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     const struct wined3d_context *context = context_get_current();
 432     IWineD3DStateBlockImpl* stateBlock = deviceImpl->stateBlock;
 433     const struct wined3d_gl_info *gl_info = context->gl_info;
 434     unsigned char i;
 435     struct shader_arb_priv *priv = deviceImpl->shader_priv;
 436     const struct arb_ps_compiled_shader *gl_shader = priv->compiled_fprog;
 437
 438     for(i = 0; i < gl_shader->numbumpenvmatconsts; i++)
 439     {
 440         int texunit = gl_shader->bumpenvmatconst[i].texunit;
 441
 442         /* The state manager takes care that this function is always called if the bump env matrix changes */
 443         const float *data = (const float *)&stateBlock->textureState[texunit][WINED3DTSS_BUMPENVMAT00];
 444         GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, gl_shader->bumpenvmatconst[i].const_num, data));
 445
 446         if (gl_shader->luminanceconst[i].const_num != WINED3D_CONST_NUM_UNUSED)
 447         {
 448             /* WINED3DTSS_BUMPENVLSCALE and WINED3DTSS_BUMPENVLOFFSET are next to each other.
 449              * point gl to the scale, and load 4 floats. x = scale, y = offset, z and w are junk, we
 450              * don't care about them. The pointers are valid for sure because the stateblock is bigger.
 451              * (they're WINED3DTSS_TEXTURETRANSFORMFLAGS and WINED3DTSS_ADDRESSW, so most likely 0 or NaN
 452             */
 453             const float *scale = (const float *)&stateBlock->textureState[texunit][WINED3DTSS_BUMPENVLSCALE];
 454             GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, gl_shader->luminanceconst[i].const_num, scale));
 455         }
 456     }
 457     checkGLcall("Load bumpmap consts");
 458
 459     if(gl_shader->ycorrection != WINED3D_CONST_NUM_UNUSED)
 460     {
 461         /* ycorrection.x: Backbuffer height(onscreen) or 0(offscreen).
 462         * ycorrection.y: -1.0(onscreen), 1.0(offscreen)
 463         * ycorrection.z: 1.0
 464         * ycorrection.w: 0.0
 465         */
 466         float val[4];
 467         val[0] = context->render_offscreen ? 0.0f
 468                 : ((IWineD3DSurfaceImpl *) deviceImpl->render_targets[0])->currentDesc.Height;
 469         val[1] = context->render_offscreen ? 1.0f : -1.0f;
 470         val[2] = 1.0f;
 471         val[3] = 0.0f;
 472         GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, gl_shader->ycorrection, val));
 473         checkGLcall("y correction loading");
 474     }
 475
 476     if(gl_shader->num_int_consts == 0) return;
 477
 478     for(i = 0; i < MAX_CONST_I; i++)
 479     {
 480         if(gl_shader->int_consts[i] != WINED3D_CONST_NUM_UNUSED)
 481         {
 482             float val[4];
 483             val[0] = stateBlock->pixelShaderConstantI[4 * i];
 484             val[1] = stateBlock->pixelShaderConstantI[4 * i + 1];
 485             val[2] = stateBlock->pixelShaderConstantI[4 * i + 2];
 486             val[3] = -1.0f;
 487
 488             GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, gl_shader->int_consts[i], val));
 489         }
 490     }
 491     checkGLcall("Load ps int consts");
 492 }
 493
 494 /* GL locking is done by the caller. */
 495 static inline void shader_arb_vs_local_constants(IWineD3DDeviceImpl* deviceImpl)
 496 {
 497     IWineD3DStateBlockImpl* stateBlock;
 498     const struct wined3d_gl_info *gl_info = &deviceImpl->adapter->gl_info;
 499     unsigned char i;
 500     struct shader_arb_priv *priv = deviceImpl->shader_priv;
 501     const struct arb_vs_compiled_shader *gl_shader = priv->compiled_vprog;
 502
 503     /* Upload the position fixup */
 504     GL_EXTCALL(glProgramLocalParameter4fvARB(GL_VERTEX_PROGRAM_ARB, gl_shader->pos_fixup, deviceImpl->posFixup));
 505
 506     if(gl_shader->num_int_consts == 0) return;
 507
 508     stateBlock = deviceImpl->stateBlock;
 509
 510     for(i = 0; i < MAX_CONST_I; i++)
 511     {
 512         if(gl_shader->int_consts[i] != WINED3D_CONST_NUM_UNUSED)
 513         {
 514             float val[4];
 515             val[0] = stateBlock->vertexShaderConstantI[4 * i];
 516             val[1] = stateBlock->vertexShaderConstantI[4 * i + 1];
 517             val[2] = stateBlock->vertexShaderConstantI[4 * i + 2];
 518             val[3] = -1.0f;
 519
 520             GL_EXTCALL(glProgramLocalParameter4fvARB(GL_VERTEX_PROGRAM_ARB, gl_shader->int_consts[i], val));
 521         }
 522     }
 523     checkGLcall("Load vs int consts");
 524 }
 525
 526 /**
 527  * Loads the app-supplied constants into the currently set ARB_[vertex/fragment]_programs.
 528  *
 529  * We only support float constants in ARB at the moment, so don't
 530  * worry about the Integers or Booleans
 531  */
 532 /* GL locking is done by the caller (state handler) */
 533 static void shader_arb_load_constants(const struct wined3d_context *context, char usePixelShader, char useVertexShader)
 534 {
 535     IWineD3DDeviceImpl *device = ((IWineD3DSurfaceImpl *)context->surface)->resource.wineD3DDevice;
 536     IWineD3DStateBlockImpl* stateBlock = device->stateBlock;
 537     const struct wined3d_gl_info *gl_info = context->gl_info;
 538
 539     if (useVertexShader) {
 540         IWineD3DBaseShaderImpl* vshader = (IWineD3DBaseShaderImpl*) stateBlock->vertexShader;
 541
 542         /* Load DirectX 9 float constants for vertex shader */
 543         device->highest_dirty_vs_const = shader_arb_load_constantsF(vshader, gl_info, GL_VERTEX_PROGRAM_ARB,
 544                 device->highest_dirty_vs_const, stateBlock->vertexShaderConstantF, context->vshader_const_dirty);
 545         shader_arb_vs_local_constants(device);
 546     }
 547
 548     if (usePixelShader) {
 549         IWineD3DBaseShaderImpl* pshader = (IWineD3DBaseShaderImpl*) stateBlock->pixelShader;
 550
 551         /* Load DirectX 9 float constants for pixel shader */
 552         device->highest_dirty_ps_const = shader_arb_load_constantsF(pshader, gl_info, GL_FRAGMENT_PROGRAM_ARB,
 553                 device->highest_dirty_ps_const, stateBlock->pixelShaderConstantF, context->pshader_const_dirty);
 554         shader_arb_ps_local_constants(device);
 555     }
 556 }
 557
 558 static void shader_arb_update_float_vertex_constants(IWineD3DDevice *iface, UINT start, UINT count)
 559 {
 560     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
 561     struct wined3d_context *context = context_get_current();
 562
 563     /* We don't want shader constant dirtification to be an O(contexts), so just dirtify the active
 564      * context. On a context switch the old context will be fully dirtified */
 565     if (!context || ((IWineD3DSurfaceImpl *)context->surface)->resource.wineD3DDevice != This) return;
 566
 567     memset(context->vshader_const_dirty + start, 1, sizeof(*context->vshader_const_dirty) * count);
 568     This->highest_dirty_vs_const = max(This->highest_dirty_vs_const, start + count);
 569 }
 570
 571 static void shader_arb_update_float_pixel_constants(IWineD3DDevice *iface, UINT start, UINT count)
 572 {
 573     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
 574     struct wined3d_context *context = context_get_current();
 575
 576     /* We don't want shader constant dirtification to be an O(contexts), so just dirtify the active
 577      * context. On a context switch the old context will be fully dirtified */
 578     if (!context || ((IWineD3DSurfaceImpl *)context->surface)->resource.wineD3DDevice != This) return;
 579
 580     memset(context->pshader_const_dirty + start, 1, sizeof(*context->pshader_const_dirty) * count);
 581     This->highest_dirty_ps_const = max(This->highest_dirty_ps_const, start + count);
 582 }
 583
 584 static DWORD *local_const_mapping(IWineD3DBaseShaderImpl *This)
 585 {
 586     DWORD *ret;
 587     DWORD idx = 0;
 588     const local_constant *lconst;
 589
 590     if(This->baseShader.load_local_constsF || list_empty(&This->baseShader.constantsF)) return NULL;
 591
 592     ret = HeapAlloc(GetProcessHeap(), 0, sizeof(DWORD) * This->baseShader.limits.constant_float);
 593     if(!ret) {
 594         ERR("Out of memory\n");
 595         return NULL;
 596     }
 597
 598     LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry) {
 599         ret[lconst->idx] = idx++;
 600     }
 601     return ret;
 602 }
 603
 604 /* Generate the variable & register declarations for the ARB_vertex_program output target */
 605 static DWORD shader_generate_arb_declarations(IWineD3DBaseShader *iface, const shader_reg_maps *reg_maps,
 606         struct wined3d_shader_buffer *buffer, const struct wined3d_gl_info *gl_info, DWORD *lconst_map,
 607         DWORD *num_clipplanes, struct shader_arb_ctx_priv *ctx)
 608 {
 609     IWineD3DBaseShaderImpl* This = (IWineD3DBaseShaderImpl*) iface;
 610     DWORD i, next_local = 0;
 611     char pshader = shader_is_pshader_version(reg_maps->shader_version.type);
 612     unsigned max_constantsF;
 613     const local_constant *lconst;
 614
 615     /* In pixel shaders, all private constants are program local, we don't need anything
 616      * from program.env. Thus we can advertise the full set of constants in pixel shaders.
 617      * If we need a private constant the GL implementation will squeeze it in somewhere
 618      *
 619      * With vertex shaders we need the posFixup and on some GL implementations 4 helper
 620      * immediate values. The posFixup is loaded using program.env for now, so always
 621      * subtract one from the number of constants. If the shader uses indirect addressing,
 622      * account for the helper const too because we have to declare all availabke d3d constants
 623      * and don't know which are actually used.
 624      */
 625     if(pshader) {
 626         max_constantsF = GL_LIMITS(pshader_constantsF);
 627     } else {
 628         if(This->baseShader.reg_maps.usesrelconstF) {
 629             DWORD highest_constf = 0, clip_limit;
 630             max_constantsF = GL_LIMITS(vshader_constantsF) - reserved_vs_const(iface, gl_info);
 631             max_constantsF -= count_bits(This->baseShader.reg_maps.integer_constants);
 632
 633             for(i = 0; i < This->baseShader.limits.constant_float; i++)
 634             {
 635                 DWORD idx = i >> 5;
 636                 DWORD shift = i & 0x1f;
 637                 if(reg_maps->constf[idx] & (1 << shift)) highest_constf = i;
 638             }
 639
 640             clip_limit = GL_LIMITS(clipplanes);
 641             if(ctx->target_version == ARB) clip_limit = min(clip_limit, 4);
 642             *num_clipplanes = min(clip_limit, max_constantsF - highest_constf - 1);
 643             max_constantsF -= *num_clipplanes;
 644             if(*num_clipplanes < clip_limit)
 645             {
 646                 WARN("Only %u clipplanes out of %u enabled\n", *num_clipplanes, GL_LIMITS(clipplanes));
 647             }
 648         }
 649         else
 650         {
 651             if(ctx->target_version >= NV2) *num_clipplanes = GL_LIMITS(clipplanes);
 652             else *num_clipplanes = min(GL_LIMITS(clipplanes), 4);
 653             max_constantsF = GL_LIMITS(vshader_constantsF);
 654         }
 655     }
 656
 657     for(i = 0; i < This->baseShader.limits.temporary; i++) {
 658         if (reg_maps->temporary[i])
 659             shader_addline(buffer, "TEMP R%u;\n", i);
 660     }
 661
 662     for (i = 0; i < This->baseShader.limits.address; i++) {
 663         if (reg_maps->address[i])
 664             shader_addline(buffer, "ADDRESS A%d;\n", i);
 665     }
 666
 667     if(pshader && reg_maps->shader_version.major == 1 && reg_maps->shader_version.minor <= 3) {
 668         for(i = 0; i < This->baseShader.limits.texcoord; i++) {
 669             if (reg_maps->texcoord[i] && pshader)
 670                 shader_addline(buffer,"TEMP T%u;\n", i);
 671         }
 672     }
 673
 674     /* Load local constants using the program-local space,
 675      * this avoids reloading them each time the shader is used
 676      */
 677     if(lconst_map) {
 678         LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry) {
 679             shader_addline(buffer, "PARAM C%u = program.local[%u];\n", lconst->idx,
 680                            lconst_map[lconst->idx]);
 681             next_local = max(next_local, lconst_map[lconst->idx] + 1);
 682         }
 683     }
 684
 685     /* we use the array-based constants array if the local constants are marked for loading,
 686      * because then we use indirect addressing, or when the local constant list is empty,
 687      * because then we don't know if we're using indirect addressing or not. If we're hardcoding
 688      * local constants do not declare the loaded constants as an array because ARB compilers usually
 689      * do not optimize unused constants away
 690      */
 691     if(This->baseShader.reg_maps.usesrelconstF) {
 692         /* Need to PARAM the environment parameters (constants) so we can use relative addressing */
 693         shader_addline(buffer, "PARAM C[%d] = { program.env[0..%d] };\n",
 694                     max_constantsF, max_constantsF - 1);
 695     } else {
 696         for(i = 0; i < max_constantsF; i++) {
 697             DWORD idx, mask;
 698             idx = i >> 5;
 699             mask = 1 << (i & 0x1f);
 700             if(!shader_constant_is_local(This, i) && (This->baseShader.reg_maps.constf[idx] & mask)) {
 701                 shader_addline(buffer, "PARAM C%d = program.env[%d];\n",i, i);
 702             }
 703         }
 704     }
 705
 706     return next_local;
 707 }
 708
 709 static const char * const shift_tab[] = {
 710     "dummy",     /*  0 (none) */
 711     "coefmul.x", /*  1 (x2)   */
 712     "coefmul.y", /*  2 (x4)   */
 713     "coefmul.z", /*  3 (x8)   */
 714     "coefmul.w", /*  4 (x16)  */
 715     "dummy",     /*  5 (x32)  */
 716     "dummy",     /*  6 (x64)  */
 717     "dummy",     /*  7 (x128) */
 718     "dummy",     /*  8 (d256) */
 719     "dummy",     /*  9 (d128) */
 720     "dummy",     /* 10 (d64)  */
 721     "dummy",     /* 11 (d32)  */
 722     "coefdiv.w", /* 12 (d16)  */
 723     "coefdiv.z", /* 13 (d8)   */
 724     "coefdiv.y", /* 14 (d4)   */
 725     "coefdiv.x"  /* 15 (d2)   */
 726 };
 727
 728 static void shader_arb_get_write_mask(const struct wined3d_shader_instruction *ins,
 729         const struct wined3d_shader_dst_param *dst, char *write_mask)
 730 {
 731     char *ptr = write_mask;
 732
 733     if (dst->write_mask != WINED3DSP_WRITEMASK_ALL)
 734     {
 735         *ptr++ = '.';
 736         if (dst->write_mask & WINED3DSP_WRITEMASK_0) *ptr++ = 'x';
 737         if (dst->write_mask & WINED3DSP_WRITEMASK_1) *ptr++ = 'y';
 738         if (dst->write_mask & WINED3DSP_WRITEMASK_2) *ptr++ = 'z';
 739         if (dst->write_mask & WINED3DSP_WRITEMASK_3) *ptr++ = 'w';
 740     }
 741
 742     *ptr = '\0';
 743 }
 744
 745 static void shader_arb_get_swizzle(const struct wined3d_shader_src_param *param, BOOL fixup, char *swizzle_str)
 746 {
 747     /* For registers of type WINED3DDECLTYPE_D3DCOLOR, data is stored as "bgra",
 748      * but addressed as "rgba". To fix this we need to swap the register's x
 749      * and z components. */
 750     const char *swizzle_chars = fixup ? "zyxw" : "xyzw";
 751     char *ptr = swizzle_str;
 752
 753     /* swizzle bits fields: wwzzyyxx */
 754     DWORD swizzle = param->swizzle;
 755     DWORD swizzle_x = swizzle & 0x03;
 756     DWORD swizzle_y = (swizzle >> 2) & 0x03;
 757     DWORD swizzle_z = (swizzle >> 4) & 0x03;
 758     DWORD swizzle_w = (swizzle >> 6) & 0x03;
 759
 760     /* If the swizzle is the default swizzle (ie, "xyzw"), we don't need to
 761      * generate a swizzle string. Unless we need to our own swizzling. */
 762     if (swizzle != WINED3DSP_NOSWIZZLE || fixup)
 763     {
 764         *ptr++ = '.';
 765         if (swizzle_x == swizzle_y && swizzle_x == swizzle_z && swizzle_x == swizzle_w) {
 766             *ptr++ = swizzle_chars[swizzle_x];
 767         } else {
 768             *ptr++ = swizzle_chars[swizzle_x];
 769             *ptr++ = swizzle_chars[swizzle_y];
 770             *ptr++ = swizzle_chars[swizzle_z];
 771             *ptr++ = swizzle_chars[swizzle_w];
 772         }
 773     }
 774
 775     *ptr = '\0';
 776 }
 777
 778 static void shader_arb_request_a0(const struct wined3d_shader_instruction *ins, const char *src)
 779 {
 780     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
 781     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
 782
 783     if(strcmp(priv->addr_reg, src) == 0) return;
 784
 785     strcpy(priv->addr_reg, src);
 786     shader_addline(buffer, "ARL A0.x, %s;\n", src);
 787 }
 788
 789 static void shader_arb_get_src_param(const struct wined3d_shader_instruction *ins,
 790         const struct wined3d_shader_src_param *src, unsigned int tmpreg, char *outregstr);
 791
 792 static void shader_arb_get_register_name(const struct wined3d_shader_instruction *ins,
 793         const struct wined3d_shader_register *reg, char *register_name, BOOL *is_color)
 794 {
 795     /* oPos, oFog and oPts in D3D */
 796     static const char * const rastout_reg_names[] = {"TMP_OUT", "result.fogcoord", "result.pointsize"};
 797     IWineD3DBaseShaderImpl *This = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
 798     BOOL pshader = shader_is_pshader_version(This->baseShader.reg_maps.shader_version.type);
 799     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
 800
 801     *is_color = FALSE;
 802
 803     switch (reg->type)
 804     {
 805         case WINED3DSPR_TEMP:
 806             sprintf(register_name, "R%u", reg->idx);
 807             break;
 808
 809         case WINED3DSPR_INPUT:
 810             if (pshader)
 811             {
 812                 if(This->baseShader.reg_maps.shader_version.major < 3)
 813                 {
 814                     if (reg->idx == 0) strcpy(register_name, "fragment.color.primary");
 815                     else strcpy(register_name, "fragment.color.secondary");
 816                 }
 817                 else
 818                 {
 819                     if(reg->rel_addr)
 820                     {
 821                         char rel_reg[50];
 822                         shader_arb_get_src_param(ins, reg->rel_addr, 0, rel_reg);
 823
 824                         if(strcmp(rel_reg, "**aL_emul**") == 0)
 825                         {
 826                             DWORD idx = ctx->aL + reg->idx;
 827                             if(idx < MAX_REG_INPUT)
 828                             {
 829                                 strcpy(register_name, ctx->ps_input[idx]);
 830                             }
 831                             else
 832                             {
 833                                 ERR("Pixel shader input register out of bounds: %u\n", idx);
 834                                 sprintf(register_name, "out_of_bounds_%u", idx);
 835                             }
 836                         }
 837                         else if(This->baseShader.reg_maps.input_registers & 0x0300)
 838                         {
 839                             /* There are two ways basically:
 840                              *
 841                              * 1) Use the unrolling code that is used for loop emulation and unroll the loop.
 842                              *    That means trouble if the loop also contains a breakc or if the control values
 843                              *    aren't local constants.
 844                              * 2) Generate an if block that checks if aL.y < 8, == 8 or == 9 and selects the
 845                              *    source dynamically. The trouble is that we cannot simply read aL.y because it
 846                              *    is an ADDRESS register. We could however push it, load .zw with a value and use
 847                              *    ADAC to load the condition code register and pop it again afterwards
 848                              */
 849                             FIXME("Relative input register addressing with more than 8 registers\n");
 850
 851                             /* This is better than nothing for now */
 852                             sprintf(register_name, "fragment.texcoord[%s + %u]", rel_reg, reg->idx);
 853                         }
 854                         else if(ctx->cur_ps_args->super.vp_mode != vertexshader)
 855                         {
 856                             /* This is problematic because we'd have to consult the ctx->ps_input strings
 857                              * for where to find the varying. Some may be "0.0", others can be texcoords or
 858                              * colors. This needs either a pipeline replacement to make the vertex shader feed
 859                              * proper varyings, or loop unrolling
 860                              *
 861                              * For now use the texcoords and hope for the best
 862                              */
 863                             FIXME("Non-vertex shader varying input with indirect addressing\n");
 864                             sprintf(register_name, "fragment.texcoord[%s + %u]", rel_reg, reg->idx);
 865                         }
 866                         else
 867                         {
 868                             /* D3D supports indirect addressing only with aL in loop registers. The loop instruction
 869                              * pulls GL_NV_fragment_program2 in
 870                              */
 871                             sprintf(register_name, "fragment.texcoord[%s + %u]", rel_reg, reg->idx);
 872                         }
 873                     }
 874                     else
 875                     {
 876                         if(reg->idx < MAX_REG_INPUT)
 877                         {
 878                             strcpy(register_name, ctx->ps_input[reg->idx]);
 879                         }
 880                         else
 881                         {
 882                             ERR("Pixel shader input register out of bounds: %u\n", reg->idx);
 883                             sprintf(register_name, "out_of_bounds_%u", reg->idx);
 884                         }
 885                     }
 886                 }
 887             }
 888             else
 889             {
 890                 if (ctx->cur_vs_args->super.swizzle_map & (1 << reg->idx)) *is_color = TRUE;
 891                 sprintf(register_name, "vertex.attrib[%u]", reg->idx);
 892             }
 893             break;
 894
 895         case WINED3DSPR_CONST:
 896             if (!pshader && reg->rel_addr)
 897             {
 898                 BOOL aL = FALSE;
 899                 char rel_reg[50];
 900                 UINT rel_offset = ((IWineD3DVertexShaderImpl *)This)->rel_offset;
 901                 if(This->baseShader.reg_maps.shader_version.major < 2) {
 902                     sprintf(rel_reg, "A0.x");
 903                 } else {
 904                     shader_arb_get_src_param(ins, reg->rel_addr, 0, rel_reg);
 905                     if(ctx->target_version == ARB) {
 906                         if(strcmp(rel_reg, "**aL_emul**") == 0) {
 907                             aL = TRUE;
 908                         } else {
 909                             shader_arb_request_a0(ins, rel_reg);
 910                             sprintf(rel_reg, "A0.x");
 911                         }
 912                     }
 913                 }
 914                 if(aL)
 915                     sprintf(register_name, "C[%u]", ctx->aL + reg->idx);
 916                 else if (reg->idx >= rel_offset)
 917                     sprintf(register_name, "C[%s + %u]", rel_reg, reg->idx - rel_offset);
 918                 else
 919                     sprintf(register_name, "C[%s - %u]", rel_reg, -reg->idx + rel_offset);
 920             }
 921             else
 922             {
 923                 if (This->baseShader.reg_maps.usesrelconstF)
 924                     sprintf(register_name, "C[%u]", reg->idx);
 925                 else
 926                     sprintf(register_name, "C%u", reg->idx);
 927             }
 928             break;
 929
 930         case WINED3DSPR_TEXTURE: /* case WINED3DSPR_ADDR: */
 931             if (pshader) {
 932                 if(This->baseShader.reg_maps.shader_version.major == 1 &&
 933                    This->baseShader.reg_maps.shader_version.minor <= 3) {
 934                     /* In ps <= 1.3, Tx is a temporary register as destination to all instructions,
 935                      * and as source to most instructions. For some instructions it is the texcoord
 936                      * input. Those instructions know about the special use
 937                      */
 938                     sprintf(register_name, "T%u", reg->idx);
 939                 } else {
 940                     /* in ps 1.4 and 2.x Tx is always a (read-only) varying */
 941                     sprintf(register_name, "fragment.texcoord[%u]", reg->idx);
 942                 }
 943             }
 944             else
 945             {
 946                 if(This->baseShader.reg_maps.shader_version.major == 1 || ctx->target_version >= NV2)
 947                 {
 948                     sprintf(register_name, "A%u", reg->idx);
 949                 }
 950                 else
 951                 {
 952                     sprintf(register_name, "A%u_SHADOW", reg->idx);
 953                 }
 954             }
 955             break;
 956
 957         case WINED3DSPR_COLOROUT:
 958             if(ctx->cur_ps_args->super.srgb_correction && reg->idx == 0)
 959             {
 960                 strcpy(register_name, "TMP_COLOR");
 961             }
 962             else
 963             {
 964                 if(ctx->cur_ps_args->super.srgb_correction) FIXME("sRGB correction on higher render targets\n");
 965                 if(This->baseShader.reg_maps.highest_render_target > 0)
 966                 {
 967                     sprintf(register_name, "result.color[%u]", reg->idx);
 968                 }
 969                 else
 970                 {
 971                     strcpy(register_name, "result.color");
 972                 }
 973             }
 974             break;
 975
 976         case WINED3DSPR_RASTOUT:
 977             if(reg->idx == 1) sprintf(register_name, "%s", ctx->fog_output);
 978             else sprintf(register_name, "%s", rastout_reg_names[reg->idx]);
 979             break;
 980
 981         case WINED3DSPR_DEPTHOUT:
 982             strcpy(register_name, "result.depth");
 983             break;
 984
 985         case WINED3DSPR_ATTROUT:
 986         /* case WINED3DSPR_OUTPUT: */
 987             if (pshader) sprintf(register_name, "oD[%u]", reg->idx);
 988             else strcpy(register_name, ctx->color_output[reg->idx]);
 989             break;
 990
 991         case WINED3DSPR_TEXCRDOUT:
 992             if (pshader)
 993             {
 994                 sprintf(register_name, "oT[%u]", reg->idx);
 995             }
 996             else
 997             {
 998                 if(This->baseShader.reg_maps.shader_version.major < 3)
 999                 {
1000                     strcpy(register_name, ctx->texcrd_output[reg->idx]);
1001                 }
1002                 else
1003                 {
1004                     strcpy(register_name, ctx->vs_output[reg->idx]);
1005                 }
1006             }
1007             break;
1008
1009         case WINED3DSPR_LOOP:
1010             if(ctx->target_version >= NV2)
1011             {
1012                 /* Pshader has an implicitly declared loop index counter A0.x that cannot be renamed */
1013                 if(pshader) sprintf(register_name, "A0.x");
1014                 else sprintf(register_name, "aL.y");
1015             }
1016             else
1017             {
1018                 /* Unfortunately this code cannot return the value of ctx->aL here. An immediate value
1019                  * would be valid, but if aL is used for indexing(its only use), there's likely an offset,
1020                  * thus the result would be something like C[15 + 30], which is not valid in the ARB program
1021                  * grammar. So return a marker for the emulated aL and intercept it in constant and varying
1022                  * indexing
1023                  */
1024                 sprintf(register_name, "**aL_emul**");
1025             }
1026
1027             break;
1028
1029         case WINED3DSPR_CONSTINT:
1030             sprintf(register_name, "I%u", reg->idx);
1031             break;
1032
1033         case WINED3DSPR_MISCTYPE:
1034             if(reg->idx == 0)
1035             {
1036                 sprintf(register_name, "vpos");
1037             }
1038             else if(reg->idx == 1)
1039             {
1040                 sprintf(register_name, "fragment.facing.x");
1041             }
1042             else
1043             {
1044                 FIXME("Unknown MISCTYPE register index %u\n", reg->idx);
1045             }
1046             break;
1047
1048         default:
1049             FIXME("Unhandled register type %#x[%u]\n", reg->type, reg->idx);
1050             sprintf(register_name, "unrecognized_register[%u]", reg->idx);
1051             break;
1052     }
1053 }
1054
1055 static void shader_arb_get_dst_param(const struct wined3d_shader_instruction *ins,
1056         const struct wined3d_shader_dst_param *wined3d_dst, char *str)
1057 {
1058     char register_name[255];
1059     char write_mask[6];
1060     BOOL is_color;
1061
1062     shader_arb_get_register_name(ins, &wined3d_dst->reg, register_name, &is_color);
1063     strcpy(str, register_name);
1064
1065     shader_arb_get_write_mask(ins, wined3d_dst, write_mask);
1066     strcat(str, write_mask);
1067 }
1068
1069 static const char *shader_arb_get_fixup_swizzle(enum fixup_channel_source channel_source)
1070 {
1071     switch(channel_source)
1072     {
1073         case CHANNEL_SOURCE_ZERO: return "0";
1074         case CHANNEL_SOURCE_ONE: return "1";
1075         case CHANNEL_SOURCE_X: return "x";
1076         case CHANNEL_SOURCE_Y: return "y";
1077         case CHANNEL_SOURCE_Z: return "z";
1078         case CHANNEL_SOURCE_W: return "w";
1079         default:
1080             FIXME("Unhandled channel source %#x\n", channel_source);
1081             return "undefined";
1082     }
1083 }
1084
1085 static void gen_color_correction(struct wined3d_shader_buffer *buffer, const char *reg,
1086         DWORD dst_mask, const char *one, const char *two, struct color_fixup_desc fixup)
1087 {
1088     DWORD mask;
1089
1090     if (is_yuv_fixup(fixup))
1091     {
1092         enum yuv_fixup yuv_fixup = get_yuv_fixup(fixup);
1093         FIXME("YUV fixup (%#x) not supported\n", yuv_fixup);
1094         return;
1095     }
1096
1097     mask = 0;
1098     if (fixup.x_source != CHANNEL_SOURCE_X) mask |= WINED3DSP_WRITEMASK_0;
1099     if (fixup.y_source != CHANNEL_SOURCE_Y) mask |= WINED3DSP_WRITEMASK_1;
1100     if (fixup.z_source != CHANNEL_SOURCE_Z) mask |= WINED3DSP_WRITEMASK_2;
1101     if (fixup.w_source != CHANNEL_SOURCE_W) mask |= WINED3DSP_WRITEMASK_3;
1102     mask &= dst_mask;
1103
1104     if (mask)
1105     {
1106         shader_addline(buffer, "SWZ %s, %s, %s, %s, %s, %s;\n", reg, reg,
1107                 shader_arb_get_fixup_swizzle(fixup.x_source), shader_arb_get_fixup_swizzle(fixup.y_source),
1108                 shader_arb_get_fixup_swizzle(fixup.z_source), shader_arb_get_fixup_swizzle(fixup.w_source));
1109     }
1110
1111     mask = 0;
1112     if (fixup.x_sign_fixup) mask |= WINED3DSP_WRITEMASK_0;
1113     if (fixup.y_sign_fixup) mask |= WINED3DSP_WRITEMASK_1;
1114     if (fixup.z_sign_fixup) mask |= WINED3DSP_WRITEMASK_2;
1115     if (fixup.w_sign_fixup) mask |= WINED3DSP_WRITEMASK_3;
1116     mask &= dst_mask;
1117
1118     if (mask)
1119     {
1120         char reg_mask[6];
1121         char *ptr = reg_mask;
1122
1123         if (mask != WINED3DSP_WRITEMASK_ALL)
1124         {
1125             *ptr++ = '.';
1126             if (mask & WINED3DSP_WRITEMASK_0) *ptr++ = 'x';
1127             if (mask & WINED3DSP_WRITEMASK_1) *ptr++ = 'y';
1128             if (mask & WINED3DSP_WRITEMASK_2) *ptr++ = 'z';
1129             if (mask & WINED3DSP_WRITEMASK_3) *ptr++ = 'w';
1130         }
1131         *ptr = '\0';
1132
1133         shader_addline(buffer, "MAD %s%s, %s, %s, -%s;\n", reg, reg_mask, reg, two, one);
1134     }
1135 }
1136
1137 static const char *shader_arb_get_modifier(const struct wined3d_shader_instruction *ins)
1138 {
1139     DWORD mod;
1140     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
1141     if (!ins->dst_count) return "";
1142
1143     mod = ins->dst[0].modifiers;
1144
1145     /* Silently ignore PARTIALPRECISION if its not supported */
1146     if(priv->target_version == ARB) mod &= ~WINED3DSPDM_PARTIALPRECISION;
1147
1148     if(mod & WINED3DSPDM_MSAMPCENTROID)
1149     {
1150         FIXME("Unhandled modifier WINED3DSPDM_MSAMPCENTROID\n");
1151         mod &= ~WINED3DSPDM_MSAMPCENTROID;
1152     }
1153
1154     switch(mod)
1155     {
1156         case WINED3DSPDM_SATURATE | WINED3DSPDM_PARTIALPRECISION:
1157             return "H_SAT";
1158
1159         case WINED3DSPDM_SATURATE:
1160             return "_SAT";
1161
1162         case WINED3DSPDM_PARTIALPRECISION:
1163             return "H";
1164
1165         case 0:
1166             return "";
1167
1168         default:
1169             FIXME("Unknown modifiers 0x%08x\n", mod);
1170             return "";
1171     }
1172 }
1173
1174 #define TEX_PROJ        0x1
1175 #define TEX_BIAS        0x2
1176 #define TEX_LOD         0x4
1177 #define TEX_DERIV       0x10
1178
1179 static void shader_hw_sample(const struct wined3d_shader_instruction *ins, DWORD sampler_idx,
1180         const char *dst_str, const char *coord_reg, WORD flags, const char *dsx, const char *dsy)
1181 {
1182     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1183     DWORD sampler_type = ins->ctx->reg_maps->sampler_type[sampler_idx];
1184     const char *tex_type;
1185     BOOL np2_fixup = FALSE;
1186     IWineD3DBaseShaderImpl *This = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
1187     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *) This->baseShader.device;
1188     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
1189     const char *mod;
1190     BOOL pshader = shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type);
1191
1192     /* D3D vertex shader sampler IDs are vertex samplers(0-3), not global d3d samplers */
1193     if(!pshader) sampler_idx += MAX_FRAGMENT_SAMPLERS;
1194
1195     switch(sampler_type) {
1196         case WINED3DSTT_1D:
1197             tex_type = "1D";
1198             break;
1199
1200         case WINED3DSTT_2D:
1201             if(device->stateBlock->textures[sampler_idx] &&
1202                IWineD3DBaseTexture_GetTextureDimensions(device->stateBlock->textures[sampler_idx]) == GL_TEXTURE_RECTANGLE_ARB) {
1203                 tex_type = "RECT";
1204             } else {
1205                 tex_type = "2D";
1206             }
1207             if (shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type))
1208             {
1209                 if (priv->cur_np2fixup_info->super.active & (1 << sampler_idx))
1210                 {
1211                     if (flags) FIXME("Only ordinary sampling from NP2 textures is supported.\n");
1212                     else np2_fixup = TRUE;
1213                 }
1214             }
1215             break;
1216
1217         case WINED3DSTT_VOLUME:
1218             tex_type = "3D";
1219             break;
1220
1221         case WINED3DSTT_CUBE:
1222             tex_type = "CUBE";
1223             break;
1224
1225         default:
1226             ERR("Unexpected texture type %d\n", sampler_type);
1227             tex_type = "";
1228     }
1229
1230     /* TEX, TXL, TXD and TXP do not support the "H" modifier,
1231      * so don't use shader_arb_get_modifier
1232      */
1233     if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE) mod = "_SAT";
1234     else mod = "";
1235
1236     /* Fragment samplers always have indentity mapping */
1237     if(sampler_idx >= MAX_FRAGMENT_SAMPLERS)
1238     {
1239         sampler_idx = priv->cur_vs_args->vertex_samplers[sampler_idx - MAX_FRAGMENT_SAMPLERS];
1240     }
1241
1242     if (flags & TEX_DERIV)
1243     {
1244         if(flags & TEX_PROJ) FIXME("Projected texture sampling with custom derivates\n");
1245         if(flags & TEX_BIAS) FIXME("Biased texture sampling with custom derivates\n");
1246         shader_addline(buffer, "TXD%s %s, %s, %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg,
1247                        dsx, dsy,sampler_idx, tex_type);
1248     }
1249     else if(flags & TEX_LOD)
1250     {
1251         if(flags & TEX_PROJ) FIXME("Projected texture sampling with explicit lod\n");
1252         if(flags & TEX_BIAS) FIXME("Biased texture sampling with explicit lod\n");
1253         shader_addline(buffer, "TXL%s %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg,
1254                        sampler_idx, tex_type);
1255     }
1256     else if (flags & TEX_BIAS)
1257     {
1258         /* Shouldn't be possible, but let's check for it */
1259         if(flags & TEX_PROJ) FIXME("Biased and Projected texture sampling\n");
1260         /* TXB takes the 4th component of the source vector automatically, as d3d. Nothing more to do */
1261         shader_addline(buffer, "TXB%s %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg, sampler_idx, tex_type);
1262     }
1263     else if (flags & TEX_PROJ)
1264     {
1265         shader_addline(buffer, "TXP%s %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg, sampler_idx, tex_type);
1266     }
1267     else
1268     {
1269         if (np2_fixup)
1270         {
1271             const unsigned char idx = priv->cur_np2fixup_info->super.idx[sampler_idx];
1272             shader_addline(buffer, "MUL TA, np2fixup[%u].%s, %s;\n", idx >> 1,
1273                            (idx % 2) ? "zwxy" : "xyzw", coord_reg);
1274
1275             shader_addline(buffer, "TEX%s %s, TA, texture[%u], %s;\n", mod, dst_str, sampler_idx, tex_type);
1276         }
1277         else
1278             shader_addline(buffer, "TEX%s %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg, sampler_idx, tex_type);
1279     }
1280
1281     if (pshader)
1282     {
1283         gen_color_correction(buffer, dst_str, ins->dst[0].write_mask,
1284                 "one", "coefmul.x", priv->cur_ps_args->super.color_fixup[sampler_idx]);
1285     }
1286 }
1287
1288 static void shader_arb_get_src_param(const struct wined3d_shader_instruction *ins,
1289         const struct wined3d_shader_src_param *src, unsigned int tmpreg, char *outregstr)
1290 {
1291     /* Generate a line that does the input modifier computation and return the input register to use */
1292     BOOL is_color = FALSE;
1293     char regstr[256];
1294     char swzstr[20];
1295     int insert_line;
1296     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1297     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
1298
1299     /* Assume a new line will be added */
1300     insert_line = 1;
1301
1302     /* Get register name */
1303     shader_arb_get_register_name(ins, &src->reg, regstr, &is_color);
1304     shader_arb_get_swizzle(src, is_color, swzstr);
1305
1306     switch (src->modifiers)
1307     {
1308     case WINED3DSPSM_NONE:
1309         sprintf(outregstr, "%s%s", regstr, swzstr);
1310         insert_line = 0;
1311         break;
1312     case WINED3DSPSM_NEG:
1313         sprintf(outregstr, "-%s%s", regstr, swzstr);
1314         insert_line = 0;
1315         break;
1316     case WINED3DSPSM_BIAS:
1317         shader_addline(buffer, "ADD T%c, %s, -coefdiv.x;\n", 'A' + tmpreg, regstr);
1318         break;
1319     case WINED3DSPSM_BIASNEG:
1320         shader_addline(buffer, "ADD T%c, -%s, coefdiv.x;\n", 'A' + tmpreg, regstr);
1321         break;
1322     case WINED3DSPSM_SIGN:
1323         shader_addline(buffer, "MAD T%c, %s, coefmul.x, -one.x;\n", 'A' + tmpreg, regstr);
1324         break;
1325     case WINED3DSPSM_SIGNNEG:
1326         shader_addline(buffer, "MAD T%c, %s, -coefmul.x, one.x;\n", 'A' + tmpreg, regstr);
1327         break;
1328     case WINED3DSPSM_COMP:
1329         shader_addline(buffer, "SUB T%c, one.x, %s;\n", 'A' + tmpreg, regstr);
1330         break;
1331     case WINED3DSPSM_X2:
1332         shader_addline(buffer, "ADD T%c, %s, %s;\n", 'A' + tmpreg, regstr, regstr);
1333         break;
1334     case WINED3DSPSM_X2NEG:
1335         shader_addline(buffer, "ADD T%c, -%s, -%s;\n", 'A' + tmpreg, regstr, regstr);
1336         break;
1337     case WINED3DSPSM_DZ:
1338         shader_addline(buffer, "RCP T%c, %s.z;\n", 'A' + tmpreg, regstr);
1339         shader_addline(buffer, "MUL T%c, %s, T%c;\n", 'A' + tmpreg, regstr, 'A' + tmpreg);
1340         break;
1341     case WINED3DSPSM_DW:
1342         shader_addline(buffer, "RCP T%c, %s.w;\n", 'A' + tmpreg, regstr);
1343         shader_addline(buffer, "MUL T%c, %s, T%c;\n", 'A' + tmpreg, regstr, 'A' + tmpreg);
1344         break;
1345     case WINED3DSPSM_ABS:
1346         if(ctx->target_version >= NV2) {
1347             sprintf(outregstr, "|%s%s|", regstr, swzstr);
1348             insert_line = 0;
1349         } else {
1350             shader_addline(buffer, "ABS T%c, %s;\n", 'A' + tmpreg, regstr);
1351         }
1352         break;
1353     case WINED3DSPSM_ABSNEG:
1354         if(ctx->target_version >= NV2) {
1355             sprintf(outregstr, "-|%s%s|", regstr, swzstr);
1356         } else {
1357             shader_addline(buffer, "ABS T%c, %s;\n", 'A' + tmpreg, regstr);
1358             sprintf(outregstr, "-T%c%s", 'A' + tmpreg, swzstr);
1359         }
1360         insert_line = 0;
1361         break;
1362     default:
1363         sprintf(outregstr, "%s%s", regstr, swzstr);
1364         insert_line = 0;
1365     }
1366
1367     /* Return modified or original register, with swizzle */
1368     if (insert_line)
1369         sprintf(outregstr, "T%c%s", 'A' + tmpreg, swzstr);
1370 }
1371
1372 static void pshader_hw_bem(const struct wined3d_shader_instruction *ins)
1373 {
1374     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1375     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1376     char dst_name[50];
1377     char src_name[2][50];
1378     DWORD sampler_code = dst->reg.idx;
1379
1380     shader_arb_get_dst_param(ins, dst, dst_name);
1381
1382     /* Sampling the perturbation map in Tsrc was done already, including the signedness correction if needed
1383      *
1384      * Keep in mind that src_name[1] can be "TB" and src_name[0] can be "TA" because modifiers like _x2 are valid
1385      * with bem. So delay loading the first parameter until after the perturbation calculation which needs two
1386      * temps is done.
