git.oblomov.eu Git - wine/blob - dlls/wined3d/arb_program_shader.c

   1 /*
   2  * Pixel and vertex shaders implementation using ARB_vertex_program
   3  * and ARB_fragment_program GL extensions.
   4  *
   5  * Copyright 2002-2003 Jason Edmeades
   6  * Copyright 2002-2003 Raphael Junqueira
   7  * Copyright 2004 Christian Costa
   8  * Copyright 2005 Oliver Stieber
   9  * Copyright 2006 Ivan Gyurdiev
  10  * Copyright 2006 Jason Green
  11  * Copyright 2006 Henri Verbeet
  12  * Copyright 2007-2008 Stefan Dösinger for CodeWeavers
  13  * Copyright 2009 Henri Verbeet for CodeWeavers
  14  *
  15  * This library is free software; you can redistribute it and/or
  16  * modify it under the terms of the GNU Lesser General Public
  17  * License as published by the Free Software Foundation; either
  18  * version 2.1 of the License, or (at your option) any later version.
  19  *
  20  * This library is distributed in the hope that it will be useful,
  21  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  22  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  23  * Lesser General Public License for more details.
  24  *
  25  * You should have received a copy of the GNU Lesser General Public
  26  * License along with this library; if not, write to the Free Software
  27  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301, USA
  28  */
  29
  30 #include "config.h"
  31
  32 #include <math.h>
  33 #include <stdio.h>
  34
  35 #include "wined3d_private.h"
  36
  37 WINE_DEFAULT_DEBUG_CHANNEL(d3d_shader);
  38 WINE_DECLARE_DEBUG_CHANNEL(d3d_constants);
  39 WINE_DECLARE_DEBUG_CHANNEL(d3d_caps);
  40 WINE_DECLARE_DEBUG_CHANNEL(d3d);
  41
  42 #define GLINFO_LOCATION      (*gl_info)
  43
  44 /* GL locking for state handlers is done by the caller. */
  45 static BOOL need_mova_const(IWineD3DBaseShader *shader, const struct wined3d_gl_info *gl_info)
  46 {
  47     IWineD3DBaseShaderImpl *This = (IWineD3DBaseShaderImpl *) shader;
  48     if(!This->baseShader.reg_maps.usesmova) return FALSE;
  49     return !GL_SUPPORT(NV_VERTEX_PROGRAM2_OPTION);
  50 }
  51
  52 /* Returns TRUE if result.clip from GL_NV_vertex_program2 should be used and FALSE otherwise */
  53 static inline BOOL use_nv_clip(const struct wined3d_gl_info *gl_info)
  54 {
  55     return GL_SUPPORT(NV_VERTEX_PROGRAM2_OPTION);
  56 }
  57
  58 static BOOL need_helper_const(const struct wined3d_gl_info *gl_info)
  59 {
  60     if (!GL_SUPPORT(NV_VERTEX_PROGRAM) /* Need to init colors. */
  61         || gl_info->quirks & WINED3D_QUIRK_ARB_VS_OFFSET_LIMIT /* Load the immval offset. */
  62         || gl_info->quirks & WINED3D_QUIRK_SET_TEXCOORD_W /* Have to init texcoords. */
  63         || (!use_nv_clip(gl_info)) /* Init the clip texcoord */)
  64     {
  65         return TRUE;
  66     }
  67     return FALSE;
  68 }
  69
  70 static unsigned int reserved_vs_const(IWineD3DBaseShader *shader, const struct wined3d_gl_info *gl_info)
  71 {
  72     unsigned int ret = 1;
  73     /* We use one PARAM for the pos fixup, and in some cases one to load
  74      * some immediate values into the shader
  75      */
  76     if(need_helper_const(gl_info)) ret++;
  77     if(need_mova_const(shader, gl_info)) ret++;
  78     return ret;
  79 }
  80
  81 static inline BOOL ffp_clip_emul(IWineD3DStateBlockImpl *stateblock)
  82 {
  83     return stateblock->lowest_disabled_stage < 7;
  84 }
  85
  86 /* Internally used shader constants. Applications can use constants 0 to GL_LIMITS(vshader_constantsF) - 1,
  87  * so upload them above that
  88  */
  89 #define ARB_SHADER_PRIVCONST_BASE (GL_LIMITS(vshader_constantsF) - 1)
  90 #define ARB_SHADER_PRIVCONST_POS ARB_SHADER_PRIVCONST_BASE + 0
  91
  92 /* ARB_program_shader private data */
  93
  94 struct control_frame
  95 {
  96     struct                          list entry;
  97     enum
  98     {
  99         IF,
 100         IFC,
 101         LOOP,
 102         REP
 103     } type;
 104     BOOL                            muting;
 105     BOOL                            outer_loop;
 106     union
 107     {
 108         unsigned int                loop_no;
 109         unsigned int                ifc_no;
 110     };
 111     struct wined3d_shader_loop_control loop_control;
 112     BOOL                            had_else;
 113 };
 114
 115 struct arb_ps_np2fixup_info
 116 {
 117     struct ps_np2fixup_info         super;
 118     /* For ARB we need a offset value:
 119      * With both GLSL and ARB mode the NP2 fixup information (the texture dimensions) are stored in a
 120      * consecutive way (GLSL uses a uniform array). Since ARB doesn't know the notion of a "standalone"
 121      * array we need an offset to the index inside the program local parameter array. */
 122     UINT                            offset;
 123 };
 124
 125 struct arb_ps_compile_args
 126 {
 127     struct ps_compile_args          super;
 128     WORD                            bools;
 129     WORD                            clip;  /* only a boolean, use a WORD for alignment */
 130     unsigned char                   loop_ctrl[MAX_CONST_I][3];
 131 };
 132
 133 struct stb_const_desc
 134 {
 135     unsigned char           texunit;
 136     UINT                    const_num;
 137 };
 138
 139 struct arb_ps_compiled_shader
 140 {
 141     struct arb_ps_compile_args      args;
 142     struct arb_ps_np2fixup_info     np2fixup_info;
 143     struct stb_const_desc           bumpenvmatconst[MAX_TEXTURES];
 144     struct stb_const_desc           luminanceconst[MAX_TEXTURES];
 145     UINT                            int_consts[MAX_CONST_I];
 146     GLuint                          prgId;
 147     UINT                            ycorrection;
 148     unsigned char                   numbumpenvmatconsts;
 149     char                            num_int_consts;
 150 };
 151
 152 struct arb_vs_compile_args
 153 {
 154     struct vs_compile_args          super;
 155     union
 156     {
 157         struct
 158         {
 159             WORD                    bools;
 160             char                    clip_control[2];
 161         }                           boolclip;
 162         DWORD                       boolclip_compare;
 163     };
 164     DWORD                           ps_signature;
 165     union
 166     {
 167         unsigned char               vertex_samplers[4];
 168         DWORD                       vertex_samplers_compare;
 169     };
 170     unsigned char                   loop_ctrl[MAX_CONST_I][3];
 171 };
 172
 173 struct arb_vs_compiled_shader
 174 {
 175     struct arb_vs_compile_args      args;
 176     GLuint                          prgId;
 177     UINT                            int_consts[MAX_CONST_I];
 178     char                            num_int_consts;
 179     char                            need_color_unclamp;
 180     UINT                            pos_fixup;
 181 };
 182
 183 struct recorded_instruction
 184 {
 185     struct wined3d_shader_instruction ins;
 186     struct list entry;
 187 };
 188
 189 struct shader_arb_ctx_priv
 190 {
 191     char addr_reg[20];
 192     enum
 193     {
 194         /* plain GL_ARB_vertex_program or GL_ARB_fragment_program */
 195         ARB,
 196         /* GL_NV_vertex_progam2_option or GL_NV_fragment_program_option */
 197         NV2,
 198         /* GL_NV_vertex_program3 or GL_NV_fragment_program2 */
 199         NV3
 200     } target_version;
 201
 202     const struct arb_vs_compile_args    *cur_vs_args;
 203     const struct arb_ps_compile_args    *cur_ps_args;
 204     const struct arb_ps_compiled_shader *compiled_fprog;
 205     const struct arb_vs_compiled_shader *compiled_vprog;
 206     struct arb_ps_np2fixup_info         *cur_np2fixup_info;
 207     struct list                         control_frames;
 208     struct list                         record;
 209     BOOL                                recording;
 210     BOOL                                muted;
 211     unsigned int                        num_loops, loop_depth, num_ifcs;
 212     int                                 aL;
 213
 214     unsigned int                        vs_clipplanes;
 215     BOOL                                footer_written;
 216     BOOL                                in_main_func;
 217
 218     /* For 3.0 vertex shaders */
 219     const char                          *vs_output[MAX_REG_OUTPUT];
 220     /* For 2.x and earlier vertex shaders */
 221     const char                          *texcrd_output[8], *color_output[2], *fog_output;
 222
 223     /* 3.0 pshader input for compatibility with fixed function */
 224     const char                          *ps_input[MAX_REG_INPUT];
 225 };
 226
 227 struct ps_signature
 228 {
 229     struct wined3d_shader_signature_element *sig;
 230     DWORD                               idx;
 231     struct wine_rb_entry                entry;
 232 };
 233
 234 struct arb_pshader_private {
 235     struct arb_ps_compiled_shader   *gl_shaders;
 236     UINT                            num_gl_shaders, shader_array_size;
 237     BOOL                            has_signature_idx;
 238     DWORD                           input_signature_idx;
 239     DWORD                           clipplane_emulation;
 240     BOOL                            clamp_consts;
 241 };
 242
 243 struct arb_vshader_private {
 244     struct arb_vs_compiled_shader   *gl_shaders;
 245     UINT                            num_gl_shaders, shader_array_size;
 246 };
 247
 248 struct shader_arb_priv
 249 {
 250     GLuint                  current_vprogram_id;
 251     GLuint                  current_fprogram_id;
 252     const struct arb_ps_compiled_shader *compiled_fprog;
 253     const struct arb_vs_compiled_shader *compiled_vprog;
 254     GLuint                  depth_blt_vprogram_id;
 255     GLuint                  depth_blt_fprogram_id[tex_type_count];
 256     BOOL                    use_arbfp_fixed_func;
 257     struct wine_rb_tree     fragment_shaders;
 258     BOOL                    last_ps_const_clamped;
 259     BOOL                    last_vs_color_unclamp;
 260
 261     struct wine_rb_tree     signature_tree;
 262     DWORD ps_sig_number;
 263 };
 264
 265 /********************************************************
 266  * ARB_[vertex/fragment]_program helper functions follow
 267  ********************************************************/
 268
 269 /* Loads floating point constants into the currently set ARB_vertex/fragment_program.
 270  * When constant_list == NULL, it will load all the constants.
 271  *
 272  * @target_type should be either GL_VERTEX_PROGRAM_ARB (for vertex shaders)
 273  *  or GL_FRAGMENT_PROGRAM_ARB (for pixel shaders)
 274  */
 275 /* GL locking is done by the caller */
 276 static unsigned int shader_arb_load_constantsF(IWineD3DBaseShaderImpl *This, const struct wined3d_gl_info *gl_info,
 277         GLuint target_type, unsigned int max_constants, const float *constants, char *dirty_consts)
 278 {
 279     local_constant* lconst;
 280     DWORD i, j;
 281     unsigned int ret;
 282
 283     if (TRACE_ON(d3d_shader)) {
 284         for(i = 0; i < max_constants; i++) {
 285             if(!dirty_consts[i]) continue;
 286             TRACE_(d3d_constants)("Loading constants %i: %f, %f, %f, %f\n", i,
 287                         constants[i * 4 + 0], constants[i * 4 + 1],
 288                         constants[i * 4 + 2], constants[i * 4 + 3]);
 289         }
 290     }
 291
 292     i = 0;
 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         /* ps 1.x supports only 8 constants, clamp only those. When switching between 1.x and higher
 299          * shaders, the first 8 constants are marked dirty for reload
 300          */
 301         for(; i < min(8, max_constants); i++) {
 302             if(!dirty_consts[i]) continue;
 303             dirty_consts[i] = 0;
 304
 305             j = 4 * i;
 306             if (constants[j + 0] > 1.0f) lcl_const[0] = 1.0f;
 307             else if (constants[j + 0] < -1.0f) lcl_const[0] = -1.0f;
 308             else lcl_const[0] = constants[j + 0];
 309
 310             if (constants[j + 1] > 1.0f) lcl_const[1] = 1.0f;
 311             else if (constants[j + 1] < -1.0f) lcl_const[1] = -1.0f;
 312             else lcl_const[1] = constants[j + 1];
 313
 314             if (constants[j + 2] > 1.0f) lcl_const[2] = 1.0f;
 315             else if (constants[j + 2] < -1.0f) lcl_const[2] = -1.0f;
 316             else lcl_const[2] = constants[j + 2];
 317
 318             if (constants[j + 3] > 1.0f) lcl_const[3] = 1.0f;
 319             else if (constants[j + 3] < -1.0f) lcl_const[3] = -1.0f;
 320             else lcl_const[3] = constants[j + 3];
 321
 322             GL_EXTCALL(glProgramEnvParameter4fvARB(target_type, i, lcl_const));
 323         }
 324
 325         /* If further constants are dirty, reload them without clamping.
 326          *
 327          * The alternative is not to touch them, but then we cannot reset the dirty constant count
 328          * to zero. That's bad for apps that only use PS 1.x shaders, because in that case the code
 329          * above would always re-check the first 8 constants since max_constant remains at the init
 330          * value
 331          */
 332     }
 333
 334     if(GL_SUPPORT(EXT_GPU_PROGRAM_PARAMETERS)) {
 335         /* TODO: Benchmark if we're better of with finding the dirty constants ourselves,
 336          * or just reloading *all* constants at once
 337          *
 338         GL_EXTCALL(glProgramEnvParameters4fvEXT(target_type, i, max_constants, constants + (i * 4)));
 339          */
 340         for(; i < max_constants; i++) {
 341             if(!dirty_consts[i]) continue;
 342
 343             /* Find the next block of dirty constants */
 344             dirty_consts[i] = 0;
 345             j = i;
 346             for(i++; (i < max_constants) && dirty_consts[i]; i++) {
 347                 dirty_consts[i] = 0;
 348             }
 349
 350             GL_EXTCALL(glProgramEnvParameters4fvEXT(target_type, j, i - j, constants + (j * 4)));
 351         }
 352     } else {
 353         for(; i < max_constants; i++) {
 354             if(dirty_consts[i]) {
 355                 dirty_consts[i] = 0;
 356                 GL_EXTCALL(glProgramEnvParameter4fvARB(target_type, i, constants + (i * 4)));
 357             }
 358         }
 359     }
 360     checkGLcall("glProgramEnvParameter4fvARB()");
 361
 362     /* Load immediate constants */
 363     if(This->baseShader.load_local_constsF) {
 364         if (TRACE_ON(d3d_shader)) {
 365             LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry) {
 366                 GLfloat* values = (GLfloat*)lconst->value;
 367                 TRACE_(d3d_constants)("Loading local constants %i: %f, %f, %f, %f\n", lconst->idx,
 368                         values[0], values[1], values[2], values[3]);
 369             }
 370         }
 371         /* Immediate constants are clamped for 1.X shaders at loading times */
 372         ret = 0;
 373         LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry) {
 374             dirty_consts[lconst->idx] = 1; /* Dirtify so the non-immediate constant overwrites it next time */
 375             ret = max(ret, lconst->idx + 1);
 376             GL_EXTCALL(glProgramEnvParameter4fvARB(target_type, lconst->idx, (GLfloat*)lconst->value));
 377         }
 378         checkGLcall("glProgramEnvParameter4fvARB()");
 379         return ret; /* The loaded immediate constants need reloading for the next shader */
 380     } else {
 381         return 0; /* No constants are dirty now */
 382     }
 383 }
 384
 385 /**
 386  * Loads the texture dimensions for NP2 fixup into the currently set ARB_[vertex/fragment]_programs.
