "SSG"; break;
1574         case WINED3DSIH_DSX: instruction = "DDX"; break;
1575         default: instruction = "";
1576             FIXME("Unhandled opcode %#x\n", ins->handler_idx);
1577             break;
1578     }
1579
1580     /* Note that shader_arb_add_dst_param() adds spaces. */
1581     arguments[0] = '\0';
1582     shader_arb_get_dst_param(ins, dst, dst_str);
1583     for (i = 0; i < ins->src_count; ++i)
1584     {
1585         char operand[100];
1586         strcat(arguments, ", ");
1587         shader_arb_get_src_param(ins, &ins->src[i], i, operand);
1588         strcat(arguments, operand);
1589     }
1590     shader_addline(buffer, "%s%s %s%s;\n", instruction, shader_arb_get_modifier(ins), dst_str, arguments);
1591 }
1592
1593 static void shader_hw_nop(const struct wined3d_shader_instruction *ins)
1594 {
1595     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1596     shader_addline(buffer, "NOP;\n");
1597 }
1598
1599 static void shader_hw_mov(const struct wined3d_shader_instruction *ins)
1600 {
1601     IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
1602     BOOL pshader = shader_is_pshader_version(shader->baseShader.reg_maps.shader_version.type);
1603     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
1604
1605     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1606     char src0_param[256];
1607
1608     if(ins->handler_idx == WINED3DSIH_MOVA) {
1609         char write_mask[6];
1610
1611         if(ctx->target_version >= NV2) {
1612             shader_hw_map2gl(ins);
1613             return;
1614         }
1615         shader_arb_get_src_param(ins, &ins->src[0], 0, src0_param);
1616         shader_arb_get_write_mask(ins, &ins->dst[0], write_mask);
1617
1618         /* This implements the mova formula used in GLSL. The first two instructions
1619          * prepare the sign() part. Note that it is fine to have my_sign(0.0) = 1.0
1620          * in this case:
1621          * mova A0.x, 0.0
1622          *
1623          * A0.x = arl(floor(abs(0.0) + 0.5) * 1.0) = floor(0.5) = 0.0 since arl does a floor
1624          *
1625          * The ARL is performed when A0 is used - the requested component is read from A0_SHADOW into
1626          * A0.x. We can use the overwritten component of A0_shadow as temporary storage for the sign.
1627          */
1628         shader_addline(buffer, "SGE A0_SHADOW%s, %s, mova_const.y;\n", write_mask, src0_param);
1629         shader_addline(buffer, "MAD A0_SHADOW%s, A0_SHADOW, mova_const.z, -mova_const.w;\n", write_mask);
1630
1631         shader_addline(buffer, "ABS TA%s, %s;\n", write_mask, src0_param);
1632         shader_addline(buffer, "ADD TA%s, TA, mova_const.x;\n", write_mask);
1633         shader_addline(buffer, "FLR TA%s, TA;\n", write_mask);
1634         if (((IWineD3DVertexShaderImpl *)shader)->rel_offset)
1635         {
1636             shader_addline(buffer, "ADD TA%s, TA, helper_const.z;\n", write_mask);
1637         }
1638         shader_addline(buffer, "MUL A0_SHADOW%s, TA, A0_SHADOW;\n", write_mask);
1639
1640         ((struct shader_arb_ctx_priv *)ins->ctx->backend_data)->addr_reg[0] = '\0';
1641     } else if (ins->ctx->reg_maps->shader_version.major == 1
1642           && !shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type)
1643           && ins->dst[0].reg.type == WINED3DSPR_ADDR)
1644     {
1645         src0_param[0] = '\0';
1646         if (((IWineD3DVertexShaderImpl *)shader)->rel_offset)
1647         {
1648             shader_arb_get_src_param(ins, &ins->src[0], 0, src0_param);
1649             shader_addline(buffer, "ADD TA.x, %s, helper_const.z;\n", src0_param);
1650             shader_addline(buffer, "ARL A0.x, TA.x;\n");
1651         }
1652         else
1653         {
1654             /* Apple's ARB_vertex_program implementation does not accept an ARL source argument
1655              * with more than one component. Thus replicate the first source argument over all
1656              * 4 components. For example, .xyzw -> .x (or better: .xxxx), .zwxy -> .z, etc) */
1657             struct wined3d_shader_src_param tmp_src = ins->src[0];
1658             tmp_src.swizzle = (tmp_src.swizzle & 0x3) * 0x55;
1659             shader_arb_get_src_param(ins, &tmp_src, 0, src0_param);
1660             shader_addline(buffer, "ARL A0.x, %s;\n", src0_param);
1661         }
1662     }
1663     else if(ins->dst[0].reg.type == WINED3DSPR_COLOROUT && ins->dst[0].reg.idx == 0 && pshader)
1664     {
1665         IWineD3DPixelShaderImpl *ps = (IWineD3DPixelShaderImpl *) shader;
1666         if(ctx->cur_ps_args->super.srgb_correction && ps->color0_mov)
1667         {
1668             shader_addline(buffer, "#mov handled in srgb write code\n");
1669             return;
1670         }
1671         shader_hw_map2gl(ins);
1672     }
1673     else
1674     {
1675         shader_hw_map2gl(ins);
1676     }
1677 }
1678
1679 static void pshader_hw_texkill(const struct wined3d_shader_instruction *ins)
1680 {
1681     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1682     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1683     char reg_dest[40];
1684
1685     /* No swizzles are allowed in d3d's texkill. PS 1.x ignores the 4th component as documented,
1686      * but >= 2.0 honors it(undocumented, but tested by the d3d9 testsuit)
1687      */
1688     shader_arb_get_dst_param(ins, dst, reg_dest);
1689
1690     if (ins->ctx->reg_maps->shader_version.major >= 2)
1691     {
1692         const char *kilsrc = "TA";
1693         BOOL is_color;
1694
1695         shader_arb_get_register_name(ins, &dst->reg, reg_dest, &is_color);
1696         if(dst->write_mask == WINED3DSP_WRITEMASK_ALL)
1697         {
1698             kilsrc = reg_dest;
1699         }
1700         else
1701         {
1702             /* Sigh. KIL doesn't support swizzles/writemasks. KIL passes a writemask, but ".xy" for example
1703              * is not valid as a swizzle in ARB (needs ".xyyy"). Use SWZ to load the register properly, and set
1704              * masked out components to 0(won't kill)
1705              */
1706             char x = '0', y = '0', z = '0', w = '0';
1707             if(dst->write_mask & WINED3DSP_WRITEMASK_0) x = 'x';
1708             if(dst->write_mask & WINED3DSP_WRITEMASK_1) y = 'y';
1709             if(dst->write_mask & WINED3DSP_WRITEMASK_2) z = 'z';
1710             if(dst->write_mask & WINED3DSP_WRITEMASK_3) w = 'w';
1711             shader_addline(buffer, "SWZ TA, %s, %c, %c, %c, %c;\n", reg_dest, x, y, z, w);
1712         }
1713         shader_addline(buffer, "KIL %s;\n", kilsrc);
1714     } else {
1715         /* ARB fp doesn't like swizzles on the parameter of the KIL instruction. To mask the 4th component,
1716          * copy the register into our general purpose TMP variable, overwrite .w and pass TMP to KIL
1717          *
1718          * ps_1_3 shaders use the texcoord incarnation of the Tx register. ps_1_4 shaders can use the same,
1719          * or pass in any temporary register(in shader phase 2)
1720          */
1721         if(ins->ctx->reg_maps->shader_version.minor <= 3) {
1722             sprintf(reg_dest, "fragment.texcoord[%u]", dst->reg.idx);
1723         } else {
1724             shader_arb_get_dst_param(ins, dst, reg_dest);
1725         }
1726         shader_addline(buffer, "SWZ TA, %s, x, y, z, 1;\n", reg_dest);
1727         shader_addline(buffer, "KIL TA;\n");
1728     }
1729 }
1730
1731 static void pshader_hw_tex(const struct wined3d_shader_instruction *ins)
1732 {
1733     IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)ins->ctx->shader;
1734     IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
1735     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1736     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
1737             ins->ctx->reg_maps->shader_version.minor);
1738     struct wined3d_shader_src_param src;
1739
1740     char reg_dest[40];
1741     char reg_coord[40];
1742     DWORD reg_sampler_code;
1743     DWORD myflags = 0;
1744
1745     /* All versions have a destination register */
1746     shader_arb_get_dst_param(ins, dst, reg_dest);
1747
1748     /* 1.0-1.4: Use destination register number as texture code.
1749        2.0+: Use provided sampler number as texure code. */
1750     if (shader_version < WINED3D_SHADER_VERSION(2,0))
1751         reg_sampler_code = dst->reg.idx;
1752     else
1753         reg_sampler_code = ins->src[1].reg.idx;
1754
1755     /* 1.0-1.3: Use the texcoord varying.
1756        1.4+: Use provided coordinate source register. */
1757     if (shader_version < WINED3D_SHADER_VERSION(1,4))
1758         sprintf(reg_coord, "fragment.texcoord[%u]", reg_sampler_code);
1759     else {
1760         /* TEX is the only instruction that can handle DW and DZ natively */
1761         src = ins->src[0];
1762         if(src.modifiers == WINED3DSPSM_DW) src.modifiers = WINED3DSPSM_NONE;
1763         if(src.modifiers == WINED3DSPSM_DZ) src.modifiers = WINED3DSPSM_NONE;
1764         shader_arb_get_src_param(ins, &src, 0, reg_coord);
1765     }
1766
1767     /* projection flag:
1768      * 1.1, 1.2, 1.3: Use WINED3DTSS_TEXTURETRANSFORMFLAGS
1769      * 1.4: Use WINED3DSPSM_DZ or WINED3DSPSM_DW on src[0]
1770      * 2.0+: Use WINED3DSI_TEXLD_PROJECT on the opcode
1771      */
1772     if (shader_version < WINED3D_SHADER_VERSION(1,4))
1773     {
1774         DWORD flags = 0;
1775         if(reg_sampler_code < MAX_TEXTURES) {
1776             flags = deviceImpl->stateBlock->textureState[reg_sampler_code][WINED3DTSS_TEXTURETRANSFORMFLAGS];
1777         }
1778         if (flags & WINED3DTTFF_PROJECTED) {
1779             myflags |= TEX_PROJ;
1780         }
1781     }
1782     else if (shader_version < WINED3D_SHADER_VERSION(2,0))
1783     {
1784         DWORD src_mod = ins->src[0].modifiers;
1785         if (src_mod == WINED3DSPSM_DZ) {
1786             /* TXP cannot handle DZ natively, so move the z coordinate to .w. reg_coord is a read-only
1787              * varying register, so we need a temp reg
1788              */
1789             shader_addline(ins->ctx->buffer, "SWZ TA, %s, x, y, z, z;\n", reg_coord);
1790             strcpy(reg_coord, "TA");
1791             myflags |= TEX_PROJ;
1792         } else if(src_mod == WINED3DSPSM_DW) {
1793             myflags |= TEX_PROJ;
1794         }
1795     } else {
1796         if (ins->flags & WINED3DSI_TEXLD_PROJECT) myflags |= TEX_PROJ;
1797         if (ins->flags & WINED3DSI_TEXLD_BIAS) myflags |= TEX_BIAS;
1798     }
1799     shader_hw_sample(ins, reg_sampler_code, reg_dest, reg_coord, myflags, NULL, NULL);
1800 }
1801
1802 static void pshader_hw_texcoord(const struct wined3d_shader_instruction *ins)
1803 {
1804     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1805     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1806     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
1807             ins->ctx->reg_maps->shader_version.minor);
1808     char dst_str[50];
1809
1810     if (shader_version < WINED3D_SHADER_VERSION(1,4))
1811     {
1812         DWORD reg = dst->reg.idx;
1813
1814         shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
1815         shader_addline(buffer, "MOV_SAT %s, fragment.texcoord[%u];\n", dst_str, reg);
1816     } else {
1817         char reg_src[40];
1818
1819         shader_arb_get_src_param(ins, &ins->src[0], 0, reg_src);
1820         shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
1821         shader_addline(buffer, "MOV %s, %s;\n", dst_str, reg_src);
1822    }
1823 }
1824
1825 static void pshader_hw_texreg2ar(const struct wined3d_shader_instruction *ins)
1826 {
1827      struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1828      IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)ins->ctx->shader;
1829      IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
1830      DWORD flags;
1831
1832      DWORD reg1 = ins->dst[0].reg.idx;
1833      char dst_str[50];
1834      char src_str[50];
1835
1836      /* Note that texreg2ar treats Tx as a temporary register, not as a varying */
1837      shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
1838      shader_arb_get_src_param(ins, &ins->src[0], 0, src_str);
1839      /* Move .x first in case src_str is "TA" */
1840      shader_addline(buffer, "MOV TA.y, %s.x;\n", src_str);
1841      shader_addline(buffer, "MOV TA.x, %s.w;\n", src_str);
1842      flags = reg1 < MAX_TEXTURES ? deviceImpl->stateBlock->textureState[reg1][WINED3DTSS_TEXTURETRANSFORMFLAGS] : 0;
1843      shader_hw_sample(ins, reg1, dst_str, "TA", flags & WINED3DTTFF_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
1844 }
1845
1846 static void pshader_hw_texreg2gb(const struct wined3d_shader_instruction *ins)
1847 {
1848      struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1849
1850      DWORD reg1 = ins->dst[0].reg.idx;
1851      char dst_str[50];
1852      char src_str[50];
1853
1854      /* Note that texreg2gb treats Tx as a temporary register, not as a varying */
1855      shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
1856      shader_arb_get_src_param(ins, &ins->src[0], 0, src_str);
1857      shader_addline(buffer, "MOV TA.x, %s.y;\n", src_str);
1858      shader_addline(buffer, "MOV TA.y, %s.z;\n", src_str);
1859      shader_hw_sample(ins, reg1, dst_str, "TA", 0, NULL, NULL);
1860 }
1861
1862 static void pshader_hw_texreg2rgb(const struct wined3d_shader_instruction *ins)
1863 {
1864     DWORD reg1 = ins->dst[0].reg.idx;
1865     char dst_str[50];
1866     char src_str[50];
1867
1868     /* Note that texreg2rg treats Tx as a temporary register, not as a varying */
1869     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
1870     shader_arb_get_src_param(ins, &ins->src[0], 0, src_str);
1871     shader_hw_sample(ins, reg1, dst_str, src_str, 0, NULL, NULL);
1872 }
1873
1874 static void pshader_hw_texbem(const struct wined3d_shader_instruction *ins)
1875 {
1876     IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)ins->ctx->shader;
1877     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1878     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1879     char reg_coord[40], dst_reg[50], src_reg[50];
1880     DWORD reg_dest_code;
1881
1882     /* All versions have a destination register. The Tx where the texture coordinates come
1883      * from is the varying incarnation of the texture register
1884      */
1885     reg_dest_code = dst->reg.idx;
1886     shader_arb_get_dst_param(ins, &ins->dst[0], dst_reg);
1887     shader_arb_get_src_param(ins, &ins->src[0], 0, src_reg);
1888     sprintf(reg_coord, "fragment.texcoord[%u]", reg_dest_code);
1889
1890     /* Sampling the perturbation map in Tsrc was done already, including the signedness correction if needed
1891      * The Tx in which the perturbation map is stored is the tempreg incarnation of the texture register
1892      *
1893      * GL_NV_fragment_program_option could handle this in one instruction via X2D:
1894      * X2D TA.xy, fragment.texcoord, T%u, bumpenvmat%u.xzyw
1895      *
1896      * However, the NV extensions are never enabled for <= 2.0 shaders because of the performance penalty that
1897      * comes with it, and texbem is an 1.x only instruction. No 1.x instruction forces us to enable the NV
1898      * extension.
1899      */
1900     shader_addline(buffer, "SWZ TB, bumpenvmat%d, x, z, 0, 0;\n", reg_dest_code);
1901     shader_addline(buffer, "DP3 TA.x, TB, %s;\n", src_reg);
1902     shader_addline(buffer, "SWZ TB, bumpenvmat%d, y, w, 0, 0;\n", reg_dest_code);
1903     shader_addline(buffer, "DP3 TA.y, TB, %s;\n", src_reg);
1904
1905     /* with projective textures, texbem only divides the static texture coord, not the displacement,
1906      * so we can't let the GL handle this.
1907      */
1908     if (((IWineD3DDeviceImpl*) This->baseShader.device)->stateBlock->textureState[reg_dest_code][WINED3DTSS_TEXTURETRANSFORMFLAGS]
1909             & WINED3DTTFF_PROJECTED) {
1910         shader_addline(buffer, "RCP TB.w, %s.w;\n", reg_coord);
1911         shader_addline(buffer, "MUL TB.xy, %s, TB.w;\n", reg_coord);
1912         shader_addline(buffer, "ADD TA.xy, TA, TB;\n");
1913     } else {
1914         shader_addline(buffer, "ADD TA.xy, TA, %s;\n", reg_coord);
1915     }
1916
1917     shader_hw_sample(ins, reg_dest_code, dst_reg, "TA", 0, NULL, NULL);
1918
1919     if (ins->handler_idx == WINED3DSIH_TEXBEML)
1920     {
1921         /* No src swizzles are allowed, so this is ok */
1922         shader_addline(buffer, "MAD TA, %s.z, luminance%d.x, luminance%d.y;\n",
1923                        src_reg, reg_dest_code, reg_dest_code);
1924         shader_addline(buffer, "MUL %s, %s, TA;\n", dst_reg, dst_reg);
1925     }
1926 }
1927
1928 static void pshader_hw_texm3x2pad(const struct wined3d_shader_instruction *ins)
1929 {
1930     DWORD reg = ins->dst[0].reg.idx;
1931     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1932     char src0_name[50], dst_name[50];
1933     BOOL is_color;
1934     struct wined3d_shader_register tmp_reg = ins->dst[0].reg;
1935
1936     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
1937     /* The next instruction will be a texm3x2tex or texm3x2depth that writes to the uninitialized
1938      * T<reg+1> register. Use this register to store the calculated vector
1939      */
1940     tmp_reg.idx = reg + 1;
1941     shader_arb_get_register_name(ins, &tmp_reg, dst_name, &is_color);
1942     shader_addline(buffer, "DP3 %s.x, fragment.texcoord[%u], %s;\n", dst_name, reg, src0_name);
1943 }
1944
1945 static void pshader_hw_texm3x2tex(const struct wined3d_shader_instruction *ins)
1946 {
1947     IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)ins->ctx->shader;
1948     IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
1949     DWORD flags;
1950     DWORD reg = ins->dst[0].reg.idx;
1951     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1952     char dst_str[50];
1953     char src0_name[50];
1954     char dst_reg[50];
1955     BOOL is_color;
1956
1957     /* We know that we're writing to the uninitialized T<reg> register, so use it for temporary storage */
1958     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_reg, &is_color);
1959
1960     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
1961     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
1962     shader_addline(buffer, "DP3 %s.y, fragment.texcoord[%u], %s;\n", dst_reg, reg, src0_name);
1963     flags = reg < MAX_TEXTURES ? deviceImpl->stateBlock->textureState[reg][WINED3DTSS_TEXTURETRANSFORMFLAGS] : 0;
1964     shader_hw_sample(ins, reg, dst_str, dst_reg, flags & WINED3DTTFF_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
1965 }
1966
1967 static void pshader_hw_texm3x3pad(const struct wined3d_shader_instruction *ins)
1968 {
1969     IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)ins->ctx->shader;
1970     DWORD reg = ins->dst[0].reg.idx;
1971     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1972     SHADER_PARSE_STATE* current_state = &This->baseShader.parse_state;
1973     char src0_name[50], dst_name[50];
1974     struct wined3d_shader_register tmp_reg = ins->dst[0].reg;
1975     BOOL is_color;
1976
1977     /* There are always 2 texm3x3pad instructions followed by one texm3x3[tex,vspec, ...] instruction, with
1978      * incrementing ins->dst[0].register_idx numbers. So the pad instruction already knows the final destination
1979      * register, and this register is uninitialized(otherwise the assembler complains that it is 'redeclared')
1980      */
1981     tmp_reg.idx = reg + 2 - current_state->current_row;
1982     shader_arb_get_register_name(ins, &tmp_reg, dst_name, &is_color);
1983
1984     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
1985     shader_addline(buffer, "DP3 %s.%c, fragment.texcoord[%u], %s;\n",
1986                    dst_name, 'x' + current_state->current_row, reg, src0_name);
1987     current_state->texcoord_w[current_state->current_row++] = reg;
1988 }
1989
1990 static void pshader_hw_texm3x3tex(const struct wined3d_shader_instruction *ins)
1991 {
1992     IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)ins->ctx->shader;
1993     IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
1994     DWORD flags;
1995     DWORD reg = ins->dst[0].reg.idx;
1996     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1997     SHADER_PARSE_STATE* current_state = &This->baseShader.parse_state;
1998     char dst_str[50];
1999     char src0_name[50], dst_name[50];
2000     BOOL is_color;
2001
2002     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2003     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2004     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_name, reg, src0_name);
2005
2006     /* Sample the texture using the calculated coordinates */
2007     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2008     flags = reg < MAX_TEXTURES ? deviceImpl->stateBlock->textureState[reg][WINED3DTSS_TEXTURETRANSFORMFLAGS] : 0;
2009     shader_hw_sample(ins, reg, dst_str, dst_name, flags & WINED3DTTFF_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2010     current_state->current_row = 0;
2011 }
2012
2013 static void pshader_hw_texm3x3vspec(const struct wined3d_shader_instruction *ins)
2014 {
2015     IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)ins->ctx->shader;
2016     IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
2017     DWORD flags;
2018     DWORD reg = ins->dst[0].reg.idx;
2019     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2020     SHADER_PARSE_STATE* current_state = &This->baseShader.parse_state;
2021     char dst_str[50];
2022     char src0_name[50];
2023     char dst_reg[8];
2024     BOOL is_color;
2025
2026     /* Get the dst reg without writemask strings. We know this register is uninitialized, so we can use all
2027      * components for temporary data storage
2028      */
2029     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_reg, &is_color);
2030     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2031     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_reg, reg, src0_name);
2032
2033     /* Construct the eye-ray vector from w coordinates */
2034     shader_addline(buffer, "MOV TB.x, fragment.texcoord[%u].w;\n", current_state->texcoord_w[0]);
2035     shader_addline(buffer, "MOV TB.y, fragment.texcoord[%u].w;\n", current_state->texcoord_w[1]);
2036     shader_addline(buffer, "MOV TB.z, fragment.texcoord[%u].w;\n", reg);
2037
2038     /* Calculate reflection vector
2039      */
2040     shader_addline(buffer, "DP3 %s.w, %s, TB;\n", dst_reg, dst_reg);
2041     /* The .w is ignored when sampling, so I can use TB.w to calculate dot(N, N) */
2042     shader_addline(buffer, "DP3 TB.w, %s, %s;\n", dst_reg, dst_reg);
2043     shader_addline(buffer, "RCP TB.w, TB.w;\n");
2044     shader_addline(buffer, "MUL %s.w, %s.w, TB.w;\n", dst_reg, dst_reg);
2045     shader_addline(buffer, "MUL %s, %s.w, %s;\n", dst_reg, dst_reg, dst_reg);
2046     shader_addline(buffer, "MAD %s, coefmul.x, %s, -TB;\n", dst_reg, dst_reg);
2047
2048     /* Sample the texture using the calculated coordinates */
2049     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2050     flags = reg < MAX_TEXTURES ? deviceImpl->stateBlock->textureState[reg][WINED3DTSS_TEXTURETRANSFORMFLAGS] : 0;
2051     shader_hw_sample(ins, reg, dst_str, dst_reg, flags & WINED3DTTFF_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2052     current_state->current_row = 0;
2053 }
2054
2055 static void pshader_hw_texm3x3spec(const struct wined3d_shader_instruction *ins)
2056 {
2057     IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)ins->ctx->shader;
2058     IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
2059     DWORD flags;
2060     DWORD reg = ins->dst[0].reg.idx;
2061     SHADER_PARSE_STATE* current_state = &This->baseShader.parse_state;
2062     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2063     char dst_str[50];
2064     char src0_name[50];
2065     char src1_name[50];
2066     char dst_reg[8];
2067     BOOL is_color;
2068
2069     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2070     shader_arb_get_src_param(ins, &ins->src[0], 1, src1_name);
2071     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_reg, &is_color);
2072     /* Note: dst_reg.xy is input here, generated by two texm3x3pad instructions */
2073     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_reg, reg, src0_name);
2074
2075     /* Calculate reflection vector.
2076      *
2077      *                   dot(N, E)
2078      * dst_reg.xyz = 2 * --------- * N - E
2079      *                   dot(N, N)
2080      *
2081      * Which normalizes the normal vector
2082      */
2083     shader_addline(buffer, "DP3 %s.w, %s, %s;\n", dst_reg, dst_reg, src1_name);
2084     shader_addline(buffer, "DP3 TC.w, %s, %s;\n", dst_reg, dst_reg);
2085     shader_addline(buffer, "RCP TC.w, TC.w;\n");
2086     shader_addline(buffer, "MUL %s.w, %s.w, TC.w;\n", dst_reg, dst_reg);
2087     shader_addline(buffer, "MUL %s, %s.w, %s;\n", dst_reg, dst_reg, dst_reg);
2088     shader_addline(buffer, "MAD %s, coefmul.x, %s, -%s;\n", dst_reg, dst_reg, src1_name);
2089
2090     /* Sample the texture using the calculated coordinates */
2091     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2092     flags = reg < MAX_TEXTURES ? deviceImpl->stateBlock->textureState[reg][WINED3DTSS_TEXTURETRANSFORMFLAGS] : 0;
2093     shader_hw_sample(ins, reg, dst_str, dst_reg, flags & WINED3DTTFF_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2094     current_state->current_row = 0;
2095 }
2096
2097 static void pshader_hw_texdepth(const struct wined3d_shader_instruction *ins)
2098 {
2099     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2100     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2101     char dst_name[50];
2102
2103     /* texdepth has an implicit destination, the fragment depth value. It's only parameter,
2104      * which is essentially an input, is the destination register because it is the first
2105      * parameter. According to the msdn, this must be register r5, but let's keep it more flexible
2106      * here(writemasks/swizzles are not valid on texdepth)
2107      */
2108     shader_arb_get_dst_param(ins, dst, dst_name);
2109
2110     /* According to the msdn, the source register(must be r5) is unusable after
2111      * the texdepth instruction, so we're free to modify it
2112      */
2113     shader_addline(buffer, "MIN %s.y, %s.y, one.y;\n", dst_name, dst_name);
2114
2115     /* How to deal with the special case dst_name.g == 0? if r != 0, then
2116      * the r * (1 / 0) will give infinity, which is clamped to 1.0, the correct
2117      * result. But if r = 0.0, then 0 * inf = 0, which is incorrect.
2118      */
2119     shader_addline(buffer, "RCP %s.y, %s.y;\n", dst_name, dst_name);
2120     shader_addline(buffer, "MUL TA.x, %s.x, %s.y;\n", dst_name, dst_name);
2121     shader_addline(buffer, "MIN TA.x, TA.x, one.x;\n");
2122     shader_addline(buffer, "MAX result.depth, TA.x, 0.0;\n");
2123 }
2124
2125 /** Process the WINED3DSIO_TEXDP3TEX instruction in ARB:
2126  * Take a 3-component dot product of the TexCoord[dstreg] and src,
2127  * then perform a 1D texture lookup from stage dstregnum, place into dst. */
2128 static void pshader_hw_texdp3tex(const struct wined3d_shader_instruction *ins)
2129 {
2130     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2131     DWORD sampler_idx = ins->dst[0].reg.idx;
2132     char src0[50];
2133     char dst_str[50];
2134
2135     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2136     shader_addline(buffer, "MOV TB, 0.0;\n");
2137     shader_addline(buffer, "DP3 TB.x, fragment.texcoord[%u], %s;\n", sampler_idx, src0);
2138
2139     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2140     shader_hw_sample(ins, sampler_idx, dst_str, "TB", 0 /* Only one coord, can't be projected */, NULL, NULL);
2141 }
2142
2143 /** Process the WINED3DSIO_TEXDP3 instruction in ARB:
2144  * Take a 3-component dot product of the TexCoord[dstreg] and src. */
2145 static void pshader_hw_texdp3(const struct wined3d_shader_instruction *ins)
2146 {
2147     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2148     char src0[50];
2149     char dst_str[50];
2150     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2151
2152     /* Handle output register */
2153     shader_arb_get_dst_param(ins, dst, dst_str);
2154     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2155     shader_addline(buffer, "DP3 %s, fragment.texcoord[%u], %s;\n", dst_str, dst->reg.idx, src0);
2156 }
2157
2158 /** Process the WINED3DSIO_TEXM3X3 instruction in ARB
2159  * Perform the 3rd row of a 3x3 matrix multiply */
2160 static void pshader_hw_texm3x3(const struct wined3d_shader_instruction *ins)
2161 {
2162     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2163     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2164     char dst_str[50], dst_name[50];
2165     char src0[50];
2166     BOOL is_color;
2167
2168     shader_arb_get_dst_param(ins, dst, dst_str);
2169     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2170     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2171     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_name, dst->reg.idx, src0);
2172     shader_addline(buffer, "MOV %s, %s;\n", dst_str, dst_name);
2173 }
2174
2175 /** Process the WINED3DSIO_TEXM3X2DEPTH instruction in ARB:
2176  * Last row of a 3x2 matrix multiply, use the result to calculate the depth:
2177  * Calculate tmp0.y = TexCoord[dstreg] . src.xyz;  (tmp0.x has already been calculated)
2178  * depth = (tmp0.y == 0.0) ? 1.0 : tmp0.x / tmp0.y
2179  */
2180 static void pshader_hw_texm3x2depth(const struct wined3d_shader_instruction *ins)
2181 {
2182     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2183     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2184     char src0[50], dst_name[50];
2185     BOOL is_color;
2186
2187     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2188     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2189     shader_addline(buffer, "DP3 %s.y, fragment.texcoord[%u], %s;\n", dst_name, dst->reg.idx, src0);
2190
2191     /* How to deal with the special case dst_name.g == 0? if r != 0, then
2192      * the r * (1 / 0) will give infinity, which is clamped to 1.0, the correct
2193      * result. But if r = 0.0, then 0 * inf = 0, which is incorrect.