1387      */
1388     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1389     shader_addline(buffer, "SWZ TA, bumpenvmat%d, x, z, 0, 0;\n", sampler_code);
1390     shader_addline(buffer, "DP3 TC.r, TA, %s;\n", src_name[1]);
1391     shader_addline(buffer, "SWZ TA, bumpenvmat%d, y, w, 0, 0;\n", sampler_code);
1392     shader_addline(buffer, "DP3 TC.g, TA, %s;\n", src_name[1]);
1393
1394     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
1395     shader_addline(buffer, "ADD %s, %s, TC;\n", dst_name, src_name[0]);
1396 }
1397
1398 static DWORD negate_modifiers(DWORD mod, char *extra_char)
1399 {
1400     *extra_char = ' ';
1401     switch(mod)
1402     {
1403         case WINED3DSPSM_NONE:      return WINED3DSPSM_NEG;
1404         case WINED3DSPSM_NEG:       return WINED3DSPSM_NONE;
1405         case WINED3DSPSM_BIAS:      return WINED3DSPSM_BIASNEG;
1406         case WINED3DSPSM_BIASNEG:   return WINED3DSPSM_BIAS;
1407         case WINED3DSPSM_SIGN:      return WINED3DSPSM_SIGNNEG;
1408         case WINED3DSPSM_SIGNNEG:   return WINED3DSPSM_SIGN;
1409         case WINED3DSPSM_COMP:      *extra_char = '-'; return WINED3DSPSM_COMP;
1410         case WINED3DSPSM_X2:        return WINED3DSPSM_X2NEG;
1411         case WINED3DSPSM_X2NEG:     return WINED3DSPSM_X2;
1412         case WINED3DSPSM_DZ:        *extra_char = '-'; return WINED3DSPSM_DZ;
1413         case WINED3DSPSM_DW:        *extra_char = '-'; return WINED3DSPSM_DW;
1414         case WINED3DSPSM_ABS:       return WINED3DSPSM_ABSNEG;
1415         case WINED3DSPSM_ABSNEG:    return WINED3DSPSM_ABS;
1416     }
1417     FIXME("Unknown modifier %u\n", mod);
1418     return mod;
1419 }
1420
1421 static void pshader_hw_cnd(const struct wined3d_shader_instruction *ins)
1422 {
1423     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1424     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1425     char dst_name[50];
1426     char src_name[3][50];
1427     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
1428             ins->ctx->reg_maps->shader_version.minor);
1429     BOOL is_color;
1430
1431     shader_arb_get_dst_param(ins, dst, dst_name);
1432     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1433
1434     /* The coissue flag changes the semantic of the cnd instruction in <= 1.3 shaders */
1435     if (shader_version <= WINED3D_SHADER_VERSION(1, 3) && ins->coissue)
1436     {
1437         shader_addline(buffer, "MOV%s %s, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name[1]);
1438     } else {
1439         struct wined3d_shader_src_param src0_copy = ins->src[0];
1440         char extra_neg;
1441
1442         /* src0 may have a negate srcmod set, so we can't blindly add "-" to the name */
1443         src0_copy.modifiers = negate_modifiers(src0_copy.modifiers, &extra_neg);
1444
1445         shader_arb_get_src_param(ins, &src0_copy, 0, src_name[0]);
1446         shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
1447         shader_addline(buffer, "ADD TA, %c%s, coefdiv.x;\n", extra_neg, src_name[0]);
1448         /* No modifiers supported on CMP */
1449         shader_addline(buffer, "CMP %s, TA, %s, %s;\n", dst_name, src_name[1], src_name[2]);
1450
1451         /* _SAT on CMP doesn't make much sense, but it is not a pure NOP */
1452         if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE)
1453         {
1454             shader_arb_get_register_name(ins, &dst->reg, src_name[0], &is_color);
1455             shader_addline(buffer, "MOV_SAT %s, %s;\n", dst_name, dst_name);
1456         }
1457     }
1458 }
1459
1460 static void pshader_hw_cmp(const struct wined3d_shader_instruction *ins)
1461 {
1462     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1463     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1464     char dst_name[50];
1465     char src_name[3][50];
1466     BOOL is_color;
1467
1468     shader_arb_get_dst_param(ins, dst, dst_name);
1469
1470     /* Generate input register names (with modifiers) */
1471     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
1472     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1473     shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
1474
1475     /* No modifiers are supported on CMP */
1476     shader_addline(buffer, "CMP %s, %s, %s, %s;\n", dst_name,
1477                    src_name[0], src_name[2], src_name[1]);
1478
1479     if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE)
1480     {
1481         shader_arb_get_register_name(ins, &dst->reg, src_name[0], &is_color);
1482         shader_addline(buffer, "MOV_SAT %s, %s;\n", dst_name, src_name[0]);
1483     }
1484 }
1485
1486 /** Process the WINED3DSIO_DP2ADD instruction in ARB.
1487  * dst = dot2(src0, src1) + src2 */
1488 static void pshader_hw_dp2add(const struct wined3d_shader_instruction *ins)
1489 {
1490     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1491     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1492     char dst_name[50];
1493     char src_name[3][50];
1494     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
1495
1496     shader_arb_get_dst_param(ins, dst, dst_name);
1497     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
1498     shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
1499
1500     if(ctx->target_version >= NV3)
1501     {
1502         /* GL_NV_fragment_program2 has a 1:1 matching instruction */
1503         shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1504         shader_addline(buffer, "DP2A%s %s, %s, %s, %s;\n", shader_arb_get_modifier(ins),
1505                        dst_name, src_name[0], src_name[1], src_name[2]);
1506     }
1507     else if(ctx->target_version >= NV2)
1508     {
1509         /* dst.x = src2.?, src0.x, src1.x + src0.y * src1.y
1510          * dst.y = src2.?, src0.x, src1.z + src0.y * src1.w
1511          * dst.z = src2.?, src0.x, src1.x + src0.y * src1.y
1512          * dst.z = src2.?, src0.x, src1.z + src0.y * src1.w
1513          *
1514          * Make sure that src1.zw = src1.xy, then we get a classic dp2add
1515          *
1516          * .xyxy and other swizzles that we could get with this are not valid in
1517          * plain ARBfp, but luckily the NV extension grammar lifts this limitation.
1518          */
1519         struct wined3d_shader_src_param tmp_param = ins->src[1];
1520         DWORD swizzle = tmp_param.swizzle & 0xf; /* Selects .xy */
1521         tmp_param.swizzle = swizzle | (swizzle << 4); /* Creates .xyxy */
1522
1523         shader_arb_get_src_param(ins, &tmp_param, 1, src_name[1]);
1524
1525         shader_addline(buffer, "X2D%s %s, %s, %s, %s;\n", shader_arb_get_modifier(ins),
1526                        dst_name, src_name[2], src_name[0], src_name[1]);
1527     }
1528     else
1529     {
1530         shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1531         /* 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]
1532         * src_name[0] can be TA, but TA is a private temp for modifiers, so it is save to overwrite
1533         */
1534         shader_addline(buffer, "MOV TA, %s;\n", src_name[0]);
1535         shader_addline(buffer, "MOV TA.z, 0.0;\n");
1536         shader_addline(buffer, "DP3 TA, TA, %s;\n", src_name[1]);
1537         shader_addline(buffer, "ADD%s %s, TA, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name[2]);
1538     }
1539 }
1540
1541 /* Map the opcode 1-to-1 to the GL code */
1542 static void shader_hw_map2gl(const struct wined3d_shader_instruction *ins)
1543 {
1544     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1545     const char *instruction;
1546     char arguments[256], dst_str[50];
1547     unsigned int i;
1548     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1549
1550     switch (ins->handler_idx)
1551     {
1552         case WINED3DSIH_ABS: instruction = "ABS"; break;
1553         case WINED3DSIH_ADD: instruction = "ADD"; break;
1554         case WINED3DSIH_CRS: instruction = "XPD"; break;
1555         case WINED3DSIH_DP3: instruction = "DP3"; break;
1556         case WINED3DSIH_DP4: instruction = "DP4"; break;
1557         case WINED3DSIH_DST: instruction = "DST"; break;
1558         case WINED3DSIH_FRC: instruction = "FRC"; break;
1559         case WINED3DSIH_LIT: instruction = "LIT"; break;
1560         case WINED3DSIH_LRP: instruction = "LRP"; break;
1561         case WINED3DSIH_MAD: instruction = "MAD"; break;
1562         case WINED3DSIH_MAX: instruction = "MAX"; break;
1563         case WINED3DSIH_MIN: instruction = "MIN"; break;
1564         case WINED3DSIH_MOV: instruction = "MOV"; break;
1565         case WINED3DSIH_MUL: instruction = "MUL"; break;
1566         case WINED3DSIH_SGE: instruction = "SGE"; break;
1567         case WINED3DSIH_SLT: instruction = "SLT"; break;
1568         case WINED3DSIH_SUB: instruction = "SUB"; break;
1569         case WINED3DSIH_MOVA:instruction = "ARR"; break;
1570         case WINED3DSIH_SGN: instruction = "SSG"; break;
1571         case WINED3DSIH_DSX: instruction = "DDX"; break;
1572         default: instruction = "";
1573             FIXME("Unhandled opcode %#x\n", ins->handler_idx);
1574             break;
1575     }
1576
1577     /* Note that shader_arb_add_dst_param() adds spaces. */
1578     arguments[0] = '\0';
1579     shader_arb_get_dst_param(ins, dst, dst_str);
1580     for (i = 0; i < ins->src_count; ++i)
1581     {
1582         char operand[100];
1583         strcat(arguments, ", ");
1584         shader_arb_get_src_param(ins, &ins->src[i], i, operand);
1585         strcat(arguments, operand);
1586     }
1587     shader_addline(buffer, "%s%s %s%s;\n", instruction, shader_arb_get_modifier(ins), dst_str, arguments);
1588 }
1589
1590 static void shader_hw_nop(const struct wined3d_shader_instruction *ins)
1591 {
1592     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1593     shader_addline(buffer, "NOP;\n");
1594 }
1595
1596 static void shader_hw_mov(const struct wined3d_shader_instruction *ins)
1597 {
1598     IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
1599     BOOL pshader = shader_is_pshader_version(shader->baseShader.reg_maps.shader_version.type);
1600     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
1601
1602     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1603     char src0_param[256];
1604
1605     if(ins->handler_idx == WINED3DSIH_MOVA) {
1606         char write_mask[6];
1607
1608         if(ctx->target_version >= NV2) {
1609             shader_hw_map2gl(ins);
1610             return;
1611         }
1612         shader_arb_get_src_param(ins, &ins->src[0], 0, src0_param);
1613         shader_arb_get_write_mask(ins, &ins->dst[0], write_mask);
1614
1615         /* This implements the mova formula used in GLSL. The first two instructions
1616          * prepare the sign() part. Note that it is fine to have my_sign(0.0) = 1.0
1617          * in this case:
1618          * mova A0.x, 0.0
1619          *
1620          * A0.x = arl(floor(abs(0.0) + 0.5) * 1.0) = floor(0.5) = 0.0 since arl does a floor
1621          *
1622          * The ARL is performed when A0 is used - the requested component is read from A0_SHADOW into
1623          * A0.x. We can use the overwritten component of A0_shadow as temporary storage for the sign.
1624          */
1625         shader_addline(buffer, "SGE A0_SHADOW%s, %s, mova_const.y;\n", write_mask, src0_param);
1626         shader_addline(buffer, "MAD A0_SHADOW%s, A0_SHADOW, mova_const.z, -mova_const.w;\n", write_mask);
1627
1628         shader_addline(buffer, "ABS TA%s, %s;\n", write_mask, src0_param);
1629         shader_addline(buffer, "ADD TA%s, TA, mova_const.x;\n", write_mask);
1630         shader_addline(buffer, "FLR TA%s, TA;\n", write_mask);
1631         if (((IWineD3DVertexShaderImpl *)shader)->rel_offset)
1632         {
1633             shader_addline(buffer, "ADD TA%s, TA, helper_const.z;\n", write_mask);
1634         }
1635         shader_addline(buffer, "MUL A0_SHADOW%s, TA, A0_SHADOW;\n", write_mask);
1636
1637         ((struct shader_arb_ctx_priv *)ins->ctx->backend_data)->addr_reg[0] = '\0';
1638     } else if (ins->ctx->reg_maps->shader_version.major == 1
1639           && !shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type)
1640           && ins->dst[0].reg.type == WINED3DSPR_ADDR)
1641     {
1642         src0_param[0] = '\0';
1643         if (((IWineD3DVertexShaderImpl *)shader)->rel_offset)
1644         {
1645             shader_arb_get_src_param(ins, &ins->src[0], 0, src0_param);
1646             shader_addline(buffer, "ADD TA.x, %s, helper_const.z;\n", src0_param);
1647             shader_addline(buffer, "ARL A0.x, TA.x;\n");
1648         }
1649         else
1650         {
1651             /* Apple's ARB_vertex_program implementation does not accept an ARL source argument
1652              * with more than one component. Thus replicate the first source argument over all
1653              * 4 components. For example, .xyzw -> .x (or better: .xxxx), .zwxy -> .z, etc) */
1654             struct wined3d_shader_src_param tmp_src = ins->src[0];
1655             tmp_src.swizzle = (tmp_src.swizzle & 0x3) * 0x55;
1656             shader_arb_get_src_param(ins, &tmp_src, 0, src0_param);
1657             shader_addline(buffer, "ARL A0.x, %s;\n", src0_param);
1658         }
1659     }
1660     else if(ins->dst[0].reg.type == WINED3DSPR_COLOROUT && ins->dst[0].reg.idx == 0 && pshader)
1661     {
1662         IWineD3DPixelShaderImpl *ps = (IWineD3DPixelShaderImpl *) shader;
1663         if(ctx->cur_ps_args->super.srgb_correction && ps->color0_mov)
1664         {
1665             shader_addline(buffer, "#mov handled in srgb write code\n");
1666             return;
1667         }
1668         shader_hw_map2gl(ins);
1669     }
1670     else
1671     {
1672         shader_hw_map2gl(ins);
1673     }
1674 }
1675
1676 static void pshader_hw_texkill(const struct wined3d_shader_instruction *ins)
1677 {
1678     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1679     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1680     char reg_dest[40];
1681
1682     /* No swizzles are allowed in d3d's texkill. PS 1.x ignores the 4th component as documented,
1683      * but >= 2.0 honors it(undocumented, but tested by the d3d9 testsuit)
1684      */
1685     shader_arb_get_dst_param(ins, dst, reg_dest);
1686
1687     if (ins->ctx->reg_maps->shader_version.major >= 2)
1688     {
1689         const char *kilsrc = "TA";
1690         BOOL is_color;
1691
1692         shader_arb_get_register_name(ins, &dst->reg, reg_dest, &is_color);
1693         if(dst->write_mask == WINED3DSP_WRITEMASK_ALL)
1694         {
1695             kilsrc = reg_dest;
1696         }
1697         else
1698         {
1699             /* Sigh. KIL doesn't support swizzles/writemasks. KIL passes a writemask, but ".xy" for example
1700              * is not valid as a swizzle in ARB (needs ".xyyy"). Use SWZ to load the register properly, and set
1701              * masked out components to 0(won't kill)
1702              */
1703             char x = '0', y = '0', z = '0', w = '0';
1704             if(dst->write_mask & WINED3DSP_WRITEMASK_0) x = 'x';
1705             if(dst->write_mask & WINED3DSP_WRITEMASK_1) y = 'y';
1706             if(dst->write_mask & WINED3DSP_WRITEMASK_2) z = 'z';
1707             if(dst->write_mask & WINED3DSP_WRITEMASK_3) w = 'w';
1708             shader_addline(buffer, "SWZ TA, %s, %c, %c, %c, %c;\n", reg_dest, x, y, z, w);
1709         }
1710         shader_addline(buffer, "KIL %s;\n", kilsrc);
1711     } else {
1712         /* ARB fp doesn't like swizzles on the parameter of the KIL instruction. To mask the 4th component,
1713          * copy the register into our general purpose TMP variable, overwrite .w and pass TMP to KIL
1714          *
1715          * ps_1_3 shaders use the texcoord incarnation of the Tx register. ps_1_4 shaders can use the same,
1716          * or pass in any temporary register(in shader phase 2)
1717          */
1718         if(ins->ctx->reg_maps->shader_version.minor <= 3) {
1719             sprintf(reg_dest, "fragment.texcoord[%u]", dst->reg.idx);
1720         } else {
1721             shader_arb_get_dst_param(ins, dst, reg_dest);
1722         }
1723         shader_addline(buffer, "SWZ TA, %s, x, y, z, 1;\n", reg_dest);
1724         shader_addline(buffer, "KIL TA;\n");
1725     }
1726 }
1727
1728 static void pshader_hw_tex(const struct wined3d_shader_instruction *ins)
1729 {
1730     IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)ins->ctx->shader;
1731     IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
1732     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1733     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
1734             ins->ctx->reg_maps->shader_version.minor);
1735     struct wined3d_shader_src_param src;
1736
1737     char reg_dest[40];
1738     char reg_coord[40];
1739     DWORD reg_sampler_code;
1740     DWORD myflags = 0;
1741
1742     /* All versions have a destination register */
1743     shader_arb_get_dst_param(ins, dst, reg_dest);
1744
1745     /* 1.0-1.4: Use destination register number as texture code.
1746        2.0+: Use provided sampler number as texure code. */
1747     if (shader_version < WINED3D_SHADER_VERSION(2,0))
1748         reg_sampler_code = dst->reg.idx;
1749     else
1750         reg_sampler_code = ins->src[1].reg.idx;
1751
1752     /* 1.0-1.3: Use the texcoord varying.
1753        1.4+: Use provided coordinate source register. */
1754     if (shader_version < WINED3D_SHADER_VERSION(1,4))
1755         sprintf(reg_coord, "fragment.texcoord[%u]", reg_sampler_code);
1756     else {
1757         /* TEX is the only instruction that can handle DW and DZ natively */
1758         src = ins->src[0];
1759         if(src.modifiers == WINED3DSPSM_DW) src.modifiers = WINED3DSPSM_NONE;
1760         if(src.modifiers == WINED3DSPSM_DZ) src.modifiers = WINED3DSPSM_NONE;
1761         shader_arb_get_src_param(ins, &src, 0, reg_coord);
1762     }
1763
1764     /* projection flag:
1765      * 1.1, 1.2, 1.3: Use WINED3DTSS_TEXTURETRANSFORMFLAGS
1766      * 1.4: Use WINED3DSPSM_DZ or WINED3DSPSM_DW on src[0]
1767      * 2.0+: Use WINED3DSI_TEXLD_PROJECT on the opcode
1768      */
1769     if (shader_version < WINED3D_SHADER_VERSION(1,4))
1770     {
1771         DWORD flags = 0;
1772         if(reg_sampler_code < MAX_TEXTURES) {
1773             flags = deviceImpl->stateBlock->textureState[reg_sampler_code][WINED3DTSS_TEXTURETRANSFORMFLAGS];
1774         }
1775         if (flags & WINED3DTTFF_PROJECTED) {
1776             myflags |= TEX_PROJ;
1777         }
1778     }
1779     else if (shader_version < WINED3D_SHADER_VERSION(2,0))
1780     {
1781         DWORD src_mod = ins->src[0].modifiers;
1782         if (src_mod == WINED3DSPSM_DZ) {
1783             /* TXP cannot handle DZ natively, so move the z coordinate to .w. reg_coord is a read-only
1784              * varying register, so we need a temp reg
1785              */
1786             shader_addline(ins->ctx->buffer, "SWZ TA, %s, x, y, z, z;\n", reg_coord);
1787             strcpy(reg_coord, "TA");
1788             myflags |= TEX_PROJ;
1789         } else if(src_mod == WINED3DSPSM_DW) {
1790             myflags |= TEX_PROJ;
1791         }
1792     } else {
1793         if (ins->flags & WINED3DSI_TEXLD_PROJECT) myflags |= TEX_PROJ;
1794         if (ins->flags & WINED3DSI_TEXLD_BIAS) myflags |= TEX_BIAS;
1795     }
1796     shader_hw_sample(ins, reg_sampler_code, reg_dest, reg_coord, myflags, NULL, NULL);
1797 }
1798
1799 static void pshader_hw_texcoord(const struct wined3d_shader_instruction *ins)
1800 {
1801     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1802     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1803     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
1804             ins->ctx->reg_maps->shader_version.minor);
1805     char dst_str[50];
1806
1807     if (shader_version < WINED3D_SHADER_VERSION(1,4))
1808     {
1809         DWORD reg = dst->reg.idx;
1810
1811         shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
1812         shader_addline(buffer, "MOV_SAT %s, fragment.texcoord[%u];\n", dst_str, reg);
1813     } else {
1814         char reg_src[40];
1815
1816         shader_arb_get_src_param(ins, &ins->src[0], 0, reg_src);
1817         shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
1818         shader_addline(buffer, "MOV %s, %s;\n", dst_str, reg_src);
1819    }
1820 }
1821
1822 static void pshader_hw_texreg2ar(const struct wined3d_shader_instruction *ins)
1823 {
1824      struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1825      IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)ins->ctx->shader;
1826      IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
1827      DWORD flags;
1828
1829      DWORD reg1 = ins->dst[0].reg.idx;
1830      char dst_str[50];
1831      char src_str[50];
1832
1833      /* Note that texreg2ar treats Tx as a temporary register, not as a varying */
1834      shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
1835      shader_arb_get_src_param(ins, &ins->src[0], 0, src_str);
1836      /* Move .x first in case src_str is "TA" */
1837      shader_addline(buffer, "MOV TA.y, %s.x;\n", src_str);
1838      shader_addline(buffer, "MOV TA.x, %s.w;\n", src_str);
1839      flags = reg1 < MAX_TEXTURES ? deviceImpl->stateBlock->textureState[reg1][WINED3DTSS_TEXTURETRANSFORMFLAGS] : 0;
1840      shader_hw_sample(ins, reg1, dst_str, "TA", flags & WINED3DTTFF_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
1841 }
1842
1843 static void pshader_hw_texreg2gb(const struct wined3d_shader_instruction *ins)
1844 {
1845      struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1846
1847      DWORD reg1 = ins->dst[0].reg.idx;
1848      char dst_str[50];
1849      char src_str[50];
1850
1851      /* Note that texreg2gb treats Tx as a temporary register, not as a varying */
1852      shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
1853      shader_arb_get_src_param(ins, &ins->src[0], 0, src_str);
1854      shader_addline(buffer, "MOV TA.x, %s.y;\n", src_str);
1855      shader_addline(buffer, "MOV TA.y, %s.z;\n", src_str);
1856      shader_hw_sample(ins, reg1, dst_str, "TA", 0, NULL, NULL);
1857 }
1858
1859 static void pshader_hw_texreg2rgb(const struct wined3d_shader_instruction *ins)
1860 {
1861     DWORD reg1 = ins->dst[0].reg.idx;
1862     char dst_str[50];
1863     char src_str[50];
1864
1865     /* Note that texreg2rg treats Tx as a temporary register, not as a varying */
1866     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
1867     shader_arb_get_src_param(ins, &ins->src[0], 0, src_str);
1868     shader_hw_sample(ins, reg1, dst_str, src_str, 0, NULL, NULL);
1869 }
1870
1871 static void pshader_hw_texbem(const struct wined3d_shader_instruction *ins)
1872 {
1873     IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)ins->ctx->shader;
1874     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1875     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1876     char reg_coord[40], dst_reg[50], src_reg[50];
1877     DWORD reg_dest_code;
1878
1879     /* All versions have a destination register. The Tx where the texture coordinates come
1880      * from is the varying incarnation of the texture register
1881      */
1882     reg_dest_code = dst->reg.idx;
1883     shader_arb_get_dst_param(ins, &ins->dst[0], dst_reg);
1884     shader_arb_get_src_param(ins, &ins->src[0], 0, src_reg);
1885     sprintf(reg_coord, "fragment.texcoord[%u]", reg_dest_code);
1886
1887     /* Sampling the perturbation map in Tsrc was done already, including the signedness correction if needed
1888      * The Tx in which the perturbation map is stored is the tempreg incarnation of the texture register
1889      *
1890      * GL_NV_fragment_program_option could handle this in one instruction via X2D:
1891      * X2D TA.xy, fragment.texcoord, T%u, bumpenvmat%u.xzyw
1892      *
1893      * However, the NV extensions are never enabled for <= 2.0 shaders because of the performance penalty that
1894      * comes with it, and texbem is an 1.x only instruction. No 1.x instruction forces us to enable the NV
1895      * extension.
1896      */
1897     shader_addline(buffer, "SWZ TB, bumpenvmat%d, x, z, 0, 0;\n", reg_dest_code);
1898     shader_addline(buffer, "DP3 TA.x, TB, %s;\n", src_reg);
1899     shader_addline(buffer, "SWZ TB, bumpenvmat%d, y, w, 0, 0;\n", reg_dest_code);
1900     shader_addline(buffer, "DP3 TA.y, TB, %s;\n", src_reg);
1901
1902     /* with projective textures, texbem only divides the static texture coord, not the displacement,
1903      * so we can't let the GL handle this.
1904      */
1905     if (((IWineD3DDeviceImpl*) This->baseShader.device)->stateBlock->textureState[reg_dest_code][WINED3DTSS_TEXTURETRANSFORMFLAGS]
1906             & WINED3DTTFF_PROJECTED) {
1907         shader_addline(buffer, "RCP TB.w, %s.w;\n", reg_coord);
1908         shader_addline(buffer, "MUL TB.xy, %s, TB.w;\n", reg_coord);
1909         shader_addline(buffer, "ADD TA.xy, TA, TB;\n");
1910     } else {
1911         shader_addline(buffer, "ADD TA.xy, TA, %s;\n", reg_coord);
1912     }
1913
1914     shader_hw_sample(ins, reg_dest_code, dst_reg, "TA", 0, NULL, NULL);
1915
1916     if (ins->handler_idx == WINED3DSIH_TEXBEML)
1917     {
1918         /* No src swizzles are allowed, so this is ok */
1919         shader_addline(buffer, "MAD TA, %s.z, luminance%d.x, luminance%d.y;\n",
1920                        src_reg, reg_dest_code, reg_dest_code);
1921         shader_addline(buffer, "MUL %s, %s, TA;\n", dst_reg, dst_reg);
1922     }
1923 }
1924
1925 static void pshader_hw_texm3x2pad(const struct wined3d_shader_instruction *ins)
1926 {
1927     DWORD reg = ins->dst[0].reg.idx;
1928     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1929     char src0_name[50], dst_name[50];
1930     BOOL is_color;
1931     struct wined3d_shader_register tmp_reg = ins->dst[0].reg;
1932
1933     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
1934     /* The next instruction will be a texm3x2tex or texm3x2depth that writes to the uninitialized
1935      * T<reg+1> register. Use this register to store the calculated vector
1936      */
1937     tmp_reg.idx = reg + 1;
1938     shader_arb_get_register_name(ins, &tmp_reg, dst_name, &is_color);
1939     shader_addline(buffer, "DP3 %s.x, fragment.texcoord[%u], %s;\n", dst_name, reg, src0_name);
1940 }
1941
1942 static void pshader_hw_texm3x2tex(const struct wined3d_shader_instruction *ins)
1943 {
1944     IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)ins->ctx->shader;
1945     IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
1946     DWORD flags;
1947     DWORD reg = ins->dst[0].reg.idx;
1948     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1949     char dst_str[50];
1950     char src0_name[50];
1951     char dst_reg[50];
1952     BOOL is_color;
1953
1954     /* We know that we're writing to the uninitialized T<reg> register, so use it for temporary storage */
1955     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_reg, &is_color);
1956
1957     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
1958     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
1959     shader_addline(buffer, "DP3 %s.y, fragment.texcoord[%u], %s;\n", dst_reg, reg, src0_name);
1960     flags = reg < MAX_TEXTURES ? deviceImpl->stateBlock->textureState[reg][WINED3DTSS_TEXTURETRANSFORMFLAGS] : 0;
1961     shader_hw_sample(ins, reg, dst_str, dst_reg, flags & WINED3DTTFF_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
1962 }
1963
1964 static void pshader_hw_texm3x3pad(const struct wined3d_shader_instruction *ins)
1965 {
1966     IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)ins->ctx->shader;
1967     DWORD reg = ins->dst[0].reg.idx;
1968     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1969     SHADER_PARSE_STATE* current_state = &This->baseShader.parse_state;
1970     char src0_name[50], dst_name[50];
1971     struct wined3d_shader_register tmp_reg = ins->dst[0].reg;
1972     BOOL is_color;
1973
1974     /* There are always 2 texm3x3pad instructions followed by one texm3x3[tex,vspec, ...] instruction, with
1975      * incrementing ins->dst[0].register_idx numbers. So the pad instruction already knows the final destination
1976      * register, and this register is uninitialized(otherwise the assembler complains that it is 'redeclared')
1977      */
1978     tmp_reg.idx = reg + 2 - current_state->current_row;
1979     shader_arb_get_register_name(ins, &tmp_reg, dst_name, &is_color);
1980
1981     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
1982     shader_addline(buffer, "DP3 %s.%c, fragment.texcoord[%u], %s;\n",
1983                    dst_name, 'x' + current_state->current_row, reg, src0_name);
1984     current_state->texcoord_w[current_state->current_row++] = reg;
1985 }
1986
1987 static void pshader_hw_texm3x3tex(const struct wined3d_shader_instruction *ins)
1988 {
1989     IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)ins->ctx->shader;
1990     IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
1991     DWORD flags;
1992     DWORD reg = ins->dst[0].reg.idx;
1993     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1994     SHADER_PARSE_STATE* current_state = &This->baseShader.parse_state;
1995     char dst_str[50];
1996     char src0_name[50], dst_name[50];
1997     BOOL is_color;
1998
1999     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2000     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2001     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_name, reg, src0_name);
2002
2003     /* Sample the texture using the calculated coordinates */
2004     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2005     flags = reg < MAX_TEXTURES ? deviceImpl->stateBlock->textureState[reg][WINED3DTSS_TEXTURETRANSFORMFLAGS] : 0;
2006     shader_hw_sample(ins, reg, dst_str, dst_name, flags & WINED3DTTFF_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2007     current_state->current_row = 0;
2008 }
2009
2010 static void pshader_hw_texm3x3vspec(const struct wined3d_shader_instruction *ins)
2011 {
2012     IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)ins->ctx->shader;
2013     IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
2014     DWORD flags;
2015     DWORD reg = ins->dst[0].reg.idx;
2016     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2017     SHADER_PARSE_STATE* current_state = &This->baseShader.parse_state;
2018     char dst_str[50];
2019     char src0_name[50];
2020     char dst_reg[8];
2021     BOOL is_color;
2022
2023     /* Get the dst reg without writemask strings. We know this register is uninitialized, so we can use all
2024      * components for temporary data storage
2025      */
2026     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_reg, &is_color);
2027     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2028     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_reg, reg, src0_name);
2029
2030     /* Construct the eye-ray vector from w coordinates */
2031     shader_addline(buffer, "MOV TB.x, fragment.texcoord[%u].w;\n", current_state->texcoord_w[0]);
2032     shader_addline(buffer, "MOV TB.y, fragment.texcoord[%u].w;\n", current_state->texcoord_w[1]);
2033     shader_addline(buffer, "MOV TB.z, fragment.texcoord[%u].w;\n", reg);
2034
2035     /* Calculate reflection vector
2036      */
2037     shader_addline(buffer, "DP3 %s.w, %s, TB;\n", dst_reg, dst_reg);
2038     /* The .w is ignored when sampling, so I can use TB.w to calculate dot(N, N) */
2039     shader_addline(buffer, "DP3 TB.w, %s, %s;\n", dst_reg, dst_reg);
2040     shader_addline(buffer, "RCP TB.w, TB.w;\n");
2041     shader_addline(buffer, "MUL %s.w, %s.w, TB.w;\n", dst_reg, dst_reg);
2042     shader_addline(buffer, "MUL %s, %s.w, %s;\n", dst_reg, dst_reg, dst_reg);
2043     shader_addline(buffer, "MAD %s, coefmul.x, %s, -TB;\n", dst_reg, dst_reg);
2044
2045     /* Sample the texture using the calculated coordinates */
2046     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2047     flags = reg < MAX_TEXTURES ? deviceImpl->stateBlock->textureState[reg][WINED3DTSS_TEXTURETRANSFORMFLAGS] : 0;
2048     shader_hw_sample(ins, reg, dst_str, dst_reg, flags & WINED3DTTFF_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2049     current_state->current_row = 0;
2050 }
2051
2052 static void pshader_hw_texm3x3spec(const struct wined3d_shader_instruction *ins)
2053 {
2054     IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)ins->ctx->shader;
2055     IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
2056     DWORD flags;
2057     DWORD reg = ins->dst[0].reg.idx;
2058     SHADER_PARSE_STATE* current_state = &This->baseShader.parse_state;
2059     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2060     char dst_str[50];
2061     char src0_name[50];
2062     char src1_name[50];
2063     char dst_reg[8];
2064     BOOL is_color;
2065
2066     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2067     shader_arb_get_src_param(ins, &ins->src[0], 1, src1_name);
2068     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_reg, &is_color);
2069     /* Note: dst_reg.xy is input here, generated by two texm3x3pad instructions */
2070     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_reg, reg, src0_name);
2071
2072     /* Calculate reflection vector.
