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 WineD3D_GL_Info *gl_info) {
  46     IWineD3DBaseShaderImpl *This = (IWineD3DBaseShaderImpl *) shader;
  47     if(!This->baseShader.reg_maps.usesmova) return FALSE;
  48     return !GL_SUPPORT(NV_VERTEX_PROGRAM2_OPTION);
  49 }
  50
  51 static BOOL need_helper_const(const WineD3D_GL_Info *gl_info) {
  52     if (!GL_SUPPORT(NV_VERTEX_PROGRAM) /* Need to init colors. */
  53             || gl_info->quirks & WINED3D_QUIRK_ARB_VS_OFFSET_LIMIT /* Load the immval offset. */
  54             || gl_info->quirks & WINED3D_QUIRK_SET_TEXCOORD_W) /* Have to init texcoords. */
  55     {
  56         return TRUE;
  57     }
  58     return FALSE;
  59 }
  60
  61 static unsigned int reserved_vs_const(IWineD3DBaseShader *shader, const WineD3D_GL_Info *gl_info) {
  62     unsigned int ret = 1;
  63     /* We use one PARAM for the pos fixup, and in some cases one to load
  64      * some immediate values into the shader
  65      */
  66     if(need_helper_const(gl_info)) ret++;
  67     if(need_mova_const(shader, gl_info)) ret++;
  68     return ret;
  69 }
  70
  71 static inline BOOL ffp_clip_emul(IWineD3DStateBlockImpl *stateblock)
  72 {
  73     return stateblock->lowest_disabled_stage < 7;
  74 }
  75
  76 /* Returns TRUE if result.clip from GL_NV_vertex_program2 should be used and FALSE otherwise */
  77 static inline BOOL use_nv_clip(const WineD3D_GL_Info *gl_info)
  78 {
  79     return GL_SUPPORT(NV_VERTEX_PROGRAM2_OPTION);
  80 }
  81
  82 /* Internally used shader constants. Applications can use constants 0 to GL_LIMITS(vshader_constantsF) - 1,
  83  * so upload them above that
  84  */
  85 #define ARB_SHADER_PRIVCONST_BASE (GL_LIMITS(vshader_constantsF) - 1)
  86 #define ARB_SHADER_PRIVCONST_POS ARB_SHADER_PRIVCONST_BASE + 0
  87
  88 /* ARB_program_shader private data */
  89
  90 struct loop_control
  91 {
  92     unsigned int count;
  93     unsigned int start;
  94     int step;
  95 };
  96
  97 struct control_frame
  98 {
  99     struct                          list entry;
 100     enum
 101     {
 102         IF,
 103         IFC,
 104         LOOP,
 105         REP
 106     } type;
 107     BOOL                            muting;
 108     BOOL                            outer_loop;
 109     union
 110     {
 111         unsigned int                loop_no;
 112         unsigned int                ifc_no;
 113     };
 114     struct loop_control             loop_control;
 115     BOOL                            had_else;
 116 };
 117
 118 struct arb_ps_np2fixup_info
 119 {
 120     struct ps_np2fixup_info         super;
 121     /* For ARB we need a offset value:
 122      * With both GLSL and ARB mode the NP2 fixup information (the texture dimensions) are stored in a
 123      * consecutive way (GLSL uses a uniform array). Since ARB doesn't know the notion of a "standalone"
 124      * array we need an offset to the index inside the program local parameter array. */
 125     UINT                            offset;
 126 };
 127
 128 struct arb_ps_compile_args
 129 {
 130     struct ps_compile_args          super;
 131     DWORD                           bools; /* WORD is enough, use DWORD for alignment */
 132     unsigned char                   loop_ctrl[MAX_CONST_I][3];
 133 };
 134
 135 struct stb_const_desc
 136 {
 137     unsigned char           texunit;
 138     UINT                    const_num;
 139 };
 140
 141 struct arb_ps_compiled_shader
 142 {
 143     struct arb_ps_compile_args      args;
 144     struct arb_ps_np2fixup_info     np2fixup_info;
 145     struct stb_const_desc           bumpenvmatconst[MAX_TEXTURES];
 146     struct stb_const_desc           luminanceconst[MAX_TEXTURES];
 147     UINT                            int_consts[MAX_CONST_I];
 148     GLuint                          prgId;
 149     UINT                            ycorrection;
 150     unsigned char                   numbumpenvmatconsts;
 151     char                            num_int_consts;
 152 };
 153
 154 struct arb_vs_compile_args
 155 {
 156     struct vs_compile_args          super;
 157     union
 158     {
 159         struct
 160         {
 161             WORD                    bools;
 162             char                    clip_control[2];
 163         }                           boolclip;
 164         DWORD                       boolclip_compare;
 165     };
 166     DWORD                           ps_signature;
 167     union
 168     {
 169         unsigned char               vertex_samplers[4];
 170         DWORD                       vertex_samplers_compare;
 171     };
 172     unsigned char                   loop_ctrl[MAX_CONST_I][3];
 173 };
 174
 175 struct arb_vs_compiled_shader
 176 {
 177     struct arb_vs_compile_args      args;
 178     GLuint                          prgId;
 179     UINT                            int_consts[MAX_CONST_I];
 180     char                            num_int_consts;
 181     char                            need_color_unclamp;
 182     UINT                            pos_fixup;
 183 };
 184
 185 struct recorded_instruction
 186 {
 187     struct wined3d_shader_instruction ins;
 188     struct list entry;
 189 };
 190
 191 struct shader_arb_ctx_priv
 192 {
 193     char addr_reg[20];
 194     enum
 195     {
 196         /* plain GL_ARB_vertex_program or GL_ARB_fragment_program */
 197         ARB,
 198         /* GL_NV_vertex_progam2_option or GL_NV_fragment_program_option */
 199         NV2,
 200         /* GL_NV_vertex_program3 or GL_NV_fragment_program2 */
 201         NV3
 202     } target_version;
 203
 204     const struct arb_vs_compile_args    *cur_vs_args;
 205     const struct arb_ps_compile_args    *cur_ps_args;
 206     const struct arb_ps_compiled_shader *compiled_fprog;
 207     const struct arb_vs_compiled_shader *compiled_vprog;
 208     struct arb_ps_np2fixup_info         *cur_np2fixup_info;
 209     struct list                         control_frames;
 210     struct list                         record;
 211     BOOL                                recording;
 212     BOOL                                muted;
 213     unsigned int                        num_loops, loop_depth, num_ifcs;
 214     int                                 aL;
 215
 216     unsigned int                        vs_clipplanes;
 217     BOOL                                footer_written;
 218     BOOL                                in_main_func;
 219
 220     /* For 3.0 vertex shaders */
 221     const char                          *vs_output[MAX_REG_OUTPUT];
 222     /* For 2.x and earlier vertex shaders */
 223     const char                          *texcrd_output[8], *color_output[2], *fog_output;
 224
 225     /* 3.0 pshader input for compatibility with fixed function */
 226     const char                          *ps_input[MAX_REG_INPUT];
 227 };
 228
 229 struct ps_signature
 230 {
 231     struct wined3d_shader_signature_element *sig;
 232     DWORD                               idx;
 233     struct wine_rb_entry                entry;
 234 };
 235
 236 struct arb_pshader_private {
 237     struct arb_ps_compiled_shader   *gl_shaders;
 238     UINT                            num_gl_shaders, shader_array_size;
 239     BOOL                            has_signature_idx;
 240     DWORD                           input_signature_idx;
 241     DWORD                           clipplane_emulation;
 242     BOOL                            clamp_consts;
 243 };
 244
 245 struct arb_vshader_private {
 246     struct arb_vs_compiled_shader   *gl_shaders;
 247     UINT                            num_gl_shaders, shader_array_size;
 248 };
 249
 250 struct shader_arb_priv
 251 {
 252     GLuint                  current_vprogram_id;
 253     GLuint                  current_fprogram_id;
 254     const struct arb_ps_compiled_shader *compiled_fprog;
 255     const struct arb_vs_compiled_shader *compiled_vprog;
 256     GLuint                  depth_blt_vprogram_id;
 257     GLuint                  depth_blt_fprogram_id[tex_type_count];
 258     BOOL                    use_arbfp_fixed_func;
 259     struct wine_rb_tree     fragment_shaders;
 260     BOOL                    last_ps_const_clamped;
 261     BOOL                    last_vs_color_unclamp;
 262
 263     struct wine_rb_tree     signature_tree;
 264     DWORD ps_sig_number;
 265 };
 266
 267 /********************************************************
 268  * ARB_[vertex/fragment]_program helper functions follow
 269  ********************************************************/
 270
 271 /**
 272  * Loads floating point constants into the currently set ARB_vertex/fragment_program.
 273  * When constant_list == NULL, it will load all the constants.
 274  *
 275  * @target_type should be either GL_VERTEX_PROGRAM_ARB (for vertex shaders)
 276  *  or GL_FRAGMENT_PROGRAM_ARB (for pixel shaders)
 277  */
 278 /* GL locking is done by the caller */
 279 static unsigned int shader_arb_load_constantsF(IWineD3DBaseShaderImpl* This, const WineD3D_GL_Info *gl_info,
 280         GLuint target_type, unsigned int max_constants, const float *constants, char *dirty_consts)
 281 {
 282     local_constant* lconst;
 283     DWORD i, j;
 284     unsigned int ret;
 285
 286     if (TRACE_ON(d3d_shader)) {
 287         for(i = 0; i < max_constants; i++) {
 288             if(!dirty_consts[i]) continue;
 289             TRACE_(d3d_constants)("Loading constants %i: %f, %f, %f, %f\n", i,
 290                         constants[i * 4 + 0], constants[i * 4 + 1],
 291                         constants[i * 4 + 2], constants[i * 4 + 3]);
 292         }
 293     }
 294     /* In 1.X pixel shaders constants are implicitly clamped in the range [-1;1] */
 295     if (target_type == GL_FRAGMENT_PROGRAM_ARB && This->baseShader.reg_maps.shader_version.major == 1)
 296     {
 297         float lcl_const[4];
 298         for(i = 0; i < max_constants; i++) {
 299             if(!dirty_consts[i]) continue;
 300             dirty_consts[i] = 0;
 301
 302             j = 4 * i;
 303             if (constants[j + 0] > 1.0f) lcl_const[0] = 1.0f;
 304             else if (constants[j + 0] < -1.0f) lcl_const[0] = -1.0f;
 305             else lcl_const[0] = constants[j + 0];
 306
 307             if (constants[j + 1] > 1.0f) lcl_const[1] = 1.0f;
 308             else if (constants[j + 1] < -1.0f) lcl_const[1] = -1.0f;
 309             else lcl_const[1] = constants[j + 1];
 310
 311             if (constants[j + 2] > 1.0f) lcl_const[2] = 1.0f;
 312             else if (constants[j + 2] < -1.0f) lcl_const[2] = -1.0f;
 313             else lcl_const[2] = constants[j + 2];
 314
 315             if (constants[j + 3] > 1.0f) lcl_const[3] = 1.0f;
 316             else if (constants[j + 3] < -1.0f) lcl_const[3] = -1.0f;
 317             else lcl_const[3] = constants[j + 3];
 318
 319             GL_EXTCALL(glProgramEnvParameter4fvARB(target_type, i, lcl_const));
 320         }
 321     } else {
 322         if(GL_SUPPORT(EXT_GPU_PROGRAM_PARAMETERS)) {
 323             /* TODO: Benchmark if we're better of with finding the dirty constants ourselves,
 324              * or just reloading *all* constants at once
 325              *
 326             GL_EXTCALL(glProgramEnvParameters4fvEXT(target_type, 0, max_constants, constants));
 327              */
 328             for(i = 0; i < max_constants; i++) {
 329                 if(!dirty_consts[i]) continue;
 330
 331                 /* Find the next block of dirty constants */
 332                 dirty_consts[i] = 0;
 333                 j = i;
 334                 for(i++; (i < max_constants) && dirty_consts[i]; i++) {
 335                     dirty_consts[i] = 0;
 336                 }
 337
 338                 GL_EXTCALL(glProgramEnvParameters4fvEXT(target_type, j, i - j, constants + (j * 4)));
 339             }
 340         } else {
 341             for(i = 0; i < max_constants; i++) {
 342                 if(dirty_consts[i]) {
 343                     dirty_consts[i] = 0;
 344                     GL_EXTCALL(glProgramEnvParameter4fvARB(target_type, i, constants + (i * 4)));
 345                 }
 346             }
 347         }
 348     }
 349     checkGLcall("glProgramEnvParameter4fvARB()");
 350
 351     /* Load immediate constants */
 352     if(This->baseShader.load_local_constsF) {
 353         if (TRACE_ON(d3d_shader)) {
 354             LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry) {
 355                 GLfloat* values = (GLfloat*)lconst->value;
 356                 TRACE_(d3d_constants)("Loading local constants %i: %f, %f, %f, %f\n", lconst->idx,
 357                         values[0], values[1], values[2], values[3]);
 358             }
 359         }
 360         /* Immediate constants are clamped for 1.X shaders at loading times */
 361         ret = 0;
 362         LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry) {
 363             dirty_consts[lconst->idx] = 1; /* Dirtify so the non-immediate constant overwrites it next time */
 364             ret = max(ret, lconst->idx + 1);
 365             GL_EXTCALL(glProgramEnvParameter4fvARB(target_type, lconst->idx, (GLfloat*)lconst->value));
 366         }
 367         checkGLcall("glProgramEnvParameter4fvARB()");
 368         return ret; /* The loaded immediate constants need reloading for the next shader */
 369     } else {
 370         return 0; /* No constants are dirty now */
 371     }
 372 }
 373
 374 /**
 375  * Loads the texture dimensions for NP2 fixup into the currently set ARB_[vertex/fragment]_programs.
 376  */
 377 static void shader_arb_load_np2fixup_constants(
 378     IWineD3DDevice* device,
 379     char usePixelShader,
 380     char useVertexShader) {
 381
 382     IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl *) device;
 383     const struct shader_arb_priv* const priv = (const struct shader_arb_priv *) deviceImpl->shader_priv;
 384     IWineD3DStateBlockImpl* stateBlock = deviceImpl->stateBlock;
 385     const WineD3D_GL_Info *gl_info = &deviceImpl->adapter->gl_info;
 386
 387     if (!usePixelShader) {
 388         /* NP2 texcoord fixup is (currently) only done for pixelshaders. */
 389         return;
 390     }
 391
 392     if (priv->compiled_fprog && priv->compiled_fprog->np2fixup_info.super.active) {
 393         const struct arb_ps_np2fixup_info* const fixup = &priv->compiled_fprog->np2fixup_info;
 394         UINT i;
 395         WORD active = fixup->super.active;
 396         GLfloat np2fixup_constants[4 * MAX_FRAGMENT_SAMPLERS];
 397
 398         for (i = 0; active; active >>= 1, ++i) {
 399             const unsigned char idx = fixup->super.idx[i];
 400             const IWineD3DTextureImpl* const tex = (const IWineD3DTextureImpl*) stateBlock->textures[i];
 401             GLfloat* tex_dim = &np2fixup_constants[(idx >> 1) * 4];
 402
 403             if (!(active & 1)) continue;
 404
 405             if (!tex) {
 406                 FIXME("Nonexistent texture is flagged for NP2 texcoord fixup\n");
 407                 continue;
 408             }
 409
 410             if (idx % 2) {
 411                 tex_dim[2] = tex->baseTexture.pow2Matrix[0]; tex_dim[3] = tex->baseTexture.pow2Matrix[5];
 412             } else {
 413                 tex_dim[0] = tex->baseTexture.pow2Matrix[0]; tex_dim[1] = tex->baseTexture.pow2Matrix[5];
 414             }
 415         }
 416
 417         for (i = 0; i < fixup->super.num_consts; ++i) {
 418             GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB,
 419                                                    fixup->offset + i, &np2fixup_constants[i * 4]));
 420         }
 421     }
 422 }
 423
 424 /* GL locking is done by the caller. */
 425 static inline void shader_arb_ps_local_constants(IWineD3DDeviceImpl* deviceImpl)
 426 {
 427     IWineD3DStateBlockImpl* stateBlock = deviceImpl->stateBlock;
 428     const WineD3D_GL_Info *gl_info = &deviceImpl->adapter->gl_info;
 429     unsigned char i;
 430     struct shader_arb_priv *priv = deviceImpl->shader_priv;
 431     const struct arb_ps_compiled_shader *gl_shader = priv->compiled_fprog;
 432
 433     for(i = 0; i < gl_shader->numbumpenvmatconsts; i++)
 434     {
 435         int texunit = gl_shader->bumpenvmatconst[i].texunit;
 436
 437         /* The state manager takes care that this function is always called if the bump env matrix changes */
 438         const float *data = (const float *)&stateBlock->textureState[texunit][WINED3DTSS_BUMPENVMAT00];
 439         GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, gl_shader->bumpenvmatconst[i].const_num, data));
 440
 441         if (gl_shader->luminanceconst[i].const_num != WINED3D_CONST_NUM_UNUSED)
 442         {
 443             /* WINED3DTSS_BUMPENVLSCALE and WINED3DTSS_BUMPENVLOFFSET are next to each other.
 444              * point gl to the scale, and load 4 floats. x = scale, y = offset, z and w are junk, we
 445              * don't care about them. The pointers are valid for sure because the stateblock is bigger.
 446              * (they're WINED3DTSS_TEXTURETRANSFORMFLAGS and WINED3DTSS_ADDRESSW, so most likely 0 or NaN
 447             */
 448             const float *scale = (const float *)&stateBlock->textureState[texunit][WINED3DTSS_BUMPENVLSCALE];
 449             GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, gl_shader->luminanceconst[i].const_num, scale));
 450         }
 451     }
 452     checkGLcall("Load bumpmap consts");
 453
 454     if(gl_shader->ycorrection != WINED3D_CONST_NUM_UNUSED)
 455     {
 456         /* ycorrection.x: Backbuffer height(onscreen) or 0(offscreen).
 457         * ycorrection.y: -1.0(onscreen), 1.0(offscreen)
 458         * ycorrection.z: 1.0
 459         * ycorrection.w: 0.0
 460         */
 461         float val[4];
 462         val[0] = deviceImpl->render_offscreen ? 0.0f : ((IWineD3DSurfaceImpl *) deviceImpl->render_targets[0])->currentDesc.Height;
 463         val[1] = deviceImpl->render_offscreen ? 1.0f : -1.0f;
 464         val[2] = 1.0f;
 465         val[3] = 0.0f;
 466         GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, gl_shader->ycorrection, val));
 467         checkGLcall("y correction loading");
 468     }
 469
 470     if(gl_shader->num_int_consts == 0) return;
 471
 472     for(i = 0; i < MAX_CONST_I; i++)
 473     {
 474         if(gl_shader->int_consts[i] != WINED3D_CONST_NUM_UNUSED)
 475         {
 476             float val[4];
 477             val[0] = stateBlock->pixelShaderConstantI[4 * i];
 478             val[1] = stateBlock->pixelShaderConstantI[4 * i + 1];
 479             val[2] = stateBlock->pixelShaderConstantI[4 * i + 2];
 480             val[3] = -1.0f;
 481
 482             GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, gl_shader->int_consts[i], val));
 483         }
 484     }
 485     checkGLcall("Load ps int consts");
 486 }
 487
 488 /* GL locking is done by the caller. */
 489 static inline void shader_arb_vs_local_constants(IWineD3DDeviceImpl* deviceImpl)
 490 {
 491     IWineD3DStateBlockImpl* stateBlock;
 492     const WineD3D_GL_Info *gl_info = &deviceImpl->adapter->gl_info;
 493     unsigned char i;
 494     struct shader_arb_priv *priv = deviceImpl->shader_priv;
 495     const struct arb_vs_compiled_shader *gl_shader = priv->compiled_vprog;
 496
 497     /* Upload the position fixup */
 498     GL_EXTCALL(glProgramLocalParameter4fvARB(GL_VERTEX_PROGRAM_ARB, gl_shader->pos_fixup, deviceImpl->posFixup));
 499
 500     if(gl_shader->num_int_consts == 0) return;
 501
 502     stateBlock = deviceImpl->stateBlock;
 503
 504     for(i = 0; i < MAX_CONST_I; i++)
 505     {
 506         if(gl_shader->int_consts[i] != WINED3D_CONST_NUM_UNUSED)
 507         {
 508             float val[4];
 509             val[0] = stateBlock->vertexShaderConstantI[4 * i];
 510             val[1] = stateBlock->vertexShaderConstantI[4 * i + 1];
 511             val[2] = stateBlock->vertexShaderConstantI[4 * i + 2];
 512             val[3] = -1.0f;
 513
 514             GL_EXTCALL(glProgramLocalParameter4fvARB(GL_VERTEX_PROGRAM_ARB, gl_shader->int_consts[i], val));
 515         }
 516     }
 517     checkGLcall("Load vs int consts");
 518 }
 519
 520 /**
 521  * Loads the app-supplied constants into the currently set ARB_[vertex/fragment]_programs.
 522  *
 523  * We only support float constants in ARB at the moment, so don't
 524  * worry about the Integers or Booleans
 525  */
 526 /* GL locking is done by the caller (state handler) */
 527 static void shader_arb_load_constants(
 528     IWineD3DDevice* device,
 529     char usePixelShader,
 530     char useVertexShader) {
 531
 532     IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) device;
 533     IWineD3DStateBlockImpl* stateBlock = deviceImpl->stateBlock;
 534     const WineD3D_GL_Info *gl_info = &deviceImpl->adapter->gl_info;
 535
 536     if (useVertexShader) {
 537         IWineD3DBaseShaderImpl* vshader = (IWineD3DBaseShaderImpl*) stateBlock->vertexShader;
 538
 539         /* Load DirectX 9 float constants for vertex shader */
 540         deviceImpl->highest_dirty_vs_const = shader_arb_load_constantsF(
 541                 vshader, gl_info, GL_VERTEX_PROGRAM_ARB,
 542                 deviceImpl->highest_dirty_vs_const,
 543                 stateBlock->vertexShaderConstantF,
 544                 deviceImpl->activeContext->vshader_const_dirty);
 545
 546         shader_arb_vs_local_constants(deviceImpl);
 547     }
 548
 549     if (usePixelShader) {
 550         IWineD3DBaseShaderImpl* pshader = (IWineD3DBaseShaderImpl*) stateBlock->pixelShader;
 551
 552         /* Load DirectX 9 float constants for pixel shader */
 553         deviceImpl->highest_dirty_ps_const = shader_arb_load_constantsF(
 554                 pshader, gl_info, GL_FRAGMENT_PROGRAM_ARB,
 555                 deviceImpl->highest_dirty_ps_const,
 556                 stateBlock->pixelShaderConstantF,
 557                 deviceImpl->activeContext->pshader_const_dirty);
 558         shader_arb_ps_local_constants(deviceImpl);
 559     }
 560 }
 561
 562 static void shader_arb_update_float_vertex_constants(IWineD3DDevice *iface, UINT start, UINT count)
 563 {
 564     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
 565
 566     /* We don't want shader constant dirtification to be an O(contexts), so just dirtify the active
 567      * context. On a context switch the old context will be fully dirtified */
 568     memset(This->activeContext->vshader_const_dirty + start, 1,
 569             sizeof(*This->activeContext->vshader_const_dirty) * count);
 570     This->highest_dirty_vs_const = max(This->highest_dirty_vs_const, start + count);
 571 }
 572
 573 static void shader_arb_update_float_pixel_constants(IWineD3DDevice *iface, UINT start, UINT count)
 574 {
 575     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
 576
 577     /* We don't want shader constant dirtification to be an O(contexts), so just dirtify the active
 578      * context. On a context switch the old context will be fully dirtified */
 579     memset(This->activeContext->pshader_const_dirty + start, 1,
 580             sizeof(*This->activeContext->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         SHADER_BUFFER *buffer, const WineD3D_GL_Info *gl_info, DWORD *lconst_map, DWORD *num_clipplanes,
 607         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     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(SHADER_BUFFER *buffer, const char *reg, DWORD dst_mask,
1086                                  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     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     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     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     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     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     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     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_EXP: instruction = "EX2"; break;
1559         case WINED3DSIH_EXPP: instruction = "EXP"; break;
1560         case WINED3DSIH_FRC: instruction = "FRC"; break;
1561         case WINED3DSIH_LIT: instruction = "LIT"; break;
1562         case WINED3DSIH_LRP: instruction = "LRP"; break;
1563         case WINED3DSIH_MAD: instruction = "MAD"; break;
1564         case WINED3DSIH_MAX: instruction = "MAX"; break;
1565         case WINED3DSIH_MIN: instruction = "MIN"; break;
1566         case WINED3DSIH_MOV: instruction = "MOV"; break;
1567         case WINED3DSIH_MUL: instruction = "MUL"; break;
1568         case WINED3DSIH_SGE: instruction = "SGE"; break;
1569         case WINED3DSIH_SLT: instruction = "SLT"; break;
1570         case WINED3DSIH_SUB: instruction = "SUB"; break;
1571         case WINED3DSIH_MOVA:instruction = "ARR"; break;
1572         case WINED3DSIH_SGN: instruction = "SSG"; break;
1573         case WINED3DSIH_DSX: instruction = "DDX"; break;
1574         default: instruction = "";
1575             FIXME("Unhandled opcode %#x\n", ins->handler_idx);
1576             break;
1577     }
1578
1579     /* Note that shader_arb_add_dst_param() adds spaces. */
1580     arguments[0] = '\0';
1581     shader_arb_get_dst_param(ins, dst, dst_str);
1582     for (i = 0; i < ins->src_count; ++i)
1583     {
1584         char operand[100];
1585         strcat(arguments, ", ");
1586         shader_arb_get_src_param(ins, &ins->src[i], i, operand);
1587         strcat(arguments, operand);
1588     }
1589     shader_addline(buffer, "%s%s %s%s;\n", instruction, shader_arb_get_modifier(ins), dst_str, arguments);
1590 }
1591
1592 static void shader_hw_nop(const struct wined3d_shader_instruction *ins)
1593 {
1594     SHADER_BUFFER *buffer = ins->ctx->buffer;
1595     shader_addline(buffer, "NOP;\n");
1596 }
1597
1598 static void shader_hw_mov(const struct wined3d_shader_instruction *ins)
1599 {
1600     IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
1601     BOOL pshader = shader_is_pshader_version(shader->baseShader.reg_maps.shader_version.type);
1602     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
1603
1604     SHADER_BUFFER *buffer = ins->ctx->buffer;
1605     char src0_param[256];
1606
1607     if(ins->handler_idx == WINED3DSIH_MOVA) {
1608         struct wined3d_shader_src_param tmp_src = ins->src[0];
1609         char write_mask[6];
1610
1611         if(ctx->target_version >= NV2) {
1612             shader_hw_map2gl(ins);
1613             return;
1614         }
1615         tmp_src.swizzle = (tmp_src.swizzle & 0x3) * 0x55;
1616         shader_arb_get_src_param(ins, &tmp_src, 0, src0_param);
1617         shader_arb_get_write_mask(ins, &ins->dst[0], write_mask);
1618
1619         /* This implements the mova formula used in GLSL. The first two instructions
1620          * prepare the sign() part. Note that it is fine to have my_sign(0.0) = 1.0
1621          * in this case:
1622          * mova A0.x, 0.0
1623          *
1624          * A0.x = arl(floor(abs(0.0) + 0.5) * 1.0) = floor(0.5) = 0.0 since arl does a floor
1625          *
1626          * The ARL is performed when A0 is used - the requested component is read from A0_SHADOW into
1627          * A0.x. We can use the overwritten component of A0_shadow as temporary storage for the sign.
1628          */
1629         shader_addline(buffer, "SGE A0_SHADOW%s, %s, mova_const.y;\n", write_mask, src0_param);
1630         shader_addline(buffer, "MAD A0_SHADOW%s, A0_SHADOW, mova_const.z, -mova_const.w;\n", write_mask);
1631
1632         shader_addline(buffer, "ABS TA%s, %s;\n", write_mask, src0_param);
1633         shader_addline(buffer, "ADD TA%s, TA, mova_const.x;\n", write_mask);
1634         shader_addline(buffer, "FLR TA%s, TA;\n", write_mask);
1635         if (((IWineD3DVertexShaderImpl *)shader)->rel_offset)
1636         {
1637             shader_addline(buffer, "ADD TA%s, TA, helper_const.z;\n", write_mask);
1638         }
1639         shader_addline(buffer, "MUL A0_SHADOW%s, TA, A0_SHADOW;\n", write_mask);
1640
1641         ((struct shader_arb_ctx_priv *)ins->ctx->backend_data)->addr_reg[0] = '\0';
1642     } else if (ins->ctx->reg_maps->shader_version.major == 1
1643           && !shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type)
1644           && ins->dst[0].reg.type == WINED3DSPR_ADDR)
1645     {
1646         src0_param[0] = '\0';
1647         if (((IWineD3DVertexShaderImpl *)shader)->rel_offset)
1648         {
1649             shader_arb_get_src_param(ins, &ins->src[0], 0, src0_param);
1650             shader_addline(buffer, "ADD TA.x, %s, helper_const.z;\n", src0_param);
1651             shader_addline(buffer, "ARL A0.x, TA.x;\n");
1652         }
1653         else
1654         {
1655             /* Apple's ARB_vertex_program implementation does not accept an ARL source argument
1656              * with more than one component. Thus replicate the first source argument over all
1657              * 4 components. For example, .xyzw -> .x (or better: .xxxx), .zwxy -> .z, etc) */
1658             struct wined3d_shader_src_param tmp_src = ins->src[0];
1659             tmp_src.swizzle = (tmp_src.swizzle & 0x3) * 0x55;
1660             shader_arb_get_src_param(ins, &tmp_src, 0, src0_param);
1661             shader_addline(buffer, "ARL A0.x, %s;\n", src0_param);
1662         }
1663     }
1664     else if(ins->dst[0].reg.type == WINED3DSPR_COLOROUT && ins->dst[0].reg.idx == 0 && pshader)
1665     {
1666         IWineD3DPixelShaderImpl *ps = (IWineD3DPixelShaderImpl *) shader;
1667         if(ctx->cur_ps_args->super.srgb_correction && ps->color0_mov)
1668         {
1669             shader_addline(buffer, "#mov handled in srgb write code\n");
1670             return;
1671         }
1672         shader_hw_map2gl(ins);
1673     }
1674     else
1675     {
1676         shader_hw_map2gl(ins);
1677     }
1678 }
1679
1680 static void pshader_hw_texkill(const struct wined3d_shader_instruction *ins)
1681 {
1682     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1683     SHADER_BUFFER *buffer = ins->ctx->buffer;
1684     char reg_dest[40];
1685
1686     /* No swizzles are allowed in d3d's texkill. PS 1.x ignores the 4th component as documented,
1687      * but >= 2.0 honors it(undocumented, but tested by the d3d9 testsuit)
1688      */
1689     shader_arb_get_dst_param(ins, dst, reg_dest);
1690
1691     if (ins->ctx->reg_maps->shader_version.major >= 2)
1692     {
1693         const char *kilsrc = "TA";
1694         BOOL is_color;
1695
1696         shader_arb_get_register_name(ins, &dst->reg, reg_dest, &is_color);
1697         if(dst->write_mask == WINED3DSP_WRITEMASK_ALL)
1698         {
1699             kilsrc = reg_dest;
1700         }
1701         else
1702         {
1703             /* Sigh. KIL doesn't support swizzles/writemasks. KIL passes a writemask, but ".xy" for example
1704              * is not valid as a swizzle in ARB (needs ".xyyy"). Use SWZ to load the register properly, and set
1705              * masked out components to 0(won't kill)
1706              */
1707             char x = '0', y = '0', z = '0', w = '0';
1708             if(dst->write_mask & WINED3DSP_WRITEMASK_0) x = 'x';
1709             if(dst->write_mask & WINED3DSP_WRITEMASK_1) y = 'y';
1710             if(dst->write_mask & WINED3DSP_WRITEMASK_2) z = 'z';
1711             if(dst->write_mask & WINED3DSP_WRITEMASK_3) w = 'w';
1712             shader_addline(buffer, "SWZ TA, %s, %c, %c, %c, %c;\n", reg_dest, x, y, z, w);
1713         }
1714         shader_addline(buffer, "KIL %s;\n", kilsrc);
1715     } else {
1716         /* ARB fp doesn't like swizzles on the parameter of the KIL instruction. To mask the 4th component,
1717          * copy the register into our general purpose TMP variable, overwrite .w and pass TMP to KIL
1718          *
1719          * ps_1_3 shaders use the texcoord incarnation of the Tx register. ps_1_4 shaders can use the same,
1720          * or pass in any temporary register(in shader phase 2)
1721          */
1722         if(ins->ctx->reg_maps->shader_version.minor <= 3) {
1723             sprintf(reg_dest, "fragment.texcoord[%u]", dst->reg.idx);
1724         } else {
1725             shader_arb_get_dst_param(ins, dst, reg_dest);
1726         }
1727         shader_addline(buffer, "SWZ TA, %s, x, y, z, 1;\n", reg_dest);
1728         shader_addline(buffer, "KIL TA;\n");
1729     }
1730 }
1731
1732 static void pshader_hw_tex(const struct wined3d_shader_instruction *ins)
1733 {
1734     IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)ins->ctx->shader;
1735     IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
1736     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1737     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
1738             ins->ctx->reg_maps->shader_version.minor);
1739     struct wined3d_shader_src_param src;
1740
1741     char reg_dest[40];
1742     char reg_coord[40];
1743     DWORD reg_sampler_code;
1744     DWORD myflags = 0;
1745
1746     /* All versions have a destination register */
1747     shader_arb_get_dst_param(ins, dst, reg_dest);
1748
1749     /* 1.0-1.4: Use destination register number as texture code.