2194      */
2195     shader_addline(buffer, "RCP %s.y, %s.y;\n", dst_name, dst_name);
2196     shader_addline(buffer, "MUL %s.x, %s.x, %s.y;\n", dst_name, dst_name, dst_name);
2197     shader_addline(buffer, "MIN %s.x, %s.x, one.x;\n", dst_name, dst_name);
2198     shader_addline(buffer, "MAX result.depth, %s.x, 0.0;\n", dst_name);
2199 }
2200
2201 /** Handles transforming all WINED3DSIO_M?x? opcodes for
2202     Vertex/Pixel shaders to ARB_vertex_program codes */
2203 static void shader_hw_mnxn(const struct wined3d_shader_instruction *ins)
2204 {
2205     int i;
2206     int nComponents = 0;
2207     struct wined3d_shader_dst_param tmp_dst = {{0}};
2208     struct wined3d_shader_src_param tmp_src[2] = {{{0}}};
2209     struct wined3d_shader_instruction tmp_ins;
2210
2211     memset(&tmp_ins, 0, sizeof(tmp_ins));
2212
2213     /* Set constants for the temporary argument */
2214     tmp_ins.ctx = ins->ctx;
2215     tmp_ins.dst_count = 1;
2216     tmp_ins.dst = &tmp_dst;
2217     tmp_ins.src_count = 2;
2218     tmp_ins.src = tmp_src;
2219
2220     switch(ins->handler_idx)
2221     {
2222         case WINED3DSIH_M4x4:
2223             nComponents = 4;
2224             tmp_ins.handler_idx = WINED3DSIH_DP4;
2225             break;
2226         case WINED3DSIH_M4x3:
2227             nComponents = 3;
2228             tmp_ins.handler_idx = WINED3DSIH_DP4;
2229             break;
2230         case WINED3DSIH_M3x4:
2231             nComponents = 4;
2232             tmp_ins.handler_idx = WINED3DSIH_DP3;
2233             break;
2234         case WINED3DSIH_M3x3:
2235             nComponents = 3;
2236             tmp_ins.handler_idx = WINED3DSIH_DP3;
2237             break;
2238         case WINED3DSIH_M3x2:
2239             nComponents = 2;
2240             tmp_ins.handler_idx = WINED3DSIH_DP3;
2241             break;
2242         default:
2243             FIXME("Unhandled opcode %#x\n", ins->handler_idx);
2244             break;
2245     }
2246
2247     tmp_dst = ins->dst[0];
2248     tmp_src[0] = ins->src[0];
2249     tmp_src[1] = ins->src[1];
2250     for (i = 0; i < nComponents; i++) {
2251         tmp_dst.write_mask = WINED3DSP_WRITEMASK_0 << i;
2252         shader_hw_map2gl(&tmp_ins);
2253         ++tmp_src[1].reg.idx;
2254     }
2255 }
2256
2257 static void shader_hw_scalar_op(const struct wined3d_shader_instruction *ins)
2258 {
2259     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2260     const char *instruction;
2261
2262     char dst[50];
2263     char src[50];
2264
2265     switch(ins->handler_idx)
2266     {
2267         case WINED3DSIH_RSQ:  instruction = "RSQ"; break;
2268         case WINED3DSIH_RCP:  instruction = "RCP"; break;
2269         case WINED3DSIH_EXP:  instruction = "EX2"; break;
2270         case WINED3DSIH_EXPP: instruction = "EXP"; break;
2271         default: instruction = "";
2272             FIXME("Unhandled opcode %#x\n", ins->handler_idx);
2273             break;
2274     }
2275
2276     shader_arb_get_dst_param(ins, &ins->dst[0], dst); /* Destination */
2277     shader_arb_get_src_param(ins, &ins->src[0], 0, src);
2278     if (ins->src[0].swizzle == WINED3DSP_NOSWIZZLE)
2279     {
2280         /* Dx sdk says .x is used if no swizzle is given, but our test shows that
2281          * .w is used
2282          */
2283         strcat(src, ".w");
2284     }
2285
2286     shader_addline(buffer, "%s%s %s, %s;\n", instruction, shader_arb_get_modifier(ins), dst, src);
2287 }
2288
2289 static void shader_hw_nrm(const struct wined3d_shader_instruction *ins)
2290 {
2291     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2292     char dst_name[50];
2293     char src_name[50];
2294     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2295     BOOL pshader = shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type);
2296
2297     shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2298     shader_arb_get_src_param(ins, &ins->src[0], 1 /* Use TB */, src_name);
2299
2300     if(pshader && priv->target_version >= NV3)
2301     {
2302         shader_addline(buffer, "NRM%s %s, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name);
2303     }
2304     else
2305     {
2306         shader_addline(buffer, "DP3 TA, %s, %s;\n", src_name, src_name);
2307         shader_addline(buffer, "RSQ TA, TA.x;\n");
2308         /* dst.w = src[0].w * 1 / (src.x^2 + src.y^2 + src.z^2)^(1/2) according to msdn*/
2309         shader_addline(buffer, "MUL%s %s, %s, TA;\n", shader_arb_get_modifier(ins), dst_name,
2310                     src_name);
2311     }
2312 }
2313
2314 static void shader_hw_lrp(const struct wined3d_shader_instruction *ins)
2315 {
2316     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2317     char dst_name[50];
2318     char src_name[3][50];
2319
2320     /* ARB_fragment_program has a convenient LRP instruction */
2321     if(shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type)) {
2322         shader_hw_map2gl(ins);
2323         return;
2324     }
2325
2326     shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2327     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
2328     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
2329     shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
2330
2331     shader_addline(buffer, "SUB TA, %s, %s;\n", src_name[1], src_name[2]);
2332     shader_addline(buffer, "MAD%s %s, %s, TA, %s;\n", shader_arb_get_modifier(ins),
2333                    dst_name, src_name[0], src_name[2]);
2334 }
2335
2336 static void shader_hw_sincos(const struct wined3d_shader_instruction *ins)
2337 {
2338     /* This instruction exists in ARB, but the d3d instruction takes two extra parameters which
2339      * must contain fixed constants. So we need a separate function to filter those constants and
2340      * can't use map2gl
2341      */
2342     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2343     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2344     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2345     char dst_name[50];
2346     char src_name0[50], src_name1[50], src_name2[50];
2347     BOOL is_color;
2348
2349     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name0);
2350     if(shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type)) {
2351         shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2352         /* No modifiers are supported on SCS */
2353         shader_addline(buffer, "SCS %s, %s;\n", dst_name, src_name0);
2354
2355         if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE)
2356         {
2357             shader_arb_get_register_name(ins, &dst->reg, src_name0, &is_color);
2358             shader_addline(buffer, "MOV_SAT %s, %s;\n", dst_name, src_name0);
2359         }
2360     } else if(priv->target_version >= NV2) {
2361         shader_arb_get_register_name(ins, &dst->reg, dst_name, &is_color);
2362
2363         /* Sincos writemask must be .x, .y or .xy */
2364         if(dst->write_mask & WINED3DSP_WRITEMASK_0)
2365             shader_addline(buffer, "COS%s %s.x, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name0);
2366         if(dst->write_mask & WINED3DSP_WRITEMASK_1)
2367             shader_addline(buffer, "SIN%s %s.y, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name0);
2368     } else {
2369         /* Approximate sine and cosine with a taylor series, as per math textbook. The application passes 8
2370          * helper constants(D3DSINCOSCONST1 and D3DSINCOSCONST2) in src1 and src2.
2371          *
2372          * sin(x) = x - x^3/3! + x^5/5! - x^7/7! + ...
2373          * cos(x) = 1 - x^2/2! + x^4/4! - x^6/6! + ...
2374          *
2375          * The constants we get are:
2376          *
2377          *  +1   +1,     -1     -1     +1      +1      -1       -1
2378          *      ---- ,  ---- , ---- , ----- , ----- , ----- , ------
2379          *      1!*2    2!*4   3!*8   4!*16   5!*32   6!*64   7!*128
2380          *
2381          * If used with x^2, x^3, x^4 etc they calculate sin(x/2) and cos(x/2):
2382          *
2383          * (x/2)^2 = x^2 / 4
2384          * (x/2)^3 = x^3 / 8
2385          * (x/2)^4 = x^4 / 16
2386          * (x/2)^5 = x^5 / 32
2387          * etc
2388          *
2389          * To get the final result:
2390          * sin(x) = 2 * sin(x/2) * cos(x/2)
2391          * cos(x) = cos(x/2)^2 - sin(x/2)^2
2392          * (from sin(x+y) and cos(x+y) rules)
2393          *
2394          * As per MSDN, dst.z is undefined after the operation, and so is
2395          * dst.x and dst.y if they're masked out by the writemask. Ie
2396          * sincos dst.y, src1, c0, c1
2397          * returns the sine in dst.y. dst.x and dst.z are undefined, dst.w is not touched. The assembler
2398          * vsa.exe also stops with an error if the dest register is the same register as the source
2399          * register. This means we can use dest.xyz as temporary storage. The assembler vsa.exe output also
2400          * indicates that sincos consumes 8 instruction slots in vs_2_0(and, strangely, in vs_3_0).
2401          */
2402         shader_arb_get_src_param(ins, &ins->src[1], 1, src_name1);
2403         shader_arb_get_src_param(ins, &ins->src[2], 2, src_name2);
2404         shader_arb_get_register_name(ins, &dst->reg, dst_name, &is_color);
2405
2406         shader_addline(buffer, "MUL %s.x, %s, %s;\n", dst_name, src_name0, src_name0);  /* x ^ 2 */
2407         shader_addline(buffer, "MUL TA.y, %s.x, %s;\n", dst_name, src_name0);           /* x ^ 3 */
2408         shader_addline(buffer, "MUL %s.y, TA.y, %s;\n", dst_name, src_name0);           /* x ^ 4 */
2409         shader_addline(buffer, "MUL TA.z, %s.y, %s;\n", dst_name, src_name0);           /* x ^ 5 */
2410         shader_addline(buffer, "MUL %s.z, TA.z, %s;\n", dst_name, src_name0);           /* x ^ 6 */
2411         shader_addline(buffer, "MUL TA.w, %s.z, %s;\n", dst_name, src_name0);           /* x ^ 7 */
2412
2413         /* sin(x/2)
2414          *
2415          * Unfortunately we don't get the constants in a DP4-capable form. Is there a way to
2416          * properly merge that with MULs in the code above?
2417          * The swizzles .yz and xw however fit into the .yzxw swizzle added to ps_2_0. Maybe
2418          * we can merge the sine and cosine MAD rows to calculate them together.
2419          */
2420         shader_addline(buffer, "MUL TA.x, %s, %s.w;\n", src_name0, src_name2); /* x^1, +1/(1!*2) */
2421         shader_addline(buffer, "MAD TA.x, TA.y, %s.x, TA.x;\n", src_name2); /* -1/(3!*8) */
2422         shader_addline(buffer, "MAD TA.x, TA.z, %s.w, TA.x;\n", src_name1); /* +1/(5!*32) */
2423         shader_addline(buffer, "MAD TA.x, TA.w, %s.x, TA.x;\n", src_name1); /* -1/(7!*128) */
2424
2425         /* cos(x/2) */
2426         shader_addline(buffer, "MAD TA.y, %s.x, %s.y, %s.z;\n", dst_name, src_name2, src_name2); /* -1/(2!*4), +1.0 */
2427         shader_addline(buffer, "MAD TA.y, %s.y, %s.z, TA.y;\n", dst_name, src_name1); /* +1/(4!*16) */
2428         shader_addline(buffer, "MAD TA.y, %s.z, %s.y, TA.y;\n", dst_name, src_name1); /* -1/(6!*64) */
2429
2430         if(dst->write_mask & WINED3DSP_WRITEMASK_0) {
2431             /* cos x */
2432             shader_addline(buffer, "MUL TA.z, TA.y, TA.y;\n");
2433             shader_addline(buffer, "MAD %s.x, -TA.x, TA.x, TA.z;\n", dst_name);
2434         }
2435         if(dst->write_mask & WINED3DSP_WRITEMASK_1) {
2436             /* sin x */
2437             shader_addline(buffer, "MUL %s.y, TA.x, TA.y;\n", dst_name);
2438             shader_addline(buffer, "ADD %s.y, %s.y, %s.y;\n", dst_name, dst_name, dst_name);
2439         }
2440     }
2441 }
2442
2443 static void shader_hw_sgn(const struct wined3d_shader_instruction *ins)
2444 {
2445     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2446     char dst_name[50];
2447     char src_name[50];
2448     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
2449
2450     /* SGN is only valid in vertex shaders */
2451     if(ctx->target_version == NV2) {
2452         shader_hw_map2gl(ins);
2453         return;
2454     }
2455     shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2456     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name);
2457
2458     /* If SRC > 0.0, -SRC < SRC = TRUE, otherwise false.
2459      * if SRC < 0.0,  SRC < -SRC = TRUE. If neither is true, src = 0.0
2460      */
2461     if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE) {
2462         shader_addline(buffer, "SLT %s, -%s, %s;\n", dst_name, src_name, src_name);
2463     } else {
2464         /* src contains TA? Write to the dest first. This won't overwrite our destination.
2465          * Then use TA, and calculate the final result
2466          *
2467          * Not reading from TA? Store the first result in TA to avoid overwriting the
2468          * destination if src reg = dst reg
2469          */
2470         if(strstr(src_name, "TA"))
2471         {
2472             shader_addline(buffer, "SLT %s,  %s, -%s;\n", dst_name, src_name, src_name);
2473             shader_addline(buffer, "SLT TA, -%s, %s;\n", src_name, src_name);
2474             shader_addline(buffer, "ADD %s, %s, -TA;\n", dst_name, dst_name);
2475         }
2476         else
2477         {
2478             shader_addline(buffer, "SLT TA, -%s, %s;\n", src_name, src_name);
2479             shader_addline(buffer, "SLT %s,  %s, -%s;\n", dst_name, src_name, src_name);
2480             shader_addline(buffer, "ADD %s, TA, -%s;\n", dst_name, dst_name);
2481         }
2482     }
2483 }
2484
2485 static void shader_hw_dsy(const struct wined3d_shader_instruction *ins)
2486 {
2487     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2488     char src[50];
2489     char dst[50];
2490     char dst_name[50];
2491     BOOL is_color;
2492
2493     shader_arb_get_dst_param(ins, &ins->dst[0], dst);
2494     shader_arb_get_src_param(ins, &ins->src[0], 0, src);
2495     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2496
2497     shader_addline(buffer, "DDY %s, %s;\n", dst, src);
2498     shader_addline(buffer, "MUL%s %s, %s, ycorrection.y;\n", shader_arb_get_modifier(ins), dst, dst_name);
2499 }
2500
2501 static DWORD abs_modifier(DWORD mod, BOOL *need_abs)
2502 {
2503     *need_abs = FALSE;
2504
2505     switch(mod)
2506     {
2507         case WINED3DSPSM_NONE:      return WINED3DSPSM_ABS;
2508         case WINED3DSPSM_NEG:       return WINED3DSPSM_ABS;
2509         case WINED3DSPSM_BIAS:      *need_abs = TRUE; return WINED3DSPSM_BIAS;
2510         case WINED3DSPSM_BIASNEG:   *need_abs = TRUE; return WINED3DSPSM_BIASNEG;
2511         case WINED3DSPSM_SIGN:      *need_abs = TRUE; return WINED3DSPSM_SIGN;
2512         case WINED3DSPSM_SIGNNEG:   *need_abs = TRUE; return WINED3DSPSM_SIGNNEG;
2513         case WINED3DSPSM_COMP:      *need_abs = TRUE; return WINED3DSPSM_COMP;
2514         case WINED3DSPSM_X2:        *need_abs = TRUE; return WINED3DSPSM_X2;
2515         case WINED3DSPSM_X2NEG:     *need_abs = TRUE; return WINED3DSPSM_X2NEG;
2516         case WINED3DSPSM_DZ:        *need_abs = TRUE; return WINED3DSPSM_DZ;
2517         case WINED3DSPSM_DW:        *need_abs = TRUE; return WINED3DSPSM_DW;
2518         case WINED3DSPSM_ABS:       return WINED3DSPSM_ABS;
2519         case WINED3DSPSM_ABSNEG:    return WINED3DSPSM_ABS;
2520     }
2521     FIXME("Unknown modifier %u\n", mod);
2522     return mod;
2523 }
2524
2525 static void shader_hw_log_pow(const struct wined3d_shader_instruction *ins)
2526 {
2527     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2528     char src0[50], src1[50], dst[50];
2529     struct wined3d_shader_src_param src0_copy = ins->src[0];
2530     BOOL need_abs = FALSE;
2531     const char *instr;
2532     BOOL arg2 = FALSE;
2533
2534     switch(ins->handler_idx)
2535     {
2536         case WINED3DSIH_LOG:  instr = "LG2"; break;
2537         case WINED3DSIH_LOGP: instr = "LOG"; break;
2538         case WINED3DSIH_POW:  instr = "POW"; arg2 = TRUE; break;
2539         default:
2540             ERR("Unexpected instruction %d\n", ins->handler_idx);
2541             return;
2542     }
2543
2544     /* LOG, LOGP and POW operate on the absolute value of the input */
2545     src0_copy.modifiers = abs_modifier(src0_copy.modifiers, &need_abs);
2546
2547     shader_arb_get_dst_param(ins, &ins->dst[0], dst);
2548     shader_arb_get_src_param(ins, &src0_copy, 0, src0);
2549     if(arg2) shader_arb_get_src_param(ins, &ins->src[1], 1, src1);
2550
2551     if(need_abs)
2552     {
2553         shader_addline(buffer, "ABS TA, %s;\n", src0);
2554         if(arg2)
2555         {
2556             shader_addline(buffer, "%s%s %s, TA, %s;\n", instr, shader_arb_get_modifier(ins), dst, src1);
2557         }
2558         else
2559         {
2560             shader_addline(buffer, "%s%s %s, TA;\n", instr, shader_arb_get_modifier(ins), dst);
2561         }
2562     }
2563     else if(arg2)
2564     {
2565         shader_addline(buffer, "%s%s %s, %s, %s;\n", instr, shader_arb_get_modifier(ins), dst, src0, src1);
2566     }
2567     else
2568     {
2569         shader_addline(buffer, "%s%s %s, %s;\n", instr, shader_arb_get_modifier(ins), dst, src0);
2570     }
2571 }
2572
2573 static void shader_hw_loop(const struct wined3d_shader_instruction *ins)
2574 {
2575     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2576     char src_name[50];
2577     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2578
2579     /* src0 is aL */
2580     shader_arb_get_src_param(ins, &ins->src[1], 0, src_name);
2581
2582     if(vshader)
2583     {
2584         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2585         struct list *e = list_head(&priv->control_frames);
2586         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2587
2588         if(priv->loop_depth > 1) shader_addline(buffer, "PUSHA aL;\n");
2589         /* The constant loader makes sure to load -1 into iX.w */
2590         shader_addline(buffer, "ARLC aL, %s.xywz;\n", src_name);
2591         shader_addline(buffer, "BRA loop_%u_end (LE.x);\n", control_frame->loop_no);
2592         shader_addline(buffer, "loop_%u_start:\n", control_frame->loop_no);
2593     }
2594     else
2595     {
2596         shader_addline(buffer, "LOOP %s;\n", src_name);
2597     }
2598 }
2599
2600 static void shader_hw_rep(const struct wined3d_shader_instruction *ins)
2601 {
2602     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2603     char src_name[50];
2604     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2605
2606     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name);
2607
2608     /* The constant loader makes sure to load -1 into iX.w */
2609     if(vshader)
2610     {
2611         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2612         struct list *e = list_head(&priv->control_frames);
2613         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2614
2615         if(priv->loop_depth > 1) shader_addline(buffer, "PUSHA aL;\n");
2616
2617         shader_addline(buffer, "ARLC aL, %s.xywz;\n", src_name);
2618         shader_addline(buffer, "BRA loop_%u_end (LE.x);\n", control_frame->loop_no);
2619         shader_addline(buffer, "loop_%u_start:\n", control_frame->loop_no);
2620     }
2621     else
2622     {
2623         shader_addline(buffer, "REP %s;\n", src_name);
2624     }
2625 }
2626
2627 static void shader_hw_endloop(const struct wined3d_shader_instruction *ins)
2628 {
2629     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2630     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2631
2632     if(vshader)
2633     {
2634         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2635         struct list *e = list_head(&priv->control_frames);
2636         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2637
2638         shader_addline(buffer, "ARAC aL.xy, aL;\n");
2639         shader_addline(buffer, "BRA loop_%u_start (GT.x);\n", control_frame->loop_no);
2640         shader_addline(buffer, "loop_%u_end:\n", control_frame->loop_no);
2641
2642         if(priv->loop_depth > 1) shader_addline(buffer, "POPA aL;\n");
2643     }
2644     else
2645     {
2646         shader_addline(buffer, "ENDLOOP;\n");
2647     }
2648 }
2649
2650 static void shader_hw_endrep(const struct wined3d_shader_instruction *ins)
2651 {
2652     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2653     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2654
2655     if(vshader)
2656     {
2657         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2658         struct list *e = list_head(&priv->control_frames);
2659         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2660
2661         shader_addline(buffer, "ARAC aL.xy, aL;\n");
2662         shader_addline(buffer, "BRA loop_%u_start (GT.x);\n", control_frame->loop_no);
2663         shader_addline(buffer, "loop_%u_end:\n", control_frame->loop_no);
2664
2665         if(priv->loop_depth > 1) shader_addline(buffer, "POPA aL;\n");
2666     }
2667     else
2668     {
2669         shader_addline(buffer, "ENDREP;\n");
2670     }
2671 }
2672
2673 static const struct control_frame *find_last_loop(const struct shader_arb_ctx_priv *priv)
2674 {
2675     struct control_frame *control_frame;
2676
2677     LIST_FOR_EACH_ENTRY(control_frame, &priv->control_frames, struct control_frame, entry)
2678     {
2679         if(control_frame->type == LOOP || control_frame->type == REP) return control_frame;
2680     }
2681     ERR("Could not find loop for break\n");
2682     return NULL;
2683 }
2684
2685 static void shader_hw_break(const struct wined3d_shader_instruction *ins)
2686 {
2687     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2688     const struct control_frame *control_frame = find_last_loop(ins->ctx->backend_data);
2689     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2690
2691     if(vshader)
2692     {
2693         shader_addline(buffer, "BRA loop_%u_end;\n", control_frame->loop_no);
2694     }
2695     else
2696     {
2697         shader_addline(buffer, "BRK;\n");
2698     }
2699 }
2700
2701 static const char *get_compare(COMPARISON_TYPE flags)
2702 {
2703     switch (flags)
2704     {
2705         case COMPARISON_GT: return "GT";
2706         case COMPARISON_EQ: return "EQ";
2707         case COMPARISON_GE: return "GE";
2708         case COMPARISON_LT: return "LT";
2709         case COMPARISON_NE: return "NE";
2710         case COMPARISON_LE: return "LE";
2711         default:
2712             FIXME("Unrecognized comparison value: %u\n", flags);
2713             return "(\?\?)";
2714     }
2715 }
2716
2717 static COMPARISON_TYPE invert_compare(COMPARISON_TYPE flags)
2718 {
2719     switch (flags)
2720     {
2721         case COMPARISON_GT: return COMPARISON_LE;
2722         case COMPARISON_EQ: return COMPARISON_NE;
2723         case COMPARISON_GE: return COMPARISON_LT;
2724         case COMPARISON_LT: return COMPARISON_GE;
2725         case COMPARISON_NE: return COMPARISON_EQ;
2726         case COMPARISON_LE: return COMPARISON_GT;
2727         default:
2728             FIXME("Unrecognized comparison value: %u\n", flags);
2729             return -1;
2730     }
2731 }
2732
2733 static void shader_hw_breakc(const struct wined3d_shader_instruction *ins)
2734 {
2735     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2736     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2737     const struct control_frame *control_frame = find_last_loop(ins->ctx->backend_data);
2738     char src_name0[50];
2739     char src_name1[50];
2740     const char *comp = get_compare(ins->flags);
2741
2742     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name0);
2743     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name1);
2744
2745     if(vshader)
2746     {
2747         /* SUBC CC, src0, src1" works only in pixel shaders, so use TA to throw
2748          * away the subtraction result
2749          */
2750         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
2751         shader_addline(buffer, "BRA loop_%u_end (%s.x);\n", control_frame->loop_no, comp);
2752     }
2753     else
2754     {
2755         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
2756         shader_addline(buffer, "BRK (%s.x);\n", comp);
2757     }
2758 }
2759
2760 static void shader_hw_ifc(const struct wined3d_shader_instruction *ins)
2761 {
2762     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2763     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2764     struct list *e = list_head(&priv->control_frames);
2765     struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2766     const char *comp;
2767     char src_name0[50];
2768     char src_name1[50];
2769     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2770
2771     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name0);
2772     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name1);
2773
2774     if(vshader)
2775     {
2776         /* Invert the flag. We jump to the else label if the condition is NOT true */
2777         comp = get_compare(invert_compare(ins->flags));
2778         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
2779         shader_addline(buffer, "BRA ifc_%u_else (%s.x);\n", control_frame->ifc_no, comp);
2780     }
2781     else
2782     {
2783         comp = get_compare(ins->flags);
2784         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
2785         shader_addline(buffer, "IF %s.x;\n", comp);
2786     }
2787 }
2788
2789 static void shader_hw_else(const struct wined3d_shader_instruction *ins)
2790 {
2791     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2792     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2793     struct list *e = list_head(&priv->control_frames);
2794     struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2795     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2796
2797     if(vshader)
2798     {
2799         shader_addline(buffer, "BRA ifc_%u_endif;\n", control_frame->ifc_no);
2800         shader_addline(buffer, "ifc_%u_else:\n", control_frame->ifc_no);
2801         control_frame->had_else = TRUE;
2802     }
2803     else
2804     {
2805         shader_addline(buffer, "ELSE;\n");
2806     }
2807 }
2808
2809 static void shader_hw_endif(const struct wined3d_shader_instruction *ins)
2810 {
2811     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2812     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2813     struct list *e = list_head(&priv->control_frames);
2814     struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2815     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2816
2817     if(vshader)
2818     {
2819         if(control_frame->had_else)
2820         {
2821             shader_addline(buffer, "ifc_%u_endif:\n", control_frame->ifc_no);
2822         }
2823         else
2824         {
2825             shader_addline(buffer, "#No else branch. else is endif\n");
2826             shader_addline(buffer, "ifc_%u_else:\n", control_frame->ifc_no);
2827         }
2828     }
2829     else
2830     {
2831         shader_addline(buffer, "ENDIF;\n");
2832     }
2833 }
2834
2835 static void shader_hw_texldd(const struct wined3d_shader_instruction *ins)
2836 {
2837     DWORD sampler_idx = ins->src[1].reg.idx;
2838     char reg_dest[40];
2839     char reg_src[3][40];
2840     DWORD flags = TEX_DERIV;
2841
2842     shader_arb_get_dst_param(ins, &ins->dst[0], reg_dest);
2843     shader_arb_get_src_param(ins, &ins->src[0], 0, reg_src[0]);
2844     shader_arb_get_src_param(ins, &ins->src[2], 1, reg_src[1]);
2845     shader_arb_get_src_param(ins, &ins->src[3], 2, reg_src[2]);
2846
2847     if (ins->flags & WINED3DSI_TEXLD_PROJECT) flags |= TEX_PROJ;
2848     if (ins->flags & WINED3DSI_TEXLD_BIAS) flags |= TEX_BIAS;
2849
2850     shader_hw_sample(ins, sampler_idx, reg_dest, reg_src[0], flags, reg_src[1], reg_src[2]);
2851 }
2852
2853 static void shader_hw_texldl(const struct wined3d_shader_instruction *ins)
2854 {
2855     DWORD sampler_idx = ins->src[1].reg.idx;
2856     char reg_dest[40];
2857     char reg_coord[40];
2858     DWORD flags = TEX_LOD;
2859
2860     shader_arb_get_dst_param(ins, &ins->dst[0], reg_dest);
2861     shader_arb_get_src_param(ins, &ins->src[0], 0, reg_coord);
2862
2863     if (ins->flags & WINED3DSI_TEXLD_PROJECT) flags |= TEX_PROJ;
2864     if (ins->flags & WINED3DSI_TEXLD_BIAS) flags |= TEX_BIAS;
2865
2866     shader_hw_sample(ins, sampler_idx, reg_dest, reg_coord, flags, NULL, NULL);
2867 }
2868
2869 static void shader_hw_label(const struct wined3d_shader_instruction *ins)
2870 {
2871     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2872     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2873
2874     priv->in_main_func = FALSE;
2875     /* Call instructions activate the NV extensions, not labels and rets. If there is an uncalled
2876      * subroutine, don't generate a label that will make GL complain
2877      */
2878     if(priv->target_version == ARB) return;
2879
2880     shader_addline(buffer, "l%u:\n", ins->src[0].reg.idx);
2881 }
2882
2883 static void vshader_add_footer(IWineD3DVertexShaderImpl *This, struct wined3d_shader_buffer *buffer,
2884         const struct arb_vs_compile_args *args, struct shader_arb_ctx_priv *priv_ctx)
2885 {
2886     const shader_reg_maps *reg_maps = &This->baseShader.reg_maps;
2887     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *)This->baseShader.device;
2888     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
2889     unsigned int i;
2890
2891     /* The D3DRS_FOGTABLEMODE render state defines if the shader-generated fog coord is used
2892      * or if the fragment depth is used. If the fragment depth is used(FOGTABLEMODE != NONE),
2893      * the fog frag coord is thrown away. If the fog frag coord is used, but not written by
2894      * the shader, it is set to 0.0(fully fogged, since start = 1.0, end = 0.0)
2895      */
2896     if(args->super.fog_src == VS_FOG_Z) {
2897         shader_addline(buffer, "MOV result.fogcoord, TMP_OUT.z;\n");
2898     } else if (!reg_maps->fog) {
2899         /* posFixup.x is always 1.0, so we can savely use it */
2900         shader_addline(buffer, "ADD result.fogcoord, posFixup.x, -posFixup.x;\n");
2901     }
2902
2903     /* Write the final position.