2073      *
2074      *                   dot(N, E)
2075      * dst_reg.xyz = 2 * --------- * N - E
2076      *                   dot(N, N)
2077      *
2078      * Which normalizes the normal vector
2079      */
2080     shader_addline(buffer, "DP3 %s.w, %s, %s;\n", dst_reg, dst_reg, src1_name);
2081     shader_addline(buffer, "DP3 TC.w, %s, %s;\n", dst_reg, dst_reg);
2082     shader_addline(buffer, "RCP TC.w, TC.w;\n");
2083     shader_addline(buffer, "MUL %s.w, %s.w, TC.w;\n", dst_reg, dst_reg);
2084     shader_addline(buffer, "MUL %s, %s.w, %s;\n", dst_reg, dst_reg, dst_reg);
2085     shader_addline(buffer, "MAD %s, coefmul.x, %s, -%s;\n", dst_reg, dst_reg, src1_name);
2086
2087     /* Sample the texture using the calculated coordinates */
2088     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2089     flags = reg < MAX_TEXTURES ? deviceImpl->stateBlock->textureState[reg][WINED3DTSS_TEXTURETRANSFORMFLAGS] : 0;
2090     shader_hw_sample(ins, reg, dst_str, dst_reg, flags & WINED3DTTFF_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2091     current_state->current_row = 0;
2092 }
2093
2094 static void pshader_hw_texdepth(const struct wined3d_shader_instruction *ins)
2095 {
2096     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2097     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2098     char dst_name[50];
2099
2100     /* texdepth has an implicit destination, the fragment depth value. It's only parameter,
2101      * which is essentially an input, is the destination register because it is the first
2102      * parameter. According to the msdn, this must be register r5, but let's keep it more flexible
2103      * here(writemasks/swizzles are not valid on texdepth)
2104      */
2105     shader_arb_get_dst_param(ins, dst, dst_name);
2106
2107     /* According to the msdn, the source register(must be r5) is unusable after
2108      * the texdepth instruction, so we're free to modify it
2109      */
2110     shader_addline(buffer, "MIN %s.y, %s.y, one.y;\n", dst_name, dst_name);
2111
2112     /* How to deal with the special case dst_name.g == 0? if r != 0, then
2113      * the r * (1 / 0) will give infinity, which is clamped to 1.0, the correct
2114      * result. But if r = 0.0, then 0 * inf = 0, which is incorrect.
2115      */
2116     shader_addline(buffer, "RCP %s.y, %s.y;\n", dst_name, dst_name);
2117     shader_addline(buffer, "MUL TA.x, %s.x, %s.y;\n", dst_name, dst_name);
2118     shader_addline(buffer, "MIN TA.x, TA.x, one.x;\n");
2119     shader_addline(buffer, "MAX result.depth, TA.x, 0.0;\n");
2120 }
2121
2122 /** Process the WINED3DSIO_TEXDP3TEX instruction in ARB:
2123  * Take a 3-component dot product of the TexCoord[dstreg] and src,
2124  * then perform a 1D texture lookup from stage dstregnum, place into dst. */
2125 static void pshader_hw_texdp3tex(const struct wined3d_shader_instruction *ins)
2126 {
2127     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2128     DWORD sampler_idx = ins->dst[0].reg.idx;
2129     char src0[50];
2130     char dst_str[50];
2131
2132     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2133     shader_addline(buffer, "MOV TB, 0.0;\n");
2134     shader_addline(buffer, "DP3 TB.x, fragment.texcoord[%u], %s;\n", sampler_idx, src0);
2135
2136     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2137     shader_hw_sample(ins, sampler_idx, dst_str, "TB", 0 /* Only one coord, can't be projected */, NULL, NULL);
2138 }
2139
2140 /** Process the WINED3DSIO_TEXDP3 instruction in ARB:
2141  * Take a 3-component dot product of the TexCoord[dstreg] and src. */
2142 static void pshader_hw_texdp3(const struct wined3d_shader_instruction *ins)
2143 {
2144     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2145     char src0[50];
2146     char dst_str[50];
2147     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2148
2149     /* Handle output register */
2150     shader_arb_get_dst_param(ins, dst, dst_str);
2151     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2152     shader_addline(buffer, "DP3 %s, fragment.texcoord[%u], %s;\n", dst_str, dst->reg.idx, src0);
2153 }
2154
2155 /** Process the WINED3DSIO_TEXM3X3 instruction in ARB
2156  * Perform the 3rd row of a 3x3 matrix multiply */
2157 static void pshader_hw_texm3x3(const struct wined3d_shader_instruction *ins)
2158 {
2159     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2160     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2161     char dst_str[50], dst_name[50];
2162     char src0[50];
2163     BOOL is_color;
2164
2165     shader_arb_get_dst_param(ins, dst, dst_str);
2166     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2167     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2168     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_name, dst->reg.idx, src0);
2169     shader_addline(buffer, "MOV %s, %s;\n", dst_str, dst_name);
2170 }
2171
2172 /** Process the WINED3DSIO_TEXM3X2DEPTH instruction in ARB:
2173  * Last row of a 3x2 matrix multiply, use the result to calculate the depth:
2174  * Calculate tmp0.y = TexCoord[dstreg] . src.xyz;  (tmp0.x has already been calculated)
2175  * depth = (tmp0.y == 0.0) ? 1.0 : tmp0.x / tmp0.y
2176  */
2177 static void pshader_hw_texm3x2depth(const struct wined3d_shader_instruction *ins)
2178 {
2179     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2180     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2181     char src0[50], dst_name[50];
2182     BOOL is_color;
2183
2184     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2185     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2186     shader_addline(buffer, "DP3 %s.y, fragment.texcoord[%u], %s;\n", dst_name, dst->reg.idx, src0);
2187
2188     /* How to deal with the special case dst_name.g == 0? if r != 0, then
2189      * the r * (1 / 0) will give infinity, which is clamped to 1.0, the correct
2190      * result. But if r = 0.0, then 0 * inf = 0, which is incorrect.
2191      */
2192     shader_addline(buffer, "RCP %s.y, %s.y;\n", dst_name, dst_name);
2193     shader_addline(buffer, "MUL %s.x, %s.x, %s.y;\n", dst_name, dst_name, dst_name);
2194     shader_addline(buffer, "MIN %s.x, %s.x, one.x;\n", dst_name, dst_name);
2195     shader_addline(buffer, "MAX result.depth, %s.x, 0.0;\n", dst_name);
2196 }
2197
2198 /** Handles transforming all WINED3DSIO_M?x? opcodes for
2199     Vertex/Pixel shaders to ARB_vertex_program codes */
2200 static void shader_hw_mnxn(const struct wined3d_shader_instruction *ins)
2201 {
2202     int i;
2203     int nComponents = 0;
2204     struct wined3d_shader_dst_param tmp_dst = {{0}};
2205     struct wined3d_shader_src_param tmp_src[2] = {{{0}}};
2206     struct wined3d_shader_instruction tmp_ins;
2207
2208     memset(&tmp_ins, 0, sizeof(tmp_ins));
2209
2210     /* Set constants for the temporary argument */
2211     tmp_ins.ctx = ins->ctx;
2212     tmp_ins.dst_count = 1;
2213     tmp_ins.dst = &tmp_dst;
2214     tmp_ins.src_count = 2;
2215     tmp_ins.src = tmp_src;
2216
2217     switch(ins->handler_idx)
2218     {
2219         case WINED3DSIH_M4x4:
2220             nComponents = 4;
2221             tmp_ins.handler_idx = WINED3DSIH_DP4;
2222             break;
2223         case WINED3DSIH_M4x3:
2224             nComponents = 3;
2225             tmp_ins.handler_idx = WINED3DSIH_DP4;
2226             break;
2227         case WINED3DSIH_M3x4:
2228             nComponents = 4;
2229             tmp_ins.handler_idx = WINED3DSIH_DP3;
2230             break;
2231         case WINED3DSIH_M3x3:
2232             nComponents = 3;
2233             tmp_ins.handler_idx = WINED3DSIH_DP3;
2234             break;
2235         case WINED3DSIH_M3x2:
2236             nComponents = 2;
2237             tmp_ins.handler_idx = WINED3DSIH_DP3;
2238             break;
2239         default:
2240             FIXME("Unhandled opcode %#x\n", ins->handler_idx);
2241             break;
2242     }
2243
2244     tmp_dst = ins->dst[0];
2245     tmp_src[0] = ins->src[0];
2246     tmp_src[1] = ins->src[1];
2247     for (i = 0; i < nComponents; i++) {
2248         tmp_dst.write_mask = WINED3DSP_WRITEMASK_0 << i;
2249         shader_hw_map2gl(&tmp_ins);
2250         ++tmp_src[1].reg.idx;
2251     }
2252 }
2253
2254 static void shader_hw_scalar_op(const struct wined3d_shader_instruction *ins)
2255 {
2256     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2257     const char *instruction;
2258
2259     char dst[50];
2260     char src[50];
2261
2262     switch(ins->handler_idx)
2263     {
2264         case WINED3DSIH_RSQ:  instruction = "RSQ"; break;
2265         case WINED3DSIH_RCP:  instruction = "RCP"; break;
2266         case WINED3DSIH_EXP:  instruction = "EX2"; break;
2267         case WINED3DSIH_EXPP: instruction = "EXP"; break;
2268         default: instruction = "";
2269             FIXME("Unhandled opcode %#x\n", ins->handler_idx);
2270             break;
2271     }
2272
2273     shader_arb_get_dst_param(ins, &ins->dst[0], dst); /* Destination */
2274     shader_arb_get_src_param(ins, &ins->src[0], 0, src);
2275     if (ins->src[0].swizzle == WINED3DSP_NOSWIZZLE)
2276     {
2277         /* Dx sdk says .x is used if no swizzle is given, but our test shows that
2278          * .w is used
2279          */
2280         strcat(src, ".w");
2281     }
2282
2283     shader_addline(buffer, "%s%s %s, %s;\n", instruction, shader_arb_get_modifier(ins), dst, src);
2284 }
2285
2286 static void shader_hw_nrm(const struct wined3d_shader_instruction *ins)
2287 {
2288     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2289     char dst_name[50];
2290     char src_name[50];
2291     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2292     BOOL pshader = shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type);
2293
2294     shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2295     shader_arb_get_src_param(ins, &ins->src[0], 1 /* Use TB */, src_name);
2296
2297     if(pshader && priv->target_version >= NV3)
2298     {
2299         shader_addline(buffer, "NRM%s %s, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name);
2300     }
2301     else
2302     {
2303         shader_addline(buffer, "DP3 TA, %s, %s;\n", src_name, src_name);
2304         shader_addline(buffer, "RSQ TA, TA.x;\n");
2305         /* dst.w = src[0].w * 1 / (src.x^2 + src.y^2 + src.z^2)^(1/2) according to msdn*/
2306         shader_addline(buffer, "MUL%s %s, %s, TA;\n", shader_arb_get_modifier(ins), dst_name,
2307                     src_name);
2308     }
2309 }
2310
2311 static void shader_hw_lrp(const struct wined3d_shader_instruction *ins)
2312 {
2313     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2314     char dst_name[50];
2315     char src_name[3][50];
2316
2317     /* ARB_fragment_program has a convenient LRP instruction */
2318     if(shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type)) {
2319         shader_hw_map2gl(ins);
2320         return;
2321     }
2322
2323     shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2324     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
2325     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
2326     shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
2327
2328     shader_addline(buffer, "SUB TA, %s, %s;\n", src_name[1], src_name[2]);
2329     shader_addline(buffer, "MAD%s %s, %s, TA, %s;\n", shader_arb_get_modifier(ins),
2330                    dst_name, src_name[0], src_name[2]);
2331 }
2332
2333 static void shader_hw_sincos(const struct wined3d_shader_instruction *ins)
2334 {
2335     /* This instruction exists in ARB, but the d3d instruction takes two extra parameters which
2336      * must contain fixed constants. So we need a separate function to filter those constants and
2337      * can't use map2gl
2338      */
2339     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2340     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2341     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2342     char dst_name[50];
2343     char src_name0[50], src_name1[50], src_name2[50];
2344     BOOL is_color;
2345
2346     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name0);
2347     if(shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type)) {
2348         shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2349         /* No modifiers are supported on SCS */
2350         shader_addline(buffer, "SCS %s, %s;\n", dst_name, src_name0);
2351
2352         if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE)
2353         {
2354             shader_arb_get_register_name(ins, &dst->reg, src_name0, &is_color);
2355             shader_addline(buffer, "MOV_SAT %s, %s;\n", dst_name, src_name0);
2356         }
2357     } else if(priv->target_version >= NV2) {
2358         shader_arb_get_register_name(ins, &dst->reg, dst_name, &is_color);
2359
2360         /* Sincos writemask must be .x, .y or .xy */
2361         if(dst->write_mask & WINED3DSP_WRITEMASK_0)
2362             shader_addline(buffer, "COS%s %s.x, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name0);
2363         if(dst->write_mask & WINED3DSP_WRITEMASK_1)
2364             shader_addline(buffer, "SIN%s %s.y, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name0);
2365     } else {
2366         /* Approximate sine and cosine with a taylor series, as per math textbook. The application passes 8
2367          * helper constants(D3DSINCOSCONST1 and D3DSINCOSCONST2) in src1 and src2.
2368          *
2369          * sin(x) = x - x^3/3! + x^5/5! - x^7/7! + ...
2370          * cos(x) = 1 - x^2/2! + x^4/4! - x^6/6! + ...
2371          *
2372          * The constants we get are:
2373          *
2374          *  +1   +1,     -1     -1     +1      +1      -1       -1
2375          *      ---- ,  ---- , ---- , ----- , ----- , ----- , ------
2376          *      1!*2    2!*4   3!*8   4!*16   5!*32   6!*64   7!*128
2377          *
2378          * If used with x^2, x^3, x^4 etc they calculate sin(x/2) and cos(x/2):
2379          *
2380          * (x/2)^2 = x^2 / 4
2381          * (x/2)^3 = x^3 / 8
2382          * (x/2)^4 = x^4 / 16
2383          * (x/2)^5 = x^5 / 32
2384          * etc
2385          *
2386          * To get the final result:
2387          * sin(x) = 2 * sin(x/2) * cos(x/2)
2388          * cos(x) = cos(x/2)^2 - sin(x/2)^2
2389          * (from sin(x+y) and cos(x+y) rules)
2390          *
2391          * As per MSDN, dst.z is undefined after the operation, and so is
2392          * dst.x and dst.y if they're masked out by the writemask. Ie
2393          * sincos dst.y, src1, c0, c1
2394          * returns the sine in dst.y. dst.x and dst.z are undefined, dst.w is not touched. The assembler
2395          * vsa.exe also stops with an error if the dest register is the same register as the source
2396          * register. This means we can use dest.xyz as temporary storage. The assembler vsa.exe output also
2397          * indicates that sincos consumes 8 instruction slots in vs_2_0(and, strangely, in vs_3_0).
2398          */
2399         shader_arb_get_src_param(ins, &ins->src[1], 1, src_name1);
2400         shader_arb_get_src_param(ins, &ins->src[2], 2, src_name2);
2401         shader_arb_get_register_name(ins, &dst->reg, dst_name, &is_color);
2402
2403         shader_addline(buffer, "MUL %s.x, %s, %s;\n", dst_name, src_name0, src_name0);  /* x ^ 2 */
2404         shader_addline(buffer, "MUL TA.y, %s.x, %s;\n", dst_name, src_name0);           /* x ^ 3 */
2405         shader_addline(buffer, "MUL %s.y, TA.y, %s;\n", dst_name, src_name0);           /* x ^ 4 */
2406         shader_addline(buffer, "MUL TA.z, %s.y, %s;\n", dst_name, src_name0);           /* x ^ 5 */
2407         shader_addline(buffer, "MUL %s.z, TA.z, %s;\n", dst_name, src_name0);           /* x ^ 6 */
2408         shader_addline(buffer, "MUL TA.w, %s.z, %s;\n", dst_name, src_name0);           /* x ^ 7 */
2409
2410         /* sin(x/2)
2411          *
2412          * Unfortunately we don't get the constants in a DP4-capable form. Is there a way to
2413          * properly merge that with MULs in the code above?
2414          * The swizzles .yz and xw however fit into the .yzxw swizzle added to ps_2_0. Maybe
2415          * we can merge the sine and cosine MAD rows to calculate them together.
2416          */
2417         shader_addline(buffer, "MUL TA.x, %s, %s.w;\n", src_name0, src_name2); /* x^1, +1/(1!*2) */
2418         shader_addline(buffer, "MAD TA.x, TA.y, %s.x, TA.x;\n", src_name2); /* -1/(3!*8) */
2419         shader_addline(buffer, "MAD TA.x, TA.z, %s.w, TA.x;\n", src_name1); /* +1/(5!*32) */
2420         shader_addline(buffer, "MAD TA.x, TA.w, %s.x, TA.x;\n", src_name1); /* -1/(7!*128) */
2421
2422         /* cos(x/2) */
2423         shader_addline(buffer, "MAD TA.y, %s.x, %s.y, %s.z;\n", dst_name, src_name2, src_name2); /* -1/(2!*4), +1.0 */
2424         shader_addline(buffer, "MAD TA.y, %s.y, %s.z, TA.y;\n", dst_name, src_name1); /* +1/(4!*16) */
2425         shader_addline(buffer, "MAD TA.y, %s.z, %s.y, TA.y;\n", dst_name, src_name1); /* -1/(6!*64) */
2426
2427         if(dst->write_mask & WINED3DSP_WRITEMASK_0) {
2428             /* cos x */
2429             shader_addline(buffer, "MUL TA.z, TA.y, TA.y;\n");
2430             shader_addline(buffer, "MAD %s.x, -TA.x, TA.x, TA.z;\n", dst_name);
2431         }
2432         if(dst->write_mask & WINED3DSP_WRITEMASK_1) {
2433             /* sin x */
2434             shader_addline(buffer, "MUL %s.y, TA.x, TA.y;\n", dst_name);
2435             shader_addline(buffer, "ADD %s.y, %s.y, %s.y;\n", dst_name, dst_name, dst_name);
2436         }
2437     }
2438 }
2439
2440 static void shader_hw_sgn(const struct wined3d_shader_instruction *ins)
2441 {
2442     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2443     char dst_name[50];
2444     char src_name[50];
2445     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
2446
2447     /* SGN is only valid in vertex shaders */
2448     if(ctx->target_version == NV2) {
2449         shader_hw_map2gl(ins);
2450         return;
2451     }
2452     shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2453     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name);
2454
2455     /* If SRC > 0.0, -SRC < SRC = TRUE, otherwise false.
2456      * if SRC < 0.0,  SRC < -SRC = TRUE. If neither is true, src = 0.0
2457      */
2458     if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE) {
2459         shader_addline(buffer, "SLT %s, -%s, %s;\n", dst_name, src_name, src_name);
2460     } else {
2461         /* src contains TA? Write to the dest first. This won't overwrite our destination.
2462          * Then use TA, and calculate the final result
2463          *
2464          * Not reading from TA? Store the first result in TA to avoid overwriting the
2465          * destination if src reg = dst reg
2466          */
2467         if(strstr(src_name, "TA"))
2468         {
2469             shader_addline(buffer, "SLT %s,  %s, -%s;\n", dst_name, src_name, src_name);
2470             shader_addline(buffer, "SLT TA, -%s, %s;\n", src_name, src_name);
2471             shader_addline(buffer, "ADD %s, %s, -TA;\n", dst_name, dst_name);
2472         }
2473         else
2474         {
2475             shader_addline(buffer, "SLT TA, -%s, %s;\n", src_name, src_name);
2476             shader_addline(buffer, "SLT %s,  %s, -%s;\n", dst_name, src_name, src_name);
2477             shader_addline(buffer, "ADD %s, TA, -%s;\n", dst_name, dst_name);
2478         }
2479     }
2480 }
2481
2482 static void shader_hw_dsy(const struct wined3d_shader_instruction *ins)
2483 {
2484     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2485     char src[50];
2486     char dst[50];
2487     char dst_name[50];
2488     BOOL is_color;
2489
2490     shader_arb_get_dst_param(ins, &ins->dst[0], dst);
2491     shader_arb_get_src_param(ins, &ins->src[0], 0, src);
2492     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2493
2494     shader_addline(buffer, "DDY %s, %s;\n", dst, src);
2495     shader_addline(buffer, "MUL%s %s, %s, ycorrection.y;\n", shader_arb_get_modifier(ins), dst, dst_name);
2496 }
2497
2498 static DWORD abs_modifier(DWORD mod, BOOL *need_abs)
2499 {
2500     *need_abs = FALSE;
2501
2502     switch(mod)
2503     {
2504         case WINED3DSPSM_NONE:      return WINED3DSPSM_ABS;
2505         case WINED3DSPSM_NEG:       return WINED3DSPSM_ABS;
2506         case WINED3DSPSM_BIAS:      *need_abs = TRUE; return WINED3DSPSM_BIAS;
2507         case WINED3DSPSM_BIASNEG:   *need_abs = TRUE; return WINED3DSPSM_BIASNEG;
2508         case WINED3DSPSM_SIGN:      *need_abs = TRUE; return WINED3DSPSM_SIGN;
2509         case WINED3DSPSM_SIGNNEG:   *need_abs = TRUE; return WINED3DSPSM_SIGNNEG;
2510         case WINED3DSPSM_COMP:      *need_abs = TRUE; return WINED3DSPSM_COMP;
2511         case WINED3DSPSM_X2:        *need_abs = TRUE; return WINED3DSPSM_X2;
2512         case WINED3DSPSM_X2NEG:     *need_abs = TRUE; return WINED3DSPSM_X2NEG;
2513         case WINED3DSPSM_DZ:        *need_abs = TRUE; return WINED3DSPSM_DZ;
2514         case WINED3DSPSM_DW:        *need_abs = TRUE; return WINED3DSPSM_DW;
2515         case WINED3DSPSM_ABS:       return WINED3DSPSM_ABS;
2516         case WINED3DSPSM_ABSNEG:    return WINED3DSPSM_ABS;
2517     }
2518     FIXME("Unknown modifier %u\n", mod);
2519     return mod;
2520 }
2521
2522 static void shader_hw_log_pow(const struct wined3d_shader_instruction *ins)
2523 {
2524     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2525     char src0[50], src1[50], dst[50];
2526     struct wined3d_shader_src_param src0_copy = ins->src[0];
2527     BOOL need_abs = FALSE;
2528     const char *instr;
2529     BOOL arg2 = FALSE;
2530
2531     switch(ins->handler_idx)
2532     {
2533         case WINED3DSIH_LOG:  instr = "LG2"; break;
2534         case WINED3DSIH_LOGP: instr = "LOG"; break;
2535         case WINED3DSIH_POW:  instr = "POW"; arg2 = TRUE; break;
2536         default:
2537             ERR("Unexpected instruction %d\n", ins->handler_idx);
2538             return;
2539     }
2540
2541     /* LOG, LOGP and POW operate on the absolute value of the input */
2542     src0_copy.modifiers = abs_modifier(src0_copy.modifiers, &need_abs);
2543
2544     shader_arb_get_dst_param(ins, &ins->dst[0], dst);
2545     shader_arb_get_src_param(ins, &src0_copy, 0, src0);
2546     if(arg2) shader_arb_get_src_param(ins, &ins->src[1], 1, src1);
2547
2548     if(need_abs)
2549     {
2550         shader_addline(buffer, "ABS TA, %s;\n", src0);
2551         if(arg2)
2552         {
2553             shader_addline(buffer, "%s%s %s, TA, %s;\n", instr, shader_arb_get_modifier(ins), dst, src1);
2554         }
2555         else
2556         {
2557             shader_addline(buffer, "%s%s %s, TA;\n", instr, shader_arb_get_modifier(ins), dst);
2558         }
2559     }
2560     else if(arg2)
2561     {
2562         shader_addline(buffer, "%s%s %s, %s, %s;\n", instr, shader_arb_get_modifier(ins), dst, src0, src1);
2563     }
2564     else
2565     {
2566         shader_addline(buffer, "%s%s %s, %s;\n", instr, shader_arb_get_modifier(ins), dst, src0);
2567     }
2568 }
2569
2570 static void shader_hw_loop(const struct wined3d_shader_instruction *ins)
2571 {
2572     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2573     char src_name[50];
2574     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2575
2576     /* src0 is aL */
2577     shader_arb_get_src_param(ins, &ins->src[1], 0, src_name);
2578
2579     if(vshader)
2580     {
2581         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2582         struct list *e = list_head(&priv->control_frames);
2583         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2584
2585         if(priv->loop_depth > 1) shader_addline(buffer, "PUSHA aL;\n");
2586         /* The constant loader makes sure to load -1 into iX.w */
2587         shader_addline(buffer, "ARLC aL, %s.xywz;\n", src_name);
2588         shader_addline(buffer, "BRA loop_%u_end (LE.x);\n", control_frame->loop_no);
2589         shader_addline(buffer, "loop_%u_start:\n", control_frame->loop_no);
2590     }
2591     else
2592     {
2593         shader_addline(buffer, "LOOP %s;\n", src_name);
2594     }
2595 }
2596
2597 static void shader_hw_rep(const struct wined3d_shader_instruction *ins)
2598 {
2599     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2600     char src_name[50];
2601     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2602
2603     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name);
2604
2605     /* The constant loader makes sure to load -1 into iX.w */
2606     if(vshader)
2607     {
2608         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2609         struct list *e = list_head(&priv->control_frames);
2610         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2611
2612         if(priv->loop_depth > 1) shader_addline(buffer, "PUSHA aL;\n");
2613
2614         shader_addline(buffer, "ARLC aL, %s.xywz;\n", src_name);
2615         shader_addline(buffer, "BRA loop_%u_end (LE.x);\n", control_frame->loop_no);
2616         shader_addline(buffer, "loop_%u_start:\n", control_frame->loop_no);
2617     }
2618     else
2619     {
2620         shader_addline(buffer, "REP %s;\n", src_name);
2621     }
2622 }
2623
2624 static void shader_hw_endloop(const struct wined3d_shader_instruction *ins)
2625 {
2626     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2627     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2628
2629     if(vshader)
2630     {
2631         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2632         struct list *e = list_head(&priv->control_frames);
2633         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2634
2635         shader_addline(buffer, "ARAC aL.xy, aL;\n");
2636         shader_addline(buffer, "BRA loop_%u_start (GT.x);\n", control_frame->loop_no);
2637         shader_addline(buffer, "loop_%u_end:\n", control_frame->loop_no);
2638
2639         if(priv->loop_depth > 1) shader_addline(buffer, "POPA aL;\n");
2640     }
2641     else
2642     {
2643         shader_addline(buffer, "ENDLOOP;\n");
2644     }
2645 }
2646
2647 static void shader_hw_endrep(const struct wined3d_shader_instruction *ins)
2648 {
2649     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2650     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2651
2652     if(vshader)
2653     {
2654         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2655         struct list *e = list_head(&priv->control_frames);
2656         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2657
2658         shader_addline(buffer, "ARAC aL.xy, aL;\n");
2659         shader_addline(buffer, "BRA loop_%u_start (GT.x);\n", control_frame->loop_no);
2660         shader_addline(buffer, "loop_%u_end:\n", control_frame->loop_no);
2661
2662         if(priv->loop_depth > 1) shader_addline(buffer, "POPA aL;\n");
2663     }
2664     else
2665     {
2666         shader_addline(buffer, "ENDREP;\n");
2667     }
2668 }
2669
2670 static const struct control_frame *find_last_loop(const struct shader_arb_ctx_priv *priv)
2671 {
2672     struct control_frame *control_frame;
2673
2674     LIST_FOR_EACH_ENTRY(control_frame, &priv->control_frames, struct control_frame, entry)
2675     {
2676         if(control_frame->type == LOOP || control_frame->type == REP) return control_frame;
2677     }
2678     ERR("Could not find loop for break\n");
2679     return NULL;
2680 }
2681
2682 static void shader_hw_break(const struct wined3d_shader_instruction *ins)
2683 {
2684     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2685     const struct control_frame *control_frame = find_last_loop(ins->ctx->backend_data);
2686     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2687
2688     if(vshader)
2689     {
2690         shader_addline(buffer, "BRA loop_%u_end;\n", control_frame->loop_no);
2691     }
2692     else
2693     {
2694         shader_addline(buffer, "BRK;\n");
2695     }
2696 }
2697
2698 static const char *get_compare(COMPARISON_TYPE flags)
2699 {
2700     switch (flags)
2701     {
2702         case COMPARISON_GT: return "GT";
2703         case COMPARISON_EQ: return "EQ";
2704         case COMPARISON_GE: return "GE";
2705         case COMPARISON_LT: return "LT";
2706         case COMPARISON_NE: return "NE";
2707         case COMPARISON_LE: return "LE";
2708         default:
2709             FIXME("Unrecognized comparison value: %u\n", flags);
2710             return "(\?\?)";
2711     }
2712 }
2713
2714 static COMPARISON_TYPE invert_compare(COMPARISON_TYPE flags)
2715 {
2716     switch (flags)
2717     {
2718         case COMPARISON_GT: return COMPARISON_LE;
2719         case COMPARISON_EQ: return COMPARISON_NE;
2720         case COMPARISON_GE: return COMPARISON_LT;
2721         case COMPARISON_LT: return COMPARISON_GE;
2722         case COMPARISON_NE: return COMPARISON_EQ;
2723         case COMPARISON_LE: return COMPARISON_GT;
2724         default:
2725             FIXME("Unrecognized comparison value: %u\n", flags);
2726             return -1;
2727     }
2728 }
2729
2730 static void shader_hw_breakc(const struct wined3d_shader_instruction *ins)
2731 {
2732     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2733     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2734     const struct control_frame *control_frame = find_last_loop(ins->ctx->backend_data);
2735     char src_name0[50];
2736     char src_name1[50];
2737     const char *comp = get_compare(ins->flags);
2738
2739     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name0);
2740     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name1);
2741
2742     if(vshader)
2743     {
2744         /* SUBC CC, src0, src1" works only in pixel shaders, so use TA to throw
2745          * away the subtraction result
2746          */
2747         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
2748         shader_addline(buffer, "BRA loop_%u_end (%s.x);\n", control_frame->loop_no, comp);
2749     }
2750     else
2751     {
2752         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
2753         shader_addline(buffer, "BRK (%s.x);\n", comp);
2754     }
2755 }
2756
2757 static void shader_hw_ifc(const struct wined3d_shader_instruction *ins)
2758 {
2759     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2760     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2761     struct list *e = list_head(&priv->control_frames);
2762     struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2763     const char *comp;
2764     char src_name0[50];
2765     char src_name1[50];
2766     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2767
2768     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name0);
2769     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name1);
2770
2771     if(vshader)
2772     {
2773         /* Invert the flag. We jump to the else label if the condition is NOT true */
2774         comp = get_compare(invert_compare(ins->flags));
2775         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
2776         shader_addline(buffer, "BRA ifc_%u_else (%s.x);\n", control_frame->ifc_no, comp);
2777     }
2778     else
2779     {
2780         comp = get_compare(ins->flags);
2781         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
2782         shader_addline(buffer, "IF %s.x;\n", comp);
2783     }
2784 }
2785
2786 static void shader_hw_else(const struct wined3d_shader_instruction *ins)
2787 {
2788     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2789     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2790     struct list *e = list_head(&priv->control_frames);
2791     struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2792     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2793
2794     if(vshader)
2795     {
2796         shader_addline(buffer, "BRA ifc_%u_endif;\n", control_frame->ifc_no);
2797         shader_addline(buffer, "ifc_%u_else:\n", control_frame->ifc_no);
2798         control_frame->had_else = TRUE;
2799     }
2800     else
2801     {
2802         shader_addline(buffer, "ELSE;\n");
2803     }
2804 }
2805
2806 static void shader_hw_endif(const struct wined3d_shader_instruction *ins)
2807 {
2808     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2809     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2810     struct list *e = list_head(&priv->control_frames);
2811     struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2812     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2813
2814     if(vshader)
2815     {
2816         if(control_frame->had_else)
2817         {
2818             shader_addline(buffer, "ifc_%u_endif:\n", control_frame->ifc_no);
2819         }
2820         else
2821         {
2822             shader_addline(buffer, "#No else branch. else is endif\n");
2823             shader_addline(buffer, "ifc_%u_else:\n", control_frame->ifc_no);
2824         }
2825     }
2826     else
2827     {
2828         shader_addline(buffer, "ENDIF;\n");
2829     }
2830 }
2831
2832 static void shader_hw_texldd(const struct wined3d_shader_instruction *ins)
2833 {
2834     DWORD sampler_idx = ins->src[1].reg.idx;
2835     char reg_dest[40];
2836     char reg_src[3][40];
2837     DWORD flags = TEX_DERIV;
2838
2839     shader_arb_get_dst_param(ins, &ins->dst[0], reg_dest);
2840     shader_arb_get_src_param(ins, &ins->src[0], 0, reg_src[0]);
2841     shader_arb_get_src_param(ins, &ins->src[2], 1, reg_src[1]);
2842     shader_arb_get_src_param(ins, &ins->src[3], 2, reg_src[2]);
2843
2844     if (ins->flags & WINED3DSI_TEXLD_PROJECT) flags |= TEX_PROJ;
2845     if (ins->flags & WINED3DSI_TEXLD_BIAS) flags |= TEX_BIAS;
2846
2847     shader_hw_sample(ins, sampler_idx, reg_dest, reg_src[0], flags, reg_src[1], reg_src[2]);
2848 }
2849
2850 static void shader_hw_texldl(const struct wined3d_shader_instruction *ins)
2851 {
2852     DWORD sampler_idx = ins->src[1].reg.idx;
2853     char reg_dest[40];
2854     char reg_coord[40];
2855     DWORD flags = TEX_LOD;
2856
2857     shader_arb_get_dst_param(ins, &ins->dst[0], reg_dest);
2858     shader_arb_get_src_param(ins, &ins->src[0], 0, reg_coord);
2859
2860     if (ins->flags & WINED3DSI_TEXLD_PROJECT) flags |= TEX_PROJ;
2861     if (ins->flags & WINED3DSI_TEXLD_BIAS) flags |= TEX_BIAS;
2862
2863     shader_hw_sample(ins, sampler_idx, reg_dest, reg_coord, flags, NULL, NULL);
2864 }
2865
2866 static void shader_hw_label(const struct wined3d_shader_instruction *ins)
2867 {
2868     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2869     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2870
2871     priv->in_main_func = FALSE;
2872     /* Call instructions activate the NV extensions, not labels and rets. If there is an uncalled
2873      * subroutine, don't generate a label that will make GL complain
2874      */
2875     if(priv->target_version == ARB) return;
2876
2877     shader_addline(buffer, "l%u:\n", ins->src[0].reg.idx);
2878 }
2879
2880 static void vshader_add_footer(IWineD3DVertexShaderImpl *This, struct wined3d_shader_buffer *buffer,
2881         const struct arb_vs_compile_args *args, struct shader_arb_ctx_priv *priv_ctx)
2882 {
2883     const shader_reg_maps *reg_maps = &This->baseShader.reg_maps;
2884     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *)This->baseShader.device;
2885     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
2886     unsigned int i;
2887
2888     /* The D3DRS_FOGTABLEMODE render state defines if the shader-generated fog coord is used
2889      * or if the fragment depth is used. If the fragment depth is used(FOGTABLEMODE != NONE),
2890      * the fog frag coord is thrown away. If the fog frag coord is used, but not written by
2891      * the shader, it is set to 0.0(fully fogged, since start = 1.0, end = 0.0)
2892      */
2893     if(args->super.fog_src == VS_FOG_Z) {
2894         shader_addline(buffer, "MOV result.fogcoord, TMP_OUT.z;\n");
2895     } else if (!reg_maps->fog) {
2896         /* posFixup.x is always 1.0, so we can savely use it */
2897         shader_addline(buffer, "ADD result.fogcoord, posFixup.x, -posFixup.x;\n");
2898     }
2899
2900     /* Write the final position.