 387  */
 388 static void shader_arb_load_np2fixup_constants(
 389     IWineD3DDevice* device,
 390     char usePixelShader,
 391     char useVertexShader) {
 392
 393     IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl *) device;
 394     const struct shader_arb_priv* const priv = (const struct shader_arb_priv *) deviceImpl->shader_priv;
 395     IWineD3DStateBlockImpl* stateBlock = deviceImpl->stateBlock;
 396     const struct wined3d_gl_info *gl_info = &deviceImpl->adapter->gl_info;
 397
 398     if (!usePixelShader) {
 399         /* NP2 texcoord fixup is (currently) only done for pixelshaders. */
 400         return;
 401     }
 402
 403     if (priv->compiled_fprog && priv->compiled_fprog->np2fixup_info.super.active) {
 404         const struct arb_ps_np2fixup_info* const fixup = &priv->compiled_fprog->np2fixup_info;
 405         UINT i;
 406         WORD active = fixup->super.active;
 407         GLfloat np2fixup_constants[4 * MAX_FRAGMENT_SAMPLERS];
 408
 409         for (i = 0; active; active >>= 1, ++i) {
 410             const unsigned char idx = fixup->super.idx[i];
 411             const IWineD3DTextureImpl* const tex = (const IWineD3DTextureImpl*) stateBlock->textures[i];
 412             GLfloat* tex_dim = &np2fixup_constants[(idx >> 1) * 4];
 413
 414             if (!(active & 1)) continue;
 415
 416             if (!tex) {
 417                 FIXME("Nonexistent texture is flagged for NP2 texcoord fixup\n");
 418                 continue;
 419             }
 420
 421             if (idx % 2) {
 422                 tex_dim[2] = tex->baseTexture.pow2Matrix[0]; tex_dim[3] = tex->baseTexture.pow2Matrix[5];
 423             } else {
 424                 tex_dim[0] = tex->baseTexture.pow2Matrix[0]; tex_dim[1] = tex->baseTexture.pow2Matrix[5];
 425             }
 426         }
 427
 428         for (i = 0; i < fixup->super.num_consts; ++i) {
 429             GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB,
 430                                                    fixup->offset + i, &np2fixup_constants[i * 4]));
 431         }
 432     }
 433 }
 434
 435 /* GL locking is done by the caller. */
 436 static inline void shader_arb_ps_local_constants(IWineD3DDeviceImpl* deviceImpl)
 437 {
 438     const struct wined3d_context *context = context_get_current();
 439     IWineD3DStateBlockImpl* stateBlock = deviceImpl->stateBlock;
 440     const struct wined3d_gl_info *gl_info = context->gl_info;
 441     unsigned char i;
 442     struct shader_arb_priv *priv = deviceImpl->shader_priv;
 443     const struct arb_ps_compiled_shader *gl_shader = priv->compiled_fprog;
 444
 445     for(i = 0; i < gl_shader->numbumpenvmatconsts; i++)
 446     {
 447         int texunit = gl_shader->bumpenvmatconst[i].texunit;
 448
 449         /* The state manager takes care that this function is always called if the bump env matrix changes */
 450         const float *data = (const float *)&stateBlock->textureState[texunit][WINED3DTSS_BUMPENVMAT00];
 451         GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, gl_shader->bumpenvmatconst[i].const_num, data));
 452
 453         if (gl_shader->luminanceconst[i].const_num != WINED3D_CONST_NUM_UNUSED)
 454         {
 455             /* WINED3DTSS_BUMPENVLSCALE and WINED3DTSS_BUMPENVLOFFSET are next to each other.
 456              * point gl to the scale, and load 4 floats. x = scale, y = offset, z and w are junk, we
 457              * don't care about them. The pointers are valid for sure because the stateblock is bigger.
 458              * (they're WINED3DTSS_TEXTURETRANSFORMFLAGS and WINED3DTSS_ADDRESSW, so most likely 0 or NaN
 459             */
 460             const float *scale = (const float *)&stateBlock->textureState[texunit][WINED3DTSS_BUMPENVLSCALE];
 461             GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, gl_shader->luminanceconst[i].const_num, scale));
 462         }
 463     }
 464     checkGLcall("Load bumpmap consts");
 465
 466     if(gl_shader->ycorrection != WINED3D_CONST_NUM_UNUSED)
 467     {
 468         /* ycorrection.x: Backbuffer height(onscreen) or 0(offscreen).
 469         * ycorrection.y: -1.0(onscreen), 1.0(offscreen)
 470         * ycorrection.z: 1.0
 471         * ycorrection.w: 0.0
 472         */
 473         float val[4];
 474         val[0] = context->render_offscreen ? 0.0f
 475                 : ((IWineD3DSurfaceImpl *) deviceImpl->render_targets[0])->currentDesc.Height;
 476         val[1] = context->render_offscreen ? 1.0f : -1.0f;
 477         val[2] = 1.0f;
 478         val[3] = 0.0f;
 479         GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, gl_shader->ycorrection, val));
 480         checkGLcall("y correction loading");
 481     }
 482
 483     if(gl_shader->num_int_consts == 0) return;
 484
 485     for(i = 0; i < MAX_CONST_I; i++)
 486     {
 487         if(gl_shader->int_consts[i] != WINED3D_CONST_NUM_UNUSED)
 488         {
 489             float val[4];
 490             val[0] = stateBlock->pixelShaderConstantI[4 * i];
 491             val[1] = stateBlock->pixelShaderConstantI[4 * i + 1];
 492             val[2] = stateBlock->pixelShaderConstantI[4 * i + 2];
 493             val[3] = -1.0f;
 494
 495             GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, gl_shader->int_consts[i], val));
 496         }
 497     }
 498     checkGLcall("Load ps int consts");
 499 }
 500
 501 /* GL locking is done by the caller. */
 502 static inline void shader_arb_vs_local_constants(IWineD3DDeviceImpl* deviceImpl)
 503 {
 504     IWineD3DStateBlockImpl* stateBlock;
 505     const struct wined3d_gl_info *gl_info = &deviceImpl->adapter->gl_info;
 506     unsigned char i;
 507     struct shader_arb_priv *priv = deviceImpl->shader_priv;
 508     const struct arb_vs_compiled_shader *gl_shader = priv->compiled_vprog;
 509
 510     /* Upload the position fixup */
 511     GL_EXTCALL(glProgramLocalParameter4fvARB(GL_VERTEX_PROGRAM_ARB, gl_shader->pos_fixup, deviceImpl->posFixup));
 512
 513     if(gl_shader->num_int_consts == 0) return;
 514
 515     stateBlock = deviceImpl->stateBlock;
 516
 517     for(i = 0; i < MAX_CONST_I; i++)
 518     {
 519         if(gl_shader->int_consts[i] != WINED3D_CONST_NUM_UNUSED)
 520         {
 521             float val[4];
 522             val[0] = stateBlock->vertexShaderConstantI[4 * i];
 523             val[1] = stateBlock->vertexShaderConstantI[4 * i + 1];
 524             val[2] = stateBlock->vertexShaderConstantI[4 * i + 2];
 525             val[3] = -1.0f;
 526
 527             GL_EXTCALL(glProgramLocalParameter4fvARB(GL_VERTEX_PROGRAM_ARB, gl_shader->int_consts[i], val));
 528         }
 529     }
 530     checkGLcall("Load vs int consts");
 531 }
 532
 533 /**
 534  * Loads the app-supplied constants into the currently set ARB_[vertex/fragment]_programs.
 535  *
 536  * We only support float constants in ARB at the moment, so don't
 537  * worry about the Integers or Booleans
 538  */
 539 /* GL locking is done by the caller (state handler) */
 540 static void shader_arb_load_constants(const struct wined3d_context *context, char usePixelShader, char useVertexShader)
 541 {
 542     IWineD3DDeviceImpl *device = ((IWineD3DSurfaceImpl *)context->surface)->resource.wineD3DDevice;
 543     IWineD3DStateBlockImpl* stateBlock = device->stateBlock;
 544     const struct wined3d_gl_info *gl_info = context->gl_info;
 545
 546     if (useVertexShader) {
 547         IWineD3DBaseShaderImpl* vshader = (IWineD3DBaseShaderImpl*) stateBlock->vertexShader;
 548
 549         /* Load DirectX 9 float constants for vertex shader */
 550         device->highest_dirty_vs_const = shader_arb_load_constantsF(vshader, gl_info, GL_VERTEX_PROGRAM_ARB,
 551                 device->highest_dirty_vs_const, stateBlock->vertexShaderConstantF, context->vshader_const_dirty);
 552         shader_arb_vs_local_constants(device);
 553     }
 554
 555     if (usePixelShader) {
 556         IWineD3DBaseShaderImpl* pshader = (IWineD3DBaseShaderImpl*) stateBlock->pixelShader;
 557
 558         /* Load DirectX 9 float constants for pixel shader */
 559         device->highest_dirty_ps_const = shader_arb_load_constantsF(pshader, gl_info, GL_FRAGMENT_PROGRAM_ARB,
 560                 device->highest_dirty_ps_const, stateBlock->pixelShaderConstantF, context->pshader_const_dirty);
 561         shader_arb_ps_local_constants(device);
 562     }
 563 }
 564
 565 static void shader_arb_update_float_vertex_constants(IWineD3DDevice *iface, UINT start, UINT count)
 566 {
 567     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
 568     struct wined3d_context *context = context_get_current();
 569
 570     /* We don't want shader constant dirtification to be an O(contexts), so just dirtify the active
 571      * context. On a context switch the old context will be fully dirtified */
 572     if (!context || ((IWineD3DSurfaceImpl *)context->surface)->resource.wineD3DDevice != This) return;
 573
 574     memset(context->vshader_const_dirty + start, 1, sizeof(*context->vshader_const_dirty) * count);
 575     This->highest_dirty_vs_const = max(This->highest_dirty_vs_const, start + count);
 576 }
 577
 578 static void shader_arb_update_float_pixel_constants(IWineD3DDevice *iface, UINT start, UINT count)
 579 {
 580     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
 581     struct wined3d_context *context = context_get_current();
 582
 583     /* We don't want shader constant dirtification to be an O(contexts), so just dirtify the active
 584      * context. On a context switch the old context will be fully dirtified */
 585     if (!context || ((IWineD3DSurfaceImpl *)context->surface)->resource.wineD3DDevice != This) return;
 586
 587     memset(context->pshader_const_dirty + start, 1, sizeof(*context->pshader_const_dirty) * count);
 588     This->highest_dirty_ps_const = max(This->highest_dirty_ps_const, start + count);
 589 }
 590
 591 static DWORD *local_const_mapping(IWineD3DBaseShaderImpl *This)
 592 {
 593     DWORD *ret;
 594     DWORD idx = 0;
 595     const local_constant *lconst;
 596
 597     if(This->baseShader.load_local_constsF || list_empty(&This->baseShader.constantsF)) return NULL;
 598
 599     ret = HeapAlloc(GetProcessHeap(), 0, sizeof(DWORD) * This->baseShader.limits.constant_float);
 600     if(!ret) {
 601         ERR("Out of memory\n");
 602         return NULL;
 603     }
 604
 605     LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry) {
 606         ret[lconst->idx] = idx++;
 607     }
 608     return ret;
 609 }
 610
 611 /* Generate the variable & register declarations for the ARB_vertex_program output target */
 612 static DWORD shader_generate_arb_declarations(IWineD3DBaseShader *iface, const shader_reg_maps *reg_maps,
 613         struct wined3d_shader_buffer *buffer, const struct wined3d_gl_info *gl_info, DWORD *lconst_map,
 614         DWORD *num_clipplanes, struct shader_arb_ctx_priv *ctx)
 615 {
 616     IWineD3DBaseShaderImpl* This = (IWineD3DBaseShaderImpl*) iface;
 617     DWORD i, next_local = 0;
 618     char pshader = shader_is_pshader_version(reg_maps->shader_version.type);
 619     unsigned max_constantsF;
 620     const local_constant *lconst;
 621     DWORD map;
 622
 623     /* In pixel shaders, all private constants are program local, we don't need anything
 624      * from program.env. Thus we can advertise the full set of constants in pixel shaders.
 625      * If we need a private constant the GL implementation will squeeze it in somewhere
 626      *
 627      * With vertex shaders we need the posFixup and on some GL implementations 4 helper
 628      * immediate values. The posFixup is loaded using program.env for now, so always
 629      * subtract one from the number of constants. If the shader uses indirect addressing,
 630      * account for the helper const too because we have to declare all availabke d3d constants
 631      * and don't know which are actually used.