1750        2.0+: Use provided sampler number as texure code. */
1751     if (shader_version < WINED3D_SHADER_VERSION(2,0))
1752         reg_sampler_code = dst->reg.idx;
1753     else
1754         reg_sampler_code = ins->src[1].reg.idx;
1755
1756     /* 1.0-1.3: Use the texcoord varying.
1757        1.4+: Use provided coordinate source register. */
1758     if (shader_version < WINED3D_SHADER_VERSION(1,4))
1759         sprintf(reg_coord, "fragment.texcoord[%u]", reg_sampler_code);
1760     else {
1761         /* TEX is the only instruction that can handle DW and DZ natively */
1762         src = ins->src[0];
1763         if(src.modifiers == WINED3DSPSM_DW) src.modifiers = WINED3DSPSM_NONE;
1764         if(src.modifiers == WINED3DSPSM_DZ) src.modifiers = WINED3DSPSM_NONE;
1765         shader_arb_get_src_param(ins, &src, 0, reg_coord);
1766     }
1767
1768     /* projection flag:
1769      * 1.1, 1.2, 1.3: Use WINED3DTSS_TEXTURETRANSFORMFLAGS
1770      * 1.4: Use WINED3DSPSM_DZ or WINED3DSPSM_DW on src[0]
1771      * 2.0+: Use WINED3DSI_TEXLD_PROJECT on the opcode
1772      */
1773     if (shader_version < WINED3D_SHADER_VERSION(1,4))
1774     {
1775         DWORD flags = 0;
1776         if(reg_sampler_code < MAX_TEXTURES) {
1777             flags = deviceImpl->stateBlock->textureState[reg_sampler_code][WINED3DTSS_TEXTURETRANSFORMFLAGS];
1778         }
1779         if (flags & WINED3DTTFF_PROJECTED) {
1780             myflags |= TEX_PROJ;
1781         }
1782     }
1783     else if (shader_version < WINED3D_SHADER_VERSION(2,0))
1784     {
1785         DWORD src_mod = ins->src[0].modifiers;
1786         if (src_mod == WINED3DSPSM_DZ) {
1787             /* TXP cannot handle DZ natively, so move the z coordinate to .w. reg_coord is a read-only
1788              * varying register, so we need a temp reg
1789              */
1790             shader_addline(ins->ctx->buffer, "SWZ TA, %s, x, y, z, z;\n", reg_coord);
1791             strcpy(reg_coord, "TA");
1792             myflags |= TEX_PROJ;
1793         } else if(src_mod == WINED3DSPSM_DW) {
1794             myflags |= TEX_PROJ;
1795         }
1796     } else {
1797         if (ins->flags & WINED3DSI_TEXLD_PROJECT) myflags |= TEX_PROJ;
1798         if (ins->flags & WINED3DSI_TEXLD_BIAS) myflags |= TEX_BIAS;
1799     }
1800     shader_hw_sample(ins, reg_sampler_code, reg_dest, reg_coord, myflags, NULL, NULL);
1801 }
1802
1803 static void pshader_hw_texcoord(const struct wined3d_shader_instruction *ins)
1804 {
1805     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1806     SHADER_BUFFER *buffer = ins->ctx->buffer;
1807     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
1808             ins->ctx->reg_maps->shader_version.minor);
1809     char dst_str[50];
1810
1811     if (shader_version < WINED3D_SHADER_VERSION(1,4))
1812     {
1813         DWORD reg = dst->reg.idx;
1814
1815         shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
1816         shader_addline(buffer, "MOV_SAT %s, fragment.texcoord[%u];\n", dst_str, reg);
1817     } else {
1818         char reg_src[40];
1819
1820         shader_arb_get_src_param(ins, &ins->src[0], 0, reg_src);
1821         shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
1822         shader_addline(buffer, "MOV %s, %s;\n", dst_str, reg_src);
1823    }
1824 }
1825
1826 static void pshader_hw_texreg2ar(const struct wined3d_shader_instruction *ins)
1827 {
1828      SHADER_BUFFER *buffer = ins->ctx->buffer;
1829      IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)ins->ctx->shader;
1830      IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
1831      DWORD flags;
1832
1833      DWORD reg1 = ins->dst[0].reg.idx;
1834      char dst_str[50];
1835      char src_str[50];
1836
1837      /* Note that texreg2ar treats Tx as a temporary register, not as a varying */
1838      shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
1839      shader_arb_get_src_param(ins, &ins->src[0], 0, src_str);
1840      /* Move .x first in case src_str is "TA" */
1841      shader_addline(buffer, "MOV TA.y, %s.x;\n", src_str);
1842      shader_addline(buffer, "MOV TA.x, %s.w;\n", src_str);
1843      flags = reg1 < MAX_TEXTURES ? deviceImpl->stateBlock->textureState[reg1][WINED3DTSS_TEXTURETRANSFORMFLAGS] : 0;
1844      shader_hw_sample(ins, reg1, dst_str, "TA", flags & WINED3DTTFF_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
1845 }
1846
1847 static void pshader_hw_texreg2gb(const struct wined3d_shader_instruction *ins)
1848 {
1849      SHADER_BUFFER *buffer = ins->ctx->buffer;
1850
1851      DWORD reg1 = ins->dst[0].reg.idx;
1852      char dst_str[50];
1853      char src_str[50];
1854
1855      /* Note that texreg2gb treats Tx as a temporary register, not as a varying */
1856      shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
1857      shader_arb_get_src_param(ins, &ins->src[0], 0, src_str);
1858      shader_addline(buffer, "MOV TA.x, %s.y;\n", src_str);
1859      shader_addline(buffer, "MOV TA.y, %s.z;\n", src_str);
1860      shader_hw_sample(ins, reg1, dst_str, "TA", 0, NULL, NULL);
1861 }
1862
1863 static void pshader_hw_texreg2rgb(const struct wined3d_shader_instruction *ins)
1864 {
1865     DWORD reg1 = ins->dst[0].reg.idx;
1866     char dst_str[50];
1867     char src_str[50];
1868
1869     /* Note that texreg2rg treats Tx as a temporary register, not as a varying */
1870     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
1871     shader_arb_get_src_param(ins, &ins->src[0], 0, src_str);
1872     shader_hw_sample(ins, reg1, dst_str, src_str, 0, NULL, NULL);
1873 }
1874
1875 static void pshader_hw_texbem(const struct wined3d_shader_instruction *ins)
1876 {
1877     IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)ins->ctx->shader;
1878     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1879     SHADER_BUFFER *buffer = ins->ctx->buffer;
1880     char reg_coord[40], dst_reg[50], src_reg[50];
1881     DWORD reg_dest_code;
1882
1883     /* All versions have a destination register. The Tx where the texture coordinates come
1884      * from is the varying incarnation of the texture register
1885      */
1886     reg_dest_code = dst->reg.idx;
1887     shader_arb_get_dst_param(ins, &ins->dst[0], dst_reg);
1888     shader_arb_get_src_param(ins, &ins->src[0], 0, src_reg);
1889     sprintf(reg_coord, "fragment.texcoord[%u]", reg_dest_code);
1890
1891     /* Sampling the perturbation map in Tsrc was done already, including the signedness correction if needed
1892      * The Tx in which the perturbation map is stored is the tempreg incarnation of the texture register
1893      *
1894      * GL_NV_fragment_program_option could handle this in one instruction via X2D:
1895      * X2D TA.xy, fragment.texcoord, T%u, bumpenvmat%u.xzyw
1896      *
1897      * However, the NV extensions are never enabled for <= 2.0 shaders because of the performance penalty that
1898      * comes with it, and texbem is an 1.x only instruction. No 1.x instruction forces us to enable the NV
1899      * extension.
1900      */
1901     shader_addline(buffer, "SWZ TB, bumpenvmat%d, x, z, 0, 0;\n", reg_dest_code);
1902     shader_addline(buffer, "DP3 TA.x, TB, %s;\n", src_reg);
1903     shader_addline(buffer, "SWZ TB, bumpenvmat%d, y, w, 0, 0;\n", reg_dest_code);
1904     shader_addline(buffer, "DP3 TA.y, TB, %s;\n", src_reg);
1905
1906     /* with projective textures, texbem only divides the static texture coord, not the displacement,
1907      * so we can't let the GL handle this.
1908      */
1909     if (((IWineD3DDeviceImpl*) This->baseShader.device)->stateBlock->textureState[reg_dest_code][WINED3DTSS_TEXTURETRANSFORMFLAGS]
1910             & WINED3DTTFF_PROJECTED) {
1911         shader_addline(buffer, "RCP TB.w, %s.w;\n", reg_coord);
1912         shader_addline(buffer, "MUL TB.xy, %s, TB.w;\n", reg_coord);
1913         shader_addline(buffer, "ADD TA.xy, TA, TB;\n");
1914     } else {
1915         shader_addline(buffer, "ADD TA.xy, TA, %s;\n", reg_coord);
1916     }
1917
1918     shader_hw_sample(ins, reg_dest_code, dst_reg, "TA", 0, NULL, NULL);
1919
1920     if (ins->handler_idx == WINED3DSIH_TEXBEML)
1921     {
1922         /* No src swizzles are allowed, so this is ok */
1923         shader_addline(buffer, "MAD TA, %s.z, luminance%d.x, luminance%d.y;\n",
1924                        src_reg, reg_dest_code, reg_dest_code);
1925         shader_addline(buffer, "MUL %s, %s, TA;\n", dst_reg, dst_reg);
1926     }
1927 }
1928
1929 static void pshader_hw_texm3x2pad(const struct wined3d_shader_instruction *ins)
1930 {
1931     DWORD reg = ins->dst[0].reg.idx;
1932     SHADER_BUFFER *buffer = ins->ctx->buffer;
1933     char src0_name[50], dst_name[50];
1934     BOOL is_color;
1935     struct wined3d_shader_register tmp_reg = ins->dst[0].reg;
1936
1937     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
1938     /* The next instruction will be a texm3x2tex or texm3x2depth that writes to the uninitialized
1939      * T<reg+1> register. Use this register to store the calculated vector
1940      */
1941     tmp_reg.idx = reg + 1;
1942     shader_arb_get_register_name(ins, &tmp_reg, dst_name, &is_color);
1943     shader_addline(buffer, "DP3 %s.x, fragment.texcoord[%u], %s;\n", dst_name, reg, src0_name);
1944 }
1945
1946 static void pshader_hw_texm3x2tex(const struct wined3d_shader_instruction *ins)
1947 {
1948     IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)ins->ctx->shader;
1949     IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
1950     DWORD flags;
1951     DWORD reg = ins->dst[0].reg.idx;
1952     SHADER_BUFFER *buffer = ins->ctx->buffer;
1953     char dst_str[50];
1954     char src0_name[50];
1955     char dst_reg[50];
1956     BOOL is_color;
1957
1958     /* We know that we're writing to the uninitialized T<reg> register, so use it for temporary storage */
1959     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_reg, &is_color);
1960
1961     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
1962     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
1963     shader_addline(buffer, "DP3 %s.y, fragment.texcoord[%u], %s;\n", dst_reg, reg, src0_name);
1964     flags = reg < MAX_TEXTURES ? deviceImpl->stateBlock->textureState[reg][WINED3DTSS_TEXTURETRANSFORMFLAGS] : 0;
1965     shader_hw_sample(ins, reg, dst_str, dst_reg, flags & WINED3DTTFF_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
1966 }
1967
1968 static void pshader_hw_texm3x3pad(const struct wined3d_shader_instruction *ins)
1969 {
1970     IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)ins->ctx->shader;
1971     DWORD reg = ins->dst[0].reg.idx;
1972     SHADER_BUFFER *buffer = ins->ctx->buffer;
1973     SHADER_PARSE_STATE* current_state = &This->baseShader.parse_state;
1974     char src0_name[50], dst_name[50];
1975     struct wined3d_shader_register tmp_reg = ins->dst[0].reg;
1976     BOOL is_color;
1977
1978     /* There are always 2 texm3x3pad instructions followed by one texm3x3[tex,vspec, ...] instruction, with
1979      * incrementing ins->dst[0].register_idx numbers. So the pad instruction already knows the final destination
1980      * register, and this register is uninitialized(otherwise the assembler complains that it is 'redeclared')
1981      */
1982     tmp_reg.idx = reg + 2 - current_state->current_row;
1983     shader_arb_get_register_name(ins, &tmp_reg, dst_name, &is_color);
1984
1985     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
1986     shader_addline(buffer, "DP3 %s.%c, fragment.texcoord[%u], %s;\n",
1987                    dst_name, 'x' + current_state->current_row, reg, src0_name);
1988     current_state->texcoord_w[current_state->current_row++] = reg;
1989 }
1990
1991 static void pshader_hw_texm3x3tex(const struct wined3d_shader_instruction *ins)
1992 {
1993     IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)ins->ctx->shader;
1994     IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
1995     DWORD flags;
1996     DWORD reg = ins->dst[0].reg.idx;
1997     SHADER_BUFFER *buffer = ins->ctx->buffer;
1998     SHADER_PARSE_STATE* current_state = &This->baseShader.parse_state;
1999     char dst_str[50];
2000     char src0_name[50], dst_name[50];
2001     BOOL is_color;
2002
2003     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2004     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2005     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_name, reg, src0_name);
2006
2007     /* Sample the texture using the calculated coordinates */
2008     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2009     flags = reg < MAX_TEXTURES ? deviceImpl->stateBlock->textureState[reg][WINED3DTSS_TEXTURETRANSFORMFLAGS] : 0;
2010     shader_hw_sample(ins, reg, dst_str, dst_name, flags & WINED3DTTFF_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2011     current_state->current_row = 0;
2012 }
2013
2014 static void pshader_hw_texm3x3vspec(const struct wined3d_shader_instruction *ins)
2015 {
2016     IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)ins->ctx->shader;
2017     IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
2018     DWORD flags;
2019     DWORD reg = ins->dst[0].reg.idx;
2020     SHADER_BUFFER *buffer = ins->ctx->buffer;
2021     SHADER_PARSE_STATE* current_state = &This->baseShader.parse_state;
2022     char dst_str[50];
2023     char src0_name[50];
2024     char dst_reg[8];
2025     BOOL is_color;
2026
2027     /* Get the dst reg without writemask strings. We know this register is uninitialized, so we can use all
2028      * components for temporary data storage
2029      */
2030     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_reg, &is_color);
2031     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2032     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_reg, reg, src0_name);
2033
2034     /* Construct the eye-ray vector from w coordinates */
2035     shader_addline(buffer, "MOV TB.x, fragment.texcoord[%u].w;\n", current_state->texcoord_w[0]);
2036     shader_addline(buffer, "MOV TB.y, fragment.texcoord[%u].w;\n", current_state->texcoord_w[1]);
2037     shader_addline(buffer, "MOV TB.z, fragment.texcoord[%u].w;\n", reg);
2038
2039     /* Calculate reflection vector
2040      */
2041     shader_addline(buffer, "DP3 %s.w, %s, TB;\n", dst_reg, dst_reg);
2042     /* The .w is ignored when sampling, so I can use TB.w to calculate dot(N, N) */
2043     shader_addline(buffer, "DP3 TB.w, %s, %s;\n", dst_reg, dst_reg);
2044     shader_addline(buffer, "RCP TB.w, TB.w;\n");
2045     shader_addline(buffer, "MUL %s.w, %s.w, TB.w;\n", dst_reg, dst_reg);
2046     shader_addline(buffer, "MUL %s, %s.w, %s;\n", dst_reg, dst_reg, dst_reg);
2047     shader_addline(buffer, "MAD %s, coefmul.x, %s, -TB;\n", dst_reg, dst_reg);
2048
2049     /* Sample the texture using the calculated coordinates */
2050     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2051     flags = reg < MAX_TEXTURES ? deviceImpl->stateBlock->textureState[reg][WINED3DTSS_TEXTURETRANSFORMFLAGS] : 0;
2052     shader_hw_sample(ins, reg, dst_str, dst_reg, flags & WINED3DTTFF_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2053     current_state->current_row = 0;
2054 }
2055
2056 static void pshader_hw_texm3x3spec(const struct wined3d_shader_instruction *ins)
2057 {
2058     IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)ins->ctx->shader;
2059     IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
2060     DWORD flags;
2061     DWORD reg = ins->dst[0].reg.idx;
2062     SHADER_PARSE_STATE* current_state = &This->baseShader.parse_state;
2063     SHADER_BUFFER *buffer = ins->ctx->buffer;
2064     char dst_str[50];
2065     char src0_name[50];
2066     char src1_name[50];
2067     char dst_reg[8];
2068     BOOL is_color;
2069
2070     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2071     shader_arb_get_src_param(ins, &ins->src[0], 1, src1_name);
2072     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_reg, &is_color);
2073     /* Note: dst_reg.xy is input here, generated by two texm3x3pad instructions */
2074     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_reg, reg, src0_name);
2075
2076     /* Calculate reflection vector.
2077      *
2078      *                   dot(N, E)
2079      * dst_reg.xyz = 2 * --------- * N - E
2080      *                   dot(N, N)
2081      *
2082      * Which normalizes the normal vector
2083      */
2084     shader_addline(buffer, "DP3 %s.w, %s, %s;\n", dst_reg, dst_reg, src1_name);
2085     shader_addline(buffer, "DP3 TC.w, %s, %s;\n", dst_reg, dst_reg);
2086     shader_addline(buffer, "RCP TC.w, TC.w;\n");
2087     shader_addline(buffer, "MUL %s.w, %s.w, TC.w;\n", dst_reg, dst_reg);
2088     shader_addline(buffer, "MUL %s, %s.w, %s;\n", dst_reg, dst_reg, dst_reg);
2089     shader_addline(buffer, "MAD %s, coefmul.x, %s, -%s;\n", dst_reg, dst_reg, src1_name);
2090
2091     /* Sample the texture using the calculated coordinates */
2092     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2093     flags = reg < MAX_TEXTURES ? deviceImpl->stateBlock->textureState[reg][WINED3DTSS_TEXTURETRANSFORMFLAGS] : 0;
2094     shader_hw_sample(ins, reg, dst_str, dst_reg, flags & WINED3DTTFF_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2095     current_state->current_row = 0;
2096 }
2097
2098 static void pshader_hw_texdepth(const struct wined3d_shader_instruction *ins)
2099 {
2100     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2101     SHADER_BUFFER *buffer = ins->ctx->buffer;
2102     char dst_name[50];
2103
2104     /* texdepth has an implicit destination, the fragment depth value. It's only parameter,
2105      * which is essentially an input, is the destination register because it is the first
2106      * parameter. According to the msdn, this must be register r5, but let's keep it more flexible
2107      * here(writemasks/swizzles are not valid on texdepth)
2108      */
2109     shader_arb_get_dst_param(ins, dst, dst_name);
2110
2111     /* According to the msdn, the source register(must be r5) is unusable after
2112      * the texdepth instruction, so we're free to modify it
2113      */
2114     shader_addline(buffer, "MIN %s.y, %s.y, one.y;\n", dst_name, dst_name);
2115
2116     /* How to deal with the special case dst_name.g == 0? if r != 0, then
2117      * the r * (1 / 0) will give infinity, which is clamped to 1.0, the correct
2118      * result. But if r = 0.0, then 0 * inf = 0, which is incorrect.
2119      */
2120     shader_addline(buffer, "RCP %s.y, %s.y;\n", dst_name, dst_name);
2121     shader_addline(buffer, "MUL TA.x, %s.x, %s.y;\n", dst_name, dst_name);
2122     shader_addline(buffer, "MIN TA.x, TA.x, one.x;\n");
2123     shader_addline(buffer, "MAX result.depth, TA.x, 0.0;\n");
2124 }
2125
2126 /** Process the WINED3DSIO_TEXDP3TEX instruction in ARB:
2127  * Take a 3-component dot product of the TexCoord[dstreg] and src,
2128  * then perform a 1D texture lookup from stage dstregnum, place into dst. */
2129 static void pshader_hw_texdp3tex(const struct wined3d_shader_instruction *ins)
2130 {
2131     SHADER_BUFFER *buffer = ins->ctx->buffer;
2132     DWORD sampler_idx = ins->dst[0].reg.idx;
2133     char src0[50];
2134     char dst_str[50];
2135
2136     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2137     shader_addline(buffer, "MOV TB, 0.0;\n");
2138     shader_addline(buffer, "DP3 TB.x, fragment.texcoord[%u], %s;\n", sampler_idx, src0);
2139
2140     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2141     shader_hw_sample(ins, sampler_idx, dst_str, "TB", 0 /* Only one coord, can't be projected */, NULL, NULL);
2142 }
2143
2144 /** Process the WINED3DSIO_TEXDP3 instruction in ARB:
2145  * Take a 3-component dot product of the TexCoord[dstreg] and src. */
2146 static void pshader_hw_texdp3(const struct wined3d_shader_instruction *ins)
2147 {
2148     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2149     char src0[50];
2150     char dst_str[50];
2151     SHADER_BUFFER *buffer = ins->ctx->buffer;
2152
2153     /* Handle output register */
2154     shader_arb_get_dst_param(ins, dst, dst_str);
2155     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2156     shader_addline(buffer, "DP3 %s, fragment.texcoord[%u], %s;\n", dst_str, dst->reg.idx, src0);
2157 }
2158
2159 /** Process the WINED3DSIO_TEXM3X3 instruction in ARB
2160  * Perform the 3rd row of a 3x3 matrix multiply */
2161 static void pshader_hw_texm3x3(const struct wined3d_shader_instruction *ins)
2162 {
2163     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2164     SHADER_BUFFER *buffer = ins->ctx->buffer;
2165     char dst_str[50], dst_name[50];
2166     char src0[50];
2167     BOOL is_color;
2168
2169     shader_arb_get_dst_param(ins, dst, dst_str);
2170     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2171     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2172     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_name, dst->reg.idx, src0);
2173     shader_addline(buffer, "MOV %s, %s;\n", dst_str, dst_name);
2174 }
2175
2176 /** Process the WINED3DSIO_TEXM3X2DEPTH instruction in ARB:
2177  * Last row of a 3x2 matrix multiply, use the result to calculate the depth:
2178  * Calculate tmp0.y = TexCoord[dstreg] . src.xyz;  (tmp0.x has already been calculated)
2179  * depth = (tmp0.y == 0.0) ? 1.0 : tmp0.x / tmp0.y
2180  */
2181 static void pshader_hw_texm3x2depth(const struct wined3d_shader_instruction *ins)
2182 {
2183     SHADER_BUFFER *buffer = ins->ctx->buffer;
2184     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2185     char src0[50], dst_name[50];
2186     BOOL is_color;
2187
2188     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2189     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2190     shader_addline(buffer, "DP3 %s.y, fragment.texcoord[%u], %s;\n", dst_name, dst->reg.idx, src0);
2191
2192     /* How to deal with the special case dst_name.g == 0? if r != 0, then
2193      * the r * (1 / 0) will give infinity, which is clamped to 1.0, the correct
2194      * result. But if r = 0.0, then 0 * inf = 0, which is incorrect.
2195      */
2196     shader_addline(buffer, "RCP %s.y, %s.y;\n", dst_name, dst_name);
2197     shader_addline(buffer, "MUL %s.x, %s.x, %s.y;\n", dst_name, dst_name, dst_name);
2198     shader_addline(buffer, "MIN %s.x, %s.x, one.x;\n", dst_name, dst_name);
2199     shader_addline(buffer, "MAX result.depth, %s.x, 0.0;\n", dst_name);
2200 }
2201
2202 /** Handles transforming all WINED3DSIO_M?x? opcodes for
2203     Vertex/Pixel shaders to ARB_vertex_program codes */
2204 static void shader_hw_mnxn(const struct wined3d_shader_instruction *ins)
2205 {
2206     int i;
2207     int nComponents = 0;
2208     struct wined3d_shader_dst_param tmp_dst = {{0}};
2209     struct wined3d_shader_src_param tmp_src[2] = {{{0}}};
2210     struct wined3d_shader_instruction tmp_ins;
2211
2212     memset(&tmp_ins, 0, sizeof(tmp_ins));
2213
2214     /* Set constants for the temporary argument */
2215     tmp_ins.ctx = ins->ctx;
2216     tmp_ins.dst_count = 1;
2217     tmp_ins.dst = &tmp_dst;
2218     tmp_ins.src_count = 2;
2219     tmp_ins.src = tmp_src;
2220
2221     switch(ins->handler_idx)
2222     {
2223         case WINED3DSIH_M4x4:
2224             nComponents = 4;
2225             tmp_ins.handler_idx = WINED3DSIH_DP4;
2226             break;
2227         case WINED3DSIH_M4x3:
2228             nComponents = 3;
2229             tmp_ins.handler_idx = WINED3DSIH_DP4;
2230             break;
2231         case WINED3DSIH_M3x4:
2232             nComponents = 4;
2233             tmp_ins.handler_idx = WINED3DSIH_DP3;
2234             break;
2235         case WINED3DSIH_M3x3:
2236             nComponents = 3;
2237             tmp_ins.handler_idx = WINED3DSIH_DP3;
2238             break;
2239         case WINED3DSIH_M3x2:
2240             nComponents = 2;
2241             tmp_ins.handler_idx = WINED3DSIH_DP3;
2242             break;
2243         default:
2244             FIXME("Unhandled opcode %#x\n", ins->handler_idx);
2245             break;
2246     }
2247
2248     tmp_dst = ins->dst[0];
2249     tmp_src[0] = ins->src[0];
2250     tmp_src[1] = ins->src[1];
2251     for (i = 0; i < nComponents; i++) {
2252         tmp_dst.write_mask = WINED3DSP_WRITEMASK_0 << i;
2253         shader_hw_map2gl(&tmp_ins);
2254         ++tmp_src[1].reg.idx;
2255     }
2256 }
2257
2258 static void shader_hw_rsq_rcp(const struct wined3d_shader_instruction *ins)
2259 {
2260     SHADER_BUFFER *buffer = ins->ctx->buffer;
2261     const char *instruction;
2262
2263     char dst[50];
2264     char src[50];
2265
2266     switch(ins->handler_idx)
2267     {
2268         case WINED3DSIH_RSQ: instruction = "RSQ"; break;
2269         case WINED3DSIH_RCP: instruction = "RCP"; break;
2270         default: instruction = "";
2271             FIXME("Unhandled opcode %#x\n", ins->handler_idx);
2272             break;
2273     }
2274
2275     shader_arb_get_dst_param(ins, &ins->dst[0], dst); /* Destination */
2276     shader_arb_get_src_param(ins, &ins->src[0], 0, src);
2277     if (ins->src[0].swizzle == WINED3DSP_NOSWIZZLE)
2278     {
2279         /* Dx sdk says .x is used if no swizzle is given, but our test shows that
2280          * .w is used
2281          */
2282         strcat(src, ".w");
2283     }
2284
2285     shader_addline(buffer, "%s%s %s, %s;\n", instruction, shader_arb_get_modifier(ins), dst, src);
2286 }
2287
2288 static void shader_hw_nrm(const struct wined3d_shader_instruction *ins)
2289 {
2290     SHADER_BUFFER *buffer = ins->ctx->buffer;
2291     char dst_name[50];
2292     char src_name[50];
2293     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2294     BOOL pshader = shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type);
2295
2296     shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2297     shader_arb_get_src_param(ins, &ins->src[0], 1 /* Use TB */, src_name);
2298
2299     if(pshader && priv->target_version >= NV3)
2300     {
2301         shader_addline(buffer, "NRM%s %s, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name);
2302     }
2303     else
2304     {
2305         shader_addline(buffer, "DP3 TA, %s, %s;\n", src_name, src_name);
2306         shader_addline(buffer, "RSQ TA, TA.x;\n");
2307         /* dst.w = src[0].w * 1 / (src.x^2 + src.y^2 + src.z^2)^(1/2) according to msdn*/
2308         shader_addline(buffer, "MUL%s %s, %s, TA;\n", shader_arb_get_modifier(ins), dst_name,
2309                     src_name);
2310     }
2311 }
2312
2313 static void shader_hw_lrp(const struct wined3d_shader_instruction *ins)
2314 {
2315     SHADER_BUFFER *buffer = ins->ctx->buffer;
2316     char dst_name[50];
2317     char src_name[3][50];
2318
2319     /* ARB_fragment_program has a convenient LRP instruction */
2320     if(shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type)) {
2321         shader_hw_map2gl(ins);
2322         return;
2323     }
2324
2325     shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2326     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
2327     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
2328     shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
2329
2330     shader_addline(buffer, "SUB TA, %s, %s;\n", src_name[1], src_name[2]);
2331     shader_addline(buffer, "MAD%s %s, %s, TA, %s;\n", shader_arb_get_modifier(ins),
2332                    dst_name, src_name[0], src_name[2]);
2333 }
2334
2335 static void shader_hw_sincos(const struct wined3d_shader_instruction *ins)
2336 {
2337     /* This instruction exists in ARB, but the d3d instruction takes two extra parameters which
2338      * must contain fixed constants. So we need a separate function to filter those constants and
2339      * can't use map2gl
2340      */
2341     SHADER_BUFFER *buffer = ins->ctx->buffer;
2342     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2343     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2344     char dst_name[50];
2345     char src_name0[50], src_name1[50], src_name2[50];
2346     BOOL is_color;
2347
2348     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name0);
2349     if(shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type)) {
2350         shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2351         /* No modifiers are supported on SCS */
2352         shader_addline(buffer, "SCS %s, %s;\n", dst_name, src_name0);
2353
2354         if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE)
2355         {
2356             shader_arb_get_register_name(ins, &dst->reg, src_name0, &is_color);
2357             shader_addline(buffer, "MOV_SAT %s, %s;\n", dst_name, src_name0);
2358         }
2359     } else if(priv->target_version >= NV2) {
2360         shader_arb_get_register_name(ins, &dst->reg, dst_name, &is_color);
2361
2362         /* Sincos writemask must be .x, .y or .xy */
2363         if(dst->write_mask & WINED3DSP_WRITEMASK_0)
2364             shader_addline(buffer, "COS%s %s.x, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name0);
2365         if(dst->write_mask & WINED3DSP_WRITEMASK_1)
2366             shader_addline(buffer, "SIN%s %s.y, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name0);
2367     } else {
2368         /* Approximate sine and cosine with a taylor series, as per math textbook. The application passes 8
2369          * helper constants(D3DSINCOSCONST1 and D3DSINCOSCONST2) in src1 and src2.
2370          *
2371          * sin(x) = x - x^3/3! + x^5/5! - x^7/7! + ...
2372          * cos(x) = 1 - x^2/2! + x^4/4! - x^6/6! + ...
2373          *
2374          * The constants we get are:
2375          *
2376          *  +1   +1,     -1     -1     +1      +1      -1       -1
2377          *      ---- ,  ---- , ---- , ----- , ----- , ----- , ------
2378          *      1!*2    2!*4   3!*8   4!*16   5!*32   6!*64   7!*128
2379          *
2380          * If used with x^2, x^3, x^4 etc they calculate sin(x/2) and cos(x/2):
2381          *
2382          * (x/2)^2 = x^2 / 4
2383          * (x/2)^3 = x^3 / 8
2384          * (x/2)^4 = x^4 / 16
2385          * (x/2)^5 = x^5 / 32
2386          * etc
2387          *
2388          * To get the final result:
2389          * sin(x) = 2 * sin(x/2) * cos(x/2)
2390          * cos(x) = cos(x/2)^2 - sin(x/2)^2
2391          * (from sin(x+y) and cos(x+y) rules)
2392          *
2393          * As per MSDN, dst.z is undefined after the operation, and so is
2394          * dst.x and dst.y if they're masked out by the writemask. Ie
2395          * sincos dst.y, src1, c0, c1
2396          * returns the sine in dst.y. dst.x and dst.z are undefined, dst.w is not touched. The assembler
2397          * vsa.exe also stops with an error if the dest register is the same register as the source
2398          * register. This means we can use dest.xyz as temporary storage. The assembler vsa.exe output also
2399          * indicates that sincos consumes 8 instruction slots in vs_2_0(and, strangely, in vs_3_0).
2400          */
2401         shader_arb_get_src_param(ins, &ins->src[1], 1, src_name1);
2402         shader_arb_get_src_param(ins, &ins->src[2], 2, src_name2);
2403         shader_arb_get_register_name(ins, &dst->reg, dst_name, &is_color);
2404
2405         shader_addline(buffer, "MUL %s.x, %s, %s;\n", dst_name, src_name0, src_name0);  /* x ^ 2 */
2406         shader_addline(buffer, "MUL TA.y, %s.x, %s;\n", dst_name, src_name0);           /* x ^ 3 */
2407         shader_addline(buffer, "MUL %s.y, TA.y, %s;\n", dst_name, src_name0);           /* x ^ 4 */
2408         shader_addline(buffer, "MUL TA.z, %s.y, %s;\n", dst_name, src_name0);           /* x ^ 5 */
2409         shader_addline(buffer, "MUL %s.z, TA.z, %s;\n", dst_name, src_name0);           /* x ^ 6 */
2410         shader_addline(buffer, "MUL TA.w, %s.z, %s;\n", dst_name, src_name0);           /* x ^ 7 */
2411
2412         /* sin(x/2)
2413          *
2414          * Unfortunately we don't get the constants in a DP4-capable form. Is there a way to
2415          * properly merge that with MULs in the code above?