2901      *
2902      * OpenGL coordinates specify the center of the pixel while d3d coords specify
2903      * the corner. The offsets are stored in z and w in posFixup. posFixup.y contains
2904      * 1.0 or -1.0 to turn the rendering upside down for offscreen rendering. PosFixup.x
2905      * contains 1.0 to allow a mad, but arb vs swizzles are too restricted for that.
2906      */
2907     shader_addline(buffer, "MUL TA, posFixup, TMP_OUT.w;\n");
2908     shader_addline(buffer, "ADD TMP_OUT.x, TMP_OUT.x, TA.z;\n");
2909     shader_addline(buffer, "MAD TMP_OUT.y, TMP_OUT.y, posFixup.y, TA.w;\n");
2910
2911     if(use_nv_clip(gl_info) && priv_ctx->target_version >= NV2)
2912     {
2913         for(i = 0; i < priv_ctx->vs_clipplanes; i++)
2914         {
2915             shader_addline(buffer, "DP4 result.clip[%u].x, TMP_OUT, state.clip[%u].plane;\n", i, i);
2916         }
2917     }
2918     else if(args->boolclip.clip_control[0])
2919     {
2920         unsigned int cur_clip = 0;
2921         char component[4] = {'x', 'y', 'z', 'w'};
2922
2923         for(i = 0; i < GL_LIMITS(clipplanes); i++)
2924         {
2925             if(args->boolclip.clip_control[1] & (1 << i))
2926             {
2927                 shader_addline(buffer, "DP4 TA.%c, TMP_OUT, state.clip[%u].plane;\n",
2928                                component[cur_clip++], i);
2929             }
2930         }
2931         switch(cur_clip)
2932         {
2933             case 0:
2934                 shader_addline(buffer, "MOV TA, -helper_const.w;\n");
2935                 break;
2936             case 1:
2937                 shader_addline(buffer, "MOV TA.yzw, -helper_const.w;\n");
2938                 break;
2939             case 2:
2940                 shader_addline(buffer, "MOV TA.zw, -helper_const.w;\n");
2941                 break;
2942             case 3:
2943                 shader_addline(buffer, "MOV TA.w, -helper_const.w;\n");
2944                 break;
2945         }
2946         shader_addline(buffer, "MOV result.texcoord[%u], TA;\n",
2947                        args->boolclip.clip_control[0] - 1);
2948     }
2949
2950     /* Z coord [0;1]->[-1;1] mapping, see comment in transform_projection in state.c
2951      * and the glsl equivalent
2952      */
2953     if(need_helper_const(gl_info)) {
2954         shader_addline(buffer, "MAD TMP_OUT.z, TMP_OUT.z, helper_const.x, -TMP_OUT.w;\n");
2955     } else {
2956         shader_addline(buffer, "ADD TMP_OUT.z, TMP_OUT.z, TMP_OUT.z;\n");
2957         shader_addline(buffer, "ADD TMP_OUT.z, TMP_OUT.z, -TMP_OUT.w;\n");
2958     }
2959
2960     shader_addline(buffer, "MOV result.position, TMP_OUT;\n");
2961
2962     priv_ctx->footer_written = TRUE;
2963 }
2964
2965 static void shader_hw_ret(const struct wined3d_shader_instruction *ins)
2966 {
2967     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2968     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2969     IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *) ins->ctx->shader;
2970     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2971
2972     if(priv->target_version == ARB) return;
2973
2974     if(vshader)
2975     {
2976         if(priv->in_main_func) vshader_add_footer((IWineD3DVertexShaderImpl *) shader, buffer, priv->cur_vs_args, priv);
2977     }
2978
2979     shader_addline(buffer, "RET;\n");
2980 }
2981
2982 static void shader_hw_call(const struct wined3d_shader_instruction *ins)
2983 {
2984     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2985     shader_addline(buffer, "CAL l%u;\n", ins->src[0].reg.idx);
2986 }
2987
2988 /* GL locking is done by the caller */
2989 static GLuint create_arb_blt_vertex_program(const struct wined3d_gl_info *gl_info)
2990 {
2991     GLuint program_id = 0;
2992     const char *blt_vprogram =
2993         "!!ARBvp1.0\n"
2994         "PARAM c[1] = { { 1, 0.5 } };\n"
2995         "MOV result.position, vertex.position;\n"
2996         "MOV result.color, c[0].x;\n"
2997         "MOV result.texcoord[0], vertex.texcoord[0];\n"
2998         "END\n";
2999
3000     GL_EXTCALL(glGenProgramsARB(1, &program_id));
3001     GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, program_id));
3002     GL_EXTCALL(glProgramStringARB(GL_VERTEX_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB, strlen(blt_vprogram), blt_vprogram));
3003
3004     if (glGetError() == GL_INVALID_OPERATION) {
3005         GLint pos;
3006         glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
3007         FIXME("Vertex program error at position %d: %s\n", pos,
3008             debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
3009     }
3010
3011     return program_id;
3012 }
3013
3014 /* GL locking is done by the caller */
3015 static GLuint create_arb_blt_fragment_program(const struct wined3d_gl_info *gl_info, enum tex_types tex_type)
3016 {
3017     GLuint program_id = 0;
3018     static const char * const blt_fprograms[tex_type_count] =
3019     {
3020         /* tex_1d */
3021         NULL,
3022         /* tex_2d */
3023         "!!ARBfp1.0\n"
3024         "TEMP R0;\n"
3025         "TEX R0.x, fragment.texcoord[0], texture[0], 2D;\n"
3026         "MOV result.depth.z, R0.x;\n"
3027         "END\n",
3028         /* tex_3d */
3029         NULL,
3030         /* tex_cube */
3031         "!!ARBfp1.0\n"
3032         "TEMP R0;\n"
3033         "TEX R0.x, fragment.texcoord[0], texture[0], CUBE;\n"
3034         "MOV result.depth.z, R0.x;\n"
3035         "END\n",
3036         /* tex_rect */
3037         "!!ARBfp1.0\n"
3038         "TEMP R0;\n"
3039         "TEX R0.x, fragment.texcoord[0], texture[0], RECT;\n"
3040         "MOV result.depth.z, R0.x;\n"
3041         "END\n",
3042     };
3043
3044     if (!blt_fprograms[tex_type])
3045     {
3046         FIXME("tex_type %#x not supported\n", tex_type);
3047         tex_type = tex_2d;
3048     }
3049
3050     GL_EXTCALL(glGenProgramsARB(1, &program_id));
3051     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, program_id));
3052     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB, strlen(blt_fprograms[tex_type]), blt_fprograms[tex_type]));
3053
3054     if (glGetError() == GL_INVALID_OPERATION) {
3055         GLint pos;
3056         glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
3057         FIXME("Fragment program error at position %d: %s\n", pos,
3058             debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
3059     }
3060
3061     return program_id;
3062 }
3063
3064 static void arbfp_add_sRGB_correction(struct wined3d_shader_buffer *buffer, const char *fragcolor,
3065         const char *tmp1, const char *tmp2, const char *tmp3, const char *tmp4, BOOL condcode)
3066 {
3067     /* Perform sRGB write correction. See GLX_EXT_framebuffer_sRGB */
3068
3069     if(condcode)
3070     {
3071         /* Sigh. MOVC CC doesn't work, so use one of the temps as dummy dest */
3072         shader_addline(buffer, "SUBC %s, %s.x, srgb_consts1.y;\n", tmp1, fragcolor);
3073         /* Calculate the > 0.0031308 case */
3074         shader_addline(buffer, "POW %s.x (GE), %s.x, srgb_consts1.z;\n", fragcolor, fragcolor);
3075         shader_addline(buffer, "POW %s.y (GE), %s.y, srgb_consts1.z;\n", fragcolor, fragcolor);
3076         shader_addline(buffer, "POW %s.z (GE), %s.z, srgb_consts1.z;\n", fragcolor, fragcolor);
3077         shader_addline(buffer, "MUL %s.xyz (GE), %s, srgb_consts1.w;\n", fragcolor, fragcolor);
3078         shader_addline(buffer, "SUB %s.xyz (GE), %s, srgb_consts2.x;\n", fragcolor, fragcolor);
3079         /* Calculate the < case */
3080         shader_addline(buffer, "MUL %s.xyz (LT), srgb_consts1.x, %s;\n", fragcolor, fragcolor);
3081     }
3082     else
3083     {
3084         /* Calculate the > 0.0031308 case */
3085         shader_addline(buffer, "POW %s.x, %s.x, srgb_consts1.z;\n", tmp1, fragcolor);
3086         shader_addline(buffer, "POW %s.y, %s.y, srgb_consts1.z;\n", tmp1, fragcolor);
3087         shader_addline(buffer, "POW %s.z, %s.z, srgb_consts1.z;\n", tmp1, fragcolor);
3088         shader_addline(buffer, "MUL %s, %s, srgb_consts1.w;\n", tmp1, tmp1);
3089         shader_addline(buffer, "SUB %s, %s, srgb_consts2.x;\n", tmp1, tmp1);
3090         /* Calculate the < case */
3091         shader_addline(buffer, "MUL %s, srgb_consts1.x, %s;\n", tmp2, fragcolor);
3092         /* Get 1.0 / 0.0 masks for > 0.0031308 and < 0.0031308 */
3093         shader_addline(buffer, "SLT %s, srgb_consts1.y, %s;\n", tmp3, fragcolor);
3094         shader_addline(buffer, "SGE %s, srgb_consts1.y, %s;\n", tmp4, fragcolor);
3095         /* Store the components > 0.0031308 in the destination */
3096         shader_addline(buffer, "MUL %s.xyz, %s, %s;\n", fragcolor, tmp1, tmp3);
3097         /* Add the components that are < 0.0031308 */
3098         shader_addline(buffer, "MAD %s.xyz, %s, %s, %s;\n", fragcolor, tmp2, tmp4, fragcolor);
3099         /* Move everything into result.color at once. Nvidia hardware cannot handle partial
3100         * result.color writes(.rgb first, then .a), or handle overwriting already written
3101         * components. The assembler uses a temporary register in this case, which is usually
3102         * not allocated from one of our registers that were used earlier.
3103         */
3104     }
3105     shader_addline(buffer, "MOV result.color, %s;\n", fragcolor);
3106     /* [0.0;1.0] clamping. Not needed, this is done implicitly */
3107 }
3108
3109 static const DWORD *find_loop_control_values(IWineD3DBaseShaderImpl *This, DWORD idx)
3110 {
3111     const local_constant *constant;
3112
3113     LIST_FOR_EACH_ENTRY(constant, &This->baseShader.constantsI, local_constant, entry)
3114     {
3115         if (constant->idx == idx)
3116         {
3117             return constant->value;
3118         }
3119     }
3120     return NULL;
3121 }
3122
3123 static void init_ps_input(const IWineD3DPixelShaderImpl *This, const struct arb_ps_compile_args *args,
3124                           struct shader_arb_ctx_priv *priv)
3125 {
3126     const char *texcoords[8] =
3127     {
3128         "fragment.texcoord[0]", "fragment.texcoord[1]", "fragment.texcoord[2]", "fragment.texcoord[3]",
3129         "fragment.texcoord[4]", "fragment.texcoord[5]", "fragment.texcoord[6]", "fragment.texcoord[7]"
3130     };
3131     unsigned int i;
3132     const struct wined3d_shader_signature_element *sig = This->input_signature;
3133     const char *semantic_name;
3134     DWORD semantic_idx;
3135
3136     switch(args->super.vp_mode)
3137     {
3138         case pretransformed:
3139         case fixedfunction:
3140             /* The pixelshader has to collect the varyings on its own. In any case properly load
3141              * color0 and color1. In the case of pretransformed vertices also load texcoords. Set
3142              * other attribs to 0.0.
3143              *
3144              * For fixedfunction this behavior is correct, according to the tests. For pretransformed
3145              * we'd either need a replacement shader that can load other attribs like BINORMAL, or
3146              * load the texcoord attrib pointers to match the pixel shader signature
3147              */
3148             for(i = 0; i < MAX_REG_INPUT; i++)
3149             {
3150                 semantic_name = sig[i].semantic_name;
3151                 semantic_idx = sig[i].semantic_idx;
3152                 if(semantic_name == NULL) continue;
3153
3154                 if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_COLOR))
3155                 {
3156                     if(semantic_idx == 0) priv->ps_input[i] = "fragment.color.primary";
3157                     else if(semantic_idx == 1) priv->ps_input[i] = "fragment.color.secondary";
3158                     else priv->ps_input[i] = "0.0";
3159                 }
3160                 else if(args->super.vp_mode == fixedfunction)
3161                 {
3162                     priv->ps_input[i] = "0.0";
3163                 }
3164                 else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_TEXCOORD))
3165                 {
3166                     if(semantic_idx < 8) priv->ps_input[i] = texcoords[semantic_idx];
3167                     else priv->ps_input[i] = "0.0";
3168                 }
3169                 else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_FOG))
3170                 {
3171                     if(semantic_idx == 0) priv->ps_input[i] = "fragment.fogcoord";
3172                     else priv->ps_input[i] = "0.0";
3173                 }
3174                 else
3175                 {
3176                     priv->ps_input[i] = "0.0";
3177                 }
3178
3179                 TRACE("v%u, semantic %s%u is %s\n", i, semantic_name, semantic_idx, priv->ps_input[i]);
3180             }
3181             break;
3182
3183         case vertexshader:
3184             /* That one is easy. The vertex shaders provide v0-v7 in fragment.texcoord and v8 and v9 in
3185              * fragment.color
3186              */
3187             for(i = 0; i < 8; i++)
3188             {
3189                 priv->ps_input[i] = texcoords[i];
3190             }
3191             priv->ps_input[8] = "fragment.color.primary";
3192             priv->ps_input[9] = "fragment.color.secondary";
3193             break;
3194     }
3195 }
3196
3197 /* GL locking is done by the caller */
3198 static GLuint shader_arb_generate_pshader(IWineD3DPixelShaderImpl *This, struct wined3d_shader_buffer *buffer,
3199         const struct arb_ps_compile_args *args, struct arb_ps_compiled_shader *compiled)
3200 {
3201     const shader_reg_maps* reg_maps = &This->baseShader.reg_maps;
3202     CONST DWORD *function = This->baseShader.function;
3203     const struct wined3d_gl_info *gl_info = &((IWineD3DDeviceImpl *)This->baseShader.device)->adapter->gl_info;
3204     const local_constant *lconst;
3205     GLuint retval;
3206     char fragcolor[16];
3207     DWORD *lconst_map = local_const_mapping((IWineD3DBaseShaderImpl *) This), next_local, cur;
3208     struct shader_arb_ctx_priv priv_ctx;
3209     BOOL dcl_tmp = args->super.srgb_correction, dcl_td = FALSE;
3210     BOOL want_nv_prog = FALSE;
3211     struct arb_pshader_private *shader_priv = This->backend_priv;
3212
3213     char srgbtmp[4][4];
3214     unsigned int i, found = 0;
3215
3216     for(i = 0; i < This->baseShader.limits.temporary; i++) {
3217
3218         /* Don't overwrite the color source */
3219         if(This->color0_mov && i == This->color0_reg) continue;
3220         else if(reg_maps->shader_version.major < 2 && i == 0) continue;
3221
3222         if(reg_maps->temporary[i]) {
3223             sprintf(srgbtmp[found], "R%u", i);
3224             found++;
3225             if(found == 4) break;
3226         }
3227     }
3228
3229     switch(found) {
3230         case 4: dcl_tmp = FALSE; break;
3231         case 0:
3232             sprintf(srgbtmp[0], "TA");
3233             sprintf(srgbtmp[1], "TB");
3234             sprintf(srgbtmp[2], "TC");
3235             sprintf(srgbtmp[3], "TD");
3236             dcl_td = TRUE;
3237             break;
3238         case 1:
3239             sprintf(srgbtmp[1], "TA");
3240             sprintf(srgbtmp[2], "TB");
3241             sprintf(srgbtmp[3], "TC");
3242             break;
3243         case 2:
3244             sprintf(srgbtmp[2], "TA");
3245             sprintf(srgbtmp[3], "TB");
3246             break;
3247         case 3:
3248             sprintf(srgbtmp[3], "TA");
3249             break;
3250     }
3251
3252     /*  Create the hw ARB shader */
3253     memset(&priv_ctx, 0, sizeof(priv_ctx));
3254     priv_ctx.cur_ps_args = args;
3255     priv_ctx.compiled_fprog = compiled;
3256     priv_ctx.cur_np2fixup_info = &compiled->np2fixup_info;
3257     init_ps_input(This, args, &priv_ctx);
3258     list_init(&priv_ctx.control_frames);
3259
3260     /* Avoid enabling NV_fragment_program* if we do not need it.
3261      *
3262      * Enabling GL_NV_fragment_program_option causes the driver to occupy a temporary register,
3263      * and it slows down the shader execution noticeably(about 5%). Usually our instruction emulation
3264      * is faster than what we gain from using higher native instructions. There are some things though
3265      * that cannot be emulated. In that case enable the extensions.
3266      * If the extension is enabled, instruction handlers that support both ways will use it.
3267      *
3268      * Testing shows no performance difference between OPTION NV_fragment_program2 and NV_fragment_program.
3269      * So enable the best we can get.
3270      */
3271     if(reg_maps->usesdsx || reg_maps->usesdsy || reg_maps->loop_depth > 0 || reg_maps->usestexldd ||
3272        reg_maps->usestexldl || reg_maps->usesfacing || reg_maps->usesifc || reg_maps->usescall)
3273     {
3274         want_nv_prog = TRUE;
3275     }
3276
3277     shader_addline(buffer, "!!ARBfp1.0\n");
3278     if(want_nv_prog && GL_SUPPORT(NV_FRAGMENT_PROGRAM2)) {
3279         shader_addline(buffer, "OPTION NV_fragment_program2;\n");
3280         priv_ctx.target_version = NV3;
3281     } else if(want_nv_prog && GL_SUPPORT(NV_FRAGMENT_PROGRAM_OPTION)) {
3282         shader_addline(buffer, "OPTION NV_fragment_program;\n");
3283         priv_ctx.target_version = NV2;
3284     } else {
3285         if(want_nv_prog)
3286         {
3287             /* This is an error - either we're advertising the wrong shader version, or aren't enforcing some
3288              * limits properly
3289              */
3290             ERR("The shader requires instructions that are not available in plain GL_ARB_fragment_program\n");
3291             ERR("Try GLSL\n");
3292         }
3293         priv_ctx.target_version = ARB;
3294     }
3295
3296     if(This->baseShader.reg_maps.highest_render_target > 0)
3297     {
3298         shader_addline(buffer, "OPTION ARB_draw_buffers;\n");
3299     }
3300
3301     if (reg_maps->shader_version.major < 3)
3302     {
3303         switch(args->super.fog) {
3304             case FOG_OFF:
3305                 break;
3306             case FOG_LINEAR:
3307                 shader_addline(buffer, "OPTION ARB_fog_linear;\n");
3308                 break;
3309             case FOG_EXP:
3310                 shader_addline(buffer, "OPTION ARB_fog_exp;\n");
3311                 break;
3312             case FOG_EXP2:
3313                 shader_addline(buffer, "OPTION ARB_fog_exp2;\n");
3314                 break;
3315         }
3316     }
3317
3318     /* For now always declare the temps. At least the Nvidia assembler optimizes completely
3319      * unused temps away(but occupies them for the whole shader if they're used once). Always
3320      * declaring them avoids tricky bookkeeping work
3321      */
3322     shader_addline(buffer, "TEMP TA;\n");      /* Used for modifiers */
3323     shader_addline(buffer, "TEMP TB;\n");      /* Used for modifiers */
3324     shader_addline(buffer, "TEMP TC;\n");      /* Used for modifiers */
3325     if(dcl_td) shader_addline(buffer, "TEMP TD;\n"); /* Used for sRGB writing */
3326     shader_addline(buffer, "PARAM coefdiv = { 0.5, 0.25, 0.125, 0.0625 };\n");
3327     shader_addline(buffer, "PARAM coefmul = { 2, 4, 8, 16 };\n");
3328     shader_addline(buffer, "PARAM one = { 1.0, 1.0, 1.0, 1.0 };\n");
3329
3330     if (reg_maps->shader_version.major < 2)
3331     {
3332         strcpy(fragcolor, "R0");
3333     } else {
3334         if(args->super.srgb_correction) {
3335             if(This->color0_mov) {
3336                 sprintf(fragcolor, "R%u", This->color0_reg);
3337             } else {
3338                 shader_addline(buffer, "TEMP TMP_COLOR;\n");
3339                 strcpy(fragcolor, "TMP_COLOR");
3340             }
3341         } else {
3342             strcpy(fragcolor, "result.color");
3343         }
3344     }
3345
3346     if(args->super.srgb_correction) {
3347         shader_addline(buffer, "PARAM srgb_consts1 = {%f, %f, %f, %f};\n",
3348                        srgb_mul_low, srgb_cmp, srgb_pow, srgb_mul_high);
3349         shader_addline(buffer, "PARAM srgb_consts2 = {%f, %f, %f, %f};\n",
3350                        srgb_sub_high, 0.0, 0.0, 0.0);
3351     }
3352
3353     /* Base Declarations */
3354     next_local = shader_generate_arb_declarations( (IWineD3DBaseShader*) This, reg_maps, buffer, &GLINFO_LOCATION,
3355             lconst_map, NULL, &priv_ctx);
3356
3357     for(i = 0; i < (sizeof(reg_maps->bumpmat) / sizeof(reg_maps->bumpmat[0])); i++) {
3358         if(!reg_maps->bumpmat[i]) continue;
3359
3360         cur = compiled->numbumpenvmatconsts;
3361         compiled->bumpenvmatconst[cur].const_num = WINED3D_CONST_NUM_UNUSED;
3362         compiled->bumpenvmatconst[cur].texunit = i;
3363         compiled->luminanceconst[cur].const_num = WINED3D_CONST_NUM_UNUSED;
3364         compiled->luminanceconst[cur].texunit = i;
3365
3366         /* We can fit the constants into the constant limit for sure because texbem, texbeml, bem and beml are only supported
3367          * in 1.x shaders, and GL_ARB_fragment_program has a constant limit of 24 constants. So in the worst case we're loading
3368          * 8 shader constants, 8 bump matrices and 8 luminance parameters and are perfectly fine. (No NP2 fixup on bumpmapped
3369          * textures due to conditional NP2 restrictions)
3370          *
3371          * Use local constants to load the bump env parameters, not program.env. This avoids collisions with d3d constants of
3372          * shaders in newer shader models. Since the bump env parameters have to share their space with NP2 fixup constants,
3373          * their location is shader dependent anyway and they cannot be loaded globally.
3374          */
3375         compiled->bumpenvmatconst[cur].const_num = next_local++;
3376         shader_addline(buffer, "PARAM bumpenvmat%d = program.local[%d];\n",
3377                        i, compiled->bumpenvmatconst[cur].const_num);
3378         compiled->numbumpenvmatconsts = cur + 1;
3379
3380         if(!reg_maps->luminanceparams[i]) continue;
3381
3382         compiled->luminanceconst[cur].const_num = next_local++;
3383         shader_addline(buffer, "PARAM luminance%d = program.local[%d];\n",
3384                        i, compiled->luminanceconst[cur].const_num);
3385     }
3386
3387     for(i = 0; i < MAX_CONST_I; i++)
3388     {
3389         compiled->int_consts[i] = WINED3D_CONST_NUM_UNUSED;
3390         if (reg_maps->integer_constants & (1 << i) && priv_ctx.target_version >= NV2)
3391         {
3392             const DWORD *control_values = find_loop_control_values((IWineD3DBaseShaderImpl *) This, i);
3393
3394             if(control_values)
3395             {
3396                 shader_addline(buffer, "PARAM I%u = {%u, %u, %u, -1};\n", i,
3397                                 control_values[0], control_values[1], control_values[2]);
3398             }
3399             else
3400             {
3401                 compiled->int_consts[i] = next_local;
3402                 compiled->num_int_consts++;
3403                 shader_addline(buffer, "PARAM I%u = program.local[%u];\n", i, next_local++);
3404             }
3405         }
3406     }
3407
3408     if(reg_maps->vpos || reg_maps->usesdsy)
3409     {
3410         compiled->ycorrection = next_local;
3411         shader_addline(buffer, "PARAM ycorrection = program.local[%u];\n", next_local++);
3412
3413         if(reg_maps->vpos)
3414         {
3415             shader_addline(buffer, "TEMP vpos;\n");
3416             /* ycorrection.x: Backbuffer height(onscreen) or 0(offscreen).
3417              * ycorrection.y: -1.0(onscreen), 1.0(offscreen)
3418              * ycorrection.z: 1.0
3419              * ycorrection.w: 0.0
3420              */
3421             shader_addline(buffer, "MAD vpos, fragment.position, ycorrection.zyww, ycorrection.wxww;\n");
3422             shader_addline(buffer, "FLR vpos.xy, vpos;\n");
3423         }
3424     }
3425     else
3426     {
3427         compiled->ycorrection = WINED3D_CONST_NUM_UNUSED;
3428     }
3429
3430     /* Load constants to fixup NP2 texcoords if there are still free constants left:
3431      * Constants (texture dimensions) for the NP2 fixup are loaded as local program parameters. This will consume
3432      * at most 8 (MAX_FRAGMENT_SAMPLERS / 2) parameters, which is highly unlikely, since the application had to
3433      * use 16 NP2 textures at the same time. In case that we run out of constants the fixup is simply not
3434      * applied / activated. This will probably result in wrong rendering of the texture, but will save us from
3435      * shader compilation errors and the subsequent errors when drawing with this shader. */
3436     if (priv_ctx.cur_ps_args->super.np2_fixup) {
3437
3438         struct arb_ps_np2fixup_info* const fixup = priv_ctx.cur_np2fixup_info;
3439         const WORD map = priv_ctx.cur_ps_args->super.np2_fixup;
3440         const UINT max_lconsts = gl_info->ps_arb_max_local_constants;
3441
3442         fixup->offset = next_local;
3443         fixup->super.active = 0;
3444
3445         cur = 0;
3446         for (i = 0; i < MAX_FRAGMENT_SAMPLERS; ++i) {
3447             if (!(map & (1 << i))) continue;
3448
3449             if (fixup->offset + (cur >> 1) < max_lconsts) {
3450                 fixup->super.active |= (1 << i);
3451                 fixup->super.idx[i] = cur++;
3452             } else {
3453                 FIXME("No free constant found to load NP2 fixup data into shader. "
3454                       "Sampling from this texture will probably look wrong.\n");
3455                 break;
3456             }
3457         }
3458
3459         fixup->super.num_consts = (cur + 1) >> 1;
3460         if (fixup->super.num_consts) {
3461             shader_addline(buffer, "PARAM np2fixup[%u] = { program.env[%u..%u] };\n",
3462                            fixup->super.num_consts, fixup->offset, fixup->super.num_consts + fixup->offset - 1);
3463         }
3464
3465         next_local += fixup->super.num_consts;
3466     }
3467
3468     if(shader_priv->clipplane_emulation)
3469     {
3470         shader_addline(buffer, "KIL fragment.texcoord[%u];\n", shader_priv->clipplane_emulation - 1);
3471     }
3472
3473     /* Base Shader Body */
3474     shader_generate_main((IWineD3DBaseShader *)This, buffer, reg_maps, function, &priv_ctx);
3475
3476     if(args->super.srgb_correction) {
3477         arbfp_add_sRGB_correction(buffer, fragcolor, srgbtmp[0], srgbtmp[1], srgbtmp[2], srgbtmp[3],
3478                                   priv_ctx.target_version >= NV2);
3479     } else if(reg_maps->shader_version.major < 2) {
3480         shader_addline(buffer, "MOV result.color, %s;\n", fragcolor);
3481     }
3482     shader_addline(buffer, "END\n");
3483
3484     /* TODO: change to resource.glObjectHandle or something like that */
3485     GL_EXTCALL(glGenProgramsARB(1, &retval));
3486
3487     TRACE("Creating a hw pixel shader, prg=%d\n", retval);
3488     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, retval));
3489
3490     TRACE("Created hw pixel shader, prg=%d\n", retval);
3491     /* Create the program and check for errors */
3492     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
3493                buffer->bsize, buffer->buffer));
3494
3495     if (glGetError() == GL_INVALID_OPERATION) {
3496         GLint errPos;
3497         glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &errPos);
3498         FIXME("HW PixelShader Error at position %d: %s\n",
3499               errPos, debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
3500         retval = 0;
3501     }
3502
3503     /* Load immediate constants */
3504     if(lconst_map) {
3505         LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry) {
3506             const float *value = (const float *)lconst->value;
3507             GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, lconst_map[lconst->idx], value));
3508             checkGLcall("glProgramLocalParameter4fvARB");
3509         }
3510         HeapFree(GetProcessHeap(), 0, lconst_map);
3511     }
3512
3513     return retval;
3514 }
3515
3516 static int compare_sig(const struct wined3d_shader_signature_element *sig1, const struct wined3d_shader_signature_element *sig2)
3517 {
3518     unsigned int i;
3519     int ret;
3520
3521     for(i = 0; i < MAX_REG_INPUT; i++)
3522     {
3523         if(sig1[i].semantic_name == NULL || sig2[i].semantic_name == NULL)
3524         {
3525             /* Compare pointers, not contents. One string is NULL(element does not exist), the other one is not NULL */
3526             if(sig1[i].semantic_name != sig2[i].semantic_name) return sig1[i].semantic_name < sig2[i].semantic_name ? -1 : 1;
3527             continue;
3528         }
3529
3530         ret = strcmp(sig1[i].semantic_name, sig2[i].semantic_name);
3531         if(ret != 0) return ret;
3532         if(sig1[i].semantic_idx    != sig2[i].semantic_idx)    return sig1[i].semantic_idx    < sig2[i].semantic_idx    ? -1 : 1;
3533         if(sig1[i].sysval_semantic != sig2[i].sysval_semantic) return sig1[i].sysval_semantic < sig2[i].sysval_semantic ? -1 : 1;
3534         if(sig1[i].component_type  != sig2[i].component_type)  return sig1[i].sysval_semantic < sig2[i].component_type  ? -1 : 1;
3535         if(sig1[i].register_idx    != sig2[i].register_idx)    return sig1[i].register_idx    < sig2[i].register_idx    ? -1 : 1;
3536         if(sig1[i].mask            != sig2->mask)              return sig1[i].mask            < sig2[i].mask            ? -1 : 1;
3537     }
3538     return 0;
3539 }
3540
3541 static struct wined3d_shader_signature_element *clone_sig(const struct wined3d_shader_signature_element *sig)
3542 {
3543     struct wined3d_shader_signature_element *new;
3544     int i;
3545     char *name;
3546
3547     new = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*new) * MAX_REG_INPUT);
3548     for(i = 0; i < MAX_REG_INPUT; i++)
3549     {
3550         if(sig[i].semantic_name == NULL)
3551         {
3552             continue;
3553         }
3554
3555         new[i] = sig[i];
3556         /* Clone the semantic string */
3557         name = HeapAlloc(GetProcessHeap(), 0, strlen(sig[i].semantic_name) + 1);
3558         strcpy(name, sig[i].semantic_name);
3559         new[i].semantic_name = name;
3560     }
3561     return new;
3562 }
3563
3564 static DWORD find_input_signature(struct shader_arb_priv *priv, const struct wined3d_shader_signature_element *sig)
3565 {
3566     struct wine_rb_entry *entry = wine_rb_get(&priv->signature_tree, sig);
3567     struct ps_signature *found_sig;
3568
3569     if(entry != NULL)
3570     {
3571         found_sig = WINE_RB_ENTRY_VALUE(entry, struct ps_signature, entry);
3572         TRACE("Found existing signature %u\n", found_sig->idx);
3573         return found_sig->idx;
3574     }
3575     found_sig = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*sig));
3576     found_sig->sig = clone_sig(sig);
3577     found_sig->idx = priv->ps_sig_number++;
3578     TRACE("New signature stored and assigned number %u\n", found_sig->idx);
3579     if(wine_rb_put(&priv->signature_tree, sig, &found_sig->entry) == -1)
3580     {
3581         ERR("Failed to insert program entry.\n");
3582     }
3583     return found_sig->idx;
3584 }
3585
3586 static void init_output_registers(IWineD3DVertexShaderImpl *shader, DWORD sig_num, struct shader_arb_ctx_priv *priv_ctx,
3587                                   struct arb_vs_compiled_shader *compiled)
3588 {
3589     unsigned int i, j;
3590     static const char *texcoords[8] =
3591     {
3592         "result.texcoord[0]", "result.texcoord[1]", "result.texcoord[2]", "result.texcoord[3]",
3593         "result.texcoord[4]", "result.texcoord[5]", "result.texcoord[6]", "result.texcoord[7]"
3594     };
3595     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *) shader->baseShader.device;
3596     const struct wined3d_shader_signature_element *sig;
3597     const char *semantic_name;
3598     DWORD semantic_idx, reg_idx;
3599
3600     /* Write generic input varyings 0 to 7 to result.texcoord[], varying 8 to result.color.primary
3601      * and varying 9 to result.color.secondary
3602      */
3603     const char *decl_idx_to_string[MAX_REG_INPUT] =
3604     {
3605         texcoords[0], texcoords[1], texcoords[2], texcoords[3],
3606         texcoords[4], texcoords[5], texcoords[6], texcoords[7],
3607         "result.color.primary", "result.color.secondary"
3608     };
3609
3610     if(sig_num == ~0)
3611     {
3612         TRACE("Pixel shader uses builtin varyings\n");
3613         /* Map builtins to builtins */
3614         for(i = 0; i < 8; i++)
3615         {
3616             priv_ctx->texcrd_output[i] = texcoords[i];
3617         }
3618         priv_ctx->color_output[0] = "result.color.primary";
3619         priv_ctx->color_output[1] = "result.color.secondary";
3620         priv_ctx->fog_output = "result.fogcoord";
3621
3622         /* Map declared regs to builtins. Use "TA" to /dev/null unread output */
3623         for(i = 0; i < (sizeof(shader->output_signature) / sizeof(*shader->output_signature)); i++)
3624         {
3625             semantic_name = shader->output_signature[i].semantic_name;
3626             if(semantic_name == NULL) continue;
3627
3628             if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_POSITION))
3629             {
3630                 TRACE("o%u is TMP_OUT\n", i);
3631                 if(shader->output_signature[i].semantic_idx == 0) priv_ctx->vs_output[i] = "TMP_OUT";
3632                 else priv_ctx->vs_output[i] = "TA";
3633             }
3634             else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_PSIZE))
3635             {
3636                 TRACE("o%u is result.pointsize\n", i);
3637                 if(shader->output_signature[i].semantic_idx == 0) priv_ctx->vs_output[i] = "result.pointsize";
3638                 else priv_ctx->vs_output[i] = "TA";
3639             }
3640             else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_COLOR))
3641             {
3642                 TRACE("o%u is result.color.?, idx %u\n", i, shader->output_signature[i].semantic_idx);
3643                 if(shader->output_signature[i].semantic_idx == 0) priv_ctx->vs_output[i] = "result.color.primary";
3644                 else if(shader->output_signature[i].semantic_idx == 1) priv_ctx->vs_output[i] = "result.color.secondary";
3645                 else priv_ctx->vs_output[i] = "TA";
3646             }
3647             else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_TEXCOORD))
3648             {
3649                 TRACE("o%u is %s\n", i, texcoords[shader->output_signature[i].semantic_idx]);
3650                 if(shader->output_signature[i].semantic_idx >= 8) priv_ctx->vs_output[i] = "TA";
3651                 else priv_ctx->vs_output[i] = texcoords[shader->output_signature[i].semantic_idx];
3652             }
3653             else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_FOG))
3654             {
3655                 TRACE("o%u is result.fogcoord\n", i);
3656                 if(shader->output_signature[i].semantic_idx > 0) priv_ctx->vs_output[i] = "TA";
3657                 else priv_ctx->vs_output[i] = "result.fogcoord";
3658             }
3659             else
3660             {
3661                 priv_ctx->vs_output[i] = "TA";
3662             }
3663         }
3664         return;
3665     }
3666
3667     /* Instead of searching for the signature in the signature list, read the one from the current pixel shader.