2904      *
2905      * OpenGL coordinates specify the center of the pixel while d3d coords specify
2906      * the corner. The offsets are stored in z and w in posFixup. posFixup.y contains
2907      * 1.0 or -1.0 to turn the rendering upside down for offscreen rendering. PosFixup.x
2908      * contains 1.0 to allow a mad, but arb vs swizzles are too restricted for that.
2909      */
2910     shader_addline(buffer, "MUL TA, posFixup, TMP_OUT.w;\n");
2911     shader_addline(buffer, "ADD TMP_OUT.x, TMP_OUT.x, TA.z;\n");
2912     shader_addline(buffer, "MAD TMP_OUT.y, TMP_OUT.y, posFixup.y, TA.w;\n");
2913
2914     if(use_nv_clip(gl_info) && priv_ctx->target_version >= NV2)
2915     {
2916         for(i = 0; i < priv_ctx->vs_clipplanes; i++)
2917         {
2918             shader_addline(buffer, "DP4 result.clip[%u].x, TMP_OUT, state.clip[%u].plane;\n", i, i);
2919         }
2920     }
2921     else if(args->boolclip.clip_control[0])
2922     {
2923         unsigned int cur_clip = 0;
2924         char component[4] = {'x', 'y', 'z', 'w'};
2925
2926         for(i = 0; i < GL_LIMITS(clipplanes); i++)
2927         {
2928             if(args->boolclip.clip_control[1] & (1 << i))
2929             {
2930                 shader_addline(buffer, "DP4 TA.%c, TMP_OUT, state.clip[%u].plane;\n",
2931                                component[cur_clip++], i);
2932             }
2933         }
2934         switch(cur_clip)
2935         {
2936             case 0:
2937                 shader_addline(buffer, "MOV TA, -helper_const.w;\n");
2938                 break;
2939             case 1:
2940                 shader_addline(buffer, "MOV TA.yzw, -helper_const.w;\n");
2941                 break;
2942             case 2:
2943                 shader_addline(buffer, "MOV TA.zw, -helper_const.w;\n");
2944                 break;
2945             case 3:
2946                 shader_addline(buffer, "MOV TA.w, -helper_const.w;\n");
2947                 break;
2948         }
2949         shader_addline(buffer, "MOV result.texcoord[%u], TA;\n",
2950                        args->boolclip.clip_control[0] - 1);
2951     }
2952
2953     /* Z coord [0;1]->[-1;1] mapping, see comment in transform_projection in state.c
2954      * and the glsl equivalent
2955      */
2956     if(need_helper_const(gl_info)) {
2957         shader_addline(buffer, "MAD TMP_OUT.z, TMP_OUT.z, helper_const.x, -TMP_OUT.w;\n");
2958     } else {
2959         shader_addline(buffer, "ADD TMP_OUT.z, TMP_OUT.z, TMP_OUT.z;\n");
2960         shader_addline(buffer, "ADD TMP_OUT.z, TMP_OUT.z, -TMP_OUT.w;\n");
2961     }
2962
2963     shader_addline(buffer, "MOV result.position, TMP_OUT;\n");
2964
2965     priv_ctx->footer_written = TRUE;
2966 }
2967
2968 static void shader_hw_ret(const struct wined3d_shader_instruction *ins)
2969 {
2970     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2971     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2972     IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *) ins->ctx->shader;
2973     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2974
2975     if(priv->target_version == ARB) return;
2976
2977     if(vshader)
2978     {
2979         if(priv->in_main_func) vshader_add_footer((IWineD3DVertexShaderImpl *) shader, buffer, priv->cur_vs_args, priv);
2980     }
2981
2982     shader_addline(buffer, "RET;\n");
2983 }
2984
2985 static void shader_hw_call(const struct wined3d_shader_instruction *ins)
2986 {
2987     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2988     shader_addline(buffer, "CAL l%u;\n", ins->src[0].reg.idx);
2989 }
2990
2991 /* GL locking is done by the caller */
2992 static GLuint create_arb_blt_vertex_program(const struct wined3d_gl_info *gl_info)
2993 {
2994     GLuint program_id = 0;
2995     const char *blt_vprogram =
2996         "!!ARBvp1.0\n"
2997         "PARAM c[1] = { { 1, 0.5 } };\n"
2998         "MOV result.position, vertex.position;\n"
2999         "MOV result.color, c[0].x;\n"
3000         "MOV result.texcoord[0], vertex.texcoord[0];\n"
3001         "END\n";
3002
3003     GL_EXTCALL(glGenProgramsARB(1, &program_id));
3004     GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, program_id));
3005     GL_EXTCALL(glProgramStringARB(GL_VERTEX_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB, strlen(blt_vprogram), blt_vprogram));
3006
3007     if (glGetError() == GL_INVALID_OPERATION) {
3008         GLint pos;
3009         glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
3010         FIXME("Vertex program error at position %d: %s\n", pos,
3011             debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
3012     }
3013
3014     return program_id;
3015 }
3016
3017 /* GL locking is done by the caller */
3018 static GLuint create_arb_blt_fragment_program(const struct wined3d_gl_info *gl_info, enum tex_types tex_type)
3019 {
3020     GLuint program_id = 0;
3021     static const char * const blt_fprograms[tex_type_count] =
3022     {
3023         /* tex_1d */
3024         NULL,
3025         /* tex_2d */
3026         "!!ARBfp1.0\n"
3027         "TEMP R0;\n"
3028         "TEX R0.x, fragment.texcoord[0], texture[0], 2D;\n"
3029         "MOV result.depth.z, R0.x;\n"
3030         "END\n",
3031         /* tex_3d */
3032         NULL,
3033         /* tex_cube */
3034         "!!ARBfp1.0\n"
3035         "TEMP R0;\n"
3036         "TEX R0.x, fragment.texcoord[0], texture[0], CUBE;\n"
3037         "MOV result.depth.z, R0.x;\n"
3038         "END\n",
3039         /* tex_rect */
3040         "!!ARBfp1.0\n"
3041         "TEMP R0;\n"
3042         "TEX R0.x, fragment.texcoord[0], texture[0], RECT;\n"
3043         "MOV result.depth.z, R0.x;\n"
3044         "END\n",
3045     };
3046
3047     if (!blt_fprograms[tex_type])
3048     {
3049         FIXME("tex_type %#x not supported\n", tex_type);
3050         tex_type = tex_2d;
3051     }
3052
3053     GL_EXTCALL(glGenProgramsARB(1, &program_id));
3054     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, program_id));
3055     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB, strlen(blt_fprograms[tex_type]), blt_fprograms[tex_type]));
3056
3057     if (glGetError() == GL_INVALID_OPERATION) {
3058         GLint pos;
3059         glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
3060         FIXME("Fragment program error at position %d: %s\n", pos,
3061             debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
3062     }
3063
3064     return program_id;
3065 }
3066
3067 static void arbfp_add_sRGB_correction(struct wined3d_shader_buffer *buffer, const char *fragcolor,
3068         const char *tmp1, const char *tmp2, const char *tmp3, const char *tmp4, BOOL condcode)
3069 {
3070     /* Perform sRGB write correction. See GLX_EXT_framebuffer_sRGB */
3071
3072     if(condcode)
3073     {
3074         /* Sigh. MOVC CC doesn't work, so use one of the temps as dummy dest */
3075         shader_addline(buffer, "SUBC %s, %s.x, srgb_consts1.y;\n", tmp1, fragcolor);
3076         /* Calculate the > 0.0031308 case */
3077         shader_addline(buffer, "POW %s.x (GE), %s.x, srgb_consts1.z;\n", fragcolor, fragcolor);
3078         shader_addline(buffer, "POW %s.y (GE), %s.y, srgb_consts1.z;\n", fragcolor, fragcolor);
3079         shader_addline(buffer, "POW %s.z (GE), %s.z, srgb_consts1.z;\n", fragcolor, fragcolor);
3080         shader_addline(buffer, "MUL %s.xyz (GE), %s, srgb_consts1.w;\n", fragcolor, fragcolor);
3081         shader_addline(buffer, "SUB %s.xyz (GE), %s, srgb_consts2.x;\n", fragcolor, fragcolor);
3082         /* Calculate the < case */
3083         shader_addline(buffer, "MUL %s.xyz (LT), srgb_consts1.x, %s;\n", fragcolor, fragcolor);
3084     }
3085     else
3086     {
3087         /* Calculate the > 0.0031308 case */
3088         shader_addline(buffer, "POW %s.x, %s.x, srgb_consts1.z;\n", tmp1, fragcolor);
3089         shader_addline(buffer, "POW %s.y, %s.y, srgb_consts1.z;\n", tmp1, fragcolor);
3090         shader_addline(buffer, "POW %s.z, %s.z, srgb_consts1.z;\n", tmp1, fragcolor);
3091         shader_addline(buffer, "MUL %s, %s, srgb_consts1.w;\n", tmp1, tmp1);
3092         shader_addline(buffer, "SUB %s, %s, srgb_consts2.x;\n", tmp1, tmp1);
3093         /* Calculate the < case */
3094         shader_addline(buffer, "MUL %s, srgb_consts1.x, %s;\n", tmp2, fragcolor);
3095         /* Get 1.0 / 0.0 masks for > 0.0031308 and < 0.0031308 */
3096         shader_addline(buffer, "SLT %s, srgb_consts1.y, %s;\n", tmp3, fragcolor);
3097         shader_addline(buffer, "SGE %s, srgb_consts1.y, %s;\n", tmp4, fragcolor);
3098         /* Store the components > 0.0031308 in the destination */
3099         shader_addline(buffer, "MUL %s.xyz, %s, %s;\n", fragcolor, tmp1, tmp3);
3100         /* Add the components that are < 0.0031308 */
3101         shader_addline(buffer, "MAD %s.xyz, %s, %s, %s;\n", fragcolor, tmp2, tmp4, fragcolor);
3102         /* Move everything into result.color at once. Nvidia hardware cannot handle partial
3103         * result.color writes(.rgb first, then .a), or handle overwriting already written
3104         * components. The assembler uses a temporary register in this case, which is usually
3105         * not allocated from one of our registers that were used earlier.
3106         */
3107     }
3108     shader_addline(buffer, "MOV result.color, %s;\n", fragcolor);
3109     /* [0.0;1.0] clamping. Not needed, this is done implicitly */
3110 }
3111
3112 static const DWORD *find_loop_control_values(IWineD3DBaseShaderImpl *This, DWORD idx)
3113 {
3114     const local_constant *constant;
3115
3116     LIST_FOR_EACH_ENTRY(constant, &This->baseShader.constantsI, local_constant, entry)
3117     {
3118         if (constant->idx == idx)
3119         {
3120             return constant->value;
3121         }
3122     }
3123     return NULL;
3124 }
3125
3126 static void init_ps_input(const IWineD3DPixelShaderImpl *This, const struct arb_ps_compile_args *args,
3127                           struct shader_arb_ctx_priv *priv)
3128 {
3129     const char *texcoords[8] =
3130     {
3131         "fragment.texcoord[0]", "fragment.texcoord[1]", "fragment.texcoord[2]", "fragment.texcoord[3]",
3132         "fragment.texcoord[4]", "fragment.texcoord[5]", "fragment.texcoord[6]", "fragment.texcoord[7]"
3133     };
3134     unsigned int i;
3135     const struct wined3d_shader_signature_element *sig = This->input_signature;
3136     const char *semantic_name;
3137     DWORD semantic_idx;
3138
3139     switch(args->super.vp_mode)
3140     {
3141         case pretransformed:
3142         case fixedfunction:
3143             /* The pixelshader has to collect the varyings on its own. In any case properly load
3144              * color0 and color1. In the case of pretransformed vertices also load texcoords. Set
3145              * other attribs to 0.0.
3146              *
3147              * For fixedfunction this behavior is correct, according to the tests. For pretransformed
3148              * we'd either need a replacement shader that can load other attribs like BINORMAL, or
3149              * load the texcoord attrib pointers to match the pixel shader signature
3150              */
3151             for(i = 0; i < MAX_REG_INPUT; i++)
3152             {
3153                 semantic_name = sig[i].semantic_name;
3154                 semantic_idx = sig[i].semantic_idx;
3155                 if(semantic_name == NULL) continue;
3156
3157                 if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_COLOR))
3158                 {
3159                     if(semantic_idx == 0) priv->ps_input[i] = "fragment.color.primary";
3160                     else if(semantic_idx == 1) priv->ps_input[i] = "fragment.color.secondary";
3161                     else priv->ps_input[i] = "0.0";
3162                 }
3163                 else if(args->super.vp_mode == fixedfunction)
3164                 {
3165                     priv->ps_input[i] = "0.0";
3166                 }
3167                 else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_TEXCOORD))
3168                 {
3169                     if(semantic_idx < 8) priv->ps_input[i] = texcoords[semantic_idx];
3170                     else priv->ps_input[i] = "0.0";
3171                 }
3172                 else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_FOG))
3173                 {
3174                     if(semantic_idx == 0) priv->ps_input[i] = "fragment.fogcoord";
3175                     else priv->ps_input[i] = "0.0";
3176                 }
3177                 else
3178                 {
3179                     priv->ps_input[i] = "0.0";
3180                 }
3181
3182                 TRACE("v%u, semantic %s%u is %s\n", i, semantic_name, semantic_idx, priv->ps_input[i]);
3183             }
3184             break;
3185
3186         case vertexshader:
3187             /* That one is easy. The vertex shaders provide v0-v7 in fragment.texcoord and v8 and v9 in
3188              * fragment.color
3189              */
3190             for(i = 0; i < 8; i++)
3191             {
3192                 priv->ps_input[i] = texcoords[i];
3193             }
3194             priv->ps_input[8] = "fragment.color.primary";
3195             priv->ps_input[9] = "fragment.color.secondary";
3196             break;
3197     }
3198 }
3199
3200 /* GL locking is done by the caller */
3201 static GLuint shader_arb_generate_pshader(IWineD3DPixelShaderImpl *This, struct wined3d_shader_buffer *buffer,
3202         const struct arb_ps_compile_args *args, struct arb_ps_compiled_shader *compiled)
3203 {
3204     const shader_reg_maps* reg_maps = &This->baseShader.reg_maps;
3205     CONST DWORD *function = This->baseShader.function;
3206     const struct wined3d_gl_info *gl_info = &((IWineD3DDeviceImpl *)This->baseShader.device)->adapter->gl_info;
3207     const local_constant *lconst;
3208     GLuint retval;
3209     char fragcolor[16];
3210     DWORD *lconst_map = local_const_mapping((IWineD3DBaseShaderImpl *) This), next_local, cur;
3211     struct shader_arb_ctx_priv priv_ctx;
3212     BOOL dcl_tmp = args->super.srgb_correction, dcl_td = FALSE;
3213     BOOL want_nv_prog = FALSE;
3214     struct arb_pshader_private *shader_priv = This->backend_priv;
3215
3216     char srgbtmp[4][4];
3217     unsigned int i, found = 0;
3218
3219     for(i = 0; i < This->baseShader.limits.temporary; i++) {
3220
3221         /* Don't overwrite the color source */
3222         if(This->color0_mov && i == This->color0_reg) continue;
3223         else if(reg_maps->shader_version.major < 2 && i == 0) continue;
3224
3225         if(reg_maps->temporary[i]) {
3226             sprintf(srgbtmp[found], "R%u", i);
3227             found++;
3228             if(found == 4) break;
3229         }
3230     }
3231
3232     switch(found) {
3233         case 4: dcl_tmp = FALSE; break;
3234         case 0:
3235             sprintf(srgbtmp[0], "TA");
3236             sprintf(srgbtmp[1], "TB");
3237             sprintf(srgbtmp[2], "TC");
3238             sprintf(srgbtmp[3], "TD");
3239             dcl_td = TRUE;
3240             break;
3241         case 1:
3242             sprintf(srgbtmp[1], "TA");
3243             sprintf(srgbtmp[2], "TB");
3244             sprintf(srgbtmp[3], "TC");
3245             break;
3246         case 2:
3247             sprintf(srgbtmp[2], "TA");
3248             sprintf(srgbtmp[3], "TB");
3249             break;
3250         case 3:
3251             sprintf(srgbtmp[3], "TA");
3252             break;
3253     }
3254
3255     /*  Create the hw ARB shader */
3256     memset(&priv_ctx, 0, sizeof(priv_ctx));
3257     priv_ctx.cur_ps_args = args;
3258     priv_ctx.compiled_fprog = compiled;
3259     priv_ctx.cur_np2fixup_info = &compiled->np2fixup_info;
3260     init_ps_input(This, args, &priv_ctx);
3261     list_init(&priv_ctx.control_frames);
3262
3263     /* Avoid enabling NV_fragment_program* if we do not need it.
3264      *
3265      * Enabling GL_NV_fragment_program_option causes the driver to occupy a temporary register,
3266      * and it slows down the shader execution noticeably(about 5%). Usually our instruction emulation
3267      * is faster than what we gain from using higher native instructions. There are some things though
3268      * that cannot be emulated. In that case enable the extensions.
3269      * If the extension is enabled, instruction handlers that support both ways will use it.
3270      *
3271      * Testing shows no performance difference between OPTION NV_fragment_program2 and NV_fragment_program.
3272      * So enable the best we can get.
3273      */
3274     if(reg_maps->usesdsx || reg_maps->usesdsy || reg_maps->loop_depth > 0 || reg_maps->usestexldd ||
3275        reg_maps->usestexldl || reg_maps->usesfacing || reg_maps->usesifc || reg_maps->usescall)
3276     {
3277         want_nv_prog = TRUE;
3278     }
3279
3280     shader_addline(buffer, "!!ARBfp1.0\n");
3281     if(want_nv_prog && GL_SUPPORT(NV_FRAGMENT_PROGRAM2)) {
3282         shader_addline(buffer, "OPTION NV_fragment_program2;\n");
3283         priv_ctx.target_version = NV3;
3284     } else if(want_nv_prog && GL_SUPPORT(NV_FRAGMENT_PROGRAM_OPTION)) {
3285         shader_addline(buffer, "OPTION NV_fragment_program;\n");
3286         priv_ctx.target_version = NV2;
3287     } else {
3288         if(want_nv_prog)
3289         {
3290             /* This is an error - either we're advertising the wrong shader version, or aren't enforcing some
3291              * limits properly
3292              */
3293             ERR("The shader requires instructions that are not available in plain GL_ARB_fragment_program\n");
3294             ERR("Try GLSL\n");
3295         }
3296         priv_ctx.target_version = ARB;
3297     }
3298
3299     if(This->baseShader.reg_maps.highest_render_target > 0)
3300     {
3301         shader_addline(buffer, "OPTION ARB_draw_buffers;\n");
3302     }
3303
3304     if (reg_maps->shader_version.major < 3)
3305     {
3306         switch(args->super.fog) {
3307             case FOG_OFF:
3308                 break;
3309             case FOG_LINEAR:
3310                 shader_addline(buffer, "OPTION ARB_fog_linear;\n");
3311                 break;
3312             case FOG_EXP:
3313                 shader_addline(buffer, "OPTION ARB_fog_exp;\n");
3314                 break;
3315             case FOG_EXP2:
3316                 shader_addline(buffer, "OPTION ARB_fog_exp2;\n");
3317                 break;
3318         }
3319     }
3320
3321     /* For now always declare the temps. At least the Nvidia assembler optimizes completely
3322      * unused temps away(but occupies them for the whole shader if they're used once). Always
3323      * declaring them avoids tricky bookkeeping work
3324      */
3325     shader_addline(buffer, "TEMP TA;\n");      /* Used for modifiers */
3326     shader_addline(buffer, "TEMP TB;\n");      /* Used for modifiers */
3327     shader_addline(buffer, "TEMP TC;\n");      /* Used for modifiers */
3328     if(dcl_td) shader_addline(buffer, "TEMP TD;\n"); /* Used for sRGB writing */
3329     shader_addline(buffer, "PARAM coefdiv = { 0.5, 0.25, 0.125, 0.0625 };\n");
3330     shader_addline(buffer, "PARAM coefmul = { 2, 4, 8, 16 };\n");
3331     shader_addline(buffer, "PARAM one = { 1.0, 1.0, 1.0, 1.0 };\n");
3332
3333     if (reg_maps->shader_version.major < 2)
3334     {
3335         strcpy(fragcolor, "R0");
3336     } else {
3337         if(args->super.srgb_correction) {
3338             if(This->color0_mov) {
3339                 sprintf(fragcolor, "R%u", This->color0_reg);
3340             } else {
3341                 shader_addline(buffer, "TEMP TMP_COLOR;\n");
3342                 strcpy(fragcolor, "TMP_COLOR");
3343             }
3344         } else {
3345             strcpy(fragcolor, "result.color");
3346         }
3347     }
3348
3349     if(args->super.srgb_correction) {
3350         shader_addline(buffer, "PARAM srgb_consts1 = {%f, %f, %f, %f};\n",
3351                        srgb_mul_low, srgb_cmp, srgb_pow, srgb_mul_high);
3352         shader_addline(buffer, "PARAM srgb_consts2 = {%f, %f, %f, %f};\n",
3353                        srgb_sub_high, 0.0, 0.0, 0.0);
3354     }
3355
3356     /* Base Declarations */
3357     next_local = shader_generate_arb_declarations( (IWineD3DBaseShader*) This, reg_maps, buffer, &GLINFO_LOCATION,
3358             lconst_map, NULL, &priv_ctx);
3359
3360     for(i = 0; i < (sizeof(reg_maps->bumpmat) / sizeof(reg_maps->bumpmat[0])); i++) {
3361         if(!reg_maps->bumpmat[i]) continue;
3362
3363         cur = compiled->numbumpenvmatconsts;
3364         compiled->bumpenvmatconst[cur].const_num = WINED3D_CONST_NUM_UNUSED;
3365         compiled->bumpenvmatconst[cur].texunit = i;
3366         compiled->luminanceconst[cur].const_num = WINED3D_CONST_NUM_UNUSED;
3367         compiled->luminanceconst[cur].texunit = i;
3368
3369         /* We can fit the constants into the constant limit for sure because texbem, texbeml, bem and beml are only supported
3370          * in 1.x shaders, and GL_ARB_fragment_program has a constant limit of 24 constants. So in the worst case we're loading
3371          * 8 shader constants, 8 bump matrices and 8 luminance parameters and are perfectly fine. (No NP2 fixup on bumpmapped
3372          * textures due to conditional NP2 restrictions)
3373          *
3374          * Use local constants to load the bump env parameters, not program.env. This avoids collisions with d3d constants of
3375          * shaders in newer shader models. Since the bump env parameters have to share their space with NP2 fixup constants,
3376          * their location is shader dependent anyway and they cannot be loaded globally.
3377          */
3378         compiled->bumpenvmatconst[cur].const_num = next_local++;
3379         shader_addline(buffer, "PARAM bumpenvmat%d = program.local[%d];\n",
3380                        i, compiled->bumpenvmatconst[cur].const_num);
3381         compiled->numbumpenvmatconsts = cur + 1;
3382
3383         if(!reg_maps->luminanceparams[i]) continue;
3384
3385         compiled->luminanceconst[cur].const_num = next_local++;
3386         shader_addline(buffer, "PARAM luminance%d = program.local[%d];\n",
3387                        i, compiled->luminanceconst[cur].const_num);
3388     }
3389
3390     for(i = 0; i < MAX_CONST_I; i++)
3391     {
3392         compiled->int_consts[i] = WINED3D_CONST_NUM_UNUSED;
3393         if (reg_maps->integer_constants & (1 << i) && priv_ctx.target_version >= NV2)
3394         {
3395             const DWORD *control_values = find_loop_control_values((IWineD3DBaseShaderImpl *) This, i);
3396
3397             if(control_values)
3398             {
3399                 shader_addline(buffer, "PARAM I%u = {%u, %u, %u, -1};\n", i,
3400                                 control_values[0], control_values[1], control_values[2]);
3401             }
3402             else
3403             {
3404                 compiled->int_consts[i] = next_local;
3405                 compiled->num_int_consts++;
3406                 shader_addline(buffer, "PARAM I%u = program.local[%u];\n", i, next_local++);
3407             }
3408         }
3409     }
3410
3411     if(reg_maps->vpos || reg_maps->usesdsy)
3412     {
3413         compiled->ycorrection = next_local;
3414         shader_addline(buffer, "PARAM ycorrection = program.local[%u];\n", next_local++);
3415
3416         if(reg_maps->vpos)
3417         {
3418             shader_addline(buffer, "TEMP vpos;\n");
3419             /* ycorrection.x: Backbuffer height(onscreen) or 0(offscreen).