2416          * The swizzles .yz and xw however fit into the .yzxw swizzle added to ps_2_0. Maybe
2417          * we can merge the sine and cosine MAD rows to calculate them together.
2418          */
2419         shader_addline(buffer, "MUL TA.x, %s, %s.w;\n", src_name0, src_name2); /* x^1, +1/(1!*2) */
2420         shader_addline(buffer, "MAD TA.x, TA.y, %s.x, TA.x;\n", src_name2); /* -1/(3!*8) */
2421         shader_addline(buffer, "MAD TA.x, TA.z, %s.w, TA.x;\n", src_name1); /* +1/(5!*32) */
2422         shader_addline(buffer, "MAD TA.x, TA.w, %s.x, TA.x;\n", src_name1); /* -1/(7!*128) */
2423
2424         /* cos(x/2) */
2425         shader_addline(buffer, "MAD TA.y, %s.x, %s.y, %s.z;\n", dst_name, src_name2, src_name2); /* -1/(2!*4), +1.0 */
2426         shader_addline(buffer, "MAD TA.y, %s.y, %s.z, TA.y;\n", dst_name, src_name1); /* +1/(4!*16) */
2427         shader_addline(buffer, "MAD TA.y, %s.z, %s.y, TA.y;\n", dst_name, src_name1); /* -1/(6!*64) */
2428
2429         if(dst->write_mask & WINED3DSP_WRITEMASK_0) {
2430             /* cos x */
2431             shader_addline(buffer, "MUL TA.z, TA.y, TA.y;\n");
2432             shader_addline(buffer, "MAD %s.x, -TA.x, TA.x, TA.z;\n", dst_name);
2433         }
2434         if(dst->write_mask & WINED3DSP_WRITEMASK_1) {
2435             /* sin x */
2436             shader_addline(buffer, "MUL %s.y, TA.x, TA.y;\n", dst_name);
2437             shader_addline(buffer, "ADD %s.y, %s.y, %s.y;\n", dst_name, dst_name, dst_name);
2438         }
2439     }
2440 }
2441
2442 static void shader_hw_sgn(const struct wined3d_shader_instruction *ins)
2443 {
2444     SHADER_BUFFER *buffer = ins->ctx->buffer;
2445     char dst_name[50];
2446     char src_name[50];
2447     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
2448
2449     /* SGN is only valid in vertex shaders */
2450     if(ctx->target_version == NV2) {
2451         shader_hw_map2gl(ins);
2452         return;
2453     }
2454     shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2455     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name);
2456
2457     /* If SRC > 0.0, -SRC < SRC = TRUE, otherwise false.
2458      * if SRC < 0.0,  SRC < -SRC = TRUE. If neither is true, src = 0.0
2459      */
2460     if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE) {
2461         shader_addline(buffer, "SLT %s, -%s, %s;\n", dst_name, src_name, src_name);
2462     } else {
2463         /* src contains TA? Write to the dest first. This won't overwrite our destination.
2464          * Then use TA, and calculate the final result
2465          *
2466          * Not reading from TA? Store the first result in TA to avoid overwriting the
2467          * destination if src reg = dst reg
2468          */
2469         if(strstr(src_name, "TA"))
2470         {
2471             shader_addline(buffer, "SLT %s,  %s, -%s;\n", dst_name, src_name, src_name);
2472             shader_addline(buffer, "SLT TA, -%s, %s;\n", src_name, src_name);
2473             shader_addline(buffer, "ADD %s, %s, -TA;\n", dst_name, dst_name);
2474         }
2475         else
2476         {
2477             shader_addline(buffer, "SLT TA, -%s, %s;\n", src_name, src_name);
2478             shader_addline(buffer, "SLT %s,  %s, -%s;\n", dst_name, src_name, src_name);
2479             shader_addline(buffer, "ADD %s, TA, -%s;\n", dst_name, dst_name);
2480         }
2481     }
2482 }
2483
2484 static void shader_hw_dsy(const struct wined3d_shader_instruction *ins)
2485 {
2486     SHADER_BUFFER *buffer = ins->ctx->buffer;
2487     char src[50];
2488     char dst[50];
2489     char dst_name[50];
2490     BOOL is_color;
2491
2492     shader_arb_get_dst_param(ins, &ins->dst[0], dst);
2493     shader_arb_get_src_param(ins, &ins->src[0], 0, src);
2494     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2495
2496     shader_addline(buffer, "DDY %s, %s;\n", dst, src);
2497     shader_addline(buffer, "MUL%s %s, %s, ycorrection.y;\n", shader_arb_get_modifier(ins), dst, dst_name);
2498 }
2499
2500 static DWORD abs_modifier(DWORD mod, BOOL *need_abs)
2501 {
2502     *need_abs = FALSE;
2503
2504     switch(mod)
2505     {
2506         case WINED3DSPSM_NONE:      return WINED3DSPSM_ABS;
2507         case WINED3DSPSM_NEG:       return WINED3DSPSM_ABS;
2508         case WINED3DSPSM_BIAS:      *need_abs = TRUE; return WINED3DSPSM_BIAS;
2509         case WINED3DSPSM_BIASNEG:   *need_abs = TRUE; return WINED3DSPSM_BIASNEG;
2510         case WINED3DSPSM_SIGN:      *need_abs = TRUE; return WINED3DSPSM_SIGN;
2511         case WINED3DSPSM_SIGNNEG:   *need_abs = TRUE; return WINED3DSPSM_SIGNNEG;
2512         case WINED3DSPSM_COMP:      *need_abs = TRUE; return WINED3DSPSM_COMP;
2513         case WINED3DSPSM_X2:        *need_abs = TRUE; return WINED3DSPSM_X2;
2514         case WINED3DSPSM_X2NEG:     *need_abs = TRUE; return WINED3DSPSM_X2NEG;
2515         case WINED3DSPSM_DZ:        *need_abs = TRUE; return WINED3DSPSM_DZ;
2516         case WINED3DSPSM_DW:        *need_abs = TRUE; return WINED3DSPSM_DW;
2517         case WINED3DSPSM_ABS:       return WINED3DSPSM_ABS;
2518         case WINED3DSPSM_ABSNEG:    return WINED3DSPSM_ABS;
2519     }
2520     FIXME("Unknown modifier %u\n", mod);
2521     return mod;
2522 }
2523
2524 static void shader_hw_log_pow(const struct wined3d_shader_instruction *ins)
2525 {
2526     SHADER_BUFFER *buffer = ins->ctx->buffer;
2527     char src0[50], src1[50], dst[50];
2528     struct wined3d_shader_src_param src0_copy = ins->src[0];
2529     BOOL need_abs = FALSE;
2530     const char *instr;
2531     BOOL arg2 = FALSE;
2532
2533     switch(ins->handler_idx)
2534     {
2535         case WINED3DSIH_LOG:  instr = "LG2"; break;
2536         case WINED3DSIH_LOGP: instr = "LOG"; break;
2537         case WINED3DSIH_POW:  instr = "POW"; arg2 = TRUE; break;
2538         default:
2539             ERR("Unexpected instruction %d\n", ins->handler_idx);
2540             return;
2541     }
2542
2543     /* LOG, LOGP and POW operate on the absolute value of the input */
2544     src0_copy.modifiers = abs_modifier(src0_copy.modifiers, &need_abs);
2545
2546     shader_arb_get_dst_param(ins, &ins->dst[0], dst);
2547     shader_arb_get_src_param(ins, &src0_copy, 0, src0);
2548     if(arg2) shader_arb_get_src_param(ins, &ins->src[1], 1, src1);
2549
2550     if(need_abs)
2551     {
2552         shader_addline(buffer, "ABS TA, %s;\n", src0);
2553         if(arg2)
2554         {
2555             shader_addline(buffer, "%s%s %s, TA, %s;\n", instr, shader_arb_get_modifier(ins), dst, src1);
2556         }
2557         else
2558         {
2559             shader_addline(buffer, "%s%s %s, TA;\n", instr, shader_arb_get_modifier(ins), dst);
2560         }
2561     }
2562     else if(arg2)
2563     {
2564         shader_addline(buffer, "%s%s %s, %s, %s;\n", instr, shader_arb_get_modifier(ins), dst, src0, src1);
2565     }
2566     else
2567     {
2568         shader_addline(buffer, "%s%s %s, %s;\n", instr, shader_arb_get_modifier(ins), dst, src0);
2569     }
2570 }
2571
2572 static void shader_hw_loop(const struct wined3d_shader_instruction *ins)
2573 {
2574     SHADER_BUFFER *buffer = ins->ctx->buffer;
2575     char src_name[50];
2576     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2577
2578     /* src0 is aL */
2579     shader_arb_get_src_param(ins, &ins->src[1], 0, src_name);
2580
2581     if(vshader)
2582     {
2583         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2584         struct list *e = list_head(&priv->control_frames);
2585         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2586
2587         if(priv->loop_depth > 1) shader_addline(buffer, "PUSHA aL;\n");
2588         /* The constant loader makes sure to load -1 into iX.w */
2589         shader_addline(buffer, "ARLC aL, %s.xywz;\n", src_name);
2590         shader_addline(buffer, "BRA loop_%u_end (LE.x);\n", control_frame->loop_no);
2591         shader_addline(buffer, "loop_%u_start:\n", control_frame->loop_no);
2592     }
2593     else
2594     {
2595         shader_addline(buffer, "LOOP %s;\n", src_name);
2596     }
2597 }
2598
2599 static void shader_hw_rep(const struct wined3d_shader_instruction *ins)
2600 {
2601     SHADER_BUFFER *buffer = ins->ctx->buffer;
2602     char src_name[50];
2603     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2604
2605     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name);
2606
2607     /* The constant loader makes sure to load -1 into iX.w */
2608     if(vshader)
2609     {
2610         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2611         struct list *e = list_head(&priv->control_frames);
2612         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2613
2614         if(priv->loop_depth > 1) shader_addline(buffer, "PUSHA aL;\n");
2615
2616         shader_addline(buffer, "ARLC aL, %s.xywz;\n", src_name);
2617         shader_addline(buffer, "BRA loop_%u_end (LE.x);\n", control_frame->loop_no);
2618         shader_addline(buffer, "loop_%u_start:\n", control_frame->loop_no);
2619     }
2620     else
2621     {
2622         shader_addline(buffer, "REP %s;\n", src_name);
2623     }
2624 }
2625
2626 static void shader_hw_endloop(const struct wined3d_shader_instruction *ins)
2627 {
2628     SHADER_BUFFER *buffer = ins->ctx->buffer;
2629     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2630
2631     if(vshader)
2632     {
2633         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2634         struct list *e = list_head(&priv->control_frames);
2635         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2636
2637         shader_addline(buffer, "ARAC aL.xy, aL;\n");
2638         shader_addline(buffer, "BRA loop_%u_start (GT.x);\n", control_frame->loop_no);
2639         shader_addline(buffer, "loop_%u_end:\n", control_frame->loop_no);
2640
2641         if(priv->loop_depth > 1) shader_addline(buffer, "POPA aL;\n");
2642     }
2643     else
2644     {
2645         shader_addline(buffer, "ENDLOOP;\n");
2646     }
2647 }
2648
2649 static void shader_hw_endrep(const struct wined3d_shader_instruction *ins)
2650 {
2651     SHADER_BUFFER *buffer = ins->ctx->buffer;
2652     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2653
2654     if(vshader)
2655     {
2656         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2657         struct list *e = list_head(&priv->control_frames);
2658         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2659
2660         shader_addline(buffer, "ARAC aL.xy, aL;\n");
2661         shader_addline(buffer, "BRA loop_%u_start (GT.x);\n", control_frame->loop_no);
2662         shader_addline(buffer, "loop_%u_end:\n", control_frame->loop_no);
2663
2664         if(priv->loop_depth > 1) shader_addline(buffer, "POPA aL;\n");
2665     }
2666     else
2667     {
2668         shader_addline(buffer, "ENDREP;\n");
2669     }
2670 }
2671
2672 static const struct control_frame *find_last_loop(const struct shader_arb_ctx_priv *priv)
2673 {
2674     struct control_frame *control_frame;
2675
2676     LIST_FOR_EACH_ENTRY(control_frame, &priv->control_frames, struct control_frame, entry)
2677     {
2678         if(control_frame->type == LOOP || control_frame->type == REP) return control_frame;
2679     }
2680     ERR("Could not find loop for break\n");
2681     return NULL;
2682 }
2683
2684 static void shader_hw_break(const struct wined3d_shader_instruction *ins)
2685 {
2686     SHADER_BUFFER *buffer = ins->ctx->buffer;
2687     const struct control_frame *control_frame = find_last_loop(ins->ctx->backend_data);
2688     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2689
2690     if(vshader)
2691     {
2692         shader_addline(buffer, "BRA loop_%u_end;\n", control_frame->loop_no);
2693     }
2694     else
2695     {
2696         shader_addline(buffer, "BRK;\n");
2697     }
2698 }
2699
2700 static const char *get_compare(COMPARISON_TYPE flags)
2701 {
2702     switch (flags)
2703     {
2704         case COMPARISON_GT: return "GT";
2705         case COMPARISON_EQ: return "EQ";
2706         case COMPARISON_GE: return "GE";
2707         case COMPARISON_LT: return "LT";
2708         case COMPARISON_NE: return "NE";
2709         case COMPARISON_LE: return "LE";
2710         default:
2711             FIXME("Unrecognized comparison value: %u\n", flags);
2712             return "(\?\?)";
2713     }
2714 }
2715
2716 static COMPARISON_TYPE invert_compare(COMPARISON_TYPE flags)
2717 {
2718     switch (flags)
2719     {
2720         case COMPARISON_GT: return COMPARISON_LE;
2721         case COMPARISON_EQ: return COMPARISON_NE;
2722         case COMPARISON_GE: return COMPARISON_LT;
2723         case COMPARISON_LT: return COMPARISON_GE;
2724         case COMPARISON_NE: return COMPARISON_EQ;
2725         case COMPARISON_LE: return COMPARISON_GT;
2726         default:
2727             FIXME("Unrecognized comparison value: %u\n", flags);
2728             return -1;
2729     }
2730 }
2731
2732 static void shader_hw_breakc(const struct wined3d_shader_instruction *ins)
2733 {
2734     SHADER_BUFFER *buffer = ins->ctx->buffer;
2735     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2736     const struct control_frame *control_frame = find_last_loop(ins->ctx->backend_data);
2737     char src_name0[50];
2738     char src_name1[50];
2739     const char *comp = get_compare(ins->flags);
2740
2741     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name0);
2742     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name1);
2743
2744     if(vshader)
2745     {
2746         /* SUBC CC, src0, src1" works only in pixel shaders, so use TA to throw
2747          * away the subtraction result
2748          */
2749         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
2750         shader_addline(buffer, "BRA loop_%u_end (%s.x);\n", control_frame->loop_no, comp);
2751     }
2752     else
2753     {
2754         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
2755         shader_addline(buffer, "BRK (%s.x);\n", comp);
2756     }
2757 }
2758
2759 static void shader_hw_ifc(const struct wined3d_shader_instruction *ins)
2760 {
2761     SHADER_BUFFER *buffer = ins->ctx->buffer;
2762     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2763     struct list *e = list_head(&priv->control_frames);
2764     struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2765     const char *comp;
2766     char src_name0[50];
2767     char src_name1[50];
2768     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2769
2770     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name0);
2771     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name1);
2772
2773     if(vshader)
2774     {
2775         /* Invert the flag. We jump to the else label if the condition is NOT true */
2776         comp = get_compare(invert_compare(ins->flags));
2777         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
2778         shader_addline(buffer, "BRA ifc_%u_else (%s.x);\n", control_frame->ifc_no, comp);
2779     }
2780     else
2781     {
2782         comp = get_compare(ins->flags);
2783         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
2784         shader_addline(buffer, "IF %s.x;\n", comp);
2785     }
2786 }
2787
2788 static void shader_hw_else(const struct wined3d_shader_instruction *ins)
2789 {
2790     SHADER_BUFFER *buffer = ins->ctx->buffer;
2791     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2792     struct list *e = list_head(&priv->control_frames);
2793     struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2794     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2795
2796     if(vshader)
2797     {
2798         shader_addline(buffer, "BRA ifc_%u_endif;\n", control_frame->ifc_no);
2799         shader_addline(buffer, "ifc_%u_else:\n", control_frame->ifc_no);
2800         control_frame->had_else = TRUE;
2801     }
2802     else
2803     {
2804         shader_addline(buffer, "ELSE;\n");
2805     }
2806 }
2807
2808 static void shader_hw_endif(const struct wined3d_shader_instruction *ins)
2809 {
2810     SHADER_BUFFER *buffer = ins->ctx->buffer;
2811     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2812     struct list *e = list_head(&priv->control_frames);
2813     struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2814     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2815
2816     if(vshader)
2817     {
2818         if(control_frame->had_else)
2819         {
2820             shader_addline(buffer, "ifc_%u_endif:\n", control_frame->ifc_no);
2821         }
2822         else
2823         {
2824             shader_addline(buffer, "#No else branch. else is endif\n");
2825             shader_addline(buffer, "ifc_%u_else:\n", control_frame->ifc_no);
2826         }
2827     }
2828     else
2829     {
2830         shader_addline(buffer, "ENDIF;\n");
2831     }
2832 }
2833
2834 static void shader_hw_texldd(const struct wined3d_shader_instruction *ins)
2835 {
2836     DWORD sampler_idx = ins->src[1].reg.idx;
2837     char reg_dest[40];
2838     char reg_src[3][40];
2839     DWORD flags = TEX_DERIV;
2840
2841     shader_arb_get_dst_param(ins, &ins->dst[0], reg_dest);
2842     shader_arb_get_src_param(ins, &ins->src[0], 0, reg_src[0]);
2843     shader_arb_get_src_param(ins, &ins->src[2], 1, reg_src[1]);
2844     shader_arb_get_src_param(ins, &ins->src[3], 2, reg_src[2]);
2845
2846     if (ins->flags & WINED3DSI_TEXLD_PROJECT) flags |= TEX_PROJ;
2847     if (ins->flags & WINED3DSI_TEXLD_BIAS) flags |= TEX_BIAS;
2848
2849     shader_hw_sample(ins, sampler_idx, reg_dest, reg_src[0], flags, reg_src[1], reg_src[2]);
2850 }
2851
2852 static void shader_hw_texldl(const struct wined3d_shader_instruction *ins)
2853 {
2854     DWORD sampler_idx = ins->src[1].reg.idx;
2855     char reg_dest[40];
2856     char reg_coord[40];
2857     DWORD flags = TEX_LOD;
2858
2859     shader_arb_get_dst_param(ins, &ins->dst[0], reg_dest);
2860     shader_arb_get_src_param(ins, &ins->src[0], 0, reg_coord);
2861
2862     if (ins->flags & WINED3DSI_TEXLD_PROJECT) flags |= TEX_PROJ;
2863     if (ins->flags & WINED3DSI_TEXLD_BIAS) flags |= TEX_BIAS;
2864
2865     shader_hw_sample(ins, sampler_idx, reg_dest, reg_coord, flags, NULL, NULL);
2866 }
2867
2868 static void shader_hw_label(const struct wined3d_shader_instruction *ins)
2869 {
2870     SHADER_BUFFER *buffer = ins->ctx->buffer;
2871     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2872
2873     priv->in_main_func = FALSE;
2874     /* Call instructions activate the NV extensions, not labels and rets. If there is an uncalled
2875      * subroutine, don't generate a label that will make GL complain
2876      */
2877     if(priv->target_version == ARB) return;
2878
2879     shader_addline(buffer, "l%u:\n", ins->src[0].reg.idx);
2880 }
2881
2882 static void vshader_add_footer(IWineD3DVertexShaderImpl *This, SHADER_BUFFER *buffer, const struct arb_vs_compile_args *args,
2883                                struct shader_arb_ctx_priv *priv_ctx)
2884 {
2885     const shader_reg_maps *reg_maps = &This->baseShader.reg_maps;
2886     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *)This->baseShader.device;
2887     const WineD3D_GL_Info *gl_info = &device->adapter->gl_info;
2888     unsigned int i;
2889
2890     /* The D3DRS_FOGTABLEMODE render state defines if the shader-generated fog coord is used
2891      * or if the fragment depth is used. If the fragment depth is used(FOGTABLEMODE != NONE),
2892      * the fog frag coord is thrown away. If the fog frag coord is used, but not written by
2893      * the shader, it is set to 0.0(fully fogged, since start = 1.0, end = 0.0)
2894      */
2895     if(args->super.fog_src == VS_FOG_Z) {
2896         shader_addline(buffer, "MOV result.fogcoord, TMP_OUT.z;\n");
2897     } else if (!reg_maps->fog) {
2898         /* posFixup.x is always 1.0, so we can savely use it */
2899         shader_addline(buffer, "ADD result.fogcoord, posFixup.x, -posFixup.x;\n");
2900     }
2901
2902     /* Write the final position.
2903      *
2904      * OpenGL coordinates specify the center of the pixel while d3d coords specify
2905      * the corner. The offsets are stored in z and w in posFixup. posFixup.y contains
2906      * 1.0 or -1.0 to turn the rendering upside down for offscreen rendering. PosFixup.x
2907      * contains 1.0 to allow a mad, but arb vs swizzles are too restricted for that.
2908      */
2909     shader_addline(buffer, "MUL TA, posFixup, TMP_OUT.w;\n");
2910     shader_addline(buffer, "ADD TMP_OUT.x, TMP_OUT.x, TA.z;\n");
2911     shader_addline(buffer, "MAD TMP_OUT.y, TMP_OUT.y, posFixup.y, TA.w;\n");
2912
2913     if(use_nv_clip(gl_info) && priv_ctx->target_version >= NV2)
2914     {
2915         for(i = 0; i < priv_ctx->vs_clipplanes; i++)
2916         {
2917             shader_addline(buffer, "DP4 result.clip[%u].x, TMP_OUT, state.clip[%u].plane;\n", i, i);
2918         }
2919     }
2920     else if(args->boolclip.clip_control[0])
2921     {
2922         unsigned int cur_clip = 0;
2923         char component[4] = {'x', 'y', 'z', 'w'};
2924
2925         for(i = 0; i < GL_LIMITS(clipplanes); i++)
2926         {
2927             if(args->boolclip.clip_control[1] & (1 << i))
2928             {
2929                 shader_addline(buffer, "DP4 TA.%c, TMP_OUT, state.clip[%u].plane;\n",
2930                                component[cur_clip++], i);
2931             }
2932         }
2933         switch(cur_clip)
2934         {
2935             case 0:
2936                 shader_addline(buffer, "MOV TA, -helper_const.w;\n");
2937                 break;
2938             case 1:
2939                 shader_addline(buffer, "MOV TA.yzw, -helper_const.w;\n");
2940                 break;
2941             case 2:
2942                 shader_addline(buffer, "MOV TA.zw, -helper_const.w;\n");
2943                 break;
2944             case 3:
2945                 shader_addline(buffer, "MOV TA.w, -helper_const.w;\n");
2946                 break;
2947         }
2948         shader_addline(buffer, "MOV result.texcoord[%u], TA;\n",
2949                        args->boolclip.clip_control[0] - 1);
2950     }
2951
2952     /* Z coord [0;1]->[-1;1] mapping, see comment in transform_projection in state.c
2953      * and the glsl equivalent
2954      */
2955     if(need_helper_const(gl_info)) {
2956         shader_addline(buffer, "MAD TMP_OUT.z, TMP_OUT.z, helper_const.x, -TMP_OUT.w;\n");
2957     } else {
2958         shader_addline(buffer, "ADD TMP_OUT.z, TMP_OUT.z, TMP_OUT.z;\n");
2959         shader_addline(buffer, "ADD TMP_OUT.z, TMP_OUT.z, -TMP_OUT.w;\n");
2960     }
2961
2962     shader_addline(buffer, "MOV result.position, TMP_OUT;\n");
2963
2964     priv_ctx->footer_written = TRUE;
2965 }
2966
2967 static void shader_hw_ret(const struct wined3d_shader_instruction *ins)
2968 {
2969     SHADER_BUFFER *buffer = ins->ctx->buffer;
2970     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2971     IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *) ins->ctx->shader;
2972     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2973
2974     if(priv->target_version == ARB) return;
2975
2976     if(vshader)
2977     {
2978         if(priv->in_main_func) vshader_add_footer((IWineD3DVertexShaderImpl *) shader, buffer, priv->cur_vs_args, priv);
2979     }
2980
2981     shader_addline(buffer, "RET;\n");
2982 }
2983
2984 static void shader_hw_call(const struct wined3d_shader_instruction *ins)
2985 {
2986     SHADER_BUFFER *buffer = ins->ctx->buffer;
2987     shader_addline(buffer, "CAL l%u;\n", ins->src[0].reg.idx);
2988 }
2989
2990 /* GL locking is done by the caller */
2991 static GLuint create_arb_blt_vertex_program(const WineD3D_GL_Info *gl_info)
2992 {
2993     GLuint program_id = 0;
2994     const char *blt_vprogram =
2995         "!!ARBvp1.0\n"
2996         "PARAM c[1] = { { 1, 0.5 } };\n"
2997         "MOV result.position, vertex.position;\n"
2998         "MOV result.color, c[0].x;\n"
2999         "MOV result.texcoord[0], vertex.texcoord[0];\n"
3000         "END\n";
3001
3002     GL_EXTCALL(glGenProgramsARB(1, &program_id));
3003     GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, program_id));
3004     GL_EXTCALL(glProgramStringARB(GL_VERTEX_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB, strlen(blt_vprogram), blt_vprogram));
3005
3006     if (glGetError() == GL_INVALID_OPERATION) {
3007         GLint pos;
3008         glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
3009         FIXME("Vertex program error at position %d: %s\n", pos,
3010             debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
3011     }
3012
3013     return program_id;
3014 }
3015
3016 /* GL locking is done by the caller */
3017 static GLuint create_arb_blt_fragment_program(const WineD3D_GL_Info *gl_info, enum tex_types tex_type)
3018 {
3019     GLuint program_id = 0;
3020     static const char * const blt_fprograms[tex_type_count] =
3021     {
3022         /* tex_1d */
3023         NULL,
3024         /* tex_2d */
3025         "!!ARBfp1.0\n"
3026         "TEMP R0;\n"
3027         "TEX R0.x, fragment.texcoord[0], texture[0], 2D;\n"
3028         "MOV result.depth.z, R0.x;\n"
3029         "END\n",
3030         /* tex_3d */
3031         NULL,
3032         /* tex_cube */
3033         "!!ARBfp1.0\n"
3034         "TEMP R0;\n"
3035         "TEX R0.x, fragment.texcoord[0], texture[0], CUBE;\n"
3036         "MOV result.depth.z, R0.x;\n"
3037         "END\n",
3038         /* tex_rect */
3039         "!!ARBfp1.0\n"
3040         "TEMP R0;\n"
3041         "TEX R0.x, fragment.texcoord[0], texture[0], RECT;\n"
3042         "MOV result.depth.z, R0.x;\n"
3043         "END\n",
3044     };
3045
3046     if (!blt_fprograms[tex_type])
3047     {
3048         FIXME("tex_type %#x not supported\n", tex_type);
3049         tex_type = tex_2d;
3050     }
3051
3052     GL_EXTCALL(glGenProgramsARB(1, &program_id));
3053     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, program_id));
3054     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB, strlen(blt_fprograms[tex_type]), blt_fprograms[tex_type]));
3055
3056     if (glGetError() == GL_INVALID_OPERATION) {
3057         GLint pos;
3058         glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
3059         FIXME("Fragment program error at position %d: %s\n", pos,
3060             debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
3061     }
3062
3063     return program_id;
3064 }
3065
3066 static void arbfp_add_sRGB_correction(SHADER_BUFFER *buffer, const char *fragcolor, const char *tmp1,
3067                                       const char *tmp2, const char *tmp3, const char *tmp4, BOOL condcode) {
3068     /* Perform sRGB write correction. See GLX_EXT_framebuffer_sRGB */
3069
3070     if(condcode)
3071     {
3072         /* Sigh. MOVC CC doesn't work, so use one of the temps as dummy dest */
3073         shader_addline(buffer, "SUBC %s, %s.x, srgb_consts1.y;\n", tmp1, fragcolor);
3074         /* Calculate the > 0.0031308 case */
3075         shader_addline(buffer, "POW %s.x (GE), %s.x, srgb_consts1.z;\n", fragcolor, fragcolor);
3076         shader_addline(buffer, "POW %s.y (GE), %s.y, srgb_consts1.z;\n", fragcolor, fragcolor);
3077         shader_addline(buffer, "POW %s.z (GE), %s.z, srgb_consts1.z;\n", fragcolor, fragcolor);
3078         shader_addline(buffer, "MUL %s.xyz (GE), %s, srgb_consts1.w;\n", fragcolor, fragcolor);
3079         shader_addline(buffer, "SUB %s.xyz (GE), %s, srgb_consts2.x;\n", fragcolor, fragcolor);
3080         /* Calculate the < case */
3081         shader_addline(buffer, "MUL %s.xyz (LT), srgb_consts1.x, %s;\n", fragcolor, fragcolor);
3082     }
3083     else
3084     {
3085         /* Calculate the > 0.0031308 case */
3086         shader_addline(buffer, "POW %s.x, %s.x, srgb_consts1.z;\n", tmp1, fragcolor);
3087         shader_addline(buffer, "POW %s.y, %s.y, srgb_consts1.z;\n", tmp1, fragcolor);
3088         shader_addline(buffer, "POW %s.z, %s.z, srgb_consts1.z;\n", tmp1, fragcolor);
3089         shader_addline(buffer, "MUL %s, %s, srgb_consts1.w;\n", tmp1, tmp1);
3090         shader_addline(buffer, "SUB %s, %s, srgb_consts2.x;\n", tmp1, tmp1);
3091         /* Calculate the < case */
3092         shader_addline(buffer, "MUL %s, srgb_consts1.x, %s;\n", tmp2, fragcolor);
3093         /* Get 1.0 / 0.0 masks for > 0.0031308 and < 0.0031308 */
3094         shader_addline(buffer, "SLT %s, srgb_consts1.y, %s;\n", tmp3, fragcolor);
3095         shader_addline(buffer, "SGE %s, srgb_consts1.y, %s;\n", tmp4, fragcolor);
3096         /* Store the components > 0.0031308 in the destination */
3097         shader_addline(buffer, "MUL %s.xyz, %s, %s;\n", fragcolor, tmp1, tmp3);
3098         /* Add the components that are < 0.0031308 */
3099         shader_addline(buffer, "MAD %s.xyz, %s, %s, %s;\n", fragcolor, tmp2, tmp4, fragcolor);
3100         /* Move everything into result.color at once. Nvidia hardware cannot handle partial
3101         * result.color writes(.rgb first, then .a), or handle overwriting already written
3102         * components. The assembler uses a temporary register in this case, which is usually
3103         * not allocated from one of our registers that were used earlier.
3104         */
3105     }
3106     shader_addline(buffer, "MOV result.color, %s;\n", fragcolor);
3107     /* [0.0;1.0] clamping. Not needed, this is done implicitly */
3108 }
3109
3110 static const DWORD *find_loop_control_values(IWineD3DBaseShaderImpl *This, DWORD idx)
3111 {
3112     const local_constant *constant;
3113
3114     LIST_FOR_EACH_ENTRY(constant, &This->baseShader.constantsI, local_constant, entry)
3115     {
3116         if (constant->idx == idx)
3117         {
3118             return constant->value;
3119         }
3120     }
3121     return NULL;
3122 }
3123
3124 static void init_ps_input(const IWineD3DPixelShaderImpl *This, const struct arb_ps_compile_args *args,
3125                           struct shader_arb_ctx_priv *priv)
3126 {
3127     const char *texcoords[8] =
3128     {
3129         "fragment.texcoord[0]", "fragment.texcoord[1]", "fragment.texcoord[2]", "fragment.texcoord[3]",
3130         "fragment.texcoord[4]", "fragment.texcoord[5]", "fragment.texcoord[6]", "fragment.texcoord[7]"
3131     };
3132     unsigned int i;
3133     const struct wined3d_shader_signature_element *sig = This->input_signature;
3134     const char *semantic_name;
3135     DWORD semantic_idx;
3136
3137     switch(args->super.vp_mode)
3138     {
3139         case pretransformed:
3140         case fixedfunction:
3141             /* The pixelshader has to collect the varyings on its own. In any case properly load
3142              * color0 and color1. In the case of pretransformed vertices also load texcoords. Set
3143              * other attribs to 0.0.