3668      * Its maybe not the shader where the signature came from, but it is the same signature and faster to find
3669      */
3670     sig = ((IWineD3DPixelShaderImpl *)device->stateBlock->pixelShader)->input_signature;
3671     TRACE("Pixel shader uses declared varyings\n");
3672
3673     /* Map builtin to declared. /dev/null the results by default to the TA temp reg */
3674     for(i = 0; i < 8; i++)
3675     {
3676         priv_ctx->texcrd_output[i] = "TA";
3677     }
3678     priv_ctx->color_output[0] = "TA";
3679     priv_ctx->color_output[1] = "TA";
3680     priv_ctx->fog_output = "TA";
3681
3682     for(i = 0; i < MAX_REG_INPUT; i++)
3683     {
3684         semantic_name = sig[i].semantic_name;
3685         semantic_idx = sig[i].semantic_idx;
3686         reg_idx = sig[i].register_idx;
3687         if(semantic_name == NULL) continue;
3688
3689         /* If a declared input register is not written by builtin arguments, don't write to it.
3690          * GL_NV_vertex_program makes sure the input defaults to 0.0, which is correct with D3D
3691          *
3692          * Don't care about POSITION and PSIZE here - this is a builtin vertex shader, position goes
3693          * to TMP_OUT in any case
3694          */
3695         if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_TEXCOORD))
3696         {
3697             if(semantic_idx < 8) priv_ctx->texcrd_output[semantic_idx] = decl_idx_to_string[reg_idx];
3698         }
3699         else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_COLOR))
3700         {
3701             if(semantic_idx < 2) priv_ctx->color_output[semantic_idx] = decl_idx_to_string[reg_idx];
3702         }
3703         else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_FOG))
3704         {
3705             if(semantic_idx == 0) priv_ctx->fog_output = decl_idx_to_string[reg_idx];
3706         }
3707         else
3708         {
3709             continue;
3710         }
3711
3712         if(strcmp(decl_idx_to_string[reg_idx], "result.color.primary") == 0 ||
3713            strcmp(decl_idx_to_string[reg_idx], "result.color.secondary") == 0)
3714         {
3715             compiled->need_color_unclamp = TRUE;
3716         }
3717     }
3718
3719     /* Map declared to declared */
3720     for(i = 0; i < (sizeof(shader->output_signature) / sizeof(*shader->output_signature)); i++)
3721     {
3722         /* Write unread output to TA to throw them away */
3723         priv_ctx->vs_output[i] = "TA";
3724         semantic_name = shader->output_signature[i].semantic_name;
3725         if(semantic_name == NULL)
3726         {
3727             continue;
3728         }
3729
3730         if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_POSITION) &&
3731            shader->output_signature[i].semantic_idx == 0)
3732         {
3733             priv_ctx->vs_output[i] = "TMP_OUT";
3734             continue;
3735         }
3736         else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_PSIZE) &&
3737            shader->output_signature[i].semantic_idx == 0)
3738         {
3739             priv_ctx->vs_output[i] = "result.pointsize";
3740             continue;
3741         }
3742
3743         for(j = 0; j < MAX_REG_INPUT; j++)
3744         {
3745             if(sig[j].semantic_name == NULL)
3746             {
3747                 continue;
3748             }
3749
3750             if(strcmp(sig[j].semantic_name, semantic_name) == 0 &&
3751                sig[j].semantic_idx == shader->output_signature[i].semantic_idx)
3752             {
3753                 priv_ctx->vs_output[i] = decl_idx_to_string[sig[j].register_idx];
3754
3755                 if(strcmp(priv_ctx->vs_output[i], "result.color.primary") == 0 ||
3756                    strcmp(priv_ctx->vs_output[i], "result.color.secondary") == 0)
3757                 {
3758                     compiled->need_color_unclamp = TRUE;
3759                 }
3760             }
3761         }
3762     }
3763 }
3764
3765 /* GL locking is done by the caller */
3766 static GLuint shader_arb_generate_vshader(IWineD3DVertexShaderImpl *This, struct wined3d_shader_buffer *buffer,
3767         const struct arb_vs_compile_args *args, struct arb_vs_compiled_shader *compiled)
3768 {
3769     const shader_reg_maps *reg_maps = &This->baseShader.reg_maps;
3770     CONST DWORD *function = This->baseShader.function;
3771     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *)This->baseShader.device;
3772     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
3773     const local_constant *lconst;
3774     GLuint ret;
3775     DWORD next_local, *lconst_map = local_const_mapping((IWineD3DBaseShaderImpl *) This);
3776     struct shader_arb_ctx_priv priv_ctx;
3777     unsigned int i;
3778
3779     memset(&priv_ctx, 0, sizeof(priv_ctx));
3780     priv_ctx.cur_vs_args = args;
3781     list_init(&priv_ctx.control_frames);
3782     init_output_registers(This, args->ps_signature, &priv_ctx, compiled);
3783
3784     /*  Create the hw ARB shader */
3785     shader_addline(buffer, "!!ARBvp1.0\n");
3786
3787     /* Always enable the NV extension if available. Unlike fragment shaders, there is no
3788      * mesurable performance penalty, and we can always make use of it for clipplanes.
3789      */
3790     if(GL_SUPPORT(NV_VERTEX_PROGRAM3)) {
3791         shader_addline(buffer, "OPTION NV_vertex_program3;\n");
3792         priv_ctx.target_version = NV3;
3793         shader_addline(buffer, "ADDRESS aL;\n");
3794     } else if(GL_SUPPORT(NV_VERTEX_PROGRAM2_OPTION)) {
3795         shader_addline(buffer, "OPTION NV_vertex_program2;\n");
3796         priv_ctx.target_version = NV2;
3797         shader_addline(buffer, "ADDRESS aL;\n");
3798     } else {
3799         priv_ctx.target_version = ARB;
3800     }
3801
3802     shader_addline(buffer, "TEMP TMP_OUT;\n");
3803     if(need_helper_const(gl_info)) {
3804         shader_addline(buffer, "PARAM helper_const = { 2.0, -1.0, %d.0, 0.0 };\n", This->rel_offset);
3805     }
3806     if(need_mova_const((IWineD3DBaseShader *) This, gl_info)) {
3807         shader_addline(buffer, "PARAM mova_const = { 0.5, 0.0, 2.0, 1.0 };\n");
3808         shader_addline(buffer, "TEMP A0_SHADOW;\n");
3809     }
3810
3811     shader_addline(buffer, "TEMP TA;\n");
3812
3813     /* Base Declarations */
3814     next_local = shader_generate_arb_declarations( (IWineD3DBaseShader*) This, reg_maps, buffer, &GLINFO_LOCATION,
3815             lconst_map, &priv_ctx.vs_clipplanes, &priv_ctx);
3816
3817     for(i = 0; i < MAX_CONST_I; i++)
3818     {
3819         compiled->int_consts[i] = WINED3D_CONST_NUM_UNUSED;
3820         if(reg_maps->integer_constants & (1 << i) && priv_ctx.target_version >= NV2)
3821         {
3822             const DWORD *control_values = find_loop_control_values((IWineD3DBaseShaderImpl *) This, i);
3823
3824             if(control_values)
3825             {
3826                 shader_addline(buffer, "PARAM I%u = {%u, %u, %u, -1};\n", i,
3827                                 control_values[0], control_values[1], control_values[2]);
3828             }
3829             else
3830             {
3831                 compiled->int_consts[i] = next_local;
3832                 compiled->num_int_consts++;
3833                 shader_addline(buffer, "PARAM I%u = program.local[%u];\n", i, next_local++);
3834             }
3835         }
3836     }
3837
3838     /* We need a constant to fixup the final position */
3839     shader_addline(buffer, "PARAM posFixup = program.local[%u];\n", next_local);
3840     compiled->pos_fixup = next_local++;
3841
3842     /* Initialize output parameters. GL_ARB_vertex_program does not require special initialization values
3843      * for output parameters. D3D in theory does not do that either, but some applications depend on a
3844      * proper initialization of the secondary color, and programs using the fixed function pipeline without
3845      * a replacement shader depend on the texcoord.w being set properly.
3846      *
3847      * GL_NV_vertex_program defines that all output values are initialized to {0.0, 0.0, 0.0, 1.0}. This
3848      * assertion is in effect even when using GL_ARB_vertex_program without any NV specific additions. So
3849      * skip this if NV_vertex_program is supported. Otherwise, initialize the secondary color. For the tex-
3850      * coords, we have a flag in the opengl caps. Many cards do not require the texcoord being set, and
3851      * this can eat a number of instructions, so skip it unless this cap is set as well
3852      */
3853     if(!GL_SUPPORT(NV_VERTEX_PROGRAM)) {
3854         shader_addline(buffer, "MOV result.color.secondary, -helper_const.wwwy;\n");
3855
3856         if ((GLINFO_LOCATION).quirks & WINED3D_QUIRK_SET_TEXCOORD_W && !device->frag_pipe->ffp_proj_control)
3857         {
3858             int i;
3859             for(i = 0; i < min(8, MAX_REG_TEXCRD); i++) {
3860                 if(This->baseShader.reg_maps.texcoord_mask[i] != 0 &&
3861                 This->baseShader.reg_maps.texcoord_mask[i] != WINED3DSP_WRITEMASK_ALL) {
3862                     shader_addline(buffer, "MOV result.texcoord[%u].w, -helper_const.y;\n", i);
3863                 }
3864             }
3865         }
3866     }
3867
3868     /* The shader starts with the main function */
3869     priv_ctx.in_main_func = TRUE;
3870     /* Base Shader Body */
3871     shader_generate_main((IWineD3DBaseShader *)This, buffer, reg_maps, function, &priv_ctx);
3872
3873     if(!priv_ctx.footer_written) vshader_add_footer(This, buffer, args, &priv_ctx);
3874
3875     shader_addline(buffer, "END\n");
3876
3877     /* TODO: change to resource.glObjectHandle or something like that */
3878     GL_EXTCALL(glGenProgramsARB(1, &ret));
3879
3880     TRACE("Creating a hw vertex shader, prg=%d\n", ret);
3881     GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, ret));
3882
3883     TRACE("Created hw vertex shader, prg=%d\n", ret);
3884     /* Create the program and check for errors */
3885     GL_EXTCALL(glProgramStringARB(GL_VERTEX_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
3886                buffer->bsize, buffer->buffer));
3887
3888     if (glGetError() == GL_INVALID_OPERATION) {
3889         GLint errPos;
3890         glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &errPos);
3891         FIXME("HW VertexShader Error at position %d: %s\n",
3892               errPos, debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
3893         ret = -1;
3894     } else {
3895         /* Load immediate constants */
3896         if(lconst_map) {
3897             LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry) {
3898                 const float *value = (const float *)lconst->value;
3899                 GL_EXTCALL(glProgramLocalParameter4fvARB(GL_VERTEX_PROGRAM_ARB, lconst_map[lconst->idx], value));
3900             }
3901         }
3902     }
3903     HeapFree(GetProcessHeap(), 0, lconst_map);
3904
3905     return ret;
3906 }
3907
3908 static void find_clip_texcoord(IWineD3DPixelShaderImpl *ps)
3909 {
3910     struct arb_pshader_private *shader_priv = ps->backend_priv;
3911     int i;
3912     const struct wined3d_gl_info *gl_info = &((IWineD3DDeviceImpl *)ps->baseShader.device)->adapter->gl_info;
3913
3914     /* See if we can use fragment.texcoord[7] for clipplane emulation
3915      *
3916      * Don't do this if it is not supported, or fragment.texcoord[7] is used
3917      */
3918     if(ps->baseShader.reg_maps.shader_version.major < 3)
3919     {
3920         for(i = GL_LIMITS(texture_stages); i > 0; i--)
3921         {
3922             if(!ps->baseShader.reg_maps.texcoord[i - 1])
3923             {
3924                 shader_priv->clipplane_emulation = i;
3925                 return;
3926             }
3927         }
3928         WARN("Did not find a free clip reg(2.0)\n");
3929     }
3930     else
3931     {
3932         for(i = GL_LIMITS(texture_stages); i > 0; i--)
3933         {
3934             if(!(ps->baseShader.reg_maps.input_registers & (1 << (i - 1))))
3935             {
3936                 shader_priv->clipplane_emulation = i;
3937                 return;
3938             }
3939         }
3940         WARN("Did not find a free clip reg(3.0)\n");
3941     }
3942 }
3943
3944 /* GL locking is done by the caller */
3945 static struct arb_ps_compiled_shader *find_arb_pshader(IWineD3DPixelShaderImpl *shader, const struct arb_ps_compile_args *args)
3946 {
3947     UINT i;
3948     DWORD new_size;
3949     struct arb_ps_compiled_shader *new_array;
3950     struct wined3d_shader_buffer buffer;
3951     struct arb_pshader_private *shader_data;
3952     GLuint ret;
3953
3954     if(!shader->backend_priv) {
3955         IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *) shader->baseShader.device;
3956         struct shader_arb_priv *priv = device->shader_priv;
3957
3958         shader->backend_priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data));
3959         shader_data = shader->backend_priv;
3960         shader_data->clamp_consts = shader->baseShader.reg_maps.shader_version.major == 1;
3961
3962         if(shader->baseShader.reg_maps.shader_version.major < 3) shader_data->input_signature_idx = ~0;
3963         else shader_data->input_signature_idx = find_input_signature(priv, shader->input_signature);
3964
3965         shader_data->has_signature_idx = TRUE;
3966         TRACE("Shader got assigned input signature index %u\n", shader_data->input_signature_idx);
3967
3968         if(!device->vs_clipping) find_clip_texcoord(shader);
3969     }
3970     shader_data = shader->backend_priv;
3971
3972     /* Usually we have very few GL shaders for each d3d shader(just 1 or maybe 2),
3973      * so a linear search is more performant than a hashmap or a binary search
3974      * (cache coherency etc)
3975      */
3976     for(i = 0; i < shader_data->num_gl_shaders; i++) {
3977         if(memcmp(&shader_data->gl_shaders[i].args, args, sizeof(*args)) == 0) {
3978             return &shader_data->gl_shaders[i];
3979         }
3980     }
3981
3982     TRACE("No matching GL shader found, compiling a new shader\n");
3983     if(shader_data->shader_array_size == shader_data->num_gl_shaders) {
3984         if (shader_data->num_gl_shaders)
3985         {
3986             new_size = shader_data->shader_array_size + max(1, shader_data->shader_array_size / 2);
3987             new_array = HeapReAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, shader_data->gl_shaders,
3988                                     new_size * sizeof(*shader_data->gl_shaders));
3989         } else {
3990             new_array = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data->gl_shaders));
3991             new_size = 1;
3992         }
3993
3994         if(!new_array) {
3995             ERR("Out of memory\n");
3996             return 0;
3997         }
3998         shader_data->gl_shaders = new_array;
3999         shader_data->shader_array_size = new_size;
4000     }
4001
4002     shader_data->gl_shaders[shader_data->num_gl_shaders].args = *args;
4003
4004     pixelshader_update_samplers(&shader->baseShader.reg_maps,
4005             ((IWineD3DDeviceImpl *)shader->baseShader.device)->stateBlock->textures);
4006
4007     if (!shader_buffer_init(&buffer))
4008     {
4009         ERR("Failed to initialize shader buffer.\n");
4010         return 0;
4011     }
4012
4013     ret = shader_arb_generate_pshader(shader, &buffer, args,
4014                                       &shader_data->gl_shaders[shader_data->num_gl_shaders]);
4015     shader_buffer_free(&buffer);
4016     shader_data->gl_shaders[shader_data->num_gl_shaders].prgId = ret;
4017
4018     return &shader_data->gl_shaders[shader_data->num_gl_shaders++];
4019 }
4020
4021 static inline BOOL vs_args_equal(const struct arb_vs_compile_args *stored, const struct arb_vs_compile_args *new,
4022                                  const DWORD use_map, BOOL skip_int) {
4023     if((stored->super.swizzle_map & use_map) != new->super.swizzle_map) return FALSE;
4024     if(stored->super.fog_src != new->super.fog_src) return FALSE;
4025     if(stored->boolclip_compare != new->boolclip_compare) return FALSE;
4026     if(stored->ps_signature != new->ps_signature) return FALSE;
4027     if(stored->vertex_samplers_compare != new->vertex_samplers_compare) return FALSE;
4028     if(skip_int) return TRUE;
4029
4030     return memcmp(stored->loop_ctrl, new->loop_ctrl, sizeof(stored->loop_ctrl)) == 0;
4031 }
4032
4033 static struct arb_vs_compiled_shader *find_arb_vshader(IWineD3DVertexShaderImpl *shader, const struct arb_vs_compile_args *args)
4034 {
4035     UINT i;
4036     DWORD new_size;
4037     struct arb_vs_compiled_shader *new_array;
4038     DWORD use_map = ((IWineD3DDeviceImpl *)shader->baseShader.device)->strided_streams.use_map;
4039     struct wined3d_shader_buffer buffer;
4040     struct arb_vshader_private *shader_data;
4041     GLuint ret;
4042     const struct wined3d_gl_info *gl_info = &((IWineD3DDeviceImpl *)shader->baseShader.device)->adapter->gl_info;
4043
4044     if(!shader->backend_priv) {
4045         shader->backend_priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data));
4046     }
4047     shader_data = shader->backend_priv;
4048
4049     /* Usually we have very few GL shaders for each d3d shader(just 1 or maybe 2),
4050      * so a linear search is more performant than a hashmap or a binary search
4051      * (cache coherency etc)
4052      */
4053     for(i = 0; i < shader_data->num_gl_shaders; i++) {
4054         if(vs_args_equal(&shader_data->gl_shaders[i].args, args, use_map, GL_SUPPORT(NV_VERTEX_PROGRAM2_OPTION))) {
4055             return &shader_data->gl_shaders[i];
4056         }
4057     }
4058
4059     TRACE("No matching GL shader found, compiling a new shader\n");
4060
4061     if(shader_data->shader_array_size == shader_data->num_gl_shaders) {
4062         if (shader_data->num_gl_shaders)
4063         {
4064             new_size = shader_data->shader_array_size + max(1, shader_data->shader_array_size / 2);
4065             new_array = HeapReAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, shader_data->gl_shaders,
4066                                     new_size * sizeof(*shader_data->gl_shaders));
4067         } else {
4068             new_array = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data->gl_shaders));
4069             new_size = 1;
4070         }
4071
4072         if(!new_array) {
4073             ERR("Out of memory\n");
4074             return 0;
4075         }
4076         shader_data->gl_shaders = new_array;
4077         shader_data->shader_array_size = new_size;
4078     }
4079
4080     shader_data->gl_shaders[shader_data->num_gl_shaders].args = *args;
4081
4082     if (!shader_buffer_init(&buffer))
4083     {
4084         ERR("Failed to initialize shader buffer.\n");
4085         return 0;
4086     }
4087
4088     ret = shader_arb_generate_vshader(shader, &buffer, args,
4089             &shader_data->gl_shaders[shader_data->num_gl_shaders]);
4090     shader_buffer_free(&buffer);
4091     shader_data->gl_shaders[shader_data->num_gl_shaders].prgId = ret;
4092
4093     return &shader_data->gl_shaders[shader_data->num_gl_shaders++];
4094 }
4095
4096 static inline void find_arb_ps_compile_args(IWineD3DPixelShaderImpl *shader, IWineD3DStateBlockImpl *stateblock,
4097         struct arb_ps_compile_args *args)
4098 {
4099     int i;
4100     WORD int_skip;
4101     const struct wined3d_gl_info *gl_info = &((IWineD3DDeviceImpl *)shader->baseShader.device)->adapter->gl_info;
4102     find_ps_compile_args(shader, stateblock, &args->super);
4103
4104     /* This forces all local boolean constants to 1 to make them stateblock independent */
4105     args->bools = shader->baseShader.reg_maps.local_bool_consts;
4106
4107     for(i = 0; i < MAX_CONST_B; i++)
4108     {
4109         if(stateblock->pixelShaderConstantB[i]) args->bools |= ( 1 << i);
4110     }
4111
4112     /* Skip if unused or local, or supported natively */
4113     int_skip = ~shader->baseShader.reg_maps.integer_constants | shader->baseShader.reg_maps.local_int_consts;
4114     if(int_skip == 0xffff || GL_SUPPORT(NV_FRAGMENT_PROGRAM_OPTION))
4115     {
4116         memset(&args->loop_ctrl, 0, sizeof(args->loop_ctrl));
4117         return;
4118     }
4119
4120     for(i = 0; i < MAX_CONST_I; i++)
4121     {
4122         if(int_skip & (1 << i))
4123         {
4124             args->loop_ctrl[i][0] = 0;
4125             args->loop_ctrl[i][1] = 0;
4126             args->loop_ctrl[i][2] = 0;
4127         }
4128         else
4129         {
4130             args->loop_ctrl[i][0] = stateblock->pixelShaderConstantI[i * 4];
4131             args->loop_ctrl[i][1] = stateblock->pixelShaderConstantI[i * 4 + 1];
4132             args->loop_ctrl[i][2] = stateblock->pixelShaderConstantI[i * 4 + 2];
4133         }
4134     }
4135 }
4136
4137 static inline void find_arb_vs_compile_args(IWineD3DVertexShaderImpl *shader, IWineD3DStateBlockImpl *stateblock,
4138         struct arb_vs_compile_args *args)
4139 {
4140     int i;
4141     WORD int_skip;
4142     IWineD3DDeviceImpl *dev = (IWineD3DDeviceImpl *)shader->baseShader.device;
4143     const struct wined3d_gl_info *gl_info = &dev->adapter->gl_info;
4144     find_vs_compile_args(shader, stateblock, &args->super);
4145
4146     args->boolclip_compare = 0;
4147     if(use_ps(stateblock))
4148     {
4149         IWineD3DPixelShaderImpl *ps = (IWineD3DPixelShaderImpl *) stateblock->pixelShader;
4150         struct arb_pshader_private *shader_priv = ps->backend_priv;
4151         args->ps_signature = shader_priv->input_signature_idx;
4152
4153         args->boolclip.clip_control[0] = shader_priv->clipplane_emulation;
4154     }
4155     else
4156     {
4157         args->ps_signature = ~0;
4158         if(!dev->vs_clipping)
4159         {
4160             args->boolclip.clip_control[0] = ffp_clip_emul(stateblock) ? GL_LIMITS(texture_stages) : 0;
4161         }
4162         /* Otherwise: Setting boolclip_compare set clip_control[0] to 0 */
4163     }
4164
4165     if(args->boolclip.clip_control[0])
4166     {
4167         if(stateblock->renderState[WINED3DRS_CLIPPING])
4168         {
4169             args->boolclip.clip_control[1] = stateblock->renderState[WINED3DRS_CLIPPLANEENABLE];
4170         }
4171         /* clip_control[1] was set to 0 by setting boolclip_compare to 0 */
4172     }
4173
4174     /* This forces all local boolean constants to 1 to make them stateblock independent */
4175     args->boolclip.bools = shader->baseShader.reg_maps.local_bool_consts;
4176     /* TODO: Figure out if it would be better to store bool constants as bitmasks in the stateblock */
4177     for(i = 0; i < MAX_CONST_B; i++)
4178     {
4179         if(stateblock->vertexShaderConstantB[i]) args->boolclip.bools |= ( 1 << i);
4180     }
4181
4182     args->vertex_samplers[0] = dev->texUnitMap[MAX_FRAGMENT_SAMPLERS + 0];
4183     args->vertex_samplers[1] = dev->texUnitMap[MAX_FRAGMENT_SAMPLERS + 1];
4184     args->vertex_samplers[2] = dev->texUnitMap[MAX_FRAGMENT_SAMPLERS + 2];
4185     args->vertex_samplers[3] = 0;
4186
4187     /* Skip if unused or local */
4188     int_skip = ~shader->baseShader.reg_maps.integer_constants | shader->baseShader.reg_maps.local_int_consts;
4189     if(int_skip == 0xffff || GL_SUPPORT(NV_VERTEX_PROGRAM2_OPTION)) /* This is about flow control, not clipping */
4190     {
4191         memset(&args->loop_ctrl, 0, sizeof(args->loop_ctrl));
4192         return;
4193     }
4194
4195     for(i = 0; i < MAX_CONST_I; i++)
4196     {
4197         if(int_skip & (1 << i))
4198         {
4199             args->loop_ctrl[i][0] = 0;
4200             args->loop_ctrl[i][1] = 0;
4201             args->loop_ctrl[i][2] = 0;
4202         }
4203         else
4204         {
4205             args->loop_ctrl[i][0] = stateblock->vertexShaderConstantI[i * 4];
4206             args->loop_ctrl[i][1] = stateblock->vertexShaderConstantI[i * 4 + 1];
4207             args->loop_ctrl[i][2] = stateblock->vertexShaderConstantI[i * 4 + 2];
4208         }
4209     }
4210 }
4211
4212 /* GL locking is done by the caller */
4213 static void shader_arb_select(IWineD3DDevice *iface, BOOL usePS, BOOL useVS) {
4214     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
4215     struct shader_arb_priv *priv = This->shader_priv;
4216     struct wined3d_context *context = context_get_current();
4217     const struct wined3d_gl_info *gl_info = context->gl_info;
4218     int i;
4219
4220     /* Deal with pixel shaders first so the vertex shader arg function has the input signature ready */
4221     if (usePS) {
4222         struct arb_ps_compile_args compile_args;
4223         struct arb_ps_compiled_shader *compiled;
4224         IWineD3DPixelShaderImpl *ps = (IWineD3DPixelShaderImpl *) This->stateBlock->pixelShader;
4225
4226         TRACE("Using pixel shader %p\n", This->stateBlock->pixelShader);
4227         find_arb_ps_compile_args(ps, This->stateBlock, &compile_args);
4228         compiled = find_arb_pshader(ps, &compile_args);
4229         priv->current_fprogram_id = compiled->prgId;
4230         priv->compiled_fprog = compiled;
4231
4232         /* Bind the fragment program */
4233         GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, priv->current_fprogram_id));
4234         checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, priv->current_fprogram_id);");
4235
4236         if(!priv->use_arbfp_fixed_func) {
4237             /* Enable OpenGL fragment programs */
4238             glEnable(GL_FRAGMENT_PROGRAM_ARB);
4239             checkGLcall("glEnable(GL_FRAGMENT_PROGRAM_ARB);");
4240         }
4241         TRACE("(%p) : Bound fragment program %u and enabled GL_FRAGMENT_PROGRAM_ARB\n", This, priv->current_fprogram_id);
4242
4243         /* Pixel Shader 1.x constants are clamped to [-1;1], Pixel Shader 2.0 constants are not. If switching between
4244          * a 1.x and newer shader, reload the first 8 constants
4245          */
4246         if(priv->last_ps_const_clamped != ((struct arb_pshader_private *) ps->backend_priv)->clamp_consts)
4247         {
4248             priv->last_ps_const_clamped = ((struct arb_pshader_private *) ps->backend_priv)->clamp_consts;
4249             This->highest_dirty_ps_const = max(This->highest_dirty_ps_const, 8);
4250             for(i = 0; i < 8; i++)
4251             {
4252                 context->pshader_const_dirty[i] = 1;
4253             }
4254             /* Also takes care of loading local constants */
4255             shader_arb_load_constants(context, TRUE, FALSE);
4256         }
4257         else
4258         {
4259             shader_arb_ps_local_constants(This);
4260         }
4261
4262         /* Force constant reloading for the NP2 fixup (see comment in shader_glsl_select for more info) */
4263         if (compiled->np2fixup_info.super.active) This->shader_backend->shader_load_np2fixup_constants(iface, usePS, useVS);
4264     } else if(GL_SUPPORT(ARB_FRAGMENT_PROGRAM) && !priv->use_arbfp_fixed_func) {
4265         /* Disable only if we're not using arbfp fixed function fragment processing. If this is used,
4266         * keep GL_FRAGMENT_PROGRAM_ARB enabled, and the fixed function pipeline will bind the fixed function
4267         * replacement shader
4268         */
4269         glDisable(GL_FRAGMENT_PROGRAM_ARB);
4270         checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
4271         priv->current_fprogram_id = 0;
4272     }
4273
4274     if (useVS) {
4275         struct arb_vs_compile_args compile_args;
4276         struct arb_vs_compiled_shader *compiled;
4277         IWineD3DVertexShaderImpl *vs = (IWineD3DVertexShaderImpl *) This->stateBlock->vertexShader;
4278
4279         TRACE("Using vertex shader %p\n", This->stateBlock->vertexShader);
4280         find_arb_vs_compile_args(vs, This->stateBlock, &compile_args);
4281         compiled = find_arb_vshader(vs, &compile_args);
4282         priv->current_vprogram_id = compiled->prgId;
4283         priv->compiled_vprog = compiled;
4284
4285         /* Bind the vertex program */
4286         GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->current_vprogram_id));
4287         checkGLcall("glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->current_vprogram_id);");
4288
4289         /* Enable OpenGL vertex programs */
4290         glEnable(GL_VERTEX_PROGRAM_ARB);
4291         checkGLcall("glEnable(GL_VERTEX_PROGRAM_ARB);");
4292         TRACE("(%p) : Bound vertex program %u and enabled GL_VERTEX_PROGRAM_ARB\n", This, priv->current_vprogram_id);
4293         shader_arb_vs_local_constants(This);
4294
4295         if(priv->last_vs_color_unclamp != compiled->need_color_unclamp) {
4296             priv->last_vs_color_unclamp = compiled->need_color_unclamp;
4297
4298             if (GL_SUPPORT(ARB_COLOR_BUFFER_FLOAT)) {
4299                 GL_EXTCALL(glClampColorARB(GL_CLAMP_VERTEX_COLOR_ARB, !compiled->need_color_unclamp));
4300                 checkGLcall("glClampColorARB");
4301             } else {
4302                 FIXME("vertex color clamp needs to be changed, but extension not supported.\n");
4303             }
4304         }
4305     } else if(GL_SUPPORT(ARB_VERTEX_PROGRAM)) {
4306         priv->current_vprogram_id = 0;
4307         glDisable(GL_VERTEX_PROGRAM_ARB);
4308         checkGLcall("glDisable(GL_VERTEX_PROGRAM_ARB)");
4309     }
4310 }
4311
4312 /* GL locking is done by the caller */
4313 static void shader_arb_select_depth_blt(IWineD3DDevice *iface, enum tex_types tex_type) {
4314     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
4315     struct shader_arb_priv *priv = This->shader_priv;
4316     GLuint *blt_fprogram = &priv->depth_blt_fprogram_id[tex_type];
4317     const struct wined3d_gl_info *gl_info = &This->adapter->gl_info;
4318
4319     if (!priv->depth_blt_vprogram_id) priv->depth_blt_vprogram_id = create_arb_blt_vertex_program(gl_info);
4320     GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->depth_blt_vprogram_id));
4321     glEnable(GL_VERTEX_PROGRAM_ARB);
4322
4323     if (!*blt_fprogram) *blt_fprogram = create_arb_blt_fragment_program(gl_info, tex_type);
4324     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, *blt_fprogram));
4325     glEnable(GL_FRAGMENT_PROGRAM_ARB);
4326 }
4327
4328 /* GL locking is done by the caller */
4329 static void shader_arb_deselect_depth_blt(IWineD3DDevice *iface) {
4330     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
4331     struct shader_arb_priv *priv = This->shader_priv;
4332     const struct wined3d_gl_info *gl_info = &This->adapter->gl_info;
4333
4334     if (priv->current_vprogram_id) {
4335         GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->current_vprogram_id));
4336         checkGLcall("glBindProgramARB(GL_VERTEX_PROGRAM_ARB, vertexShader->prgId);");
4337
4338         TRACE("(%p) : Bound vertex program %u and enabled GL_VERTEX_PROGRAM_ARB\n", This, priv->current_vprogram_id);
4339     } else {
4340         glDisable(GL_VERTEX_PROGRAM_ARB);
4341         checkGLcall("glDisable(GL_VERTEX_PROGRAM_ARB)");
4342     }
4343
4344     if (priv->current_fprogram_id) {
4345         GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, priv->current_fprogram_id));
4346         checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, pixelShader->prgId);");
4347
4348         TRACE("(%p) : Bound fragment program %u and enabled GL_FRAGMENT_PROGRAM_ARB\n", This, priv->current_fprogram_id);
4349     } else if(!priv->use_arbfp_fixed_func) {
4350         glDisable(GL_FRAGMENT_PROGRAM_ARB);
4351         checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
4352     }
4353 }
4354
4355 static void shader_arb_destroy(IWineD3DBaseShader *iface) {
4356     IWineD3DBaseShaderImpl *baseShader = (IWineD3DBaseShaderImpl *) iface;
4357     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *)baseShader->baseShader.device;
4358     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4359
4360     if (shader_is_pshader_version(baseShader->baseShader.reg_maps.shader_version.type))
4361     {
4362         IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *) iface;
4363         struct arb_pshader_private *shader_data = This->backend_priv;
4364         UINT i;
4365
4366         if(!shader_data) return; /* This can happen if a shader was never compiled */
4367         ENTER_GL();
4368
4369         if(shader_data->num_gl_shaders) ActivateContext(device, NULL, CTXUSAGE_RESOURCELOAD);
4370
4371         for(i = 0; i < shader_data->num_gl_shaders; i++) {
4372             GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId));
4373             checkGLcall("GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId))");
4374         }
4375         LEAVE_GL();
4376         HeapFree(GetProcessHeap(), 0, shader_data->gl_shaders);
4377         HeapFree(GetProcessHeap(), 0, shader_data);
4378         This->backend_priv = NULL;
4379     } else {
4380         IWineD3DVertexShaderImpl *This = (IWineD3DVertexShaderImpl *) iface;
4381         struct arb_vshader_private *shader_data = This->backend_priv;
4382         UINT i;
4383
4384         if(!shader_data) return; /* This can happen if a shader was never compiled */
4385         ENTER_GL();
4386
4387         if(shader_data->num_gl_shaders) ActivateContext(device, NULL, CTXUSAGE_RESOURCELOAD);
4388
4389         for(i = 0; i < shader_data->num_gl_shaders; i++) {