3420              * ycorrection.y: -1.0(onscreen), 1.0(offscreen)
3421              * ycorrection.z: 1.0
3422              * ycorrection.w: 0.0
3423              */
3424             shader_addline(buffer, "MAD vpos, fragment.position, ycorrection.zyww, ycorrection.wxww;\n");
3425             shader_addline(buffer, "FLR vpos.xy, vpos;\n");
3426         }
3427     }
3428     else
3429     {
3430         compiled->ycorrection = WINED3D_CONST_NUM_UNUSED;
3431     }
3432
3433     /* Load constants to fixup NP2 texcoords if there are still free constants left:
3434      * Constants (texture dimensions) for the NP2 fixup are loaded as local program parameters. This will consume
3435      * at most 8 (MAX_FRAGMENT_SAMPLERS / 2) parameters, which is highly unlikely, since the application had to
3436      * use 16 NP2 textures at the same time. In case that we run out of constants the fixup is simply not
3437      * applied / activated. This will probably result in wrong rendering of the texture, but will save us from
3438      * shader compilation errors and the subsequent errors when drawing with this shader. */
3439     if (priv_ctx.cur_ps_args->super.np2_fixup) {
3440
3441         struct arb_ps_np2fixup_info* const fixup = priv_ctx.cur_np2fixup_info;
3442         const WORD map = priv_ctx.cur_ps_args->super.np2_fixup;
3443         const UINT max_lconsts = gl_info->ps_arb_max_local_constants;
3444
3445         fixup->offset = next_local;
3446         fixup->super.active = 0;
3447
3448         cur = 0;
3449         for (i = 0; i < MAX_FRAGMENT_SAMPLERS; ++i) {
3450             if (!(map & (1 << i))) continue;
3451
3452             if (fixup->offset + (cur >> 1) < max_lconsts) {
3453                 fixup->super.active |= (1 << i);
3454                 fixup->super.idx[i] = cur++;
3455             } else {
3456                 FIXME("No free constant found to load NP2 fixup data into shader. "
3457                       "Sampling from this texture will probably look wrong.\n");
3458                 break;
3459             }
3460         }
3461
3462         fixup->super.num_consts = (cur + 1) >> 1;
3463         if (fixup->super.num_consts) {
3464             shader_addline(buffer, "PARAM np2fixup[%u] = { program.env[%u..%u] };\n",
3465                            fixup->super.num_consts, fixup->offset, fixup->super.num_consts + fixup->offset - 1);
3466         }
3467
3468         next_local += fixup->super.num_consts;
3469     }
3470
3471     if(shader_priv->clipplane_emulation)
3472     {
3473         shader_addline(buffer, "KIL fragment.texcoord[%u];\n", shader_priv->clipplane_emulation - 1);
3474     }
3475
3476     /* Base Shader Body */
3477     shader_generate_main((IWineD3DBaseShader *)This, buffer, reg_maps, function, &priv_ctx);
3478
3479     if(args->super.srgb_correction) {
3480         arbfp_add_sRGB_correction(buffer, fragcolor, srgbtmp[0], srgbtmp[1], srgbtmp[2], srgbtmp[3],
3481                                   priv_ctx.target_version >= NV2);
3482     } else if(reg_maps->shader_version.major < 2) {
3483         shader_addline(buffer, "MOV result.color, %s;\n", fragcolor);
3484     }
3485     shader_addline(buffer, "END\n");
3486
3487     /* TODO: change to resource.glObjectHandle or something like that */
3488     GL_EXTCALL(glGenProgramsARB(1, &retval));
3489
3490     TRACE("Creating a hw pixel shader, prg=%d\n", retval);
3491     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, retval));
3492
3493     TRACE("Created hw pixel shader, prg=%d\n", retval);
3494     /* Create the program and check for errors */
3495     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
3496                buffer->bsize, buffer->buffer));
3497
3498     if (glGetError() == GL_INVALID_OPERATION) {
3499         GLint errPos;
3500         glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &errPos);
3501         FIXME("HW PixelShader Error at position %d: %s\n",
3502               errPos, debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
3503         retval = 0;
3504     }
3505
3506     /* Load immediate constants */
3507     if(lconst_map) {
3508         LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry) {
3509             const float *value = (const float *)lconst->value;
3510             GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, lconst_map[lconst->idx], value));
3511             checkGLcall("glProgramLocalParameter4fvARB");
3512         }
3513         HeapFree(GetProcessHeap(), 0, lconst_map);
3514     }
3515
3516     return retval;
3517 }
3518
3519 static int compare_sig(const struct wined3d_shader_signature_element *sig1, const struct wined3d_shader_signature_element *sig2)
3520 {
3521     unsigned int i;
3522     int ret;
3523
3524     for(i = 0; i < MAX_REG_INPUT; i++)
3525     {
3526         if(sig1[i].semantic_name == NULL || sig2[i].semantic_name == NULL)
3527         {
3528             /* Compare pointers, not contents. One string is NULL(element does not exist), the other one is not NULL */
3529             if(sig1[i].semantic_name != sig2[i].semantic_name) return sig1[i].semantic_name < sig2[i].semantic_name ? -1 : 1;
3530             continue;
3531         }
3532
3533         ret = strcmp(sig1[i].semantic_name, sig2[i].semantic_name);
3534         if(ret != 0) return ret;
3535         if(sig1[i].semantic_idx    != sig2[i].semantic_idx)    return sig1[i].semantic_idx    < sig2[i].semantic_idx    ? -1 : 1;
3536         if(sig1[i].sysval_semantic != sig2[i].sysval_semantic) return sig1[i].sysval_semantic < sig2[i].sysval_semantic ? -1 : 1;
3537         if(sig1[i].component_type  != sig2[i].component_type)  return sig1[i].sysval_semantic < sig2[i].component_type  ? -1 : 1;
3538         if(sig1[i].register_idx    != sig2[i].register_idx)    return sig1[i].register_idx    < sig2[i].register_idx    ? -1 : 1;
3539         if(sig1[i].mask            != sig2->mask)              return sig1[i].mask            < sig2[i].mask            ? -1 : 1;
3540     }
3541     return 0;
3542 }
3543
3544 static struct wined3d_shader_signature_element *clone_sig(const struct wined3d_shader_signature_element *sig)
3545 {
3546     struct wined3d_shader_signature_element *new;
3547     int i;
3548     char *name;
3549
3550     new = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*new) * MAX_REG_INPUT);
3551     for(i = 0; i < MAX_REG_INPUT; i++)
3552     {
3553         if(sig[i].semantic_name == NULL)
3554         {
3555             continue;
3556         }
3557
3558         new[i] = sig[i];
3559         /* Clone the semantic string */
3560         name = HeapAlloc(GetProcessHeap(), 0, strlen(sig[i].semantic_name) + 1);
3561         strcpy(name, sig[i].semantic_name);
3562         new[i].semantic_name = name;
3563     }
3564     return new;
3565 }
3566
3567 static DWORD find_input_signature(struct shader_arb_priv *priv, const struct wined3d_shader_signature_element *sig)
3568 {
3569     struct wine_rb_entry *entry = wine_rb_get(&priv->signature_tree, sig);
3570     struct ps_signature *found_sig;
3571
3572     if(entry != NULL)
3573     {
3574         found_sig = WINE_RB_ENTRY_VALUE(entry, struct ps_signature, entry);
3575         TRACE("Found existing signature %u\n", found_sig->idx);
3576         return found_sig->idx;
3577     }
3578     found_sig = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*sig));
3579     found_sig->sig = clone_sig(sig);
3580     found_sig->idx = priv->ps_sig_number++;
3581     TRACE("New signature stored and assigned number %u\n", found_sig->idx);
3582     if(wine_rb_put(&priv->signature_tree, sig, &found_sig->entry) == -1)
3583     {
3584         ERR("Failed to insert program entry.\n");
3585     }
3586     return found_sig->idx;
3587 }
3588
3589 static void init_output_registers(IWineD3DVertexShaderImpl *shader, DWORD sig_num, struct shader_arb_ctx_priv *priv_ctx,
3590                                   struct arb_vs_compiled_shader *compiled)
3591 {
3592     unsigned int i, j;
3593     static const char *texcoords[8] =
3594     {
3595         "result.texcoord[0]", "result.texcoord[1]", "result.texcoord[2]", "result.texcoord[3]",
3596         "result.texcoord[4]", "result.texcoord[5]", "result.texcoord[6]", "result.texcoord[7]"
3597     };
3598     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *) shader->baseShader.device;
3599     const struct wined3d_shader_signature_element *sig;
3600     const char *semantic_name;
3601     DWORD semantic_idx, reg_idx;
3602
3603     /* Write generic input varyings 0 to 7 to result.texcoord[], varying 8 to result.color.primary
3604      * and varying 9 to result.color.secondary
3605      */
3606     const char *decl_idx_to_string[MAX_REG_INPUT] =
3607     {
3608         texcoords[0], texcoords[1], texcoords[2], texcoords[3],
3609         texcoords[4], texcoords[5], texcoords[6], texcoords[7],
3610         "result.color.primary", "result.color.secondary"
3611     };
3612
3613     if(sig_num == ~0)
3614     {
3615         TRACE("Pixel shader uses builtin varyings\n");
3616         /* Map builtins to builtins */
3617         for(i = 0; i < 8; i++)
3618         {
3619             priv_ctx->texcrd_output[i] = texcoords[i];
3620         }
3621         priv_ctx->color_output[0] = "result.color.primary";
3622         priv_ctx->color_output[1] = "result.color.secondary";
3623         priv_ctx->fog_output = "result.fogcoord";
3624
3625         /* Map declared regs to builtins. Use "TA" to /dev/null unread output */
3626         for(i = 0; i < (sizeof(shader->output_signature) / sizeof(*shader->output_signature)); i++)
3627         {
3628             semantic_name = shader->output_signature[i].semantic_name;
3629             if(semantic_name == NULL) continue;
3630
3631             if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_POSITION))
3632             {
3633                 TRACE("o%u is TMP_OUT\n", i);
3634                 if(shader->output_signature[i].semantic_idx == 0) priv_ctx->vs_output[i] = "TMP_OUT";
3635                 else priv_ctx->vs_output[i] = "TA";
3636             }
3637             else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_PSIZE))
3638             {
3639                 TRACE("o%u is result.pointsize\n", i);
3640                 if(shader->output_signature[i].semantic_idx == 0) priv_ctx->vs_output[i] = "result.pointsize";
3641                 else priv_ctx->vs_output[i] = "TA";
3642             }
3643             else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_COLOR))
3644             {
3645                 TRACE("o%u is result.color.?, idx %u\n", i, shader->output_signature[i].semantic_idx);
3646                 if(shader->output_signature[i].semantic_idx == 0) priv_ctx->vs_output[i] = "result.color.primary";
3647                 else if(shader->output_signature[i].semantic_idx == 1) priv_ctx->vs_output[i] = "result.color.secondary";
3648                 else priv_ctx->vs_output[i] = "TA";
3649             }
3650             else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_TEXCOORD))
3651             {
3652                 TRACE("o%u is %s\n", i, texcoords[shader->output_signature[i].semantic_idx]);
3653                 if(shader->output_signature[i].semantic_idx >= 8) priv_ctx->vs_output[i] = "TA";
3654                 else priv_ctx->vs_output[i] = texcoords[shader->output_signature[i].semantic_idx];
3655             }
3656             else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_FOG))
3657             {
3658                 TRACE("o%u is result.fogcoord\n", i);
3659                 if(shader->output_signature[i].semantic_idx > 0) priv_ctx->vs_output[i] = "TA";
3660                 else priv_ctx->vs_output[i] = "result.fogcoord";
3661             }
3662             else
3663             {
3664                 priv_ctx->vs_output[i] = "TA";
3665             }
3666         }
3667         return;
3668     }
3669
3670     /* Instead of searching for the signature in the signature list, read the one from the current pixel shader.
3671      * Its maybe not the shader where the signature came from, but it is the same signature and faster to find
3672      */
3673     sig = ((IWineD3DPixelShaderImpl *)device->stateBlock->pixelShader)->input_signature;
3674     TRACE("Pixel shader uses declared varyings\n");
3675
3676     /* Map builtin to declared. /dev/null the results by default to the TA temp reg */
3677     for(i = 0; i < 8; i++)
3678     {
3679         priv_ctx->texcrd_output[i] = "TA";
3680     }
3681     priv_ctx->color_output[0] = "TA";
3682     priv_ctx->color_output[1] = "TA";
3683     priv_ctx->fog_output = "TA";
3684
3685     for(i = 0; i < MAX_REG_INPUT; i++)
3686     {
3687         semantic_name = sig[i].semantic_name;
3688         semantic_idx = sig[i].semantic_idx;
3689         reg_idx = sig[i].register_idx;
3690         if(semantic_name == NULL) continue;
3691
3692         /* If a declared input register is not written by builtin arguments, don't write to it.
3693          * GL_NV_vertex_program makes sure the input defaults to 0.0, which is correct with D3D
3694          *
3695          * Don't care about POSITION and PSIZE here - this is a builtin vertex shader, position goes
3696          * to TMP_OUT in any case
3697          */
3698         if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_TEXCOORD))
3699         {
3700             if(semantic_idx < 8) priv_ctx->texcrd_output[semantic_idx] = decl_idx_to_string[reg_idx];
3701         }
3702         else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_COLOR))
3703         {
3704             if(semantic_idx < 2) priv_ctx->color_output[semantic_idx] = decl_idx_to_string[reg_idx];
3705         }
3706         else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_FOG))
3707         {
3708             if(semantic_idx == 0) priv_ctx->fog_output = decl_idx_to_string[reg_idx];
3709         }
3710         else
3711         {
3712             continue;
3713         }
3714
3715         if(strcmp(decl_idx_to_string[reg_idx], "result.color.primary") == 0 ||
3716            strcmp(decl_idx_to_string[reg_idx], "result.color.secondary") == 0)
3717         {
3718             compiled->need_color_unclamp = TRUE;
3719         }
3720     }
3721
3722     /* Map declared to declared */
3723     for(i = 0; i < (sizeof(shader->output_signature) / sizeof(*shader->output_signature)); i++)
3724     {
3725         /* Write unread output to TA to throw them away */
3726         priv_ctx->vs_output[i] = "TA";
3727         semantic_name = shader->output_signature[i].semantic_name;
3728         if(semantic_name == NULL)
3729         {
3730             continue;
3731         }
3732
3733         if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_POSITION) &&
3734            shader->output_signature[i].semantic_idx == 0)
3735         {
3736             priv_ctx->vs_output[i] = "TMP_OUT";
3737             continue;
3738         }
3739         else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_PSIZE) &&
3740            shader->output_signature[i].semantic_idx == 0)
3741         {
3742             priv_ctx->vs_output[i] = "result.pointsize";
3743             continue;
3744         }
3745
3746         for(j = 0; j < MAX_REG_INPUT; j++)
3747         {
3748             if(sig[j].semantic_name == NULL)
3749             {
3750                 continue;
3751             }
3752
3753             if(strcmp(sig[j].semantic_name, semantic_name) == 0 &&
3754                sig[j].semantic_idx == shader->output_signature[i].semantic_idx)
3755             {
3756                 priv_ctx->vs_output[i] = decl_idx_to_string[sig[j].register_idx];
3757
3758                 if(strcmp(priv_ctx->vs_output[i], "result.color.primary") == 0 ||
3759                    strcmp(priv_ctx->vs_output[i], "result.color.secondary") == 0)
3760                 {
3761                     compiled->need_color_unclamp = TRUE;
3762                 }
3763             }
3764         }
3765     }
3766 }
3767
3768 /* GL locking is done by the caller */
3769 static GLuint shader_arb_generate_vshader(IWineD3DVertexShaderImpl *This, struct wined3d_shader_buffer *buffer,
3770         const struct arb_vs_compile_args *args, struct arb_vs_compiled_shader *compiled)
3771 {
3772     const shader_reg_maps *reg_maps = &This->baseShader.reg_maps;
3773     CONST DWORD *function = This->baseShader.function;
3774     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *)This->baseShader.device;
3775     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
3776     const local_constant *lconst;
3777     GLuint ret;
3778     DWORD next_local, *lconst_map = local_const_mapping((IWineD3DBaseShaderImpl *) This);
3779     struct shader_arb_ctx_priv priv_ctx;
3780     unsigned int i;
3781
3782     memset(&priv_ctx, 0, sizeof(priv_ctx));
3783     priv_ctx.cur_vs_args = args;
3784     list_init(&priv_ctx.control_frames);
3785     init_output_registers(This, args->ps_signature, &priv_ctx, compiled);
3786
3787     /*  Create the hw ARB shader */
3788     shader_addline(buffer, "!!ARBvp1.0\n");
3789
3790     /* Always enable the NV extension if available. Unlike fragment shaders, there is no
3791      * mesurable performance penalty, and we can always make use of it for clipplanes.
3792      */
3793     if(GL_SUPPORT(NV_VERTEX_PROGRAM3)) {
3794         shader_addline(buffer, "OPTION NV_vertex_program3;\n");
3795         priv_ctx.target_version = NV3;
3796         shader_addline(buffer, "ADDRESS aL;\n");
3797     } else if(GL_SUPPORT(NV_VERTEX_PROGRAM2_OPTION)) {
3798         shader_addline(buffer, "OPTION NV_vertex_program2;\n");
3799         priv_ctx.target_version = NV2;
3800         shader_addline(buffer, "ADDRESS aL;\n");
3801     } else {
3802         priv_ctx.target_version = ARB;
3803     }
3804
3805     shader_addline(buffer, "TEMP TMP_OUT;\n");
3806     if(need_helper_const(gl_info)) {
3807         shader_addline(buffer, "PARAM helper_const = { 2.0, -1.0, %d.0, 0.0 };\n", This->rel_offset);
3808     }
3809     if(need_mova_const((IWineD3DBaseShader *) This, gl_info)) {
3810         shader_addline(buffer, "PARAM mova_const = { 0.5, 0.0, 2.0, 1.0 };\n");
3811         shader_addline(buffer, "TEMP A0_SHADOW;\n");
3812     }
3813
3814     shader_addline(buffer, "TEMP TA;\n");
3815
3816     /* Base Declarations */
3817     next_local = shader_generate_arb_declarations( (IWineD3DBaseShader*) This, reg_maps, buffer, &GLINFO_LOCATION,
3818             lconst_map, &priv_ctx.vs_clipplanes, &priv_ctx);
3819
3820     for(i = 0; i < MAX_CONST_I; i++)
3821     {
3822         compiled->int_consts[i] = WINED3D_CONST_NUM_UNUSED;
3823         if(reg_maps->integer_constants & (1 << i) && priv_ctx.target_version >= NV2)
3824         {
3825             const DWORD *control_values = find_loop_control_values((IWineD3DBaseShaderImpl *) This, i);
3826
3827             if(control_values)
3828             {
3829                 shader_addline(buffer, "PARAM I%u = {%u, %u, %u, -1};\n", i,
3830                                 control_values[0], control_values[1], control_values[2]);
3831             }
3832             else
3833             {
3834                 compiled->int_consts[i] = next_local;
3835                 compiled->num_int_consts++;
3836                 shader_addline(buffer, "PARAM I%u = program.local[%u];\n", i, next_local++);
3837             }
3838         }
3839     }
3840
3841     /* We need a constant to fixup the final position */
3842     shader_addline(buffer, "PARAM posFixup = program.local[%u];\n", next_local);
3843     compiled->pos_fixup = next_local++;
3844
3845     /* Initialize output parameters. GL_ARB_vertex_program does not require special initialization values
3846      * for output parameters. D3D in theory does not do that either, but some applications depend on a
3847      * proper initialization of the secondary color, and programs using the fixed function pipeline without
3848      * a replacement shader depend on the texcoord.w being set properly.
3849      *
3850      * GL_NV_vertex_program defines that all output values are initialized to {0.0, 0.0, 0.0, 1.0}. This
3851      * assertion is in effect even when using GL_ARB_vertex_program without any NV specific additions. So
3852      * skip this if NV_vertex_program is supported. Otherwise, initialize the secondary color. For the tex-
3853      * coords, we have a flag in the opengl caps. Many cards do not require the texcoord being set, and
3854      * this can eat a number of instructions, so skip it unless this cap is set as well
3855      */
3856     if(!GL_SUPPORT(NV_VERTEX_PROGRAM)) {
3857         shader_addline(buffer, "MOV result.color.secondary, -helper_const.wwwy;\n");
3858
3859         if ((GLINFO_LOCATION).quirks & WINED3D_QUIRK_SET_TEXCOORD_W && !device->frag_pipe->ffp_proj_control)
3860         {
3861             int i;
3862             for(i = 0; i < min(8, MAX_REG_TEXCRD); i++) {
3863                 if(This->baseShader.reg_maps.texcoord_mask[i] != 0 &&
3864                 This->baseShader.reg_maps.texcoord_mask[i] != WINED3DSP_WRITEMASK_ALL) {
3865                     shader_addline(buffer, "MOV result.texcoord[%u].w, -helper_const.y;\n", i);
3866                 }
3867             }
3868         }
3869     }
3870
3871     /* The shader starts with the main function */
3872     priv_ctx.in_main_func = TRUE;
3873     /* Base Shader Body */
3874     shader_generate_main((IWineD3DBaseShader *)This, buffer, reg_maps, function, &priv_ctx);
3875
3876     if(!priv_ctx.footer_written) vshader_add_footer(This, buffer, args, &priv_ctx);
3877
3878     shader_addline(buffer, "END\n");
3879
3880     /* TODO: change to resource.glObjectHandle or something like that */
3881     GL_EXTCALL(glGenProgramsARB(1, &ret));
3882
3883     TRACE("Creating a hw vertex shader, prg=%d\n", ret);
3884     GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, ret));
3885
3886     TRACE("Created hw vertex shader, prg=%d\n", ret);
3887     /* Create the program and check for errors */
3888     GL_EXTCALL(glProgramStringARB(GL_VERTEX_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
3889                buffer->bsize, buffer->buffer));
3890
3891     if (glGetError() == GL_INVALID_OPERATION) {
3892         GLint errPos;
3893         glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &errPos);
3894         FIXME("HW VertexShader Error at position %d: %s\n",
3895               errPos, debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
3896         ret = -1;
3897     } else {
3898         /* Load immediate constants */
3899         if(lconst_map) {
3900             LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry) {
3901                 const float *value = (const float *)lconst->value;
3902                 GL_EXTCALL(glProgramLocalParameter4fvARB(GL_VERTEX_PROGRAM_ARB, lconst_map[lconst->idx], value));
3903             }
3904         }
3905     }
3906     HeapFree(GetProcessHeap(), 0, lconst_map);
3907
3908     return ret;
3909 }
3910
3911 static void find_clip_texcoord(IWineD3DPixelShaderImpl *ps)
3912 {
3913     struct arb_pshader_private *shader_priv = ps->backend_priv;
3914     int i;
3915     const struct wined3d_gl_info *gl_info = &((IWineD3DDeviceImpl *)ps->baseShader.device)->adapter->gl_info;
3916
3917     /* See if we can use fragment.texcoord[7] for clipplane emulation
3918      *
3919      * Don't do this if it is not supported, or fragment.texcoord[7] is used
3920      */
3921     if(ps->baseShader.reg_maps.shader_version.major < 3)
3922     {
3923         for(i = GL_LIMITS(texture_stages); i > 0; i--)
3924         {
3925             if(!ps->baseShader.reg_maps.texcoord[i - 1])
3926             {
3927                 shader_priv->clipplane_emulation = i;
3928                 return;
3929             }
3930         }
3931         WARN("Did not find a free clip reg(2.0)\n");
3932     }
3933     else
3934     {
3935         for(i = GL_LIMITS(texture_stages); i > 0; i--)
3936         {
3937             if(!(ps->baseShader.reg_maps.input_registers & (1 << (i - 1))))
3938             {
3939                 shader_priv->clipplane_emulation = i;
3940                 return;
3941             }
3942         }
3943         WARN("Did not find a free clip reg(3.0)\n");
3944     }
3945 }
3946
3947 /* GL locking is done by the caller */
3948 static struct arb_ps_compiled_shader *find_arb_pshader(IWineD3DPixelShaderImpl *shader, const struct arb_ps_compile_args *args)
3949 {
3950     UINT i;
3951     DWORD new_size;
3952     struct arb_ps_compiled_shader *new_array;
3953     struct wined3d_shader_buffer buffer;
3954     struct arb_pshader_private *shader_data;
3955     GLuint ret;
3956
3957     if(!shader->backend_priv) {
3958         IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *) shader->baseShader.device;
3959         struct shader_arb_priv *priv = device->shader_priv;
3960
3961         shader->backend_priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data));
3962         shader_data = shader->backend_priv;
3963         shader_data->clamp_consts = shader->baseShader.reg_maps.shader_version.major == 1;
3964
3965         if(shader->baseShader.reg_maps.shader_version.major < 3) shader_data->input_signature_idx = ~0;
3966         else shader_data->input_signature_idx = find_input_signature(priv, shader->input_signature);
3967
3968         shader_data->has_signature_idx = TRUE;
3969         TRACE("Shader got assigned input signature index %u\n", shader_data->input_signature_idx);
3970
3971         if(!device->vs_clipping) find_clip_texcoord(shader);
3972     }
3973     shader_data = shader->backend_priv;
3974
3975     /* Usually we have very few GL shaders for each d3d shader(just 1 or maybe 2),
3976      * so a linear search is more performant than a hashmap or a binary search
3977      * (cache coherency etc)
3978      */
3979     for(i = 0; i < shader_data->num_gl_shaders; i++) {
3980         if(memcmp(&shader_data->gl_shaders[i].args, args, sizeof(*args)) == 0) {
3981             return &shader_data->gl_shaders[i];
3982         }
3983     }
3984
3985     TRACE("No matching GL shader found, compiling a new shader\n");
3986     if(shader_data->shader_array_size == shader_data->num_gl_shaders) {
3987         if (shader_data->num_gl_shaders)
3988         {
3989             new_size = shader_data->shader_array_size + max(1, shader_data->shader_array_size / 2);
3990             new_array = HeapReAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, shader_data->gl_shaders,
3991                                     new_size * sizeof(*shader_data->gl_shaders));
3992         } else {
3993             new_array = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data->gl_shaders));
3994             new_size = 1;
3995         }
3996
3997         if(!new_array) {
3998             ERR("Out of memory\n");
3999             return 0;
4000         }
4001         shader_data->gl_shaders = new_array;
4002         shader_data->shader_array_size = new_size;
4003     }
4004
4005     shader_data->gl_shaders[shader_data->num_gl_shaders].args = *args;
4006
4007     pixelshader_update_samplers(&shader->baseShader.reg_maps,
4008             ((IWineD3DDeviceImpl *)shader->baseShader.device)->stateBlock->textures);
4009
4010     if (!shader_buffer_init(&buffer))
4011     {
4012         ERR("Failed to initialize shader buffer.\n");
4013         return 0;
4014     }
4015
4016     ret = shader_arb_generate_pshader(shader, &buffer, args,
4017                                       &shader_data->gl_shaders[shader_data->num_gl_shaders]);
4018     shader_buffer_free(&buffer);
4019     shader_data->gl_shaders[shader_data->num_gl_shaders].prgId = ret;
4020
4021     return &shader_data->gl_shaders[shader_data->num_gl_shaders++];
4022 }
4023
4024 static inline BOOL vs_args_equal(const struct arb_vs_compile_args *stored, const struct arb_vs_compile_args *new,
4025                                  const DWORD use_map, BOOL skip_int) {
4026     if((stored->super.swizzle_map & use_map) != new->super.swizzle_map) return FALSE;
4027     if(stored->super.fog_src != new->super.fog_src) return FALSE;
4028     if(stored->boolclip_compare != new->boolclip_compare) return FALSE;
4029     if(stored->ps_signature != new->ps_signature) return FALSE;
4030     if(stored->vertex_samplers_compare != new->vertex_samplers_compare) return FALSE;
4031     if(skip_int) return TRUE;
4032
4033     return memcmp(stored->loop_ctrl, new->loop_ctrl, sizeof(stored->loop_ctrl)) == 0;
4034 }
4035
4036 static struct arb_vs_compiled_shader *find_arb_vshader(IWineD3DVertexShaderImpl *shader, const struct arb_vs_compile_args *args)
4037 {
4038     UINT i;
4039     DWORD new_size;
4040     struct arb_vs_compiled_shader *new_array;
4041     DWORD use_map = ((IWineD3DDeviceImpl *)shader->baseShader.device)->strided_streams.use_map;
4042     struct wined3d_shader_buffer buffer;
4043     struct arb_vshader_private *shader_data;
4044     GLuint ret;
4045     const struct wined3d_gl_info *gl_info = &((IWineD3DDeviceImpl *)shader->baseShader.device)->adapter->gl_info;
4046
4047     if(!shader->backend_priv) {
4048         shader->backend_priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data));
4049     }
4050     shader_data = shader->backend_priv;
4051
4052     /* Usually we have very few GL shaders for each d3d shader(just 1 or maybe 2),
4053      * so a linear search is more performant than a hashmap or a binary search
4054      * (cache coherency etc)
4055      */
4056     for(i = 0; i < shader_data->num_gl_shaders; i++) {
4057         if(vs_args_equal(&shader_data->gl_shaders[i].args, args, use_map, GL_SUPPORT(NV_VERTEX_PROGRAM2_OPTION))) {
4058             return &shader_data->gl_shaders[i];
4059         }
4060     }
4061
4062     TRACE("No matching GL shader found, compiling a new shader\n");
4063
4064     if(shader_data->shader_array_size == shader_data->num_gl_shaders) {
4065         if (shader_data->num_gl_shaders)
4066         {
4067             new_size = shader_data->shader_array_size + max(1, shader_data->shader_array_size / 2);
4068             new_array = HeapReAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, shader_data->gl_shaders,
4069                                     new_size * sizeof(*shader_data->gl_shaders));
4070         } else {
4071             new_array = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data->gl_shaders));
4072             new_size = 1;
4073         }
4074
4075         if(!new_array) {
4076             ERR("Out of memory\n");
4077             return 0;
4078         }
4079         shader_data->gl_shaders = new_array;
4080         shader_data->shader_array_size = new_size;
4081     }
4082
4083     shader_data->gl_shaders[shader_data->num_gl_shaders].args = *args;
4084
4085     if (!shader_buffer_init(&buffer))
4086     {
4087         ERR("Failed to initialize shader buffer.\n");
4088         return 0;
4089     }
4090
4091     ret = shader_arb_generate_vshader(shader, &buffer, args,
4092             &shader_data->gl_shaders[shader_data->num_gl_shaders]);
4093     shader_buffer_free(&buffer);
4094     shader_data->gl_shaders[shader_data->num_gl_shaders].prgId = ret;
4095
4096     return &shader_data->gl_shaders[shader_data->num_gl_shaders++];
4097 }
4098
4099 static inline void find_arb_ps_compile_args(IWineD3DPixelShaderImpl *shader, IWineD3DStateBlockImpl *stateblock,
4100         struct arb_ps_compile_args *args)
4101 {
4102     int i;
4103     WORD int_skip;
4104     const struct wined3d_gl_info *gl_info = &((IWineD3DDeviceImpl *)shader->baseShader.device)->adapter->gl_info;
4105     find_ps_compile_args(shader, stateblock, &args->super);
4106
4107     /* This forces all local boolean constants to 1 to make them stateblock independent */
4108     args->bools = shader->baseShader.reg_maps.local_bool_consts;
4109
4110     for(i = 0; i < MAX_CONST_B; i++)
4111     {
4112         if(stateblock->pixelShaderConstantB[i]) args->bools |= ( 1 << i);
4113     }
4114
4115     /* Skip if unused or local, or supported natively */
4116     int_skip = ~shader->baseShader.reg_maps.integer_constants | shader->baseShader.reg_maps.local_int_consts;
4117     if(int_skip == 0xffff || GL_SUPPORT(NV_FRAGMENT_PROGRAM_OPTION))
4118     {
4119         memset(&args->loop_ctrl, 0, sizeof(args->loop_ctrl));
4120         return;
4121     }
4122
4123     for(i = 0; i < MAX_CONST_I; i++)
4124     {
4125         if(int_skip & (1 << i))
4126         {
4127             args->loop_ctrl[i][0] = 0;
4128             args->loop_ctrl[i][1] = 0;
4129             args->loop_ctrl[i][2] = 0;
4130         }
4131         else
4132         {
4133             args->loop_ctrl[i][0] = stateblock->pixelShaderConstantI[i * 4];
4134             args->loop_ctrl[i][1] = stateblock->pixelShaderConstantI[i * 4 + 1];
4135             args->loop_ctrl[i][2] = stateblock->pixelShaderConstantI[i * 4 + 2];
4136         }
4137     }
4138 }
4139
4140 static inline void find_arb_vs_compile_args(IWineD3DVertexShaderImpl *shader, IWineD3DStateBlockImpl *stateblock,