3144              *
3145              * For fixedfunction this behavior is correct, according to the tests. For pretransformed
3146              * we'd either need a replacement shader that can load other attribs like BINORMAL, or
3147              * load the texcoord attrib pointers to match the pixel shader signature
3148              */
3149             for(i = 0; i < MAX_REG_INPUT; i++)
3150             {
3151                 semantic_name = sig[i].semantic_name;
3152                 semantic_idx = sig[i].semantic_idx;
3153                 if(semantic_name == NULL) continue;
3154
3155                 if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_COLOR))
3156                 {
3157                     if(semantic_idx == 0) priv->ps_input[i] = "fragment.color.primary";
3158                     else if(semantic_idx == 1) priv->ps_input[i] = "fragment.color.secondary";
3159                     else priv->ps_input[i] = "0.0";
3160                 }
3161                 else if(args->super.vp_mode == fixedfunction)
3162                 {
3163                     priv->ps_input[i] = "0.0";
3164                 }
3165                 else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_TEXCOORD))
3166                 {
3167                     if(semantic_idx < 8) priv->ps_input[i] = texcoords[semantic_idx];
3168                     else priv->ps_input[i] = "0.0";
3169                 }
3170                 else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_FOG))
3171                 {
3172                     if(semantic_idx == 0) priv->ps_input[i] = "fragment.fogcoord";
3173                     else priv->ps_input[i] = "0.0";
3174                 }
3175                 else
3176                 {
3177                     priv->ps_input[i] = "0.0";
3178                 }
3179
3180                 TRACE("v%u, semantic %s%u is %s\n", i, semantic_name, semantic_idx, priv->ps_input[i]);
3181             }
3182             break;
3183
3184         case vertexshader:
3185             /* That one is easy. The vertex shaders provide v0-v7 in fragment.texcoord and v8 and v9 in
3186              * fragment.color
3187              */
3188             for(i = 0; i < 8; i++)
3189             {
3190                 priv->ps_input[i] = texcoords[i];
3191             }
3192             priv->ps_input[8] = "fragment.color.primary";
3193             priv->ps_input[9] = "fragment.color.secondary";
3194             break;
3195     }
3196 }
3197
3198 /* GL locking is done by the caller */
3199 static GLuint shader_arb_generate_pshader(IWineD3DPixelShaderImpl *This,
3200         SHADER_BUFFER *buffer, const struct arb_ps_compile_args *args, struct arb_ps_compiled_shader *compiled)
3201 {
3202     const shader_reg_maps* reg_maps = &This->baseShader.reg_maps;
3203     CONST DWORD *function = This->baseShader.function;
3204     const WineD3D_GL_Info *gl_info = &((IWineD3DDeviceImpl *)This->baseShader.device)->adapter->gl_info;
3205     const local_constant *lconst;
3206     GLuint retval;
3207     char fragcolor[16];
3208     DWORD *lconst_map = local_const_mapping((IWineD3DBaseShaderImpl *) This), next_local, cur;
3209     struct shader_arb_ctx_priv priv_ctx;
3210     BOOL dcl_tmp = args->super.srgb_correction, dcl_td = FALSE;
3211     BOOL want_nv_prog = FALSE;
3212     struct arb_pshader_private *shader_priv = This->backend_priv;
3213
3214     char srgbtmp[4][4];
3215     unsigned int i, found = 0;
3216
3217     for(i = 0; i < This->baseShader.limits.temporary; i++) {
3218
3219         /* Don't overwrite the color source */
3220         if(This->color0_mov && i == This->color0_reg) continue;
3221         else if(reg_maps->shader_version.major < 2 && i == 0) continue;
3222
3223         if(reg_maps->temporary[i]) {
3224             sprintf(srgbtmp[found], "R%u", i);
3225             found++;
3226             if(found == 4) break;
3227         }
3228     }
3229
3230     switch(found) {
3231         case 4: dcl_tmp = FALSE; break;
3232         case 0:
3233             sprintf(srgbtmp[0], "TA");
3234             sprintf(srgbtmp[1], "TB");
3235             sprintf(srgbtmp[2], "TC");
3236             sprintf(srgbtmp[3], "TD");
3237             dcl_td = TRUE;
3238             break;
3239         case 1:
3240             sprintf(srgbtmp[1], "TA");
3241             sprintf(srgbtmp[2], "TB");
3242             sprintf(srgbtmp[3], "TC");
3243             break;
3244         case 2:
3245             sprintf(srgbtmp[2], "TA");
3246             sprintf(srgbtmp[3], "TB");
3247             break;
3248         case 3:
3249             sprintf(srgbtmp[3], "TA");
3250             break;
3251     }
3252
3253     /*  Create the hw ARB shader */
3254     memset(&priv_ctx, 0, sizeof(priv_ctx));
3255     priv_ctx.cur_ps_args = args;
3256     priv_ctx.compiled_fprog = compiled;
3257     priv_ctx.cur_np2fixup_info = &compiled->np2fixup_info;
3258     init_ps_input(This, args, &priv_ctx);
3259     list_init(&priv_ctx.control_frames);
3260
3261     /* Avoid enabling NV_fragment_program* if we do not need it.
3262      *
3263      * Enabling GL_NV_fragment_program_option causes the driver to occupy a temporary register,
3264      * and it slows down the shader execution noticeably(about 5%). Usually our instruction emulation
3265      * is faster than what we gain from using higher native instructions. There are some things though
3266      * that cannot be emulated. In that case enable the extensions.
3267      * If the extension is enabled, instruction handlers that support both ways will use it.
3268      *
3269      * Testing shows no performance difference between OPTION NV_fragment_program2 and NV_fragment_program.
3270      * So enable the best we can get.
3271      */
3272     if(reg_maps->usesdsx || reg_maps->usesdsy || reg_maps->loop_depth > 0 || reg_maps->usestexldd ||
3273        reg_maps->usestexldl || reg_maps->usesfacing || reg_maps->usesifc || reg_maps->usescall)
3274     {
3275         want_nv_prog = TRUE;
3276     }
3277
3278     shader_addline(buffer, "!!ARBfp1.0\n");
3279     if(want_nv_prog && GL_SUPPORT(NV_FRAGMENT_PROGRAM2)) {
3280         shader_addline(buffer, "OPTION NV_fragment_program2;\n");
3281         priv_ctx.target_version = NV3;
3282     } else if(want_nv_prog && GL_SUPPORT(NV_FRAGMENT_PROGRAM_OPTION)) {
3283         shader_addline(buffer, "OPTION NV_fragment_program;\n");
3284         priv_ctx.target_version = NV2;
3285     } else {
3286         if(want_nv_prog)
3287         {
3288             /* This is an error - either we're advertising the wrong shader version, or aren't enforcing some
3289              * limits properly
3290              */
3291             ERR("The shader requires instructions that are not available in plain GL_ARB_fragment_program\n");
3292             ERR("Try GLSL\n");
3293         }
3294         priv_ctx.target_version = ARB;
3295     }
3296
3297     if(This->baseShader.reg_maps.highest_render_target > 0)
3298     {
3299         shader_addline(buffer, "OPTION ARB_draw_buffers;\n");
3300     }
3301
3302     if (reg_maps->shader_version.major < 3)
3303     {
3304         switch(args->super.fog) {
3305             case FOG_OFF:
3306                 break;
3307             case FOG_LINEAR:
3308                 shader_addline(buffer, "OPTION ARB_fog_linear;\n");
3309                 break;
3310             case FOG_EXP:
3311                 shader_addline(buffer, "OPTION ARB_fog_exp;\n");
3312                 break;
3313             case FOG_EXP2:
3314                 shader_addline(buffer, "OPTION ARB_fog_exp2;\n");
3315                 break;
3316         }
3317     }
3318
3319     /* For now always declare the temps. At least the Nvidia assembler optimizes completely
3320      * unused temps away(but occupies them for the whole shader if they're used once). Always
3321      * declaring them avoids tricky bookkeeping work
3322      */
3323     shader_addline(buffer, "TEMP TA;\n");      /* Used for modifiers */
3324     shader_addline(buffer, "TEMP TB;\n");      /* Used for modifiers */
3325     shader_addline(buffer, "TEMP TC;\n");      /* Used for modifiers */
3326     if(dcl_td) shader_addline(buffer, "TEMP TD;\n"); /* Used for sRGB writing */
3327     shader_addline(buffer, "PARAM coefdiv = { 0.5, 0.25, 0.125, 0.0625 };\n");
3328     shader_addline(buffer, "PARAM coefmul = { 2, 4, 8, 16 };\n");
3329     shader_addline(buffer, "PARAM one = { 1.0, 1.0, 1.0, 1.0 };\n");
3330
3331     if (reg_maps->shader_version.major < 2)
3332     {
3333         strcpy(fragcolor, "R0");
3334     } else {
3335         if(args->super.srgb_correction) {
3336             if(This->color0_mov) {
3337                 sprintf(fragcolor, "R%u", This->color0_reg);
3338             } else {
3339                 shader_addline(buffer, "TEMP TMP_COLOR;\n");
3340                 strcpy(fragcolor, "TMP_COLOR");
3341             }
3342         } else {
3343             strcpy(fragcolor, "result.color");
3344         }
3345     }
3346
3347     if(args->super.srgb_correction) {
3348         shader_addline(buffer, "PARAM srgb_consts1 = {%f, %f, %f, %f};\n",
3349                        srgb_mul_low, srgb_cmp, srgb_pow, srgb_mul_high);
3350         shader_addline(buffer, "PARAM srgb_consts2 = {%f, %f, %f, %f};\n",
3351                        srgb_sub_high, 0.0, 0.0, 0.0);
3352     }
3353
3354     /* Base Declarations */
3355     next_local = shader_generate_arb_declarations( (IWineD3DBaseShader*) This, reg_maps, buffer, &GLINFO_LOCATION,
3356             lconst_map, NULL, &priv_ctx);
3357
3358     for(i = 0; i < (sizeof(reg_maps->bumpmat) / sizeof(reg_maps->bumpmat[0])); i++) {
3359         if(!reg_maps->bumpmat[i]) continue;
3360
3361         cur = compiled->numbumpenvmatconsts;
3362         compiled->bumpenvmatconst[cur].const_num = WINED3D_CONST_NUM_UNUSED;
3363         compiled->bumpenvmatconst[cur].texunit = i;
3364         compiled->luminanceconst[cur].const_num = WINED3D_CONST_NUM_UNUSED;
3365         compiled->luminanceconst[cur].texunit = i;
3366
3367         /* We can fit the constants into the constant limit for sure because texbem, texbeml, bem and beml are only supported
3368          * in 1.x shaders, and GL_ARB_fragment_program has a constant limit of 24 constants. So in the worst case we're loading
3369          * 8 shader constants, 8 bump matrices and 8 luminance parameters and are perfectly fine. (No NP2 fixup on bumpmapped
3370          * textures due to conditional NP2 restrictions)
3371          *
3372          * Use local constants to load the bump env parameters, not program.env. This avoids collisions with d3d constants of
3373          * shaders in newer shader models. Since the bump env parameters have to share their space with NP2 fixup constants,
3374          * their location is shader dependent anyway and they cannot be loaded globally.
3375          */
3376         compiled->bumpenvmatconst[cur].const_num = next_local++;
3377         shader_addline(buffer, "PARAM bumpenvmat%d = program.local[%d];\n",
3378                        i, compiled->bumpenvmatconst[cur].const_num);
3379         compiled->numbumpenvmatconsts = cur + 1;
3380
3381         if(!reg_maps->luminanceparams[i]) continue;
3382
3383         compiled->luminanceconst[cur].const_num = next_local++;
3384         shader_addline(buffer, "PARAM luminance%d = program.local[%d];\n",
3385                        i, compiled->luminanceconst[cur].const_num);
3386     }
3387
3388     for(i = 0; i < MAX_CONST_I; i++)
3389     {
3390         compiled->int_consts[i] = WINED3D_CONST_NUM_UNUSED;
3391         if (reg_maps->integer_constants & (1 << i) && priv_ctx.target_version >= NV2)
3392         {
3393             const DWORD *control_values = find_loop_control_values((IWineD3DBaseShaderImpl *) This, i);
3394
3395             if(control_values)
3396             {
3397                 shader_addline(buffer, "PARAM I%u = {%u, %u, %u, -1};\n", i,
3398                                 control_values[0], control_values[1], control_values[2]);
3399             }
3400             else
3401             {
3402                 compiled->int_consts[i] = next_local;
3403                 compiled->num_int_consts++;
3404                 shader_addline(buffer, "PARAM I%u = program.local[%u];\n", i, next_local++);
3405             }
3406         }
3407     }
3408
3409     if(reg_maps->vpos || reg_maps->usesdsy)
3410     {
3411         compiled->ycorrection = next_local;
3412         shader_addline(buffer, "PARAM ycorrection = program.local[%u];\n", next_local++);
3413
3414         if(reg_maps->vpos)
3415         {
3416             shader_addline(buffer, "TEMP vpos;\n");
3417             /* ycorrection.x: Backbuffer height(onscreen) or 0(offscreen).
3418              * ycorrection.y: -1.0(onscreen), 1.0(offscreen)
3419              * ycorrection.z: 1.0
3420              * ycorrection.w: 0.0
3421              */
3422             shader_addline(buffer, "MAD vpos, fragment.position, ycorrection.zyww, ycorrection.wxww;\n");
3423             shader_addline(buffer, "FLR vpos.xy, vpos;\n");
3424         }
3425     }
3426     else
3427     {
3428         compiled->ycorrection = WINED3D_CONST_NUM_UNUSED;
3429     }
3430
3431     /* Load constants to fixup NP2 texcoords if there are still free constants left:
3432      * Constants (texture dimensions) for the NP2 fixup are loaded as local program parameters. This will consume
3433      * at most 8 (MAX_FRAGMENT_SAMPLERS / 2) parameters, which is highly unlikely, since the application had to
3434      * use 16 NP2 textures at the same time. In case that we run out of constants the fixup is simply not
3435      * applied / activated. This will probably result in wrong rendering of the texture, but will save us from
3436      * shader compilation errors and the subsequent errors when drawing with this shader. */
3437     if (priv_ctx.cur_ps_args->super.np2_fixup) {
3438
3439         struct arb_ps_np2fixup_info* const fixup = priv_ctx.cur_np2fixup_info;
3440         const WORD map = priv_ctx.cur_ps_args->super.np2_fixup;
3441         const UINT max_lconsts = gl_info->ps_arb_max_local_constants;
3442
3443         fixup->offset = next_local;
3444         fixup->super.active = 0;
3445
3446         cur = 0;
3447         for (i = 0; i < MAX_FRAGMENT_SAMPLERS; ++i) {
3448             if (!(map & (1 << i))) continue;
3449
3450             if (fixup->offset + (cur >> 1) < max_lconsts) {
3451                 fixup->super.active |= (1 << i);
3452                 fixup->super.idx[i] = cur++;
3453             } else {
3454                 FIXME("No free constant found to load NP2 fixup data into shader. "
3455                       "Sampling from this texture will probably look wrong.\n");
3456                 break;
3457             }
3458         }
3459
3460         fixup->super.num_consts = (cur + 1) >> 1;
3461         if (fixup->super.num_consts) {
3462             shader_addline(buffer, "PARAM np2fixup[%u] = { program.env[%u..%u] };\n",
3463                            fixup->super.num_consts, fixup->offset, fixup->super.num_consts + fixup->offset - 1);
3464         }
3465
3466         next_local += fixup->super.num_consts;
3467     }
3468
3469     if(shader_priv->clipplane_emulation)
3470     {
3471         shader_addline(buffer, "KIL fragment.texcoord[%u];\n", shader_priv->clipplane_emulation - 1);
3472     }
3473
3474     /* Base Shader Body */
3475     shader_generate_main((IWineD3DBaseShader *)This, buffer, reg_maps, function, &priv_ctx);
3476
3477     if(args->super.srgb_correction) {
3478         arbfp_add_sRGB_correction(buffer, fragcolor, srgbtmp[0], srgbtmp[1], srgbtmp[2], srgbtmp[3],
3479                                   priv_ctx.target_version >= NV2);
3480     } else if(reg_maps->shader_version.major < 2) {
3481         shader_addline(buffer, "MOV result.color, %s;\n", fragcolor);
3482     }
3483     shader_addline(buffer, "END\n");
3484
3485     /* TODO: change to resource.glObjectHandle or something like that */
3486     GL_EXTCALL(glGenProgramsARB(1, &retval));
3487
3488     TRACE("Creating a hw pixel shader, prg=%d\n", retval);
3489     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, retval));
3490
3491     TRACE("Created hw pixel shader, prg=%d\n", retval);
3492     /* Create the program and check for errors */
3493     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
3494                buffer->bsize, buffer->buffer));
3495
3496     if (glGetError() == GL_INVALID_OPERATION) {
3497         GLint errPos;
3498         glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &errPos);
3499         FIXME("HW PixelShader Error at position %d: %s\n",
3500               errPos, debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
3501         retval = 0;
3502     }
3503
3504     /* Load immediate constants */
3505     if(lconst_map) {
3506         LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry) {
3507             const float *value = (const float *)lconst->value;
3508             GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, lconst_map[lconst->idx], value));
3509             checkGLcall("glProgramLocalParameter4fvARB");
3510         }
3511         HeapFree(GetProcessHeap(), 0, lconst_map);
3512     }
3513
3514     return retval;
3515 }
3516
3517 static int compare_sig(const struct wined3d_shader_signature_element *sig1, const struct wined3d_shader_signature_element *sig2)
3518 {
3519     unsigned int i;
3520     int ret;
3521
3522     for(i = 0; i < MAX_REG_INPUT; i++)
3523     {
3524         if(sig1[i].semantic_name == NULL || sig2[i].semantic_name == NULL)
3525         {
3526             /* Compare pointers, not contents. One string is NULL(element does not exist), the other one is not NULL */
3527             if(sig1[i].semantic_name != sig2[i].semantic_name) return sig1[i].semantic_name < sig2[i].semantic_name ? -1 : 1;
3528             continue;
3529         }
3530
3531         ret = strcmp(sig1[i].semantic_name, sig2[i].semantic_name);
3532         if(ret != 0) return ret;
3533         if(sig1[i].semantic_idx    != sig2[i].semantic_idx)    return sig1[i].semantic_idx    < sig2[i].semantic_idx    ? -1 : 1;
3534         if(sig1[i].sysval_semantic != sig2[i].sysval_semantic) return sig1[i].sysval_semantic < sig2[i].sysval_semantic ? -1 : 1;
3535         if(sig1[i].component_type  != sig2[i].component_type)  return sig1[i].sysval_semantic < sig2[i].component_type  ? -1 : 1;
3536         if(sig1[i].register_idx    != sig2[i].register_idx)    return sig1[i].register_idx    < sig2[i].register_idx    ? -1 : 1;
3537         if(sig1[i].mask            != sig2->mask)              return sig1[i].mask            < sig2[i].mask            ? -1 : 1;
3538     }
3539     return 0;
3540 }
3541
3542 static struct wined3d_shader_signature_element *clone_sig(const struct wined3d_shader_signature_element *sig)
3543 {
3544     struct wined3d_shader_signature_element *new;
3545     int i;
3546     char *name;
3547
3548     new = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*new) * MAX_REG_INPUT);
3549     for(i = 0; i < MAX_REG_INPUT; i++)
3550     {
3551         if(sig[i].semantic_name == NULL)
3552         {
3553             continue;
3554         }
3555
3556         new[i] = sig[i];
3557         /* Clone the semantic string */
3558         name = HeapAlloc(GetProcessHeap(), 0, strlen(sig[i].semantic_name) + 1);
3559         strcpy(name, sig[i].semantic_name);
3560         new[i].semantic_name = name;
3561     }
3562     return new;
3563 }
3564
3565 static DWORD find_input_signature(struct shader_arb_priv *priv, const struct wined3d_shader_signature_element *sig)
3566 {
3567     struct wine_rb_entry *entry = wine_rb_get(&priv->signature_tree, sig);
3568     struct ps_signature *found_sig;
3569
3570     if(entry != NULL)
3571     {
3572         found_sig = WINE_RB_ENTRY_VALUE(entry, struct ps_signature, entry);
3573         TRACE("Found existing signature %u\n", found_sig->idx);
3574         return found_sig->idx;
3575     }
3576     found_sig = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*sig));
3577     found_sig->sig = clone_sig(sig);
3578     found_sig->idx = priv->ps_sig_number++;
3579     TRACE("New signature stored and assigned number %u\n", found_sig->idx);
3580     if(wine_rb_put(&priv->signature_tree, sig, &found_sig->entry) == -1)
3581     {
3582         ERR("Failed to insert program entry.\n");
3583     }
3584     return found_sig->idx;
3585 }
3586
3587 static void init_output_registers(IWineD3DVertexShaderImpl *shader, DWORD sig_num, struct shader_arb_ctx_priv *priv_ctx,
3588                                   struct arb_vs_compiled_shader *compiled)
3589 {
3590     unsigned int i, j;
3591     static const char *texcoords[8] =
3592     {
3593         "result.texcoord[0]", "result.texcoord[1]", "result.texcoord[2]", "result.texcoord[3]",
3594         "result.texcoord[4]", "result.texcoord[5]", "result.texcoord[6]", "result.texcoord[7]"
3595     };
3596     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *) shader->baseShader.device;
3597     const struct wined3d_shader_signature_element *sig;
3598     const char *semantic_name;
3599     DWORD semantic_idx, reg_idx;
3600
3601     /* Write generic input varyings 0 to 7 to result.texcoord[], varying 8 to result.color.primary
3602      * and varying 9 to result.color.secondary
3603      */
3604     const char *decl_idx_to_string[MAX_REG_INPUT] =
3605     {
3606         texcoords[0], texcoords[1], texcoords[2], texcoords[3],
3607         texcoords[4], texcoords[5], texcoords[6], texcoords[7],
3608         "result.color.primary", "result.color.secondary"
3609     };
3610
3611     if(sig_num == ~0)
3612     {
3613         TRACE("Pixel shader uses builtin varyings\n");
3614         /* Map builtins to builtins */
3615         for(i = 0; i < 8; i++)
3616         {
3617             priv_ctx->texcrd_output[i] = texcoords[i];
3618         }
3619         priv_ctx->color_output[0] = "result.color.primary";
3620         priv_ctx->color_output[1] = "result.color.secondary";
3621         priv_ctx->fog_output = "result.fogcoord";
3622
3623         /* Map declared regs to builtins. Use "TA" to /dev/null unread output */
3624         for(i = 0; i < (sizeof(shader->output_signature) / sizeof(*shader->output_signature)); i++)
3625         {
3626             semantic_name = shader->output_signature[i].semantic_name;
3627             if(semantic_name == NULL) continue;
3628
3629             if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_POSITION))
3630             {
3631                 TRACE("o%u is TMP_OUT\n", i);
3632                 if(shader->output_signature[i].semantic_idx == 0) priv_ctx->vs_output[i] = "TMP_OUT";
3633                 else priv_ctx->vs_output[i] = "TA";
3634             }
3635             else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_PSIZE))
3636             {
3637                 TRACE("o%u is result.pointsize\n", i);
3638                 if(shader->output_signature[i].semantic_idx == 0) priv_ctx->vs_output[i] = "result.pointsize";
3639                 else priv_ctx->vs_output[i] = "TA";
3640             }
3641             else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_COLOR))
3642             {
3643                 TRACE("o%u is result.color.?, idx %u\n", i, shader->output_signature[i].semantic_idx);
3644                 if(shader->output_signature[i].semantic_idx == 0) priv_ctx->vs_output[i] = "result.color.primary";
3645                 else if(shader->output_signature[i].semantic_idx == 1) priv_ctx->vs_output[i] = "result.color.secondary";
3646                 else priv_ctx->vs_output[i] = "TA";
3647             }
3648             else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_TEXCOORD))
3649             {
3650                 TRACE("o%u is %s\n", i, texcoords[shader->output_signature[i].semantic_idx]);
3651                 if(shader->output_signature[i].semantic_idx >= 8) priv_ctx->vs_output[i] = "TA";
3652                 else priv_ctx->vs_output[i] = texcoords[shader->output_signature[i].semantic_idx];
3653             }
3654             else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_FOG))
3655             {
3656                 TRACE("o%u is result.fogcoord\n", i);
3657                 if(shader->output_signature[i].semantic_idx > 0) priv_ctx->vs_output[i] = "TA";
3658                 else priv_ctx->vs_output[i] = "result.fogcoord";
3659             }
3660             else
3661             {
3662                 priv_ctx->vs_output[i] = "TA";
3663             }
3664         }
3665         return;
3666     }
3667
3668     /* Instead of searching for the signature in the signature list, read the one from the current pixel shader.
3669      * Its maybe not the shader where the signature came from, but it is the same signature and faster to find
3670      */
3671     sig = ((IWineD3DPixelShaderImpl *)device->stateBlock->pixelShader)->input_signature;
3672     TRACE("Pixel shader uses declared varyings\n");
3673
3674     /* Map builtin to declared. /dev/null the results by default to the TA temp reg */
3675     for(i = 0; i < 8; i++)
3676     {
3677         priv_ctx->texcrd_output[i] = "TA";
3678     }
3679     priv_ctx->color_output[0] = "TA";
3680     priv_ctx->color_output[1] = "TA";
3681     priv_ctx->fog_output = "TA";
3682
3683     for(i = 0; i < MAX_REG_INPUT; i++)
3684     {
3685         semantic_name = sig[i].semantic_name;
3686         semantic_idx = sig[i].semantic_idx;
3687         reg_idx = sig[i].register_idx;
3688         if(semantic_name == NULL) continue;
3689
3690         /* If a declared input register is not written by builtin arguments, don't write to it.
3691          * GL_NV_vertex_program makes sure the input defaults to 0.0, which is correct with D3D
3692          *
3693          * Don't care about POSITION and PSIZE here - this is a builtin vertex shader, position goes
3694          * to TMP_OUT in any case
3695          */
3696         if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_TEXCOORD))
3697         {
3698             if(semantic_idx < 8) priv_ctx->texcrd_output[semantic_idx] = decl_idx_to_string[reg_idx];
3699         }
3700         else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_COLOR))
3701         {
3702             if(semantic_idx < 2) priv_ctx->color_output[semantic_idx] = decl_idx_to_string[reg_idx];
3703         }
3704         else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_FOG))
3705         {
3706             if(semantic_idx == 0) priv_ctx->fog_output = decl_idx_to_string[reg_idx];
3707         }
3708         else
3709         {
3710             continue;
3711         }
3712
3713         if(strcmp(decl_idx_to_string[reg_idx], "result.color.primary") == 0 ||
3714            strcmp(decl_idx_to_string[reg_idx], "result.color.secondary") == 0)
3715         {
3716             compiled->need_color_unclamp = TRUE;
3717         }
3718     }
3719
3720     /* Map declared to declared */
3721     for(i = 0; i < (sizeof(shader->output_signature) / sizeof(*shader->output_signature)); i++)
3722     {
3723         /* Write unread output to TA to throw them away */
3724         priv_ctx->vs_output[i] = "TA";
3725         semantic_name = shader->output_signature[i].semantic_name;
3726         if(semantic_name == NULL)
3727         {
3728             continue;
3729         }
3730
3731         if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_POSITION) &&
3732            shader->output_signature[i].semantic_idx == 0)
3733         {
3734             priv_ctx->vs_output[i] = "TMP_OUT";
3735             continue;
3736         }
3737         else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_PSIZE) &&
3738            shader->output_signature[i].semantic_idx == 0)
3739         {
3740             priv_ctx->vs_output[i] = "result.pointsize";
3741             continue;
3742         }
3743
3744         for(j = 0; j < MAX_REG_INPUT; j++)
3745         {
3746             if(sig[j].semantic_name == NULL)
3747             {
3748                 continue;
3749             }
3750
3751             if(strcmp(sig[j].semantic_name, semantic_name) == 0 &&
3752                sig[j].semantic_idx == shader->output_signature[i].semantic_idx)
3753             {
3754                 priv_ctx->vs_output[i] = decl_idx_to_string[sig[j].register_idx];
3755
3756                 if(strcmp(priv_ctx->vs_output[i], "result.color.primary") == 0 ||
3757                    strcmp(priv_ctx->vs_output[i], "result.color.secondary") == 0)
3758                 {
3759                     compiled->need_color_unclamp = TRUE;
3760                 }
3761             }
3762         }
3763     }
3764 }
3765
3766 /* GL locking is done by the caller */
3767 static GLuint shader_arb_generate_vshader(IWineD3DVertexShaderImpl *This,
3768         SHADER_BUFFER *buffer, const struct arb_vs_compile_args *args, struct arb_vs_compiled_shader *compiled)
3769 {
3770     const shader_reg_maps *reg_maps = &This->baseShader.reg_maps;
3771     CONST DWORD *function = This->baseShader.function;
3772     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *)This->baseShader.device;
3773     const WineD3D_GL_Info *gl_info = &device->adapter->gl_info;
3774     const local_constant *lconst;
3775     GLuint ret;
3776     DWORD next_local, *lconst_map = local_const_mapping((IWineD3DBaseShaderImpl *) This);
3777     struct shader_arb_ctx_priv priv_ctx;
3778     unsigned int i;
3779
3780     memset(&priv_ctx, 0, sizeof(priv_ctx));
3781     priv_ctx.cur_vs_args = args;
3782     list_init(&priv_ctx.control_frames);
3783     init_output_registers(This, args->ps_signature, &priv_ctx, compiled);
3784
3785     /*  Create the hw ARB shader */
3786     shader_addline(buffer, "!!ARBvp1.0\n");
3787
3788     /* Always enable the NV extension if available. Unlike fragment shaders, there is no
3789      * mesurable performance penalty, and we can always make use of it for clipplanes.
3790      */
3791     if(GL_SUPPORT(NV_VERTEX_PROGRAM3)) {
3792         shader_addline(buffer, "OPTION NV_vertex_program3;\n");
3793         priv_ctx.target_version = NV3;
3794         shader_addline(buffer, "ADDRESS aL;\n");
3795     } else if(GL_SUPPORT(NV_VERTEX_PROGRAM2_OPTION)) {
3796         shader_addline(buffer, "OPTION NV_vertex_program2;\n");
3797         priv_ctx.target_version = NV2;
3798         shader_addline(buffer, "ADDRESS aL;\n");
3799     } else {
3800         priv_ctx.target_version = ARB;
3801     }
3802
3803     shader_addline(buffer, "TEMP TMP_OUT;\n");
3804     if(need_helper_const(gl_info)) {
3805         shader_addline(buffer, "PARAM helper_const = { 2.0, -1.0, %d.0, 0.0 };\n", This->rel_offset);
3806     }
3807     if(need_mova_const((IWineD3DBaseShader *) This, gl_info)) {
3808         shader_addline(buffer, "PARAM mova_const = { 0.5, 0.0, 2.0, 1.0 };\n");
3809         shader_addline(buffer, "TEMP A0_SHADOW;\n");
3810     }
3811
3812     shader_addline(buffer, "TEMP TA;\n");
3813
3814     /* Base Declarations */
3815     next_local = shader_generate_arb_declarations( (IWineD3DBaseShader*) This, reg_maps, buffer, &GLINFO_LOCATION,
3816             lconst_map, &priv_ctx.vs_clipplanes, &priv_ctx);
3817
3818     for(i = 0; i < MAX_CONST_I; i++)
3819     {
3820         compiled->int_consts[i] = WINED3D_CONST_NUM_UNUSED;
3821         if(reg_maps->integer_constants & (1 << i) && priv_ctx.target_version >= NV2)
3822         {
3823             const DWORD *control_values = find_loop_control_values((IWineD3DBaseShaderImpl *) This, i);
3824
3825             if(control_values)
3826             {
3827                 shader_addline(buffer, "PARAM I%u = {%u, %u, %u, -1};\n", i,
3828                                 control_values[0], control_values[1], control_values[2]);
3829             }
3830             else
3831             {
3832                 compiled->int_consts[i] = next_local;
3833                 compiled->num_int_consts++;
3834                 shader_addline(buffer, "PARAM I%u = program.local[%u];\n", i, next_local++);
3835             }
3836         }
3837     }
3838
3839     /* We need a constant to fixup the final position */
3840     shader_addline(buffer, "PARAM posFixup = program.local[%u];\n", next_local);
3841     compiled->pos_fixup = next_local++;
3842
3843     /* Initialize output parameters. GL_ARB_vertex_program does not require special initialization values
3844      * for output parameters. D3D in theory does not do that either, but some applications depend on a
3845      * proper initialization of the secondary color, and programs using the fixed function pipeline without
3846      * a replacement shader depend on the texcoord.w being set properly.