4390             GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId));
4391             checkGLcall("GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId))");
4392         }
4393         LEAVE_GL();
4394         HeapFree(GetProcessHeap(), 0, shader_data->gl_shaders);
4395         HeapFree(GetProcessHeap(), 0, shader_data);
4396         This->backend_priv = NULL;
4397     }
4398 }
4399
4400 static int sig_tree_compare(const void *key, const struct wine_rb_entry *entry)
4401 {
4402     struct ps_signature *e = WINE_RB_ENTRY_VALUE(entry, struct ps_signature, entry);
4403     return compare_sig(key, e->sig);
4404 }
4405
4406 struct wine_rb_functions sig_tree_functions =
4407 {
4408     wined3d_rb_alloc,
4409     wined3d_rb_realloc,
4410     wined3d_rb_free,
4411     sig_tree_compare
4412 };
4413
4414 static HRESULT shader_arb_alloc(IWineD3DDevice *iface) {
4415     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
4416     struct shader_arb_priv *priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*priv));
4417     if(wine_rb_init(&priv->signature_tree, &sig_tree_functions) == -1)
4418     {
4419         ERR("RB tree init failed\n");
4420         HeapFree(GetProcessHeap(), 0, priv);
4421         return E_OUTOFMEMORY;
4422     }
4423     This->shader_priv = priv;
4424     return WINED3D_OK;
4425 }
4426
4427 static void release_signature(struct wine_rb_entry *entry, void *context)
4428 {
4429     struct ps_signature *sig = WINE_RB_ENTRY_VALUE(entry, struct ps_signature, entry);
4430     int i;
4431     for(i = 0; i < MAX_REG_INPUT; i++)
4432     {
4433         HeapFree(GetProcessHeap(), 0, (char *) sig->sig[i].semantic_name);
4434     }
4435     HeapFree(GetProcessHeap(), 0, sig->sig);
4436     HeapFree(GetProcessHeap(), 0, sig);
4437 }
4438
4439 /* Context activation is done by the caller. */
4440 static void shader_arb_free(IWineD3DDevice *iface) {
4441     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
4442     const struct wined3d_gl_info *gl_info = &This->adapter->gl_info;
4443     struct shader_arb_priv *priv = This->shader_priv;
4444     int i;
4445
4446     ENTER_GL();
4447     if(priv->depth_blt_vprogram_id) {
4448         GL_EXTCALL(glDeleteProgramsARB(1, &priv->depth_blt_vprogram_id));
4449     }
4450     for (i = 0; i < tex_type_count; ++i) {
4451         if (priv->depth_blt_fprogram_id[i]) {
4452             GL_EXTCALL(glDeleteProgramsARB(1, &priv->depth_blt_fprogram_id[i]));
4453         }
4454     }
4455     LEAVE_GL();
4456
4457     wine_rb_destroy(&priv->signature_tree, release_signature, NULL);
4458     HeapFree(GetProcessHeap(), 0, This->shader_priv);
4459 }
4460
4461 static BOOL shader_arb_dirty_const(IWineD3DDevice *iface) {
4462     return TRUE;
4463 }
4464
4465 static void shader_arb_get_caps(WINED3DDEVTYPE devtype, const struct wined3d_gl_info *gl_info,
4466         struct shader_caps *pCaps)
4467 {
4468     /* We don't have an ARB fixed function pipeline yet, so let the none backend set its caps,
4469      * then overwrite the shader specific ones
4470      */
4471     none_shader_backend.shader_get_caps(devtype, gl_info, pCaps);
4472
4473     if(GL_SUPPORT(ARB_VERTEX_PROGRAM)) {
4474         if(GL_SUPPORT(NV_VERTEX_PROGRAM3))
4475         {
4476             pCaps->VertexShaderVersion = WINED3DVS_VERSION(3,0);
4477             TRACE_(d3d_caps)("Hardware vertex shader version 3.0 enabled (NV_VERTEX_PROGRAM3)\n");
4478         }
4479         else if(GL_LIMITS(vshader_constantsF) >= 256)
4480         {
4481             /* Shader Model 2.0 requires at least 256 vertex shader constants */
4482             pCaps->VertexShaderVersion = WINED3DVS_VERSION(2,0);
4483             TRACE_(d3d_caps)("Hardware vertex shader version 2.0 enabled (ARB_PROGRAM)\n");
4484         }
4485         else
4486         {
4487             pCaps->VertexShaderVersion = WINED3DVS_VERSION(1,1);
4488             TRACE_(d3d_caps)("Hardware vertex shader version 1.1 enabled (ARB_PROGRAM)\n");
4489         }
4490         pCaps->MaxVertexShaderConst = GL_LIMITS(vshader_constantsF);
4491     }
4492
4493     if(GL_SUPPORT(ARB_FRAGMENT_PROGRAM)) {
4494         if(GL_SUPPORT(NV_FRAGMENT_PROGRAM2))
4495         {
4496             pCaps->PixelShaderVersion    = WINED3DPS_VERSION(3,0);
4497             TRACE_(d3d_caps)("Hardware pixel shader version 3.0 enabled (NV_FRAGMENT_PROGRAM2)\n");
4498         }
4499         else if(GL_LIMITS(pshader_constantsF) >= 32)
4500         {
4501             /* Shader Model 2.0 requires at least 32 pixel shader constants */
4502             pCaps->PixelShaderVersion    = WINED3DPS_VERSION(2,0);
4503             TRACE_(d3d_caps)("Hardware pixel shader version 2.0 enabled (ARB_PROGRAM)\n");
4504         }
4505         else
4506         {
4507             pCaps->PixelShaderVersion    = WINED3DPS_VERSION(1,4);
4508             TRACE_(d3d_caps)("Hardware pixel shader version 1.4 enabled (ARB_PROGRAM)\n");
4509         }
4510         pCaps->PixelShader1xMaxValue = 8.0f;
4511         pCaps->MaxPixelShaderConst = GL_LIMITS(pshader_constantsF);
4512     }
4513
4514     pCaps->VSClipping = use_nv_clip(gl_info);
4515 }
4516
4517 static BOOL shader_arb_color_fixup_supported(struct color_fixup_desc fixup)
4518 {
4519     if (TRACE_ON(d3d_shader) && TRACE_ON(d3d))
4520     {
4521         TRACE("Checking support for color_fixup:\n");
4522         dump_color_fixup_desc(fixup);
4523     }
4524
4525     /* We support everything except YUV conversions. */
4526     if (!is_yuv_fixup(fixup))
4527     {
4528         TRACE("[OK]\n");
4529         return TRUE;
4530     }
4531
4532     TRACE("[FAILED]\n");
4533     return FALSE;
4534 }
4535
4536 static void shader_arb_add_instruction_modifiers(const struct wined3d_shader_instruction *ins) {
4537     DWORD shift;
4538     char write_mask[20], regstr[50];
4539     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
4540     BOOL is_color = FALSE;
4541     const struct wined3d_shader_dst_param *dst;
4542
4543     if (!ins->dst_count) return;
4544
4545     dst = &ins->dst[0];
4546     shift = dst->shift;
4547     if(shift == 0) return; /* Saturate alone is handled by the instructions */
4548
4549     shader_arb_get_write_mask(ins, dst, write_mask);
4550     shader_arb_get_register_name(ins, &dst->reg, regstr, &is_color);
4551
4552     /* Generate a line that does the output modifier computation
4553      * FIXME: _SAT vs shift? _SAT alone is already handled in the instructions, if this
4554      * maps problems in e.g. _d4_sat modify shader_arb_get_modifier
4555      */
4556     shader_addline(buffer, "MUL%s %s%s, %s, %s;\n", shader_arb_get_modifier(ins),
4557                    regstr, write_mask, regstr, shift_tab[shift]);
4558 }
4559
4560 static const SHADER_HANDLER shader_arb_instruction_handler_table[WINED3DSIH_TABLE_SIZE] =
4561 {
4562     /* WINED3DSIH_ABS           */ shader_hw_map2gl,
4563     /* WINED3DSIH_ADD           */ shader_hw_map2gl,
4564     /* WINED3DSIH_BEM           */ pshader_hw_bem,
4565     /* WINED3DSIH_BREAK         */ shader_hw_break,
4566     /* WINED3DSIH_BREAKC        */ shader_hw_breakc,
4567     /* WINED3DSIH_BREAKP        */ NULL,
4568     /* WINED3DSIH_CALL          */ shader_hw_call,
4569     /* WINED3DSIH_CALLNZ        */ NULL,
4570     /* WINED3DSIH_CMP           */ pshader_hw_cmp,
4571     /* WINED3DSIH_CND           */ pshader_hw_cnd,
4572     /* WINED3DSIH_CRS           */ shader_hw_map2gl,
4573     /* WINED3DSIH_DCL           */ NULL,
4574     /* WINED3DSIH_DEF           */ NULL,
4575     /* WINED3DSIH_DEFB          */ NULL,
4576     /* WINED3DSIH_DEFI          */ NULL,
4577     /* WINED3DSIH_DP2ADD        */ pshader_hw_dp2add,
4578     /* WINED3DSIH_DP3           */ shader_hw_map2gl,
4579     /* WINED3DSIH_DP4           */ shader_hw_map2gl,
4580     /* WINED3DSIH_DST           */ shader_hw_map2gl,
4581     /* WINED3DSIH_DSX           */ shader_hw_map2gl,
4582     /* WINED3DSIH_DSY           */ shader_hw_dsy,
4583     /* WINED3DSIH_ELSE          */ shader_hw_else,
4584     /* WINED3DSIH_ENDIF         */ shader_hw_endif,
4585     /* WINED3DSIH_ENDLOOP       */ shader_hw_endloop,
4586     /* WINED3DSIH_ENDREP        */ shader_hw_endrep,
4587     /* WINED3DSIH_EXP           */ shader_hw_scalar_op,
4588     /* WINED3DSIH_EXPP          */ shader_hw_scalar_op,
4589     /* WINED3DSIH_FRC           */ shader_hw_map2gl,
4590     /* WINED3DSIH_IF            */ NULL /* Hardcoded into the shader */,
4591     /* WINED3DSIH_IFC           */ shader_hw_ifc,
4592     /* WINED3DSIH_LABEL         */ shader_hw_label,
4593     /* WINED3DSIH_LIT           */ shader_hw_map2gl,
4594     /* WINED3DSIH_LOG           */ shader_hw_log_pow,
4595     /* WINED3DSIH_LOGP          */ shader_hw_log_pow,
4596     /* WINED3DSIH_LOOP          */ shader_hw_loop,
4597     /* WINED3DSIH_LRP           */ shader_hw_lrp,
4598     /* WINED3DSIH_M3x2          */ shader_hw_mnxn,
4599     /* WINED3DSIH_M3x3          */ shader_hw_mnxn,
4600     /* WINED3DSIH_M3x4          */ shader_hw_mnxn,
4601     /* WINED3DSIH_M4x3          */ shader_hw_mnxn,
4602     /* WINED3DSIH_M4x4          */ shader_hw_mnxn,
4603     /* WINED3DSIH_MAD           */ shader_hw_map2gl,
4604     /* WINED3DSIH_MAX           */ shader_hw_map2gl,
4605     /* WINED3DSIH_MIN           */ shader_hw_map2gl,
4606     /* WINED3DSIH_MOV           */ shader_hw_mov,
4607     /* WINED3DSIH_MOVA          */ shader_hw_mov,
4608     /* WINED3DSIH_MUL           */ shader_hw_map2gl,
4609     /* WINED3DSIH_NOP           */ shader_hw_nop,
4610     /* WINED3DSIH_NRM           */ shader_hw_nrm,
4611     /* WINED3DSIH_PHASE         */ NULL,
4612     /* WINED3DSIH_POW           */ shader_hw_log_pow,
4613     /* WINED3DSIH_RCP           */ shader_hw_scalar_op,
4614     /* WINED3DSIH_REP           */ shader_hw_rep,
4615     /* WINED3DSIH_RET           */ shader_hw_ret,
4616     /* WINED3DSIH_RSQ           */ shader_hw_scalar_op,
4617     /* WINED3DSIH_SETP          */ NULL,
4618     /* WINED3DSIH_SGE           */ shader_hw_map2gl,
4619     /* WINED3DSIH_SGN           */ shader_hw_sgn,
4620     /* WINED3DSIH_SINCOS        */ shader_hw_sincos,
4621     /* WINED3DSIH_SLT           */ shader_hw_map2gl,
4622     /* WINED3DSIH_SUB           */ shader_hw_map2gl,
4623     /* WINED3DSIH_TEX           */ pshader_hw_tex,
4624     /* WINED3DSIH_TEXBEM        */ pshader_hw_texbem,
4625     /* WINED3DSIH_TEXBEML       */ pshader_hw_texbem,
4626     /* WINED3DSIH_TEXCOORD      */ pshader_hw_texcoord,
4627     /* WINED3DSIH_TEXDEPTH      */ pshader_hw_texdepth,
4628     /* WINED3DSIH_TEXDP3        */ pshader_hw_texdp3,
4629     /* WINED3DSIH_TEXDP3TEX     */ pshader_hw_texdp3tex,
4630     /* WINED3DSIH_TEXKILL       */ pshader_hw_texkill,
4631     /* WINED3DSIH_TEXLDD        */ shader_hw_texldd,
4632     /* WINED3DSIH_TEXLDL        */ shader_hw_texldl,
4633     /* WINED3DSIH_TEXM3x2DEPTH  */ pshader_hw_texm3x2depth,
4634     /* WINED3DSIH_TEXM3x2PAD    */ pshader_hw_texm3x2pad,
4635     /* WINED3DSIH_TEXM3x2TEX    */ pshader_hw_texm3x2tex,
4636     /* WINED3DSIH_TEXM3x3       */ pshader_hw_texm3x3,
4637     /* WINED3DSIH_TEXM3x3DIFF   */ NULL,
4638     /* WINED3DSIH_TEXM3x3PAD    */ pshader_hw_texm3x3pad,
4639     /* WINED3DSIH_TEXM3x3SPEC   */ pshader_hw_texm3x3spec,
4640     /* WINED3DSIH_TEXM3x3TEX    */ pshader_hw_texm3x3tex,
4641     /* WINED3DSIH_TEXM3x3VSPEC  */ pshader_hw_texm3x3vspec,
4642     /* WINED3DSIH_TEXREG2AR     */ pshader_hw_texreg2ar,
4643     /* WINED3DSIH_TEXREG2GB     */ pshader_hw_texreg2gb,
4644     /* WINED3DSIH_TEXREG2RGB    */ pshader_hw_texreg2rgb,
4645 };
4646
4647 static inline BOOL get_bool_const(const struct wined3d_shader_instruction *ins, IWineD3DBaseShaderImpl *This, DWORD idx)
4648 {
4649     BOOL vshader = shader_is_vshader_version(This->baseShader.reg_maps.shader_version.type);
4650     WORD bools = 0;
4651     WORD flag = (1 << idx);
4652     const local_constant *constant;
4653     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
4654
4655     if(This->baseShader.reg_maps.local_bool_consts & flag)
4656     {
4657         /* What good is a if(bool) with a hardcoded local constant? I don't know, but handle it */
4658         LIST_FOR_EACH_ENTRY(constant, &This->baseShader.constantsB, local_constant, entry)
4659         {
4660             if (constant->idx == idx)
4661             {
4662                 return constant->value[0];
4663             }
4664         }
4665         ERR("Local constant not found\n");
4666         return FALSE;
4667     }
4668     else
4669     {
4670         if(vshader) bools = priv->cur_vs_args->boolclip.bools;
4671         else bools = priv->cur_ps_args->bools;
4672         return bools & flag;
4673     }
4674 }
4675
4676 static void get_loop_control_const(const struct wined3d_shader_instruction *ins,
4677         IWineD3DBaseShaderImpl *This, UINT idx, struct loop_control *loop_control)
4678 {
4679     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
4680
4681     /* Integer constants can either be a local constant, or they can be stored in the shader
4682      * type specific compile args. */
4683     if (This->baseShader.reg_maps.local_int_consts & (1 << idx))
4684     {
4685         const local_constant *constant;
4686
4687         LIST_FOR_EACH_ENTRY(constant, &This->baseShader.constantsI, local_constant, entry)
4688         {
4689             if (constant->idx == idx)
4690             {
4691                 loop_control->count = constant->value[0];
4692                 loop_control->start = constant->value[1];
4693                 /* Step is signed. */
4694                 loop_control->step = (int)constant->value[2];
4695                 return;
4696             }
4697         }
4698         /* If this happens the flag was set incorrectly */
4699         ERR("Local constant not found\n");
4700         loop_control->count = 0;
4701         loop_control->start = 0;
4702         loop_control->step = 0;
4703         return;
4704     }
4705
4706     switch (This->baseShader.reg_maps.shader_version.type)
4707     {
4708         case WINED3D_SHADER_TYPE_VERTEX:
4709             /* Count and aL start value are unsigned */
4710             loop_control->count = priv->cur_vs_args->loop_ctrl[idx][0];
4711             loop_control->start = priv->cur_vs_args->loop_ctrl[idx][1];
4712             /* Step is signed. */
4713             loop_control->step = ((char)priv->cur_vs_args->loop_ctrl[idx][2]);
4714             break;
4715
4716         case WINED3D_SHADER_TYPE_PIXEL:
4717             loop_control->count = priv->cur_ps_args->loop_ctrl[idx][0];
4718             loop_control->start = priv->cur_ps_args->loop_ctrl[idx][1];
4719             loop_control->step = ((char)priv->cur_ps_args->loop_ctrl[idx][2]);
4720             break;
4721
4722         default:
4723             FIXME("Unhandled shader type %#x.\n", This->baseShader.reg_maps.shader_version.type);
4724             break;
4725     }
4726 }
4727
4728 static void record_instruction(struct list *list, const struct wined3d_shader_instruction *ins)
4729 {
4730     unsigned int i;
4731     struct wined3d_shader_dst_param *dst_param = NULL;
4732     struct wined3d_shader_src_param *src_param = NULL, *rel_addr = NULL;
4733     struct recorded_instruction *rec = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*rec));
4734     if(!rec)
4735     {
4736         ERR("Out of memory\n");
4737         return;
4738     }
4739
4740     rec->ins = *ins;
4741     dst_param = HeapAlloc(GetProcessHeap(), 0, sizeof(*dst_param));
4742     if(!dst_param) goto free;
4743     *dst_param = *ins->dst;
4744     if(ins->dst->reg.rel_addr)
4745     {
4746         rel_addr = HeapAlloc(GetProcessHeap(), 0, sizeof(*dst_param->reg.rel_addr));
4747         if(!rel_addr) goto free;
4748         *rel_addr = *ins->dst->reg.rel_addr;
4749         dst_param->reg.rel_addr = rel_addr;
4750     }
4751     rec->ins.dst = dst_param;
4752
4753     src_param = HeapAlloc(GetProcessHeap(), 0, sizeof(*src_param) * ins->src_count);
4754     if(!src_param) goto free;
4755     for(i = 0; i < ins->src_count; i++)
4756     {
4757         src_param[i] = ins->src[i];
4758         if(ins->src[i].reg.rel_addr)
4759         {
4760             rel_addr = HeapAlloc(GetProcessHeap(), 0, sizeof(*rel_addr));
4761             if(!rel_addr) goto free;
4762             *rel_addr = *ins->src[i].reg.rel_addr;
4763             src_param[i].reg.rel_addr = rel_addr;
4764         }
4765     }
4766     rec->ins.src = src_param;
4767     list_add_tail(list, &rec->entry);
4768     return;
4769
4770 free:
4771     ERR("Out of memory\n");
4772     if(dst_param)
4773     {
4774         HeapFree(GetProcessHeap(), 0, (void *) dst_param->reg.rel_addr);
4775         HeapFree(GetProcessHeap(), 0, dst_param);
4776     }
4777     if(src_param)
4778     {
4779         for(i = 0; i < ins->src_count; i++)
4780         {
4781             HeapFree(GetProcessHeap(), 0, (void *) src_param[i].reg.rel_addr);
4782         }
4783         HeapFree(GetProcessHeap(), 0, src_param);
4784     }
4785     HeapFree(GetProcessHeap(), 0, rec);
4786 }
4787
4788 static void free_recorded_instruction(struct list *list)
4789 {
4790     struct recorded_instruction *rec_ins, *entry2;
4791     unsigned int i;
4792
4793     LIST_FOR_EACH_ENTRY_SAFE(rec_ins, entry2, list, struct recorded_instruction, entry)
4794     {
4795         list_remove(&rec_ins->entry);
4796         if(rec_ins->ins.dst)
4797         {
4798             HeapFree(GetProcessHeap(), 0, (void *) rec_ins->ins.dst->reg.rel_addr);
4799             HeapFree(GetProcessHeap(), 0, (void *) rec_ins->ins.dst);
4800         }
4801         if(rec_ins->ins.src)
4802         {
4803             for(i = 0; i < rec_ins->ins.src_count; i++)
4804             {
4805                 HeapFree(GetProcessHeap(), 0, (void *) rec_ins->ins.src[i].reg.rel_addr);
4806             }
4807             HeapFree(GetProcessHeap(), 0, (void *) rec_ins->ins.src);
4808         }
4809         HeapFree(GetProcessHeap(), 0, rec_ins);
4810     }
4811 }
4812
4813 static void shader_arb_handle_instruction(const struct wined3d_shader_instruction *ins) {
4814     SHADER_HANDLER hw_fct;
4815     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
4816     IWineD3DBaseShaderImpl *This = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
4817     struct control_frame *control_frame;
4818     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
4819     BOOL bool_const;
4820
4821     if(ins->handler_idx == WINED3DSIH_LOOP || ins->handler_idx == WINED3DSIH_REP)
4822     {
4823         control_frame = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*control_frame));
4824         list_add_head(&priv->control_frames, &control_frame->entry);
4825
4826         if(ins->handler_idx == WINED3DSIH_LOOP) control_frame->type = LOOP;
4827         if(ins->handler_idx == WINED3DSIH_REP) control_frame->type = REP;
4828
4829         if(priv->target_version >= NV2)
4830         {
4831             control_frame->loop_no = priv->num_loops++;
4832             priv->loop_depth++;
4833         }
4834         else
4835         {
4836             /* Don't bother recording when we're in a not used if branch */
4837             if(priv->muted)
4838             {
4839                 return;
4840             }
4841
4842             if(!priv->recording)
4843             {
4844                 list_init(&priv->record);
4845                 priv->recording = TRUE;
4846                 control_frame->outer_loop = TRUE;
4847                 get_loop_control_const(ins, This, ins->src[0].reg.idx, &control_frame->loop_control);
4848                 return; /* Instruction is handled */
4849             }
4850             /* Record this loop in the outer loop's recording */
4851         }
4852     }
4853     else if(ins->handler_idx == WINED3DSIH_ENDLOOP || ins->handler_idx == WINED3DSIH_ENDREP)
4854     {
4855         if(priv->target_version >= NV2)
4856         {
4857             /* Nothing to do. The control frame is popped after the HW instr handler */
4858         }
4859         else
4860         {
4861             struct list *e = list_head(&priv->control_frames);
4862             control_frame = LIST_ENTRY(e, struct control_frame, entry);
4863             list_remove(&control_frame->entry);
4864
4865             if(control_frame->outer_loop)
4866             {
4867                 int iteration, aL = 0;
4868                 struct list copy;
4869
4870                 /* Turn off recording before playback */
4871                 priv->recording = FALSE;
4872
4873                 /* Move the recorded instructions to a separate list and get them out of the private data
4874                  * structure. If there are nested loops, the shader_arb_handle_instruction below will
4875                  * be recorded again, thus priv->record might be overwritten
4876                  */
4877                 list_init(&copy);
4878                 list_move_tail(&copy, &priv->record);
4879                 list_init(&priv->record);
4880
4881                 if(ins->handler_idx == WINED3DSIH_ENDLOOP)
4882                 {
4883                     shader_addline(buffer, "#unrolling loop: %u iterations, aL=%u, inc %d\n",
4884                                    control_frame->loop_control.count, control_frame->loop_control.start,
4885                                    control_frame->loop_control.step);
4886                     aL = control_frame->loop_control.start;
4887                 }
4888                 else
4889                 {
4890                     shader_addline(buffer, "#unrolling rep: %u iterations\n", control_frame->loop_control.count);
4891                 }
4892
4893                 for (iteration = 0; iteration < control_frame->loop_control.count; ++iteration)
4894                 {
4895                     struct recorded_instruction *rec_ins;
4896                     if(ins->handler_idx == WINED3DSIH_ENDLOOP)
4897                     {
4898                         priv->aL = aL;
4899                         shader_addline(buffer, "#Iteration %d, aL=%d\n", iteration, aL);
4900                     }
4901                     else
4902                     {
4903                         shader_addline(buffer, "#Iteration %d\n", iteration);
4904                     }
4905
4906                     LIST_FOR_EACH_ENTRY(rec_ins, &copy, struct recorded_instruction, entry)
4907                     {
4908                         shader_arb_handle_instruction(&rec_ins->ins);
4909                     }
4910
4911                     if(ins->handler_idx == WINED3DSIH_ENDLOOP)
4912                     {
4913                         aL += control_frame->loop_control.step;
4914                     }
4915                 }
4916                 shader_addline(buffer, "#end loop/rep\n");
4917
4918                 free_recorded_instruction(&copy);
4919                 HeapFree(GetProcessHeap(), 0, control_frame);
4920                 return; /* Instruction is handled */
4921             }
4922             else
4923             {
4924                 /* This is a nested loop. Proceed to the normal recording function */
4925                 HeapFree(GetProcessHeap(), 0, control_frame);
4926             }
4927         }
4928     }
4929
4930     if(priv->recording)
4931     {
4932         record_instruction(&priv->record, ins);
4933         return;
4934     }
4935
4936     /* boolean if */
4937     if(ins->handler_idx == WINED3DSIH_IF)
4938     {
4939         control_frame = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*control_frame));
4940         list_add_head(&priv->control_frames, &control_frame->entry);
4941         control_frame->type = IF;
4942
4943         bool_const = get_bool_const(ins, This, ins->src[0].reg.idx);
4944         if(ins->src[0].modifiers == WINED3DSPSM_NOT) bool_const = !bool_const;
4945         if(!priv->muted && bool_const == FALSE)
4946         {
4947             shader_addline(buffer, "#if(FALSE){\n");
4948             priv->muted = TRUE;
4949             control_frame->muting = TRUE;
4950         }
4951         else shader_addline(buffer, "#if(TRUE) {\n");
4952
4953         return; /* Instruction is handled */
4954     }
4955     else if(ins->handler_idx == WINED3DSIH_IFC)
4956     {
4957         /* IF(bool) and if_cond(a, b) use the same ELSE and ENDIF tokens */
4958         control_frame = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*control_frame));
4959         control_frame->type = IFC;
4960         control_frame->ifc_no = priv->num_ifcs++;
4961         list_add_head(&priv->control_frames, &control_frame->entry);
4962     }
4963     else if(ins->handler_idx == WINED3DSIH_ELSE)
4964     {
4965         struct list *e = list_head(&priv->control_frames);
4966         control_frame = LIST_ENTRY(e, struct control_frame, entry);
4967
4968         if(control_frame->type == IF)
4969         {
4970             shader_addline(buffer, "#} else {\n");
4971             if(!priv->muted && !control_frame->muting)
4972             {
4973                 priv->muted = TRUE;
4974                 control_frame->muting = TRUE;
4975             }
4976             else if(control_frame->muting) priv->muted = FALSE;
4977             return; /* Instruction is handled. */
4978         }
4979         /* In case of an ifc, generate a HW shader instruction */
4980     }
4981     else if(ins->handler_idx == WINED3DSIH_ENDIF)
4982     {
4983         struct list *e = list_head(&priv->control_frames);
4984         control_frame = LIST_ENTRY(e, struct control_frame, entry);
4985
4986         if(control_frame->type == IF)
4987         {
4988             shader_addline(buffer, "#} endif\n");
4989             if(control_frame->muting) priv->muted = FALSE;
4990             list_remove(&control_frame->entry);
4991             HeapFree(GetProcessHeap(), 0, control_frame);
4992             return; /* Instruction is handled */
4993         }
4994     }
4995
4996     if(priv->muted) return;
4997
4998     /* Select handler */
4999     hw_fct = shader_arb_instruction_handler_table[ins->handler_idx];
5000
5001     /* Unhandled opcode */
5002     if (!hw_fct)
5003     {
5004         FIXME("Backend can't handle opcode %#x\n", ins->handler_idx);
5005         return;
5006     }
5007     hw_fct(ins);
5008
5009     if(ins->handler_idx == WINED3DSIH_ENDLOOP || ins->handler_idx == WINED3DSIH_ENDREP)
5010     {
5011         struct list *e = list_head(&priv->control_frames);
5012         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5013         list_remove(&control_frame->entry);
5014         HeapFree(GetProcessHeap(), 0, control_frame);
5015         priv->loop_depth--;
5016     }
5017     else if(ins->handler_idx == WINED3DSIH_ENDIF)
5018     {
5019         /* Non-ifc ENDIFs don't reach that place because of the return in the if block above */
5020         struct list *e = list_head(&priv->control_frames);
5021         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5022         list_remove(&control_frame->entry);
5023         HeapFree(GetProcessHeap(), 0, control_frame);
5024     }
5025
5026
5027     shader_arb_add_instruction_modifiers(ins);
5028 }
5029
5030 const shader_backend_t arb_program_shader_backend = {
5031     shader_arb_handle_instruction,
5032     shader_arb_select,
5033     shader_arb_select_depth_blt,
5034     shader_arb_deselect_depth_blt,
5035     shader_arb_update_float_vertex_constants,
5036     shader_arb_update_float_pixel_constants,
5037     shader_arb_load_constants,
5038     shader_arb_load_np2fixup_constants,
5039     shader_arb_destroy,
5040     shader_arb_alloc,
5041     shader_arb_free,
5042     shader_arb_dirty_const,
5043     shader_arb_get_caps,
5044     shader_arb_color_fixup_supported,
5045 };
5046
5047 /* ARB_fragment_program fixed function pipeline replacement definitions */
5048 #define ARB_FFP_CONST_TFACTOR           0
5049 #define ARB_FFP_CONST_SPECULAR_ENABLE   ((ARB_FFP_CONST_TFACTOR) + 1)
5050 #define ARB_FFP_CONST_CONSTANT(i)       ((ARB_FFP_CONST_SPECULAR_ENABLE) + 1 + i)
5051 #define ARB_FFP_CONST_BUMPMAT(i)        ((ARB_FFP_CONST_CONSTANT(7)) + 1 + i)
5052 #define ARB_FFP_CONST_LUMINANCE(i)      ((ARB_FFP_CONST_BUMPMAT(7)) + 1 + i)
5053
5054 struct arbfp_ffp_desc
5055 {
5056     struct ffp_frag_desc parent;
5057     GLuint shader;
5058     unsigned int num_textures_used;
5059 };
5060
5061 /* Context activation is done by the caller. */
5062 static void arbfp_enable(IWineD3DDevice *iface, BOOL enable) {
5063     ENTER_GL();
5064     if(enable) {
5065         glEnable(GL_FRAGMENT_PROGRAM_ARB);
5066         checkGLcall("glEnable(GL_FRAGMENT_PROGRAM_ARB)");
5067     } else {
5068         glDisable(GL_FRAGMENT_PROGRAM_ARB);
5069         checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
5070     }
5071     LEAVE_GL();
5072 }
5073
5074 static HRESULT arbfp_alloc(IWineD3DDevice *iface) {
5075     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *) iface;
5076     struct shader_arb_priv *priv;
5077     /* Share private data between the shader backend and the pipeline replacement, if both
5078      * are the arb implementation. This is needed to figure out whether ARBfp should be disabled
5079      * if no pixel shader is bound or not
5080      */
5081     if(This->shader_backend == &arb_program_shader_backend) {
5082         This->fragment_priv = This->shader_priv;
5083     } else {
5084         This->fragment_priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(struct shader_arb_priv));
5085         if(!This->fragment_priv) return E_OUTOFMEMORY;
5086     }
5087     priv = This->fragment_priv;
5088     if (wine_rb_init(&priv->fragment_shaders, &wined3d_ffp_frag_program_rb_functions) == -1)
5089     {
5090         ERR("Failed to initialize rbtree.\n");
5091         HeapFree(GetProcessHeap(), 0, This->fragment_priv);
5092         return E_OUTOFMEMORY;
5093     }
5094     priv->use_arbfp_fixed_func = TRUE;
5095     return WINED3D_OK;
5096 }
5097
5098 /* Context activation is done by the caller. */
5099 static void arbfp_free_ffpshader(struct wine_rb_entry *entry, void *context)
5100 {
5101     const struct wined3d_gl_info *gl_info = context;
5102     struct arbfp_ffp_desc *entry_arb = WINE_RB_ENTRY_VALUE(entry, struct arbfp_ffp_desc, parent.entry);
5103
5104     ENTER_GL();
5105     GL_EXTCALL(glDeleteProgramsARB(1, &entry_arb->shader));
5106     checkGLcall("glDeleteProgramsARB(1, &entry_arb->shader)");
5107     HeapFree(GetProcessHeap(), 0, entry_arb);
5108     LEAVE_GL();
5109 }
5110
5111 /* Context activation is done by the caller. */
5112 static void arbfp_free(IWineD3DDevice *iface) {
5113     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *) iface;
5114     struct shader_arb_priv *priv = This->fragment_priv;
5115
5116     wine_rb_destroy(&priv->fragment_shaders, arbfp_free_ffpshader, &This->adapter->gl_info);
5117     priv->use_arbfp_fixed_func = FALSE;
5118
5119     if(This->shader_backend != &arb_program_shader_backend) {
5120         HeapFree(GetProcessHeap(), 0, This->fragment_priv);
5121     }
5122 }
5123
5124 static void arbfp_get_caps(WINED3DDEVTYPE devtype, const struct wined3d_gl_info *gl_info, struct fragment_caps *caps)
5125 {
5126     caps->TextureOpCaps =  WINED3DTEXOPCAPS_DISABLE                     |
5127                            WINED3DTEXOPCAPS_SELECTARG1                  |
5128                            WINED3DTEXOPCAPS_SELECTARG2                  |
5129                            WINED3DTEXOPCAPS_MODULATE4X                  |
5130                            WINED3DTEXOPCAPS_MODULATE2X                  |
5131                            WINED3DTEXOPCAPS_MODULATE                    |
5132                            WINED3DTEXOPCAPS_ADDSIGNED2X                 |
5133                            WINED3DTEXOPCAPS_ADDSIGNED                   |
5134                            WINED3DTEXOPCAPS_ADD                         |
5135                            WINED3DTEXOPCAPS_SUBTRACT                    |
5136                            WINED3DTEXOPCAPS_ADDSMOOTH                   |
5137                            WINED3DTEXOPCAPS_BLENDCURRENTALPHA           |
5138                            WINED3DTEXOPCAPS_BLENDFACTORALPHA            |
5139                            WINED3DTEXOPCAPS_BLENDTEXTUREALPHA           |
5140                            WINED3DTEXOPCAPS_BLENDDIFFUSEALPHA           |
5141                            WINED3DTEXOPCAPS_BLENDTEXTUREALPHAPM         |
5142                            WINED3DTEXOPCAPS_MODULATEALPHA_ADDCOLOR      |
5143                            WINED3DTEXOPCAPS_MODULATECOLOR_ADDALPHA      |
5144                            WINED3DTEXOPCAPS_MODULATEINVCOLOR_ADDALPHA   |
5145                            WINED3DTEXOPCAPS_MODULATEINVALPHA_ADDCOLOR   |
5146                            WINED3DTEXOPCAPS_DOTPRODUCT3                 |
5147                            WINED3DTEXOPCAPS_MULTIPLYADD                 |
5148                            WINED3DTEXOPCAPS_LERP                        |
5149                            WINED3DTEXOPCAPS_BUMPENVMAP                  |
5150                            WINED3DTEXOPCAPS_BUMPENVMAPLUMINANCE;
5151
5152     /* TODO: Implement WINED3DTEXOPCAPS_PREMODULATE */
5153
5154     caps->MaxTextureBlendStages   = 8;
5155     caps->MaxSimultaneousTextures = min(GL_LIMITS(fragment_samplers), 8);
5156
5157     caps->PrimitiveMiscCaps |= WINED3DPMISCCAPS_TSSARGTEMP;
5158 }