 632      */
 633     if(pshader) {
 634         max_constantsF = GL_LIMITS(pshader_constantsF);
 635     } else {
 636         if(This->baseShader.reg_maps.usesrelconstF) {
 637             DWORD highest_constf = 0, clip_limit;
 638             max_constantsF = GL_LIMITS(vshader_constantsF) - reserved_vs_const(iface, gl_info);
 639             max_constantsF -= count_bits(This->baseShader.reg_maps.integer_constants);
 640
 641             for(i = 0; i < This->baseShader.limits.constant_float; i++)
 642             {
 643                 DWORD idx = i >> 5;
 644                 DWORD shift = i & 0x1f;
 645                 if(reg_maps->constf[idx] & (1 << shift)) highest_constf = i;
 646             }
 647
 648             clip_limit = GL_LIMITS(clipplanes);
 649             if(ctx->target_version == ARB) clip_limit = min(clip_limit, 4);
 650             *num_clipplanes = min(clip_limit, max_constantsF - highest_constf - 1);
 651             max_constantsF -= *num_clipplanes;
 652             if(*num_clipplanes < clip_limit)
 653             {
 654                 WARN("Only %u clipplanes out of %u enabled\n", *num_clipplanes, GL_LIMITS(clipplanes));
 655             }
 656         }
 657         else
 658         {
 659             if(ctx->target_version >= NV2) *num_clipplanes = GL_LIMITS(clipplanes);
 660             else *num_clipplanes = min(GL_LIMITS(clipplanes), 4);
 661             max_constantsF = GL_LIMITS(vshader_constantsF);
 662         }
 663     }
 664
 665     for (i = 0, map = reg_maps->temporary; map; map >>= 1, ++i)
 666     {
 667         if (map & 1) shader_addline(buffer, "TEMP R%u;\n", i);
 668     }
 669
 670     for (i = 0, map = reg_maps->address; map; map >>= 1, ++i)
 671     {
 672         if (map & 1) shader_addline(buffer, "ADDRESS A%u;\n", i);
 673     }
 674
 675     if (pshader && reg_maps->shader_version.major == 1 && reg_maps->shader_version.minor <= 3)
 676     {
 677         for (i = 0, map = reg_maps->texcoord; map; map >>= 1, ++i)
 678         {
 679             if (map & 1) shader_addline(buffer, "TEMP T%u;\n", i);
 680         }
 681     }
 682
 683     /* Load local constants using the program-local space,
 684      * this avoids reloading them each time the shader is used
 685      */
 686     if(lconst_map) {
 687         LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry) {
 688             shader_addline(buffer, "PARAM C%u = program.local[%u];\n", lconst->idx,
 689                            lconst_map[lconst->idx]);
 690             next_local = max(next_local, lconst_map[lconst->idx] + 1);
 691         }
 692     }
 693
 694     /* we use the array-based constants array if the local constants are marked for loading,
 695      * because then we use indirect addressing, or when the local constant list is empty,
 696      * because then we don't know if we're using indirect addressing or not. If we're hardcoding
 697      * local constants do not declare the loaded constants as an array because ARB compilers usually
 698      * do not optimize unused constants away
 699      */
 700     if(This->baseShader.reg_maps.usesrelconstF) {
 701         /* Need to PARAM the environment parameters (constants) so we can use relative addressing */
 702         shader_addline(buffer, "PARAM C[%d] = { program.env[0..%d] };\n",
 703                     max_constantsF, max_constantsF - 1);
 704     } else {
 705         for(i = 0; i < max_constantsF; i++) {
 706             DWORD idx, mask;
 707             idx = i >> 5;
 708             mask = 1 << (i & 0x1f);
 709             if(!shader_constant_is_local(This, i) && (This->baseShader.reg_maps.constf[idx] & mask)) {
 710                 shader_addline(buffer, "PARAM C%d = program.env[%d];\n",i, i);
 711             }
 712         }
 713     }
 714
 715     return next_local;
 716 }
 717
 718 static const char * const shift_tab[] = {
 719     "dummy",     /*  0 (none) */
 720     "coefmul.x", /*  1 (x2)   */
 721     "coefmul.y", /*  2 (x4)   */
 722     "coefmul.z", /*  3 (x8)   */
 723     "coefmul.w", /*  4 (x16)  */
 724     "dummy",     /*  5 (x32)  */
 725     "dummy",     /*  6 (x64)  */
 726     "dummy",     /*  7 (x128) */
 727     "dummy",     /*  8 (d256) */
 728     "dummy",     /*  9 (d128) */
 729     "dummy",     /* 10 (d64)  */
 730     "dummy",     /* 11 (d32)  */
 731     "coefdiv.w", /* 12 (d16)  */
 732     "coefdiv.z", /* 13 (d8)   */
 733     "coefdiv.y", /* 14 (d4)   */
 734     "coefdiv.x"  /* 15 (d2)   */
 735 };
 736
 737 static void shader_arb_get_write_mask(const struct wined3d_shader_instruction *ins,
 738         const struct wined3d_shader_dst_param *dst, char *write_mask)
 739 {
 740     char *ptr = write_mask;
 741
 742     if (dst->write_mask != WINED3DSP_WRITEMASK_ALL)
 743     {
 744         *ptr++ = '.';
 745         if (dst->write_mask & WINED3DSP_WRITEMASK_0) *ptr++ = 'x';
 746         if (dst->write_mask & WINED3DSP_WRITEMASK_1) *ptr++ = 'y';
 747         if (dst->write_mask & WINED3DSP_WRITEMASK_2) *ptr++ = 'z';
 748         if (dst->write_mask & WINED3DSP_WRITEMASK_3) *ptr++ = 'w';
 749     }
 750
 751     *ptr = '\0';
 752 }
 753
 754 static void shader_arb_get_swizzle(const struct wined3d_shader_src_param *param, BOOL fixup, char *swizzle_str)
 755 {
 756     /* For registers of type WINED3DDECLTYPE_D3DCOLOR, data is stored as "bgra",
 757      * but addressed as "rgba". To fix this we need to swap the register's x
 758      * and z components. */
 759     const char *swizzle_chars = fixup ? "zyxw" : "xyzw";
 760     char *ptr = swizzle_str;
 761
 762     /* swizzle bits fields: wwzzyyxx */
 763     DWORD swizzle = param->swizzle;
 764     DWORD swizzle_x = swizzle & 0x03;
 765     DWORD swizzle_y = (swizzle >> 2) & 0x03;
 766     DWORD swizzle_z = (swizzle >> 4) & 0x03;
 767     DWORD swizzle_w = (swizzle >> 6) & 0x03;
 768
 769     /* If the swizzle is the default swizzle (ie, "xyzw"), we don't need to
 770      * generate a swizzle string. Unless we need to our own swizzling. */
 771     if (swizzle != WINED3DSP_NOSWIZZLE || fixup)
 772     {
 773         *ptr++ = '.';
 774         if (swizzle_x == swizzle_y && swizzle_x == swizzle_z && swizzle_x == swizzle_w) {
 775             *ptr++ = swizzle_chars[swizzle_x];
 776         } else {
 777             *ptr++ = swizzle_chars[swizzle_x];
 778             *ptr++ = swizzle_chars[swizzle_y];
 779             *ptr++ = swizzle_chars[swizzle_z];
 780             *ptr++ = swizzle_chars[swizzle_w];
 781         }
 782     }
 783
 784     *ptr = '\0';
 785 }
 786
 787 static void shader_arb_request_a0(const struct wined3d_shader_instruction *ins, const char *src)
 788 {
 789     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
 790     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
 791
 792     if(strcmp(priv->addr_reg, src) == 0) return;
 793
 794     strcpy(priv->addr_reg, src);
 795     shader_addline(buffer, "ARL A0.x, %s;\n", src);
 796 }
 797
 798 static void shader_arb_get_src_param(const struct wined3d_shader_instruction *ins,
 799         const struct wined3d_shader_src_param *src, unsigned int tmpreg, char *outregstr);
 800
 801 static void shader_arb_get_register_name(const struct wined3d_shader_instruction *ins,
 802         const struct wined3d_shader_register *reg, char *register_name, BOOL *is_color)
 803 {
 804     /* oPos, oFog and oPts in D3D */
 805     static const char * const rastout_reg_names[] = {"TMP_OUT", "result.fogcoord", "result.pointsize"};
 806     IWineD3DBaseShaderImpl *This = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
 807     BOOL pshader = shader_is_pshader_version(This->baseShader.reg_maps.shader_version.type);
 808     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
 809
 810     *is_color = FALSE;
 811
 812     switch (reg->type)
 813     {
 814         case WINED3DSPR_TEMP:
 815             sprintf(register_name, "R%u", reg->idx);
 816             break;
 817
 818         case WINED3DSPR_INPUT:
 819             if (pshader)
 820             {
 821                 if(This->baseShader.reg_maps.shader_version.major < 3)
 822                 {
 823                     if (reg->idx == 0) strcpy(register_name, "fragment.color.primary");
 824                     else strcpy(register_name, "fragment.color.secondary");
 825                 }
 826                 else
 827                 {
 828                     if(reg->rel_addr)
 829                     {
 830                         char rel_reg[50];
 831                         shader_arb_get_src_param(ins, reg->rel_addr, 0, rel_reg);
 832
 833                         if(strcmp(rel_reg, "**aL_emul**") == 0)
 834                         {
 835                             DWORD idx = ctx->aL + reg->idx;
 836                             if(idx < MAX_REG_INPUT)
 837                             {
 838                                 strcpy(register_name, ctx->ps_input[idx]);
 839                             }
 840                             else
 841                             {
 842                                 ERR("Pixel shader input register out of bounds: %u\n", idx);
 843                                 sprintf(register_name, "out_of_bounds_%u", idx);
 844                             }
 845                         }
 846                         else if(This->baseShader.reg_maps.input_registers & 0x0300)
 847                         {
 848                             /* There are two ways basically:
 849                              *
 850                              * 1) Use the unrolling code that is used for loop emulation and unroll the loop.
 851                              *    That means trouble if the loop also contains a breakc or if the control values
 852                              *    aren't local constants.
 853                              * 2) Generate an if block that checks if aL.y < 8, == 8 or == 9 and selects the
 854                              *    source dynamically. The trouble is that we cannot simply read aL.y because it
 855                              *    is an ADDRESS register. We could however push it, load .zw with a value and use
 856                              *    ADAC to load the condition code register and pop it again afterwards
 857                              */
 858                             FIXME("Relative input register addressing with more than 8 registers\n");
 859
 860                             /* This is better than nothing for now */
 861                             sprintf(register_name, "fragment.texcoord[%s + %u]", rel_reg, reg->idx);
 862                         }
 863                         else if(ctx->cur_ps_args->super.vp_mode != vertexshader)
 864                         {
 865                             /* This is problematic because we'd have to consult the ctx->ps_input strings
 866                              * for where to find the varying. Some may be "0.0", others can be texcoords or
 867                              * colors. This needs either a pipeline replacement to make the vertex shader feed
 868                              * proper varyings, or loop unrolling
 869                              *
 870                              * For now use the texcoords and hope for the best
 871                              */
 872                             FIXME("Non-vertex shader varying input with indirect addressing\n");
 873                             sprintf(register_name, "fragment.texcoord[%s + %u]", rel_reg, reg->idx);
 874                         }
 875                         else
 876                         {
 877                             /* D3D supports indirect addressing only with aL in loop registers. The loop instruction
 878                              * pulls GL_NV_fragment_program2 in
 879                              */
 880                             sprintf(register_name, "fragment.texcoord[%s + %u]", rel_reg, reg->idx);
 881                         }
 882                     }
 883                     else
 884                     {
 885                         if(reg->idx < MAX_REG_INPUT)
 886                         {
 887                             strcpy(register_name, ctx->ps_input[reg->idx]);
 888                         }
 889                         else
 890                         {
 891                             ERR("Pixel shader input register out of bounds: %u\n", reg->idx);
 892                             sprintf(register_name, "out_of_bounds_%u", reg->idx);
 893                         }
 894                     }
 895                 }
 896             }
 897             else
 898             {
 899                 if (ctx->cur_vs_args->super.swizzle_map & (1 << reg->idx)) *is_color = TRUE;
 900                 sprintf(register_name, "vertex.attrib[%u]", reg->idx);
 901             }
 902             break;
 903
 904         case WINED3DSPR_CONST:
 905             if (!pshader && reg->rel_addr)
 906             {
 907                 BOOL aL = FALSE;
 908                 char rel_reg[50];
 909                 UINT rel_offset = ((IWineD3DVertexShaderImpl *)This)->rel_offset;
 910                 if(This->baseShader.reg_maps.shader_version.major < 2) {
 911                     sprintf(rel_reg, "A0.x");
 912                 } else {
 913                     shader_arb_get_src_param(ins, reg->rel_addr, 0, rel_reg);
 914                     if(ctx->target_version == ARB) {
 915                         if(strcmp(rel_reg, "**aL_emul**") == 0) {
 916                             aL = TRUE;
 917                         } else {
 918                             shader_arb_request_a0(ins, rel_reg);
 919                             sprintf(rel_reg, "A0.x");
 920                         }
 921                     }
 922                 }
 923                 if(aL)
 924                     sprintf(register_name, "C[%u]", ctx->aL + reg->idx);
 925                 else if (reg->idx >= rel_offset)
 926                     sprintf(register_name, "C[%s + %u]", rel_reg, reg->idx - rel_offset);
 927                 else
 928                     sprintf(register_name, "C[%s - %u]", rel_reg, -reg->idx + rel_offset);
 929             }
 930             else
 931             {
 932                 if (This->baseShader.reg_maps.usesrelconstF)
 933                     sprintf(register_name, "C[%u]", reg->idx);
 934                 else
 935                     sprintf(register_name, "C%u", reg->idx);
 936             }
 937             break;
 938
 939         case WINED3DSPR_TEXTURE: /* case WINED3DSPR_ADDR: */
 940             if (pshader) {
 941                 if(This->baseShader.reg_maps.shader_version.major == 1 &&
 942                    This->baseShader.reg_maps.shader_version.minor <= 3) {
 943                     /* In ps <= 1.3, Tx is a temporary register as destination to all instructions,
 944                      * and as source to most instructions. For some instructions it is the texcoord
 945                      * input. Those instructions know about the special use
 946                      */
 947                     sprintf(register_name, "T%u", reg->idx);
 948                 } else {
 949                     /* in ps 1.4 and 2.x Tx is always a (read-only) varying */
 950                     sprintf(register_name, "fragment.texcoord[%u]", reg->idx);
 951                 }
 952             }
 953             else
 954             {
 955                 if(This->baseShader.reg_maps.shader_version.major == 1 || ctx->target_version >= NV2)
 956                 {
 957                     sprintf(register_name, "A%u", reg->idx);
 958                 }
 959                 else
 960                 {
 961                     sprintf(register_name, "A%u_SHADOW", reg->idx);
 962                 }
 963             }
 964             break;
 965
 966         case WINED3DSPR_COLOROUT:
 967             if(ctx->cur_ps_args->super.srgb_correction && reg->idx == 0)
 968             {
 969                 strcpy(register_name, "TMP_COLOR");
 970             }
 971             else
 972             {
 973                 if(ctx->cur_ps_args->super.srgb_correction) FIXME("sRGB correction on higher render targets\n");
 974                 if(This->baseShader.reg_maps.highest_render_target > 0)
 975                 {
 976                     sprintf(register_name, "result.color[%u]", reg->idx);
 977                 }
 978                 else
 979                 {
 980                     strcpy(register_name, "result.color");
 981                 }
 982             }
 983             break;
 984
 985         case WINED3DSPR_RASTOUT:
 986             if(reg->idx == 1) sprintf(register_name, "%s", ctx->fog_output);
 987             else sprintf(register_name, "%s", rastout_reg_names[reg->idx]);
 988             break;
 989
 990         case WINED3DSPR_DEPTHOUT:
 991             strcpy(register_name, "result.depth");
 992             break;
 993
 994         case WINED3DSPR_ATTROUT:
 995         /* case WINED3DSPR_OUTPUT: */
 996             if (pshader) sprintf(register_name, "oD[%u]", reg->idx);
 997             else strcpy(register_name, ctx->color_output[reg->idx]);
 998             break;
 999
1000         case WINED3DSPR_TEXCRDOUT:
1001             if (pshader)
1002             {
1003                 sprintf(register_name, "oT[%u]", reg->idx);
1004             }
1005             else
1006             {
1007                 if(This->baseShader.reg_maps.shader_version.major < 3)
1008                 {
1009                     strcpy(register_name, ctx->texcrd_output[reg->idx]);
1010                 }
1011                 else
1012                 {
1013                     strcpy(register_name, ctx->vs_output[reg->idx]);
1014                 }
1015             }
1016             break;
1017
1018         case WINED3DSPR_LOOP:
1019             if(ctx->target_version >= NV2)
1020             {
1021                 /* Pshader has an implicitly declared loop index counter A0.x that cannot be renamed */
1022                 if(pshader) sprintf(register_name, "A0.x");
1023                 else sprintf(register_name, "aL.y");
1024             }
1025             else
1026             {
1027                 /* Unfortunately this code cannot return the value of ctx->aL here. An immediate value
1028                  * would be valid, but if aL is used for indexing(its only use), there's likely an offset,
1029                  * thus the result would be something like C[15 + 30], which is not valid in the ARB program
1030                  * grammar. So return a marker for the emulated aL and intercept it in constant and varying
1031                  * indexing
1032                  */
1033                 sprintf(register_name, "**aL_emul**");
1034             }
1035
1036             break;
1037
1038         case WINED3DSPR_CONSTINT:
1039             sprintf(register_name, "I%u", reg->idx);
1040             break;
1041
1042         case WINED3DSPR_MISCTYPE:
1043             if(reg->idx == 0)
1044             {
1045                 sprintf(register_name, "vpos");
1046             }
1047             else if(reg->idx == 1)
1048             {
1049                 sprintf(register_name, "fragment.facing.x");
1050             }
1051             else
1052             {
1053                 FIXME("Unknown MISCTYPE register index %u\n", reg->idx);
1054             }
1055             break;
1056
1057         default:
1058             FIXME("Unhandled register type %#x[%u]\n", reg->type, reg->idx);
1059             sprintf(register_name, "unrecognized_register[%u]", reg->idx);
1060             break;
1061     }
1062 }
1063
1064 static void shader_arb_get_dst_param(const struct wined3d_shader_instruction *ins,
1065         const struct wined3d_shader_dst_param *wined3d_dst, char *str)
1066 {
1067     char register_name[255];
1068     char write_mask[6];
1069     BOOL is_color;
1070
1071     shader_arb_get_register_name(ins, &wined3d_dst->reg, register_name, &is_color);
1072     strcpy(str, register_name);
1073
1074     shader_arb_get_write_mask(ins, wined3d_dst, write_mask);
1075     strcat(str, write_mask);
1076 }
1077
1078 static const char *shader_arb_get_fixup_swizzle(enum fixup_channel_source channel_source)
1079 {
1080     switch(channel_source)
1081     {
1082         case CHANNEL_SOURCE_ZERO: return "0";
1083         case CHANNEL_SOURCE_ONE: return "1";
1084         case CHANNEL_SOURCE_X: return "x";
1085         case CHANNEL_SOURCE_Y: return "y";
1086         case CHANNEL_SOURCE_Z: return "z";
1087         case CHANNEL_SOURCE_W: return "w";
1088         default:
1089             FIXME("Unhandled channel source %#x\n", channel_source);
1090             return "undefined";
1091     }
1092 }
1093
1094 static void gen_color_correction(struct wined3d_shader_buffer *buffer, const char *reg,
1095         DWORD dst_mask, const char *one, const char *two, struct color_fixup_desc fixup)
1096 {
1097     DWORD mask;
1098
1099     if (is_yuv_fixup(fixup))
1100     {
1101         enum yuv_fixup yuv_fixup = get_yuv_fixup(fixup);
1102         FIXME("YUV fixup (%#x) not supported\n", yuv_fixup);
1103         return;
1104     }
1105
1106     mask = 0;
1107     if (fixup.x_source != CHANNEL_SOURCE_X) mask |= WINED3DSP_WRITEMASK_0;
1108     if (fixup.y_source != CHANNEL_SOURCE_Y) mask |= WINED3DSP_WRITEMASK_1;
1109     if (fixup.z_source != CHANNEL_SOURCE_Z) mask |= WINED3DSP_WRITEMASK_2;
1110     if (fixup.w_source != CHANNEL_SOURCE_W) mask |= WINED3DSP_WRITEMASK_3;
1111     mask &= dst_mask;
1112
1113     if (mask)
1114     {
1115         shader_addline(buffer, "SWZ %s, %s, %s, %s, %s, %s;\n", reg, reg,
1116                 shader_arb_get_fixup_swizzle(fixup.x_source), shader_arb_get_fixup_swizzle(fixup.y_source),
1117                 shader_arb_get_fixup_swizzle(fixup.z_source), shader_arb_get_fixup_swizzle(fixup.w_source));
1118     }
1119
1120     mask = 0;
1121     if (fixup.x_sign_fixup) mask |= WINED3DSP_WRITEMASK_0;
1122     if (fixup.y_sign_fixup) mask |= WINED3DSP_WRITEMASK_1;
1123     if (fixup.z_sign_fixup) mask |= WINED3DSP_WRITEMASK_2;
1124     if (fixup.w_sign_fixup) mask |= WINED3DSP_WRITEMASK_3;
1125     mask &= dst_mask;
1126
1127     if (mask)
1128     {
1129         char reg_mask[6];
1130         char *ptr = reg_mask;
1131
1132         if (mask != WINED3DSP_WRITEMASK_ALL)
1133         {
1134             *ptr++ = '.';
1135             if (mask & WINED3DSP_WRITEMASK_0) *ptr++ = 'x';
1136             if (mask & WINED3DSP_WRITEMASK_1) *ptr++ = 'y';
1137             if (mask & WINED3DSP_WRITEMASK_2) *ptr++ = 'z';
1138             if (mask & WINED3DSP_WRITEMASK_3) *ptr++ = 'w';
1139         }
1140         *ptr = '\0';
1141
1142         shader_addline(buffer, "MAD %s%s, %s, %s, -%s;\n", reg, reg_mask, reg, two, one);
1143     }
1144 }
1145
1146 static const char *shader_arb_get_modifier(const struct wined3d_shader_instruction *ins)
1147 {
1148     DWORD mod;
1149     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
1150     if (!ins->dst_count) return "";
1151
1152     mod = ins->dst[0].modifiers;
1153
1154     /* Silently ignore PARTIALPRECISION if its not supported */
1155     if(priv->target_version == ARB) mod &= ~WINED3DSPDM_PARTIALPRECISION;
1156
1157     if(mod & WINED3DSPDM_MSAMPCENTROID)
1158     {
1159         FIXME("Unhandled modifier WINED3DSPDM_MSAMPCENTROID\n");
1160         mod &= ~WINED3DSPDM_MSAMPCENTROID;
1161     }
1162
1163     switch(mod)
1164     {
1165         case WINED3DSPDM_SATURATE | WINED3DSPDM_PARTIALPRECISION:
1166             return "H_SAT";
1167
1168         case WINED3DSPDM_SATURATE:
1169             return "_SAT";
1170
1171         case WINED3DSPDM_PARTIALPRECISION:
1172             return "H";
1173
1174         case 0:
1175             return "";
1176
1177         default:
1178             FIXME("Unknown modifiers 0x%08x\n", mod);
1179             return "";
1180     }
1181 }
1182
1183 #define TEX_PROJ        0x1
1184 #define TEX_BIAS        0x2
1185 #define TEX_LOD         0x4
1186 #define TEX_DERIV       0x10
1187
1188 static void shader_hw_sample(const struct wined3d_shader_instruction *ins, DWORD sampler_idx,
1189         const char *dst_str, const char *coord_reg, WORD flags, const char *dsx, const char *dsy)
1190 {
1191     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1192     DWORD sampler_type = ins->ctx->reg_maps->sampler_type[sampler_idx];
1193     const char *tex_type;
1194     BOOL np2_fixup = FALSE;
1195     IWineD3DBaseShaderImpl *This = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
1196     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *) This->baseShader.device;
1197     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
1198     const char *mod;
1199     BOOL pshader = shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type);
1200
1201     /* D3D vertex shader sampler IDs are vertex samplers(0-3), not global d3d samplers */
1202     if(!pshader) sampler_idx += MAX_FRAGMENT_SAMPLERS;
1203
1204     switch(sampler_type) {
1205         case WINED3DSTT_1D:
1206             tex_type = "1D";
1207             break;
1208
1209         case WINED3DSTT_2D:
1210             if(device->stateBlock->textures[sampler_idx] &&
1211                IWineD3DBaseTexture_GetTextureDimensions(device->stateBlock->textures[sampler_idx]) == GL_TEXTURE_RECTANGLE_ARB) {
1212                 tex_type = "RECT";
1213             } else {
1214                 tex_type = "2D";
1215             }
1216             if (shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type))
1217             {
1218                 if (priv->cur_np2fixup_info->super.active & (1 << sampler_idx))
1219                 {
1220                     if (flags) FIXME("Only ordinary sampling from NP2 textures is supported.\n");
1221                     else np2_fixup = TRUE;
1222                 }
1223             }
1224             break;
1225
1226         case WINED3DSTT_VOLUME:
1227             tex_type = "3D";
1228             break;
1229
1230         case WINED3DSTT_CUBE:
1231             tex_type = "CUBE";
1232             break;
1233
1234         default:
1235             ERR("Unexpected texture type %d\n", sampler_type);
1236             tex_type = "";
1237     }
1238
1239     /* TEX, TXL, TXD and TXP do not support the "H" modifier,
1240      * so don't use shader_arb_get_modifier
1241      */
1242     if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE) mod = "_SAT";
1243     else mod = "";
1244
1245     /* Fragment samplers always have indentity mapping */
1246     if(sampler_idx >= MAX_FRAGMENT_SAMPLERS)
1247     {
1248         sampler_idx = priv->cur_vs_args->vertex_samplers[sampler_idx - MAX_FRAGMENT_SAMPLERS];
1249     }
1250
1251     if (flags & TEX_DERIV)
1252     {
1253         if(flags & TEX_PROJ) FIXME("Projected texture sampling with custom derivatives\n");
1254         if(flags & TEX_BIAS) FIXME("Biased texture sampling with custom derivatives\n");
1255         shader_addline(buffer, "TXD%s %s, %s, %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg,
1256                        dsx, dsy,sampler_idx, tex_type);
1257     }
1258     else if(flags & TEX_LOD)
1259     {
1260         if(flags & TEX_PROJ) FIXME("Projected texture sampling with explicit lod\n");
1261         if(flags & TEX_BIAS) FIXME("Biased texture sampling with explicit lod\n");
1262         shader_addline(buffer, "TXL%s %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg,
1263                        sampler_idx, tex_type);
1264     }
1265     else if (flags & TEX_BIAS)
1266     {
1267         /* Shouldn't be possible, but let's check for it */
1268         if(flags & TEX_PROJ) FIXME("Biased and Projected texture sampling\n");
1269         /* TXB takes the 4th component of the source vector automatically, as d3d. Nothing more to do */
1270         shader_addline(buffer, "TXB%s %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg, sampler_idx, tex_type);
1271     }
1272     else if (flags & TEX_PROJ)
1273     {
1274         shader_addline(buffer, "TXP%s %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg, sampler_idx, tex_type);
1275     }
1276     else
1277     {
1278         if (np2_fixup)
1279         {
1280             const unsigned char idx = priv->cur_np2fixup_info->super.idx[sampler_idx];
1281             shader_addline(buffer, "MUL TA, np2fixup[%u].%s, %s;\n", idx >> 1,
1282                            (idx % 2) ? "zwxy" : "xyzw", coord_reg);
1283
1284             shader_addline(buffer, "TEX%s %s, TA, texture[%u], %s;\n", mod, dst_str, sampler_idx, tex_type);
1285         }
1286         else
1287             shader_addline(buffer, "TEX%s %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg, sampler_idx, tex_type);
1288     }
1289
1290     if (pshader)
1291     {
1292         gen_color_correction(buffer, dst_str, ins->dst[0].write_mask,
1293                 "one", "coefmul.x", priv->cur_ps_args->super.color_fixup[sampler_idx]);
1294     }
1295 }
1296
1297 static void shader_arb_get_src_param(const struct wined3d_shader_instruction *ins,
1298         const struct wined3d_shader_src_param *src, unsigned int tmpreg, char *outregstr)
1299 {
1300     /* Generate a line that does the input modifier computation and return the input register to use */
1301     BOOL is_color = FALSE;
1302     char regstr[256];
1303     char swzstr[20];
1304     int insert_line;
1305     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1306     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
1307
1308     /* Assume a new line will be added */
1309     insert_line = 1;
1310
1311     /* Get register name */
1312     shader_arb_get_register_name(ins, &src->reg, regstr, &is_color);
1313     shader_arb_get_swizzle(src, is_color, swzstr);
1314
1315     switch (src->modifiers)
1316     {
1317     case WINED3DSPSM_NONE:
1318         sprintf(outregstr, "%s%s", regstr, swzstr);
1319         insert_line = 0;
1320         break;
1321     case WINED3DSPSM_NEG:
1322         sprintf(outregstr, "-%s%s", regstr, swzstr);
1323         insert_line = 0;
1324         break;
1325     case WINED3DSPSM_BIAS:
1326         shader_addline(buffer, "ADD T%c, %s, -coefdiv.x;\n", 'A' + tmpreg, regstr);
1327         break;
1328     case WINED3DSPSM_BIASNEG:
1329         shader_addline(buffer, "ADD T%c, -%s, coefdiv.x;\n", 'A' + tmpreg, regstr);
1330         break;
1331     case WINED3DSPSM_SIGN:
1332         shader_addline(buffer, "MAD T%c, %s, coefmul.x, -one.x;\n", 'A' + tmpreg, regstr);
1333         break;
1334     case WINED3DSPSM_SIGNNEG:
1335         shader_addline(buffer, "MAD T%c, %s, -coefmul.x, one.x;\n", 'A' + tmpreg, regstr);
1336         break;
1337     case WINED3DSPSM_COMP:
1338         shader_addline(buffer, "SUB T%c, one.x, %s;\n", 'A' + tmpreg, regstr);
1339         break;
1340     case WINED3DSPSM_X2:
1341         shader_addline(buffer, "ADD T%c, %s, %s;\n", 'A' + tmpreg, regstr, regstr);
1342         break;
1343     case WINED3DSPSM_X2NEG:
1344         shader_addline(buffer, "ADD T%c, -%s, -%s;\n", 'A' + tmpreg, regstr, regstr);
1345         break;
1346     case WINED3DSPSM_DZ:
1347         shader_addline(buffer, "RCP T%c, %s.z;\n", 'A' + tmpreg, regstr);
1348         shader_addline(buffer, "MUL T%c, %s, T%c;\n", 'A' + tmpreg, regstr, 'A' + tmpreg);
1349         break;
1350     case WINED3DSPSM_DW:
1351         shader_addline(buffer, "RCP T%c, %s.w;\n", 'A' + tmpreg, regstr);
1352         shader_addline(buffer, "MUL T%c, %s, T%c;\n", 'A' + tmpreg, regstr, 'A' + tmpreg);
1353         break;
1354     case WINED3DSPSM_ABS:
1355         if(ctx->target_version >= NV2) {
1356             sprintf(outregstr, "|%s%s|", regstr, swzstr);
1357             insert_line = 0;
1358         } else {
1359             shader_addline(buffer, "ABS T%c, %s;\n", 'A' + tmpreg, regstr);
1360         }
1361         break;
1362     case WINED3DSPSM_ABSNEG:
1363         if(ctx->target_version >= NV2) {
1364             sprintf(outregstr, "-|%s%s|", regstr, swzstr);
1365         } else {
1366             shader_addline(buffer, "ABS T%c, %s;\n", 'A' + tmpreg, regstr);
1367             sprintf(outregstr, "-T%c%s", 'A' + tmpreg, swzstr);
1368         }
1369         insert_line = 0;
1370         break;
1371     default:
1372         sprintf(outregstr, "%s%s", regstr, swzstr);
1373         insert_line = 0;
1374     }
1375
1376     /* Return modified or original register, with swizzle */
1377     if (insert_line)
1378         sprintf(outregstr, "T%c%s", 'A' + tmpreg, swzstr);
1379 }
1380
1381 static void pshader_hw_bem(const struct wined3d_shader_instruction *ins)
1382 {
1383     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1384     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1385     char dst_name[50];
1386     char src_name[2][50];
1387     DWORD sampler_code = dst->reg.idx;
1388
1389     shader_arb_get_dst_param(ins, dst, dst_name);
1390
1391     /* Sampling the perturbation map in Tsrc was done already, including the signedness correction if needed
1392      *
1393      * Keep in mind that src_name[1] can be "TB" and src_name[0] can be "TA" because modifiers like _x2 are valid
1394      * with bem. So delay loading the first parameter until after the perturbation calculation which needs two
1395      * temps is done.
1396      */
1397     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1398     shader_addline(buffer, "SWZ TA, bumpenvmat%d, x, z, 0, 0;\n", sampler_code);
1399     shader_addline(buffer, "DP3 TC.r, TA, %s;\n", src_name[1]);
1400     shader_addline(buffer, "SWZ TA, bumpenvmat%d, y, w, 0, 0;\n", sampler_code);
1401     shader_addline(buffer, "DP3 TC.g, TA, %s;\n", src_name[1]);
1402
1403     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
1404     shader_addline(buffer, "ADD %s, %s, TC;\n", dst_name, src_name[0]);
1405 }
1406
1407 static DWORD negate_modifiers(DWORD mod, char *extra_char)
1408 {
1409     *extra_char = ' ';
1410     switch(mod)
1411     {
1412         case WINED3DSPSM_NONE:      return WINED3DSPSM_NEG;
1413         case WINED3DSPSM_NEG:       return WINED3DSPSM_NONE;
1414         case WINED3DSPSM_BIAS:      return WINED3DSPSM_BIASNEG;
1415         case WINED3DSPSM_BIASNEG:   return WINED3DSPSM_BIAS;
1416         case WINED3DSPSM_SIGN:      return WINED3DSPSM_SIGNNEG;
1417         case WINED3DSPSM_SIGNNEG:   return WINED3DSPSM_SIGN;
1418         case WINED3DSPSM_COMP:      *extra_char = '-'; return WINED3DSPSM_COMP;
1419         case WINED3DSPSM_X2:        return WINED3DSPSM_X2NEG;
1420         case WINED3DSPSM_X2NEG:     return WINED3DSPSM_X2;
1421         case WINED3DSPSM_DZ:        *extra_char = '-'; return WINED3DSPSM_DZ;
1422         case WINED3DSPSM_DW:        *extra_char = '-'; return WINED3DSPSM_DW;
1423         case WINED3DSPSM_ABS:       return WINED3DSPSM_ABSNEG;
1424         case WINED3DSPSM_ABSNEG:    return WINED3DSPSM_ABS;
1425     }
1426     FIXME("Unknown modifier %u\n", mod);
1427     return mod;
1428 }
1429
1430 static void pshader_hw_cnd(const struct wined3d_shader_instruction *ins)
1431 {
1432     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1433     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1434     char dst_name[50];
1435     char src_name[3][50];
1436     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
1437             ins->ctx->reg_maps->shader_version.minor);
1438     BOOL is_color;
1439
1440     shader_arb_get_dst_param(ins, dst, dst_name);
1441     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1442
1443     /* The coissue flag changes the semantic of the cnd instruction in <= 1.3 shaders */
1444     if (shader_version <= WINED3D_SHADER_VERSION(1, 3) && ins->coissue)
1445     {
1446         shader_addline(buffer, "MOV%s %s, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name[1]);
1447     } else {
1448         struct wined3d_shader_src_param src0_copy = ins->src[0];
1449         char extra_neg;
1450
1451         /* src0 may have a negate srcmod set, so we can't blindly add "-" to the name */
1452         src0_copy.modifiers = negate_modifiers(src0_copy.modifiers, &extra_neg);
1453
1454         shader_arb_get_src_param(ins, &src0_copy, 0, src_name[0]);
1455         shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
1456         shader_addline(buffer, "ADD TA, %c%s, coefdiv.x;\n", extra_neg, src_name[0]);
1457         /* No modifiers supported on CMP */
1458         shader_addline(buffer, "CMP %s, TA, %s, %s;\n", dst_name, src_name[1], src_name[2]);
1459
1460         /* _SAT on CMP doesn't make much sense, but it is not a pure NOP */
1461         if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE)
1462         {
1463             shader_arb_get_register_name(ins, &dst->reg, src_name[0], &is_color);
1464             shader_addline(buffer, "MOV_SAT %s, %s;\n", dst_name, dst_name);
1465         }
1466     }
1467 }
1468
1469 static void pshader_hw_cmp(const struct wined3d_shader_instruction *ins)
1470 {
1471     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1472     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1473     char dst_name[50];
1474     char src_name[3][50];
1475     BOOL is_color;
1476
1477     shader_arb_get_dst_param(ins, dst, dst_name);
1478
1479     /* Generate input register names (with modifiers) */
1480     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
1481     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1482     shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
1483
1484     /* No modifiers are supported on CMP */
1485     shader_addline(buffer, "CMP %s, %s, %s, %s;\n", dst_name,
1486                    src_name[0], src_name[2], src_name[1]);
1487
1488     if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE)
1489     {
1490         shader_arb_get_register_name(ins, &dst->reg, src_name[0], &is_color);
1491         shader_addline(buffer, "MOV_SAT %s, %s;\n", dst_name, src_name[0]);
1492     }
1493 }
1494
1495 /** Process the WINED3DSIO_DP2ADD instruction in ARB.