4141         struct arb_vs_compile_args *args)
4142 {
4143     int i;
4144     WORD int_skip;
4145     IWineD3DDeviceImpl *dev = (IWineD3DDeviceImpl *)shader->baseShader.device;
4146     const struct wined3d_gl_info *gl_info = &dev->adapter->gl_info;
4147     find_vs_compile_args(shader, stateblock, &args->super);
4148
4149     args->boolclip_compare = 0;
4150     if(use_ps(stateblock))
4151     {
4152         IWineD3DPixelShaderImpl *ps = (IWineD3DPixelShaderImpl *) stateblock->pixelShader;
4153         struct arb_pshader_private *shader_priv = ps->backend_priv;
4154         args->ps_signature = shader_priv->input_signature_idx;
4155
4156         args->boolclip.clip_control[0] = shader_priv->clipplane_emulation;
4157     }
4158     else
4159     {
4160         args->ps_signature = ~0;
4161         if(!dev->vs_clipping)
4162         {
4163             args->boolclip.clip_control[0] = ffp_clip_emul(stateblock) ? GL_LIMITS(texture_stages) : 0;
4164         }
4165         /* Otherwise: Setting boolclip_compare set clip_control[0] to 0 */
4166     }
4167
4168     if(args->boolclip.clip_control[0])
4169     {
4170         if(stateblock->renderState[WINED3DRS_CLIPPING])
4171         {
4172             args->boolclip.clip_control[1] = stateblock->renderState[WINED3DRS_CLIPPLANEENABLE];
4173         }
4174         /* clip_control[1] was set to 0 by setting boolclip_compare to 0 */
4175     }
4176
4177     /* This forces all local boolean constants to 1 to make them stateblock independent */
4178     args->boolclip.bools = shader->baseShader.reg_maps.local_bool_consts;
4179     /* TODO: Figure out if it would be better to store bool constants as bitmasks in the stateblock */
4180     for(i = 0; i < MAX_CONST_B; i++)
4181     {
4182         if(stateblock->vertexShaderConstantB[i]) args->boolclip.bools |= ( 1 << i);
4183     }
4184
4185     args->vertex_samplers[0] = dev->texUnitMap[MAX_FRAGMENT_SAMPLERS + 0];
4186     args->vertex_samplers[1] = dev->texUnitMap[MAX_FRAGMENT_SAMPLERS + 1];
4187     args->vertex_samplers[2] = dev->texUnitMap[MAX_FRAGMENT_SAMPLERS + 2];
4188     args->vertex_samplers[3] = 0;
4189
4190     /* Skip if unused or local */
4191     int_skip = ~shader->baseShader.reg_maps.integer_constants | shader->baseShader.reg_maps.local_int_consts;
4192     if(int_skip == 0xffff || GL_SUPPORT(NV_VERTEX_PROGRAM2_OPTION)) /* This is about flow control, not clipping */
4193     {
4194         memset(&args->loop_ctrl, 0, sizeof(args->loop_ctrl));
4195         return;
4196     }
4197
4198     for(i = 0; i < MAX_CONST_I; i++)
4199     {
4200         if(int_skip & (1 << i))
4201         {
4202             args->loop_ctrl[i][0] = 0;
4203             args->loop_ctrl[i][1] = 0;
4204             args->loop_ctrl[i][2] = 0;
4205         }
4206         else
4207         {
4208             args->loop_ctrl[i][0] = stateblock->vertexShaderConstantI[i * 4];
4209             args->loop_ctrl[i][1] = stateblock->vertexShaderConstantI[i * 4 + 1];
4210             args->loop_ctrl[i][2] = stateblock->vertexShaderConstantI[i * 4 + 2];
4211         }
4212     }
4213 }
4214
4215 /* GL locking is done by the caller */
4216 static void shader_arb_select(IWineD3DDevice *iface, BOOL usePS, BOOL useVS) {
4217     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
4218     struct shader_arb_priv *priv = This->shader_priv;
4219     struct wined3d_context *context = context_get_current();
4220     const struct wined3d_gl_info *gl_info = context->gl_info;
4221     int i;
4222
4223     /* Deal with pixel shaders first so the vertex shader arg function has the input signature ready */
4224     if (usePS) {
4225         struct arb_ps_compile_args compile_args;
4226         struct arb_ps_compiled_shader *compiled;
4227         IWineD3DPixelShaderImpl *ps = (IWineD3DPixelShaderImpl *) This->stateBlock->pixelShader;
4228
4229         TRACE("Using pixel shader %p\n", This->stateBlock->pixelShader);
4230         find_arb_ps_compile_args(ps, This->stateBlock, &compile_args);
4231         compiled = find_arb_pshader(ps, &compile_args);
4232         priv->current_fprogram_id = compiled->prgId;
4233         priv->compiled_fprog = compiled;
4234
4235         /* Bind the fragment program */
4236         GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, priv->current_fprogram_id));
4237         checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, priv->current_fprogram_id);");
4238
4239         if(!priv->use_arbfp_fixed_func) {
4240             /* Enable OpenGL fragment programs */
4241             glEnable(GL_FRAGMENT_PROGRAM_ARB);
4242             checkGLcall("glEnable(GL_FRAGMENT_PROGRAM_ARB);");
4243         }
4244         TRACE("(%p) : Bound fragment program %u and enabled GL_FRAGMENT_PROGRAM_ARB\n", This, priv->current_fprogram_id);
4245
4246         /* Pixel Shader 1.x constants are clamped to [-1;1], Pixel Shader 2.0 constants are not. If switching between
4247          * a 1.x and newer shader, reload the first 8 constants
4248          */
4249         if(priv->last_ps_const_clamped != ((struct arb_pshader_private *) ps->backend_priv)->clamp_consts)
4250         {
4251             priv->last_ps_const_clamped = ((struct arb_pshader_private *) ps->backend_priv)->clamp_consts;
4252             This->highest_dirty_ps_const = max(This->highest_dirty_ps_const, 8);
4253             for(i = 0; i < 8; i++)
4254             {
4255                 context->pshader_const_dirty[i] = 1;
4256             }
4257             /* Also takes care of loading local constants */
4258             shader_arb_load_constants(iface, TRUE, FALSE);
4259         }
4260         else
4261         {
4262             shader_arb_ps_local_constants(This);
4263         }
4264
4265         /* Force constant reloading for the NP2 fixup (see comment in shader_glsl_select for more info) */
4266         if (compiled->np2fixup_info.super.active) This->shader_backend->shader_load_np2fixup_constants(iface, usePS, useVS);
4267     } else if(GL_SUPPORT(ARB_FRAGMENT_PROGRAM) && !priv->use_arbfp_fixed_func) {
4268         /* Disable only if we're not using arbfp fixed function fragment processing. If this is used,
4269         * keep GL_FRAGMENT_PROGRAM_ARB enabled, and the fixed function pipeline will bind the fixed function
4270         * replacement shader
4271         */
4272         glDisable(GL_FRAGMENT_PROGRAM_ARB);
4273         checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
4274         priv->current_fprogram_id = 0;
4275     }
4276
4277     if (useVS) {
4278         struct arb_vs_compile_args compile_args;
4279         struct arb_vs_compiled_shader *compiled;
4280         IWineD3DVertexShaderImpl *vs = (IWineD3DVertexShaderImpl *) This->stateBlock->vertexShader;
4281
4282         TRACE("Using vertex shader %p\n", This->stateBlock->vertexShader);
4283         find_arb_vs_compile_args(vs, This->stateBlock, &compile_args);
4284         compiled = find_arb_vshader(vs, &compile_args);
4285         priv->current_vprogram_id = compiled->prgId;
4286         priv->compiled_vprog = compiled;
4287
4288         /* Bind the vertex program */
4289         GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->current_vprogram_id));
4290         checkGLcall("glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->current_vprogram_id);");
4291
4292         /* Enable OpenGL vertex programs */
4293         glEnable(GL_VERTEX_PROGRAM_ARB);
4294         checkGLcall("glEnable(GL_VERTEX_PROGRAM_ARB);");
4295         TRACE("(%p) : Bound vertex program %u and enabled GL_VERTEX_PROGRAM_ARB\n", This, priv->current_vprogram_id);
4296         shader_arb_vs_local_constants(This);
4297
4298         if(priv->last_vs_color_unclamp != compiled->need_color_unclamp) {
4299             priv->last_vs_color_unclamp = compiled->need_color_unclamp;
4300
4301             if (GL_SUPPORT(ARB_COLOR_BUFFER_FLOAT)) {
4302                 GL_EXTCALL(glClampColorARB(GL_CLAMP_VERTEX_COLOR_ARB, !compiled->need_color_unclamp));
4303                 checkGLcall("glClampColorARB");
4304             } else {
4305                 FIXME("vertex color clamp needs to be changed, but extension not supported.\n");
4306             }
4307         }
4308     } else if(GL_SUPPORT(ARB_VERTEX_PROGRAM)) {
4309         priv->current_vprogram_id = 0;
4310         glDisable(GL_VERTEX_PROGRAM_ARB);
4311         checkGLcall("glDisable(GL_VERTEX_PROGRAM_ARB)");
4312     }
4313 }
4314
4315 /* GL locking is done by the caller */
4316 static void shader_arb_select_depth_blt(IWineD3DDevice *iface, enum tex_types tex_type) {
4317     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
4318     struct shader_arb_priv *priv = This->shader_priv;
4319     GLuint *blt_fprogram = &priv->depth_blt_fprogram_id[tex_type];
4320     const struct wined3d_gl_info *gl_info = &This->adapter->gl_info;
4321
4322     if (!priv->depth_blt_vprogram_id) priv->depth_blt_vprogram_id = create_arb_blt_vertex_program(gl_info);
4323     GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->depth_blt_vprogram_id));
4324     glEnable(GL_VERTEX_PROGRAM_ARB);
4325
4326     if (!*blt_fprogram) *blt_fprogram = create_arb_blt_fragment_program(gl_info, tex_type);
4327     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, *blt_fprogram));
4328     glEnable(GL_FRAGMENT_PROGRAM_ARB);
4329 }
4330
4331 /* GL locking is done by the caller */
4332 static void shader_arb_deselect_depth_blt(IWineD3DDevice *iface) {
4333     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
4334     struct shader_arb_priv *priv = This->shader_priv;
4335     const struct wined3d_gl_info *gl_info = &This->adapter->gl_info;
4336
4337     if (priv->current_vprogram_id) {
4338         GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->current_vprogram_id));
4339         checkGLcall("glBindProgramARB(GL_VERTEX_PROGRAM_ARB, vertexShader->prgId);");
4340
4341         TRACE("(%p) : Bound vertex program %u and enabled GL_VERTEX_PROGRAM_ARB\n", This, priv->current_vprogram_id);
4342     } else {
4343         glDisable(GL_VERTEX_PROGRAM_ARB);
4344         checkGLcall("glDisable(GL_VERTEX_PROGRAM_ARB)");
4345     }
4346
4347     if (priv->current_fprogram_id) {
4348         GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, priv->current_fprogram_id));
4349         checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, pixelShader->prgId);");
4350
4351         TRACE("(%p) : Bound fragment program %u and enabled GL_FRAGMENT_PROGRAM_ARB\n", This, priv->current_fprogram_id);
4352     } else if(!priv->use_arbfp_fixed_func) {
4353         glDisable(GL_FRAGMENT_PROGRAM_ARB);
4354         checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
4355     }
4356 }
4357
4358 static void shader_arb_destroy(IWineD3DBaseShader *iface) {
4359     IWineD3DBaseShaderImpl *baseShader = (IWineD3DBaseShaderImpl *) iface;
4360     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *)baseShader->baseShader.device;
4361     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4362
4363     if (shader_is_pshader_version(baseShader->baseShader.reg_maps.shader_version.type))
4364     {
4365         IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *) iface;
4366         struct arb_pshader_private *shader_data = This->backend_priv;
4367         UINT i;
4368
4369         if(!shader_data) return; /* This can happen if a shader was never compiled */
4370         ENTER_GL();
4371
4372         if(shader_data->num_gl_shaders) ActivateContext(device, NULL, CTXUSAGE_RESOURCELOAD);
4373
4374         for(i = 0; i < shader_data->num_gl_shaders; i++) {
4375             GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId));
4376             checkGLcall("GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId))");
4377         }
4378         LEAVE_GL();
4379         HeapFree(GetProcessHeap(), 0, shader_data->gl_shaders);
4380         HeapFree(GetProcessHeap(), 0, shader_data);
4381         This->backend_priv = NULL;
4382     } else {
4383         IWineD3DVertexShaderImpl *This = (IWineD3DVertexShaderImpl *) iface;
4384         struct arb_vshader_private *shader_data = This->backend_priv;
4385         UINT i;
4386
4387         if(!shader_data) return; /* This can happen if a shader was never compiled */
4388         ENTER_GL();
4389
4390         if(shader_data->num_gl_shaders) ActivateContext(device, NULL, CTXUSAGE_RESOURCELOAD);
4391
4392         for(i = 0; i < shader_data->num_gl_shaders; i++) {
4393             GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId));
4394             checkGLcall("GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId))");
4395         }
4396         LEAVE_GL();
4397         HeapFree(GetProcessHeap(), 0, shader_data->gl_shaders);
4398         HeapFree(GetProcessHeap(), 0, shader_data);
4399         This->backend_priv = NULL;
4400     }
4401 }
4402
4403 static int sig_tree_compare(const void *key, const struct wine_rb_entry *entry)
4404 {
4405     struct ps_signature *e = WINE_RB_ENTRY_VALUE(entry, struct ps_signature, entry);
4406     return compare_sig(key, e->sig);
4407 }
4408
4409 struct wine_rb_functions sig_tree_functions =
4410 {
4411     wined3d_rb_alloc,
4412     wined3d_rb_realloc,
4413     wined3d_rb_free,
4414     sig_tree_compare
4415 };
4416
4417 static HRESULT shader_arb_alloc(IWineD3DDevice *iface) {
4418     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
4419     struct shader_arb_priv *priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*priv));
4420     if(wine_rb_init(&priv->signature_tree, &sig_tree_functions) == -1)
4421     {
4422         ERR("RB tree init failed\n");
4423         HeapFree(GetProcessHeap(), 0, priv);
4424         return E_OUTOFMEMORY;
4425     }
4426     This->shader_priv = priv;
4427     return WINED3D_OK;
4428 }
4429
4430 static void release_signature(struct wine_rb_entry *entry, void *context)
4431 {
4432     struct ps_signature *sig = WINE_RB_ENTRY_VALUE(entry, struct ps_signature, entry);
4433     int i;
4434     for(i = 0; i < MAX_REG_INPUT; i++)
4435     {
4436         HeapFree(GetProcessHeap(), 0, (char *) sig->sig[i].semantic_name);
4437     }
4438     HeapFree(GetProcessHeap(), 0, sig->sig);
4439     HeapFree(GetProcessHeap(), 0, sig);
4440 }
4441
4442 /* Context activation is done by the caller. */
4443 static void shader_arb_free(IWineD3DDevice *iface) {
4444     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
4445     const struct wined3d_gl_info *gl_info = &This->adapter->gl_info;
4446     struct shader_arb_priv *priv = This->shader_priv;
4447     int i;
4448
4449     ENTER_GL();
4450     if(priv->depth_blt_vprogram_id) {
4451         GL_EXTCALL(glDeleteProgramsARB(1, &priv->depth_blt_vprogram_id));
4452     }
4453     for (i = 0; i < tex_type_count; ++i) {
4454         if (priv->depth_blt_fprogram_id[i]) {
4455             GL_EXTCALL(glDeleteProgramsARB(1, &priv->depth_blt_fprogram_id[i]));
4456         }
4457     }
4458     LEAVE_GL();
4459
4460     wine_rb_destroy(&priv->signature_tree, release_signature, NULL);
4461     HeapFree(GetProcessHeap(), 0, This->shader_priv);
4462 }
4463
4464 static BOOL shader_arb_dirty_const(IWineD3DDevice *iface) {
4465     return TRUE;
4466 }
4467
4468 static void shader_arb_get_caps(WINED3DDEVTYPE devtype, const struct wined3d_gl_info *gl_info,
4469         struct shader_caps *pCaps)
4470 {
4471     /* We don't have an ARB fixed function pipeline yet, so let the none backend set its caps,
4472      * then overwrite the shader specific ones
4473      */
4474     none_shader_backend.shader_get_caps(devtype, gl_info, pCaps);
4475
4476     if(GL_SUPPORT(ARB_VERTEX_PROGRAM)) {
4477         if(GL_SUPPORT(NV_VERTEX_PROGRAM3))
4478         {
4479             pCaps->VertexShaderVersion = WINED3DVS_VERSION(3,0);
4480             TRACE_(d3d_caps)("Hardware vertex shader version 3.0 enabled (NV_VERTEX_PROGRAM3)\n");
4481         }
4482         else if(GL_LIMITS(vshader_constantsF) >= 256)
4483         {
4484             /* Shader Model 2.0 requires at least 256 vertex shader constants */
4485             pCaps->VertexShaderVersion = WINED3DVS_VERSION(2,0);
4486             TRACE_(d3d_caps)("Hardware vertex shader version 2.0 enabled (ARB_PROGRAM)\n");
4487         }
4488         else
4489         {
4490             pCaps->VertexShaderVersion = WINED3DVS_VERSION(1,1);
4491             TRACE_(d3d_caps)("Hardware vertex shader version 1.1 enabled (ARB_PROGRAM)\n");
4492         }
4493         pCaps->MaxVertexShaderConst = GL_LIMITS(vshader_constantsF);
4494     }
4495
4496     if(GL_SUPPORT(ARB_FRAGMENT_PROGRAM)) {
4497         if(GL_SUPPORT(NV_FRAGMENT_PROGRAM2))
4498         {
4499             pCaps->PixelShaderVersion    = WINED3DPS_VERSION(3,0);
4500             TRACE_(d3d_caps)("Hardware pixel shader version 3.0 enabled (NV_FRAGMENT_PROGRAM2)\n");
4501         }
4502         else if(GL_LIMITS(pshader_constantsF) >= 32)
4503         {
4504             /* Shader Model 2.0 requires at least 32 pixel shader constants */
4505             pCaps->PixelShaderVersion    = WINED3DPS_VERSION(2,0);
4506             TRACE_(d3d_caps)("Hardware pixel shader version 2.0 enabled (ARB_PROGRAM)\n");
4507         }
4508         else
4509         {
4510             pCaps->PixelShaderVersion    = WINED3DPS_VERSION(1,4);
4511             TRACE_(d3d_caps)("Hardware pixel shader version 1.4 enabled (ARB_PROGRAM)\n");
4512         }
4513         pCaps->PixelShader1xMaxValue = 8.0f;
4514         pCaps->MaxPixelShaderConst = GL_LIMITS(pshader_constantsF);
4515     }
4516
4517     pCaps->VSClipping = use_nv_clip(gl_info);
4518 }
4519
4520 static BOOL shader_arb_color_fixup_supported(struct color_fixup_desc fixup)
4521 {
4522     if (TRACE_ON(d3d_shader) && TRACE_ON(d3d))
4523     {
4524         TRACE("Checking support for color_fixup:\n");
4525         dump_color_fixup_desc(fixup);
4526     }
4527
4528     /* We support everything except YUV conversions. */
4529     if (!is_yuv_fixup(fixup))
4530     {
4531         TRACE("[OK]\n");
4532         return TRUE;
4533     }
4534
4535     TRACE("[FAILED]\n");
4536     return FALSE;
4537 }
4538
4539 static void shader_arb_add_instruction_modifiers(const struct wined3d_shader_instruction *ins) {
4540     DWORD shift;
4541     char write_mask[20], regstr[50];
4542     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
4543     BOOL is_color = FALSE;
4544     const struct wined3d_shader_dst_param *dst;
4545
4546     if (!ins->dst_count) return;
4547
4548     dst = &ins->dst[0];
4549     shift = dst->shift;
4550     if(shift == 0) return; /* Saturate alone is handled by the instructions */
4551
4552     shader_arb_get_write_mask(ins, dst, write_mask);
4553     shader_arb_get_register_name(ins, &dst->reg, regstr, &is_color);
4554
4555     /* Generate a line that does the output modifier computation
4556      * FIXME: _SAT vs shift? _SAT alone is already handled in the instructions, if this
4557      * maps problems in e.g. _d4_sat modify shader_arb_get_modifier
4558      */
4559     shader_addline(buffer, "MUL%s %s%s, %s, %s;\n", shader_arb_get_modifier(ins),
4560                    regstr, write_mask, regstr, shift_tab[shift]);
4561 }
4562
4563 static const SHADER_HANDLER shader_arb_instruction_handler_table[WINED3DSIH_TABLE_SIZE] =
4564 {
4565     /* WINED3DSIH_ABS           */ shader_hw_map2gl,
4566     /* WINED3DSIH_ADD           */ shader_hw_map2gl,
4567     /* WINED3DSIH_BEM           */ pshader_hw_bem,
4568     /* WINED3DSIH_BREAK         */ shader_hw_break,
4569     /* WINED3DSIH_BREAKC        */ shader_hw_breakc,
4570     /* WINED3DSIH_BREAKP        */ NULL,
4571     /* WINED3DSIH_CALL          */ shader_hw_call,
4572     /* WINED3DSIH_CALLNZ        */ NULL,
4573     /* WINED3DSIH_CMP           */ pshader_hw_cmp,
4574     /* WINED3DSIH_CND           */ pshader_hw_cnd,
4575     /* WINED3DSIH_CRS           */ shader_hw_map2gl,
4576     /* WINED3DSIH_DCL           */ NULL,
4577     /* WINED3DSIH_DEF           */ NULL,
4578     /* WINED3DSIH_DEFB          */ NULL,
4579     /* WINED3DSIH_DEFI          */ NULL,
4580     /* WINED3DSIH_DP2ADD        */ pshader_hw_dp2add,
4581     /* WINED3DSIH_DP3           */ shader_hw_map2gl,
4582     /* WINED3DSIH_DP4           */ shader_hw_map2gl,
4583     /* WINED3DSIH_DST           */ shader_hw_map2gl,
4584     /* WINED3DSIH_DSX           */ shader_hw_map2gl,
4585     /* WINED3DSIH_DSY           */ shader_hw_dsy,
4586     /* WINED3DSIH_ELSE          */ shader_hw_else,
4587     /* WINED3DSIH_ENDIF         */ shader_hw_endif,
4588     /* WINED3DSIH_ENDLOOP       */ shader_hw_endloop,
4589     /* WINED3DSIH_ENDREP        */ shader_hw_endrep,
4590     /* WINED3DSIH_EXP           */ shader_hw_scalar_op,
4591     /* WINED3DSIH_EXPP          */ shader_hw_scalar_op,
4592     /* WINED3DSIH_FRC           */ shader_hw_map2gl,
4593     /* WINED3DSIH_IF            */ NULL /* Hardcoded into the shader */,
4594     /* WINED3DSIH_IFC           */ shader_hw_ifc,
4595     /* WINED3DSIH_LABEL         */ shader_hw_label,
4596     /* WINED3DSIH_LIT           */ shader_hw_map2gl,
4597     /* WINED3DSIH_LOG           */ shader_hw_log_pow,
4598     /* WINED3DSIH_LOGP          */ shader_hw_log_pow,
4599     /* WINED3DSIH_LOOP          */ shader_hw_loop,
4600     /* WINED3DSIH_LRP           */ shader_hw_lrp,
4601     /* WINED3DSIH_M3x2          */ shader_hw_mnxn,
4602     /* WINED3DSIH_M3x3          */ shader_hw_mnxn,
4603     /* WINED3DSIH_M3x4          */ shader_hw_mnxn,
4604     /* WINED3DSIH_M4x3          */ shader_hw_mnxn,
4605     /* WINED3DSIH_M4x4          */ shader_hw_mnxn,
4606     /* WINED3DSIH_MAD           */ shader_hw_map2gl,
4607     /* WINED3DSIH_MAX           */ shader_hw_map2gl,
4608     /* WINED3DSIH_MIN           */ shader_hw_map2gl,
4609     /* WINED3DSIH_MOV           */ shader_hw_mov,
4610     /* WINED3DSIH_MOVA          */ shader_hw_mov,
4611     /* WINED3DSIH_MUL           */ shader_hw_map2gl,
4612     /* WINED3DSIH_NOP           */ shader_hw_nop,
4613     /* WINED3DSIH_NRM           */ shader_hw_nrm,
4614     /* WINED3DSIH_PHASE         */ NULL,
4615     /* WINED3DSIH_POW           */ shader_hw_log_pow,
4616     /* WINED3DSIH_RCP           */ shader_hw_scalar_op,
4617     /* WINED3DSIH_REP           */ shader_hw_rep,
4618     /* WINED3DSIH_RET           */ shader_hw_ret,
4619     /* WINED3DSIH_RSQ           */ shader_hw_scalar_op,
4620     /* WINED3DSIH_SETP          */ NULL,
4621     /* WINED3DSIH_SGE           */ shader_hw_map2gl,
4622     /* WINED3DSIH_SGN           */ shader_hw_sgn,
4623     /* WINED3DSIH_SINCOS        */ shader_hw_sincos,
4624     /* WINED3DSIH_SLT           */ shader_hw_map2gl,
4625     /* WINED3DSIH_SUB           */ shader_hw_map2gl,
4626     /* WINED3DSIH_TEX           */ pshader_hw_tex,
4627     /* WINED3DSIH_TEXBEM        */ pshader_hw_texbem,
4628     /* WINED3DSIH_TEXBEML       */ pshader_hw_texbem,
4629     /* WINED3DSIH_TEXCOORD      */ pshader_hw_texcoord,
4630     /* WINED3DSIH_TEXDEPTH      */ pshader_hw_texdepth,
4631     /* WINED3DSIH_TEXDP3        */ pshader_hw_texdp3,
4632     /* WINED3DSIH_TEXDP3TEX     */ pshader_hw_texdp3tex,
4633     /* WINED3DSIH_TEXKILL       */ pshader_hw_texkill,
4634     /* WINED3DSIH_TEXLDD        */ shader_hw_texldd,
4635     /* WINED3DSIH_TEXLDL        */ shader_hw_texldl,
4636     /* WINED3DSIH_TEXM3x2DEPTH  */ pshader_hw_texm3x2depth,
4637     /* WINED3DSIH_TEXM3x2PAD    */ pshader_hw_texm3x2pad,
4638     /* WINED3DSIH_TEXM3x2TEX    */ pshader_hw_texm3x2tex,
4639     /* WINED3DSIH_TEXM3x3       */ pshader_hw_texm3x3,
4640     /* WINED3DSIH_TEXM3x3DIFF   */ NULL,
4641     /* WINED3DSIH_TEXM3x3PAD    */ pshader_hw_texm3x3pad,
4642     /* WINED3DSIH_TEXM3x3SPEC   */ pshader_hw_texm3x3spec,
4643     /* WINED3DSIH_TEXM3x3TEX    */ pshader_hw_texm3x3tex,
4644     /* WINED3DSIH_TEXM3x3VSPEC  */ pshader_hw_texm3x3vspec,
4645     /* WINED3DSIH_TEXREG2AR     */ pshader_hw_texreg2ar,
4646     /* WINED3DSIH_TEXREG2GB     */ pshader_hw_texreg2gb,
4647     /* WINED3DSIH_TEXREG2RGB    */ pshader_hw_texreg2rgb,
4648 };
4649
4650 static inline BOOL get_bool_const(const struct wined3d_shader_instruction *ins, IWineD3DBaseShaderImpl *This, DWORD idx)
4651 {
4652     BOOL vshader = shader_is_vshader_version(This->baseShader.reg_maps.shader_version.type);
4653     WORD bools = 0;
4654     WORD flag = (1 << idx);
4655     const local_constant *constant;
4656     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
4657
4658     if(This->baseShader.reg_maps.local_bool_consts & flag)
4659     {
4660         /* What good is a if(bool) with a hardcoded local constant? I don't know, but handle it */
4661         LIST_FOR_EACH_ENTRY(constant, &This->baseShader.constantsB, local_constant, entry)
4662         {
4663             if (constant->idx == idx)
4664             {
4665                 return constant->value[0];
4666             }
4667         }
4668         ERR("Local constant not found\n");
4669         return FALSE;
4670     }
4671     else
4672     {
4673         if(vshader) bools = priv->cur_vs_args->boolclip.bools;
4674         else bools = priv->cur_ps_args->bools;
4675         return bools & flag;
4676     }
4677 }
4678
4679 static void get_loop_control_const(const struct wined3d_shader_instruction *ins,
4680         IWineD3DBaseShaderImpl *This, UINT idx, struct loop_control *loop_control)
4681 {
4682     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
4683
4684     /* Integer constants can either be a local constant, or they can be stored in the shader
4685      * type specific compile args. */
4686     if (This->baseShader.reg_maps.local_int_consts & (1 << idx))
4687     {
4688         const local_constant *constant;
4689
4690         LIST_FOR_EACH_ENTRY(constant, &This->baseShader.constantsI, local_constant, entry)
4691         {
4692             if (constant->idx == idx)
4693             {
4694                 loop_control->count = constant->value[0];
4695                 loop_control->start = constant->value[1];
4696                 /* Step is signed. */
4697                 loop_control->step = (int)constant->value[2];
4698                 return;
4699             }
4700         }
4701         /* If this happens the flag was set incorrectly */
4702         ERR("Local constant not found\n");
4703         loop_control->count = 0;
4704         loop_control->start = 0;
4705         loop_control->step = 0;
4706         return;
4707     }
4708
4709     switch (This->baseShader.reg_maps.shader_version.type)
4710     {
4711         case WINED3D_SHADER_TYPE_VERTEX:
4712             /* Count and aL start value are unsigned */
4713             loop_control->count = priv->cur_vs_args->loop_ctrl[idx][0];
4714             loop_control->start = priv->cur_vs_args->loop_ctrl[idx][1];
4715             /* Step is signed. */
4716             loop_control->step = ((char)priv->cur_vs_args->loop_ctrl[idx][2]);
4717             break;
4718
4719         case WINED3D_SHADER_TYPE_PIXEL:
4720             loop_control->count = priv->cur_ps_args->loop_ctrl[idx][0];
4721             loop_control->start = priv->cur_ps_args->loop_ctrl[idx][1];
4722             loop_control->step = ((char)priv->cur_ps_args->loop_ctrl[idx][2]);
4723             break;
4724
4725         default:
4726             FIXME("Unhandled shader type %#x.\n", This->baseShader.reg_maps.shader_version.type);
4727             break;
4728     }
4729 }
4730
4731 static void record_instruction(struct list *list, const struct wined3d_shader_instruction *ins)
4732 {
4733     unsigned int i;
4734     struct wined3d_shader_dst_param *dst_param = NULL;
4735     struct wined3d_shader_src_param *src_param = NULL, *rel_addr = NULL;
4736     struct recorded_instruction *rec = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*rec));
4737     if(!rec)
4738     {
4739         ERR("Out of memory\n");
4740         return;
4741     }
4742
4743     rec->ins = *ins;
4744     dst_param = HeapAlloc(GetProcessHeap(), 0, sizeof(*dst_param));
4745     if(!dst_param) goto free;
4746     *dst_param = *ins->dst;
4747     if(ins->dst->reg.rel_addr)
4748     {
4749         rel_addr = HeapAlloc(GetProcessHeap(), 0, sizeof(*dst_param->reg.rel_addr));
4750         if(!rel_addr) goto free;
4751         *rel_addr = *ins->dst->reg.rel_addr;
4752         dst_param->reg.rel_addr = rel_addr;
4753     }
4754     rec->ins.dst = dst_param;
4755
4756     src_param = HeapAlloc(GetProcessHeap(), 0, sizeof(*src_param) * ins->src_count);
4757     if(!src_param) goto free;
4758     for(i = 0; i < ins->src_count; i++)
4759     {
4760         src_param[i] = ins->src[i];
4761         if(ins->src[i].reg.rel_addr)
4762         {
4763             rel_addr = HeapAlloc(GetProcessHeap(), 0, sizeof(*rel_addr));
4764             if(!rel_addr) goto free;
4765             *rel_addr = *ins->src[i].reg.rel_addr;
4766             src_param[i].reg.rel_addr = rel_addr;
4767         }
4768     }
4769     rec->ins.src = src_param;
4770     list_add_tail(list, &rec->entry);
4771     return;
4772
4773 free:
4774     ERR("Out of memory\n");
4775     if(dst_param)
4776     {
4777         HeapFree(GetProcessHeap(), 0, (void *) dst_param->reg.rel_addr);
4778         HeapFree(GetProcessHeap(), 0, dst_param);
4779     }
4780     if(src_param)
4781     {
4782         for(i = 0; i < ins->src_count; i++)
4783         {
4784             HeapFree(GetProcessHeap(), 0, (void *) src_param[i].reg.rel_addr);
4785         }
4786         HeapFree(GetProcessHeap(), 0, src_param);
4787     }
4788     HeapFree(GetProcessHeap(), 0, rec);
4789 }
4790
4791 static void free_recorded_instruction(struct list *list)
4792 {
4793     struct recorded_instruction *rec_ins, *entry2;
4794     unsigned int i;
4795
4796     LIST_FOR_EACH_ENTRY_SAFE(rec_ins, entry2, list, struct recorded_instruction, entry)
4797     {
4798         list_remove(&rec_ins->entry);
4799         if(rec_ins->ins.dst)
4800         {
4801             HeapFree(GetProcessHeap(), 0, (void *) rec_ins->ins.dst->reg.rel_addr);
4802             HeapFree(GetProcessHeap(), 0, (void *) rec_ins->ins.dst);
4803         }
4804         if(rec_ins->ins.src)
4805         {
4806             for(i = 0; i < rec_ins->ins.src_count; i++)
4807             {
4808                 HeapFree(GetProcessHeap(), 0, (void *) rec_ins->ins.src[i].reg.rel_addr);
4809             }
4810             HeapFree(GetProcessHeap(), 0, (void *) rec_ins->ins.src);
4811         }
4812         HeapFree(GetProcessHeap(), 0, rec_ins);
4813     }
4814 }
4815
4816 static void shader_arb_handle_instruction(const struct wined3d_shader_instruction *ins) {
4817     SHADER_HANDLER hw_fct;
4818     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
4819     IWineD3DBaseShaderImpl *This = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
4820     struct control_frame *control_frame;
4821     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
4822     BOOL bool_const;
4823
4824     if(ins->handler_idx == WINED3DSIH_LOOP || ins->handler_idx == WINED3DSIH_REP)
4825     {
4826         control_frame = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*control_frame));
4827         list_add_head(&priv->control_frames, &control_frame->entry);
4828
4829         if(ins->handler_idx == WINED3DSIH_LOOP) control_frame->type = LOOP;
4830         if(ins->handler_idx == WINED3DSIH_REP) control_frame->type = REP;
4831
4832         if(priv->target_version >= NV2)
4833         {
4834             control_frame->loop_no = priv->num_loops++;
4835             priv->loop_depth++;
4836         }
4837         else
4838         {
4839             /* Don't bother recording when we're in a not used if branch */
4840             if(priv->muted)
4841             {
4842                 return;
4843             }
4844
4845             if(!priv->recording)
4846             {
4847                 list_init(&priv->record);
4848                 priv->recording = TRUE;
4849                 control_frame->outer_loop = TRUE;
4850                 get_loop_control_const(ins, This, ins->src[0].reg.idx, &control_frame->loop_control);
4851                 return; /* Instruction is handled */
4852             }
4853             /* Record this loop in the outer loop's recording */
4854         }
4855     }
4856     else if(ins->handler_idx == WINED3DSIH_ENDLOOP || ins->handler_idx == WINED3DSIH_ENDREP)
4857     {
4858         if(priv->target_version >= NV2)
4859         {
4860             /* Nothing to do. The control frame is popped after the HW instr handler */
4861         }
4862         else
4863         {
4864             struct list *e = list_head(&priv->control_frames);
4865             control_frame = LIST_ENTRY(e, struct control_frame, entry);
4866             list_remove(&control_frame->entry);
4867
4868             if(control_frame->outer_loop)
4869             {
4870                 int iteration, aL = 0;
4871                 struct list copy;
4872
4873                 /* Turn off recording before playback */
4874                 priv->recording = FALSE;
4875
4876                 /* Move the recorded instructions to a separate list and get them out of the private data