5159 #undef GLINFO_LOCATION
5160
5161 #define GLINFO_LOCATION stateblock->wineD3DDevice->adapter->gl_info
5162 static void state_texfactor_arbfp(DWORD state, IWineD3DStateBlockImpl *stateblock, struct wined3d_context *context)
5163 {
5164     float col[4];
5165     IWineD3DDeviceImpl *device = stateblock->wineD3DDevice;
5166
5167     /* Don't load the parameter if we're using an arbfp pixel shader, otherwise we'll overwrite
5168      * application provided constants
5169      */
5170     if(device->shader_backend == &arb_program_shader_backend) {
5171         if (use_ps(stateblock)) return;
5172
5173         device = stateblock->wineD3DDevice;
5174         context->pshader_const_dirty[ARB_FFP_CONST_TFACTOR] = 1;
5175         device->highest_dirty_ps_const = max(device->highest_dirty_ps_const, ARB_FFP_CONST_TFACTOR + 1);
5176     }
5177
5178     D3DCOLORTOGLFLOAT4(stateblock->renderState[WINED3DRS_TEXTUREFACTOR], col);
5179     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_TFACTOR, col));
5180     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_TFACTOR, col)");
5181
5182 }
5183
5184 static void state_arb_specularenable(DWORD state, IWineD3DStateBlockImpl *stateblock, struct wined3d_context *context)
5185 {
5186     float col[4];
5187     IWineD3DDeviceImpl *device = stateblock->wineD3DDevice;
5188
5189     /* Don't load the parameter if we're using an arbfp pixel shader, otherwise we'll overwrite
5190      * application provided constants
5191      */
5192     if(device->shader_backend == &arb_program_shader_backend) {
5193         if (use_ps(stateblock)) return;
5194
5195         device = stateblock->wineD3DDevice;
5196         context->pshader_const_dirty[ARB_FFP_CONST_SPECULAR_ENABLE] = 1;
5197         device->highest_dirty_ps_const = max(device->highest_dirty_ps_const, ARB_FFP_CONST_SPECULAR_ENABLE + 1);
5198     }
5199
5200     if(stateblock->renderState[WINED3DRS_SPECULARENABLE]) {
5201         /* The specular color has no alpha */
5202         col[0] = 1.0f; col[1] = 1.0f;
5203         col[2] = 1.0f; col[3] = 0.0f;
5204     } else {
5205         col[0] = 0.0f; col[1] = 0.0f;
5206         col[2] = 0.0f; col[3] = 0.0f;
5207     }
5208     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_SPECULAR_ENABLE, col));
5209     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_SPECULAR_ENABLE, col)");
5210 }
5211
5212 static void set_bumpmat_arbfp(DWORD state, IWineD3DStateBlockImpl *stateblock, struct wined3d_context *context)
5213 {
5214     DWORD stage = (state - STATE_TEXTURESTAGE(0, 0)) / (WINED3D_HIGHEST_TEXTURE_STATE + 1);
5215     IWineD3DDeviceImpl *device = stateblock->wineD3DDevice;
5216     float mat[2][2];
5217
5218     if (use_ps(stateblock))
5219     {
5220         if(stage != 0 &&
5221            ((IWineD3DPixelShaderImpl *) stateblock->pixelShader)->baseShader.reg_maps.bumpmat[stage]) {
5222             /* The pixel shader has to know the bump env matrix. Do a constants update if it isn't scheduled
5223              * anyway
5224              */
5225             if(!isStateDirty(context, STATE_PIXELSHADERCONSTANT)) {
5226                 device->StateTable[STATE_PIXELSHADERCONSTANT].apply(STATE_PIXELSHADERCONSTANT, stateblock, context);
5227             }
5228         }
5229
5230         if(device->shader_backend == &arb_program_shader_backend) {
5231             /* Exit now, don't set the bumpmat below, otherwise we may overwrite pixel shader constants */
5232             return;
5233         }
5234     } else if(device->shader_backend == &arb_program_shader_backend) {
5235         context->pshader_const_dirty[ARB_FFP_CONST_BUMPMAT(stage)] = 1;
5236         device->highest_dirty_ps_const = max(device->highest_dirty_ps_const, ARB_FFP_CONST_BUMPMAT(stage) + 1);
5237     }
5238
5239     mat[0][0] = *((float *) &stateblock->textureState[stage][WINED3DTSS_BUMPENVMAT00]);
5240     mat[0][1] = *((float *) &stateblock->textureState[stage][WINED3DTSS_BUMPENVMAT01]);
5241     mat[1][0] = *((float *) &stateblock->textureState[stage][WINED3DTSS_BUMPENVMAT10]);
5242     mat[1][1] = *((float *) &stateblock->textureState[stage][WINED3DTSS_BUMPENVMAT11]);
5243
5244     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_BUMPMAT(stage), &mat[0][0]));
5245     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_BUMPMAT(stage), &mat[0][0])");
5246 }
5247
5248 static void tex_bumpenvlum_arbfp(DWORD state, IWineD3DStateBlockImpl *stateblock, struct wined3d_context *context)
5249 {
5250     DWORD stage = (state - STATE_TEXTURESTAGE(0, 0)) / (WINED3D_HIGHEST_TEXTURE_STATE + 1);
5251     IWineD3DDeviceImpl *device = stateblock->wineD3DDevice;
5252     float param[4];
5253
5254     if (use_ps(stateblock))
5255     {
5256         if(stage != 0 &&
5257            ((IWineD3DPixelShaderImpl *) stateblock->pixelShader)->baseShader.reg_maps.luminanceparams[stage]) {
5258             /* The pixel shader has to know the luminance offset. Do a constants update if it
5259              * isn't scheduled anyway
5260              */
5261             if(!isStateDirty(context, STATE_PIXELSHADERCONSTANT)) {
5262                 device->StateTable[STATE_PIXELSHADERCONSTANT].apply(STATE_PIXELSHADERCONSTANT, stateblock, context);
5263             }
5264         }
5265
5266         if(device->shader_backend == &arb_program_shader_backend) {
5267             /* Exit now, don't set the bumpmat below, otherwise we may overwrite pixel shader constants */
5268             return;
5269         }
5270     } else if(device->shader_backend == &arb_program_shader_backend) {
5271         context->pshader_const_dirty[ARB_FFP_CONST_LUMINANCE(stage)] = 1;
5272         device->highest_dirty_ps_const = max(device->highest_dirty_ps_const, ARB_FFP_CONST_LUMINANCE(stage) + 1);
5273     }
5274
5275     param[0] = *((float *) &stateblock->textureState[stage][WINED3DTSS_BUMPENVLSCALE]);
5276     param[1] = *((float *) &stateblock->textureState[stage][WINED3DTSS_BUMPENVLOFFSET]);
5277     param[2] = 0.0f;
5278     param[3] = 0.0f;
5279
5280     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_LUMINANCE(stage), param));
5281     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_LUMINANCE(stage), param)");
5282 }
5283
5284 static const char *get_argreg(struct wined3d_shader_buffer *buffer, DWORD argnum, unsigned int stage, DWORD arg)
5285 {
5286     const char *ret;
5287
5288     if(arg == ARG_UNUSED) return "unused"; /* This is the marker for unused registers */
5289
5290     switch(arg & WINED3DTA_SELECTMASK) {
5291         case WINED3DTA_DIFFUSE:
5292             ret = "fragment.color.primary"; break;
5293
5294         case WINED3DTA_CURRENT:
5295             if(stage == 0) ret = "fragment.color.primary";
5296             else ret = "ret";
5297             break;
5298
5299         case WINED3DTA_TEXTURE:
5300             switch(stage) {
5301                 case 0: ret = "tex0"; break;
5302                 case 1: ret = "tex1"; break;
5303                 case 2: ret = "tex2"; break;
5304                 case 3: ret = "tex3"; break;
5305                 case 4: ret = "tex4"; break;
5306                 case 5: ret = "tex5"; break;
5307                 case 6: ret = "tex6"; break;
5308                 case 7: ret = "tex7"; break;
5309                 default: ret = "unknown texture";
5310             }
5311             break;
5312
5313         case WINED3DTA_TFACTOR:
5314             ret = "tfactor"; break;
5315
5316         case WINED3DTA_SPECULAR:
5317             ret = "fragment.color.secondary"; break;
5318
5319         case WINED3DTA_TEMP:
5320             ret = "tempreg"; break;
5321
5322         case WINED3DTA_CONSTANT:
5323             FIXME("Implement perstage constants\n");
5324             switch(stage) {
5325                 case 0: ret = "const0"; break;
5326                 case 1: ret = "const1"; break;
5327                 case 2: ret = "const2"; break;
5328                 case 3: ret = "const3"; break;
5329                 case 4: ret = "const4"; break;
5330                 case 5: ret = "const5"; break;
5331                 case 6: ret = "const6"; break;
5332                 case 7: ret = "const7"; break;
5333                 default: ret = "unknown constant";
5334             }
5335             break;
5336
5337         default:
5338             return "unknown";
5339     }
5340
5341     if(arg & WINED3DTA_COMPLEMENT) {
5342         shader_addline(buffer, "SUB arg%u, const.x, %s;\n", argnum, ret);
5343         if(argnum == 0) ret = "arg0";
5344         if(argnum == 1) ret = "arg1";
5345         if(argnum == 2) ret = "arg2";
5346     }
5347     if(arg & WINED3DTA_ALPHAREPLICATE) {
5348         shader_addline(buffer, "MOV arg%u, %s.w;\n", argnum, ret);
5349         if(argnum == 0) ret = "arg0";
5350         if(argnum == 1) ret = "arg1";
5351         if(argnum == 2) ret = "arg2";
5352     }
5353     return ret;
5354 }
5355
5356 static void gen_ffp_instr(struct wined3d_shader_buffer *buffer, unsigned int stage, BOOL color,
5357         BOOL alpha, DWORD dst, DWORD op, DWORD dw_arg0, DWORD dw_arg1, DWORD dw_arg2)
5358 {
5359     const char *dstmask, *dstreg, *arg0, *arg1, *arg2;
5360     unsigned int mul = 1;
5361     BOOL mul_final_dest = FALSE;
5362
5363     if(color && alpha) dstmask = "";
5364     else if(color) dstmask = ".xyz";
5365     else dstmask = ".w";
5366
5367     if(dst == tempreg) dstreg = "tempreg";
5368     else dstreg = "ret";
5369
5370     arg0 = get_argreg(buffer, 0, stage, dw_arg0);
5371     arg1 = get_argreg(buffer, 1, stage, dw_arg1);
5372     arg2 = get_argreg(buffer, 2, stage, dw_arg2);
5373
5374     switch(op) {
5375         case WINED3DTOP_DISABLE:
5376             if(stage == 0) shader_addline(buffer, "MOV %s%s, fragment.color.primary;\n", dstreg, dstmask);
5377             break;
5378
5379         case WINED3DTOP_SELECTARG2:
5380             arg1 = arg2;
5381         case WINED3DTOP_SELECTARG1:
5382             shader_addline(buffer, "MOV %s%s, %s;\n", dstreg, dstmask, arg1);
5383             break;
5384
5385         case WINED3DTOP_MODULATE4X:
5386             mul = 2;
5387         case WINED3DTOP_MODULATE2X:
5388             mul *= 2;
5389             if(strcmp(dstreg, "result.color") == 0) {
5390                 dstreg = "ret";
5391                 mul_final_dest = TRUE;
5392             }
5393         case WINED3DTOP_MODULATE:
5394             shader_addline(buffer, "MUL %s%s, %s, %s;\n", dstreg, dstmask, arg1, arg2);
5395             break;
5396
5397         case WINED3DTOP_ADDSIGNED2X:
5398             mul = 2;
5399             if(strcmp(dstreg, "result.color") == 0) {
5400                 dstreg = "ret";
5401                 mul_final_dest = TRUE;
5402             }
5403         case WINED3DTOP_ADDSIGNED:
5404             shader_addline(buffer, "SUB arg2, %s, const.w;\n", arg2);
5405             arg2 = "arg2";
5406         case WINED3DTOP_ADD:
5407             shader_addline(buffer, "ADD_SAT %s%s, %s, %s;\n", dstreg, dstmask, arg1, arg2);
5408             break;
5409
5410         case WINED3DTOP_SUBTRACT:
5411             shader_addline(buffer, "SUB_SAT %s%s, %s, %s;\n", dstreg, dstmask, arg1, arg2);
5412             break;
5413
5414         case WINED3DTOP_ADDSMOOTH:
5415             shader_addline(buffer, "SUB arg1, const.x, %s;\n", arg1);
5416             shader_addline(buffer, "MAD_SAT %s%s, arg1, %s, %s;\n", dstreg, dstmask, arg2, arg1);
5417             break;
5418
5419         case WINED3DTOP_BLENDCURRENTALPHA:
5420             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_CURRENT);
5421             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
5422             break;
5423         case WINED3DTOP_BLENDFACTORALPHA:
5424             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_TFACTOR);
5425             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
5426             break;
5427         case WINED3DTOP_BLENDTEXTUREALPHA:
5428             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_TEXTURE);
5429             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
5430             break;
5431         case WINED3DTOP_BLENDDIFFUSEALPHA:
5432             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_DIFFUSE);
5433             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
5434             break;
5435
5436         case WINED3DTOP_BLENDTEXTUREALPHAPM:
5437             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_TEXTURE);
5438             shader_addline(buffer, "SUB arg0.w, const.x, %s.w;\n", arg0);
5439             shader_addline(buffer, "MAD_SAT %s%s, %s, arg0.w, %s;\n", dstreg, dstmask, arg2, arg1);
5440             break;
5441
5442         /* D3DTOP_PREMODULATE ???? */
5443
5444         case WINED3DTOP_MODULATEINVALPHA_ADDCOLOR:
5445             shader_addline(buffer, "SUB arg0.w, const.x, %s;\n", arg1);
5446             shader_addline(buffer, "MAD_SAT %s%s, arg0.w, %s, %s;\n", dstreg, dstmask, arg2, arg1);
5447             break;
5448         case WINED3DTOP_MODULATEALPHA_ADDCOLOR:
5449             shader_addline(buffer, "MAD_SAT %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg1, arg2, arg1);
5450             break;
5451         case WINED3DTOP_MODULATEINVCOLOR_ADDALPHA:
5452             shader_addline(buffer, "SUB arg0, const.x, %s;\n", arg1);
5453             shader_addline(buffer, "MAD_SAT %s%s, arg0, %s, %s.w;\n", dstreg, dstmask, arg2, arg1);
5454             break;
5455         case WINED3DTOP_MODULATECOLOR_ADDALPHA:
5456             shader_addline(buffer, "MAD_SAT %s%s, %s, %s, %s.w;\n", dstreg, dstmask, arg1, arg2, arg1);
5457             break;
5458
5459         case WINED3DTOP_DOTPRODUCT3:
5460             mul = 4;
5461             if(strcmp(dstreg, "result.color") == 0) {
5462                 dstreg = "ret";
5463                 mul_final_dest = TRUE;
5464             }
5465             shader_addline(buffer, "SUB arg1, %s, const.w;\n", arg1);
5466             shader_addline(buffer, "SUB arg2, %s, const.w;\n", arg2);
5467             shader_addline(buffer, "DP3_SAT %s%s, arg1, arg2;\n", dstreg, dstmask);
5468             break;
5469
5470         case WINED3DTOP_MULTIPLYADD:
5471             shader_addline(buffer, "MAD_SAT %s%s, %s, %s, %s;\n", dstreg, dstmask, arg1, arg2, arg0);
5472             break;
5473
5474         case WINED3DTOP_LERP:
5475             /* The msdn is not quite right here */
5476             shader_addline(buffer, "LRP %s%s, %s, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
5477             break;
5478
5479         case WINED3DTOP_BUMPENVMAP:
5480         case WINED3DTOP_BUMPENVMAPLUMINANCE:
5481             /* Those are handled in the first pass of the shader(generation pass 1 and 2) already */
5482             break;
5483
5484         default:
5485             FIXME("Unhandled texture op %08x\n", op);
5486     }
5487
5488     if(mul == 2) {
5489         shader_addline(buffer, "MUL_SAT %s%s, %s, const.y;\n", mul_final_dest ? "result.color" : dstreg, dstmask, dstreg);
5490     } else if(mul == 4) {
5491         shader_addline(buffer, "MUL_SAT %s%s, %s, const.z;\n", mul_final_dest ? "result.color" : dstreg, dstmask, dstreg);
5492     }
5493 }
5494
5495 /* The stateblock is passed for GLINFO_LOCATION */
5496 static GLuint gen_arbfp_ffp_shader(const struct ffp_frag_settings *settings, IWineD3DStateBlockImpl *stateblock)
5497 {
5498     unsigned int stage;
5499     struct wined3d_shader_buffer buffer;
5500     BOOL tex_read[MAX_TEXTURES] = {FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE};
5501     BOOL bump_used[MAX_TEXTURES] = {FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE};
5502     BOOL luminance_used[MAX_TEXTURES] = {FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE};
5503     const char *textype;
5504     const char *instr, *sat;
5505     char colorcor_dst[8];
5506     GLuint ret;
5507     DWORD arg0, arg1, arg2;
5508     BOOL tempreg_used = FALSE, tfactor_used = FALSE;
5509     BOOL op_equal;
5510     const char *final_combiner_src = "ret";
5511
5512     /* Find out which textures are read */
5513     for(stage = 0; stage < MAX_TEXTURES; stage++) {
5514         if(settings->op[stage].cop == WINED3DTOP_DISABLE) break;
5515         arg0 = settings->op[stage].carg0 & WINED3DTA_SELECTMASK;
5516         arg1 = settings->op[stage].carg1 & WINED3DTA_SELECTMASK;
5517         arg2 = settings->op[stage].carg2 & WINED3DTA_SELECTMASK;
5518         if(arg0 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
5519         if(arg1 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
5520         if(arg2 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
5521
5522         if(settings->op[stage].cop == WINED3DTOP_BLENDTEXTUREALPHA) tex_read[stage] = TRUE;
5523         if(settings->op[stage].cop == WINED3DTOP_BLENDTEXTUREALPHAPM) tex_read[stage] = TRUE;
5524         if(settings->op[stage].cop == WINED3DTOP_BUMPENVMAP) {
5525             bump_used[stage] = TRUE;
5526             tex_read[stage] = TRUE;
5527         }
5528         if(settings->op[stage].cop == WINED3DTOP_BUMPENVMAPLUMINANCE) {
5529             bump_used[stage] = TRUE;
5530             tex_read[stage] = TRUE;
5531             luminance_used[stage] = TRUE;
5532         } else if(settings->op[stage].cop == WINED3DTOP_BLENDFACTORALPHA) {
5533             tfactor_used = TRUE;
5534         }
5535
5536         if(arg0 == WINED3DTA_TFACTOR || arg1 == WINED3DTA_TFACTOR || arg2 == WINED3DTA_TFACTOR) {
5537             tfactor_used = TRUE;
5538         }
5539
5540         if(settings->op[stage].dst == tempreg) tempreg_used = TRUE;
5541         if(arg0 == WINED3DTA_TEMP || arg1 == WINED3DTA_TEMP || arg2 == WINED3DTA_TEMP) {
5542             tempreg_used = TRUE;
5543         }
5544
5545         if(settings->op[stage].aop == WINED3DTOP_DISABLE) continue;
5546         arg0 = settings->op[stage].aarg0 & WINED3DTA_SELECTMASK;
5547         arg1 = settings->op[stage].aarg1 & WINED3DTA_SELECTMASK;
5548         arg2 = settings->op[stage].aarg2 & WINED3DTA_SELECTMASK;
5549         if(arg0 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
5550         if(arg1 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
5551         if(arg2 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
5552
5553         if(arg0 == WINED3DTA_TEMP || arg1 == WINED3DTA_TEMP || arg2 == WINED3DTA_TEMP) {
5554             tempreg_used = TRUE;
5555         }
5556         if(arg0 == WINED3DTA_TFACTOR || arg1 == WINED3DTA_TFACTOR || arg2 == WINED3DTA_TFACTOR) {
5557             tfactor_used = TRUE;
5558         }
5559     }
5560
5561     /* Shader header */
5562     if (!shader_buffer_init(&buffer))
5563     {
5564         ERR("Failed to initialize shader buffer.\n");
5565         return 0;
5566     }
5567
5568     shader_addline(&buffer, "!!ARBfp1.0\n");
5569
5570     switch(settings->fog) {
5571         case FOG_OFF:                                                         break;
5572         case FOG_LINEAR: shader_addline(&buffer, "OPTION ARB_fog_linear;\n"); break;
5573         case FOG_EXP:    shader_addline(&buffer, "OPTION ARB_fog_exp;\n");    break;
5574         case FOG_EXP2:   shader_addline(&buffer, "OPTION ARB_fog_exp2;\n");   break;
5575         default: FIXME("Unexpected fog setting %d\n", settings->fog);
5576     }
5577
5578     shader_addline(&buffer, "PARAM const = {1, 2, 4, 0.5};\n");
5579     shader_addline(&buffer, "TEMP TMP;\n");
5580     shader_addline(&buffer, "TEMP ret;\n");
5581     if(tempreg_used || settings->sRGB_write) shader_addline(&buffer, "TEMP tempreg;\n");
5582     shader_addline(&buffer, "TEMP arg0;\n");
5583     shader_addline(&buffer, "TEMP arg1;\n");
5584     shader_addline(&buffer, "TEMP arg2;\n");
5585     for(stage = 0; stage < MAX_TEXTURES; stage++) {
5586         if(!tex_read[stage]) continue;
5587         shader_addline(&buffer, "TEMP tex%u;\n", stage);
5588         if(!bump_used[stage]) continue;
5589         shader_addline(&buffer, "PARAM bumpmat%u = program.env[%u];\n", stage, ARB_FFP_CONST_BUMPMAT(stage));
5590         if(!luminance_used[stage]) continue;
5591         shader_addline(&buffer, "PARAM luminance%u = program.env[%u];\n", stage, ARB_FFP_CONST_LUMINANCE(stage));
5592     }
5593     if(tfactor_used) {
5594         shader_addline(&buffer, "PARAM tfactor = program.env[%u];\n", ARB_FFP_CONST_TFACTOR);
5595     }
5596         shader_addline(&buffer, "PARAM specular_enable = program.env[%u];\n", ARB_FFP_CONST_SPECULAR_ENABLE);
5597
5598     if(settings->sRGB_write) {
5599         shader_addline(&buffer, "PARAM srgb_consts1 = {%f, %f, %f, %f};\n",
5600                        srgb_mul_low, srgb_cmp, srgb_pow, srgb_mul_high);
5601         shader_addline(&buffer, "PARAM srgb_consts2 = {%f, %f, %f, %f};\n",
5602                        srgb_sub_high, 0.0, 0.0, 0.0);
5603     }
5604
5605     if(ffp_clip_emul(stateblock) && settings->emul_clipplanes) shader_addline(&buffer, "KIL fragment.texcoord[7];\n");
5606
5607     /* Generate texture sampling instructions) */
5608     for(stage = 0; stage < MAX_TEXTURES && settings->op[stage].cop != WINED3DTOP_DISABLE; stage++) {
5609         if(!tex_read[stage]) continue;
5610
5611         switch(settings->op[stage].tex_type) {
5612             case tex_1d:                    textype = "1D";     break;
5613             case tex_2d:                    textype = "2D";     break;
5614             case tex_3d:                    textype = "3D";     break;
5615             case tex_cube:                  textype = "CUBE";   break;
5616             case tex_rect:                  textype = "RECT";   break;
5617             default: textype = "unexpected_textype";   break;
5618         }
5619
5620         if(settings->op[stage].cop == WINED3DTOP_BUMPENVMAP ||
5621            settings->op[stage].cop == WINED3DTOP_BUMPENVMAPLUMINANCE) {
5622             sat = "";
5623         } else {
5624             sat = "_SAT";
5625         }
5626
5627         if(settings->op[stage].projected == proj_none) {
5628             instr = "TEX";
5629         } else if(settings->op[stage].projected == proj_count4 ||
5630                   settings->op[stage].projected == proj_count3) {
5631             instr = "TXP";
5632         } else {
5633             FIXME("Unexpected projection mode %d\n", settings->op[stage].projected);
5634             instr = "TXP";
5635         }
5636
5637         if(stage > 0 &&
5638            (settings->op[stage - 1].cop == WINED3DTOP_BUMPENVMAP ||
5639             settings->op[stage - 1].cop == WINED3DTOP_BUMPENVMAPLUMINANCE)) {
5640             shader_addline(&buffer, "SWZ arg1, bumpmat%u, x, z, 0, 0;\n", stage - 1);
5641             shader_addline(&buffer, "DP3 ret.x, arg1, tex%u;\n", stage - 1);
5642             shader_addline(&buffer, "SWZ arg1, bumpmat%u, y, w, 0, 0;\n", stage - 1);
5643             shader_addline(&buffer, "DP3 ret.y, arg1, tex%u;\n", stage - 1);
5644
5645             /* with projective textures, texbem only divides the static texture coord, not the displacement,
5646              * so multiply the displacement with the dividing parameter before passing it to TXP
5647              */
5648             if (settings->op[stage].projected != proj_none) {
5649                 if(settings->op[stage].projected == proj_count4) {
5650                     shader_addline(&buffer, "MOV ret.w, fragment.texcoord[%u].w;\n", stage);
5651                     shader_addline(&buffer, "MUL ret.xyz, ret, fragment.texcoord[%u].w, fragment.texcoord[%u];\n", stage, stage);
5652                 } else {
5653                     shader_addline(&buffer, "MOV ret.w, fragment.texcoord[%u].z;\n", stage);
5654                     shader_addline(&buffer, "MAD ret.xyz, ret, fragment.texcoord[%u].z, fragment.texcoord[%u];\n", stage, stage);
5655                 }
5656             } else {
5657                 shader_addline(&buffer, "ADD ret, ret, fragment.texcoord[%u];\n", stage);
5658             }
5659
5660             shader_addline(&buffer, "%s%s tex%u, ret, texture[%u], %s;\n",
5661                            instr, sat, stage, stage, textype);
5662             if(settings->op[stage - 1].cop == WINED3DTOP_BUMPENVMAPLUMINANCE) {
5663                 shader_addline(&buffer, "MAD_SAT ret.x, tex%u.z, luminance%u.x, luminance%u.y;\n",
5664                                stage - 1, stage - 1, stage - 1);
5665                 shader_addline(&buffer, "MUL tex%u, tex%u, ret.x;\n", stage, stage);
5666             }
5667         } else if(settings->op[stage].projected == proj_count3) {
5668             shader_addline(&buffer, "MOV ret, fragment.texcoord[%u];\n", stage);
5669             shader_addline(&buffer, "MOV ret.w, ret.z;\n");
5670             shader_addline(&buffer, "%s%s tex%u, ret, texture[%u], %s;\n",
5671                             instr, sat, stage, stage, textype);
5672         } else {
5673             shader_addline(&buffer, "%s%s tex%u, fragment.texcoord[%u], texture[%u], %s;\n",
5674                             instr, sat, stage, stage, stage, textype);
5675         }
5676
5677         sprintf(colorcor_dst, "tex%u", stage);
5678         gen_color_correction(&buffer, colorcor_dst, WINED3DSP_WRITEMASK_ALL, "const.x", "const.y",
5679                 settings->op[stage].color_fixup);
5680     }
5681
5682     /* Generate the main shader */
5683     for(stage = 0; stage < MAX_TEXTURES; stage++) {
5684         if(settings->op[stage].cop == WINED3DTOP_DISABLE) {
5685             if(stage == 0) {
5686                 final_combiner_src = "fragment.color.primary";
5687             }
5688             break;
5689         }
5690
5691         if(settings->op[stage].cop == WINED3DTOP_SELECTARG1 &&
5692            settings->op[stage].aop == WINED3DTOP_SELECTARG1) {
5693             op_equal = settings->op[stage].carg1 == settings->op[stage].aarg1;
5694         } else if(settings->op[stage].cop == WINED3DTOP_SELECTARG1 &&
5695                   settings->op[stage].aop == WINED3DTOP_SELECTARG2) {
5696             op_equal = settings->op[stage].carg1 == settings->op[stage].aarg2;
5697         } else if(settings->op[stage].cop == WINED3DTOP_SELECTARG2 &&
5698                   settings->op[stage].aop == WINED3DTOP_SELECTARG1) {
5699             op_equal = settings->op[stage].carg2 == settings->op[stage].aarg1;
5700         } else if(settings->op[stage].cop == WINED3DTOP_SELECTARG2 &&
5701                   settings->op[stage].aop == WINED3DTOP_SELECTARG2) {
5702             op_equal = settings->op[stage].carg2 == settings->op[stage].aarg2;
5703         } else {
5704             op_equal = settings->op[stage].aop   == settings->op[stage].cop &&
5705                        settings->op[stage].carg0 == settings->op[stage].aarg0 &&
5706                        settings->op[stage].carg1 == settings->op[stage].aarg1 &&
5707                        settings->op[stage].carg2 == settings->op[stage].aarg2;
5708         }
5709
5710         if(settings->op[stage].aop == WINED3DTOP_DISABLE) {
5711             gen_ffp_instr(&buffer, stage, TRUE, FALSE, settings->op[stage].dst,
5712                           settings->op[stage].cop, settings->op[stage].carg0,
5713                           settings->op[stage].carg1, settings->op[stage].carg2);
5714             if(stage == 0) {
5715                 shader_addline(&buffer, "MOV ret.w, fragment.color.primary.w;\n");
5716             }
5717         } else if(op_equal) {
5718             gen_ffp_instr(&buffer, stage, TRUE, TRUE, settings->op[stage].dst,
5719                           settings->op[stage].cop, settings->op[stage].carg0,
5720                           settings->op[stage].carg1, settings->op[stage].carg2);
5721         } else {
5722             gen_ffp_instr(&buffer, stage, TRUE, FALSE, settings->op[stage].dst,
5723                           settings->op[stage].cop, settings->op[stage].carg0,
5724                           settings->op[stage].carg1, settings->op[stage].carg2);
5725             gen_ffp_instr(&buffer, stage, FALSE, TRUE, settings->op[stage].dst,
5726                           settings->op[stage].aop, settings->op[stage].aarg0,
5727                           settings->op[stage].aarg1, settings->op[stage].aarg2);
5728         }
5729     }
5730
5731     if(settings->sRGB_write) {
5732         shader_addline(&buffer, "MAD ret, fragment.color.secondary, specular_enable, %s;\n", final_combiner_src);
5733         arbfp_add_sRGB_correction(&buffer, "ret", "arg0", "arg1", "arg2", "tempreg", FALSE);
5734     } else {
5735         shader_addline(&buffer, "MAD result.color, fragment.color.secondary, specular_enable, %s;\n", final_combiner_src);
5736     }
5737
5738     /* Footer */
5739     shader_addline(&buffer, "END\n");
5740
5741     /* Generate the shader */
5742     GL_EXTCALL(glGenProgramsARB(1, &ret));
5743     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, ret));
5744     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB, strlen(buffer.buffer), buffer.buffer));
5745
5746     if (glGetError() == GL_INVALID_OPERATION) {
5747         GLint pos;
5748         glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
5749         FIXME("Fragment program error at position %d: %s\n", pos,
5750               debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
5751     }
5752     shader_buffer_free(&buffer);
5753     return ret;
5754 }
5755
5756 static void fragment_prog_arbfp(DWORD state, IWineD3DStateBlockImpl *stateblock, struct wined3d_context *context)
5757 {
5758     IWineD3DDeviceImpl *device = stateblock->wineD3DDevice;
5759     struct shader_arb_priv *priv = device->fragment_priv;
5760     BOOL use_pshader = use_ps(stateblock);
5761     BOOL use_vshader = use_vs(stateblock);
5762     struct ffp_frag_settings settings;
5763     const struct arbfp_ffp_desc *desc;
5764     unsigned int i;
5765
5766     TRACE("state %#x, stateblock %p, context %p\n", state, stateblock, context);
5767
5768     if(isStateDirty(context, STATE_RENDER(WINED3DRS_FOGENABLE))) {
5769         if(!use_pshader && device->shader_backend == &arb_program_shader_backend && context->last_was_pshader) {
5770             /* Reload fixed function constants since they collide with the pixel shader constants */
5771             for(i = 0; i < MAX_TEXTURES; i++) {
5772                 set_bumpmat_arbfp(STATE_TEXTURESTAGE(i, WINED3DTSS_BUMPENVMAT00), stateblock, context);
5773             }
5774             state_texfactor_arbfp(STATE_RENDER(WINED3DRS_TEXTUREFACTOR), stateblock, context);
5775             state_arb_specularenable(STATE_RENDER(WINED3DRS_SPECULARENABLE), stateblock, context);
5776         } else if(use_pshader && !isStateDirty(context, device->StateTable[STATE_VSHADER].representative)) {
5777             device->shader_backend->shader_select((IWineD3DDevice *)stateblock->wineD3DDevice, use_pshader, use_vshader);
5778         }
5779         return;
5780     }
5781
5782     if(!use_pshader) {
5783         /* Find or create a shader implementing the fixed function pipeline settings, then activate it */
5784         gen_ffp_frag_op(stateblock, &settings, FALSE);
5785         desc = (const struct arbfp_ffp_desc *)find_ffp_frag_shader(&priv->fragment_shaders, &settings);
5786         if(!desc) {
5787             struct arbfp_ffp_desc *new_desc = HeapAlloc(GetProcessHeap(), 0, sizeof(*new_desc));
5788             if (!new_desc)
5789             {
5790                 ERR("Out of memory\n");
5791                 return;
5792             }
5793             new_desc->num_textures_used = 0;
5794             for(i = 0; i < GL_LIMITS(texture_stages); i++) {
5795                 if(settings.op[i].cop == WINED3DTOP_DISABLE) break;
5796                 new_desc->num_textures_used = i;
5797             }
5798
5799             memcpy(&new_desc->parent.settings, &settings, sizeof(settings));
5800             new_desc->shader = gen_arbfp_ffp_shader(&settings, stateblock);
5801             add_ffp_frag_shader(&priv->fragment_shaders, &new_desc->parent);
5802             TRACE("Allocated fixed function replacement shader descriptor %p\n", new_desc);
5803             desc = new_desc;
5804         }
5805
5806         /* Now activate the replacement program. GL_FRAGMENT_PROGRAM_ARB is already active(however, note the
5807          * comment above the shader_select call below). If e.g. GLSL is active, the shader_select call will
5808          * deactivate it.
5809          */
5810         GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, desc->shader));
5811         checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, desc->shader)");
5812         priv->current_fprogram_id = desc->shader;
5813
5814         if(device->shader_backend == &arb_program_shader_backend && context->last_was_pshader) {
5815             /* Reload fixed function constants since they collide with the pixel shader constants */
5816             for(i = 0; i < MAX_TEXTURES; i++) {
5817                 set_bumpmat_arbfp(STATE_TEXTURESTAGE(i, WINED3DTSS_BUMPENVMAT00), stateblock, context);
5818             }
5819             state_texfactor_arbfp(STATE_RENDER(WINED3DRS_TEXTUREFACTOR), stateblock, context);
5820             state_arb_specularenable(STATE_RENDER(WINED3DRS_SPECULARENABLE), stateblock, context);
5821         }
5822         context->last_was_pshader = FALSE;
5823     } else {
5824         context->last_was_pshader = TRUE;
5825     }
5826
5827     /* Finally, select the shader. If a pixel shader is used, it will be set and enabled by the shader backend.