3847      *
3848      * GL_NV_vertex_program defines that all output values are initialized to {0.0, 0.0, 0.0, 1.0}. This
3849      * assertion is in effect even when using GL_ARB_vertex_program without any NV specific additions. So
3850      * skip this if NV_vertex_program is supported. Otherwise, initialize the secondary color. For the tex-
3851      * coords, we have a flag in the opengl caps. Many cards do not require the texcoord being set, and
3852      * this can eat a number of instructions, so skip it unless this cap is set as well
3853      */
3854     if(!GL_SUPPORT(NV_VERTEX_PROGRAM)) {
3855         shader_addline(buffer, "MOV result.color.secondary, -helper_const.wwwy;\n");
3856
3857         if ((GLINFO_LOCATION).quirks & WINED3D_QUIRK_SET_TEXCOORD_W && !device->frag_pipe->ffp_proj_control)
3858         {
3859             int i;
3860             for(i = 0; i < min(8, MAX_REG_TEXCRD); i++) {
3861                 if(This->baseShader.reg_maps.texcoord_mask[i] != 0 &&
3862                 This->baseShader.reg_maps.texcoord_mask[i] != WINED3DSP_WRITEMASK_ALL) {
3863                     shader_addline(buffer, "MOV result.texcoord[%u].w, -helper_const.y;\n", i);
3864                 }
3865             }
3866         }
3867     }
3868
3869     /* The shader starts with the main function */
3870     priv_ctx.in_main_func = TRUE;
3871     /* Base Shader Body */
3872     shader_generate_main((IWineD3DBaseShader *)This, buffer, reg_maps, function, &priv_ctx);
3873
3874     if(!priv_ctx.footer_written) vshader_add_footer(This, buffer, args, &priv_ctx);
3875
3876     shader_addline(buffer, "END\n");
3877
3878     /* TODO: change to resource.glObjectHandle or something like that */
3879     GL_EXTCALL(glGenProgramsARB(1, &ret));
3880
3881     TRACE("Creating a hw vertex shader, prg=%d\n", ret);
3882     GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, ret));
3883
3884     TRACE("Created hw vertex shader, prg=%d\n", ret);
3885     /* Create the program and check for errors */
3886     GL_EXTCALL(glProgramStringARB(GL_VERTEX_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
3887                buffer->bsize, buffer->buffer));
3888
3889     if (glGetError() == GL_INVALID_OPERATION) {
3890         GLint errPos;
3891         glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &errPos);
3892         FIXME("HW VertexShader Error at position %d: %s\n",
3893               errPos, debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
3894         ret = -1;
3895     } else {
3896         /* Load immediate constants */
3897         if(lconst_map) {
3898             LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry) {
3899                 const float *value = (const float *)lconst->value;
3900                 GL_EXTCALL(glProgramLocalParameter4fvARB(GL_VERTEX_PROGRAM_ARB, lconst_map[lconst->idx], value));
3901             }
3902         }
3903     }
3904     HeapFree(GetProcessHeap(), 0, lconst_map);
3905
3906     return ret;
3907 }
3908
3909 static void find_clip_texcoord(IWineD3DPixelShaderImpl *ps)
3910 {
3911     struct arb_pshader_private *shader_priv = ps->backend_priv;
3912     int i;
3913     const WineD3D_GL_Info *gl_info = &((IWineD3DDeviceImpl *)ps->baseShader.device)->adapter->gl_info;
3914
3915     /* See if we can use fragment.texcoord[7] for clipplane emulation
3916      *
3917      * Don't do this if it is not supported, or fragment.texcoord[7] is used
3918      */
3919     if(ps->baseShader.reg_maps.shader_version.major < 3)
3920     {
3921         for(i = GL_LIMITS(texture_stages); i > 0; i--)
3922         {
3923             if(!ps->baseShader.reg_maps.texcoord[i - 1])
3924             {
3925                 shader_priv->clipplane_emulation = i;
3926                 return;
3927             }
3928         }
3929         WARN("Did not find a free clip reg(2.0)\n");
3930     }
3931     else
3932     {
3933         for(i = GL_LIMITS(texture_stages); i > 0; i--)
3934         {
3935             if(!(ps->baseShader.reg_maps.input_registers & (1 << (i - 1))))
3936             {
3937                 shader_priv->clipplane_emulation = i;
3938                 return;
3939             }
3940         }
3941         WARN("Did not find a free clip reg(3.0)\n");
3942     }
3943 }
3944
3945 /* GL locking is done by the caller */
3946 static struct arb_ps_compiled_shader *find_arb_pshader(IWineD3DPixelShaderImpl *shader, const struct arb_ps_compile_args *args)
3947 {
3948     UINT i;
3949     DWORD new_size;
3950     struct arb_ps_compiled_shader *new_array;
3951     SHADER_BUFFER buffer;
3952     struct arb_pshader_private *shader_data;
3953     GLuint ret;
3954
3955     if(!shader->backend_priv) {
3956         IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *) shader->baseShader.device;
3957         struct shader_arb_priv *priv = device->shader_priv;
3958
3959         shader->backend_priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data));
3960         shader_data = shader->backend_priv;
3961         shader_data->clamp_consts = shader->baseShader.reg_maps.shader_version.major == 1;
3962
3963         if(shader->baseShader.reg_maps.shader_version.major < 3) shader_data->input_signature_idx = ~0;
3964         else shader_data->input_signature_idx = find_input_signature(priv, shader->input_signature);
3965
3966         shader_data->has_signature_idx = TRUE;
3967         TRACE("Shader got assigned input signature index %u\n", shader_data->input_signature_idx);
3968
3969         if(!device->vs_clipping) find_clip_texcoord(shader);
3970     }
3971     shader_data = shader->backend_priv;
3972
3973     /* Usually we have very few GL shaders for each d3d shader(just 1 or maybe 2),
3974      * so a linear search is more performant than a hashmap or a binary search
3975      * (cache coherency etc)
3976      */
3977     for(i = 0; i < shader_data->num_gl_shaders; i++) {
3978         if(memcmp(&shader_data->gl_shaders[i].args, args, sizeof(*args)) == 0) {
3979             return &shader_data->gl_shaders[i];
3980         }
3981     }
3982
3983     TRACE("No matching GL shader found, compiling a new shader\n");
3984     if(shader_data->shader_array_size == shader_data->num_gl_shaders) {
3985         if (shader_data->num_gl_shaders)
3986         {
3987             new_size = shader_data->shader_array_size + max(1, shader_data->shader_array_size / 2);
3988             new_array = HeapReAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, shader_data->gl_shaders,
3989                                     new_size * sizeof(*shader_data->gl_shaders));
3990         } else {
3991             new_array = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data->gl_shaders));
3992             new_size = 1;
3993         }
3994
3995         if(!new_array) {
3996             ERR("Out of memory\n");
3997             return 0;
3998         }
3999         shader_data->gl_shaders = new_array;
4000         shader_data->shader_array_size = new_size;
4001     }
4002
4003     shader_data->gl_shaders[shader_data->num_gl_shaders].args = *args;
4004
4005     pixelshader_update_samplers(&shader->baseShader.reg_maps,
4006             ((IWineD3DDeviceImpl *)shader->baseShader.device)->stateBlock->textures);
4007
4008     shader_buffer_init(&buffer);
4009     ret = shader_arb_generate_pshader(shader, &buffer, args,
4010                                       &shader_data->gl_shaders[shader_data->num_gl_shaders]);
4011     shader_buffer_free(&buffer);
4012     shader_data->gl_shaders[shader_data->num_gl_shaders].prgId = ret;
4013
4014     return &shader_data->gl_shaders[shader_data->num_gl_shaders++];
4015 }
4016
4017 static inline BOOL vs_args_equal(const struct arb_vs_compile_args *stored, const struct arb_vs_compile_args *new,
4018                                  const DWORD use_map, BOOL skip_int) {
4019     if((stored->super.swizzle_map & use_map) != new->super.swizzle_map) return FALSE;
4020     if(stored->super.fog_src != new->super.fog_src) return FALSE;
4021     if(stored->boolclip_compare != new->boolclip_compare) return FALSE;
4022     if(stored->ps_signature != new->ps_signature) return FALSE;
4023     if(stored->vertex_samplers_compare != new->vertex_samplers_compare) return FALSE;
4024     if(skip_int) return TRUE;
4025
4026     return memcmp(stored->loop_ctrl, new->loop_ctrl, sizeof(stored->loop_ctrl)) == 0;
4027 }
4028
4029 static struct arb_vs_compiled_shader *find_arb_vshader(IWineD3DVertexShaderImpl *shader, const struct arb_vs_compile_args *args)
4030 {
4031     UINT i;
4032     DWORD new_size;
4033     struct arb_vs_compiled_shader *new_array;
4034     DWORD use_map = ((IWineD3DDeviceImpl *)shader->baseShader.device)->strided_streams.use_map;
4035     SHADER_BUFFER buffer;
4036     struct arb_vshader_private *shader_data;
4037     GLuint ret;
4038     const WineD3D_GL_Info *gl_info = &((IWineD3DDeviceImpl *)shader->baseShader.device)->adapter->gl_info;
4039
4040     if(!shader->backend_priv) {
4041         shader->backend_priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data));
4042     }
4043     shader_data = shader->backend_priv;
4044
4045     /* Usually we have very few GL shaders for each d3d shader(just 1 or maybe 2),
4046      * so a linear search is more performant than a hashmap or a binary search
4047      * (cache coherency etc)
4048      */
4049     for(i = 0; i < shader_data->num_gl_shaders; i++) {
4050         if(vs_args_equal(&shader_data->gl_shaders[i].args, args, use_map, GL_SUPPORT(NV_VERTEX_PROGRAM2_OPTION))) {
4051             return &shader_data->gl_shaders[i];
4052         }
4053     }
4054
4055     TRACE("No matching GL shader found, compiling a new shader\n");
4056
4057     if(shader_data->shader_array_size == shader_data->num_gl_shaders) {
4058         if (shader_data->num_gl_shaders)
4059         {
4060             new_size = shader_data->shader_array_size + max(1, shader_data->shader_array_size / 2);
4061             new_array = HeapReAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, shader_data->gl_shaders,
4062                                     new_size * sizeof(*shader_data->gl_shaders));
4063         } else {
4064             new_array = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data->gl_shaders));
4065             new_size = 1;
4066         }
4067
4068         if(!new_array) {
4069             ERR("Out of memory\n");
4070             return 0;
4071         }
4072         shader_data->gl_shaders = new_array;
4073         shader_data->shader_array_size = new_size;
4074     }
4075
4076     shader_data->gl_shaders[shader_data->num_gl_shaders].args = *args;
4077
4078     shader_buffer_init(&buffer);
4079     ret = shader_arb_generate_vshader(shader, &buffer, args,
4080             &shader_data->gl_shaders[shader_data->num_gl_shaders]);
4081     shader_buffer_free(&buffer);
4082     shader_data->gl_shaders[shader_data->num_gl_shaders].prgId = ret;
4083
4084     return &shader_data->gl_shaders[shader_data->num_gl_shaders++];
4085 }
4086
4087 static inline void find_arb_ps_compile_args(IWineD3DPixelShaderImpl *shader, IWineD3DStateBlockImpl *stateblock,
4088         struct arb_ps_compile_args *args)
4089 {
4090     int i;
4091     WORD int_skip;
4092     const WineD3D_GL_Info *gl_info = &((IWineD3DDeviceImpl *)shader->baseShader.device)->adapter->gl_info;
4093     find_ps_compile_args(shader, stateblock, &args->super);
4094
4095     /* This forces all local boolean constants to 1 to make them stateblock independent */
4096     args->bools = shader->baseShader.reg_maps.local_bool_consts;
4097
4098     for(i = 0; i < MAX_CONST_B; i++)
4099     {
4100         if(stateblock->pixelShaderConstantB[i]) args->bools |= ( 1 << i);
4101     }
4102
4103     /* Skip if unused or local, or supported natively */
4104     int_skip = ~shader->baseShader.reg_maps.integer_constants | shader->baseShader.reg_maps.local_int_consts;
4105     if(int_skip == 0xffff || GL_SUPPORT(NV_FRAGMENT_PROGRAM_OPTION))
4106     {
4107         memset(&args->loop_ctrl, 0, sizeof(args->loop_ctrl));
4108         return;
4109     }
4110
4111     for(i = 0; i < MAX_CONST_I; i++)
4112     {
4113         if(int_skip & (1 << i))
4114         {
4115             args->loop_ctrl[i][0] = 0;
4116             args->loop_ctrl[i][1] = 0;
4117             args->loop_ctrl[i][2] = 0;
4118         }
4119         else
4120         {
4121             args->loop_ctrl[i][0] = stateblock->pixelShaderConstantI[i * 4];
4122             args->loop_ctrl[i][1] = stateblock->pixelShaderConstantI[i * 4 + 1];
4123             args->loop_ctrl[i][2] = stateblock->pixelShaderConstantI[i * 4 + 2];
4124         }
4125     }
4126 }
4127
4128 static inline void find_arb_vs_compile_args(IWineD3DVertexShaderImpl *shader, IWineD3DStateBlockImpl *stateblock,
4129         struct arb_vs_compile_args *args)
4130 {
4131     int i;
4132     WORD int_skip;
4133     IWineD3DDeviceImpl *dev = (IWineD3DDeviceImpl *)shader->baseShader.device;
4134     const WineD3D_GL_Info *gl_info = &dev->adapter->gl_info;
4135     find_vs_compile_args(shader, stateblock, &args->super);
4136
4137     args->boolclip_compare = 0;
4138     if(use_ps(stateblock))
4139     {
4140         IWineD3DPixelShaderImpl *ps = (IWineD3DPixelShaderImpl *) stateblock->pixelShader;
4141         struct arb_pshader_private *shader_priv = ps->backend_priv;
4142         args->ps_signature = shader_priv->input_signature_idx;
4143
4144         args->boolclip.clip_control[0] = shader_priv->clipplane_emulation;
4145     }
4146     else
4147     {
4148         args->ps_signature = ~0;
4149         if(!dev->vs_clipping)
4150         {
4151             args->boolclip.clip_control[0] = ffp_clip_emul(stateblock) ? GL_LIMITS(texture_stages) : 0;
4152         }
4153         /* Otherwise: Setting boolclip_compare set clip_control[0] to 0 */
4154     }
4155
4156     if(args->boolclip.clip_control[0])
4157     {
4158         if(stateblock->renderState[WINED3DRS_CLIPPING])
4159         {
4160             args->boolclip.clip_control[1] = stateblock->renderState[WINED3DRS_CLIPPLANEENABLE];
4161         }
4162         /* clip_control[1] was set to 0 by setting boolclip_compare to 0 */
4163     }
4164
4165     /* This forces all local boolean constants to 1 to make them stateblock independent */
4166     args->boolclip.bools = shader->baseShader.reg_maps.local_bool_consts;
4167     /* TODO: Figure out if it would be better to store bool constants as bitmasks in the stateblock */
4168     for(i = 0; i < MAX_CONST_B; i++)
4169     {
4170         if(stateblock->vertexShaderConstantB[i]) args->boolclip.bools |= ( 1 << i);
4171     }
4172
4173     args->vertex_samplers[0] = dev->texUnitMap[MAX_FRAGMENT_SAMPLERS + 0];
4174     args->vertex_samplers[1] = dev->texUnitMap[MAX_FRAGMENT_SAMPLERS + 1];
4175     args->vertex_samplers[2] = dev->texUnitMap[MAX_FRAGMENT_SAMPLERS + 2];
4176     args->vertex_samplers[3] = 0;
4177
4178     /* Skip if unused or local */
4179     int_skip = ~shader->baseShader.reg_maps.integer_constants | shader->baseShader.reg_maps.local_int_consts;
4180     if(int_skip == 0xffff || GL_SUPPORT(NV_VERTEX_PROGRAM2_OPTION)) /* This is about flow control, not clipping */
4181     {
4182         memset(&args->loop_ctrl, 0, sizeof(args->loop_ctrl));
4183         return;
4184     }
4185
4186     for(i = 0; i < MAX_CONST_I; i++)
4187     {
4188         if(int_skip & (1 << i))
4189         {
4190             args->loop_ctrl[i][0] = 0;
4191             args->loop_ctrl[i][1] = 0;
4192             args->loop_ctrl[i][2] = 0;
4193         }
4194         else
4195         {
4196             args->loop_ctrl[i][0] = stateblock->vertexShaderConstantI[i * 4];
4197             args->loop_ctrl[i][1] = stateblock->vertexShaderConstantI[i * 4 + 1];
4198             args->loop_ctrl[i][2] = stateblock->vertexShaderConstantI[i * 4 + 2];
4199         }
4200     }
4201 }
4202
4203 /* GL locking is done by the caller */
4204 static void shader_arb_select(IWineD3DDevice *iface, BOOL usePS, BOOL useVS) {
4205     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
4206     struct shader_arb_priv *priv = This->shader_priv;
4207     const WineD3D_GL_Info *gl_info = &This->adapter->gl_info;
4208     int i;
4209
4210     /* Deal with pixel shaders first so the vertex shader arg function has the input signature ready */
4211     if (usePS) {
4212         struct arb_ps_compile_args compile_args;
4213         struct arb_ps_compiled_shader *compiled;
4214         IWineD3DPixelShaderImpl *ps = (IWineD3DPixelShaderImpl *) This->stateBlock->pixelShader;
4215
4216         TRACE("Using pixel shader %p\n", This->stateBlock->pixelShader);
4217         find_arb_ps_compile_args(ps, This->stateBlock, &compile_args);
4218         compiled = find_arb_pshader(ps, &compile_args);
4219         priv->current_fprogram_id = compiled->prgId;
4220         priv->compiled_fprog = compiled;
4221
4222         /* Bind the fragment program */
4223         GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, priv->current_fprogram_id));
4224         checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, priv->current_fprogram_id);");
4225
4226         if(!priv->use_arbfp_fixed_func) {
4227             /* Enable OpenGL fragment programs */
4228             glEnable(GL_FRAGMENT_PROGRAM_ARB);
4229             checkGLcall("glEnable(GL_FRAGMENT_PROGRAM_ARB);");
4230         }
4231         TRACE("(%p) : Bound fragment program %u and enabled GL_FRAGMENT_PROGRAM_ARB\n", This, priv->current_fprogram_id);
4232
4233         /* Pixel Shader 1.x constants are clamped to [-1;1], Pixel Shader 2.0 constants are not. If switching between
4234          * a 1.x and newer shader, reload the first 8 constants
4235          */
4236         if(priv->last_ps_const_clamped != ((struct arb_pshader_private *) ps->backend_priv)->clamp_consts)
4237         {
4238             priv->last_ps_const_clamped = ((struct arb_pshader_private *) ps->backend_priv)->clamp_consts;
4239             This->highest_dirty_ps_const = max(This->highest_dirty_ps_const, 8);
4240             for(i = 0; i < 8; i++)
4241             {
4242                 This->activeContext->pshader_const_dirty[i] = 1;
4243             }
4244             /* Also takes care of loading local constants */
4245             shader_arb_load_constants(iface, TRUE, FALSE);
4246         }
4247         else
4248         {
4249             shader_arb_ps_local_constants(This);
4250         }
4251
4252         /* Force constant reloading for the NP2 fixup (see comment in shader_glsl_select for more info) */
4253         if (compiled->np2fixup_info.super.active) This->shader_backend->shader_load_np2fixup_constants(iface, usePS, useVS);
4254     } else if(GL_SUPPORT(ARB_FRAGMENT_PROGRAM) && !priv->use_arbfp_fixed_func) {
4255         /* Disable only if we're not using arbfp fixed function fragment processing. If this is used,
4256         * keep GL_FRAGMENT_PROGRAM_ARB enabled, and the fixed function pipeline will bind the fixed function
4257         * replacement shader
4258         */
4259         glDisable(GL_FRAGMENT_PROGRAM_ARB);
4260         checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
4261         priv->current_fprogram_id = 0;
4262     }
4263
4264     if (useVS) {
4265         struct arb_vs_compile_args compile_args;
4266         struct arb_vs_compiled_shader *compiled;
4267         IWineD3DVertexShaderImpl *vs = (IWineD3DVertexShaderImpl *) This->stateBlock->vertexShader;
4268
4269         TRACE("Using vertex shader %p\n", This->stateBlock->vertexShader);
4270         find_arb_vs_compile_args(vs, This->stateBlock, &compile_args);
4271         compiled = find_arb_vshader(vs, &compile_args);
4272         priv->current_vprogram_id = compiled->prgId;
4273         priv->compiled_vprog = compiled;
4274
4275         /* Bind the vertex program */
4276         GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->current_vprogram_id));
4277         checkGLcall("glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->current_vprogram_id);");
4278
4279         /* Enable OpenGL vertex programs */
4280         glEnable(GL_VERTEX_PROGRAM_ARB);
4281         checkGLcall("glEnable(GL_VERTEX_PROGRAM_ARB);");
4282         TRACE("(%p) : Bound vertex program %u and enabled GL_VERTEX_PROGRAM_ARB\n", This, priv->current_vprogram_id);
4283         shader_arb_vs_local_constants(This);
4284
4285         if(priv->last_vs_color_unclamp != compiled->need_color_unclamp) {
4286             priv->last_vs_color_unclamp = compiled->need_color_unclamp;
4287
4288             if (GL_SUPPORT(ARB_COLOR_BUFFER_FLOAT)) {
4289                 GL_EXTCALL(glClampColorARB(GL_CLAMP_VERTEX_COLOR_ARB, !compiled->need_color_unclamp));
4290                 checkGLcall("glClampColorARB");
4291             } else {
4292                 FIXME("vertex color clamp needs to be changed, but extension not supported.\n");
4293             }
4294         }
4295     } else if(GL_SUPPORT(ARB_VERTEX_PROGRAM)) {
4296         priv->current_vprogram_id = 0;
4297         glDisable(GL_VERTEX_PROGRAM_ARB);
4298         checkGLcall("glDisable(GL_VERTEX_PROGRAM_ARB)");
4299     }
4300 }
4301
4302 /* GL locking is done by the caller */
4303 static void shader_arb_select_depth_blt(IWineD3DDevice *iface, enum tex_types tex_type) {
4304     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
4305     struct shader_arb_priv *priv = This->shader_priv;
4306     GLuint *blt_fprogram = &priv->depth_blt_fprogram_id[tex_type];
4307     const WineD3D_GL_Info *gl_info = &This->adapter->gl_info;
4308
4309     if (!priv->depth_blt_vprogram_id) priv->depth_blt_vprogram_id = create_arb_blt_vertex_program(gl_info);
4310     GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->depth_blt_vprogram_id));
4311     glEnable(GL_VERTEX_PROGRAM_ARB);
4312
4313     if (!*blt_fprogram) *blt_fprogram = create_arb_blt_fragment_program(gl_info, tex_type);
4314     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, *blt_fprogram));
4315     glEnable(GL_FRAGMENT_PROGRAM_ARB);
4316 }
4317
4318 /* GL locking is done by the caller */
4319 static void shader_arb_deselect_depth_blt(IWineD3DDevice *iface) {
4320     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
4321     struct shader_arb_priv *priv = This->shader_priv;
4322     const WineD3D_GL_Info *gl_info = &This->adapter->gl_info;
4323
4324     if (priv->current_vprogram_id) {
4325         GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->current_vprogram_id));
4326         checkGLcall("glBindProgramARB(GL_VERTEX_PROGRAM_ARB, vertexShader->prgId);");
4327
4328         TRACE("(%p) : Bound vertex program %u and enabled GL_VERTEX_PROGRAM_ARB\n", This, priv->current_vprogram_id);
4329     } else {
4330         glDisable(GL_VERTEX_PROGRAM_ARB);
4331         checkGLcall("glDisable(GL_VERTEX_PROGRAM_ARB)");
4332     }
4333
4334     if (priv->current_fprogram_id) {
4335         GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, priv->current_fprogram_id));
4336         checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, pixelShader->prgId);");
4337
4338         TRACE("(%p) : Bound fragment program %u and enabled GL_FRAGMENT_PROGRAM_ARB\n", This, priv->current_fprogram_id);
4339     } else if(!priv->use_arbfp_fixed_func) {
4340         glDisable(GL_FRAGMENT_PROGRAM_ARB);
4341         checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
4342     }
4343 }
4344
4345 static void shader_arb_destroy(IWineD3DBaseShader *iface) {
4346     IWineD3DBaseShaderImpl *baseShader = (IWineD3DBaseShaderImpl *) iface;
4347     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *)baseShader->baseShader.device;
4348     const WineD3D_GL_Info *gl_info = &device->adapter->gl_info;
4349
4350     ActivateContext(device, device->lastActiveRenderTarget, CTXUSAGE_RESOURCELOAD);
4351
4352     if (shader_is_pshader_version(baseShader->baseShader.reg_maps.shader_version.type))
4353     {
4354         IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *) iface;
4355         struct arb_pshader_private *shader_data = This->backend_priv;
4356         UINT i;
4357
4358         if(!shader_data) return; /* This can happen if a shader was never compiled */
4359         ENTER_GL();
4360         for(i = 0; i < shader_data->num_gl_shaders; i++) {
4361             GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId));
4362             checkGLcall("GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId))");
4363         }
4364         LEAVE_GL();
4365         HeapFree(GetProcessHeap(), 0, shader_data->gl_shaders);
4366         HeapFree(GetProcessHeap(), 0, shader_data);
4367         This->backend_priv = NULL;
4368     } else {
4369         IWineD3DVertexShaderImpl *This = (IWineD3DVertexShaderImpl *) iface;
4370         struct arb_vshader_private *shader_data = This->backend_priv;
4371         UINT i;
4372
4373         if(!shader_data) return; /* This can happen if a shader was never compiled */
4374         ENTER_GL();
4375         for(i = 0; i < shader_data->num_gl_shaders; i++) {
4376             GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId));
4377             checkGLcall("GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId))");
4378         }
4379         LEAVE_GL();
4380         HeapFree(GetProcessHeap(), 0, shader_data->gl_shaders);
4381         HeapFree(GetProcessHeap(), 0, shader_data);
4382         This->backend_priv = NULL;
4383     }
4384 }
4385
4386 static int sig_tree_compare(const void *key, const struct wine_rb_entry *entry)
4387 {
4388     struct ps_signature *e = WINE_RB_ENTRY_VALUE(entry, struct ps_signature, entry);
4389     return compare_sig(key, e->sig);
4390 }
4391
4392 struct wine_rb_functions sig_tree_functions =
4393 {
4394     wined3d_rb_alloc,
4395     wined3d_rb_realloc,
4396     wined3d_rb_free,
4397     sig_tree_compare
4398 };
4399
4400 static HRESULT shader_arb_alloc(IWineD3DDevice *iface) {
4401     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
4402     struct shader_arb_priv *priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*priv));
4403     if(wine_rb_init(&priv->signature_tree, &sig_tree_functions) == -1)
4404     {
4405         ERR("RB tree init failed\n");
4406         HeapFree(GetProcessHeap(), 0, priv);
4407         return E_OUTOFMEMORY;
4408     }
4409     This->shader_priv = priv;
4410     return WINED3D_OK;
4411 }
4412
4413 static void release_signature(struct wine_rb_entry *entry, void *context)
4414 {
4415     struct ps_signature *sig = WINE_RB_ENTRY_VALUE(entry, struct ps_signature, entry);
4416     int i;
4417     for(i = 0; i < MAX_REG_INPUT; i++)
4418     {
4419         HeapFree(GetProcessHeap(), 0, (char *) sig->sig[i].semantic_name);
4420     }
4421     HeapFree(GetProcessHeap(), 0, sig->sig);
4422     HeapFree(GetProcessHeap(), 0, sig);
4423 }
4424
4425 /* Context activation is done by the caller. */
4426 static void shader_arb_free(IWineD3DDevice *iface) {
4427     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
4428     const WineD3D_GL_Info *gl_info = &This->adapter->gl_info;
4429     struct shader_arb_priv *priv = This->shader_priv;
4430     int i;
4431
4432     ENTER_GL();
4433     if(priv->depth_blt_vprogram_id) {
4434         GL_EXTCALL(glDeleteProgramsARB(1, &priv->depth_blt_vprogram_id));
4435     }
4436     for (i = 0; i < tex_type_count; ++i) {
4437         if (priv->depth_blt_fprogram_id[i]) {
4438             GL_EXTCALL(glDeleteProgramsARB(1, &priv->depth_blt_fprogram_id[i]));
4439         }
4440     }
4441     LEAVE_GL();
4442
4443     wine_rb_destroy(&priv->signature_tree, release_signature, NULL);
4444     HeapFree(GetProcessHeap(), 0, This->shader_priv);
4445 }
4446
4447 static BOOL shader_arb_dirty_const(IWineD3DDevice *iface) {
4448     return TRUE;
4449 }
4450
4451 static void shader_arb_get_caps(WINED3DDEVTYPE devtype, const WineD3D_GL_Info *gl_info, struct shader_caps *pCaps)
4452 {
4453     /* We don't have an ARB fixed function pipeline yet, so let the none backend set its caps,
4454      * then overwrite the shader specific ones
4455      */
4456     none_shader_backend.shader_get_caps(devtype, gl_info, pCaps);
4457
4458     if(GL_SUPPORT(ARB_VERTEX_PROGRAM)) {
4459         if(GL_SUPPORT(NV_VERTEX_PROGRAM3))
4460         {
4461             pCaps->VertexShaderVersion = WINED3DVS_VERSION(3,0);
4462             TRACE_(d3d_caps)("Hardware vertex shader version 3.0 enabled (NV_VERTEX_PROGRAM3)\n");
4463         }
4464         else if(GL_LIMITS(vshader_constantsF) >= 256)
4465         {
4466             /* Shader Model 2.0 requires at least 256 vertex shader constants */
4467             pCaps->VertexShaderVersion = WINED3DVS_VERSION(2,0);
4468             TRACE_(d3d_caps)("Hardware vertex shader version 2.0 enabled (ARB_PROGRAM)\n");
4469         }
4470         else
4471         {
4472             pCaps->VertexShaderVersion = WINED3DVS_VERSION(1,1);
4473             TRACE_(d3d_caps)("Hardware vertex shader version 1.1 enabled (ARB_PROGRAM)\n");
4474         }
4475         pCaps->MaxVertexShaderConst = GL_LIMITS(vshader_constantsF);
4476     }
4477
4478     if(GL_SUPPORT(ARB_FRAGMENT_PROGRAM)) {
4479         if(GL_SUPPORT(NV_FRAGMENT_PROGRAM2))
4480         {
4481             pCaps->PixelShaderVersion    = WINED3DPS_VERSION(3,0);
4482             TRACE_(d3d_caps)("Hardware pixel shader version 3.0 enabled (NV_FRAGMENT_PROGRAM2)\n");
4483         }
4484         else if(GL_LIMITS(vshader_constantsF) >= 32)
4485         {
4486             /* Shader Model 2.0 requires at least 32 pixel shader constants */
4487             pCaps->PixelShaderVersion    = WINED3DPS_VERSION(2,0);
4488             TRACE_(d3d_caps)("Hardware pixel shader version 2.0 enabled (ARB_PROGRAM)\n");
4489         }
4490         else
4491         {
4492             pCaps->PixelShaderVersion    = WINED3DPS_VERSION(1,4);
4493             TRACE_(d3d_caps)("Hardware pixel shader version 1.4 enabled (ARB_PROGRAM)\n");
4494         }
4495         pCaps->PixelShader1xMaxValue = 8.0f;
4496         pCaps->MaxPixelShaderConst = GL_LIMITS(pshader_constantsF);
4497     }
4498
4499     pCaps->VSClipping = use_nv_clip(gl_info);
4500 }
4501
4502 static BOOL shader_arb_color_fixup_supported(struct color_fixup_desc fixup)
4503 {
4504     if (TRACE_ON(d3d_shader) && TRACE_ON(d3d))
4505     {
4506         TRACE("Checking support for color_fixup:\n");
4507         dump_color_fixup_desc(fixup);
4508     }
4509
4510     /* We support everything except YUV conversions. */
4511     if (!is_yuv_fixup(fixup))
4512     {
4513         TRACE("[OK]\n");