4877                  * structure. If there are nested loops, the shader_arb_handle_instruction below will
4878                  * be recorded again, thus priv->record might be overwritten
4879                  */
4880                 list_init(&copy);
4881                 list_move_tail(&copy, &priv->record);
4882                 list_init(&priv->record);
4883
4884                 if(ins->handler_idx == WINED3DSIH_ENDLOOP)
4885                 {
4886                     shader_addline(buffer, "#unrolling loop: %u iterations, aL=%u, inc %d\n",
4887                                    control_frame->loop_control.count, control_frame->loop_control.start,
4888                                    control_frame->loop_control.step);
4889                     aL = control_frame->loop_control.start;
4890                 }
4891                 else
4892                 {
4893                     shader_addline(buffer, "#unrolling rep: %u iterations\n", control_frame->loop_control.count);
4894                 }
4895
4896                 for (iteration = 0; iteration < control_frame->loop_control.count; ++iteration)
4897                 {
4898                     struct recorded_instruction *rec_ins;
4899                     if(ins->handler_idx == WINED3DSIH_ENDLOOP)
4900                     {
4901                         priv->aL = aL;
4902                         shader_addline(buffer, "#Iteration %d, aL=%d\n", iteration, aL);
4903                     }
4904                     else
4905                     {
4906                         shader_addline(buffer, "#Iteration %d\n", iteration);
4907                     }
4908
4909                     LIST_FOR_EACH_ENTRY(rec_ins, &copy, struct recorded_instruction, entry)
4910                     {
4911                         shader_arb_handle_instruction(&rec_ins->ins);
4912                     }
4913
4914                     if(ins->handler_idx == WINED3DSIH_ENDLOOP)
4915                     {
4916                         aL += control_frame->loop_control.step;
4917                     }
4918                 }
4919                 shader_addline(buffer, "#end loop/rep\n");
4920
4921                 free_recorded_instruction(&copy);
4922                 HeapFree(GetProcessHeap(), 0, control_frame);
4923                 return; /* Instruction is handled */
4924             }
4925             else
4926             {
4927                 /* This is a nested loop. Proceed to the normal recording function */
4928                 HeapFree(GetProcessHeap(), 0, control_frame);
4929             }
4930         }
4931     }
4932
4933     if(priv->recording)
4934     {
4935         record_instruction(&priv->record, ins);
4936         return;
4937     }
4938
4939     /* boolean if */
4940     if(ins->handler_idx == WINED3DSIH_IF)
4941     {
4942         control_frame = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*control_frame));
4943         list_add_head(&priv->control_frames, &control_frame->entry);
4944         control_frame->type = IF;
4945
4946         bool_const = get_bool_const(ins, This, ins->src[0].reg.idx);
4947         if(ins->src[0].modifiers == WINED3DSPSM_NOT) bool_const = !bool_const;
4948         if(!priv->muted && bool_const == FALSE)
4949         {
4950             shader_addline(buffer, "#if(FALSE){\n");
4951             priv->muted = TRUE;
4952             control_frame->muting = TRUE;
4953         }
4954         else shader_addline(buffer, "#if(TRUE) {\n");
4955
4956         return; /* Instruction is handled */
4957     }
4958     else if(ins->handler_idx == WINED3DSIH_IFC)
4959     {
4960         /* IF(bool) and if_cond(a, b) use the same ELSE and ENDIF tokens */
4961         control_frame = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*control_frame));
4962         control_frame->type = IFC;
4963         control_frame->ifc_no = priv->num_ifcs++;
4964         list_add_head(&priv->control_frames, &control_frame->entry);
4965     }
4966     else if(ins->handler_idx == WINED3DSIH_ELSE)
4967     {
4968         struct list *e = list_head(&priv->control_frames);
4969         control_frame = LIST_ENTRY(e, struct control_frame, entry);
4970
4971         if(control_frame->type == IF)
4972         {
4973             shader_addline(buffer, "#} else {\n");
4974             if(!priv->muted && !control_frame->muting)
4975             {
4976                 priv->muted = TRUE;
4977                 control_frame->muting = TRUE;
4978             }
4979             else if(control_frame->muting) priv->muted = FALSE;
4980             return; /* Instruction is handled. */
4981         }
4982         /* In case of an ifc, generate a HW shader instruction */
4983     }
4984     else if(ins->handler_idx == WINED3DSIH_ENDIF)
4985     {
4986         struct list *e = list_head(&priv->control_frames);
4987         control_frame = LIST_ENTRY(e, struct control_frame, entry);
4988
4989         if(control_frame->type == IF)
4990         {
4991             shader_addline(buffer, "#} endif\n");
4992             if(control_frame->muting) priv->muted = FALSE;
4993             list_remove(&control_frame->entry);
4994             HeapFree(GetProcessHeap(), 0, control_frame);
4995             return; /* Instruction is handled */
4996         }
4997     }
4998
4999     if(priv->muted) return;
5000
5001     /* Select handler */
5002     hw_fct = shader_arb_instruction_handler_table[ins->handler_idx];
5003
5004     /* Unhandled opcode */
5005     if (!hw_fct)
5006     {
5007         FIXME("Backend can't handle opcode %#x\n", ins->handler_idx);
5008         return;
5009     }
5010     hw_fct(ins);
5011
5012     if(ins->handler_idx == WINED3DSIH_ENDLOOP || ins->handler_idx == WINED3DSIH_ENDREP)
5013     {
5014         struct list *e = list_head(&priv->control_frames);
5015         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5016         list_remove(&control_frame->entry);
5017         HeapFree(GetProcessHeap(), 0, control_frame);
5018         priv->loop_depth--;
5019     }
5020     else if(ins->handler_idx == WINED3DSIH_ENDIF)
5021     {
5022         /* Non-ifc ENDIFs don't reach that place because of the return in the if block above */
5023         struct list *e = list_head(&priv->control_frames);
5024         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5025         list_remove(&control_frame->entry);
5026         HeapFree(GetProcessHeap(), 0, control_frame);
5027     }
5028
5029
5030     shader_arb_add_instruction_modifiers(ins);
5031 }
5032
5033 const shader_backend_t arb_program_shader_backend = {
5034     shader_arb_handle_instruction,
5035     shader_arb_select,
5036     shader_arb_select_depth_blt,
5037     shader_arb_deselect_depth_blt,
5038     shader_arb_update_float_vertex_constants,
5039     shader_arb_update_float_pixel_constants,
5040     shader_arb_load_constants,
5041     shader_arb_load_np2fixup_constants,
5042     shader_arb_destroy,
5043     shader_arb_alloc,
5044     shader_arb_free,
5045     shader_arb_dirty_const,
5046     shader_arb_get_caps,
5047     shader_arb_color_fixup_supported,
5048 };
5049
5050 /* ARB_fragment_program fixed function pipeline replacement definitions */
5051 #define ARB_FFP_CONST_TFACTOR           0
5052 #define ARB_FFP_CONST_SPECULAR_ENABLE   ((ARB_FFP_CONST_TFACTOR) + 1)
5053 #define ARB_FFP_CONST_CONSTANT(i)       ((ARB_FFP_CONST_SPECULAR_ENABLE) + 1 + i)
5054 #define ARB_FFP_CONST_BUMPMAT(i)        ((ARB_FFP_CONST_CONSTANT(7)) + 1 + i)
5055 #define ARB_FFP_CONST_LUMINANCE(i)      ((ARB_FFP_CONST_BUMPMAT(7)) + 1 + i)
5056
5057 struct arbfp_ffp_desc
5058 {
5059     struct ffp_frag_desc parent;
5060     GLuint shader;
5061     unsigned int num_textures_used;
5062 };
5063
5064 /* Context activation is done by the caller. */
5065 static void arbfp_enable(IWineD3DDevice *iface, BOOL enable) {
5066     ENTER_GL();
5067     if(enable) {
5068         glEnable(GL_FRAGMENT_PROGRAM_ARB);
5069         checkGLcall("glEnable(GL_FRAGMENT_PROGRAM_ARB)");
5070     } else {
5071         glDisable(GL_FRAGMENT_PROGRAM_ARB);
5072         checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
5073     }
5074     LEAVE_GL();
5075 }
5076
5077 static HRESULT arbfp_alloc(IWineD3DDevice *iface) {
5078     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *) iface;
5079     struct shader_arb_priv *priv;
5080     /* Share private data between the shader backend and the pipeline replacement, if both
5081      * are the arb implementation. This is needed to figure out whether ARBfp should be disabled
5082      * if no pixel shader is bound or not
5083      */
5084     if(This->shader_backend == &arb_program_shader_backend) {
5085         This->fragment_priv = This->shader_priv;
5086     } else {
5087         This->fragment_priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(struct shader_arb_priv));
5088         if(!This->fragment_priv) return E_OUTOFMEMORY;
5089     }
5090     priv = This->fragment_priv;
5091     if (wine_rb_init(&priv->fragment_shaders, &wined3d_ffp_frag_program_rb_functions) == -1)
5092     {
5093         ERR("Failed to initialize rbtree.\n");
5094         HeapFree(GetProcessHeap(), 0, This->fragment_priv);
5095         return E_OUTOFMEMORY;
5096     }
5097     priv->use_arbfp_fixed_func = TRUE;
5098     return WINED3D_OK;
5099 }
5100
5101 /* Context activation is done by the caller. */
5102 static void arbfp_free_ffpshader(struct wine_rb_entry *entry, void *context)
5103 {
5104     const struct wined3d_gl_info *gl_info = context;
5105     struct arbfp_ffp_desc *entry_arb = WINE_RB_ENTRY_VALUE(entry, struct arbfp_ffp_desc, parent.entry);
5106
5107     ENTER_GL();
5108     GL_EXTCALL(glDeleteProgramsARB(1, &entry_arb->shader));
5109     checkGLcall("glDeleteProgramsARB(1, &entry_arb->shader)");
5110     HeapFree(GetProcessHeap(), 0, entry_arb);
5111     LEAVE_GL();
5112 }
5113
5114 /* Context activation is done by the caller. */
5115 static void arbfp_free(IWineD3DDevice *iface) {
5116     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *) iface;
5117     struct shader_arb_priv *priv = This->fragment_priv;
5118
5119     wine_rb_destroy(&priv->fragment_shaders, arbfp_free_ffpshader, &This->adapter->gl_info);
5120     priv->use_arbfp_fixed_func = FALSE;
5121
5122     if(This->shader_backend != &arb_program_shader_backend) {
5123         HeapFree(GetProcessHeap(), 0, This->fragment_priv);
5124     }
5125 }
5126
5127 static void arbfp_get_caps(WINED3DDEVTYPE devtype, const struct wined3d_gl_info *gl_info, struct fragment_caps *caps)
5128 {
5129     caps->TextureOpCaps =  WINED3DTEXOPCAPS_DISABLE                     |
5130                            WINED3DTEXOPCAPS_SELECTARG1                  |
5131                            WINED3DTEXOPCAPS_SELECTARG2                  |
5132                            WINED3DTEXOPCAPS_MODULATE4X                  |
5133                            WINED3DTEXOPCAPS_MODULATE2X                  |
5134                            WINED3DTEXOPCAPS_MODULATE                    |
5135                            WINED3DTEXOPCAPS_ADDSIGNED2X                 |
5136                            WINED3DTEXOPCAPS_ADDSIGNED                   |
5137                            WINED3DTEXOPCAPS_ADD                         |
5138                            WINED3DTEXOPCAPS_SUBTRACT                    |
5139                            WINED3DTEXOPCAPS_ADDSMOOTH                   |
5140                            WINED3DTEXOPCAPS_BLENDCURRENTALPHA           |
5141                            WINED3DTEXOPCAPS_BLENDFACTORALPHA            |
5142                            WINED3DTEXOPCAPS_BLENDTEXTUREALPHA           |
5143                            WINED3DTEXOPCAPS_BLENDDIFFUSEALPHA           |
5144                            WINED3DTEXOPCAPS_BLENDTEXTUREALPHAPM         |
5145                            WINED3DTEXOPCAPS_MODULATEALPHA_ADDCOLOR      |
5146                            WINED3DTEXOPCAPS_MODULATECOLOR_ADDALPHA      |
5147                            WINED3DTEXOPCAPS_MODULATEINVCOLOR_ADDALPHA   |
5148                            WINED3DTEXOPCAPS_MODULATEINVALPHA_ADDCOLOR   |
5149                            WINED3DTEXOPCAPS_DOTPRODUCT3                 |
5150                            WINED3DTEXOPCAPS_MULTIPLYADD                 |
5151                            WINED3DTEXOPCAPS_LERP                        |
5152                            WINED3DTEXOPCAPS_BUMPENVMAP                  |
5153                            WINED3DTEXOPCAPS_BUMPENVMAPLUMINANCE;
5154
5155     /* TODO: Implement WINED3DTEXOPCAPS_PREMODULATE */
5156
5157     caps->MaxTextureBlendStages   = 8;
5158     caps->MaxSimultaneousTextures = min(GL_LIMITS(fragment_samplers), 8);
5159
5160     caps->PrimitiveMiscCaps |= WINED3DPMISCCAPS_TSSARGTEMP;
5161 }
5162 #undef GLINFO_LOCATION
5163
5164 #define GLINFO_LOCATION stateblock->wineD3DDevice->adapter->gl_info
5165 static void state_texfactor_arbfp(DWORD state, IWineD3DStateBlockImpl *stateblock, struct wined3d_context *context)
5166 {
5167     float col[4];
5168     IWineD3DDeviceImpl *device = stateblock->wineD3DDevice;
5169
5170     /* Don't load the parameter if we're using an arbfp pixel shader, otherwise we'll overwrite
5171      * application provided constants
5172      */
5173     if(device->shader_backend == &arb_program_shader_backend) {
5174         if (use_ps(stateblock)) return;
5175
5176         device = stateblock->wineD3DDevice;
5177         context->pshader_const_dirty[ARB_FFP_CONST_TFACTOR] = 1;
5178         device->highest_dirty_ps_const = max(device->highest_dirty_ps_const, ARB_FFP_CONST_TFACTOR + 1);
5179     }
5180
5181     D3DCOLORTOGLFLOAT4(stateblock->renderState[WINED3DRS_TEXTUREFACTOR], col);
5182     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_TFACTOR, col));
5183     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_TFACTOR, col)");
5184
5185 }
5186
5187 static void state_arb_specularenable(DWORD state, IWineD3DStateBlockImpl *stateblock, struct wined3d_context *context)
5188 {
5189     float col[4];
5190     IWineD3DDeviceImpl *device = stateblock->wineD3DDevice;
5191
5192     /* Don't load the parameter if we're using an arbfp pixel shader, otherwise we'll overwrite
5193      * application provided constants
5194      */
5195     if(device->shader_backend == &arb_program_shader_backend) {
5196         if (use_ps(stateblock)) return;
5197
5198         device = stateblock->wineD3DDevice;
5199         context->pshader_const_dirty[ARB_FFP_CONST_SPECULAR_ENABLE] = 1;
5200         device->highest_dirty_ps_const = max(device->highest_dirty_ps_const, ARB_FFP_CONST_SPECULAR_ENABLE + 1);
5201     }
5202
5203     if(stateblock->renderState[WINED3DRS_SPECULARENABLE]) {
5204         /* The specular color has no alpha */
5205         col[0] = 1.0f; col[1] = 1.0f;
5206         col[2] = 1.0f; col[3] = 0.0f;
5207     } else {
5208         col[0] = 0.0f; col[1] = 0.0f;
5209         col[2] = 0.0f; col[3] = 0.0f;
5210     }
5211     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_SPECULAR_ENABLE, col));
5212     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_SPECULAR_ENABLE, col)");
5213 }
5214
5215 static void set_bumpmat_arbfp(DWORD state, IWineD3DStateBlockImpl *stateblock, struct wined3d_context *context)
5216 {
5217     DWORD stage = (state - STATE_TEXTURESTAGE(0, 0)) / (WINED3D_HIGHEST_TEXTURE_STATE + 1);
5218     IWineD3DDeviceImpl *device = stateblock->wineD3DDevice;
5219     float mat[2][2];
5220
5221     if (use_ps(stateblock))
5222     {
5223         if(stage != 0 &&
5224            ((IWineD3DPixelShaderImpl *) stateblock->pixelShader)->baseShader.reg_maps.bumpmat[stage]) {
5225             /* The pixel shader has to know the bump env matrix. Do a constants update if it isn't scheduled
5226              * anyway
5227              */
5228             if(!isStateDirty(context, STATE_PIXELSHADERCONSTANT)) {
5229                 device->StateTable[STATE_PIXELSHADERCONSTANT].apply(STATE_PIXELSHADERCONSTANT, stateblock, context);
5230             }
5231         }
5232
5233         if(device->shader_backend == &arb_program_shader_backend) {
5234             /* Exit now, don't set the bumpmat below, otherwise we may overwrite pixel shader constants */
5235             return;
5236         }
5237     } else if(device->shader_backend == &arb_program_shader_backend) {
5238         context->pshader_const_dirty[ARB_FFP_CONST_BUMPMAT(stage)] = 1;
5239         device->highest_dirty_ps_const = max(device->highest_dirty_ps_const, ARB_FFP_CONST_BUMPMAT(stage) + 1);
5240     }
5241
5242     mat[0][0] = *((float *) &stateblock->textureState[stage][WINED3DTSS_BUMPENVMAT00]);
5243     mat[0][1] = *((float *) &stateblock->textureState[stage][WINED3DTSS_BUMPENVMAT01]);
5244     mat[1][0] = *((float *) &stateblock->textureState[stage][WINED3DTSS_BUMPENVMAT10]);
5245     mat[1][1] = *((float *) &stateblock->textureState[stage][WINED3DTSS_BUMPENVMAT11]);
5246
5247     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_BUMPMAT(stage), &mat[0][0]));
5248     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_BUMPMAT(stage), &mat[0][0])");
5249 }
5250
5251 static void tex_bumpenvlum_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 param[4];
5256
5257     if (use_ps(stateblock))
5258     {
5259         if(stage != 0 &&
5260            ((IWineD3DPixelShaderImpl *) stateblock->pixelShader)->baseShader.reg_maps.luminanceparams[stage]) {
5261             /* The pixel shader has to know the luminance offset. Do a constants update if it
5262              * isn't scheduled anyway
5263              */
5264             if(!isStateDirty(context, STATE_PIXELSHADERCONSTANT)) {
5265                 device->StateTable[STATE_PIXELSHADERCONSTANT].apply(STATE_PIXELSHADERCONSTANT, stateblock, context);
5266             }
5267         }
5268
5269         if(device->shader_backend == &arb_program_shader_backend) {
5270             /* Exit now, don't set the bumpmat below, otherwise we may overwrite pixel shader constants */
5271             return;
5272         }
5273     } else if(device->shader_backend == &arb_program_shader_backend) {
5274         context->pshader_const_dirty[ARB_FFP_CONST_LUMINANCE(stage)] = 1;
5275         device->highest_dirty_ps_const = max(device->highest_dirty_ps_const, ARB_FFP_CONST_LUMINANCE(stage) + 1);
5276     }
5277
5278     param[0] = *((float *) &stateblock->textureState[stage][WINED3DTSS_BUMPENVLSCALE]);
5279     param[1] = *((float *) &stateblock->textureState[stage][WINED3DTSS_BUMPENVLOFFSET]);
5280     param[2] = 0.0f;
5281     param[3] = 0.0f;
5282
5283     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_LUMINANCE(stage), param));
5284     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_LUMINANCE(stage), param)");
5285 }
5286
5287 static const char *get_argreg(struct wined3d_shader_buffer *buffer, DWORD argnum, unsigned int stage, DWORD arg)
5288 {
5289     const char *ret;
5290
5291     if(arg == ARG_UNUSED) return "unused"; /* This is the marker for unused registers */
5292
5293     switch(arg & WINED3DTA_SELECTMASK) {
5294         case WINED3DTA_DIFFUSE:
5295             ret = "fragment.color.primary"; break;
5296
5297         case WINED3DTA_CURRENT:
5298             if(stage == 0) ret = "fragment.color.primary";
5299             else ret = "ret";
5300             break;
5301
5302         case WINED3DTA_TEXTURE:
5303             switch(stage) {
5304                 case 0: ret = "tex0"; break;
5305                 case 1: ret = "tex1"; break;
5306                 case 2: ret = "tex2"; break;
5307                 case 3: ret = "tex3"; break;
5308                 case 4: ret = "tex4"; break;
5309                 case 5: ret = "tex5"; break;
5310                 case 6: ret = "tex6"; break;
5311                 case 7: ret = "tex7"; break;
5312                 default: ret = "unknown texture";
5313             }
5314             break;
5315
5316         case WINED3DTA_TFACTOR:
5317             ret = "tfactor"; break;
5318
5319         case WINED3DTA_SPECULAR:
5320             ret = "fragment.color.secondary"; break;
5321
5322         case WINED3DTA_TEMP:
5323             ret = "tempreg"; break;
5324
5325         case WINED3DTA_CONSTANT:
5326             FIXME("Implement perstage constants\n");
5327             switch(stage) {
5328                 case 0: ret = "const0"; break;
5329                 case 1: ret = "const1"; break;
5330                 case 2: ret = "const2"; break;
5331                 case 3: ret = "const3"; break;
5332                 case 4: ret = "const4"; break;
5333                 case 5: ret = "const5"; break;
5334                 case 6: ret = "const6"; break;
5335                 case 7: ret = "const7"; break;
5336                 default: ret = "unknown constant";
5337             }
5338             break;
5339
5340         default:
5341             return "unknown";
5342     }
5343
5344     if(arg & WINED3DTA_COMPLEMENT) {
5345         shader_addline(buffer, "SUB arg%u, const.x, %s;\n", argnum, ret);
5346         if(argnum == 0) ret = "arg0";
5347         if(argnum == 1) ret = "arg1";
5348         if(argnum == 2) ret = "arg2";
5349     }
5350     if(arg & WINED3DTA_ALPHAREPLICATE) {
5351         shader_addline(buffer, "MOV arg%u, %s.w;\n", argnum, ret);
5352         if(argnum == 0) ret = "arg0";
5353         if(argnum == 1) ret = "arg1";
5354         if(argnum == 2) ret = "arg2";
5355     }
5356     return ret;
5357 }
5358
5359 static void gen_ffp_instr(struct wined3d_shader_buffer *buffer, unsigned int stage, BOOL color,
5360         BOOL alpha, DWORD dst, DWORD op, DWORD dw_arg0, DWORD dw_arg1, DWORD dw_arg2)
5361 {
5362     const char *dstmask, *dstreg, *arg0, *arg1, *arg2;
5363     unsigned int mul = 1;
5364     BOOL mul_final_dest = FALSE;
5365
5366     if(color && alpha) dstmask = "";
5367     else if(color) dstmask = ".xyz";
5368     else dstmask = ".w";
5369
5370     if(dst == tempreg) dstreg = "tempreg";
5371     else dstreg = "ret";
5372
5373     arg0 = get_argreg(buffer, 0, stage, dw_arg0);
5374     arg1 = get_argreg(buffer, 1, stage, dw_arg1);
5375     arg2 = get_argreg(buffer, 2, stage, dw_arg2);
5376
5377     switch(op) {
5378         case WINED3DTOP_DISABLE:
5379             if(stage == 0) shader_addline(buffer, "MOV %s%s, fragment.color.primary;\n", dstreg, dstmask);
5380             break;
5381
5382         case WINED3DTOP_SELECTARG2:
5383             arg1 = arg2;
5384         case WINED3DTOP_SELECTARG1:
5385             shader_addline(buffer, "MOV %s%s, %s;\n", dstreg, dstmask, arg1);
5386             break;
5387
5388         case WINED3DTOP_MODULATE4X:
5389             mul = 2;
5390         case WINED3DTOP_MODULATE2X:
5391             mul *= 2;
5392             if(strcmp(dstreg, "result.color") == 0) {
5393                 dstreg = "ret";
5394                 mul_final_dest = TRUE;
5395             }
5396         case WINED3DTOP_MODULATE:
5397             shader_addline(buffer, "MUL %s%s, %s, %s;\n", dstreg, dstmask, arg1, arg2);
5398             break;
5399
5400         case WINED3DTOP_ADDSIGNED2X:
5401             mul = 2;
5402             if(strcmp(dstreg, "result.color") == 0) {
5403                 dstreg = "ret";
5404                 mul_final_dest = TRUE;
5405             }
5406         case WINED3DTOP_ADDSIGNED:
5407             shader_addline(buffer, "SUB arg2, %s, const.w;\n", arg2);
5408             arg2 = "arg2";
5409         case WINED3DTOP_ADD:
5410             shader_addline(buffer, "ADD_SAT %s%s, %s, %s;\n", dstreg, dstmask, arg1, arg2);
5411             break;
5412
5413         case WINED3DTOP_SUBTRACT:
5414             shader_addline(buffer, "SUB_SAT %s%s, %s, %s;\n", dstreg, dstmask, arg1, arg2);
5415             break;
5416
5417         case WINED3DTOP_ADDSMOOTH:
5418             shader_addline(buffer, "SUB arg1, const.x, %s;\n", arg1);
5419             shader_addline(buffer, "MAD_SAT %s%s, arg1, %s, %s;\n", dstreg, dstmask, arg2, arg1);
5420             break;
5421
5422         case WINED3DTOP_BLENDCURRENTALPHA:
5423             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_CURRENT);
5424             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
5425             break;
5426         case WINED3DTOP_BLENDFACTORALPHA:
5427             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_TFACTOR);
5428             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
5429             break;
5430         case WINED3DTOP_BLENDTEXTUREALPHA:
5431             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_TEXTURE);
5432             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
5433             break;
5434         case WINED3DTOP_BLENDDIFFUSEALPHA:
5435             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_DIFFUSE);
5436             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
5437             break;
5438
5439         case WINED3DTOP_BLENDTEXTUREALPHAPM:
5440             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_TEXTURE);
5441             shader_addline(buffer, "SUB arg0.w, const.x, %s.w;\n", arg0);
5442             shader_addline(buffer, "MAD_SAT %s%s, %s, arg0.w, %s;\n", dstreg, dstmask, arg2, arg1);
5443             break;
5444
5445         /* D3DTOP_PREMODULATE ???? */
5446
5447         case WINED3DTOP_MODULATEINVALPHA_ADDCOLOR:
5448             shader_addline(buffer, "SUB arg0.w, const.x, %s;\n", arg1);
5449             shader_addline(buffer, "MAD_SAT %s%s, arg0.w, %s, %s;\n", dstreg, dstmask, arg2, arg1);
5450             break;
5451         case WINED3DTOP_MODULATEALPHA_ADDCOLOR:
5452             shader_addline(buffer, "MAD_SAT %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg1, arg2, arg1);
5453             break;
5454         case WINED3DTOP_MODULATEINVCOLOR_ADDALPHA:
5455             shader_addline(buffer, "SUB arg0, const.x, %s;\n", arg1);
5456             shader_addline(buffer, "MAD_SAT %s%s, arg0, %s, %s.w;\n", dstreg, dstmask, arg2, arg1);
5457             break;
5458         case WINED3DTOP_MODULATECOLOR_ADDALPHA:
5459             shader_addline(buffer, "MAD_SAT %s%s, %s, %s, %s.w;\n", dstreg, dstmask, arg1, arg2, arg1);
5460             break;
5461
5462         case WINED3DTOP_DOTPRODUCT3:
5463             mul = 4;
5464             if(strcmp(dstreg, "result.color") == 0) {
5465                 dstreg = "ret";
5466                 mul_final_dest = TRUE;
5467             }
5468             shader_addline(buffer, "SUB arg1, %s, const.w;\n", arg1);
5469             shader_addline(buffer, "SUB arg2, %s, const.w;\n", arg2);
5470             shader_addline(buffer, "DP3_SAT %s%s, arg1, arg2;\n", dstreg, dstmask);
5471             break;
5472
5473         case WINED3DTOP_MULTIPLYADD:
5474             shader_addline(buffer, "MAD_SAT %s%s, %s, %s, %s;\n", dstreg, dstmask, arg1, arg2, arg0);
5475             break;
5476
5477         case WINED3DTOP_LERP:
5478             /* The msdn is not quite right here */
5479             shader_addline(buffer, "LRP %s%s, %s, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
5480             break;
5481
5482         case WINED3DTOP_BUMPENVMAP:
5483         case WINED3DTOP_BUMPENVMAPLUMINANCE:
5484             /* Those are handled in the first pass of the shader(generation pass 1 and 2) already */
5485             break;
5486
5487         default:
5488             FIXME("Unhandled texture op %08x\n", op);
5489     }
5490
5491     if(mul == 2) {
5492         shader_addline(buffer, "MUL_SAT %s%s, %s, const.y;\n", mul_final_dest ? "result.color" : dstreg, dstmask, dstreg);
5493     } else if(mul == 4) {
5494         shader_addline(buffer, "MUL_SAT %s%s, %s, const.z;\n", mul_final_dest ? "result.color" : dstreg, dstmask, dstreg);
5495     }
5496 }
5497
5498 /* The stateblock is passed for GLINFO_LOCATION */
5499 static GLuint gen_arbfp_ffp_shader(const struct ffp_frag_settings *settings, IWineD3DStateBlockImpl *stateblock)
5500 {
5501     unsigned int stage;
5502     struct wined3d_shader_buffer buffer;
5503     BOOL tex_read[MAX_TEXTURES] = {FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE};
5504     BOOL bump_used[MAX_TEXTURES] = {FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE};
5505     BOOL luminance_used[MAX_TEXTURES] = {FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE};
5506     const char *textype;
5507     const char *instr, *sat;
5508     char colorcor_dst[8];
5509     GLuint ret;
5510     DWORD arg0, arg1, arg2;
5511     BOOL tempreg_used = FALSE, tfactor_used = FALSE;
5512     BOOL op_equal;
5513     const char *final_combiner_src = "ret";
5514
5515     /* Find out which textures are read */
5516     for(stage = 0; stage < MAX_TEXTURES; stage++) {
5517         if(settings->op[stage].cop == WINED3DTOP_DISABLE) break;
5518         arg0 = settings->op[stage].carg0 & WINED3DTA_SELECTMASK;
5519         arg1 = settings->op[stage].carg1 & WINED3DTA_SELECTMASK;
5520         arg2 = settings->op[stage].carg2 & WINED3DTA_SELECTMASK;
5521         if(arg0 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
5522         if(arg1 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
5523         if(arg2 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
5524
5525         if(settings->op[stage].cop == WINED3DTOP_BLENDTEXTUREALPHA) tex_read[stage] = TRUE;
5526         if(settings->op[stage].cop == WINED3DTOP_BLENDTEXTUREALPHAPM) tex_read[stage] = TRUE;
5527         if(settings->op[stage].cop == WINED3DTOP_BUMPENVMAP) {
5528             bump_used[stage] = TRUE;
5529             tex_read[stage] = TRUE;
5530         }
5531         if(settings->op[stage].cop == WINED3DTOP_BUMPENVMAPLUMINANCE) {
5532             bump_used[stage] = TRUE;
5533             tex_read[stage] = TRUE;
5534             luminance_used[stage] = TRUE;
5535         } else if(settings->op[stage].cop == WINED3DTOP_BLENDFACTORALPHA) {
5536             tfactor_used = TRUE;
5537         }
5538
5539         if(arg0 == WINED3DTA_TFACTOR || arg1 == WINED3DTA_TFACTOR || arg2 == WINED3DTA_TFACTOR) {
5540             tfactor_used = TRUE;
5541         }
5542
5543         if(settings->op[stage].dst == tempreg) tempreg_used = TRUE;
5544         if(arg0 == WINED3DTA_TEMP || arg1 == WINED3DTA_TEMP || arg2 == WINED3DTA_TEMP) {
5545             tempreg_used = TRUE;
5546         }
5547
5548         if(settings->op[stage].aop == WINED3DTOP_DISABLE) continue;
5549         arg0 = settings->op[stage].aarg0 & WINED3DTA_SELECTMASK;
5550         arg1 = settings->op[stage].aarg1 & WINED3DTA_SELECTMASK;
5551         arg2 = settings->op[stage].aarg2 & WINED3DTA_SELECTMASK;
5552         if(arg0 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
5553         if(arg1 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
5554         if(arg2 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
5555
5556         if(arg0 == WINED3DTA_TEMP || arg1 == WINED3DTA_TEMP || arg2 == WINED3DTA_TEMP) {
5557             tempreg_used = TRUE;
5558         }
5559         if(arg0 == WINED3DTA_TFACTOR || arg1 == WINED3DTA_TFACTOR || arg2 == WINED3DTA_TFACTOR) {
5560             tfactor_used = TRUE;
5561         }
5562     }
5563
5564     /* Shader header */
5565     if (!shader_buffer_init(&buffer))
5566     {
5567         ERR("Failed to initialize shader buffer.\n");
5568         return 0;
5569     }
5570
5571     shader_addline(&buffer, "!!ARBfp1.0\n");
5572
5573     switch(settings->fog) {
5574         case FOG_OFF:                                                         break;
5575         case FOG_LINEAR: shader_addline(&buffer, "OPTION ARB_fog_linear;\n"); break;
5576         case FOG_EXP:    shader_addline(&buffer, "OPTION ARB_fog_exp;\n");    break;
5577         case FOG_EXP2:   shader_addline(&buffer, "OPTION ARB_fog_exp2;\n");   break;
5578         default: FIXME("Unexpected fog setting %d\n", settings->fog);
5579     }
5580
5581     shader_addline(&buffer, "PARAM const = {1, 2, 4, 0.5};\n");
5582     shader_addline(&buffer, "TEMP TMP;\n");
5583     shader_addline(&buffer, "TEMP ret;\n");
5584     if(tempreg_used || settings->sRGB_write) shader_addline(&buffer, "TEMP tempreg;\n");
5585     shader_addline(&buffer, "TEMP arg0;\n");
5586     shader_addline(&buffer, "TEMP arg1;\n");
5587     shader_addline(&buffer, "TEMP arg2;\n");
5588     for(stage = 0; stage < MAX_TEXTURES; stage++) {
5589         if(!tex_read[stage]) continue;
5590         shader_addline(&buffer, "TEMP tex%u;\n", stage);
5591         if(!bump_used[stage]) continue;
5592         shader_addline(&buffer, "PARAM bumpmat%u = program.env[%u];\n", stage, ARB_FFP_CONST_BUMPMAT(stage));
5593         if(!luminance_used[stage]) continue;
5594         shader_addline(&buffer, "PARAM luminance%u = program.env[%u];\n", stage, ARB_FFP_CONST_LUMINANCE(stage));
5595     }
5596     if(tfactor_used) {
5597         shader_addline(&buffer, "PARAM tfactor = program.env[%u];\n", ARB_FFP_CONST_TFACTOR);
5598     }
5599         shader_addline(&buffer, "PARAM specular_enable = program.env[%u];\n", ARB_FFP_CONST_SPECULAR_ENABLE);
5600
5601     if(settings->sRGB_write) {