1496  * dst = dot2(src0, src1) + src2 */
1497 static void pshader_hw_dp2add(const struct wined3d_shader_instruction *ins)
1498 {
1499     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1500     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1501     char dst_name[50];
1502     char src_name[3][50];
1503     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
1504
1505     shader_arb_get_dst_param(ins, dst, dst_name);
1506     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
1507     shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
1508
1509     if(ctx->target_version >= NV3)
1510     {
1511         /* GL_NV_fragment_program2 has a 1:1 matching instruction */
1512         shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1513         shader_addline(buffer, "DP2A%s %s, %s, %s, %s;\n", shader_arb_get_modifier(ins),
1514                        dst_name, src_name[0], src_name[1], src_name[2]);
1515     }
1516     else if(ctx->target_version >= NV2)
1517     {
1518         /* dst.x = src2.?, src0.x, src1.x + src0.y * src1.y
1519          * dst.y = src2.?, src0.x, src1.z + src0.y * src1.w
1520          * dst.z = src2.?, src0.x, src1.x + src0.y * src1.y
1521          * dst.z = src2.?, src0.x, src1.z + src0.y * src1.w
1522          *
1523          * Make sure that src1.zw = src1.xy, then we get a classic dp2add
1524          *
1525          * .xyxy and other swizzles that we could get with this are not valid in
1526          * plain ARBfp, but luckily the NV extension grammar lifts this limitation.
1527          */
1528         struct wined3d_shader_src_param tmp_param = ins->src[1];
1529         DWORD swizzle = tmp_param.swizzle & 0xf; /* Selects .xy */
1530         tmp_param.swizzle = swizzle | (swizzle << 4); /* Creates .xyxy */
1531
1532         shader_arb_get_src_param(ins, &tmp_param, 1, src_name[1]);
1533
1534         shader_addline(buffer, "X2D%s %s, %s, %s, %s;\n", shader_arb_get_modifier(ins),
1535                        dst_name, src_name[2], src_name[0], src_name[1]);
1536     }
1537     else
1538     {
1539         shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1540         /* 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]
1541         * src_name[0] can be TA, but TA is a private temp for modifiers, so it is save to overwrite
1542         */
1543         shader_addline(buffer, "MOV TA, %s;\n", src_name[0]);
1544         shader_addline(buffer, "MOV TA.z, 0.0;\n");
1545         shader_addline(buffer, "DP3 TA, TA, %s;\n", src_name[1]);
1546         shader_addline(buffer, "ADD%s %s, TA, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name[2]);
1547     }
1548 }
1549
1550 /* Map the opcode 1-to-1 to the GL code */
1551 static void shader_hw_map2gl(const struct wined3d_shader_instruction *ins)
1552 {
1553     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1554     const char *instruction;
1555     char arguments[256], dst_str[50];
1556     unsigned int i;
1557     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1558
1559     switch (ins->handler_idx)
1560     {
1561         case WINED3DSIH_ABS: instruction = "ABS"; break;
1562         case WINED3DSIH_ADD: instruction = "ADD"; break;
1563         case WINED3DSIH_CRS: instruction = "XPD"; break;
1564         case WINED3DSIH_DP3: instruction = "DP3"; break;
1565         case WINED3DSIH_DP4: instruction = "DP4"; break;
1566         case WINED3DSIH_DST: instruction = "DST"; break;
1567         case WINED3DSIH_FRC: instruction = "FRC"; break;
1568         case WINED3DSIH_LIT: instruction = "LIT"; break;
1569         case WINED3DSIH_LRP: instruction = "LRP"; break;
1570         case WINED3DSIH_MAD: instruction = "MAD"; break;
1571         case WINED3DSIH_MAX: instruction = "MAX"; break;
1572         case WINED3DSIH_MIN: instruction = "MIN"; break;
1573         case WINED3DSIH_MOV: instruction = "MOV"; break;
1574         case WINED3DSIH_MUL: instruction = "MUL"; break;
1575         case WINED3DSIH_SGE: instruction = "SGE"; break;
1576         case WINED3DSIH_SLT: instruction = "SLT"; break;
1577         case WINED3DSIH_SUB: instruction = "SUB"; break;
1578         case WINED3DSIH_MOVA:instruction = "ARR"; break;
1579         case WINED3DSIH_SGN: instruction = "SSG"; break;
1580         case WINED3DSIH_DSX: instruction = "DDX"; break;
1581         default: instruction = "";
1582             FIXME("Unhandled opcode %#x\n", ins->handler_idx);
1583             break;
1584     }
1585
1586     /* Note that shader_arb_add_dst_param() adds spaces. */
1587     arguments[0] = '\0';
1588     shader_arb_get_dst_param(ins, dst, dst_str);
1589     for (i = 0; i < ins->src_count; ++i)
1590     {
1591         char operand[100];
1592         strcat(arguments, ", ");
1593         shader_arb_get_src_param(ins, &ins->src[i], i, operand);
1594         strcat(arguments, operand);
1595     }
1596     shader_addline(buffer, "%s%s %s%s;\n", instruction, shader_arb_get_modifier(ins), dst_str, arguments);
1597 }
1598
1599 static void shader_hw_nop(const struct wined3d_shader_instruction *ins)
1600 {
1601     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1602     shader_addline(buffer, "NOP;\n");
1603 }
1604
1605 static void shader_hw_mov(const struct wined3d_shader_instruction *ins)
1606 {
1607     IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
1608     BOOL pshader = shader_is_pshader_version(shader->baseShader.reg_maps.shader_version.type);
1609     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
1610
1611     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1612     char src0_param[256];
1613
1614     if(ins->handler_idx == WINED3DSIH_MOVA) {
1615         char write_mask[6];
1616
1617         if(ctx->target_version >= NV2) {
1618             shader_hw_map2gl(ins);
1619             return;
1620         }
1621         shader_arb_get_src_param(ins, &ins->src[0], 0, src0_param);
1622         shader_arb_get_write_mask(ins, &ins->dst[0], write_mask);
1623
1624         /* This implements the mova formula used in GLSL. The first two instructions
1625          * prepare the sign() part. Note that it is fine to have my_sign(0.0) = 1.0
1626          * in this case:
1627          * mova A0.x, 0.0
1628          *
1629          * A0.x = arl(floor(abs(0.0) + 0.5) * 1.0) = floor(0.5) = 0.0 since arl does a floor
1630          *
1631          * The ARL is performed when A0 is used - the requested component is read from A0_SHADOW into
1632          * A0.x. We can use the overwritten component of A0_shadow as temporary storage for the sign.
1633          */
1634         shader_addline(buffer, "SGE A0_SHADOW%s, %s, mova_const.y;\n", write_mask, src0_param);
1635         shader_addline(buffer, "MAD A0_SHADOW%s, A0_SHADOW, mova_const.z, -mova_const.w;\n", write_mask);
1636
1637         shader_addline(buffer, "ABS TA%s, %s;\n", write_mask, src0_param);
1638         shader_addline(buffer, "ADD TA%s, TA, mova_const.x;\n", write_mask);
1639         shader_addline(buffer, "FLR TA%s, TA;\n", write_mask);
1640         if (((IWineD3DVertexShaderImpl *)shader)->rel_offset)
1641         {
1642             shader_addline(buffer, "ADD TA%s, TA, helper_const.z;\n", write_mask);
1643         }
1644         shader_addline(buffer, "MUL A0_SHADOW%s, TA, A0_SHADOW;\n", write_mask);
1645
1646         ((struct shader_arb_ctx_priv *)ins->ctx->backend_data)->addr_reg[0] = '\0';
1647     } else if (ins->ctx->reg_maps->shader_version.major == 1
1648           && !shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type)
1649           && ins->dst[0].reg.type == WINED3DSPR_ADDR)
1650     {
1651         src0_param[0] = '\0';
1652         if (((IWineD3DVertexShaderImpl *)shader)->rel_offset)
1653         {
1654             shader_arb_get_src_param(ins, &ins->src[0], 0, src0_param);
1655             shader_addline(buffer, "ADD TA.x, %s, helper_const.z;\n", src0_param);
1656             shader_addline(buffer, "ARL A0.x, TA.x;\n");
1657         }
1658         else
1659         {
1660             /* Apple's ARB_vertex_program implementation does not accept an ARL source argument
1661              * with more than one component. Thus replicate the first source argument over all
1662              * 4 components. For example, .xyzw -> .x (or better: .xxxx), .zwxy -> .z, etc) */
1663             struct wined3d_shader_src_param tmp_src = ins->src[0];
1664             tmp_src.swizzle = (tmp_src.swizzle & 0x3) * 0x55;
1665             shader_arb_get_src_param(ins, &tmp_src, 0, src0_param);
1666             shader_addline(buffer, "ARL A0.x, %s;\n", src0_param);
1667         }
1668     }
1669     else if(ins->dst[0].reg.type == WINED3DSPR_COLOROUT && ins->dst[0].reg.idx == 0 && pshader)
1670     {
1671         IWineD3DPixelShaderImpl *ps = (IWineD3DPixelShaderImpl *) shader;
1672         if(ctx->cur_ps_args->super.srgb_correction && ps->color0_mov)
1673         {
1674             shader_addline(buffer, "#mov handled in srgb write code\n");
1675             return;
1676         }
1677         shader_hw_map2gl(ins);
1678     }
1679     else
1680     {
1681         shader_hw_map2gl(ins);
1682     }
1683 }
1684
1685 static void pshader_hw_texkill(const struct wined3d_shader_instruction *ins)
1686 {
1687     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1688     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1689     char reg_dest[40];
1690
1691     /* No swizzles are allowed in d3d's texkill. PS 1.x ignores the 4th component as documented,
1692      * but >= 2.0 honors it(undocumented, but tested by the d3d9 testsuit)
1693      */
1694     shader_arb_get_dst_param(ins, dst, reg_dest);
1695
1696     if (ins->ctx->reg_maps->shader_version.major >= 2)
1697     {
1698         const char *kilsrc = "TA";
1699         BOOL is_color;
1700
1701         shader_arb_get_register_name(ins, &dst->reg, reg_dest, &is_color);
1702         if(dst->write_mask == WINED3DSP_WRITEMASK_ALL)
1703         {
1704             kilsrc = reg_dest;
1705         }
1706         else
1707         {
1708             /* Sigh. KIL doesn't support swizzles/writemasks. KIL passes a writemask, but ".xy" for example
1709              * is not valid as a swizzle in ARB (needs ".xyyy"). Use SWZ to load the register properly, and set
1710              * masked out components to 0(won't kill)
1711              */
1712             char x = '0', y = '0', z = '0', w = '0';
1713             if(dst->write_mask & WINED3DSP_WRITEMASK_0) x = 'x';
1714             if(dst->write_mask & WINED3DSP_WRITEMASK_1) y = 'y';
1715             if(dst->write_mask & WINED3DSP_WRITEMASK_2) z = 'z';
1716             if(dst->write_mask & WINED3DSP_WRITEMASK_3) w = 'w';
1717             shader_addline(buffer, "SWZ TA, %s, %c, %c, %c, %c;\n", reg_dest, x, y, z, w);
1718         }
1719         shader_addline(buffer, "KIL %s;\n", kilsrc);
1720     } else {
1721         /* ARB fp doesn't like swizzles on the parameter of the KIL instruction. To mask the 4th component,
1722          * copy the register into our general purpose TMP variable, overwrite .w and pass TMP to KIL
1723          *
1724          * ps_1_3 shaders use the texcoord incarnation of the Tx register. ps_1_4 shaders can use the same,
1725          * or pass in any temporary register(in shader phase 2)
1726          */
1727         if(ins->ctx->reg_maps->shader_version.minor <= 3) {
1728             sprintf(reg_dest, "fragment.texcoord[%u]", dst->reg.idx);
1729         } else {
1730             shader_arb_get_dst_param(ins, dst, reg_dest);
1731         }
1732         shader_addline(buffer, "SWZ TA, %s, x, y, z, 1;\n", reg_dest);
1733         shader_addline(buffer, "KIL TA;\n");
1734     }
1735 }
1736
1737 static void pshader_hw_tex(const struct wined3d_shader_instruction *ins)
1738 {
1739     IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)ins->ctx->shader;
1740     IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
1741     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1742     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
1743             ins->ctx->reg_maps->shader_version.minor);
1744     struct wined3d_shader_src_param src;
1745
1746     char reg_dest[40];
1747     char reg_coord[40];
1748     DWORD reg_sampler_code;
1749     DWORD myflags = 0;
1750
1751     /* All versions have a destination register */
1752     shader_arb_get_dst_param(ins, dst, reg_dest);
1753
1754     /* 1.0-1.4: Use destination register number as texture code.
1755        2.0+: Use provided sampler number as texure code. */
1756     if (shader_version < WINED3D_SHADER_VERSION(2,0))
1757         reg_sampler_code = dst->reg.idx;
1758     else
1759         reg_sampler_code = ins->src[1].reg.idx;
1760
1761     /* 1.0-1.3: Use the texcoord varying.