4514         return TRUE;
4515     }
4516
4517     TRACE("[FAILED]\n");
4518     return FALSE;
4519 }
4520
4521 static void shader_arb_add_instruction_modifiers(const struct wined3d_shader_instruction *ins) {
4522     DWORD shift;
4523     char write_mask[20], regstr[50];
4524     SHADER_BUFFER *buffer = ins->ctx->buffer;
4525     BOOL is_color = FALSE;
4526     const struct wined3d_shader_dst_param *dst;
4527
4528     if (!ins->dst_count) return;
4529
4530     dst = &ins->dst[0];
4531     shift = dst->shift;
4532     if(shift == 0) return; /* Saturate alone is handled by the instructions */
4533
4534     shader_arb_get_write_mask(ins, dst, write_mask);
4535     shader_arb_get_register_name(ins, &dst->reg, regstr, &is_color);
4536
4537     /* Generate a line that does the output modifier computation
4538      * FIXME: _SAT vs shift? _SAT alone is already handled in the instructions, if this
4539      * maps problems in e.g. _d4_sat modify shader_arb_get_modifier
4540      */
4541     shader_addline(buffer, "MUL%s %s%s, %s, %s;\n", shader_arb_get_modifier(ins),
4542                    regstr, write_mask, regstr, shift_tab[shift]);
4543 }
4544
4545 static const SHADER_HANDLER shader_arb_instruction_handler_table[WINED3DSIH_TABLE_SIZE] =
4546 {
4547     /* WINED3DSIH_ABS           */ shader_hw_map2gl,
4548     /* WINED3DSIH_ADD           */ shader_hw_map2gl,
4549     /* WINED3DSIH_BEM           */ pshader_hw_bem,
4550     /* WINED3DSIH_BREAK         */ shader_hw_break,
4551     /* WINED3DSIH_BREAKC        */ shader_hw_breakc,
4552     /* WINED3DSIH_BREAKP        */ NULL,
4553     /* WINED3DSIH_CALL          */ shader_hw_call,
4554     /* WINED3DSIH_CALLNZ        */ NULL,
4555     /* WINED3DSIH_CMP           */ pshader_hw_cmp,
4556     /* WINED3DSIH_CND           */ pshader_hw_cnd,
4557     /* WINED3DSIH_CRS           */ shader_hw_map2gl,
4558     /* WINED3DSIH_DCL           */ NULL,
4559     /* WINED3DSIH_DEF           */ NULL,
4560     /* WINED3DSIH_DEFB          */ NULL,
4561     /* WINED3DSIH_DEFI          */ NULL,
4562     /* WINED3DSIH_DP2ADD        */ pshader_hw_dp2add,
4563     /* WINED3DSIH_DP3           */ shader_hw_map2gl,
4564     /* WINED3DSIH_DP4           */ shader_hw_map2gl,
4565     /* WINED3DSIH_DST           */ shader_hw_map2gl,
4566     /* WINED3DSIH_DSX           */ shader_hw_map2gl,
4567     /* WINED3DSIH_DSY           */ shader_hw_dsy,
4568     /* WINED3DSIH_ELSE          */ shader_hw_else,
4569     /* WINED3DSIH_ENDIF         */ shader_hw_endif,
4570     /* WINED3DSIH_ENDLOOP       */ shader_hw_endloop,
4571     /* WINED3DSIH_ENDREP        */ shader_hw_endrep,
4572     /* WINED3DSIH_EXP           */ shader_hw_map2gl,
4573     /* WINED3DSIH_EXPP          */ shader_hw_map2gl,
4574     /* WINED3DSIH_FRC           */ shader_hw_map2gl,
4575     /* WINED3DSIH_IF            */ NULL /* Hardcoded into the shader */,
4576     /* WINED3DSIH_IFC           */ shader_hw_ifc,
4577     /* WINED3DSIH_LABEL         */ shader_hw_label,
4578     /* WINED3DSIH_LIT           */ shader_hw_map2gl,
4579     /* WINED3DSIH_LOG           */ shader_hw_log_pow,
4580     /* WINED3DSIH_LOGP          */ shader_hw_log_pow,
4581     /* WINED3DSIH_LOOP          */ shader_hw_loop,
4582     /* WINED3DSIH_LRP           */ shader_hw_lrp,
4583     /* WINED3DSIH_M3x2          */ shader_hw_mnxn,
4584     /* WINED3DSIH_M3x3          */ shader_hw_mnxn,
4585     /* WINED3DSIH_M3x4          */ shader_hw_mnxn,
4586     /* WINED3DSIH_M4x3          */ shader_hw_mnxn,
4587     /* WINED3DSIH_M4x4          */ shader_hw_mnxn,
4588     /* WINED3DSIH_MAD           */ shader_hw_map2gl,
4589     /* WINED3DSIH_MAX           */ shader_hw_map2gl,
4590     /* WINED3DSIH_MIN           */ shader_hw_map2gl,
4591     /* WINED3DSIH_MOV           */ shader_hw_mov,
4592     /* WINED3DSIH_MOVA          */ shader_hw_mov,
4593     /* WINED3DSIH_MUL           */ shader_hw_map2gl,
4594     /* WINED3DSIH_NOP           */ shader_hw_nop,
4595     /* WINED3DSIH_NRM           */ shader_hw_nrm,
4596     /* WINED3DSIH_PHASE         */ NULL,
4597     /* WINED3DSIH_POW           */ shader_hw_log_pow,
4598     /* WINED3DSIH_RCP           */ shader_hw_rsq_rcp,
4599     /* WINED3DSIH_REP           */ shader_hw_rep,
4600     /* WINED3DSIH_RET           */ shader_hw_ret,
4601     /* WINED3DSIH_RSQ           */ shader_hw_rsq_rcp,
4602     /* WINED3DSIH_SETP          */ NULL,
4603     /* WINED3DSIH_SGE           */ shader_hw_map2gl,
4604     /* WINED3DSIH_SGN           */ shader_hw_sgn,
4605     /* WINED3DSIH_SINCOS        */ shader_hw_sincos,
4606     /* WINED3DSIH_SLT           */ shader_hw_map2gl,
4607     /* WINED3DSIH_SUB           */ shader_hw_map2gl,
4608     /* WINED3DSIH_TEX           */ pshader_hw_tex,
4609     /* WINED3DSIH_TEXBEM        */ pshader_hw_texbem,
4610     /* WINED3DSIH_TEXBEML       */ pshader_hw_texbem,
4611     /* WINED3DSIH_TEXCOORD      */ pshader_hw_texcoord,
4612     /* WINED3DSIH_TEXDEPTH      */ pshader_hw_texdepth,
4613     /* WINED3DSIH_TEXDP3        */ pshader_hw_texdp3,
4614     /* WINED3DSIH_TEXDP3TEX     */ pshader_hw_texdp3tex,
4615     /* WINED3DSIH_TEXKILL       */ pshader_hw_texkill,
4616     /* WINED3DSIH_TEXLDD        */ shader_hw_texldd,
4617     /* WINED3DSIH_TEXLDL        */ shader_hw_texldl,
4618     /* WINED3DSIH_TEXM3x2DEPTH  */ pshader_hw_texm3x2depth,
4619     /* WINED3DSIH_TEXM3x2PAD    */ pshader_hw_texm3x2pad,
4620     /* WINED3DSIH_TEXM3x2TEX    */ pshader_hw_texm3x2tex,
4621     /* WINED3DSIH_TEXM3x3       */ pshader_hw_texm3x3,
4622     /* WINED3DSIH_TEXM3x3DIFF   */ NULL,
4623     /* WINED3DSIH_TEXM3x3PAD    */ pshader_hw_texm3x3pad,
4624     /* WINED3DSIH_TEXM3x3SPEC   */ pshader_hw_texm3x3spec,
4625     /* WINED3DSIH_TEXM3x3TEX    */ pshader_hw_texm3x3tex,
4626     /* WINED3DSIH_TEXM3x3VSPEC  */ pshader_hw_texm3x3vspec,
4627     /* WINED3DSIH_TEXREG2AR     */ pshader_hw_texreg2ar,
4628     /* WINED3DSIH_TEXREG2GB     */ pshader_hw_texreg2gb,
4629     /* WINED3DSIH_TEXREG2RGB    */ pshader_hw_texreg2rgb,
4630 };
4631
4632 static inline BOOL get_bool_const(const struct wined3d_shader_instruction *ins, IWineD3DBaseShaderImpl *This, DWORD idx)
4633 {
4634     BOOL vshader = shader_is_vshader_version(This->baseShader.reg_maps.shader_version.type);
4635     WORD bools = 0;
4636     WORD flag = (1 << idx);
4637     const local_constant *constant;
4638     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
4639
4640     if(This->baseShader.reg_maps.local_bool_consts & flag)
4641     {
4642         /* What good is a if(bool) with a hardcoded local constant? I don't know, but handle it */
4643         LIST_FOR_EACH_ENTRY(constant, &This->baseShader.constantsB, local_constant, entry)
4644         {
4645             if (constant->idx == idx)
4646             {
4647                 return constant->value[0];
4648             }
4649         }
4650         ERR("Local constant not found\n");
4651         return FALSE;
4652     }
4653     else
4654     {
4655         if(vshader) bools = priv->cur_vs_args->boolclip.bools;
4656         else bools = priv->cur_ps_args->bools;
4657         return bools & flag;
4658     }
4659 }
4660
4661 static void get_loop_control_const(const struct wined3d_shader_instruction *ins,
4662         IWineD3DBaseShaderImpl *This, UINT idx, struct loop_control *loop_control)
4663 {
4664     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
4665
4666     /* Integer constants can either be a local constant, or they can be stored in the shader
4667      * type specific compile args. */
4668     if (This->baseShader.reg_maps.local_int_consts & (1 << idx))
4669     {
4670         const local_constant *constant;
4671
4672         LIST_FOR_EACH_ENTRY(constant, &This->baseShader.constantsI, local_constant, entry)
4673         {
4674             if (constant->idx == idx)
4675             {
4676                 loop_control->count = constant->value[0];
4677                 loop_control->start = constant->value[1];
4678                 /* Step is signed. */
4679                 loop_control->step = (int)constant->value[2];
4680                 return;
4681             }
4682         }
4683         /* If this happens the flag was set incorrectly */
4684         ERR("Local constant not found\n");
4685         loop_control->count = 0;
4686         loop_control->start = 0;
4687         loop_control->step = 0;
4688         return;
4689     }
4690
4691     switch (This->baseShader.reg_maps.shader_version.type)
4692     {
4693         case WINED3D_SHADER_TYPE_VERTEX:
4694             /* Count and aL start value are unsigned */
4695             loop_control->count = priv->cur_vs_args->loop_ctrl[idx][0];
4696             loop_control->start = priv->cur_vs_args->loop_ctrl[idx][1];
4697             /* Step is signed. */
4698             loop_control->step = ((char)priv->cur_vs_args->loop_ctrl[idx][2]);
4699             break;
4700
4701         case WINED3D_SHADER_TYPE_PIXEL:
4702             loop_control->count = priv->cur_ps_args->loop_ctrl[idx][0];
4703             loop_control->start = priv->cur_ps_args->loop_ctrl[idx][1];
4704             loop_control->step = ((char)priv->cur_ps_args->loop_ctrl[idx][2]);
4705             break;
4706
4707         default:
4708             FIXME("Unhandled shader type %#x.\n", This->baseShader.reg_maps.shader_version.type);
4709             break;
4710     }
4711 }
4712
4713 static void record_instruction(struct list *list, const struct wined3d_shader_instruction *ins)
4714 {
4715     unsigned int i;
4716     struct wined3d_shader_dst_param *dst_param = NULL;
4717     struct wined3d_shader_src_param *src_param = NULL, *rel_addr = NULL;
4718     struct recorded_instruction *rec = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*rec));
4719     if(!rec)
4720     {
4721         ERR("Out of memory\n");
4722         return;
4723     }
4724
4725     rec->ins = *ins;
4726     dst_param = HeapAlloc(GetProcessHeap(), 0, sizeof(*dst_param));
4727     if(!dst_param) goto free;
4728     *dst_param = *ins->dst;
4729     if(ins->dst->reg.rel_addr)
4730     {
4731         rel_addr = HeapAlloc(GetProcessHeap(), 0, sizeof(*dst_param->reg.rel_addr));
4732         if(!rel_addr) goto free;
4733         *rel_addr = *ins->dst->reg.rel_addr;
4734         dst_param->reg.rel_addr = rel_addr;
4735     }
4736     rec->ins.dst = dst_param;
4737
4738     src_param = HeapAlloc(GetProcessHeap(), 0, sizeof(*src_param) * ins->src_count);
4739     if(!src_param) goto free;
4740     for(i = 0; i < ins->src_count; i++)
4741     {
4742         src_param[i] = ins->src[i];
4743         if(ins->src[i].reg.rel_addr)
4744         {
4745             rel_addr = HeapAlloc(GetProcessHeap(), 0, sizeof(*rel_addr));
4746             if(!rel_addr) goto free;
4747             *rel_addr = *ins->src[i].reg.rel_addr;
4748             src_param[i].reg.rel_addr = rel_addr;
4749         }
4750     }
4751     rec->ins.src = src_param;
4752     list_add_tail(list, &rec->entry);
4753     return;
4754
4755 free:
4756     ERR("Out of memory\n");
4757     if(dst_param)
4758     {
4759         HeapFree(GetProcessHeap(), 0, (void *) dst_param->reg.rel_addr);
4760         HeapFree(GetProcessHeap(), 0, dst_param);
4761     }
4762     if(src_param)
4763     {
4764         for(i = 0; i < ins->src_count; i++)
4765         {
4766             HeapFree(GetProcessHeap(), 0, (void *) src_param[i].reg.rel_addr);
4767         }
4768         HeapFree(GetProcessHeap(), 0, src_param);
4769     }
4770     HeapFree(GetProcessHeap(), 0, rec);
4771 }
4772
4773 static void free_recorded_instruction(struct list *list)
4774 {
4775     struct recorded_instruction *rec_ins, *entry2;
4776     unsigned int i;
4777
4778     LIST_FOR_EACH_ENTRY_SAFE(rec_ins, entry2, list, struct recorded_instruction, entry)
4779     {
4780         list_remove(&rec_ins->entry);
4781         if(rec_ins->ins.dst)
4782         {
4783             HeapFree(GetProcessHeap(), 0, (void *) rec_ins->ins.dst->reg.rel_addr);
4784             HeapFree(GetProcessHeap(), 0, (void *) rec_ins->ins.dst);
4785         }
4786         if(rec_ins->ins.src)
4787         {
4788             for(i = 0; i < rec_ins->ins.src_count; i++)
4789             {
4790                 HeapFree(GetProcessHeap(), 0, (void *) rec_ins->ins.src[i].reg.rel_addr);
4791             }
4792             HeapFree(GetProcessHeap(), 0, (void *) rec_ins->ins.src);
4793         }
4794         HeapFree(GetProcessHeap(), 0, rec_ins);
4795     }
4796 }
4797
4798 static void shader_arb_handle_instruction(const struct wined3d_shader_instruction *ins) {
4799     SHADER_HANDLER hw_fct;
4800     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
4801     IWineD3DBaseShaderImpl *This = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
4802     struct control_frame *control_frame;
4803     SHADER_BUFFER *buffer = ins->ctx->buffer;
4804     BOOL bool_const;
4805
4806     if(ins->handler_idx == WINED3DSIH_LOOP || ins->handler_idx == WINED3DSIH_REP)
4807     {
4808         control_frame = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*control_frame));
4809         list_add_head(&priv->control_frames, &control_frame->entry);
4810
4811         if(ins->handler_idx == WINED3DSIH_LOOP) control_frame->type = LOOP;
4812         if(ins->handler_idx == WINED3DSIH_REP) control_frame->type = REP;
4813
4814         if(priv->target_version >= NV2)
4815         {
4816             control_frame->loop_no = priv->num_loops++;
4817             priv->loop_depth++;
4818         }
4819         else
4820         {
4821             /* Don't bother recording when we're in a not used if branch */
4822             if(priv->muted)
4823             {
4824                 return;
4825             }
4826
4827             if(!priv->recording)
4828             {
4829                 list_init(&priv->record);
4830                 priv->recording = TRUE;
4831                 control_frame->outer_loop = TRUE;
4832                 get_loop_control_const(ins, This, ins->src[0].reg.idx, &control_frame->loop_control);
4833                 return; /* Instruction is handled */
4834             }
4835             /* Record this loop in the outer loop's recording */
4836         }
4837     }
4838     else if(ins->handler_idx == WINED3DSIH_ENDLOOP || ins->handler_idx == WINED3DSIH_ENDREP)
4839     {
4840         if(priv->target_version >= NV2)
4841         {
4842             /* Nothing to do. The control frame is popped after the HW instr handler */
4843         }
4844         else
4845         {
4846             struct list *e = list_head(&priv->control_frames);
4847             control_frame = LIST_ENTRY(e, struct control_frame, entry);
4848             list_remove(&control_frame->entry);
4849
4850             if(control_frame->outer_loop)
4851             {
4852                 int iteration, aL = 0;
4853                 struct list copy;
4854
4855                 /* Turn off recording before playback */
4856                 priv->recording = FALSE;
4857
4858                 /* Move the recorded instructions to a separate list and get them out of the private data
4859                  * structure. If there are nested loops, the shader_arb_handle_instruction below will
4860                  * be recorded again, thus priv->record might be overwritten
4861                  */
4862                 list_init(&copy);
4863                 list_move_tail(&copy, &priv->record);
4864                 list_init(&priv->record);
4865
4866                 if(ins->handler_idx == WINED3DSIH_ENDLOOP)
4867                 {
4868                     shader_addline(buffer, "#unrolling loop: %u iterations, aL=%u, inc %d\n",
4869                                    control_frame->loop_control.count, control_frame->loop_control.start,
4870                                    control_frame->loop_control.step);
4871                     aL = control_frame->loop_control.start;
4872                 }
4873                 else
4874                 {
4875                     shader_addline(buffer, "#unrolling rep: %u iterations\n", control_frame->loop_control.count);
4876                 }
4877
4878                 for (iteration = 0; iteration < control_frame->loop_control.count; ++iteration)
4879                 {
4880                     struct recorded_instruction *rec_ins;
4881                     if(ins->handler_idx == WINED3DSIH_ENDLOOP)
4882                     {
4883                         priv->aL = aL;
4884                         shader_addline(buffer, "#Iteration %d, aL=%d\n", iteration, aL);
4885                     }
4886                     else
4887                     {
4888                         shader_addline(buffer, "#Iteration %d\n", iteration);
4889                     }
4890
4891                     LIST_FOR_EACH_ENTRY(rec_ins, &copy, struct recorded_instruction, entry)
4892                     {
4893                         shader_arb_handle_instruction(&rec_ins->ins);
4894                     }
4895
4896                     if(ins->handler_idx == WINED3DSIH_ENDLOOP)
4897                     {
4898                         aL += control_frame->loop_control.step;
4899                     }
4900                 }
4901                 shader_addline(buffer, "#end loop/rep\n");
4902
4903                 free_recorded_instruction(&copy);
4904                 HeapFree(GetProcessHeap(), 0, control_frame);
4905                 return; /* Instruction is handled */
4906             }
4907             else
4908             {
4909                 /* This is a nested loop. Proceed to the normal recording function */
4910                 HeapFree(GetProcessHeap(), 0, control_frame);
4911             }
4912         }
4913     }
4914
4915     if(priv->recording)
4916     {
4917         record_instruction(&priv->record, ins);
4918         return;
4919     }
4920
4921     /* boolean if */
4922     if(ins->handler_idx == WINED3DSIH_IF)
4923     {
4924         control_frame = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*control_frame));
4925         list_add_head(&priv->control_frames, &control_frame->entry);
4926         control_frame->type = IF;
4927
4928         bool_const = get_bool_const(ins, This, ins->src[0].reg.idx);
4929         if(ins->src[0].modifiers == WINED3DSPSM_NOT) bool_const = !bool_const;
4930         if(!priv->muted && bool_const == FALSE)
4931         {
4932             shader_addline(buffer, "#if(FALSE){\n");
4933             priv->muted = TRUE;
4934             control_frame->muting = TRUE;
4935         }
4936         else shader_addline(buffer, "#if(TRUE) {\n");
4937
4938         return; /* Instruction is handled */
4939     }
4940     else if(ins->handler_idx == WINED3DSIH_IFC)
4941     {
4942         /* IF(bool) and if_cond(a, b) use the same ELSE and ENDIF tokens */
4943         control_frame = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*control_frame));
4944         control_frame->type = IFC;
4945         control_frame->ifc_no = priv->num_ifcs++;
4946         list_add_head(&priv->control_frames, &control_frame->entry);
4947     }
4948     else if(ins->handler_idx == WINED3DSIH_ELSE)
4949     {
4950         struct list *e = list_head(&priv->control_frames);
4951         control_frame = LIST_ENTRY(e, struct control_frame, entry);
4952
4953         if(control_frame->type == IF)
4954         {
4955             shader_addline(buffer, "#} else {\n");
4956             if(!priv->muted && !control_frame->muting)
4957             {
4958                 priv->muted = TRUE;
4959                 control_frame->muting = TRUE;
4960             }
4961             else if(control_frame->muting) priv->muted = FALSE;
4962             return; /* Instruction is handled. */
4963         }
4964         /* In case of an ifc, generate a HW shader instruction */
4965     }
4966     else if(ins->handler_idx == WINED3DSIH_ENDIF)
4967     {
4968         struct list *e = list_head(&priv->control_frames);
4969         control_frame = LIST_ENTRY(e, struct control_frame, entry);
4970
4971         if(control_frame->type == IF)
4972         {
4973             shader_addline(buffer, "#} endif\n");
4974             if(control_frame->muting) priv->muted = FALSE;
4975             list_remove(&control_frame->entry);
4976             HeapFree(GetProcessHeap(), 0, control_frame);
4977             return; /* Instruction is handled */
4978         }
4979     }
4980
4981     if(priv->muted) return;
4982
4983     /* Select handler */
4984     hw_fct = shader_arb_instruction_handler_table[ins->handler_idx];
4985
4986     /* Unhandled opcode */
4987     if (!hw_fct)
4988     {
4989         FIXME("Backend can't handle opcode %#x\n", ins->handler_idx);
4990         return;
4991     }
4992     hw_fct(ins);
4993
4994     if(ins->handler_idx == WINED3DSIH_ENDLOOP || ins->handler_idx == WINED3DSIH_ENDREP)
4995     {
4996         struct list *e = list_head(&priv->control_frames);
4997         control_frame = LIST_ENTRY(e, struct control_frame, entry);
4998         list_remove(&control_frame->entry);
4999         HeapFree(GetProcessHeap(), 0, control_frame);
5000         priv->loop_depth--;
5001     }
5002     else if(ins->handler_idx == WINED3DSIH_ENDIF)
5003     {
5004         /* Non-ifc ENDIFs don't reach that place because of the return in the if block above */
5005         struct list *e = list_head(&priv->control_frames);
5006         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5007         list_remove(&control_frame->entry);
5008         HeapFree(GetProcessHeap(), 0, control_frame);
5009     }
5010
5011
5012     shader_arb_add_instruction_modifiers(ins);
5013 }
5014
5015 const shader_backend_t arb_program_shader_backend = {
5016     shader_arb_handle_instruction,
5017     shader_arb_select,
5018     shader_arb_select_depth_blt,
5019     shader_arb_deselect_depth_blt,
5020     shader_arb_update_float_vertex_constants,
5021     shader_arb_update_float_pixel_constants,
5022     shader_arb_load_constants,
5023     shader_arb_load_np2fixup_constants,
5024     shader_arb_destroy,
5025     shader_arb_alloc,
5026     shader_arb_free,
5027     shader_arb_dirty_const,
5028     shader_arb_get_caps,
5029     shader_arb_color_fixup_supported,
5030 };
5031
5032 /* ARB_fragment_program fixed function pipeline replacement definitions */
5033 #define ARB_FFP_CONST_TFACTOR           0
5034 #define ARB_FFP_CONST_SPECULAR_ENABLE   ((ARB_FFP_CONST_TFACTOR) + 1)
5035 #define ARB_FFP_CONST_CONSTANT(i)       ((ARB_FFP_CONST_SPECULAR_ENABLE) + 1 + i)
5036 #define ARB_FFP_CONST_BUMPMAT(i)        ((ARB_FFP_CONST_CONSTANT(7)) + 1 + i)
5037 #define ARB_FFP_CONST_LUMINANCE(i)      ((ARB_FFP_CONST_BUMPMAT(7)) + 1 + i)
5038
5039 struct arbfp_ffp_desc
5040 {
5041     struct ffp_frag_desc parent;
5042     GLuint shader;
5043     unsigned int num_textures_used;
5044 };
5045
5046 /* Context activation is done by the caller. */
5047 static void arbfp_enable(IWineD3DDevice *iface, BOOL enable) {
5048     ENTER_GL();
5049     if(enable) {
5050         glEnable(GL_FRAGMENT_PROGRAM_ARB);
5051         checkGLcall("glEnable(GL_FRAGMENT_PROGRAM_ARB)");
5052     } else {
5053         glDisable(GL_FRAGMENT_PROGRAM_ARB);
5054         checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
5055     }
5056     LEAVE_GL();
5057 }
5058
5059 static HRESULT arbfp_alloc(IWineD3DDevice *iface) {
5060     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *) iface;
5061     struct shader_arb_priv *priv;
5062     /* Share private data between the shader backend and the pipeline replacement, if both
5063      * are the arb implementation. This is needed to figure out whether ARBfp should be disabled
5064      * if no pixel shader is bound or not
5065      */
5066     if(This->shader_backend == &arb_program_shader_backend) {
5067         This->fragment_priv = This->shader_priv;
5068     } else {
5069         This->fragment_priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(struct shader_arb_priv));
5070         if(!This->fragment_priv) return E_OUTOFMEMORY;
5071     }
5072     priv = This->fragment_priv;
5073     if (wine_rb_init(&priv->fragment_shaders, &wined3d_ffp_frag_program_rb_functions) == -1)
5074     {
5075         ERR("Failed to initialize rbtree.\n");
5076         HeapFree(GetProcessHeap(), 0, This->fragment_priv);
5077         return E_OUTOFMEMORY;
5078     }
5079     priv->use_arbfp_fixed_func = TRUE;
5080     return WINED3D_OK;
5081 }
5082
5083 /* Context activation is done by the caller. */
5084 static void arbfp_free_ffpshader(struct wine_rb_entry *entry, void *context)
5085 {
5086     const WineD3D_GL_Info *gl_info = context;
5087     struct arbfp_ffp_desc *entry_arb = WINE_RB_ENTRY_VALUE(entry, struct arbfp_ffp_desc, parent.entry);
5088
5089     ENTER_GL();
5090     GL_EXTCALL(glDeleteProgramsARB(1, &entry_arb->shader));
5091     checkGLcall("glDeleteProgramsARB(1, &entry_arb->shader)");
5092     HeapFree(GetProcessHeap(), 0, entry_arb);
5093     LEAVE_GL();
5094 }
5095
5096 /* Context activation is done by the caller. */
5097 static void arbfp_free(IWineD3DDevice *iface) {
5098     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *) iface;
5099     struct shader_arb_priv *priv = This->fragment_priv;
5100
5101     wine_rb_destroy(&priv->fragment_shaders, arbfp_free_ffpshader, &This->adapter->gl_info);
5102     priv->use_arbfp_fixed_func = FALSE;
5103
5104     if(This->shader_backend != &arb_program_shader_backend) {
5105         HeapFree(GetProcessHeap(), 0, This->fragment_priv);
5106     }
5107 }
5108
5109 static void arbfp_get_caps(WINED3DDEVTYPE devtype, const WineD3D_GL_Info *gl_info, struct fragment_caps *caps)
5110 {
5111     caps->TextureOpCaps =  WINED3DTEXOPCAPS_DISABLE                     |
5112                            WINED3DTEXOPCAPS_SELECTARG1                  |
5113                            WINED3DTEXOPCAPS_SELECTARG2                  |
5114                            WINED3DTEXOPCAPS_MODULATE4X                  |
5115                            WINED3DTEXOPCAPS_MODULATE2X                  |
5116                            WINED3DTEXOPCAPS_MODULATE                    |
5117                            WINED3DTEXOPCAPS_ADDSIGNED2X                 |
5118                            WINED3DTEXOPCAPS_ADDSIGNED                   |
5119                            WINED3DTEXOPCAPS_ADD                         |
5120                            WINED3DTEXOPCAPS_SUBTRACT                    |
5121                            WINED3DTEXOPCAPS_ADDSMOOTH                   |
5122                            WINED3DTEXOPCAPS_BLENDCURRENTALPHA           |
5123                            WINED3DTEXOPCAPS_BLENDFACTORALPHA            |
5124                            WINED3DTEXOPCAPS_BLENDTEXTUREALPHA           |
5125                            WINED3DTEXOPCAPS_BLENDDIFFUSEALPHA           |
5126                            WINED3DTEXOPCAPS_BLENDTEXTUREALPHAPM         |
5127                            WINED3DTEXOPCAPS_MODULATEALPHA_ADDCOLOR      |
5128                            WINED3DTEXOPCAPS_MODULATECOLOR_ADDALPHA      |
5129                            WINED3DTEXOPCAPS_MODULATEINVCOLOR_ADDALPHA   |
5130                            WINED3DTEXOPCAPS_MODULATEINVALPHA_ADDCOLOR   |
5131                            WINED3DTEXOPCAPS_DOTPRODUCT3                 |
5132                            WINED3DTEXOPCAPS_MULTIPLYADD                 |
5133                            WINED3DTEXOPCAPS_LERP                        |
5134                            WINED3DTEXOPCAPS_BUMPENVMAP                  |
5135                            WINED3DTEXOPCAPS_BUMPENVMAPLUMINANCE;
5136
5137     /* TODO: Implement WINED3DTEXOPCAPS_PREMODULATE */
5138
5139     caps->MaxTextureBlendStages   = 8;
5140     caps->MaxSimultaneousTextures = min(GL_LIMITS(fragment_samplers), 8);
5141
5142     caps->PrimitiveMiscCaps |= WINED3DPMISCCAPS_TSSARGTEMP;
5143 }
5144 #undef GLINFO_LOCATION
5145
5146 #define GLINFO_LOCATION stateblock->wineD3DDevice->adapter->gl_info
5147 static void state_texfactor_arbfp(DWORD state, IWineD3DStateBlockImpl *stateblock, WineD3DContext *context) {
5148     float col[4];
5149     IWineD3DDeviceImpl *device = stateblock->wineD3DDevice;
5150
5151     /* Don't load the parameter if we're using an arbfp pixel shader, otherwise we'll overwrite
5152      * application provided constants
5153      */
5154     if(device->shader_backend == &arb_program_shader_backend) {
5155         if (use_ps(stateblock)) return;
5156
5157         device = stateblock->wineD3DDevice;
5158         device->activeContext->pshader_const_dirty[ARB_FFP_CONST_TFACTOR] = 1;
5159         device->highest_dirty_ps_const = max(device->highest_dirty_ps_const, ARB_FFP_CONST_TFACTOR + 1);
5160     }
5161
5162     D3DCOLORTOGLFLOAT4(stateblock->renderState[WINED3DRS_TEXTUREFACTOR], col);
5163     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_TFACTOR, col));
5164     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_TFACTOR, col)");
5165
5166 }
5167
5168 static void state_arb_specularenable(DWORD state, IWineD3DStateBlockImpl *stateblock, WineD3DContext *context) {
5169     float col[4];
5170     IWineD3DDeviceImpl *device = stateblock->wineD3DDevice;
5171
5172     /* Don't load the parameter if we're using an arbfp pixel shader, otherwise we'll overwrite
5173      * application provided constants
5174      */
5175     if(device->shader_backend == &arb_program_shader_backend) {
5176         if (use_ps(stateblock)) return;
5177
5178         device = stateblock->wineD3DDevice;
5179         device->activeContext->pshader_const_dirty[ARB_FFP_CONST_SPECULAR_ENABLE] = 1;
5180         device->highest_dirty_ps_const = max(device->highest_dirty_ps_const, ARB_FFP_CONST_SPECULAR_ENABLE + 1);
5181     }
5182
5183     if(stateblock->renderState[WINED3DRS_SPECULARENABLE]) {
5184         /* The specular color has no alpha */
5185         col[0] = 1.0f; col[1] = 1.0f;
5186         col[2] = 1.0f; col[3] = 0.0f;
5187     } else {
5188         col[0] = 0.0f; col[1] = 0.0f;
5189         col[2] = 0.0f; col[3] = 0.0f;
5190     }
5191     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_SPECULAR_ENABLE, col));
5192     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_SPECULAR_ENABLE, col)");
5193 }
5194
5195 static void set_bumpmat_arbfp(DWORD state, IWineD3DStateBlockImpl *stateblock, WineD3DContext *context) {
5196     DWORD stage = (state - STATE_TEXTURESTAGE(0, 0)) / (WINED3D_HIGHEST_TEXTURE_STATE + 1);
5197     IWineD3DDeviceImpl *device = stateblock->wineD3DDevice;
5198     float mat[2][2];
5199
5200     if (use_ps(stateblock))
5201     {
5202         if(stage != 0 &&