5602         shader_addline(&buffer, "PARAM srgb_consts1 = {%f, %f, %f, %f};\n",
5603                        srgb_mul_low, srgb_cmp, srgb_pow, srgb_mul_high);
5604         shader_addline(&buffer, "PARAM srgb_consts2 = {%f, %f, %f, %f};\n",
5605                        srgb_sub_high, 0.0, 0.0, 0.0);
5606     }
5607
5608     if(ffp_clip_emul(stateblock) && settings->emul_clipplanes) shader_addline(&buffer, "KIL fragment.texcoord[7];\n");
5609
5610     /* Generate texture sampling instructions) */
5611     for(stage = 0; stage < MAX_TEXTURES && settings->op[stage].cop != WINED3DTOP_DISABLE; stage++) {
5612         if(!tex_read[stage]) continue;
5613
5614         switch(settings->op[stage].tex_type) {
5615             case tex_1d:                    textype = "1D";     break;
5616             case tex_2d:                    textype = "2D";     break;
5617             case tex_3d:                    textype = "3D";     break;
5618             case tex_cube:                  textype = "CUBE";   break;
5619             case tex_rect:                  textype = "RECT";   break;
5620             default: textype = "unexpected_textype";   break;
5621         }
5622
5623         if(settings->op[stage].cop == WINED3DTOP_BUMPENVMAP ||
5624            settings->op[stage].cop == WINED3DTOP_BUMPENVMAPLUMINANCE) {
5625             sat = "";
5626         } else {
5627             sat = "_SAT";
5628         }
5629
5630         if(settings->op[stage].projected == proj_none) {
5631             instr = "TEX";
5632         } else if(settings->op[stage].projected == proj_count4 ||
5633                   settings->op[stage].projected == proj_count3) {
5634             instr = "TXP";
5635         } else {
5636             FIXME("Unexpected projection mode %d\n", settings->op[stage].projected);
5637             instr = "TXP";
5638         }
5639
5640         if(stage > 0 &&
5641            (settings->op[stage - 1].cop == WINED3DTOP_BUMPENVMAP ||
5642             settings->op[stage - 1].cop == WINED3DTOP_BUMPENVMAPLUMINANCE)) {
5643             shader_addline(&buffer, "SWZ arg1, bumpmat%u, x, z, 0, 0;\n", stage - 1);
5644             shader_addline(&buffer, "DP3 ret.x, arg1, tex%u;\n", stage - 1);
5645             shader_addline(&buffer, "SWZ arg1, bumpmat%u, y, w, 0, 0;\n", stage - 1);
5646             shader_addline(&buffer, "DP3 ret.y, arg1, tex%u;\n", stage - 1);
5647
5648             /* with projective textures, texbem only divides the static texture coord, not the displacement,
5649              * so multiply the displacement with the dividing parameter before passing it to TXP
5650              */
5651             if (settings->op[stage].projected != proj_none) {
5652                 if(settings->op[stage].projected == proj_count4) {
5653                     shader_addline(&buffer, "MOV ret.w, fragment.texcoord[%u].w;\n", stage);
5654                     shader_addline(&buffer, "MUL ret.xyz, ret, fragment.texcoord[%u].w, fragment.texcoord[%u];\n", stage, stage);
5655                 } else {
5656                     shader_addline(&buffer, "MOV ret.w, fragment.texcoord[%u].z;\n", stage);
5657                     shader_addline(&buffer, "MAD ret.xyz, ret, fragment.texcoord[%u].z, fragment.texcoord[%u];\n", stage, stage);
5658                 }
5659             } else {
5660                 shader_addline(&buffer, "ADD ret, ret, fragment.texcoord[%u];\n", stage);
5661             }
5662
5663             shader_addline(&buffer, "%s%s tex%u, ret, texture[%u], %s;\n",
5664                            instr, sat, stage, stage, textype);
5665             if(settings->op[stage - 1].cop == WINED3DTOP_BUMPENVMAPLUMINANCE) {
5666                 shader_addline(&buffer, "MAD_SAT ret.x, tex%u.z, luminance%u.x, luminance%u.y;\n",
5667                                stage - 1, stage - 1, stage - 1);
5668                 shader_addline(&buffer, "MUL tex%u, tex%u, ret.x;\n", stage, stage);
5669             }
5670         } else if(settings->op[stage].projected == proj_count3) {
5671             shader_addline(&buffer, "MOV ret, fragment.texcoord[%u];\n", stage);
5672             shader_addline(&buffer, "MOV ret.w, ret.z;\n");
5673             shader_addline(&buffer, "%s%s tex%u, ret, texture[%u], %s;\n",
5674                             instr, sat, stage, stage, textype);
5675         } else {
5676             shader_addline(&buffer, "%s%s tex%u, fragment.texcoord[%u], texture[%u], %s;\n",
5677                             instr, sat, stage, stage, stage, textype);
5678         }
5679
5680         sprintf(colorcor_dst, "tex%u", stage);
5681         gen_color_correction(&buffer, colorcor_dst, WINED3DSP_WRITEMASK_ALL, "const.x", "const.y",
5682                 settings->op[stage].color_fixup);
5683     }
5684
5685     /* Generate the main shader */
5686     for(stage = 0; stage < MAX_TEXTURES; stage++) {
5687         if(settings->op[stage].cop == WINED3DTOP_DISABLE) {
5688             if(stage == 0) {
5689                 final_combiner_src = "fragment.color.primary";
5690             }
5691             break;
5692         }
5693
5694         if(settings->op[stage].cop == WINED3DTOP_SELECTARG1 &&
5695            settings->op[stage].aop == WINED3DTOP_SELECTARG1) {
5696             op_equal = settings->op[stage].carg1 == settings->op[stage].aarg1;
5697         } else if(settings->op[stage].cop == WINED3DTOP_SELECTARG1 &&
5698                   settings->op[stage].aop == WINED3DTOP_SELECTARG2) {
5699             op_equal = settings->op[stage].carg1 == settings->op[stage].aarg2;
5700         } else if(settings->op[stage].cop == WINED3DTOP_SELECTARG2 &&
5701                   settings->op[stage].aop == WINED3DTOP_SELECTARG1) {
5702             op_equal = settings->op[stage].carg2 == settings->op[stage].aarg1;
5703         } else if(settings->op[stage].cop == WINED3DTOP_SELECTARG2 &&
5704                   settings->op[stage].aop == WINED3DTOP_SELECTARG2) {
5705             op_equal = settings->op[stage].carg2 == settings->op[stage].aarg2;
5706         } else {
5707             op_equal = settings->op[stage].aop   == settings->op[stage].cop &&
5708                        settings->op[stage].carg0 == settings->op[stage].aarg0 &&
5709                        settings->op[stage].carg1 == settings->op[stage].aarg1 &&
5710                        settings->op[stage].carg2 == settings->op[stage].aarg2;
5711         }
5712
5713         if(settings->op[stage].aop == WINED3DTOP_DISABLE) {
5714             gen_ffp_instr(&buffer, stage, TRUE, FALSE, settings->op[stage].dst,
5715                           settings->op[stage].cop, settings->op[stage].carg0,
5716                           settings->op[stage].carg1, settings->op[stage].carg2);
5717             if(stage == 0) {
5718                 shader_addline(&buffer, "MOV ret.w, fragment.color.primary.w;\n");
5719             }
5720         } else if(op_equal) {
5721             gen_ffp_instr(&buffer, stage, TRUE, TRUE, settings->op[stage].dst,
5722                           settings->op[stage].cop, settings->op[stage].carg0,
5723                           settings->op[stage].carg1, settings->op[stage].carg2);
5724         } else {
5725             gen_ffp_instr(&buffer, stage, TRUE, FALSE, settings->op[stage].dst,
5726                           settings->op[stage].cop, settings->op[stage].carg0,
5727                           settings->op[stage].carg1, settings->op[stage].carg2);
5728             gen_ffp_instr(&buffer, stage, FALSE, TRUE, settings->op[stage].dst,
5729                           settings->op[stage].aop, settings->op[stage].aarg0,
5730                           settings->op[stage].aarg1, settings->op[stage].aarg2);
5731         }
5732     }
5733
5734     if(settings->sRGB_write) {
5735         shader_addline(&buffer, "MAD ret, fragment.color.secondary, specular_enable, %s;\n", final_combiner_src);
5736         arbfp_add_sRGB_correction(&buffer, "ret", "arg0", "arg1", "arg2", "tempreg", FALSE);
5737     } else {
5738         shader_addline(&buffer, "MAD result.color, fragment.color.secondary, specular_enable, %s;\n", final_combiner_src);
5739     }
5740
5741     /* Footer */
5742     shader_addline(&buffer, "END\n");
5743
5744     /* Generate the shader */
5745     GL_EXTCALL(glGenProgramsARB(1, &ret));
5746     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, ret));
5747     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB, strlen(buffer.buffer), buffer.buffer));
5748
5749     if (glGetError() == GL_INVALID_OPERATION) {
5750         GLint pos;
5751         glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
5752         FIXME("Fragment program error at position %d: %s\n", pos,
5753               debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
5754     }
5755     shader_buffer_free(&buffer);
5756     return ret;
5757 }
5758
5759 static void fragment_prog_arbfp(DWORD state, IWineD3DStateBlockImpl *stateblock, struct wined3d_context *context)
5760 {
5761     IWineD3DDeviceImpl *device = stateblock->wineD3DDevice;
5762     struct shader_arb_priv *priv = device->fragment_priv;
5763     BOOL use_pshader = use_ps(stateblock);
5764     BOOL use_vshader = use_vs(stateblock);
5765     struct ffp_frag_settings settings;
5766     const struct arbfp_ffp_desc *desc;
5767     unsigned int i;
5768
5769     TRACE("state %#x, stateblock %p, context %p\n", state, stateblock, context);
5770
5771     if(isStateDirty(context, STATE_RENDER(WINED3DRS_FOGENABLE))) {
5772         if(!use_pshader && device->shader_backend == &arb_program_shader_backend && context->last_was_pshader) {
5773             /* Reload fixed function constants since they collide with the pixel shader constants */
5774             for(i = 0; i < MAX_TEXTURES; i++) {
5775                 set_bumpmat_arbfp(STATE_TEXTURESTAGE(i, WINED3DTSS_BUMPENVMAT00), stateblock, context);
5776             }
5777             state_texfactor_arbfp(STATE_RENDER(WINED3DRS_TEXTUREFACTOR), stateblock, context);
5778             state_arb_specularenable(STATE_RENDER(WINED3DRS_SPECULARENABLE), stateblock, context);
5779         } else if(use_pshader && !isStateDirty(context, device->StateTable[STATE_VSHADER].representative)) {
5780             device->shader_backend->shader_select((IWineD3DDevice *)stateblock->wineD3DDevice, use_pshader, use_vshader);
5781         }
5782         return;
5783     }
5784
5785     if(!use_pshader) {
5786         /* Find or create a shader implementing the fixed function pipeline settings, then activate it */
5787         gen_ffp_frag_op(stateblock, &settings, FALSE);
5788         desc = (const struct arbfp_ffp_desc *)find_ffp_frag_shader(&priv->fragment_shaders, &settings);
5789         if(!desc) {
5790             struct arbfp_ffp_desc *new_desc = HeapAlloc(GetProcessHeap(), 0, sizeof(*new_desc));
5791             if (!new_desc)
5792             {
5793                 ERR("Out of memory\n");
5794                 return;
5795             }
5796             new_desc->num_textures_used = 0;
5797             for(i = 0; i < GL_LIMITS(texture_stages); i++) {
5798                 if(settings.op[i].cop == WINED3DTOP_DISABLE) break;
5799                 new_desc->num_textures_used = i;
5800             }
5801
5802             memcpy(&new_desc->parent.settings, &settings, sizeof(settings));
5803             new_desc->shader = gen_arbfp_ffp_shader(&settings, stateblock);
5804             add_ffp_frag_shader(&priv->fragment_shaders, &new_desc->parent);
5805             TRACE("Allocated fixed function replacement shader descriptor %p\n", new_desc);
5806             desc = new_desc;
5807         }
5808
5809         /* Now activate the replacement program. GL_FRAGMENT_PROGRAM_ARB is already active(however, note the
5810          * comment above the shader_select call below). If e.g. GLSL is active, the shader_select call will
5811          * deactivate it.
5812          */
5813         GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, desc->shader));
5814         checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, desc->shader)");
5815         priv->current_fprogram_id = desc->shader;
5816
5817         if(device->shader_backend == &arb_program_shader_backend && context->last_was_pshader) {
5818             /* Reload fixed function constants since they collide with the pixel shader constants */
5819             for(i = 0; i < MAX_TEXTURES; i++) {
5820                 set_bumpmat_arbfp(STATE_TEXTURESTAGE(i, WINED3DTSS_BUMPENVMAT00), stateblock, context);
5821             }
5822             state_texfactor_arbfp(STATE_RENDER(WINED3DRS_TEXTUREFACTOR), stateblock, context);
5823             state_arb_specularenable(STATE_RENDER(WINED3DRS_SPECULARENABLE), stateblock, context);
5824         }
5825         context->last_was_pshader = FALSE;
5826     } else {
5827         context->last_was_pshader = TRUE;
5828     }
5829
5830     /* Finally, select the shader. If a pixel shader is used, it will be set and enabled by the shader backend.
5831      * If this shader backend is arbfp(most likely), then it will simply overwrite the last fixed function replace-
5832      * ment shader. If the shader backend is not ARB, it currently is important that the opengl implementation
5833      * type overwrites GL_ARB_fragment_program. This is currently the case with GLSL. If we really want to use
5834      * atifs or nvrc pixel shaders with arb fragment programs we'd have to disable GL_FRAGMENT_PROGRAM_ARB here
5835      *
5836      * Don't call shader_select if the vertex shader is dirty, because it will be called later on by the vertex
5837      * shader handler
5838      */
5839     if(!isStateDirty(context, device->StateTable[STATE_VSHADER].representative)) {
5840         device->shader_backend->shader_select((IWineD3DDevice *)stateblock->wineD3DDevice, use_pshader, use_vshader);
5841
5842         if (!isStateDirty(context, STATE_VERTEXSHADERCONSTANT) && (use_vshader || use_pshader)) {
5843             device->StateTable[STATE_VERTEXSHADERCONSTANT].apply(STATE_VERTEXSHADERCONSTANT, stateblock, context);
5844         }
5845     }
5846     if(use_pshader) {
5847         device->StateTable[STATE_PIXELSHADERCONSTANT].apply(STATE_PIXELSHADERCONSTANT, stateblock, context);
5848     }
5849 }
5850
5851 /* We can't link the fog states to the fragment state directly since the vertex pipeline links them
5852  * to FOGENABLE. A different linking in different pipeline parts can't be expressed in the combined
5853  * state table, so we need to handle that with a forwarding function. The other invisible side effect
5854  * is that changing the fog start and fog end(which links to FOGENABLE in vertex) results in the
5855  * fragment_prog_arbfp function being called because FOGENABLE is dirty, which calls this function here
5856  */
5857 static void state_arbfp_fog(DWORD state, IWineD3DStateBlockImpl *stateblock, struct wined3d_context *context)
5858 {
5859     enum fogsource new_source;
5860
5861     TRACE("state %#x, stateblock %p, context %p\n", state, stateblock, context);
5862
5863     if(!isStateDirty(context, STATE_PIXELSHADER)) {
5864         fragment_prog_arbfp(state, stateblock, context);
5865     }
5866
5867     if(!stateblock->renderState[WINED3DRS_FOGENABLE]) return;
5868
5869     if(stateblock->renderState[WINED3DRS_FOGTABLEMODE] == WINED3DFOG_NONE) {
5870         if(use_vs(stateblock)) {
5871             new_source = FOGSOURCE_VS;
5872         } else {
5873             if(stateblock->renderState[WINED3DRS_FOGVERTEXMODE] == WINED3DFOG_NONE || context->last_was_rhw) {
5874                 new_source = FOGSOURCE_COORD;
5875             } else {
5876                 new_source = FOGSOURCE_FFP;
5877             }
5878         }
5879     } else {
5880         new_source = FOGSOURCE_FFP;
5881     }
5882     if(new_source != context->fog_source) {
5883         context->fog_source = new_source;
5884         state_fogstartend(STATE_RENDER(WINED3DRS_FOGSTART), stateblock, context);
5885     }
5886 }
5887
5888 static void textransform(DWORD state, IWineD3DStateBlockImpl *stateblock, struct wined3d_context *context)
5889 {
5890     if(!isStateDirty(context, STATE_PIXELSHADER)) {
5891         fragment_prog_arbfp(state, stateblock, context);
5892     }
5893 }
5894
5895 #undef GLINFO_LOCATION
5896
5897 static const struct StateEntryTemplate arbfp_fragmentstate_template[] = {
5898     {STATE_RENDER(WINED3DRS_TEXTUREFACTOR),               { STATE_RENDER(WINED3DRS_TEXTUREFACTOR),              state_texfactor_arbfp   }, WINED3D_GL_EXT_NONE             },
5899     {STATE_TEXTURESTAGE(0, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5900     {STATE_TEXTURESTAGE(0, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5901     {STATE_TEXTURESTAGE(0, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5902     {STATE_TEXTURESTAGE(0, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5903     {STATE_TEXTURESTAGE(0, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5904     {STATE_TEXTURESTAGE(0, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5905     {STATE_TEXTURESTAGE(0, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5906     {STATE_TEXTURESTAGE(0, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5907     {STATE_TEXTURESTAGE(0, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5908     {STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5909     {STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5910     {STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5911     {STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5912     {STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
5913     {STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
5914     {STATE_TEXTURESTAGE(1, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5915     {STATE_TEXTURESTAGE(1, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5916     {STATE_TEXTURESTAGE(1, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5917     {STATE_TEXTURESTAGE(1, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5918     {STATE_TEXTURESTAGE(1, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5919     {STATE_TEXTURESTAGE(1, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5920     {STATE_TEXTURESTAGE(1, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5921     {STATE_TEXTURESTAGE(1, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5922     {STATE_TEXTURESTAGE(1, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5923     {STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5924     {STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5925     {STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5926     {STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5927     {STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
5928     {STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
5929     {STATE_TEXTURESTAGE(2, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5930     {STATE_TEXTURESTAGE(2, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5931     {STATE_TEXTURESTAGE(2, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5932     {STATE_TEXTURESTAGE(2, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5933     {STATE_TEXTURESTAGE(2, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5934     {STATE_TEXTURESTAGE(2, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5935     {STATE_TEXTURESTAGE(2, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5936     {STATE_TEXTURESTAGE(2, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5937     {STATE_TEXTURESTAGE(2, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5938     {STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5939     {STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5940     {STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5941     {STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5942     {STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
5943     {STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
5944     {STATE_TEXTURESTAGE(3, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5945     {STATE_TEXTURESTAGE(3, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5946     {STATE_TEXTURESTAGE(3, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5947     {STATE_TEXTURESTAGE(3, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5948     {STATE_TEXTURESTAGE(3, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5949     {STATE_TEXTURESTAGE(3, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5950     {STATE_TEXTURESTAGE(3, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5951     {STATE_TEXTURESTAGE(3, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5952     {STATE_TEXTURESTAGE(3, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5953     {STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5954     {STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5955     {STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5956     {STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5957     {STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
5958     {STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
5959     {STATE_TEXTURESTAGE(4, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5960     {STATE_TEXTURESTAGE(4, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5961     {STATE_TEXTURESTAGE(4, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5962     {STATE_TEXTURESTAGE(4, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5963     {STATE_TEXTURESTAGE(4, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5964     {STATE_TEXTURESTAGE(4, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5965     {STATE_TEXTURESTAGE(4, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5966     {STATE_TEXTURESTAGE(4, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5967     {STATE_TEXTURESTAGE(4, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5968     {STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5969     {STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5970     {STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5971     {STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5972     {STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
5973     {STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
5974     {STATE_TEXTURESTAGE(5, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5975     {STATE_TEXTURESTAGE(5, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5976     {STATE_TEXTURESTAGE(5, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5977     {STATE_TEXTURESTAGE(5, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5978     {STATE_TEXTURESTAGE(5, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5979     {STATE_TEXTURESTAGE(5, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5980     {STATE_TEXTURESTAGE(5, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5981     {STATE_TEXTURESTAGE(5, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5982     {STATE_TEXTURESTAGE(5, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5983     {STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5984     {STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5985     {STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5986     {STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5987     {STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
5988     {STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
5989     {STATE_TEXTURESTAGE(6, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5990     {STATE_TEXTURESTAGE(6, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5991     {STATE_TEXTURESTAGE(6, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5992     {STATE_TEXTURESTAGE(6, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5993     {STATE_TEXTURESTAGE(6, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5994     {STATE_TEXTURESTAGE(6, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5995     {STATE_TEXTURESTAGE(6, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5996     {STATE_TEXTURESTAGE(6, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5997     {STATE_TEXTURESTAGE(6, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5998     {STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5999     {STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6000     {STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6001     {STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6002     {STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6003     {STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6004     {STATE_TEXTURESTAGE(7, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6005     {STATE_TEXTURESTAGE(7, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6006     {STATE_TEXTURESTAGE(7, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6007     {STATE_TEXTURESTAGE(7, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6008     {STATE_TEXTURESTAGE(7, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6009     {STATE_TEXTURESTAGE(7, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6010     {STATE_TEXTURESTAGE(7, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6011     {STATE_TEXTURESTAGE(7, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6012     {STATE_TEXTURESTAGE(7, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6013     {STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6014     {STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6015     {STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6016     {STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6017     {STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6018     {STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6019     {STATE_SAMPLER(0),                                    { STATE_SAMPLER(0),                                   sampler_texdim          }, WINED3D_GL_EXT_NONE             },
6020     {STATE_SAMPLER(1),                                    { STATE_SAMPLER(1),                                   sampler_texdim          }, WINED3D_GL_EXT_NONE             },
6021     {STATE_SAMPLER(2),                                    { STATE_SAMPLER(2),                                   sampler_texdim          }, WINED3D_GL_EXT_NONE             },
6022     {STATE_SAMPLER(3),                                    { STATE_SAMPLER(3),                                   sampler_texdim          }, WINED3D_GL_EXT_NONE             },
6023     {STATE_SAMPLER(4),                                    { STATE_SAMPLER(4),                                   sampler_texdim          }, WINED3D_GL_EXT_NONE             },
6024     {STATE_SAMPLER(5),                                    { STATE_SAMPLER(5),                                   sampler_texdim          }, WINED3D_GL_EXT_NONE             },
6025     {STATE_SAMPLER(6),                                    { STATE_SAMPLER(6),                                   sampler_texdim          }, WINED3D_GL_EXT_NONE             },
6026     {STATE_SAMPLER(7),                                    { STATE_SAMPLER(7),                                   sampler_texdim          }, WINED3D_GL_EXT_NONE             },
6027     {STATE_PIXELSHADER,                                   { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6028     {STATE_RENDER(WINED3DRS_FOGENABLE),                   { STATE_RENDER(WINED3DRS_FOGENABLE),                  state_arbfp_fog         }, WINED3D_GL_EXT_NONE             },
6029     {STATE_RENDER(WINED3DRS_FOGTABLEMODE),                { STATE_RENDER(WINED3DRS_FOGENABLE),                  state_arbfp_fog         }, WINED3D_GL_EXT_NONE             },
6030     {STATE_RENDER(WINED3DRS_FOGVERTEXMODE),               { STATE_RENDER(WINED3DRS_FOGENABLE),                  state_arbfp_fog         }, WINED3D_GL_EXT_NONE             },
6031     {STATE_RENDER(WINED3DRS_FOGSTART),                    { STATE_RENDER(WINED3DRS_FOGSTART),                   state_fogstartend       }, WINED3D_GL_EXT_NONE             },
6032     {STATE_RENDER(WINED3DRS_FOGEND),                      { STATE_RENDER(WINED3DRS_FOGSTART),                   state_fogstartend       }, WINED3D_GL_EXT_NONE             },
6033     {STATE_RENDER(WINED3DRS_SRGBWRITEENABLE),             { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6034     {STATE_RENDER(WINED3DRS_FOGCOLOR),                    { STATE_RENDER(WINED3DRS_FOGCOLOR),                   state_fogcolor          }, WINED3D_GL_EXT_NONE             },
6035     {STATE_RENDER(WINED3DRS_FOGDENSITY),                  { STATE_RENDER(WINED3DRS_FOGDENSITY),                 state_fogdensity        }, WINED3D_GL_EXT_NONE             },
6036     {STATE_TEXTURESTAGE(0,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(0, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6037     {STATE_TEXTURESTAGE(1,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(1, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6038     {STATE_TEXTURESTAGE(2,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(2, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6039     {STATE_TEXTURESTAGE(3,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(3, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6040     {STATE_TEXTURESTAGE(4,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(4, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6041     {STATE_TEXTURESTAGE(5,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(5, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6042     {STATE_TEXTURESTAGE(6,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(6, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6043     {STATE_TEXTURESTAGE(7,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(7, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6044     {STATE_RENDER(WINED3DRS_SPECULARENABLE),              { STATE_RENDER(WINED3DRS_SPECULARENABLE),             state_arb_specularenable}, WINED3D_GL_EXT_NONE             },
6045     {0 /* Terminate */,                                   { 0,                                                  0                       }, WINED3D_GL_EXT_NONE             },
6046 };
6047
6048 const struct fragment_pipeline arbfp_fragment_pipeline = {
6049     arbfp_enable,
6050     arbfp_get_caps,
6051     arbfp_alloc,
6052     arbfp_free,
6053     shader_arb_color_fixup_supported,
6054     arbfp_fragmentstate_template,
6055     TRUE /* We can disable projected textures */
6056 };
6057
6058 #define GLINFO_LOCATION device->adapter->gl_info
6059
6060 struct arbfp_blit_priv {
6061     GLenum yuy2_rect_shader, yuy2_2d_shader;
6062     GLenum uyvy_rect_shader, uyvy_2d_shader;
6063     GLenum yv12_rect_shader, yv12_2d_shader;
6064 };
6065
6066 static HRESULT arbfp_blit_alloc(IWineD3DDevice *iface) {
6067     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *) iface;
6068     device->blit_priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(struct arbfp_blit_priv));
6069     if(!device->blit_priv) {
6070         ERR("Out of memory\n");
6071         return E_OUTOFMEMORY;
6072     }
6073     return WINED3D_OK;
6074 }
6075
6076 /* Context activation is done by the caller. */
6077 static void arbfp_blit_free(IWineD3DDevice *iface) {
6078     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *) iface;
6079     struct arbfp_blit_priv *priv = device->blit_priv;
6080
6081     ENTER_GL();
6082     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yuy2_rect_shader));
6083     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yuy2_2d_shader));
6084     GL_EXTCALL(glDeleteProgramsARB(1, &priv->uyvy_rect_shader));
6085     GL_EXTCALL(glDeleteProgramsARB(1, &priv->uyvy_2d_shader));
6086     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yv12_rect_shader));
6087     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yv12_2d_shader));
6088     checkGLcall("Delete yuv programs");
6089     LEAVE_GL();
6090 }
6091
6092 static BOOL gen_planar_yuv_read(struct wined3d_shader_buffer *buffer, enum yuv_fixup yuv_fixup,
6093         GLenum textype, char *luminance)
6094 {
6095     char chroma;
6096     const char *tex, *texinstr;
6097
6098     if (yuv_fixup == YUV_FIXUP_UYVY) {
6099         chroma = 'x';
6100         *luminance = 'w';
6101     } else {
6102         chroma = 'w';
6103         *luminance = 'x';
6104     }
6105     switch(textype) {
6106         case GL_TEXTURE_2D:             tex = "2D";     texinstr = "TXP"; break;
6107         case GL_TEXTURE_RECTANGLE_ARB:  tex = "RECT";   texinstr = "TEX"; break;
6108         default:
6109             /* This is more tricky than just replacing the texture type - we have to navigate
6110              * properly in the texture to find the correct chroma values
6111              */
6112             FIXME("Implement yuv correction for non-2d, non-rect textures\n");
6113             return FALSE;
6114     }
6115
6116     /* First we have to read the chroma values. This means we need at least two pixels(no filtering),
6117      * or 4 pixels(with filtering). To get the unmodified chromas, we have to rid ourselves of the
6118      * filtering when we sample the texture.