1762        1.4+: Use provided coordinate source register. */
1763     if (shader_version < WINED3D_SHADER_VERSION(1,4))
1764         sprintf(reg_coord, "fragment.texcoord[%u]", reg_sampler_code);
1765     else {
1766         /* TEX is the only instruction that can handle DW and DZ natively */
1767         src = ins->src[0];
1768         if(src.modifiers == WINED3DSPSM_DW) src.modifiers = WINED3DSPSM_NONE;
1769         if(src.modifiers == WINED3DSPSM_DZ) src.modifiers = WINED3DSPSM_NONE;
1770         shader_arb_get_src_param(ins, &src, 0, reg_coord);
1771     }
1772
1773     /* projection flag:
1774      * 1.1, 1.2, 1.3: Use WINED3DTSS_TEXTURETRANSFORMFLAGS
1775      * 1.4: Use WINED3DSPSM_DZ or WINED3DSPSM_DW on src[0]
1776      * 2.0+: Use WINED3DSI_TEXLD_PROJECT on the opcode
1777      */
1778     if (shader_version < WINED3D_SHADER_VERSION(1,4))
1779     {
1780         DWORD flags = 0;
1781         if(reg_sampler_code < MAX_TEXTURES) {
1782             flags = deviceImpl->stateBlock->textureState[reg_sampler_code][WINED3DTSS_TEXTURETRANSFORMFLAGS];
1783         }
1784         if (flags & WINED3DTTFF_PROJECTED) {
1785             myflags |= TEX_PROJ;
1786         }
1787     }
1788     else if (shader_version < WINED3D_SHADER_VERSION(2,0))
1789     {
1790         DWORD src_mod = ins->src[0].modifiers;
1791         if (src_mod == WINED3DSPSM_DZ) {
1792             /* TXP cannot handle DZ natively, so move the z coordinate to .w. reg_coord is a read-only
1793              * varying register, so we need a temp reg
1794              */
1795             shader_addline(ins->ctx->buffer, "SWZ TA, %s, x, y, z, z;\n", reg_coord);
1796             strcpy(reg_coord, "TA");
1797             myflags |= TEX_PROJ;
1798         } else if(src_mod == WINED3DSPSM_DW) {
1799             myflags |= TEX_PROJ;
1800         }
1801     } else {
1802         if (ins->flags & WINED3DSI_TEXLD_PROJECT) myflags |= TEX_PROJ;
1803         if (ins->flags & WINED3DSI_TEXLD_BIAS) myflags |= TEX_BIAS;
1804     }
1805     shader_hw_sample(ins, reg_sampler_code, reg_dest, reg_coord, myflags, NULL, NULL);
1806 }
1807
1808 static void pshader_hw_texcoord(const struct wined3d_shader_instruction *ins)
1809 {
1810     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1811     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1812     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
1813             ins->ctx->reg_maps->shader_version.minor);
1814     char dst_str[50];
1815
1816     if (shader_version < WINED3D_SHADER_VERSION(1,4))
1817     {
1818         DWORD reg = dst->reg.idx;
1819
1820         shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
1821         shader_addline(buffer, "MOV_SAT %s, fragment.texcoord[%u];\n", dst_str, reg);
1822     } else {
1823         char reg_src[40];
1824
1825         shader_arb_get_src_param(ins, &ins->src[0], 0, reg_src);
1826         shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
1827         shader_addline(buffer, "MOV %s, %s;\n", dst_str, reg_src);
1828    }
1829 }
1830
1831 static void pshader_hw_texreg2ar(const struct wined3d_shader_instruction *ins)
1832 {
1833      struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1834      IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)ins->ctx->shader;
1835      IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
1836      DWORD flags;
1837
1838      DWORD reg1 = ins->dst[0].reg.idx;
1839      char dst_str[50];
1840      char src_str[50];
1841
1842      /* Note that texreg2ar treats Tx as a temporary register, not as a varying */
1843      shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
1844      shader_arb_get_src_param(ins, &ins->src[0], 0, src_str);
1845      /* Move .x first in case src_str is "TA" */
1846      shader_addline(buffer, "MOV TA.y, %s.x;\n", src_str);
1847      shader_addline(buffer, "MOV TA.x, %s.w;\n", src_str);
1848      flags = reg1 < MAX_TEXTURES ? deviceImpl->stateBlock->textureState[reg1][WINED3DTSS_TEXTURETRANSFORMFLAGS] : 0;
1849      shader_hw_sample(ins, reg1, dst_str, "TA", flags & WINED3DTTFF_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
1850 }
1851
1852 static void pshader_hw_texreg2gb(const struct wined3d_shader_instruction *ins)
1853 {
1854      struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1855
1856      DWORD reg1 = ins->dst[0].reg.idx;
1857      char dst_str[50];
1858      char src_str[50];
1859
1860      /* Note that texreg2gb treats Tx as a temporary register, not as a varying */
1861      shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
1862      shader_arb_get_src_param(ins, &ins->src[0], 0, src_str);
1863      shader_addline(buffer, "MOV TA.x, %s.y;\n", src_str);
1864      shader_addline(buffer, "MOV TA.y, %s.z;\n", src_str);
1865      shader_hw_sample(ins, reg1, dst_str, "TA", 0, NULL, NULL);
1866 }
1867
1868 static void pshader_hw_texreg2rgb(const struct wined3d_shader_instruction *ins)
1869 {
1870     DWORD reg1 = ins->dst[0].reg.idx;
1871     char dst_str[50];
1872     char src_str[50];
1873
1874     /* Note that texreg2rg treats Tx as a temporary register, not as a varying */
1875     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
1876     shader_arb_get_src_param(ins, &ins->src[0], 0, src_str);
1877     shader_hw_sample(ins, reg1, dst_str, src_str, 0, NULL, NULL);
1878 }
1879
1880 static void pshader_hw_texbem(const struct wined3d_shader_instruction *ins)
1881 {
1882     IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)ins->ctx->shader;
1883     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1884     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1885     char reg_coord[40], dst_reg[50], src_reg[50];
1886     DWORD reg_dest_code;
1887
1888     /* All versions have a destination register. The Tx where the texture coordinates come
1889      * from is the varying incarnation of the texture register
1890      */
1891     reg_dest_code = dst->reg.idx;
1892     shader_arb_get_dst_param(ins, &ins->dst[0], dst_reg);
1893     shader_arb_get_src_param(ins, &ins->src[0], 0, src_reg);
1894     sprintf(reg_coord, "fragment.texcoord[%u]", reg_dest_code);
1895
1896     /* Sampling the perturbation map in Tsrc was done already, including the signedness correction if needed
1897      * The Tx in which the perturbation map is stored is the tempreg incarnation of the texture register
1898      *
1899      * GL_NV_fragment_program_option could handle this in one instruction via X2D:
1900      * X2D TA.xy, fragment.texcoord, T%u, bumpenvmat%u.xzyw
1901      *
1902      * However, the NV extensions are never enabled for <= 2.0 shaders because of the performance penalty that
1903      * comes with it, and texbem is an 1.x only instruction. No 1.x instruction forces us to enable the NV
1904      * extension.
1905      */
1906     shader_addline(buffer, "SWZ TB, bumpenvmat%d, x, z, 0, 0;\n", reg_dest_code);
1907     shader_addline(buffer, "DP3 TA.x, TB, %s;\n", src_reg);
1908     shader_addline(buffer, "SWZ TB, bumpenvmat%d, y, w, 0, 0;\n", reg_dest_code);
1909     shader_addline(buffer, "DP3 TA.y, TB, %s;\n", src_reg);
1910
1911     /* with projective textures, texbem only divides the static texture coord, not the displacement,
1912      * so we can't let the GL handle this.
1913      */
1914     if (((IWineD3DDeviceImpl*) This->baseShader.device)->stateBlock->textureState[reg_dest_code][WINED3DTSS_TEXTURETRANSFORMFLAGS]
1915             & WINED3DTTFF_PROJECTED) {
1916         shader_addline(buffer, "RCP TB.w, %s.w;\n", reg_coord);
1917         shader_addline(buffer, "MUL TB.xy, %s, TB.w;\n", reg_coord);
1918         shader_addline(buffer, "ADD TA.xy, TA, TB;\n");
1919     } else {
1920         shader_addline(buffer, "ADD TA.xy, TA, %s;\n", reg_coord);
1921     }
1922
1923     shader_hw_sample(ins, reg_dest_code, dst_reg, "TA", 0, NULL, NULL);
1924
1925     if (ins->handler_idx == WINED3DSIH_TEXBEML)
1926     {
1927         /* No src swizzles are allowed, so this is ok */
1928         shader_addline(buffer, "MAD TA, %s.z, luminance%d.x, luminance%d.y;\n",
1929                        src_reg, reg_dest_code, reg_dest_code);
1930         shader_addline(buffer, "MUL %s, %s, TA;\n", dst_reg, dst_reg);
1931     }
1932 }
1933
1934 static void pshader_hw_texm3x2pad(const struct wined3d_shader_instruction *ins)
1935 {
1936     DWORD reg = ins->dst[0].reg.idx;
1937     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1938     char src0_name[50], dst_name[50];
1939     BOOL is_color;
1940     struct wined3d_shader_register tmp_reg = ins->dst[0].reg;
1941
1942     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
1943     /* The next instruction will be a texm3x2tex or texm3x2depth that writes to the uninitialized
1944      * T<reg+1> register. Use this register to store the calculated vector
1945      */
1946     tmp_reg.idx = reg + 1;
1947     shader_arb_get_register_name(ins, &tmp_reg, dst_name, &is_color);
1948     shader_addline(buffer, "DP3 %s.x, fragment.texcoord[%u], %s;\n", dst_name, reg, src0_name);
1949 }
1950
1951 static void pshader_hw_texm3x2tex(const struct wined3d_shader_instruction *ins)
1952 {
1953     IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)ins->ctx->shader;
1954     IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
1955     DWORD flags;
1956     DWORD reg = ins->dst[0].reg.idx;
1957     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1958     char dst_str[50];
1959     char src0_name[50];
1960     char dst_reg[50];
1961     BOOL is_color;
1962
1963     /* We know that we're writing to the uninitialized T<reg> register, so use it for temporary storage */
1964     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_reg, &is_color);
1965
1966     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
1967     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
1968     shader_addline(buffer, "DP3 %s.y, fragment.texcoord[%u], %s;\n", dst_reg, reg, src0_name);
1969     flags = reg < MAX_TEXTURES ? deviceImpl->stateBlock->textureState[reg][WINED3DTSS_TEXTURETRANSFORMFLAGS] : 0;
1970     shader_hw_sample(ins, reg, dst_str, dst_reg, flags & WINED3DTTFF_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
1971 }
1972
1973 static void pshader_hw_texm3x3pad(const struct wined3d_shader_instruction *ins)
1974 {
1975     IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)ins->ctx->shader;
1976     DWORD reg = ins->dst[0].reg.idx;
1977     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1978     SHADER_PARSE_STATE* current_state = &This->baseShader.parse_state;
1979     char src0_name[50], dst_name[50];
1980     struct wined3d_shader_register tmp_reg = ins->dst[0].reg;
1981     BOOL is_color;
1982
1983     /* There are always 2 texm3x3pad instructions followed by one texm3x3[tex,vspec, ...] instruction, with
1984      * incrementing ins->dst[0].register_idx numbers. So the pad instruction already knows the final destination
1985      * register, and this register is uninitialized(otherwise the assembler complains that it is 'redeclared')
1986      */
1987     tmp_reg.idx = reg + 2 - current_state->current_row;
1988     shader_arb_get_register_name(ins, &tmp_reg, dst_name, &is_color);
1989
1990     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
1991     shader_addline(buffer, "DP3 %s.%c, fragment.texcoord[%u], %s;\n",
1992                    dst_name, 'x' + current_state->current_row, reg, src0_name);
1993     current_state->texcoord_w[current_state->current_row++] = reg;
1994 }
1995
1996 static void pshader_hw_texm3x3tex(const struct wined3d_shader_instruction *ins)
1997 {
1998     IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)ins->ctx->shader;
1999     IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
2000     DWORD flags;
2001     DWORD reg = ins->dst[0].reg.idx;
2002     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2003     SHADER_PARSE_STATE* current_state = &This->baseShader.parse_state;
2004     char dst_str[50];
2005     char src0_name[50], dst_name[50];
2006     BOOL is_color;
2007
2008     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2009     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2010     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_name, reg, src0_name);
2011
2012     /* Sample the texture using the calculated coordinates */
2013     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2014     flags = reg < MAX_TEXTURES ? deviceImpl->stateBlock->textureState[reg][WINED3DTSS_TEXTURETRANSFORMFLAGS] : 0;
2015     shader_hw_sample(ins, reg, dst_str, dst_name, flags & WINED3DTTFF_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2016     current_state->current_row = 0;
2017 }
2018
2019 static void pshader_hw_texm3x3vspec(const struct wined3d_shader_instruction *ins)
2020 {
2021     IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)ins->ctx->shader;
2022     IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
2023     DWORD flags;
2024     DWORD reg = ins->dst[0].reg.idx;
2025     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2026     SHADER_PARSE_STATE* current_state = &This->baseShader.parse_state;
2027     char dst_str[50];
2028     char src0_name[50];
2029     char dst_reg[50];
2030     BOOL is_color;
2031
2032     /* Get the dst reg without writemask strings. We know this register is uninitialized, so we can use all
2033      * components for temporary data storage
2034      */
2035     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_reg, &is_color);
2036     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2037     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_reg, reg, src0_name);
2038
2039     /* Construct the eye-ray vector from w coordinates */
2040     shader_addline(buffer, "MOV TB.x, fragment.texcoord[%u].w;\n", current_state->texcoord_w[0]);
2041     shader_addline(buffer, "MOV TB.y, fragment.texcoord[%u].w;\n", current_state->texcoord_w[1]);
2042     shader_addline(buffer, "MOV TB.z, fragment.texcoord[%u].w;\n", reg);
2043
2044     /* Calculate reflection vector
2045      */
2046     shader_addline(buffer, "DP3 %s.w, %s, TB;\n", dst_reg, dst_reg);
2047     /* The .w is ignored when sampling, so I can use TB.w to calculate dot(N, N) */
2048     shader_addline(buffer, "DP3 TB.w, %s, %s;\n", dst_reg, dst_reg);
2049     shader_addline(buffer, "RCP TB.w, TB.w;\n");
2050     shader_addline(buffer, "MUL %s.w, %s.w, TB.w;\n", dst_reg, dst_reg);
2051     shader_addline(buffer, "MUL %s, %s.w, %s;\n", dst_reg, dst_reg, dst_reg);
2052     shader_addline(buffer, "MAD %s, coefmul.x, %s, -TB;\n", dst_reg, dst_reg);
2053
2054     /* Sample the texture using the calculated coordinates */
2055     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2056     flags = reg < MAX_TEXTURES ? deviceImpl->stateBlock->textureState[reg][WINED3DTSS_TEXTURETRANSFORMFLAGS] : 0;
2057     shader_hw_sample(ins, reg, dst_str, dst_reg, flags & WINED3DTTFF_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2058     current_state->current_row = 0;
2059 }
2060
2061 static void pshader_hw_texm3x3spec(const struct wined3d_shader_instruction *ins)
2062 {
2063     IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)ins->ctx->shader;
2064     IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
2065     DWORD flags;
2066     DWORD reg = ins->dst[0].reg.idx;
2067     SHADER_PARSE_STATE* current_state = &This->baseShader.parse_state;
2068     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2069     char dst_str[50];
2070     char src0_name[50];
2071     char src1_name[50];
2072     char dst_reg[50];
2073     BOOL is_color;
2074
2075     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2076     shader_arb_get_src_param(ins, &ins->src[0], 1, src1_name);
2077     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_reg, &is_color);
2078     /* Note: dst_reg.xy is input here, generated by two texm3x3pad instructions */
2079     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_reg, reg, src0_name);
2080
2081     /* Calculate reflection vector.
2082      *
2083      *                   dot(N, E)
2084      * dst_reg.xyz = 2 * --------- * N - E
2085      *                   dot(N, N)
2086      *
2087      * Which normalizes the normal vector
2088      */
2089     shader_addline(buffer, "DP3 %s.w, %s, %s;\n", dst_reg, dst_reg, src1_name);
2090     shader_addline(buffer, "DP3 TC.w, %s, %s;\n", dst_reg, dst_reg);
2091     shader_addline(buffer, "RCP TC.w, TC.w;\n");
2092     shader_addline(buffer, "MUL %s.w, %s.w, TC.w;\n", dst_reg, dst_reg);
2093     shader_addline(buffer, "MUL %s, %s.w, %s;\n", dst_reg, dst_reg, dst_reg);
2094     shader_addline(buffer, "MAD %s, coefmul.x, %s, -%s;\n", dst_reg, dst_reg, src1_name);
2095
2096     /* Sample the texture using the calculated coordinates */
2097     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2098     flags = reg < MAX_TEXTURES ? deviceImpl->stateBlock->textureState[reg][WINED3DTSS_TEXTURETRANSFORMFLAGS] : 0;
2099     shader_hw_sample(ins, reg, dst_str, dst_reg, flags & WINED3DTTFF_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2100     current_state->current_row = 0;
2101 }
2102
2103 static void pshader_hw_texdepth(const struct wined3d_shader_instruction *ins)
2104 {
2105     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2106     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2107     char dst_name[50];
2108
2109     /* texdepth has an implicit destination, the fragment depth value. It's only parameter,
2110      * which is essentially an input, is the destination register because it is the first
2111      * parameter. According to the msdn, this must be register r5, but let's keep it more flexible
2112      * here(writemasks/swizzles are not valid on texdepth)
2113      */
2114     shader_arb_get_dst_param(ins, dst, dst_name);
2115
2116     /* According to the msdn, the source register(must be r5) is unusable after
2117      * the texdepth instruction, so we're free to modify it
2118      */
2119     shader_addline(buffer, "MIN %s.y, %s.y, one.y;\n", dst_name, dst_name);
2120
2121     /* How to deal with the special case dst_name.g == 0? if r != 0, then
2122      * the r * (1 / 0) will give infinity, which is clamped to 1.0, the correct
2123      * result. But if r = 0.0, then 0 * inf = 0, which is incorrect.