5203            ((IWineD3DPixelShaderImpl *) stateblock->pixelShader)->baseShader.reg_maps.bumpmat[stage]) {
5204             /* The pixel shader has to know the bump env matrix. Do a constants update if it isn't scheduled
5205              * anyway
5206              */
5207             if(!isStateDirty(context, STATE_PIXELSHADERCONSTANT)) {
5208                 device->StateTable[STATE_PIXELSHADERCONSTANT].apply(STATE_PIXELSHADERCONSTANT, stateblock, context);
5209             }
5210         }
5211
5212         if(device->shader_backend == &arb_program_shader_backend) {
5213             /* Exit now, don't set the bumpmat below, otherwise we may overwrite pixel shader constants */
5214             return;
5215         }
5216     } else if(device->shader_backend == &arb_program_shader_backend) {
5217         device->activeContext->pshader_const_dirty[ARB_FFP_CONST_BUMPMAT(stage)] = 1;
5218         device->highest_dirty_ps_const = max(device->highest_dirty_ps_const, ARB_FFP_CONST_BUMPMAT(stage) + 1);
5219     }
5220
5221     mat[0][0] = *((float *) &stateblock->textureState[stage][WINED3DTSS_BUMPENVMAT00]);
5222     mat[0][1] = *((float *) &stateblock->textureState[stage][WINED3DTSS_BUMPENVMAT01]);
5223     mat[1][0] = *((float *) &stateblock->textureState[stage][WINED3DTSS_BUMPENVMAT10]);
5224     mat[1][1] = *((float *) &stateblock->textureState[stage][WINED3DTSS_BUMPENVMAT11]);
5225
5226     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_BUMPMAT(stage), &mat[0][0]));
5227     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_BUMPMAT(stage), &mat[0][0])");
5228 }
5229
5230 static void tex_bumpenvlum_arbfp(DWORD state, IWineD3DStateBlockImpl *stateblock, WineD3DContext *context) {
5231     DWORD stage = (state - STATE_TEXTURESTAGE(0, 0)) / (WINED3D_HIGHEST_TEXTURE_STATE + 1);
5232     IWineD3DDeviceImpl *device = stateblock->wineD3DDevice;
5233     float param[4];
5234
5235     if (use_ps(stateblock))
5236     {
5237         if(stage != 0 &&
5238            ((IWineD3DPixelShaderImpl *) stateblock->pixelShader)->baseShader.reg_maps.luminanceparams[stage]) {
5239             /* The pixel shader has to know the luminance offset. Do a constants update if it
5240              * isn't scheduled anyway
5241              */
5242             if(!isStateDirty(context, STATE_PIXELSHADERCONSTANT)) {
5243                 device->StateTable[STATE_PIXELSHADERCONSTANT].apply(STATE_PIXELSHADERCONSTANT, stateblock, context);
5244             }
5245         }
5246
5247         if(device->shader_backend == &arb_program_shader_backend) {
5248             /* Exit now, don't set the bumpmat below, otherwise we may overwrite pixel shader constants */
5249             return;
5250         }
5251     } else if(device->shader_backend == &arb_program_shader_backend) {
5252         device->activeContext->pshader_const_dirty[ARB_FFP_CONST_LUMINANCE(stage)] = 1;
5253         device->highest_dirty_ps_const = max(device->highest_dirty_ps_const, ARB_FFP_CONST_LUMINANCE(stage) + 1);
5254     }
5255
5256     param[0] = *((float *) &stateblock->textureState[stage][WINED3DTSS_BUMPENVLSCALE]);
5257     param[1] = *((float *) &stateblock->textureState[stage][WINED3DTSS_BUMPENVLOFFSET]);
5258     param[2] = 0.0f;
5259     param[3] = 0.0f;
5260
5261     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_LUMINANCE(stage), param));
5262     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_LUMINANCE(stage), param)");
5263 }
5264
5265 static const char *get_argreg(SHADER_BUFFER *buffer, DWORD argnum, unsigned int stage, DWORD arg) {
5266     const char *ret;
5267
5268     if(arg == ARG_UNUSED) return "unused"; /* This is the marker for unused registers */
5269
5270     switch(arg & WINED3DTA_SELECTMASK) {
5271         case WINED3DTA_DIFFUSE:
5272             ret = "fragment.color.primary"; break;
5273
5274         case WINED3DTA_CURRENT:
5275             if(stage == 0) ret = "fragment.color.primary";
5276             else ret = "ret";
5277             break;
5278
5279         case WINED3DTA_TEXTURE:
5280             switch(stage) {
5281                 case 0: ret = "tex0"; break;
5282                 case 1: ret = "tex1"; break;
5283                 case 2: ret = "tex2"; break;
5284                 case 3: ret = "tex3"; break;
5285                 case 4: ret = "tex4"; break;
5286                 case 5: ret = "tex5"; break;
5287                 case 6: ret = "tex6"; break;
5288                 case 7: ret = "tex7"; break;
5289                 default: ret = "unknown texture";
5290             }
5291             break;
5292
5293         case WINED3DTA_TFACTOR:
5294             ret = "tfactor"; break;
5295
5296         case WINED3DTA_SPECULAR:
5297             ret = "fragment.color.secondary"; break;
5298
5299         case WINED3DTA_TEMP:
5300             ret = "tempreg"; break;
5301
5302         case WINED3DTA_CONSTANT:
5303             FIXME("Implement perstage constants\n");
5304             switch(stage) {
5305                 case 0: ret = "const0"; break;
5306                 case 1: ret = "const1"; break;
5307                 case 2: ret = "const2"; break;
5308                 case 3: ret = "const3"; break;
5309                 case 4: ret = "const4"; break;
5310                 case 5: ret = "const5"; break;
5311                 case 6: ret = "const6"; break;
5312                 case 7: ret = "const7"; break;
5313                 default: ret = "unknown constant";
5314             }
5315             break;
5316
5317         default:
5318             return "unknown";
5319     }
5320
5321     if(arg & WINED3DTA_COMPLEMENT) {
5322         shader_addline(buffer, "SUB arg%u, const.x, %s;\n", argnum, ret);
5323         if(argnum == 0) ret = "arg0";
5324         if(argnum == 1) ret = "arg1";
5325         if(argnum == 2) ret = "arg2";
5326     }
5327     if(arg & WINED3DTA_ALPHAREPLICATE) {
5328         shader_addline(buffer, "MOV arg%u, %s.w;\n", argnum, ret);
5329         if(argnum == 0) ret = "arg0";
5330         if(argnum == 1) ret = "arg1";
5331         if(argnum == 2) ret = "arg2";
5332     }
5333     return ret;
5334 }
5335
5336 static void gen_ffp_instr(SHADER_BUFFER *buffer, unsigned int stage, BOOL color, BOOL alpha,
5337                           DWORD dst, DWORD op, DWORD dw_arg0, DWORD dw_arg1, DWORD dw_arg2) {
5338     const char *dstmask, *dstreg, *arg0, *arg1, *arg2;
5339     unsigned int mul = 1;
5340     BOOL mul_final_dest = FALSE;
5341
5342     if(color && alpha) dstmask = "";
5343     else if(color) dstmask = ".xyz";
5344     else dstmask = ".w";
5345
5346     if(dst == tempreg) dstreg = "tempreg";
5347     else dstreg = "ret";
5348
5349     arg0 = get_argreg(buffer, 0, stage, dw_arg0);
5350     arg1 = get_argreg(buffer, 1, stage, dw_arg1);
5351     arg2 = get_argreg(buffer, 2, stage, dw_arg2);
5352
5353     switch(op) {
5354         case WINED3DTOP_DISABLE:
5355             if(stage == 0) shader_addline(buffer, "MOV %s%s, fragment.color.primary;\n", dstreg, dstmask);
5356             break;
5357
5358         case WINED3DTOP_SELECTARG2:
5359             arg1 = arg2;
5360         case WINED3DTOP_SELECTARG1:
5361             shader_addline(buffer, "MOV %s%s, %s;\n", dstreg, dstmask, arg1);
5362             break;
5363
5364         case WINED3DTOP_MODULATE4X:
5365             mul = 2;
5366         case WINED3DTOP_MODULATE2X:
5367             mul *= 2;
5368             if(strcmp(dstreg, "result.color") == 0) {
5369                 dstreg = "ret";
5370                 mul_final_dest = TRUE;
5371             }
5372         case WINED3DTOP_MODULATE:
5373             shader_addline(buffer, "MUL %s%s, %s, %s;\n", dstreg, dstmask, arg1, arg2);
5374             break;
5375
5376         case WINED3DTOP_ADDSIGNED2X:
5377             mul = 2;
5378             if(strcmp(dstreg, "result.color") == 0) {
5379                 dstreg = "ret";
5380                 mul_final_dest = TRUE;
5381             }
5382         case WINED3DTOP_ADDSIGNED:
5383             shader_addline(buffer, "SUB arg2, %s, const.w;\n", arg2);
5384             arg2 = "arg2";
5385         case WINED3DTOP_ADD:
5386             shader_addline(buffer, "ADD_SAT %s%s, %s, %s;\n", dstreg, dstmask, arg1, arg2);
5387             break;
5388
5389         case WINED3DTOP_SUBTRACT:
5390             shader_addline(buffer, "SUB_SAT %s%s, %s, %s;\n", dstreg, dstmask, arg1, arg2);
5391             break;
5392
5393         case WINED3DTOP_ADDSMOOTH:
5394             shader_addline(buffer, "SUB arg1, const.x, %s;\n", arg1);
5395             shader_addline(buffer, "MAD_SAT %s%s, arg1, %s, %s;\n", dstreg, dstmask, arg2, arg1);
5396             break;
5397
5398         case WINED3DTOP_BLENDCURRENTALPHA:
5399             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_CURRENT);
5400             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
5401             break;
5402         case WINED3DTOP_BLENDFACTORALPHA:
5403             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_TFACTOR);
5404             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
5405             break;
5406         case WINED3DTOP_BLENDTEXTUREALPHA:
5407             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_TEXTURE);
5408             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
5409             break;
5410         case WINED3DTOP_BLENDDIFFUSEALPHA:
5411             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_DIFFUSE);
5412             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
5413             break;
5414
5415         case WINED3DTOP_BLENDTEXTUREALPHAPM:
5416             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_TEXTURE);
5417             shader_addline(buffer, "SUB arg0.w, const.x, %s.w;\n", arg0);
5418             shader_addline(buffer, "MAD_SAT %s%s, %s, arg0.w, %s;\n", dstreg, dstmask, arg2, arg1);
5419             break;
5420
5421         /* D3DTOP_PREMODULATE ???? */
5422
5423         case WINED3DTOP_MODULATEINVALPHA_ADDCOLOR:
5424             shader_addline(buffer, "SUB arg0.w, const.x, %s;\n", arg1);
5425             shader_addline(buffer, "MAD_SAT %s%s, arg0.w, %s, %s;\n", dstreg, dstmask, arg2, arg1);
5426             break;
5427         case WINED3DTOP_MODULATEALPHA_ADDCOLOR:
5428             shader_addline(buffer, "MAD_SAT %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg1, arg2, arg1);
5429             break;
5430         case WINED3DTOP_MODULATEINVCOLOR_ADDALPHA:
5431             shader_addline(buffer, "SUB arg0, const.x, %s;\n", arg1);
5432             shader_addline(buffer, "MAD_SAT %s%s, arg0, %s, %s.w;\n", dstreg, dstmask, arg2, arg1);
5433             break;
5434         case WINED3DTOP_MODULATECOLOR_ADDALPHA:
5435             shader_addline(buffer, "MAD_SAT %s%s, %s, %s, %s.w;\n", dstreg, dstmask, arg1, arg2, arg1);
5436             break;
5437
5438         case WINED3DTOP_DOTPRODUCT3:
5439             mul = 4;
5440             if(strcmp(dstreg, "result.color") == 0) {
5441                 dstreg = "ret";
5442                 mul_final_dest = TRUE;
5443             }
5444             shader_addline(buffer, "SUB arg1, %s, const.w;\n", arg1);
5445             shader_addline(buffer, "SUB arg2, %s, const.w;\n", arg2);
5446             shader_addline(buffer, "DP3_SAT %s%s, arg1, arg2;\n", dstreg, dstmask);
5447             break;
5448
5449         case WINED3DTOP_MULTIPLYADD:
5450             shader_addline(buffer, "MAD_SAT %s%s, %s, %s, %s;\n", dstreg, dstmask, arg1, arg2, arg0);
5451             break;
5452
5453         case WINED3DTOP_LERP:
5454             /* The msdn is not quite right here */
5455             shader_addline(buffer, "LRP %s%s, %s, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
5456             break;
5457
5458         case WINED3DTOP_BUMPENVMAP:
5459         case WINED3DTOP_BUMPENVMAPLUMINANCE:
5460             /* Those are handled in the first pass of the shader(generation pass 1 and 2) already */
5461             break;
5462
5463         default:
5464             FIXME("Unhandled texture op %08x\n", op);
5465     }
5466
5467     if(mul == 2) {
5468         shader_addline(buffer, "MUL_SAT %s%s, %s, const.y;\n", mul_final_dest ? "result.color" : dstreg, dstmask, dstreg);
5469     } else if(mul == 4) {
5470         shader_addline(buffer, "MUL_SAT %s%s, %s, const.z;\n", mul_final_dest ? "result.color" : dstreg, dstmask, dstreg);
5471     }
5472 }
5473
5474 /* The stateblock is passed for GLINFO_LOCATION */
5475 static GLuint gen_arbfp_ffp_shader(const struct ffp_frag_settings *settings, IWineD3DStateBlockImpl *stateblock)
5476 {
5477     unsigned int stage;
5478     SHADER_BUFFER buffer;
5479     BOOL tex_read[MAX_TEXTURES] = {FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE};
5480     BOOL bump_used[MAX_TEXTURES] = {FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE};
5481     BOOL luminance_used[MAX_TEXTURES] = {FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE};
5482     const char *textype;
5483     const char *instr, *sat;
5484     char colorcor_dst[8];
5485     GLuint ret;
5486     DWORD arg0, arg1, arg2;
5487     BOOL tempreg_used = FALSE, tfactor_used = FALSE;
5488     BOOL op_equal;
5489     const char *final_combiner_src = "ret";
5490
5491     /* Find out which textures are read */
5492     for(stage = 0; stage < MAX_TEXTURES; stage++) {
5493         if(settings->op[stage].cop == WINED3DTOP_DISABLE) break;
5494         arg0 = settings->op[stage].carg0 & WINED3DTA_SELECTMASK;
5495         arg1 = settings->op[stage].carg1 & WINED3DTA_SELECTMASK;
5496         arg2 = settings->op[stage].carg2 & WINED3DTA_SELECTMASK;
5497         if(arg0 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
5498         if(arg1 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
5499         if(arg2 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
5500
5501         if(settings->op[stage].cop == WINED3DTOP_BLENDTEXTUREALPHA) tex_read[stage] = TRUE;
5502         if(settings->op[stage].cop == WINED3DTOP_BLENDTEXTUREALPHAPM) tex_read[stage] = TRUE;
5503         if(settings->op[stage].cop == WINED3DTOP_BUMPENVMAP) {
5504             bump_used[stage] = TRUE;
5505             tex_read[stage] = TRUE;
5506         }
5507         if(settings->op[stage].cop == WINED3DTOP_BUMPENVMAPLUMINANCE) {
5508             bump_used[stage] = TRUE;
5509             tex_read[stage] = TRUE;
5510             luminance_used[stage] = TRUE;
5511         } else if(settings->op[stage].cop == WINED3DTOP_BLENDFACTORALPHA) {
5512             tfactor_used = TRUE;
5513         }
5514
5515         if(arg0 == WINED3DTA_TFACTOR || arg1 == WINED3DTA_TFACTOR || arg2 == WINED3DTA_TFACTOR) {
5516             tfactor_used = TRUE;
5517         }
5518
5519         if(settings->op[stage].dst == tempreg) tempreg_used = TRUE;
5520         if(arg0 == WINED3DTA_TEMP || arg1 == WINED3DTA_TEMP || arg2 == WINED3DTA_TEMP) {
5521             tempreg_used = TRUE;
5522         }
5523
5524         if(settings->op[stage].aop == WINED3DTOP_DISABLE) continue;
5525         arg0 = settings->op[stage].aarg0 & WINED3DTA_SELECTMASK;
5526         arg1 = settings->op[stage].aarg1 & WINED3DTA_SELECTMASK;
5527         arg2 = settings->op[stage].aarg2 & WINED3DTA_SELECTMASK;
5528         if(arg0 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
5529         if(arg1 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
5530         if(arg2 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
5531
5532         if(arg0 == WINED3DTA_TEMP || arg1 == WINED3DTA_TEMP || arg2 == WINED3DTA_TEMP) {
5533             tempreg_used = TRUE;
5534         }
5535         if(arg0 == WINED3DTA_TFACTOR || arg1 == WINED3DTA_TFACTOR || arg2 == WINED3DTA_TFACTOR) {
5536             tfactor_used = TRUE;
5537         }
5538     }
5539
5540     /* Shader header */
5541     shader_buffer_init(&buffer);
5542
5543     shader_addline(&buffer, "!!ARBfp1.0\n");
5544
5545     switch(settings->fog) {
5546         case FOG_OFF:                                                         break;
5547         case FOG_LINEAR: shader_addline(&buffer, "OPTION ARB_fog_linear;\n"); break;
5548         case FOG_EXP:    shader_addline(&buffer, "OPTION ARB_fog_exp;\n");    break;
5549         case FOG_EXP2:   shader_addline(&buffer, "OPTION ARB_fog_exp2;\n");   break;
5550         default: FIXME("Unexpected fog setting %d\n", settings->fog);
5551     }
5552
5553     shader_addline(&buffer, "PARAM const = {1, 2, 4, 0.5};\n");
5554     shader_addline(&buffer, "TEMP TMP;\n");
5555     shader_addline(&buffer, "TEMP ret;\n");
5556     if(tempreg_used || settings->sRGB_write) shader_addline(&buffer, "TEMP tempreg;\n");
5557     shader_addline(&buffer, "TEMP arg0;\n");
5558     shader_addline(&buffer, "TEMP arg1;\n");
5559     shader_addline(&buffer, "TEMP arg2;\n");
5560     for(stage = 0; stage < MAX_TEXTURES; stage++) {
5561         if(!tex_read[stage]) continue;
5562         shader_addline(&buffer, "TEMP tex%u;\n", stage);
5563         if(!bump_used[stage]) continue;
5564         shader_addline(&buffer, "PARAM bumpmat%u = program.env[%u];\n", stage, ARB_FFP_CONST_BUMPMAT(stage));
5565         if(!luminance_used[stage]) continue;
5566         shader_addline(&buffer, "PARAM luminance%u = program.env[%u];\n", stage, ARB_FFP_CONST_LUMINANCE(stage));
5567     }
5568     if(tfactor_used) {
5569         shader_addline(&buffer, "PARAM tfactor = program.env[%u];\n", ARB_FFP_CONST_TFACTOR);
5570     }
5571         shader_addline(&buffer, "PARAM specular_enable = program.env[%u];\n", ARB_FFP_CONST_SPECULAR_ENABLE);
5572
5573     if(settings->sRGB_write) {
5574         shader_addline(&buffer, "PARAM srgb_consts1 = {%f, %f, %f, %f};\n",
5575                        srgb_mul_low, srgb_cmp, srgb_pow, srgb_mul_high);
5576         shader_addline(&buffer, "PARAM srgb_consts2 = {%f, %f, %f, %f};\n",
5577                        srgb_sub_high, 0.0, 0.0, 0.0);
5578     }
5579
5580     if(ffp_clip_emul(stateblock) && settings->emul_clipplanes) shader_addline(&buffer, "KIL fragment.texcoord[7];\n");
5581
5582     /* Generate texture sampling instructions) */
5583     for(stage = 0; stage < MAX_TEXTURES && settings->op[stage].cop != WINED3DTOP_DISABLE; stage++) {
5584         if(!tex_read[stage]) continue;
5585
5586         switch(settings->op[stage].tex_type) {
5587             case tex_1d:                    textype = "1D";     break;
5588             case tex_2d:                    textype = "2D";     break;
5589             case tex_3d:                    textype = "3D";     break;
5590             case tex_cube:                  textype = "CUBE";   break;
5591             case tex_rect:                  textype = "RECT";   break;
5592             default: textype = "unexpected_textype";   break;
5593         }
5594
5595         if(settings->op[stage].cop == WINED3DTOP_BUMPENVMAP ||
5596            settings->op[stage].cop == WINED3DTOP_BUMPENVMAPLUMINANCE) {
5597             sat = "";
5598         } else {
5599             sat = "_SAT";
5600         }
5601
5602         if(settings->op[stage].projected == proj_none) {
5603             instr = "TEX";
5604         } else if(settings->op[stage].projected == proj_count4 ||
5605                   settings->op[stage].projected == proj_count3) {
5606             instr = "TXP";
5607         } else {
5608             FIXME("Unexpected projection mode %d\n", settings->op[stage].projected);
5609             instr = "TXP";
5610         }
5611
5612         if(stage > 0 &&
5613            (settings->op[stage - 1].cop == WINED3DTOP_BUMPENVMAP ||
5614             settings->op[stage - 1].cop == WINED3DTOP_BUMPENVMAPLUMINANCE)) {
5615             shader_addline(&buffer, "SWZ arg1, bumpmat%u, x, z, 0, 0;\n", stage - 1);
5616             shader_addline(&buffer, "DP3 ret.x, arg1, tex%u;\n", stage - 1);
5617             shader_addline(&buffer, "SWZ arg1, bumpmat%u, y, w, 0, 0;\n", stage - 1);
5618             shader_addline(&buffer, "DP3 ret.y, arg1, tex%u;\n", stage - 1);
5619
5620             /* with projective textures, texbem only divides the static texture coord, not the displacement,
5621              * so multiply the displacement with the dividing parameter before passing it to TXP
5622              */
5623             if (settings->op[stage].projected != proj_none) {
5624                 if(settings->op[stage].projected == proj_count4) {
5625                     shader_addline(&buffer, "MOV ret.w, fragment.texcoord[%u].w;\n", stage);
5626                     shader_addline(&buffer, "MUL ret.xyz, ret, fragment.texcoord[%u].w, fragment.texcoord[%u];\n", stage, stage);
5627                 } else {
5628                     shader_addline(&buffer, "MOV ret.w, fragment.texcoord[%u].z;\n", stage);
5629                     shader_addline(&buffer, "MAD ret.xyz, ret, fragment.texcoord[%u].z, fragment.texcoord[%u];\n", stage, stage);
5630                 }
5631             } else {
5632                 shader_addline(&buffer, "ADD ret, ret, fragment.texcoord[%u];\n", stage);
5633             }
5634
5635             shader_addline(&buffer, "%s%s tex%u, ret, texture[%u], %s;\n",
5636                            instr, sat, stage, stage, textype);
5637             if(settings->op[stage - 1].cop == WINED3DTOP_BUMPENVMAPLUMINANCE) {
5638                 shader_addline(&buffer, "MAD_SAT ret.x, tex%u.z, luminance%u.x, luminance%u.y;\n",
5639                                stage - 1, stage - 1, stage - 1);
5640                 shader_addline(&buffer, "MUL tex%u, tex%u, ret.x;\n", stage, stage);
5641             }
5642         } else if(settings->op[stage].projected == proj_count3) {
5643             shader_addline(&buffer, "MOV ret, fragment.texcoord[%u];\n", stage);
5644             shader_addline(&buffer, "MOV ret.w, ret.z;\n");
5645             shader_addline(&buffer, "%s%s tex%u, ret, texture[%u], %s;\n",
5646                             instr, sat, stage, stage, textype);
5647         } else {
5648             shader_addline(&buffer, "%s%s tex%u, fragment.texcoord[%u], texture[%u], %s;\n",
5649                             instr, sat, stage, stage, stage, textype);
5650         }
5651
5652         sprintf(colorcor_dst, "tex%u", stage);
5653         gen_color_correction(&buffer, colorcor_dst, WINED3DSP_WRITEMASK_ALL, "const.x", "const.y",
5654                 settings->op[stage].color_fixup);
5655     }
5656
5657     /* Generate the main shader */
5658     for(stage = 0; stage < MAX_TEXTURES; stage++) {
5659         if(settings->op[stage].cop == WINED3DTOP_DISABLE) {
5660             if(stage == 0) {
5661                 final_combiner_src = "fragment.color.primary";
5662             }
5663             break;
5664         }
5665
5666         if(settings->op[stage].cop == WINED3DTOP_SELECTARG1 &&
5667            settings->op[stage].aop == WINED3DTOP_SELECTARG1) {
5668             op_equal = settings->op[stage].carg1 == settings->op[stage].aarg1;
5669         } else if(settings->op[stage].cop == WINED3DTOP_SELECTARG1 &&
5670                   settings->op[stage].aop == WINED3DTOP_SELECTARG2) {
5671             op_equal = settings->op[stage].carg1 == settings->op[stage].aarg2;
5672         } else if(settings->op[stage].cop == WINED3DTOP_SELECTARG2 &&
5673                   settings->op[stage].aop == WINED3DTOP_SELECTARG1) {
5674             op_equal = settings->op[stage].carg2 == settings->op[stage].aarg1;
5675         } else if(settings->op[stage].cop == WINED3DTOP_SELECTARG2 &&
5676                   settings->op[stage].aop == WINED3DTOP_SELECTARG2) {
5677             op_equal = settings->op[stage].carg2 == settings->op[stage].aarg2;
5678         } else {
5679             op_equal = settings->op[stage].aop   == settings->op[stage].cop &&
5680                        settings->op[stage].carg0 == settings->op[stage].aarg0 &&
5681                        settings->op[stage].carg1 == settings->op[stage].aarg1 &&
5682                        settings->op[stage].carg2 == settings->op[stage].aarg2;
5683         }
5684
5685         if(settings->op[stage].aop == WINED3DTOP_DISABLE) {
5686             gen_ffp_instr(&buffer, stage, TRUE, FALSE, settings->op[stage].dst,
5687                           settings->op[stage].cop, settings->op[stage].carg0,
5688                           settings->op[stage].carg1, settings->op[stage].carg2);
5689             if(stage == 0) {
5690                 shader_addline(&buffer, "MOV ret.w, fragment.color.primary.w;\n");
5691             }
5692         } else if(op_equal) {
5693             gen_ffp_instr(&buffer, stage, TRUE, TRUE, settings->op[stage].dst,
5694                           settings->op[stage].cop, settings->op[stage].carg0,
5695                           settings->op[stage].carg1, settings->op[stage].carg2);
5696         } else {
5697             gen_ffp_instr(&buffer, stage, TRUE, FALSE, settings->op[stage].dst,
5698                           settings->op[stage].cop, settings->op[stage].carg0,
5699                           settings->op[stage].carg1, settings->op[stage].carg2);
5700             gen_ffp_instr(&buffer, stage, FALSE, TRUE, settings->op[stage].dst,
5701                           settings->op[stage].aop, settings->op[stage].aarg0,
5702                           settings->op[stage].aarg1, settings->op[stage].aarg2);
5703         }
5704     }
5705
5706     if(settings->sRGB_write) {
5707         shader_addline(&buffer, "MAD ret, fragment.color.secondary, specular_enable, %s;\n", final_combiner_src);
5708         arbfp_add_sRGB_correction(&buffer, "ret", "arg0", "arg1", "arg2", "tempreg", FALSE);
5709     } else {
5710         shader_addline(&buffer, "MAD result.color, fragment.color.secondary, specular_enable, %s;\n", final_combiner_src);
5711     }
5712
5713     /* Footer */
5714     shader_addline(&buffer, "END\n");
5715
5716     /* Generate the shader */
5717     GL_EXTCALL(glGenProgramsARB(1, &ret));
5718     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, ret));
5719     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB, strlen(buffer.buffer), buffer.buffer));
5720
5721     if (glGetError() == GL_INVALID_OPERATION) {
5722         GLint pos;
5723         glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
5724         FIXME("Fragment program error at position %d: %s\n", pos,
5725               debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
5726     }
5727     shader_buffer_free(&buffer);
5728     return ret;
5729 }
5730
5731 static void fragment_prog_arbfp(DWORD state, IWineD3DStateBlockImpl *stateblock, WineD3DContext *context) {
5732     IWineD3DDeviceImpl *device = stateblock->wineD3DDevice;
5733     struct shader_arb_priv *priv = device->fragment_priv;
5734     BOOL use_pshader = use_ps(stateblock);
5735     BOOL use_vshader = use_vs(stateblock);
5736     struct ffp_frag_settings settings;
5737     const struct arbfp_ffp_desc *desc;
5738     unsigned int i;
5739
5740     TRACE("state %#x, stateblock %p, context %p\n", state, stateblock, context);
5741
5742     if(isStateDirty(context, STATE_RENDER(WINED3DRS_FOGENABLE))) {
5743         if(!use_pshader && device->shader_backend == &arb_program_shader_backend && context->last_was_pshader) {
5744             /* Reload fixed function constants since they collide with the pixel shader constants */
5745             for(i = 0; i < MAX_TEXTURES; i++) {
5746                 set_bumpmat_arbfp(STATE_TEXTURESTAGE(i, WINED3DTSS_BUMPENVMAT00), stateblock, context);
5747             }
5748             state_texfactor_arbfp(STATE_RENDER(WINED3DRS_TEXTUREFACTOR), stateblock, context);
5749             state_arb_specularenable(STATE_RENDER(WINED3DRS_SPECULARENABLE), stateblock, context);
5750         } else if(use_pshader && !isStateDirty(context, device->StateTable[STATE_VSHADER].representative)) {
5751             device->shader_backend->shader_select((IWineD3DDevice *)stateblock->wineD3DDevice, use_pshader, use_vshader);
5752         }
5753         return;
5754     }
5755
5756     if(!use_pshader) {
5757         /* Find or create a shader implementing the fixed function pipeline settings, then activate it */
5758         gen_ffp_frag_op(stateblock, &settings, FALSE);
5759         desc = (const struct arbfp_ffp_desc *)find_ffp_frag_shader(&priv->fragment_shaders, &settings);
5760         if(!desc) {
5761             struct arbfp_ffp_desc *new_desc = HeapAlloc(GetProcessHeap(), 0, sizeof(*new_desc));
5762             if (!new_desc)
5763             {
5764                 ERR("Out of memory\n");
5765                 return;
5766             }
5767             new_desc->num_textures_used = 0;
5768             for(i = 0; i < GL_LIMITS(texture_stages); i++) {
5769                 if(settings.op[i].cop == WINED3DTOP_DISABLE) break;
5770                 new_desc->num_textures_used = i;
5771             }
5772
5773             memcpy(&new_desc->parent.settings, &settings, sizeof(settings));
5774             new_desc->shader = gen_arbfp_ffp_shader(&settings, stateblock);
5775             add_ffp_frag_shader(&priv->fragment_shaders, &new_desc->parent);
5776             TRACE("Allocated fixed function replacement shader descriptor %p\n", new_desc);
5777             desc = new_desc;
5778         }
5779
5780         /* Now activate the replacement program. GL_FRAGMENT_PROGRAM_ARB is already active(however, note the
5781          * comment above the shader_select call below). If e.g. GLSL is active, the shader_select call will
5782          * deactivate it.
5783          */
5784         GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, desc->shader));
5785         checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, desc->shader)");
5786         priv->current_fprogram_id = desc->shader;
5787
5788         if(device->shader_backend == &arb_program_shader_backend && context->last_was_pshader) {
5789             /* Reload fixed function constants since they collide with the pixel shader constants */
5790             for(i = 0; i < MAX_TEXTURES; i++) {
5791                 set_bumpmat_arbfp(STATE_TEXTURESTAGE(i, WINED3DTSS_BUMPENVMAT00), stateblock, context);
5792             }
5793             state_texfactor_arbfp(STATE_RENDER(WINED3DRS_TEXTUREFACTOR), stateblock, context);
5794             state_arb_specularenable(STATE_RENDER(WINED3DRS_SPECULARENABLE), stateblock, context);
5795         }
5796         context->last_was_pshader = FALSE;
5797     } else {
5798         context->last_was_pshader = TRUE;
5799     }
5800
5801     /* Finally, select the shader. If a pixel shader is used, it will be set and enabled by the shader backend.