5828      * If this shader backend is arbfp(most likely), then it will simply overwrite the last fixed function replace-
5829      * ment shader. If the shader backend is not ARB, it currently is important that the opengl implementation
5830      * type overwrites GL_ARB_fragment_program. This is currently the case with GLSL. If we really want to use
5831      * atifs or nvrc pixel shaders with arb fragment programs we'd have to disable GL_FRAGMENT_PROGRAM_ARB here
5832      *
5833      * Don't call shader_select if the vertex shader is dirty, because it will be called later on by the vertex
5834      * shader handler
5835      */
5836     if(!isStateDirty(context, device->StateTable[STATE_VSHADER].representative)) {
5837         device->shader_backend->shader_select((IWineD3DDevice *)stateblock->wineD3DDevice, use_pshader, use_vshader);
5838
5839         if (!isStateDirty(context, STATE_VERTEXSHADERCONSTANT) && (use_vshader || use_pshader)) {
5840             device->StateTable[STATE_VERTEXSHADERCONSTANT].apply(STATE_VERTEXSHADERCONSTANT, stateblock, context);
5841         }
5842     }
5843     if(use_pshader) {
5844         device->StateTable[STATE_PIXELSHADERCONSTANT].apply(STATE_PIXELSHADERCONSTANT, stateblock, context);
5845     }
5846 }
5847
5848 /* We can't link the fog states to the fragment state directly since the vertex pipeline links them
5849  * to FOGENABLE. A different linking in different pipeline parts can't be expressed in the combined
5850  * state table, so we need to handle that with a forwarding function. The other invisible side effect
5851  * is that changing the fog start and fog end(which links to FOGENABLE in vertex) results in the
5852  * fragment_prog_arbfp function being called because FOGENABLE is dirty, which calls this function here
5853  */
5854 static void state_arbfp_fog(DWORD state, IWineD3DStateBlockImpl *stateblock, struct wined3d_context *context)
5855 {
5856     enum fogsource new_source;
5857
5858     TRACE("state %#x, stateblock %p, context %p\n", state, stateblock, context);
5859
5860     if(!isStateDirty(context, STATE_PIXELSHADER)) {
5861         fragment_prog_arbfp(state, stateblock, context);
5862     }
5863
5864     if(!stateblock->renderState[WINED3DRS_FOGENABLE]) return;
5865
5866     if(stateblock->renderState[WINED3DRS_FOGTABLEMODE] == WINED3DFOG_NONE) {
5867         if(use_vs(stateblock)) {
5868             new_source = FOGSOURCE_VS;
5869         } else {
5870             if(stateblock->renderState[WINED3DRS_FOGVERTEXMODE] == WINED3DFOG_NONE || context->last_was_rhw) {
5871                 new_source = FOGSOURCE_COORD;
5872             } else {
5873                 new_source = FOGSOURCE_FFP;
5874             }
5875         }
5876     } else {
5877         new_source = FOGSOURCE_FFP;
5878     }
5879     if(new_source != context->fog_source) {
5880         context->fog_source = new_source;
5881         state_fogstartend(STATE_RENDER(WINED3DRS_FOGSTART), stateblock, context);
5882     }
5883 }
5884
5885 static void textransform(DWORD state, IWineD3DStateBlockImpl *stateblock, struct wined3d_context *context)
5886 {
5887     if(!isStateDirty(context, STATE_PIXELSHADER)) {
5888         fragment_prog_arbfp(state, stateblock, context);
5889     }
5890 }
5891
5892 #undef GLINFO_LOCATION
5893
5894 static const struct StateEntryTemplate arbfp_fragmentstate_template[] = {
5895     {STATE_RENDER(WINED3DRS_TEXTUREFACTOR),               { STATE_RENDER(WINED3DRS_TEXTUREFACTOR),              state_texfactor_arbfp   }, WINED3D_GL_EXT_NONE             },
5896     {STATE_TEXTURESTAGE(0, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5897     {STATE_TEXTURESTAGE(0, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5898     {STATE_TEXTURESTAGE(0, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5899     {STATE_TEXTURESTAGE(0, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5900     {STATE_TEXTURESTAGE(0, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5901     {STATE_TEXTURESTAGE(0, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5902     {STATE_TEXTURESTAGE(0, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5903     {STATE_TEXTURESTAGE(0, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5904     {STATE_TEXTURESTAGE(0, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5905     {STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5906     {STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5907     {STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5908     {STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5909     {STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
5910     {STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
5911     {STATE_TEXTURESTAGE(1, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5912     {STATE_TEXTURESTAGE(1, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5913     {STATE_TEXTURESTAGE(1, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5914     {STATE_TEXTURESTAGE(1, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5915     {STATE_TEXTURESTAGE(1, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5916     {STATE_TEXTURESTAGE(1, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5917     {STATE_TEXTURESTAGE(1, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5918     {STATE_TEXTURESTAGE(1, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5919     {STATE_TEXTURESTAGE(1, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5920     {STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5921     {STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5922     {STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5923     {STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5924     {STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
5925     {STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
5926     {STATE_TEXTURESTAGE(2, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5927     {STATE_TEXTURESTAGE(2, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5928     {STATE_TEXTURESTAGE(2, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5929     {STATE_TEXTURESTAGE(2, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5930     {STATE_TEXTURESTAGE(2, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5931     {STATE_TEXTURESTAGE(2, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5932     {STATE_TEXTURESTAGE(2, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5933     {STATE_TEXTURESTAGE(2, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5934     {STATE_TEXTURESTAGE(2, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5935     {STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5936     {STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5937     {STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5938     {STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5939     {STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
5940     {STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
5941     {STATE_TEXTURESTAGE(3, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5942     {STATE_TEXTURESTAGE(3, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5943     {STATE_TEXTURESTAGE(3, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5944     {STATE_TEXTURESTAGE(3, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5945     {STATE_TEXTURESTAGE(3, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5946     {STATE_TEXTURESTAGE(3, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5947     {STATE_TEXTURESTAGE(3, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5948     {STATE_TEXTURESTAGE(3, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5949     {STATE_TEXTURESTAGE(3, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5950     {STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5951     {STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5952     {STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5953     {STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5954     {STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
5955     {STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
5956     {STATE_TEXTURESTAGE(4, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5957     {STATE_TEXTURESTAGE(4, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5958     {STATE_TEXTURESTAGE(4, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5959     {STATE_TEXTURESTAGE(4, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5960     {STATE_TEXTURESTAGE(4, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5961     {STATE_TEXTURESTAGE(4, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5962     {STATE_TEXTURESTAGE(4, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5963     {STATE_TEXTURESTAGE(4, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5964     {STATE_TEXTURESTAGE(4, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5965     {STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5966     {STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5967     {STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5968     {STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5969     {STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
5970     {STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
5971     {STATE_TEXTURESTAGE(5, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5972     {STATE_TEXTURESTAGE(5, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5973     {STATE_TEXTURESTAGE(5, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5974     {STATE_TEXTURESTAGE(5, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5975     {STATE_TEXTURESTAGE(5, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5976     {STATE_TEXTURESTAGE(5, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5977     {STATE_TEXTURESTAGE(5, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5978     {STATE_TEXTURESTAGE(5, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5979     {STATE_TEXTURESTAGE(5, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5980     {STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5981     {STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5982     {STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5983     {STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5984     {STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
5985     {STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
5986     {STATE_TEXTURESTAGE(6, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5987     {STATE_TEXTURESTAGE(6, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5988     {STATE_TEXTURESTAGE(6, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5989     {STATE_TEXTURESTAGE(6, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5990     {STATE_TEXTURESTAGE(6, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5991     {STATE_TEXTURESTAGE(6, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5992     {STATE_TEXTURESTAGE(6, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5993     {STATE_TEXTURESTAGE(6, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5994     {STATE_TEXTURESTAGE(6, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5995     {STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5996     {STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5997     {STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5998     {STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5999     {STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6000     {STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6001     {STATE_TEXTURESTAGE(7, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6002     {STATE_TEXTURESTAGE(7, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6003     {STATE_TEXTURESTAGE(7, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6004     {STATE_TEXTURESTAGE(7, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6005     {STATE_TEXTURESTAGE(7, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6006     {STATE_TEXTURESTAGE(7, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6007     {STATE_TEXTURESTAGE(7, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6008     {STATE_TEXTURESTAGE(7, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6009     {STATE_TEXTURESTAGE(7, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6010     {STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6011     {STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6012     {STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6013     {STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6014     {STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6015     {STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6016     {STATE_SAMPLER(0),                                    { STATE_SAMPLER(0),                                   sampler_texdim          }, WINED3D_GL_EXT_NONE             },
6017     {STATE_SAMPLER(1),                                    { STATE_SAMPLER(1),                                   sampler_texdim          }, WINED3D_GL_EXT_NONE             },
6018     {STATE_SAMPLER(2),                                    { STATE_SAMPLER(2),                                   sampler_texdim          }, WINED3D_GL_EXT_NONE             },
6019     {STATE_SAMPLER(3),                                    { STATE_SAMPLER(3),                                   sampler_texdim          }, WINED3D_GL_EXT_NONE             },
6020     {STATE_SAMPLER(4),                                    { STATE_SAMPLER(4),                                   sampler_texdim          }, WINED3D_GL_EXT_NONE             },
6021     {STATE_SAMPLER(5),                                    { STATE_SAMPLER(5),                                   sampler_texdim          }, WINED3D_GL_EXT_NONE             },
6022     {STATE_SAMPLER(6),                                    { STATE_SAMPLER(6),                                   sampler_texdim          }, WINED3D_GL_EXT_NONE             },
6023     {STATE_SAMPLER(7),                                    { STATE_SAMPLER(7),                                   sampler_texdim          }, WINED3D_GL_EXT_NONE             },
6024     {STATE_PIXELSHADER,                                   { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6025     {STATE_RENDER(WINED3DRS_FOGENABLE),                   { STATE_RENDER(WINED3DRS_FOGENABLE),                  state_arbfp_fog         }, WINED3D_GL_EXT_NONE             },
6026     {STATE_RENDER(WINED3DRS_FOGTABLEMODE),                { STATE_RENDER(WINED3DRS_FOGENABLE),                  state_arbfp_fog         }, WINED3D_GL_EXT_NONE             },
6027     {STATE_RENDER(WINED3DRS_FOGVERTEXMODE),               { STATE_RENDER(WINED3DRS_FOGENABLE),                  state_arbfp_fog         }, WINED3D_GL_EXT_NONE             },
6028     {STATE_RENDER(WINED3DRS_FOGSTART),                    { STATE_RENDER(WINED3DRS_FOGSTART),                   state_fogstartend       }, WINED3D_GL_EXT_NONE             },
6029     {STATE_RENDER(WINED3DRS_FOGEND),                      { STATE_RENDER(WINED3DRS_FOGSTART),                   state_fogstartend       }, WINED3D_GL_EXT_NONE             },
6030     {STATE_RENDER(WINED3DRS_SRGBWRITEENABLE),             { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6031     {STATE_RENDER(WINED3DRS_FOGCOLOR),                    { STATE_RENDER(WINED3DRS_FOGCOLOR),                   state_fogcolor          }, WINED3D_GL_EXT_NONE             },
6032     {STATE_RENDER(WINED3DRS_FOGDENSITY),                  { STATE_RENDER(WINED3DRS_FOGDENSITY),                 state_fogdensity        }, WINED3D_GL_EXT_NONE             },
6033     {STATE_TEXTURESTAGE(0,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(0, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6034     {STATE_TEXTURESTAGE(1,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(1, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6035     {STATE_TEXTURESTAGE(2,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(2, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6036     {STATE_TEXTURESTAGE(3,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(3, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6037     {STATE_TEXTURESTAGE(4,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(4, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6038     {STATE_TEXTURESTAGE(5,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(5, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6039     {STATE_TEXTURESTAGE(6,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(6, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6040     {STATE_TEXTURESTAGE(7,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(7, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6041     {STATE_RENDER(WINED3DRS_SPECULARENABLE),              { STATE_RENDER(WINED3DRS_SPECULARENABLE),             state_arb_specularenable}, WINED3D_GL_EXT_NONE             },
6042     {0 /* Terminate */,                                   { 0,                                                  0                       }, WINED3D_GL_EXT_NONE             },
6043 };
6044
6045 const struct fragment_pipeline arbfp_fragment_pipeline = {
6046     arbfp_enable,
6047     arbfp_get_caps,
6048     arbfp_alloc,
6049     arbfp_free,
6050     shader_arb_color_fixup_supported,
6051     arbfp_fragmentstate_template,
6052     TRUE /* We can disable projected textures */
6053 };
6054
6055 #define GLINFO_LOCATION device->adapter->gl_info
6056
6057 struct arbfp_blit_priv {
6058     GLenum yuy2_rect_shader, yuy2_2d_shader;
6059     GLenum uyvy_rect_shader, uyvy_2d_shader;
6060     GLenum yv12_rect_shader, yv12_2d_shader;
6061 };
6062
6063 static HRESULT arbfp_blit_alloc(IWineD3DDevice *iface) {
6064     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *) iface;
6065     device->blit_priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(struct arbfp_blit_priv));
6066     if(!device->blit_priv) {
6067         ERR("Out of memory\n");
6068         return E_OUTOFMEMORY;
6069     }
6070     return WINED3D_OK;
6071 }
6072
6073 /* Context activation is done by the caller. */
6074 static void arbfp_blit_free(IWineD3DDevice *iface) {
6075     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *) iface;
6076     struct arbfp_blit_priv *priv = device->blit_priv;
6077
6078     ENTER_GL();
6079     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yuy2_rect_shader));
6080     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yuy2_2d_shader));
6081     GL_EXTCALL(glDeleteProgramsARB(1, &priv->uyvy_rect_shader));
6082     GL_EXTCALL(glDeleteProgramsARB(1, &priv->uyvy_2d_shader));
6083     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yv12_rect_shader));
6084     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yv12_2d_shader));
6085     checkGLcall("Delete yuv programs");
6086     LEAVE_GL();
6087 }
6088
6089 static BOOL gen_planar_yuv_read(struct wined3d_shader_buffer *buffer, enum yuv_fixup yuv_fixup,
6090         GLenum textype, char *luminance)
6091 {
6092     char chroma;
6093     const char *tex, *texinstr;
6094
6095     if (yuv_fixup == YUV_FIXUP_UYVY) {
6096         chroma = 'x';
6097         *luminance = 'w';
6098     } else {
6099         chroma = 'w';
6100         *luminance = 'x';
6101     }
6102     switch(textype) {
6103         case GL_TEXTURE_2D:             tex = "2D";     texinstr = "TXP"; break;
6104         case GL_TEXTURE_RECTANGLE_ARB:  tex = "RECT";   texinstr = "TEX"; break;
6105         default:
6106             /* This is more tricky than just replacing the texture type - we have to navigate
6107              * properly in the texture to find the correct chroma values
6108              */
6109             FIXME("Implement yuv correction for non-2d, non-rect textures\n");
6110             return FALSE;
6111     }
6112
6113     /* First we have to read the chroma values. This means we need at least two pixels(no filtering),
6114      * or 4 pixels(with filtering). To get the unmodified chromas, we have to rid ourselves of the
6115      * filtering when we sample the texture.
6116      *
6117      * These are the rules for reading the chroma:
6118      *
6119      * Even pixel: Cr
6120      * Even pixel: U
6121      * Odd pixel: V
6122      *
6123      * So we have to get the sampling x position in non-normalized coordinates in integers
6124      */
6125     if(textype != GL_TEXTURE_RECTANGLE_ARB) {
6126         shader_addline(buffer, "MUL texcrd.xy, fragment.texcoord[0], size.x;\n");
6127         shader_addline(buffer, "MOV texcrd.w, size.x;\n");
6128     } else {
6129         shader_addline(buffer, "MOV texcrd, fragment.texcoord[0];\n");
6130     }
6131     /* We must not allow filtering between pixel x and x+1, this would mix U and V
6132      * Vertical filtering is ok. However, bear in mind that the pixel center is at
6133      * 0.5, so add 0.5.
6134      */
6135     shader_addline(buffer, "FLR texcrd.x, texcrd.x;\n");
6136     shader_addline(buffer, "ADD texcrd.x, texcrd.x, coef.y;\n");
6137
6138     /* Divide the x coordinate by 0.5 and get the fraction. This gives 0.25 and 0.75 for the
6139      * even and odd pixels respectively
6140      */
6141     shader_addline(buffer, "MUL texcrd2, texcrd, coef.y;\n");
6142     shader_addline(buffer, "FRC texcrd2, texcrd2;\n");
6143
6144     /* Sample Pixel 1 */
6145     shader_addline(buffer, "%s luminance, texcrd, texture[0], %s;\n", texinstr, tex);
6146
6147     /* Put the value into either of the chroma values */
6148     shader_addline(buffer, "SGE temp.x, texcrd2.x, coef.y;\n");
6149     shader_addline(buffer, "MUL chroma.x, luminance.%c, temp.x;\n", chroma);
6150     shader_addline(buffer, "SLT temp.x, texcrd2.x, coef.y;\n");
6151     shader_addline(buffer, "MUL chroma.y, luminance.%c, temp.x;\n", chroma);
6152
6153     /* Sample pixel 2. If we read an even pixel(SLT above returned 1), sample
6154      * the pixel right to the current one. Otherwise, sample the left pixel.
6155      * Bias and scale the SLT result to -1;1 and add it to the texcrd.x.
6156      */
6157     shader_addline(buffer, "MAD temp.x, temp.x, coef.z, -coef.x;\n");
6158     shader_addline(buffer, "ADD texcrd.x, texcrd, temp.x;\n");
6159     shader_addline(buffer, "%s luminance, texcrd, texture[0], %s;\n", texinstr, tex);
6160
6161     /* Put the value into the other chroma */
6162     shader_addline(buffer, "SGE temp.x, texcrd2.x, coef.y;\n");
6163     shader_addline(buffer, "MAD chroma.y, luminance.%c, temp.x, chroma.y;\n", chroma);
6164     shader_addline(buffer, "SLT temp.x, texcrd2.x, coef.y;\n");
6165     shader_addline(buffer, "MAD chroma.x, luminance.%c, temp.x, chroma.x;\n", chroma);
6166
6167     /* TODO: If filtering is enabled, sample a 2nd pair of pixels left or right of
6168      * the current one and lerp the two U and V values
6169      */
6170
6171     /* This gives the correctly filtered luminance value */
6172     shader_addline(buffer, "TEX luminance, fragment.texcoord[0], texture[0], %s;\n", tex);
6173
6174     return TRUE;
6175 }
6176
6177 static BOOL gen_yv12_read(struct wined3d_shader_buffer *buffer, GLenum textype, char *luminance)
6178 {
6179     const char *tex;
6180
6181     switch(textype) {
6182         case GL_TEXTURE_2D:             tex = "2D";     break;
6183         case GL_TEXTURE_RECTANGLE_ARB:  tex = "RECT";   break;
6184         default:
6185             FIXME("Implement yv12 correction for non-2d, non-rect textures\n");
6186             return FALSE;
6187     }
6188
6189     /* YV12 surfaces contain a WxH sized luminance plane, followed by a (W/2)x(H/2)
6190      * V and a (W/2)x(H/2) U plane, each with 8 bit per pixel. So the effective
6191      * bitdepth is 12 bits per pixel. Since the U and V planes have only half the
6192      * pitch of the luminance plane, the packing into the gl texture is a bit
6193      * unfortunate. If the whole texture is interpreted as luminance data it looks
6194      * approximately like this:
6195      *
6196      *        +----------------------------------+----
6197      *        |                                  |
6198      *        |                                  |
6199      *        |                                  |
6200      *        |                                  |
6201      *        |                                  |   2
6202      *        |            LUMINANCE             |   -
6203      *        |                                  |   3
6204      *        |                                  |
6205      *        |                                  |
6206      *        |                                  |
6207      *        |                                  |
6208      *        +----------------+-----------------+----
6209      *        |                |                 |
6210      *        |  U even rows   |  U odd rows     |
6211      *        |                |                 |   1
6212      *        +----------------+------------------   -
6213      *        |                |                 |   3
6214      *        |  V even rows   |  V odd rows     |
6215      *        |                |                 |
6216      *        +----------------+-----------------+----
6217      *        |                |                 |
6218      *        |     0.5        |       0.5       |
6219      *
6220      * So it appears as if there are 4 chroma images, but in fact the odd rows
6221      * in the chroma images are in the same row as the even ones. So its is
6222      * kinda tricky to read
6223      *
6224      * When reading from rectangle textures, keep in mind that the input y coordinates
6225      * go from 0 to d3d_height, whereas the opengl texture height is 1.5 * d3d_height
6226      */
6227     shader_addline(buffer, "PARAM yv12_coef = {%f, %f, %f, %f};\n",
6228             2.0f / 3.0f, 1.0f / 6.0f, (2.0f / 3.0f) + (1.0f / 6.0f), 1.0f / 3.0f);
6229
6230     shader_addline(buffer, "MOV texcrd, fragment.texcoord[0];\n");
6231     /* the chroma planes have only half the width */
6232     shader_addline(buffer, "MUL texcrd.x, texcrd.x, coef.y;\n");
6233
6234     /* The first value is between 2/3 and 5/6th of the texture's height, so scale+bias
6235      * the coordinate. Also read the right side of the image when reading odd lines
6236      *
6237      * Don't forget to clamp the y values in into the range, otherwise we'll get filtering
6238      * bleeding
6239      */
6240     if(textype == GL_TEXTURE_2D) {
6241
6242         shader_addline(buffer, "RCP chroma.w, size.y;\n");
6243
6244         shader_addline(buffer, "MUL texcrd2.y, texcrd.y, size.y;\n");
6245
6246         shader_addline(buffer, "FLR texcrd2.y, texcrd2.y;\n");
6247         shader_addline(buffer, "MAD texcrd.y, texcrd.y, yv12_coef.y, yv12_coef.x;\n");
6248
6249         /* Read odd lines from the right side(add size * 0.5 to the x coordinate */
6250         shader_addline(buffer, "ADD texcrd2.x, texcrd2.y, yv12_coef.y;\n"); /* To avoid 0.5 == 0.5 comparisons */
6251         shader_addline(buffer, "FRC texcrd2.x, texcrd2.x;\n");
6252         shader_addline(buffer, "SGE texcrd2.x, texcrd2.x, coef.y;\n");
6253         shader_addline(buffer, "MAD texcrd.x, texcrd2.x, coef.y, texcrd.x;\n");
6254
6255         /* clamp, keep the half pixel origin in mind */
6256         shader_addline(buffer, "MAD temp.y, coef.y, chroma.w, yv12_coef.x;\n");
6257         shader_addline(buffer, "MAX texcrd.y, temp.y, texcrd.y;\n");
6258         shader_addline(buffer, "MAD temp.y, -coef.y, chroma.w, yv12_coef.z;\n");
6259         shader_addline(buffer, "MIN texcrd.y, temp.y, texcrd.y;\n");
6260     } else {
6261         /* Read from [size - size+size/4] */
6262         shader_addline(buffer, "FLR texcrd.y, texcrd.y;\n");
6263         shader_addline(buffer, "MAD texcrd.y, texcrd.y, coef.w, size.y;\n");
6264
6265         /* Read odd lines from the right side(add size * 0.5 to the x coordinate */
6266         shader_addline(buffer, "ADD texcrd2.x, texcrd.y, yv12_coef.y;\n"); /* To avoid 0.5 == 0.5 comparisons */
6267         shader_addline(buffer, "FRC texcrd2.x, texcrd2.x;\n");
6268         shader_addline(buffer, "SGE texcrd2.x, texcrd2.x, coef.y;\n");
6269         shader_addline(buffer, "MUL texcrd2.x, texcrd2.x, size.x;\n");
6270         shader_addline(buffer, "MAD texcrd.x, texcrd2.x, coef.y, texcrd.x;\n");
6271
6272         /* Make sure to read exactly from the pixel center */
6273         shader_addline(buffer, "FLR texcrd.y, texcrd.y;\n");
6274         shader_addline(buffer, "ADD texcrd.y, texcrd.y, coef.y;\n");
6275
6276         /* Clamp */
6277         shader_addline(buffer, "MAD temp.y, size.y, coef.w, size.y;\n");
6278         shader_addline(buffer, "ADD temp.y, temp.y, -coef.y;\n");
6279         shader_addline(buffer, "MIN texcrd.y, temp.y, texcrd.y;\n");
6280         shader_addline(buffer, "ADD temp.y, size.y, -coef.y;\n");
6281         shader_addline(buffer, "MAX texcrd.y, temp.y, texcrd.y;\n");
6282     }
6283     /* Read the texture, put the result into the output register */
6284     shader_addline(buffer, "TEX temp, texcrd, texture[0], %s;\n", tex);
6285     shader_addline(buffer, "MOV chroma.x, temp.w;\n");
6286
6287     /* The other chroma value is 1/6th of the texture lower, from 5/6th to 6/6th
6288      * No need to clamp because we're just reusing the already clamped value from above
6289      */
6290     if(textype == GL_TEXTURE_2D) {
6291         shader_addline(buffer, "ADD texcrd.y, texcrd.y, yv12_coef.y;\n");
6292     } else {
6293         shader_addline(buffer, "MAD texcrd.y, size.y, coef.w, texcrd.y;\n");
6294     }
6295     shader_addline(buffer, "TEX temp, texcrd, texture[0], %s;\n", tex);
6296     shader_addline(buffer, "MOV chroma.y, temp.w;\n");
6297
6298     /* Sample the luminance value. It is in the top 2/3rd of the texture, so scale the y coordinate.
6299      * Clamp the y coordinate to prevent the chroma values from bleeding into the sampled luminance
6300      * values due to filtering
6301      */
6302     shader_addline(buffer, "MOV texcrd, fragment.texcoord[0];\n");
6303     if(textype == GL_TEXTURE_2D) {
6304         /* Multiply the y coordinate by 2/3 and clamp it */
6305         shader_addline(buffer, "MUL texcrd.y, texcrd.y, yv12_coef.x;\n");
6306         shader_addline(buffer, "MAD temp.y, -coef.y, chroma.w, yv12_coef.x;\n");
6307         shader_addline(buffer, "MIN texcrd.y, temp.y, texcrd.y;\n");
6308         shader_addline(buffer, "TEX luminance, texcrd, texture[0], %s;\n", tex);
6309     } else {
6310         /* Reading from texture_rectangles is pretty straightforward, just use the unmodified
6311          * texture coordinate. It is still a good idea to clamp it though, since the opengl texture
6312          * is bigger
6313          */
6314         shader_addline(buffer, "ADD temp.x, size.y, -coef.y;\n");
6315         shader_addline(buffer, "MIN texcrd.y, texcrd.y, size.x;\n");
6316         shader_addline(buffer, "TEX luminance, texcrd, texture[0], %s;\n", tex);
6317     }
6318     *luminance = 'a';
6319
6320     return TRUE;
6321 }
6322
6323 /* Context activation is done by the caller. */
6324 static GLuint gen_yuv_shader(IWineD3DDeviceImpl *device, enum yuv_fixup yuv_fixup, GLenum textype)
6325 {
6326     GLenum shader;
6327     struct wined3d_shader_buffer buffer;
6328     char luminance_component;
6329     struct arbfp_blit_priv *priv = device->blit_priv;
6330
6331     /* Shader header */
6332     if (!shader_buffer_init(&buffer))
6333     {
6334         ERR("Failed to initialize shader buffer.\n");
6335         return 0;
6336     }
6337
6338     ENTER_GL();
6339     GL_EXTCALL(glGenProgramsARB(1, &shader));
6340     checkGLcall("GL_EXTCALL(glGenProgramsARB(1, &shader))");
6341     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader));
6342     checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader)");
6343     LEAVE_GL();
6344     if(!shader) {
6345         shader_buffer_free(&buffer);
6346         return 0;
6347     }
6348
6349     /* The YUY2 and UYVY formats contain two pixels packed into a 32 bit macropixel,
6350      * giving effectively 16 bit per pixel. The color consists of a luminance(Y) and
6351      * two chroma(U and V) values. Each macropixel has two luminance values, one for
6352      * each single pixel it contains, and one U and one V value shared between both
6353      * pixels.
6354      *
6355      * The data is loaded into an A8L8 texture. With YUY2, the luminance component
6356      * contains the luminance and alpha the chroma. With UYVY it is vice versa. Thus
6357      * take the format into account when generating the read swizzles
6358      *
6359      * Reading the Y value is straightforward - just sample the texture. The hardware
6360      * takes care of filtering in the horizontal and vertical direction.
6361      *
6362      * Reading the U and V values is harder. We have to avoid filtering horizontally,
6363      * because that would mix the U and V values of one pixel or two adjacent pixels.
6364      * Thus floor the texture coordinate and add 0.5 to get an unfiltered read,
6365      * regardless of the filtering setting. Vertical filtering works automatically
6366      * though - the U and V values of two rows are mixed nicely.
6367      *
6368      * Appart of avoiding filtering issues, the code has to know which value it just
6369      * read, and where it can find the other one. To determine this, it checks if
6370      * it sampled an even or odd pixel, and shifts the 2nd read accordingly.
6371      *
6372      * Handling horizontal filtering of U and V values requires reading a 2nd pair
6373      * of pixels, extracting U and V and mixing them. This is not implemented yet.
6374      *
6375      * An alternative implementation idea is to load the texture as A8R8G8B8 texture,
6376      * with width / 2. This way one read gives all 3 values, finding U and V is easy
6377      * in an unfiltered situation. Finding the luminance on the other hand requires
6378      * finding out if it is an odd or even pixel. The real drawback of this approach
6379      * is filtering. This would have to be emulated completely in the shader, reading
6380      * up two 2 packed pixels in up to 2 rows and interpolating both horizontally and
6381      * vertically. Beyond that it would require adjustments to the texture handling
6382      * code to deal with the width scaling
6383      */
6384     shader_addline(&buffer, "!!ARBfp1.0\n");
6385     shader_addline(&buffer, "TEMP luminance;\n");
6386     shader_addline(&buffer, "TEMP temp;\n");
6387     shader_addline(&buffer, "TEMP chroma;\n");
6388     shader_addline(&buffer, "TEMP texcrd;\n");
6389     shader_addline(&buffer, "TEMP texcrd2;\n");
6390     shader_addline(&buffer, "PARAM coef = {1.0, 0.5, 2.0, 0.25};\n");
6391     shader_addline(&buffer, "PARAM yuv_coef = {1.403, 0.344, 0.714, 1.770};\n");
6392     shader_addline(&buffer, "PARAM size = program.local[0];\n");
6393
6394     switch (yuv_fixup)
6395     {
6396         case YUV_FIXUP_UYVY:
6397         case YUV_FIXUP_YUY2:
6398             if (!gen_planar_yuv_read(&buffer, yuv_fixup, textype, &luminance_component))
6399             {
6400                 shader_buffer_free(&buffer);
6401                 return 0;
6402             }
6403             break;
6404
6405         case YUV_FIXUP_YV12:
6406             if (!gen_yv12_read(&buffer, textype, &luminance_component))
6407             {
6408                 shader_buffer_free(&buffer);
6409                 return 0;
6410             }
6411             break;
6412
6413         default:
6414             FIXME("Unsupported YUV fixup %#x\n", yuv_fixup);
6415             shader_buffer_free(&buffer);
6416             return 0;
6417     }
6418
6419     /* Calculate the final result. Formula is taken from
6420      * http://www.fourcc.org/fccyvrgb.php. Note that the chroma
6421      * ranges from -0.5 to 0.5
6422      */
6423     shader_addline(&buffer, "SUB chroma.xy, chroma, coef.y;\n");
6424
6425     shader_addline(&buffer, "MAD result.color.x, chroma.x, yuv_coef.x, luminance.%c;\n", luminance_component);
6426     shader_addline(&buffer, "MAD temp.x, -chroma.y, yuv_coef.y, luminance.%c;\n", luminance_component);
6427     shader_addline(&buffer, "MAD result.color.y, -chroma.x, yuv_coef.z, temp.x;\n");
6428     shader_addline(&buffer, "MAD result.color.z, chroma.y, yuv_coef.w, luminance.%c;\n", luminance_component);
6429     shader_addline(&buffer, "END\n");
6430
6431     ENTER_GL();
6432     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB, strlen(buffer.buffer), buffer.buffer));
6433
6434     if (glGetError() == GL_INVALID_OPERATION) {
6435         GLint pos;
6436         glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
6437         FIXME("Fragment program error at position %d: %s\n", pos,
6438               debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
6439     }
6440     shader_buffer_free(&buffer);
6441     LEAVE_GL();
6442
6443     switch (yuv_fixup)
6444     {
6445         case YUV_FIXUP_YUY2:
6446             if (textype == GL_TEXTURE_RECTANGLE_ARB) priv->yuy2_rect_shader = shader;
6447             else priv->yuy2_2d_shader = shader;
6448             break;
6449
6450         case YUV_FIXUP_UYVY:
6451             if (textype == GL_TEXTURE_RECTANGLE_ARB) priv->uyvy_rect_shader = shader;
6452             else priv->uyvy_2d_shader = shader;
6453             break;
6454
6455         case YUV_FIXUP_YV12:
6456             if (textype == GL_TEXTURE_RECTANGLE_ARB) priv->yv12_rect_shader = shader;
6457             else priv->yv12_2d_shader = shader;
6458             break;
6459     }
6460
6461     return shader;
6462 }
6463
6464 /* Context activation is done by the caller. */
6465 static HRESULT arbfp_blit_set(IWineD3DDevice *iface, const struct GlPixelFormatDesc *format_desc,
6466         GLenum textype, UINT width, UINT height)
6467 {
6468     GLenum shader;
6469     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *) iface;
6470     float size[4] = {width, height, 1, 1};
6471     struct arbfp_blit_priv *priv = device->blit_priv;
6472     enum yuv_fixup yuv_fixup;
6473
6474     if (!is_yuv_fixup(format_desc->color_fixup))
6475     {
6476         TRACE("Fixup:\n");
6477         dump_color_fixup_desc(format_desc->color_fixup);
6478         /* Don't bother setting up a shader for unconverted formats */
6479         ENTER_GL();
6480         glEnable(textype);
6481         checkGLcall("glEnable(textype)");
6482         LEAVE_GL();
6483         return WINED3D_OK;
6484     }
6485
6486     yuv_fixup = get_yuv_fixup(format_desc->color_fixup);
6487
6488     switch(yuv_fixup)
6489     {
6490         case YUV_FIXUP_YUY2:
6491             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->yuy2_rect_shader : priv->yuy2_2d_shader;
6492             break;
6493
6494         case YUV_FIXUP_UYVY:
6495             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->uyvy_rect_shader : priv->uyvy_2d_shader;
6496             break;
6497
6498         case YUV_FIXUP_YV12:
6499             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->yv12_rect_shader : priv->yv12_2d_shader;
6500             break;
6501
6502         default:
6503             FIXME("Unsupported YUV fixup %#x, not setting a shader\n", yuv_fixup);
6504             ENTER_GL();
6505             glEnable(textype);
6506             checkGLcall("glEnable(textype)");
6507             LEAVE_GL();
6508             return E_NOTIMPL;
6509     }
6510
6511     if (!shader) shader = gen_yuv_shader(device, yuv_fixup, textype);
6512
6513     ENTER_GL();
6514     glEnable(GL_FRAGMENT_PROGRAM_ARB);
6515     checkGLcall("glEnable(GL_FRAGMENT_PROGRAM_ARB)");
6516     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader));
6517     checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader)");
6518     GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, 0, size));
6519     checkGLcall("glProgramLocalParameter4fvARB");
6520     LEAVE_GL();
6521
6522     return WINED3D_OK;
6523 }
6524
6525 /* Context activation is done by the caller. */
6526 static void arbfp_blit_unset(IWineD3DDevice *iface) {
6527     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *) iface;
6528
6529     ENTER_GL();
6530     glDisable(GL_FRAGMENT_PROGRAM_ARB);
6531     checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
6532     glDisable(GL_TEXTURE_2D);
6533     checkGLcall("glDisable(GL_TEXTURE_2D)");
6534     if(GL_SUPPORT(ARB_TEXTURE_CUBE_MAP)) {
6535         glDisable(GL_TEXTURE_CUBE_MAP_ARB);
6536         checkGLcall("glDisable(GL_TEXTURE_CUBE_MAP_ARB)");
6537     }
6538     if(GL_SUPPORT(ARB_TEXTURE_RECTANGLE)) {
6539         glDisable(GL_TEXTURE_RECTANGLE_ARB);
6540         checkGLcall("glDisable(GL_TEXTURE_RECTANGLE_ARB)");
6541     }
6542     LEAVE_GL();
6543 }
6544
6545 static BOOL arbfp_blit_color_fixup_supported(struct color_fixup_desc fixup)
6546 {
6547     enum yuv_fixup yuv_fixup;
6548
6549     if (TRACE_ON(d3d_shader) && TRACE_ON(d3d))
6550     {
6551         TRACE("Checking support for fixup:\n");
6552         dump_color_fixup_desc(fixup);
6553     }
6554
6555     if (is_identity_fixup(fixup))
6556     {
6557         TRACE("[OK]\n");
6558         return TRUE;
6559     }
6560
6561     /* We only support YUV conversions. */
6562     if (!is_yuv_fixup(fixup))
6563     {
6564         TRACE("[FAILED]\n");
6565         return FALSE;
6566     }
6567
6568     yuv_fixup = get_yuv_fixup(fixup);
6569     switch(yuv_fixup)
6570     {
6571         case YUV_FIXUP_YUY2:
6572         case YUV_FIXUP_UYVY:
6573         case YUV_FIXUP_YV12:
6574             TRACE("[OK]\n");
6575             return TRUE;
6576
6577         default:
6578             FIXME("Unsupported YUV fixup %#x\n", yuv_fixup);
6579             TRACE("[FAILED]\n");
6580             return FALSE;
6581     }
6582 }
6583
6584 const struct blit_shader arbfp_blit = {
6585     arbfp_blit_alloc,
6586     arbfp_blit_free,
6587     arbfp_blit_set,
6588     arbfp_blit_unset,
6589     arbfp_blit_color_fixup_supported,
6590 };
6591
6592 #undef GLINFO_LOCATION