2124      */
2125     shader_addline(buffer, "RCP %s.y, %s.y;\n", dst_name, dst_name);
2126     shader_addline(buffer, "MUL TA.x, %s.x, %s.y;\n", dst_name, dst_name);
2127     shader_addline(buffer, "MIN TA.x, TA.x, one.x;\n");
2128     shader_addline(buffer, "MAX result.depth, TA.x, 0.0;\n");
2129 }
2130
2131 /** Process the WINED3DSIO_TEXDP3TEX instruction in ARB:
2132  * Take a 3-component dot product of the TexCoord[dstreg] and src,
2133  * then perform a 1D texture lookup from stage dstregnum, place into dst. */
2134 static void pshader_hw_texdp3tex(const struct wined3d_shader_instruction *ins)
2135 {
2136     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2137     DWORD sampler_idx = ins->dst[0].reg.idx;
2138     char src0[50];
2139     char dst_str[50];
2140
2141     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2142     shader_addline(buffer, "MOV TB, 0.0;\n");
2143     shader_addline(buffer, "DP3 TB.x, fragment.texcoord[%u], %s;\n", sampler_idx, src0);
2144
2145     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2146     shader_hw_sample(ins, sampler_idx, dst_str, "TB", 0 /* Only one coord, can't be projected */, NULL, NULL);
2147 }
2148
2149 /** Process the WINED3DSIO_TEXDP3 instruction in ARB:
2150  * Take a 3-component dot product of the TexCoord[dstreg] and src. */
2151 static void pshader_hw_texdp3(const struct wined3d_shader_instruction *ins)
2152 {
2153     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2154     char src0[50];
2155     char dst_str[50];
2156     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2157
2158     /* Handle output register */
2159     shader_arb_get_dst_param(ins, dst, dst_str);
2160     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2161     shader_addline(buffer, "DP3 %s, fragment.texcoord[%u], %s;\n", dst_str, dst->reg.idx, src0);
2162 }
2163
2164 /** Process the WINED3DSIO_TEXM3X3 instruction in ARB
2165  * Perform the 3rd row of a 3x3 matrix multiply */
2166 static void pshader_hw_texm3x3(const struct wined3d_shader_instruction *ins)
2167 {
2168     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2169     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2170     char dst_str[50], dst_name[50];
2171     char src0[50];
2172     BOOL is_color;
2173
2174     shader_arb_get_dst_param(ins, dst, dst_str);
2175     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2176     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2177     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_name, dst->reg.idx, src0);
2178     shader_addline(buffer, "MOV %s, %s;\n", dst_str, dst_name);
2179 }
2180
2181 /** Process the WINED3DSIO_TEXM3X2DEPTH instruction in ARB:
2182  * Last row of a 3x2 matrix multiply, use the result to calculate the depth:
2183  * Calculate tmp0.y = TexCoord[dstreg] . src.xyz;  (tmp0.x has already been calculated)
2184  * depth = (tmp0.y == 0.0) ? 1.0 : tmp0.x / tmp0.y
2185  */
2186 static void pshader_hw_texm3x2depth(const struct wined3d_shader_instruction *ins)
2187 {
2188     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2189     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2190     char src0[50], dst_name[50];
2191     BOOL is_color;
2192
2193     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2194     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2195     shader_addline(buffer, "DP3 %s.y, fragment.texcoord[%u], %s;\n", dst_name, dst->reg.idx, src0);
2196
2197     /* How to deal with the special case dst_name.g == 0? if r != 0, then
2198      * the r * (1 / 0) will give infinity, which is clamped to 1.0, the correct
2199      * result. But if r = 0.0, then 0 * inf = 0, which is incorrect.
2200      */
2201     shader_addline(buffer, "RCP %s.y, %s.y;\n", dst_name, dst_name);
2202     shader_addline(buffer, "MUL %s.x, %s.x, %s.y;\n", dst_name, dst_name, dst_name);
2203     shader_addline(buffer, "MIN %s.x, %s.x, one.x;\n", dst_name, dst_name);
2204     shader_addline(buffer, "MAX result.depth, %s.x, 0.0;\n", dst_name);
2205 }
2206
2207 /** Handles transforming all WINED3DSIO_M?x? opcodes for
2208     Vertex/Pixel shaders to ARB_vertex_program codes */
2209 static void shader_hw_mnxn(const struct wined3d_shader_instruction *ins)
2210 {
2211     int i;
2212     int nComponents = 0;
2213     struct wined3d_shader_dst_param tmp_dst = {{0}};
2214     struct wined3d_shader_src_param tmp_src[2] = {{{0}}};
2215     struct wined3d_shader_instruction tmp_ins;
2216
2217     memset(&tmp_ins, 0, sizeof(tmp_ins));
2218
2219     /* Set constants for the temporary argument */
2220     tmp_ins.ctx = ins->ctx;
2221     tmp_ins.dst_count = 1;
2222     tmp_ins.dst = &tmp_dst;
2223     tmp_ins.src_count = 2;
2224     tmp_ins.src = tmp_src;
2225
2226     switch(ins->handler_idx)
2227     {
2228         case WINED3DSIH_M4x4:
2229             nComponents = 4;
2230             tmp_ins.handler_idx = WINED3DSIH_DP4;
2231             break;
2232         case WINED3DSIH_M4x3:
2233             nComponents = 3;
2234             tmp_ins.handler_idx = WINED3DSIH_DP4;
2235             break;
2236         case WINED3DSIH_M3x4:
2237             nComponents = 4;
2238             tmp_ins.handler_idx = WINED3DSIH_DP3;
2239             break;
2240         case WINED3DSIH_M3x3:
2241             nComponents = 3;
2242             tmp_ins.handler_idx = WINED3DSIH_DP3;
2243             break;
2244         case WINED3DSIH_M3x2:
2245             nComponents = 2;
2246             tmp_ins.handler_idx = WINED3DSIH_DP3;
2247             break;
2248         default:
2249             FIXME("Unhandled opcode %#x\n", ins->handler_idx);
2250             break;
2251     }
2252
2253     tmp_dst = ins->dst[0];
2254     tmp_src[0] = ins->src[0];
2255     tmp_src[1] = ins->src[1];
2256     for (i = 0; i < nComponents; i++) {
2257         tmp_dst.write_mask = WINED3DSP_WRITEMASK_0 << i;
2258         shader_hw_map2gl(&tmp_ins);
2259         ++tmp_src[1].reg.idx;
2260     }
2261 }
2262
2263 static void shader_hw_scalar_op(const struct wined3d_shader_instruction *ins)
2264 {
2265     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2266     const char *instruction;
2267
2268     char dst[50];
2269     char src[50];
2270
2271     switch(ins->handler_idx)
2272     {
2273         case WINED3DSIH_RSQ:  instruction = "RSQ"; break;
2274         case WINED3DSIH_RCP:  instruction = "RCP"; break;
2275         case WINED3DSIH_EXP:  instruction = "EX2"; break;
2276         case WINED3DSIH_EXPP: instruction = "EXP"; break;
2277         default: instruction = "";
2278             FIXME("Unhandled opcode %#x\n", ins->handler_idx);
2279             break;
2280     }
2281
2282     shader_arb_get_dst_param(ins, &ins->dst[0], dst); /* Destination */
2283     shader_arb_get_src_param(ins, &ins->src[0], 0, src);
2284     if (ins->src[0].swizzle == WINED3DSP_NOSWIZZLE)
2285     {
2286         /* Dx sdk says .x is used if no swizzle is given, but our test shows that
2287          * .w is used
2288          */
2289         strcat(src, ".w");
2290     }
2291
2292     shader_addline(buffer, "%s%s %s, %s;\n", instruction, shader_arb_get_modifier(ins), dst, src);
2293 }
2294
2295 static void shader_hw_nrm(const struct wined3d_shader_instruction *ins)
2296 {
2297     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2298     char dst_name[50];
2299     char src_name[50];
2300     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2301     BOOL pshader = shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type);
2302
2303     shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2304     shader_arb_get_src_param(ins, &ins->src[0], 1 /* Use TB */, src_name);
2305
2306     if(pshader && priv->target_version >= NV3)
2307     {
2308         shader_addline(buffer, "NRM%s %s, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name);
2309     }
2310     else
2311     {
2312         shader_addline(buffer, "DP3 TA, %s, %s;\n", src_name, src_name);
2313         shader_addline(buffer, "RSQ TA, TA.x;\n");
2314         /* dst.w = src[0].w * 1 / (src.x^2 + src.y^2 + src.z^2)^(1/2) according to msdn*/
2315         shader_addline(buffer, "MUL%s %s, %s, TA;\n", shader_arb_get_modifier(ins), dst_name,
2316                     src_name);
2317     }
2318 }
2319
2320 static void shader_hw_lrp(const struct wined3d_shader_instruction *ins)
2321 {
2322     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2323     char dst_name[50];
2324     char src_name[3][50];
2325
2326     /* ARB_fragment_program has a convenient LRP instruction */
2327     if(shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type)) {
2328         shader_hw_map2gl(ins);
2329         return;
2330     }
2331
2332     shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2333     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
2334     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
2335     shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
2336
2337     shader_addline(buffer, "SUB TA, %s, %s;\n", src_name[1], src_name[2]);
2338     shader_addline(buffer, "MAD%s %s, %s, TA, %s;\n", shader_arb_get_modifier(ins),
2339                    dst_name, src_name[0], src_name[2]);
2340 }
2341
2342 static void shader_hw_sincos(const struct wined3d_shader_instruction *ins)
2343 {
2344     /* This instruction exists in ARB, but the d3d instruction takes two extra parameters which
2345      * must contain fixed constants. So we need a separate function to filter those constants and
2346      * can't use map2gl
2347      */
2348     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2349     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2350     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2351     char dst_name[50];
2352     char src_name0[50], src_name1[50], src_name2[50];
2353     BOOL is_color;
2354
2355     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name0);
2356     if(shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type)) {
2357         shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2358         /* No modifiers are supported on SCS */
2359         shader_addline(buffer, "SCS %s, %s;\n", dst_name, src_name0);
2360
2361         if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE)
2362         {
2363             shader_arb_get_register_name(ins, &dst->reg, src_name0, &is_color);
2364             shader_addline(buffer, "MOV_SAT %s, %s;\n", dst_name, src_name0);
2365         }
2366     } else if(priv->target_version >= NV2) {
2367         shader_arb_get_register_name(ins, &dst->reg, dst_name, &is_color);
2368
2369         /* Sincos writemask must be .x, .y or .xy */
2370         if(dst->write_mask & WINED3DSP_WRITEMASK_0)
2371             shader_addline(buffer, "COS%s %s.x, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name0);
2372         if(dst->write_mask & WINED3DSP_WRITEMASK_1)
2373             shader_addline(buffer, "SIN%s %s.y, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name0);
2374     } else {
2375         /* Approximate sine and cosine with a taylor series, as per math textbook. The application passes 8
2376          * helper constants(D3DSINCOSCONST1 and D3DSINCOSCONST2) in src1 and src2.
2377          *
2378          * sin(x) = x - x^3/3! + x^5/5! - x^7/7! + ...
2379          * cos(x) = 1 - x^2/2! + x^4/4! - x^6/6! + ...
2380          *
2381          * The constants we get are:
2382          *
2383          *  +1   +1,     -1     -1     +1      +1      -1       -1
2384          *      ---- ,  ---- , ---- , ----- , ----- , ----- , ------
2385          *      1!*2    2!*4   3!*8   4!*16   5!*32   6!*64   7!*128
2386          *
2387          * If used with x^2, x^3, x^4 etc they calculate sin(x/2) and cos(x/2):
2388          *
2389          * (x/2)^2 = x^2 / 4
2390          * (x/2)^3 = x^3 / 8
2391          * (x/2)^4 = x^4 / 16
2392          * (x/2)^5 = x^5 / 32
2393          * etc
2394          *
2395          * To get the final result:
2396          * sin(x) = 2 * sin(x/2) * cos(x/2)
2397          * cos(x) = cos(x/2)^2 - sin(x/2)^2
2398          * (from sin(x+y) and cos(x+y) rules)
2399          *
2400          * As per MSDN, dst.z is undefined after the operation, and so is
2401          * dst.x and dst.y if they're masked out by the writemask. Ie
2402          * sincos dst.y, src1, c0, c1
2403          * returns the sine in dst.y. dst.x and dst.z are undefined, dst.w is not touched. The assembler
2404          * vsa.exe also stops with an error if the dest register is the same register as the source
2405          * register. This means we can use dest.xyz as temporary storage. The assembler vsa.exe output also
2406          * indicates that sincos consumes 8 instruction slots in vs_2_0(and, strangely, in vs_3_0).
2407          */
2408         shader_arb_get_src_param(ins, &ins->src[1], 1, src_name1);
2409         shader_arb_get_src_param(ins, &ins->src[2], 2, src_name2);
2410         shader_arb_get_register_name(ins, &dst->reg, dst_name, &is_color);
2411
2412         shader_addline(buffer, "MUL %s.x, %s, %s;\n", dst_name, src_name0, src_name0);  /* x ^ 2 */
2413         shader_addline(buffer, "MUL TA.y, %s.x, %s;\n", dst_name, src_name0);           /* x ^ 3 */
2414         shader_addline(buffer, "MUL %s.y, TA.y, %s;\n", dst_name, src_name0);           /* x ^ 4 */
2415         shader_addline(buffer, "MUL TA.z, %s.y, %s;\n", dst_name, src_name0);           /* x ^ 5 */
2416         shader_addline(buffer, "MUL %s.z, TA.z, %s;\n", dst_name, src_name0);           /* x ^ 6 */
2417         shader_addline(buffer, "MUL TA.w, %s.z, %s;\n", dst_name, src_name0);           /* x ^ 7 */
2418
2419         /* sin(x/2)
2420          *
2421          * Unfortunately we don't get the constants in a DP4-capable form. Is there a way to
2422          * properly merge that with MULs in the code above?
2423          * The swizzles .yz and xw however fit into the .yzxw swizzle added to ps_2_0. Maybe
2424          * we can merge the sine and cosine MAD rows to calculate them together.
2425          */
2426         shader_addline(buffer, "MUL TA.x, %s, %s.w;\n", src_name0, src_name2); /* x^1, +1/(1!*2) */
2427         shader_addline(buffer, "MAD TA.x, TA.y, %s.x, TA.x;\n", src_name2); /* -1/(3!*8) */
2428         shader_addline(buffer, "MAD TA.x, TA.z, %s.w, TA.x;\n", src_name1); /* +1/(5!*32) */
2429         shader_addline(buffer, "MAD TA.x, TA.w, %s.x, TA.x;\n", src_name1); /* -1/(7!*128) */
2430
2431         /* cos(x/2) */
2432         shader_addline(buffer, "MAD TA.y, %s.x, %s.y, %s.z;\n", dst_name, src_name2, src_name2); /* -1/(2!*4), +1.0 */
2433         shader_addline(buffer, "MAD TA.y, %s.y, %s.z, TA.y;\n", dst_name, src_name1); /* +1/(4!*16) */
2434         shader_addline(buffer, "MAD TA.y, %s.z, %s.y, TA.y;\n", dst_name, src_name1); /* -1/(6!*64) */
2435
2436         if(dst->write_mask & WINED3DSP_WRITEMASK_0) {
2437             /* cos x */
2438             shader_addline(buffer, "MUL TA.z, TA.y, TA.y;\n");
2439             shader_addline(buffer, "MAD %s.x, -TA.x, TA.x, TA.z;\n", dst_name);
2440         }
2441         if(dst->write_mask & WINED3DSP_WRITEMASK_1) {
2442             /* sin x */
2443             shader_addline(buffer, "MUL %s.y, TA.x, TA.y;\n", dst_name);
2444             shader_addline(buffer, "ADD %s.y, %s.y, %s.y;\n", dst_name, dst_name, dst_name);
2445         }
2446     }
2447 }
2448
2449 static void shader_hw_sgn(const struct wined3d_shader_instruction *ins)
2450 {
2451     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2452     char dst_name[50];
2453     char src_name[50];
2454     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
2455
2456     /* SGN is only valid in vertex shaders */
2457     if(ctx->target_version == NV2) {
2458         shader_hw_map2gl(ins);
2459         return;
2460     }
2461     shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2462     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name);
2463
2464     /* If SRC > 0.0, -SRC < SRC = TRUE, otherwise false.
2465      * if SRC < 0.0,  SRC < -SRC = TRUE. If neither is true, src = 0.0
2466      */
2467     if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE) {
2468         shader_addline(buffer, "SLT %s, -%s, %s;\n", dst_name, src_name, src_name);
2469     } else {
2470         /* src contains TA? Write to the dest first. This won't overwrite our destination.