6119      *
6120      * These are the rules for reading the chroma:
6121      *
6122      * Even pixel: Cr
6123      * Even pixel: U
6124      * Odd pixel: V
6125      *
6126      * So we have to get the sampling x position in non-normalized coordinates in integers
6127      */
6128     if(textype != GL_TEXTURE_RECTANGLE_ARB) {
6129         shader_addline(buffer, "MUL texcrd.xy, fragment.texcoord[0], size.x;\n");
6130         shader_addline(buffer, "MOV texcrd.w, size.x;\n");
6131     } else {
6132         shader_addline(buffer, "MOV texcrd, fragment.texcoord[0];\n");
6133     }
6134     /* We must not allow filtering between pixel x and x+1, this would mix U and V
6135      * Vertical filtering is ok. However, bear in mind that the pixel center is at
6136      * 0.5, so add 0.5.
6137      */
6138     shader_addline(buffer, "FLR texcrd.x, texcrd.x;\n");
6139     shader_addline(buffer, "ADD texcrd.x, texcrd.x, coef.y;\n");
6140
6141     /* Divide the x coordinate by 0.5 and get the fraction. This gives 0.25 and 0.75 for the
6142      * even and odd pixels respectively
6143      */
6144     shader_addline(buffer, "MUL texcrd2, texcrd, coef.y;\n");
6145     shader_addline(buffer, "FRC texcrd2, texcrd2;\n");
6146
6147     /* Sample Pixel 1 */
6148     shader_addline(buffer, "%s luminance, texcrd, texture[0], %s;\n", texinstr, tex);
6149
6150     /* Put the value into either of the chroma values */
6151     shader_addline(buffer, "SGE temp.x, texcrd2.x, coef.y;\n");
6152     shader_addline(buffer, "MUL chroma.x, luminance.%c, temp.x;\n", chroma);
6153     shader_addline(buffer, "SLT temp.x, texcrd2.x, coef.y;\n");
6154     shader_addline(buffer, "MUL chroma.y, luminance.%c, temp.x;\n", chroma);
6155
6156     /* Sample pixel 2. If we read an even pixel(SLT above returned 1), sample
6157      * the pixel right to the current one. Otherwise, sample the left pixel.
6158      * Bias and scale the SLT result to -1;1 and add it to the texcrd.x.
6159      */
6160     shader_addline(buffer, "MAD temp.x, temp.x, coef.z, -coef.x;\n");
6161     shader_addline(buffer, "ADD texcrd.x, texcrd, temp.x;\n");
6162     shader_addline(buffer, "%s luminance, texcrd, texture[0], %s;\n", texinstr, tex);
6163
6164     /* Put the value into the other chroma */
6165     shader_addline(buffer, "SGE temp.x, texcrd2.x, coef.y;\n");
6166     shader_addline(buffer, "MAD chroma.y, luminance.%c, temp.x, chroma.y;\n", chroma);
6167     shader_addline(buffer, "SLT temp.x, texcrd2.x, coef.y;\n");
6168     shader_addline(buffer, "MAD chroma.x, luminance.%c, temp.x, chroma.x;\n", chroma);
6169
6170     /* TODO: If filtering is enabled, sample a 2nd pair of pixels left or right of
6171      * the current one and lerp the two U and V values
6172      */
6173
6174     /* This gives the correctly filtered luminance value */
6175     shader_addline(buffer, "TEX luminance, fragment.texcoord[0], texture[0], %s;\n", tex);
6176
6177     return TRUE;
6178 }
6179
6180 static BOOL gen_yv12_read(struct wined3d_shader_buffer *buffer, GLenum textype, char *luminance)
6181 {
6182     const char *tex;
6183
6184     switch(textype) {
6185         case GL_TEXTURE_2D:             tex = "2D";     break;
6186         case GL_TEXTURE_RECTANGLE_ARB:  tex = "RECT";   break;
6187         default:
6188             FIXME("Implement yv12 correction for non-2d, non-rect textures\n");
6189             return FALSE;
6190     }
6191
6192     /* YV12 surfaces contain a WxH sized luminance plane, followed by a (W/2)x(H/2)
6193      * V and a (W/2)x(H/2) U plane, each with 8 bit per pixel. So the effective
6194      * bitdepth is 12 bits per pixel. Since the U and V planes have only half the
6195      * pitch of the luminance plane, the packing into the gl texture is a bit
6196      * unfortunate. If the whole texture is interpreted as luminance data it looks
6197      * approximately like this:
6198      *
6199      *        +----------------------------------+----
6200      *        |                                  |
6201      *        |                                  |
6202      *        |                                  |
6203      *        |                                  |
6204      *        |                                  |   2
6205      *        |            LUMINANCE             |   -
6206      *        |                                  |   3
6207      *        |                                  |
6208      *        |                                  |
6209      *        |                                  |
6210      *        |                                  |
6211      *        +----------------+-----------------+----
6212      *        |                |                 |
6213      *        |  U even rows   |  U odd rows     |
6214      *        |                |                 |   1
6215      *        +----------------+------------------   -
6216      *        |                |                 |   3
6217      *        |  V even rows   |  V odd rows     |
6218      *        |                |                 |
6219      *        +----------------+-----------------+----
6220      *        |                |                 |
6221      *        |     0.5        |       0.5       |
6222      *
6223      * So it appears as if there are 4 chroma images, but in fact the odd rows
6224      * in the chroma images are in the same row as the even ones. So its is
6225      * kinda tricky to read
6226      *
6227      * When reading from rectangle textures, keep in mind that the input y coordinates
6228      * go from 0 to d3d_height, whereas the opengl texture height is 1.5 * d3d_height
6229      */
6230     shader_addline(buffer, "PARAM yv12_coef = {%f, %f, %f, %f};\n",
6231             2.0f / 3.0f, 1.0f / 6.0f, (2.0f / 3.0f) + (1.0f / 6.0f), 1.0f / 3.0f);
6232
6233     shader_addline(buffer, "MOV texcrd, fragment.texcoord[0];\n");
6234     /* the chroma planes have only half the width */
6235     shader_addline(buffer, "MUL texcrd.x, texcrd.x, coef.y;\n");
6236
6237     /* The first value is between 2/3 and 5/6th of the texture's height, so scale+bias
6238      * the coordinate. Also read the right side of the image when reading odd lines
6239      *
6240      * Don't forget to clamp the y values in into the range, otherwise we'll get filtering
6241      * bleeding
6242      */
6243     if(textype == GL_TEXTURE_2D) {
6244
6245         shader_addline(buffer, "RCP chroma.w, size.y;\n");
6246
6247         shader_addline(buffer, "MUL texcrd2.y, texcrd.y, size.y;\n");
6248
6249         shader_addline(buffer, "FLR texcrd2.y, texcrd2.y;\n");
6250         shader_addline(buffer, "MAD texcrd.y, texcrd.y, yv12_coef.y, yv12_coef.x;\n");
6251
6252         /* Read odd lines from the right side(add size * 0.5 to the x coordinate */
6253         shader_addline(buffer, "ADD texcrd2.x, texcrd2.y, yv12_coef.y;\n"); /* To avoid 0.5 == 0.5 comparisons */
6254         shader_addline(buffer, "FRC texcrd2.x, texcrd2.x;\n");
6255         shader_addline(buffer, "SGE texcrd2.x, texcrd2.x, coef.y;\n");
6256         shader_addline(buffer, "MAD texcrd.x, texcrd2.x, coef.y, texcrd.x;\n");
6257
6258         /* clamp, keep the half pixel origin in mind */
6259         shader_addline(buffer, "MAD temp.y, coef.y, chroma.w, yv12_coef.x;\n");
6260         shader_addline(buffer, "MAX texcrd.y, temp.y, texcrd.y;\n");
6261         shader_addline(buffer, "MAD temp.y, -coef.y, chroma.w, yv12_coef.z;\n");
6262         shader_addline(buffer, "MIN texcrd.y, temp.y, texcrd.y;\n");
6263     } else {
6264         /* Read from [size - size+size/4] */
6265         shader_addline(buffer, "FLR texcrd.y, texcrd.y;\n");
6266         shader_addline(buffer, "MAD texcrd.y, texcrd.y, coef.w, size.y;\n");
6267
6268         /* Read odd lines from the right side(add size * 0.5 to the x coordinate */
6269         shader_addline(buffer, "ADD texcrd2.x, texcrd.y, yv12_coef.y;\n"); /* To avoid 0.5 == 0.5 comparisons */
6270         shader_addline(buffer, "FRC texcrd2.x, texcrd2.x;\n");
6271         shader_addline(buffer, "SGE texcrd2.x, texcrd2.x, coef.y;\n");
6272         shader_addline(buffer, "MUL texcrd2.x, texcrd2.x, size.x;\n");
6273         shader_addline(buffer, "MAD texcrd.x, texcrd2.x, coef.y, texcrd.x;\n");
6274
6275         /* Make sure to read exactly from the pixel center */
6276         shader_addline(buffer, "FLR texcrd.y, texcrd.y;\n");
6277         shader_addline(buffer, "ADD texcrd.y, texcrd.y, coef.y;\n");
6278
6279         /* Clamp */
6280         shader_addline(buffer, "MAD temp.y, size.y, coef.w, size.y;\n");
6281         shader_addline(buffer, "ADD temp.y, temp.y, -coef.y;\n");
6282         shader_addline(buffer, "MIN texcrd.y, temp.y, texcrd.y;\n");
6283         shader_addline(buffer, "ADD temp.y, size.y, -coef.y;\n");
6284         shader_addline(buffer, "MAX texcrd.y, temp.y, texcrd.y;\n");
6285     }
6286     /* Read the texture, put the result into the output register */
6287     shader_addline(buffer, "TEX temp, texcrd, texture[0], %s;\n", tex);
6288     shader_addline(buffer, "MOV chroma.x, temp.w;\n");
6289
6290     /* The other chroma value is 1/6th of the texture lower, from 5/6th to 6/6th
6291      * No need to clamp because we're just reusing the already clamped value from above
6292      */
6293     if(textype == GL_TEXTURE_2D) {
6294         shader_addline(buffer, "ADD texcrd.y, texcrd.y, yv12_coef.y;\n");
6295     } else {
6296         shader_addline(buffer, "MAD texcrd.y, size.y, coef.w, texcrd.y;\n");
6297     }
6298     shader_addline(buffer, "TEX temp, texcrd, texture[0], %s;\n", tex);
6299     shader_addline(buffer, "MOV chroma.y, temp.w;\n");
6300
6301     /* Sample the luminance value. It is in the top 2/3rd of the texture, so scale the y coordinate.
6302      * Clamp the y coordinate to prevent the chroma values from bleeding into the sampled luminance
6303      * values due to filtering
6304      */
6305     shader_addline(buffer, "MOV texcrd, fragment.texcoord[0];\n");
6306     if(textype == GL_TEXTURE_2D) {
6307         /* Multiply the y coordinate by 2/3 and clamp it */
6308         shader_addline(buffer, "MUL texcrd.y, texcrd.y, yv12_coef.x;\n");
6309         shader_addline(buffer, "MAD temp.y, -coef.y, chroma.w, yv12_coef.x;\n");
6310         shader_addline(buffer, "MIN texcrd.y, temp.y, texcrd.y;\n");
6311         shader_addline(buffer, "TEX luminance, texcrd, texture[0], %s;\n", tex);
6312     } else {
6313         /* Reading from texture_rectangles is pretty straightforward, just use the unmodified
6314          * texture coordinate. It is still a good idea to clamp it though, since the opengl texture
6315          * is bigger
6316          */
6317         shader_addline(buffer, "ADD temp.x, size.y, -coef.y;\n");
6318         shader_addline(buffer, "MIN texcrd.y, texcrd.y, size.x;\n");
6319         shader_addline(buffer, "TEX luminance, texcrd, texture[0], %s;\n", tex);
6320     }
6321     *luminance = 'a';
6322
6323     return TRUE;
6324 }
6325
6326 /* Context activation is done by the caller. */
6327 static GLuint gen_yuv_shader(IWineD3DDeviceImpl *device, enum yuv_fixup yuv_fixup, GLenum textype)
6328 {
6329     GLenum shader;
6330     struct wined3d_shader_buffer buffer;
6331     char luminance_component;
6332     struct arbfp_blit_priv *priv = device->blit_priv;
6333
6334     /* Shader header */
6335     if (!shader_buffer_init(&buffer))
6336     {
6337         ERR("Failed to initialize shader buffer.\n");
6338         return 0;
6339     }
6340
6341     ENTER_GL();
6342     GL_EXTCALL(glGenProgramsARB(1, &shader));
6343     checkGLcall("GL_EXTCALL(glGenProgramsARB(1, &shader))");
6344     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader));
6345     checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader)");
6346     LEAVE_GL();
6347     if(!shader) {
6348         shader_buffer_free(&buffer);
6349         return 0;
6350     }
6351
6352     /* The YUY2 and UYVY formats contain two pixels packed into a 32 bit macropixel,
6353      * giving effectively 16 bit per pixel. The color consists of a luminance(Y) and
6354      * two chroma(U and V) values. Each macropixel has two luminance values, one for
6355      * each single pixel it contains, and one U and one V value shared between both
6356      * pixels.
6357      *
6358      * The data is loaded into an A8L8 texture. With YUY2, the luminance component
6359      * contains the luminance and alpha the chroma. With UYVY it is vice versa. Thus
6360      * take the format into account when generating the read swizzles
6361      *
6362      * Reading the Y value is straightforward - just sample the texture. The hardware
6363      * takes care of filtering in the horizontal and vertical direction.
6364      *
6365      * Reading the U and V values is harder. We have to avoid filtering horizontally,
6366      * because that would mix the U and V values of one pixel or two adjacent pixels.
6367      * Thus floor the texture coordinate and add 0.5 to get an unfiltered read,
6368      * regardless of the filtering setting. Vertical filtering works automatically
6369      * though - the U and V values of two rows are mixed nicely.
6370      *
6371      * Appart of avoiding filtering issues, the code has to know which value it just
6372      * read, and where it can find the other one. To determine this, it checks if
6373      * it sampled an even or odd pixel, and shifts the 2nd read accordingly.
6374      *
6375      * Handling horizontal filtering of U and V values requires reading a 2nd pair
6376      * of pixels, extracting U and V and mixing them. This is not implemented yet.
6377      *
6378      * An alternative implementation idea is to load the texture as A8R8G8B8 texture,
6379      * with width / 2. This way one read gives all 3 values, finding U and V is easy
6380      * in an unfiltered situation. Finding the luminance on the other hand requires
6381      * finding out if it is an odd or even pixel. The real drawback of this approach
6382      * is filtering. This would have to be emulated completely in the shader, reading
6383      * up two 2 packed pixels in up to 2 rows and interpolating both horizontally and
6384      * vertically. Beyond that it would require adjustments to the texture handling
6385      * code to deal with the width scaling
6386      */
6387     shader_addline(&buffer, "!!ARBfp1.0\n");
6388     shader_addline(&buffer, "TEMP luminance;\n");
6389     shader_addline(&buffer, "TEMP temp;\n");
6390     shader_addline(&buffer, "TEMP chroma;\n");
6391     shader_addline(&buffer, "TEMP texcrd;\n");
6392     shader_addline(&buffer, "TEMP texcrd2;\n");
6393     shader_addline(&buffer, "PARAM coef = {1.0, 0.5, 2.0, 0.25};\n");
6394     shader_addline(&buffer, "PARAM yuv_coef = {1.403, 0.344, 0.714, 1.770};\n");
6395     shader_addline(&buffer, "PARAM size = program.local[0];\n");
6396
6397     switch (yuv_fixup)
6398     {
6399         case YUV_FIXUP_UYVY:
6400         case YUV_FIXUP_YUY2:
6401             if (!gen_planar_yuv_read(&buffer, yuv_fixup, textype, &luminance_component))
6402             {
6403                 shader_buffer_free(&buffer);
6404                 return 0;
6405             }
6406             break;
6407
6408         case YUV_FIXUP_YV12:
6409             if (!gen_yv12_read(&buffer, textype, &luminance_component))
6410             {
6411                 shader_buffer_free(&buffer);
6412                 return 0;
6413             }
6414             break;
6415
6416         default:
6417             FIXME("Unsupported YUV fixup %#x\n", yuv_fixup);
6418             shader_buffer_free(&buffer);
6419             return 0;
6420     }
6421
6422     /* Calculate the final result. Formula is taken from
6423      * http://www.fourcc.org/fccyvrgb.php. Note that the chroma
6424      * ranges from -0.5 to 0.5
6425      */
6426     shader_addline(&buffer, "SUB chroma.xy, chroma, coef.y;\n");
6427
6428     shader_addline(&buffer, "MAD result.color.x, chroma.x, yuv_coef.x, luminance.%c;\n", luminance_component);
6429     shader_addline(&buffer, "MAD temp.x, -chroma.y, yuv_coef.y, luminance.%c;\n", luminance_component);
6430     shader_addline(&buffer, "MAD result.color.y, -chroma.x, yuv_coef.z, temp.x;\n");
6431     shader_addline(&buffer, "MAD result.color.z, chroma.y, yuv_coef.w, luminance.%c;\n", luminance_component);
6432     shader_addline(&buffer, "END\n");
6433
6434     ENTER_GL();
6435     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB, strlen(buffer.buffer), buffer.buffer));
6436
6437     if (glGetError() == GL_INVALID_OPERATION) {
6438         GLint pos;
6439         glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
6440         FIXME("Fragment program error at position %d: %s\n", pos,
6441               debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
6442     }
6443     shader_buffer_free(&buffer);
6444     LEAVE_GL();
6445
6446     switch (yuv_fixup)
6447     {
6448         case YUV_FIXUP_YUY2:
6449             if (textype == GL_TEXTURE_RECTANGLE_ARB) priv->yuy2_rect_shader = shader;
6450             else priv->yuy2_2d_shader = shader;
6451             break;
6452
6453         case YUV_FIXUP_UYVY:
6454             if (textype == GL_TEXTURE_RECTANGLE_ARB) priv->uyvy_rect_shader = shader;
6455             else priv->uyvy_2d_shader = shader;
6456             break;
6457
6458         case YUV_FIXUP_YV12:
6459             if (textype == GL_TEXTURE_RECTANGLE_ARB) priv->yv12_rect_shader = shader;
6460             else priv->yv12_2d_shader = shader;
6461             break;
6462     }
6463
6464     return shader;
6465 }
6466
6467 /* Context activation is done by the caller. */
6468 static HRESULT arbfp_blit_set(IWineD3DDevice *iface, const struct GlPixelFormatDesc *format_desc,
6469         GLenum textype, UINT width, UINT height)
6470 {
6471     GLenum shader;
6472     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *) iface;
6473     float size[4] = {width, height, 1, 1};
6474     struct arbfp_blit_priv *priv = device->blit_priv;
6475     enum yuv_fixup yuv_fixup;
6476
6477     if (!is_yuv_fixup(format_desc->color_fixup))
6478     {
6479         TRACE("Fixup:\n");
6480         dump_color_fixup_desc(format_desc->color_fixup);
6481         /* Don't bother setting up a shader for unconverted formats */
6482         ENTER_GL();
6483         glEnable(textype);
6484         checkGLcall("glEnable(textype)");
6485         LEAVE_GL();
6486         return WINED3D_OK;
6487     }
6488
6489     yuv_fixup = get_yuv_fixup(format_desc->color_fixup);
6490
6491     switch(yuv_fixup)
6492     {
6493         case YUV_FIXUP_YUY2:
6494             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->yuy2_rect_shader : priv->yuy2_2d_shader;
6495             break;
6496
6497         case YUV_FIXUP_UYVY:
6498             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->uyvy_rect_shader : priv->uyvy_2d_shader;
6499             break;
6500
6501         case YUV_FIXUP_YV12:
6502             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->yv12_rect_shader : priv->yv12_2d_shader;
6503             break;
6504
6505         default:
6506             FIXME("Unsupported YUV fixup %#x, not setting a shader\n", yuv_fixup);
6507             ENTER_GL();
6508             glEnable(textype);
6509             checkGLcall("glEnable(textype)");
6510             LEAVE_GL();
6511             return E_NOTIMPL;
6512     }
6513
6514     if (!shader) shader = gen_yuv_shader(device, yuv_fixup, textype);
6515
6516     ENTER_GL();
6517     glEnable(GL_FRAGMENT_PROGRAM_ARB);
6518     checkGLcall("glEnable(GL_FRAGMENT_PROGRAM_ARB)");
6519     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader));
6520     checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader)");
6521     GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, 0, size));
6522     checkGLcall("glProgramLocalParameter4fvARB");
6523     LEAVE_GL();
6524
6525     return WINED3D_OK;
6526 }
6527
6528 /* Context activation is done by the caller. */
6529 static void arbfp_blit_unset(IWineD3DDevice *iface) {
6530     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *) iface;
6531
6532     ENTER_GL();
6533     glDisable(GL_FRAGMENT_PROGRAM_ARB);
6534     checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
6535     glDisable(GL_TEXTURE_2D);
6536     checkGLcall("glDisable(GL_TEXTURE_2D)");
6537     if(GL_SUPPORT(ARB_TEXTURE_CUBE_MAP)) {
6538         glDisable(GL_TEXTURE_CUBE_MAP_ARB);
6539         checkGLcall("glDisable(GL_TEXTURE_CUBE_MAP_ARB)");
6540     }
6541     if(GL_SUPPORT(ARB_TEXTURE_RECTANGLE)) {
6542         glDisable(GL_TEXTURE_RECTANGLE_ARB);
6543         checkGLcall("glDisable(GL_TEXTURE_RECTANGLE_ARB)");
6544     }
6545     LEAVE_GL();
6546 }
6547
6548 static BOOL arbfp_blit_color_fixup_supported(struct color_fixup_desc fixup)
6549 {
6550     enum yuv_fixup yuv_fixup;
6551
6552     if (TRACE_ON(d3d_shader) && TRACE_ON(d3d))
6553     {
6554         TRACE("Checking support for fixup:\n");
6555         dump_color_fixup_desc(fixup);
6556     }
6557
6558     if (is_identity_fixup(fixup))
6559     {
6560         TRACE("[OK]\n");
6561         return TRUE;
6562     }
6563
6564     /* We only support YUV conversions. */
6565     if (!is_yuv_fixup(fixup))
6566     {
6567         TRACE("[FAILED]\n");
6568         return FALSE;
6569     }
6570
6571     yuv_fixup = get_yuv_fixup(fixup);
6572     switch(yuv_fixup)
6573     {
6574         case YUV_FIXUP_YUY2:
6575         case YUV_FIXUP_UYVY:
6576         case YUV_FIXUP_YV12:
6577             TRACE("[OK]\n");
6578             return TRUE;
6579
6580         default:
6581             FIXME("Unsupported YUV fixup %#x\n", yuv_fixup);
6582             TRACE("[FAILED]\n");
6583             return FALSE;
6584     }
6585 }
6586
6587 const struct blit_shader arbfp_blit = {
6588     arbfp_blit_alloc,
6589     arbfp_blit_free,
6590     arbfp_blit_set,
6591     arbfp_blit_unset,
6592     arbfp_blit_color_fixup_supported,
6593 };
6594
6595 #undef GLINFO_LOCATION