5802      * If this shader backend is arbfp(most likely), then it will simply overwrite the last fixed function replace-
5803      * ment shader. If the shader backend is not ARB, it currently is important that the opengl implementation
5804      * type overwrites GL_ARB_fragment_program. This is currently the case with GLSL. If we really want to use
5805      * atifs or nvrc pixel shaders with arb fragment programs we'd have to disable GL_FRAGMENT_PROGRAM_ARB here
5806      *
5807      * Don't call shader_select if the vertex shader is dirty, because it will be called later on by the vertex
5808      * shader handler
5809      */
5810     if(!isStateDirty(context, device->StateTable[STATE_VSHADER].representative)) {
5811         device->shader_backend->shader_select((IWineD3DDevice *)stateblock->wineD3DDevice, use_pshader, use_vshader);
5812
5813         if (!isStateDirty(context, STATE_VERTEXSHADERCONSTANT) && (use_vshader || use_pshader)) {
5814             device->StateTable[STATE_VERTEXSHADERCONSTANT].apply(STATE_VERTEXSHADERCONSTANT, stateblock, context);
5815         }
5816     }
5817     if(use_pshader) {
5818         device->StateTable[STATE_PIXELSHADERCONSTANT].apply(STATE_PIXELSHADERCONSTANT, stateblock, context);
5819     }
5820 }
5821
5822 /* We can't link the fog states to the fragment state directly since the vertex pipeline links them
5823  * to FOGENABLE. A different linking in different pipeline parts can't be expressed in the combined
5824  * state table, so we need to handle that with a forwarding function. The other invisible side effect
5825  * is that changing the fog start and fog end(which links to FOGENABLE in vertex) results in the
5826  * fragment_prog_arbfp function being called because FOGENABLE is dirty, which calls this function here
5827  */
5828 static void state_arbfp_fog(DWORD state, IWineD3DStateBlockImpl *stateblock, WineD3DContext *context) {
5829     enum fogsource new_source;
5830
5831     TRACE("state %#x, stateblock %p, context %p\n", state, stateblock, context);
5832
5833     if(!isStateDirty(context, STATE_PIXELSHADER)) {
5834         fragment_prog_arbfp(state, stateblock, context);
5835     }
5836
5837     if(!stateblock->renderState[WINED3DRS_FOGENABLE]) return;
5838
5839     if(stateblock->renderState[WINED3DRS_FOGTABLEMODE] == WINED3DFOG_NONE) {
5840         if(use_vs(stateblock)) {
5841             new_source = FOGSOURCE_VS;
5842         } else {
5843             if(stateblock->renderState[WINED3DRS_FOGVERTEXMODE] == WINED3DFOG_NONE || context->last_was_rhw) {
5844                 new_source = FOGSOURCE_COORD;
5845             } else {
5846                 new_source = FOGSOURCE_FFP;
5847             }
5848         }
5849     } else {
5850         new_source = FOGSOURCE_FFP;
5851     }
5852     if(new_source != context->fog_source) {
5853         context->fog_source = new_source;
5854         state_fogstartend(STATE_RENDER(WINED3DRS_FOGSTART), stateblock, context);
5855     }
5856 }
5857
5858 static void textransform(DWORD state, IWineD3DStateBlockImpl *stateblock, WineD3DContext *context) {
5859     if(!isStateDirty(context, STATE_PIXELSHADER)) {
5860         fragment_prog_arbfp(state, stateblock, context);
5861     }
5862 }
5863
5864 #undef GLINFO_LOCATION
5865
5866 static const struct StateEntryTemplate arbfp_fragmentstate_template[] = {
5867     {STATE_RENDER(WINED3DRS_TEXTUREFACTOR),               { STATE_RENDER(WINED3DRS_TEXTUREFACTOR),              state_texfactor_arbfp   }, WINED3D_GL_EXT_NONE             },
5868     {STATE_TEXTURESTAGE(0, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5869     {STATE_TEXTURESTAGE(0, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5870     {STATE_TEXTURESTAGE(0, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5871     {STATE_TEXTURESTAGE(0, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5872     {STATE_TEXTURESTAGE(0, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5873     {STATE_TEXTURESTAGE(0, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5874     {STATE_TEXTURESTAGE(0, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5875     {STATE_TEXTURESTAGE(0, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5876     {STATE_TEXTURESTAGE(0, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5877     {STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5878     {STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5879     {STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5880     {STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5881     {STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
5882     {STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
5883     {STATE_TEXTURESTAGE(1, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5884     {STATE_TEXTURESTAGE(1, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5885     {STATE_TEXTURESTAGE(1, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5886     {STATE_TEXTURESTAGE(1, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5887     {STATE_TEXTURESTAGE(1, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5888     {STATE_TEXTURESTAGE(1, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5889     {STATE_TEXTURESTAGE(1, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5890     {STATE_TEXTURESTAGE(1, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5891     {STATE_TEXTURESTAGE(1, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5892     {STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5893     {STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5894     {STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5895     {STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5896     {STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
5897     {STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
5898     {STATE_TEXTURESTAGE(2, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5899     {STATE_TEXTURESTAGE(2, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5900     {STATE_TEXTURESTAGE(2, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5901     {STATE_TEXTURESTAGE(2, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5902     {STATE_TEXTURESTAGE(2, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5903     {STATE_TEXTURESTAGE(2, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5904     {STATE_TEXTURESTAGE(2, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5905     {STATE_TEXTURESTAGE(2, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5906     {STATE_TEXTURESTAGE(2, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5907     {STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5908     {STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5909     {STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5910     {STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5911     {STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
5912     {STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
5913     {STATE_TEXTURESTAGE(3, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5914     {STATE_TEXTURESTAGE(3, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5915     {STATE_TEXTURESTAGE(3, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5916     {STATE_TEXTURESTAGE(3, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5917     {STATE_TEXTURESTAGE(3, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5918     {STATE_TEXTURESTAGE(3, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5919     {STATE_TEXTURESTAGE(3, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5920     {STATE_TEXTURESTAGE(3, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5921     {STATE_TEXTURESTAGE(3, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5922     {STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5923     {STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5924     {STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5925     {STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5926     {STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
5927     {STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
5928     {STATE_TEXTURESTAGE(4, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5929     {STATE_TEXTURESTAGE(4, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5930     {STATE_TEXTURESTAGE(4, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5931     {STATE_TEXTURESTAGE(4, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5932     {STATE_TEXTURESTAGE(4, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5933     {STATE_TEXTURESTAGE(4, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5934     {STATE_TEXTURESTAGE(4, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5935     {STATE_TEXTURESTAGE(4, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5936     {STATE_TEXTURESTAGE(4, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5937     {STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5938     {STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5939     {STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5940     {STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5941     {STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
5942     {STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
5943     {STATE_TEXTURESTAGE(5, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5944     {STATE_TEXTURESTAGE(5, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5945     {STATE_TEXTURESTAGE(5, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5946     {STATE_TEXTURESTAGE(5, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5947     {STATE_TEXTURESTAGE(5, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5948     {STATE_TEXTURESTAGE(5, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5949     {STATE_TEXTURESTAGE(5, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5950     {STATE_TEXTURESTAGE(5, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5951     {STATE_TEXTURESTAGE(5, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5952     {STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5953     {STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5954     {STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5955     {STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5956     {STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
5957     {STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
5958     {STATE_TEXTURESTAGE(6, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5959     {STATE_TEXTURESTAGE(6, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5960     {STATE_TEXTURESTAGE(6, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5961     {STATE_TEXTURESTAGE(6, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5962     {STATE_TEXTURESTAGE(6, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5963     {STATE_TEXTURESTAGE(6, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5964     {STATE_TEXTURESTAGE(6, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5965     {STATE_TEXTURESTAGE(6, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5966     {STATE_TEXTURESTAGE(6, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5967     {STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5968     {STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5969     {STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5970     {STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5971     {STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
5972     {STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
5973     {STATE_TEXTURESTAGE(7, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5974     {STATE_TEXTURESTAGE(7, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5975     {STATE_TEXTURESTAGE(7, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5976     {STATE_TEXTURESTAGE(7, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5977     {STATE_TEXTURESTAGE(7, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5978     {STATE_TEXTURESTAGE(7, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5979     {STATE_TEXTURESTAGE(7, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5980     {STATE_TEXTURESTAGE(7, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5981     {STATE_TEXTURESTAGE(7, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5982     {STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5983     {STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5984     {STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5985     {STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5986     {STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
5987     {STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
5988     {STATE_SAMPLER(0),                                    { STATE_SAMPLER(0),                                   sampler_texdim          }, WINED3D_GL_EXT_NONE             },
5989     {STATE_SAMPLER(1),                                    { STATE_SAMPLER(1),                                   sampler_texdim          }, WINED3D_GL_EXT_NONE             },
5990     {STATE_SAMPLER(2),                                    { STATE_SAMPLER(2),                                   sampler_texdim          }, WINED3D_GL_EXT_NONE             },
5991     {STATE_SAMPLER(3),                                    { STATE_SAMPLER(3),                                   sampler_texdim          }, WINED3D_GL_EXT_NONE             },
5992     {STATE_SAMPLER(4),                                    { STATE_SAMPLER(4),                                   sampler_texdim          }, WINED3D_GL_EXT_NONE             },
5993     {STATE_SAMPLER(5),                                    { STATE_SAMPLER(5),                                   sampler_texdim          }, WINED3D_GL_EXT_NONE             },
5994     {STATE_SAMPLER(6),                                    { STATE_SAMPLER(6),                                   sampler_texdim          }, WINED3D_GL_EXT_NONE             },
5995     {STATE_SAMPLER(7),                                    { STATE_SAMPLER(7),                                   sampler_texdim          }, WINED3D_GL_EXT_NONE             },
5996     {STATE_PIXELSHADER,                                   { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5997     {STATE_RENDER(WINED3DRS_FOGENABLE),                   { STATE_RENDER(WINED3DRS_FOGENABLE),                  state_arbfp_fog         }, WINED3D_GL_EXT_NONE             },
5998     {STATE_RENDER(WINED3DRS_FOGTABLEMODE),                { STATE_RENDER(WINED3DRS_FOGENABLE),                  state_arbfp_fog         }, WINED3D_GL_EXT_NONE             },
5999     {STATE_RENDER(WINED3DRS_FOGVERTEXMODE),               { STATE_RENDER(WINED3DRS_FOGENABLE),                  state_arbfp_fog         }, WINED3D_GL_EXT_NONE             },
6000     {STATE_RENDER(WINED3DRS_FOGSTART),                    { STATE_RENDER(WINED3DRS_FOGSTART),                   state_fogstartend       }, WINED3D_GL_EXT_NONE             },
6001     {STATE_RENDER(WINED3DRS_FOGEND),                      { STATE_RENDER(WINED3DRS_FOGSTART),                   state_fogstartend       }, WINED3D_GL_EXT_NONE             },
6002     {STATE_RENDER(WINED3DRS_SRGBWRITEENABLE),             { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6003     {STATE_RENDER(WINED3DRS_FOGCOLOR),                    { STATE_RENDER(WINED3DRS_FOGCOLOR),                   state_fogcolor          }, WINED3D_GL_EXT_NONE             },
6004     {STATE_RENDER(WINED3DRS_FOGDENSITY),                  { STATE_RENDER(WINED3DRS_FOGDENSITY),                 state_fogdensity        }, WINED3D_GL_EXT_NONE             },
6005     {STATE_TEXTURESTAGE(0,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(0, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6006     {STATE_TEXTURESTAGE(1,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(1, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6007     {STATE_TEXTURESTAGE(2,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(2, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6008     {STATE_TEXTURESTAGE(3,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(3, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6009     {STATE_TEXTURESTAGE(4,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(4, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6010     {STATE_TEXTURESTAGE(5,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(5, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6011     {STATE_TEXTURESTAGE(6,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(6, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6012     {STATE_TEXTURESTAGE(7,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(7, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6013     {STATE_RENDER(WINED3DRS_SPECULARENABLE),              { STATE_RENDER(WINED3DRS_SPECULARENABLE),             state_arb_specularenable}, WINED3D_GL_EXT_NONE             },
6014     {0 /* Terminate */,                                   { 0,                                                  0                       }, WINED3D_GL_EXT_NONE             },
6015 };
6016
6017 const struct fragment_pipeline arbfp_fragment_pipeline = {
6018     arbfp_enable,
6019     arbfp_get_caps,
6020     arbfp_alloc,
6021     arbfp_free,
6022     shader_arb_color_fixup_supported,
6023     arbfp_fragmentstate_template,
6024     TRUE /* We can disable projected textures */
6025 };
6026
6027 #define GLINFO_LOCATION device->adapter->gl_info
6028
6029 struct arbfp_blit_priv {
6030     GLenum yuy2_rect_shader, yuy2_2d_shader;
6031     GLenum uyvy_rect_shader, uyvy_2d_shader;
6032     GLenum yv12_rect_shader, yv12_2d_shader;
6033 };
6034
6035 static HRESULT arbfp_blit_alloc(IWineD3DDevice *iface) {
6036     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *) iface;
6037     device->blit_priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(struct arbfp_blit_priv));
6038     if(!device->blit_priv) {
6039         ERR("Out of memory\n");
6040         return E_OUTOFMEMORY;
6041     }
6042     return WINED3D_OK;
6043 }
6044
6045 /* Context activation is done by the caller. */
6046 static void arbfp_blit_free(IWineD3DDevice *iface) {
6047     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *) iface;
6048     struct arbfp_blit_priv *priv = device->blit_priv;
6049
6050     ENTER_GL();
6051     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yuy2_rect_shader));
6052     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yuy2_2d_shader));
6053     GL_EXTCALL(glDeleteProgramsARB(1, &priv->uyvy_rect_shader));
6054     GL_EXTCALL(glDeleteProgramsARB(1, &priv->uyvy_2d_shader));
6055     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yv12_rect_shader));
6056     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yv12_2d_shader));
6057     checkGLcall("Delete yuv programs");
6058     LEAVE_GL();
6059 }
6060
6061 static BOOL gen_planar_yuv_read(SHADER_BUFFER *buffer, enum yuv_fixup yuv_fixup, GLenum textype, char *luminance)
6062 {
6063     char chroma;
6064     const char *tex, *texinstr;
6065
6066     if (yuv_fixup == YUV_FIXUP_UYVY) {
6067         chroma = 'x';
6068         *luminance = 'w';
6069     } else {
6070         chroma = 'w';
6071         *luminance = 'x';
6072     }
6073     switch(textype) {
6074         case GL_TEXTURE_2D:             tex = "2D";     texinstr = "TXP"; break;
6075         case GL_TEXTURE_RECTANGLE_ARB:  tex = "RECT";   texinstr = "TEX"; break;
6076         default:
6077             /* This is more tricky than just replacing the texture type - we have to navigate
6078              * properly in the texture to find the correct chroma values
6079              */
6080             FIXME("Implement yuv correction for non-2d, non-rect textures\n");
6081             return FALSE;
6082     }
6083
6084     /* First we have to read the chroma values. This means we need at least two pixels(no filtering),
6085      * or 4 pixels(with filtering). To get the unmodified chromas, we have to rid ourselves of the
6086      * filtering when we sample the texture.
6087      *
6088      * These are the rules for reading the chroma:
6089      *
6090      * Even pixel: Cr
6091      * Even pixel: U
6092      * Odd pixel: V
6093      *
6094      * So we have to get the sampling x position in non-normalized coordinates in integers
6095      */
6096     if(textype != GL_TEXTURE_RECTANGLE_ARB) {
6097         shader_addline(buffer, "MUL texcrd.xy, fragment.texcoord[0], size.x;\n");
6098         shader_addline(buffer, "MOV texcrd.w, size.x;\n");
6099     } else {
6100         shader_addline(buffer, "MOV texcrd, fragment.texcoord[0];\n");
6101     }
6102     /* We must not allow filtering between pixel x and x+1, this would mix U and V
6103      * Vertical filtering is ok. However, bear in mind that the pixel center is at
6104      * 0.5, so add 0.5.
6105      */
6106     shader_addline(buffer, "FLR texcrd.x, texcrd.x;\n");
6107     shader_addline(buffer, "ADD texcrd.x, texcrd.x, coef.y;\n");
6108
6109     /* Divide the x coordinate by 0.5 and get the fraction. This gives 0.25 and 0.75 for the
6110      * even and odd pixels respectively
6111      */
6112     shader_addline(buffer, "MUL texcrd2, texcrd, coef.y;\n");
6113     shader_addline(buffer, "FRC texcrd2, texcrd2;\n");
6114
6115     /* Sample Pixel 1 */
6116     shader_addline(buffer, "%s luminance, texcrd, texture[0], %s;\n", texinstr, tex);
6117
6118     /* Put the value into either of the chroma values */
6119     shader_addline(buffer, "SGE temp.x, texcrd2.x, coef.y;\n");
6120     shader_addline(buffer, "MUL chroma.x, luminance.%c, temp.x;\n", chroma);
6121     shader_addline(buffer, "SLT temp.x, texcrd2.x, coef.y;\n");
6122     shader_addline(buffer, "MUL chroma.y, luminance.%c, temp.x;\n", chroma);
6123
6124     /* Sample pixel 2. If we read an even pixel(SLT above returned 1), sample
6125      * the pixel right to the current one. Otherwise, sample the left pixel.
6126      * Bias and scale the SLT result to -1;1 and add it to the texcrd.x.
6127      */
6128     shader_addline(buffer, "MAD temp.x, temp.x, coef.z, -coef.x;\n");
6129     shader_addline(buffer, "ADD texcrd.x, texcrd, temp.x;\n");
6130     shader_addline(buffer, "%s luminance, texcrd, texture[0], %s;\n", texinstr, tex);
6131
6132     /* Put the value into the other chroma */
6133     shader_addline(buffer, "SGE temp.x, texcrd2.x, coef.y;\n");
6134     shader_addline(buffer, "MAD chroma.y, luminance.%c, temp.x, chroma.y;\n", chroma);
6135     shader_addline(buffer, "SLT temp.x, texcrd2.x, coef.y;\n");
6136     shader_addline(buffer, "MAD chroma.x, luminance.%c, temp.x, chroma.x;\n", chroma);
6137
6138     /* TODO: If filtering is enabled, sample a 2nd pair of pixels left or right of
6139      * the current one and lerp the two U and V values
6140      */
6141
6142     /* This gives the correctly filtered luminance value */
6143     shader_addline(buffer, "TEX luminance, fragment.texcoord[0], texture[0], %s;\n", tex);
6144
6145     return TRUE;
6146 }
6147
6148 static BOOL gen_yv12_read(SHADER_BUFFER *buffer, GLenum textype, char *luminance)
6149 {
6150     const char *tex;
6151
6152     switch(textype) {
6153         case GL_TEXTURE_2D:             tex = "2D";     break;
6154         case GL_TEXTURE_RECTANGLE_ARB:  tex = "RECT";   break;
6155         default:
6156             FIXME("Implement yv12 correction for non-2d, non-rect textures\n");
6157             return FALSE;
6158     }
6159
6160     /* YV12 surfaces contain a WxH sized luminance plane, followed by a (W/2)x(H/2)
6161      * V and a (W/2)x(H/2) U plane, each with 8 bit per pixel. So the effective
6162      * bitdepth is 12 bits per pixel. Since the U and V planes have only half the
6163      * pitch of the luminance plane, the packing into the gl texture is a bit
6164      * unfortunate. If the whole texture is interpreted as luminance data it looks
6165      * approximately like this:
6166      *
6167      *        +----------------------------------+----
6168      *        |                                  |
6169      *        |                                  |
6170      *        |                                  |
6171      *        |                                  |
6172      *        |                                  |   2
6173      *        |            LUMINANCE             |   -
6174      *        |                                  |   3
6175      *        |                                  |
6176      *        |                                  |
6177      *        |                                  |
6178      *        |                                  |
6179      *        +----------------+-----------------+----
6180      *        |                |                 |
6181      *        |  U even rows   |  U odd rows     |
6182      *        |                |                 |   1
6183      *        +----------------+------------------   -
6184      *        |                |                 |   3
6185      *        |  V even rows   |  V odd rows     |
6186      *        |                |                 |
6187      *        +----------------+-----------------+----
6188      *        |                |                 |
6189      *        |     0.5        |       0.5       |
6190      *
6191      * So it appears as if there are 4 chroma images, but in fact the odd rows
6192      * in the chroma images are in the same row as the even ones. So its is
6193      * kinda tricky to read
6194      *
6195      * When reading from rectangle textures, keep in mind that the input y coordinates
6196      * go from 0 to d3d_height, whereas the opengl texture height is 1.5 * d3d_height
6197      */
6198     shader_addline(buffer, "PARAM yv12_coef = {%f, %f, %f, %f};\n",
6199             2.0f / 3.0f, 1.0f / 6.0f, (2.0f / 3.0f) + (1.0f / 6.0f), 1.0f / 3.0f);
6200
6201     shader_addline(buffer, "MOV texcrd, fragment.texcoord[0];\n");
6202     /* the chroma planes have only half the width */
6203     shader_addline(buffer, "MUL texcrd.x, texcrd.x, coef.y;\n");
6204
6205     /* The first value is between 2/3 and 5/6th of the texture's height, so scale+bias
6206      * the coordinate. Also read the right side of the image when reading odd lines
6207      *
6208      * Don't forget to clamp the y values in into the range, otherwise we'll get filtering
6209      * bleeding
6210      */
6211     if(textype == GL_TEXTURE_2D) {
6212
6213         shader_addline(buffer, "RCP chroma.w, size.y;\n");
6214
6215         shader_addline(buffer, "MUL texcrd2.y, texcrd.y, size.y;\n");
6216
6217         shader_addline(buffer, "FLR texcrd2.y, texcrd2.y;\n");
6218         shader_addline(buffer, "MAD texcrd.y, texcrd.y, yv12_coef.y, yv12_coef.x;\n");
6219
6220         /* Read odd lines from the right side(add size * 0.5 to the x coordinate */
6221         shader_addline(buffer, "ADD texcrd2.x, texcrd2.y, yv12_coef.y;\n"); /* To avoid 0.5 == 0.5 comparisons */
6222         shader_addline(buffer, "FRC texcrd2.x, texcrd2.x;\n");
6223         shader_addline(buffer, "SGE texcrd2.x, texcrd2.x, coef.y;\n");
6224         shader_addline(buffer, "MAD texcrd.x, texcrd2.x, coef.y, texcrd.x;\n");
6225
6226         /* clamp, keep the half pixel origin in mind */
6227         shader_addline(buffer, "MAD temp.y, coef.y, chroma.w, yv12_coef.x;\n");
6228         shader_addline(buffer, "MAX texcrd.y, temp.y, texcrd.y;\n");
6229         shader_addline(buffer, "MAD temp.y, -coef.y, chroma.w, yv12_coef.z;\n");
6230         shader_addline(buffer, "MIN texcrd.y, temp.y, texcrd.y;\n");
6231     } else {
6232         /* Read from [size - size+size/4] */
6233         shader_addline(buffer, "FLR texcrd.y, texcrd.y;\n");
6234         shader_addline(buffer, "MAD texcrd.y, texcrd.y, coef.w, size.y;\n");
6235
6236         /* Read odd lines from the right side(add size * 0.5 to the x coordinate */
6237         shader_addline(buffer, "ADD texcrd2.x, texcrd.y, yv12_coef.y;\n"); /* To avoid 0.5 == 0.5 comparisons */
6238         shader_addline(buffer, "FRC texcrd2.x, texcrd2.x;\n");
6239         shader_addline(buffer, "SGE texcrd2.x, texcrd2.x, coef.y;\n");
6240         shader_addline(buffer, "MUL texcrd2.x, texcrd2.x, size.x;\n");
6241         shader_addline(buffer, "MAD texcrd.x, texcrd2.x, coef.y, texcrd.x;\n");
6242
6243         /* Make sure to read exactly from the pixel center */
6244         shader_addline(buffer, "FLR texcrd.y, texcrd.y;\n");
6245         shader_addline(buffer, "ADD texcrd.y, texcrd.y, coef.y;\n");
6246
6247         /* Clamp */
6248         shader_addline(buffer, "MAD temp.y, size.y, coef.w, size.y;\n");
6249         shader_addline(buffer, "ADD temp.y, temp.y, -coef.y;\n");
6250         shader_addline(buffer, "MIN texcrd.y, temp.y, texcrd.y;\n");
6251         shader_addline(buffer, "ADD temp.y, size.y, -coef.y;\n");
6252         shader_addline(buffer, "MAX texcrd.y, temp.y, texcrd.y;\n");
6253     }
6254     /* Read the texture, put the result into the output register */
6255     shader_addline(buffer, "TEX temp, texcrd, texture[0], %s;\n", tex);
6256     shader_addline(buffer, "MOV chroma.x, temp.w;\n");
6257
6258     /* The other chroma value is 1/6th of the texture lower, from 5/6th to 6/6th
6259      * No need to clamp because we're just reusing the already clamped value from above
6260      */
6261     if(textype == GL_TEXTURE_2D) {
6262         shader_addline(buffer, "ADD texcrd.y, texcrd.y, yv12_coef.y;\n");
6263     } else {
6264         shader_addline(buffer, "MAD texcrd.y, size.y, coef.w, texcrd.y;\n");
6265     }
6266     shader_addline(buffer, "TEX temp, texcrd, texture[0], %s;\n", tex);
6267     shader_addline(buffer, "MOV chroma.y, temp.w;\n");
6268
6269     /* Sample the luminance value. It is in the top 2/3rd of the texture, so scale the y coordinate.
6270      * Clamp the y coordinate to prevent the chroma values from bleeding into the sampled luminance
6271      * values due to filtering
6272      */
6273     shader_addline(buffer, "MOV texcrd, fragment.texcoord[0];\n");
6274     if(textype == GL_TEXTURE_2D) {
6275         /* Multiply the y coordinate by 2/3 and clamp it */
6276         shader_addline(buffer, "MUL texcrd.y, texcrd.y, yv12_coef.x;\n");
6277         shader_addline(buffer, "MAD temp.y, -coef.y, chroma.w, yv12_coef.x;\n");
6278         shader_addline(buffer, "MIN texcrd.y, temp.y, texcrd.y;\n");
6279         shader_addline(buffer, "TEX luminance, texcrd, texture[0], %s;\n", tex);
6280     } else {
6281         /* Reading from texture_rectangles is pretty straightforward, just use the unmodified
6282          * texture coordinate. It is still a good idea to clamp it though, since the opengl texture
6283          * is bigger
6284          */
6285         shader_addline(buffer, "ADD temp.x, size.y, -coef.y;\n");
6286         shader_addline(buffer, "MIN texcrd.y, texcrd.y, size.x;\n");
6287         shader_addline(buffer, "TEX luminance, texcrd, texture[0], %s;\n", tex);
6288     }
6289     *luminance = 'a';
6290
6291     return TRUE;
6292 }
6293
6294 /* Context activation is done by the caller. */
6295 static GLuint gen_yuv_shader(IWineD3DDeviceImpl *device, enum yuv_fixup yuv_fixup, GLenum textype)
6296 {
6297     GLenum shader;
6298     SHADER_BUFFER buffer;
6299     char luminance_component;
6300     struct arbfp_blit_priv *priv = device->blit_priv;
6301
6302     /* Shader header */
6303     shader_buffer_init(&buffer);
6304
6305     ENTER_GL();
6306     GL_EXTCALL(glGenProgramsARB(1, &shader));
6307     checkGLcall("GL_EXTCALL(glGenProgramsARB(1, &shader))");
6308     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader));
6309     checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader)");
6310     LEAVE_GL();
6311     if(!shader) {
6312         shader_buffer_free(&buffer);
6313         return 0;
6314     }
6315
6316     /* The YUY2 and UYVY formats contain two pixels packed into a 32 bit macropixel,
6317      * giving effectively 16 bit per pixel. The color consists of a luminance(Y) and
6318      * two chroma(U and V) values. Each macropixel has two luminance values, one for
6319      * each single pixel it contains, and one U and one V value shared between both
6320      * pixels.
6321      *
6322      * The data is loaded into an A8L8 texture. With YUY2, the luminance component
6323      * contains the luminance and alpha the chroma. With UYVY it is vice versa. Thus
6324      * take the format into account when generating the read swizzles
6325      *
6326      * Reading the Y value is straightforward - just sample the texture. The hardware
6327      * takes care of filtering in the horizontal and vertical direction.
6328      *
6329      * Reading the U and V values is harder. We have to avoid filtering horizontally,
6330      * because that would mix the U and V values of one pixel or two adjacent pixels.
6331      * Thus floor the texture coordinate and add 0.5 to get an unfiltered read,
6332      * regardless of the filtering setting. Vertical filtering works automatically
6333      * though - the U and V values of two rows are mixed nicely.
6334      *
6335      * Appart of avoiding filtering issues, the code has to know which value it just
6336      * read, and where it can find the other one. To determine this, it checks if
6337      * it sampled an even or odd pixel, and shifts the 2nd read accordingly.
6338      *
6339      * Handling horizontal filtering of U and V values requires reading a 2nd pair
6340      * of pixels, extracting U and V and mixing them. This is not implemented yet.
6341      *
6342      * An alternative implementation idea is to load the texture as A8R8G8B8 texture,
6343      * with width / 2. This way one read gives all 3 values, finding U and V is easy
6344      * in an unfiltered situation. Finding the luminance on the other hand requires
6345      * finding out if it is an odd or even pixel. The real drawback of this approach
6346      * is filtering. This would have to be emulated completely in the shader, reading
6347      * up two 2 packed pixels in up to 2 rows and interpolating both horizontally and
6348      * vertically. Beyond that it would require adjustments to the texture handling
6349      * code to deal with the width scaling
6350      */
6351     shader_addline(&buffer, "!!ARBfp1.0\n");
6352     shader_addline(&buffer, "TEMP luminance;\n");
6353     shader_addline(&buffer, "TEMP temp;\n");
6354     shader_addline(&buffer, "TEMP chroma;\n");
6355     shader_addline(&buffer, "TEMP texcrd;\n");
6356     shader_addline(&buffer, "TEMP texcrd2;\n");
6357     shader_addline(&buffer, "PARAM coef = {1.0, 0.5, 2.0, 0.25};\n");
6358     shader_addline(&buffer, "PARAM yuv_coef = {1.403, 0.344, 0.714, 1.770};\n");
6359     shader_addline(&buffer, "PARAM size = program.local[0];\n");
6360
6361     switch (yuv_fixup)
6362     {
6363         case YUV_FIXUP_UYVY:
6364         case YUV_FIXUP_YUY2:
6365             if (!gen_planar_yuv_read(&buffer, yuv_fixup, textype, &luminance_component))
6366             {
6367                 shader_buffer_free(&buffer);
6368                 return 0;
6369             }
6370             break;
6371
6372         case YUV_FIXUP_YV12:
6373             if (!gen_yv12_read(&buffer, textype, &luminance_component))
6374             {
6375                 shader_buffer_free(&buffer);
6376                 return 0;
6377             }
6378             break;
6379
6380         default:
6381             FIXME("Unsupported YUV fixup %#x\n", yuv_fixup);
6382             shader_buffer_free(&buffer);
6383             return 0;
6384     }
6385
6386     /* Calculate the final result. Formula is taken from
6387      * http://www.fourcc.org/fccyvrgb.php. Note that the chroma
6388      * ranges from -0.5 to 0.5
6389      */
6390     shader_addline(&buffer, "SUB chroma.xy, chroma, coef.y;\n");
6391
6392     shader_addline(&buffer, "MAD result.color.x, chroma.x, yuv_coef.x, luminance.%c;\n", luminance_component);
6393     shader_addline(&buffer, "MAD temp.x, -chroma.y, yuv_coef.y, luminance.%c;\n", luminance_component);
6394     shader_addline(&buffer, "MAD result.color.y, -chroma.x, yuv_coef.z, temp.x;\n");
6395     shader_addline(&buffer, "MAD result.color.z, chroma.y, yuv_coef.w, luminance.%c;\n", luminance_component);
6396     shader_addline(&buffer, "END\n");
6397
6398     ENTER_GL();
6399     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB, strlen(buffer.buffer), buffer.buffer));
6400
6401     if (glGetError() == GL_INVALID_OPERATION) {
6402         GLint pos;
6403         glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
6404         FIXME("Fragment program error at position %d: %s\n", pos,
6405               debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
6406     }
6407     shader_buffer_free(&buffer);
6408     LEAVE_GL();
6409
6410     switch (yuv_fixup)
6411     {
6412         case YUV_FIXUP_YUY2:
6413             if (textype == GL_TEXTURE_RECTANGLE_ARB) priv->yuy2_rect_shader = shader;
6414             else priv->yuy2_2d_shader = shader;
6415             break;
6416
6417         case YUV_FIXUP_UYVY:
6418             if (textype == GL_TEXTURE_RECTANGLE_ARB) priv->uyvy_rect_shader = shader;
6419             else priv->uyvy_2d_shader = shader;
6420             break;
6421
6422         case YUV_FIXUP_YV12:
6423             if (textype == GL_TEXTURE_RECTANGLE_ARB) priv->yv12_rect_shader = shader;
6424             else priv->yv12_2d_shader = shader;
6425             break;
6426     }
6427
6428     return shader;
6429 }
6430
6431 /* Context activation is done by the caller. */
6432 static HRESULT arbfp_blit_set(IWineD3DDevice *iface, const struct GlPixelFormatDesc *format_desc,
6433         GLenum textype, UINT width, UINT height)
6434 {
6435     GLenum shader;
6436     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *) iface;
6437     float size[4] = {width, height, 1, 1};
6438     struct arbfp_blit_priv *priv = device->blit_priv;
6439     enum yuv_fixup yuv_fixup;
6440
6441     if (!is_yuv_fixup(format_desc->color_fixup))
6442     {
6443         TRACE("Fixup:\n");
6444         dump_color_fixup_desc(format_desc->color_fixup);
6445         /* Don't bother setting up a shader for unconverted formats */
6446         ENTER_GL();
6447         glEnable(textype);
6448         checkGLcall("glEnable(textype)");
6449         LEAVE_GL();
6450         return WINED3D_OK;
6451     }
6452
6453     yuv_fixup = get_yuv_fixup(format_desc->color_fixup);
6454
6455     switch(yuv_fixup)
6456     {
6457         case YUV_FIXUP_YUY2:
6458             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->yuy2_rect_shader : priv->yuy2_2d_shader;
6459             break;
6460
6461         case YUV_FIXUP_UYVY:
6462             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->uyvy_rect_shader : priv->uyvy_2d_shader;
6463             break;
6464
6465         case YUV_FIXUP_YV12:
6466             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->yv12_rect_shader : priv->yv12_2d_shader;
6467             break;
6468
6469         default:
6470             FIXME("Unsupported YUV fixup %#x, not setting a shader\n", yuv_fixup);
6471             ENTER_GL();
6472             glEnable(textype);
6473             checkGLcall("glEnable(textype)");
6474             LEAVE_GL();
6475             return E_NOTIMPL;
6476     }
6477
6478     if (!shader) shader = gen_yuv_shader(device, yuv_fixup, textype);
6479
6480     ENTER_GL();
6481     glEnable(GL_FRAGMENT_PROGRAM_ARB);
6482     checkGLcall("glEnable(GL_FRAGMENT_PROGRAM_ARB)");
6483     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader));
6484     checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader)");
6485     GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, 0, size));
6486     checkGLcall("glProgramLocalParameter4fvARB");
6487     LEAVE_GL();
6488
6489     return WINED3D_OK;
6490 }
6491
6492 /* Context activation is done by the caller. */
6493 static void arbfp_blit_unset(IWineD3DDevice *iface) {
6494     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *) iface;
6495
6496     ENTER_GL();
6497     glDisable(GL_FRAGMENT_PROGRAM_ARB);
6498     checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
6499     glDisable(GL_TEXTURE_2D);
6500     checkGLcall("glDisable(GL_TEXTURE_2D)");
6501     if(GL_SUPPORT(ARB_TEXTURE_CUBE_MAP)) {
6502         glDisable(GL_TEXTURE_CUBE_MAP_ARB);
6503         checkGLcall("glDisable(GL_TEXTURE_CUBE_MAP_ARB)");
6504     }
6505     if(GL_SUPPORT(ARB_TEXTURE_RECTANGLE)) {
6506         glDisable(GL_TEXTURE_RECTANGLE_ARB);
6507         checkGLcall("glDisable(GL_TEXTURE_RECTANGLE_ARB)");
6508     }
6509     LEAVE_GL();
6510 }
6511
6512 static BOOL arbfp_blit_color_fixup_supported(struct color_fixup_desc fixup)
6513 {
6514     enum yuv_fixup yuv_fixup;
6515
6516     if (TRACE_ON(d3d_shader) && TRACE_ON(d3d))
6517     {
6518         TRACE("Checking support for fixup:\n");
6519         dump_color_fixup_desc(fixup);
6520     }
6521
6522     if (is_identity_fixup(fixup))
6523     {
6524         TRACE("[OK]\n");
6525         return TRUE;
6526     }
6527
6528     /* We only support YUV conversions. */
6529     if (!is_yuv_fixup(fixup))
6530     {
6531         TRACE("[FAILED]\n");
6532         return FALSE;
6533     }
6534
6535     yuv_fixup = get_yuv_fixup(fixup);
6536     switch(yuv_fixup)
6537     {
6538         case YUV_FIXUP_YUY2:
6539         case YUV_FIXUP_UYVY:
6540         case YUV_FIXUP_YV12:
6541             TRACE("[OK]\n");
6542             return TRUE;
6543
6544         default:
6545             FIXME("Unsupported YUV fixup %#x\n", yuv_fixup);
6546             TRACE("[FAILED]\n");
6547             return FALSE;
6548     }
6549 }
6550
6551 const struct blit_shader arbfp_blit = {
6552     arbfp_blit_alloc,
6553     arbfp_blit_free,
6554     arbfp_blit_set,
6555     arbfp_blit_unset,
6556     arbfp_blit_color_fixup_supported,
6557 };
6558
6559 #undef GLINFO_LOCATION