2471          * Then use TA, and calculate the final result
2472          *
2473          * Not reading from TA? Store the first result in TA to avoid overwriting the
2474          * destination if src reg = dst reg
2475          */
2476         if(strstr(src_name, "TA"))
2477         {
2478             shader_addline(buffer, "SLT %s,  %s, -%s;\n", dst_name, src_name, src_name);
2479             shader_addline(buffer, "SLT TA, -%s, %s;\n", src_name, src_name);
2480             shader_addline(buffer, "ADD %s, %s, -TA;\n", dst_name, dst_name);
2481         }
2482         else
2483         {
2484             shader_addline(buffer, "SLT TA, -%s, %s;\n", src_name, src_name);
2485             shader_addline(buffer, "SLT %s,  %s, -%s;\n", dst_name, src_name, src_name);
2486             shader_addline(buffer, "ADD %s, TA, -%s;\n", dst_name, dst_name);
2487         }
2488     }
2489 }
2490
2491 static void shader_hw_dsy(const struct wined3d_shader_instruction *ins)
2492 {
2493     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2494     char src[50];
2495     char dst[50];
2496     char dst_name[50];
2497     BOOL is_color;
2498
2499     shader_arb_get_dst_param(ins, &ins->dst[0], dst);
2500     shader_arb_get_src_param(ins, &ins->src[0], 0, src);
2501     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2502
2503     shader_addline(buffer, "DDY %s, %s;\n", dst, src);
2504     shader_addline(buffer, "MUL%s %s, %s, ycorrection.y;\n", shader_arb_get_modifier(ins), dst, dst_name);
2505 }
2506
2507 static DWORD abs_modifier(DWORD mod, BOOL *need_abs)
2508 {
2509     *need_abs = FALSE;
2510
2511     switch(mod)
2512     {
2513         case WINED3DSPSM_NONE:      return WINED3DSPSM_ABS;
2514         case WINED3DSPSM_NEG:       return WINED3DSPSM_ABS;
2515         case WINED3DSPSM_BIAS:      *need_abs = TRUE; return WINED3DSPSM_BIAS;
2516         case WINED3DSPSM_BIASNEG:   *need_abs = TRUE; return WINED3DSPSM_BIASNEG;
2517         case WINED3DSPSM_SIGN:      *need_abs = TRUE; return WINED3DSPSM_SIGN;
2518         case WINED3DSPSM_SIGNNEG:   *need_abs = TRUE; return WINED3DSPSM_SIGNNEG;
2519         case WINED3DSPSM_COMP:      *need_abs = TRUE; return WINED3DSPSM_COMP;
2520         case WINED3DSPSM_X2:        *need_abs = TRUE; return WINED3DSPSM_X2;
2521         case WINED3DSPSM_X2NEG:     *need_abs = TRUE; return WINED3DSPSM_X2NEG;
2522         case WINED3DSPSM_DZ:        *need_abs = TRUE; return WINED3DSPSM_DZ;
2523         case WINED3DSPSM_DW:        *need_abs = TRUE; return WINED3DSPSM_DW;
2524         case WINED3DSPSM_ABS:       return WINED3DSPSM_ABS;
2525         case WINED3DSPSM_ABSNEG:    return WINED3DSPSM_ABS;
2526     }
2527     FIXME("Unknown modifier %u\n", mod);
2528     return mod;
2529 }
2530
2531 static void shader_hw_log_pow(const struct wined3d_shader_instruction *ins)
2532 {
2533     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2534     char src0[50], src1[50], dst[50];
2535     struct wined3d_shader_src_param src0_copy = ins->src[0];
2536     BOOL need_abs = FALSE;
2537     const char *instr;
2538     BOOL arg2 = FALSE;
2539
2540     switch(ins->handler_idx)
2541     {
2542         case WINED3DSIH_LOG:  instr = "LG2"; break;
2543         case WINED3DSIH_LOGP: instr = "LOG"; break;
2544         case WINED3DSIH_POW:  instr = "POW"; arg2 = TRUE; break;
2545         default:
2546             ERR("Unexpected instruction %d\n", ins->handler_idx);
2547             return;
2548     }
2549
2550     /* LOG, LOGP and POW operate on the absolute value of the input */
2551     src0_copy.modifiers = abs_modifier(src0_copy.modifiers, &need_abs);
2552
2553     shader_arb_get_dst_param(ins, &ins->dst[0], dst);
2554     shader_arb_get_src_param(ins, &src0_copy, 0, src0);
2555     if(arg2) shader_arb_get_src_param(ins, &ins->src[1], 1, src1);
2556
2557     if(need_abs)
2558     {
2559         shader_addline(buffer, "ABS TA, %s;\n", src0);
2560         if(arg2)
2561         {
2562             shader_addline(buffer, "%s%s %s, TA, %s;\n", instr, shader_arb_get_modifier(ins), dst, src1);
2563         }
2564         else
2565         {
2566             shader_addline(buffer, "%s%s %s, TA;\n", instr, shader_arb_get_modifier(ins), dst);
2567         }
2568     }
2569     else if(arg2)
2570     {
2571         shader_addline(buffer, "%s%s %s, %s, %s;\n", instr, shader_arb_get_modifier(ins), dst, src0, src1);
2572     }
2573     else
2574     {
2575         shader_addline(buffer, "%s%s %s, %s;\n", instr, shader_arb_get_modifier(ins), dst, src0);
2576     }
2577 }
2578
2579 static void shader_hw_loop(const struct wined3d_shader_instruction *ins)
2580 {
2581     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2582     char src_name[50];
2583     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2584
2585     /* src0 is aL */
2586     shader_arb_get_src_param(ins, &ins->src[1], 0, src_name);
2587
2588     if(vshader)
2589     {
2590         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2591         struct list *e = list_head(&priv->control_frames);
2592         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2593
2594         if(priv->loop_depth > 1) shader_addline(buffer, "PUSHA aL;\n");
2595         /* The constant loader makes sure to load -1 into iX.w */
2596         shader_addline(buffer, "ARLC aL, %s.xywz;\n", src_name);
2597         shader_addline(buffer, "BRA loop_%u_end (LE.x);\n", control_frame->loop_no);
2598         shader_addline(buffer, "loop_%u_start:\n", control_frame->loop_no);
2599     }
2600     else
2601     {
2602         shader_addline(buffer, "LOOP %s;\n", src_name);
2603     }
2604 }
2605
2606 static void shader_hw_rep(const struct wined3d_shader_instruction *ins)
2607 {
2608     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2609     char src_name[50];
2610     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2611
2612     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name);
2613
2614     /* The constant loader makes sure to load -1 into iX.w */
2615     if(vshader)
2616     {
2617         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2618         struct list *e = list_head(&priv->control_frames);
2619         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2620
2621         if(priv->loop_depth > 1) shader_addline(buffer, "PUSHA aL;\n");
2622
2623         shader_addline(buffer, "ARLC aL, %s.xywz;\n", src_name);
2624         shader_addline(buffer, "BRA loop_%u_end (LE.x);\n", control_frame->loop_no);
2625         shader_addline(buffer, "loop_%u_start:\n", control_frame->loop_no);
2626     }
2627     else
2628     {
2629         shader_addline(buffer, "REP %s;\n", src_name);
2630     }
2631 }
2632
2633 static void shader_hw_endloop(const struct wined3d_shader_instruction *ins)
2634 {
2635     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2636     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2637
2638     if(vshader)
2639     {
2640         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2641         struct list *e = list_head(&priv->control_frames);
2642         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2643
2644         shader_addline(buffer, "ARAC aL.xy, aL;\n");
2645         shader_addline(buffer, "BRA loop_%u_start (GT.x);\n", control_frame->loop_no);
2646         shader_addline(buffer, "loop_%u_end:\n", control_frame->loop_no);
2647
2648         if(priv->loop_depth > 1) shader_addline(buffer, "POPA aL;\n");
2649     }
2650     else
2651     {
2652         shader_addline(buffer, "ENDLOOP;\n");
2653     }
2654 }
2655
2656 static void shader_hw_endrep(const struct wined3d_shader_instruction *ins)
2657 {
2658     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2659     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2660
2661     if(vshader)
2662     {
2663         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2664         struct list *e = list_head(&priv->control_frames);
2665         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2666
2667         shader_addline(buffer, "ARAC aL.xy, aL;\n");
2668         shader_addline(buffer, "BRA loop_%u_start (GT.x);\n", control_frame->loop_no);
2669         shader_addline(buffer, "loop_%u_end:\n", control_frame->loop_no);
2670
2671         if(priv->loop_depth > 1) shader_addline(buffer, "POPA aL;\n");
2672     }
2673     else
2674     {
2675         shader_addline(buffer, "ENDREP;\n");
2676     }
2677 }
2678
2679 static const struct control_frame *find_last_loop(const struct shader_arb_ctx_priv *priv)
2680 {
2681     struct control_frame *control_frame;
2682
2683     LIST_FOR_EACH_ENTRY(control_frame, &priv->control_frames, struct control_frame, entry)
2684     {
2685         if(control_frame->type == LOOP || control_frame->type == REP) return control_frame;
2686     }
2687     ERR("Could not find loop for break\n");
2688     return NULL;
2689 }
2690
2691 static void shader_hw_break(const struct wined3d_shader_instruction *ins)
2692 {
2693     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2694     const struct control_frame *control_frame = find_last_loop(ins->ctx->backend_data);
2695     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2696
2697     if(vshader)
2698     {
2699         shader_addline(buffer, "BRA loop_%u_end;\n", control_frame->loop_no);
2700     }
2701     else
2702     {
2703         shader_addline(buffer, "BRK;\n");
2704     }
2705 }
2706
2707 static const char *get_compare(COMPARISON_TYPE flags)
2708 {
2709     switch (flags)
2710     {
2711         case COMPARISON_GT: return "GT";
2712         case COMPARISON_EQ: return "EQ";
2713         case COMPARISON_GE: return "GE";
2714         case COMPARISON_LT: return "LT";
2715         case COMPARISON_NE: return "NE";
2716         case COMPARISON_LE: return "LE";
2717         default:
2718             FIXME("Unrecognized comparison value: %u\n", flags);
2719             return "(\?\?)";
2720     }
2721 }
2722
2723 static COMPARISON_TYPE invert_compare(COMPARISON_TYPE flags)
2724 {
2725     switch (flags)
2726     {
2727         case COMPARISON_GT: return COMPARISON_LE;
2728         case COMPARISON_EQ: return COMPARISON_NE;
2729         case COMPARISON_GE: return COMPARISON_LT;
2730         case COMPARISON_LT: return COMPARISON_GE;
2731         case COMPARISON_NE: return COMPARISON_EQ;
2732         case COMPARISON_LE: return COMPARISON_GT;
2733         default:
2734             FIXME("Unrecognized comparison value: %u\n", flags);
2735             return -1;
2736     }
2737 }
2738
2739 static void shader_hw_breakc(const struct wined3d_shader_instruction *ins)
2740 {
2741     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2742     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2743     const struct control_frame *control_frame = find_last_loop(ins->ctx->backend_data);
2744     char src_name0[50];
2745     char src_name1[50];
2746     const char *comp = get_compare(ins->flags);
2747
2748     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name0);
2749     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name1);
2750
2751     if(vshader)
2752     {
2753         /* SUBC CC, src0, src1" works only in pixel shaders, so use TA to throw
2754          * away the subtraction result
2755          */
2756         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
2757         shader_addline(buffer, "BRA loop_%u_end (%s.x);\n", control_frame->loop_no, comp);
2758     }
2759     else
2760     {
2761         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
2762         shader_addline(buffer, "BRK (%s.x);\n", comp);
2763     }
2764 }
2765
2766 static void shader_hw_ifc(const struct wined3d_shader_instruction *ins)
2767 {
2768     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2769     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2770     struct list *e = list_head(&priv->control_frames);
2771     struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2772     const char *comp;
2773     char src_name0[50];
2774     char src_name1[50];
2775     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2776
2777     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name0);
2778     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name1);
2779
2780     if(vshader)
2781     {
2782         /* Invert the flag. We jump to the else label if the condition is NOT true */
2783         comp = get_compare(invert_compare(ins->flags));
2784         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
2785         shader_addline(buffer, "BRA ifc_%u_else (%s.x);\n", control_frame->ifc_no, comp);
2786     }
2787     else
2788     {
2789         comp = get_compare(ins->flags);
2790         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
2791         shader_addline(buffer, "IF %s.x;\n", comp);
2792     }
2793 }
2794
2795 static void shader_hw_else(const struct wined3d_shader_instruction *ins)
2796 {
2797     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2798     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2799     struct list *e = list_head(&priv->control_frames);
2800     struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2801     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2802
2803     if(vshader)
2804     {
2805         shader_addline(buffer, "BRA ifc_%u_endif;\n", control_frame->ifc_no);
2806         shader_addline(buffer, "ifc_%u_else:\n", control_frame->ifc_no);
2807         control_frame->had_else = TRUE;
2808     }
2809     else
2810     {
2811         shader_addline(buffer, "ELSE;\n");
2812     }
2813 }
2814
2815 static void shader_hw_endif(const struct wined3d_shader_instruction *ins)
2816 {
2817     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2818     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2819     struct list *e = list_head(&priv->control_frames);
2820     struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2821     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2822
2823     if(vshader)
2824     {
2825         if(control_frame->had_else)
2826         {
2827             shader_addline(buffer, "ifc_%u_endif:\n", control_frame->ifc_no);
2828         }
2829         else
2830         {
2831             shader_addline(buffer, "#No else branch. else is endif\n");
2832             shader_addline(buffer, "ifc_%u_else:\n", control_frame->ifc_no);
2833         }
2834     }
2835     else
2836     {
2837         shader_addline(buffer, "ENDIF;\n");
2838     }
2839 }
2840
2841 static void shader_hw_texldd(const struct wined3d_shader_instruction *ins)
2842 {
2843     DWORD sampler_idx = ins->src[1].reg.idx;
2844     char reg_dest[40];
2845     char reg_src[3][40];
2846     DWORD flags = TEX_DERIV;
2847
2848     shader_arb_get_dst_param(ins, &ins->dst[0], reg_dest);
2849     shader_arb_get_src_param(ins, &ins->src[0], 0, reg_src[0]);
2850     shader_arb_get_src_param(ins, &ins->src[2], 1, reg_src[1]);
2851     shader_arb_get_src_param(ins, &ins->src[3], 2, reg_src[2]);
2852
2853     if (ins->flags & WINED3DSI_TEXLD_PROJECT) flags |= TEX_PROJ;
2854     if (ins->flags & WINED3DSI_TEXLD_BIAS) flags |= TEX_BIAS;
2855
2856     shader_hw_sample(ins, sampler_idx, reg_dest, reg_src[0], flags, reg_src[1], reg_src[2]);
2857 }
2858
2859 static void shader_hw_texldl(const struct wined3d_shader_instruction *ins)
2860 {
2861     DWORD sampler_idx = ins->src[1].reg.idx;
2862     char reg_dest[40];
2863     char reg_coord[40];
2864     DWORD flags = TEX_LOD;
2865
2866     shader_arb_get_dst_param(ins, &ins->dst[0], reg_dest);
2867     shader_arb_get_src_param(ins, &ins->src[0], 0, reg_coord);
2868
2869     if (ins->flags & WINED3DSI_TEXLD_PROJECT) flags |= TEX_PROJ;
2870     if (ins->flags & WINED3DSI_TEXLD_BIAS) flags |= TEX_BIAS;
2871
2872     shader_hw_sample(ins, sampler_idx, reg_dest, reg_coord, flags, NULL, NULL);
2873 }
2874
2875 static void shader_hw_label(const struct wined3d_shader_instruction *ins)
2876 {
2877     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2878     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2879
2880     priv->in_main_func = FALSE;
2881     /* Call instructions activate the NV extensions, not labels and rets. If there is an uncalled
2882      * subroutine, don't generate a label that will make GL complain
2883      */
2884     if(priv->target_version == ARB) return;
2885
2886     shader_addline(buffer, "l%u:\n", ins->src[0].reg.idx);
2887 }
2888
2889 static void vshader_add_footer(IWineD3DVertexShaderImpl *This, struct wined3d_shader_buffer *buffer,
2890         const struct arb_vs_compile_args *args, struct shader_arb_ctx_priv *priv_ctx)
2891 {
2892     const shader_reg_maps *reg_maps = &This->baseShader.reg_maps;
2893     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *)This->baseShader.device;
2894     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
2895     unsigned int i;
2896
2897     /* The D3DRS_FOGTABLEMODE render state defines if the shader-generated fog coord is used
2898      * or if the fragment depth is used. If the fragment depth is used(FOGTABLEMODE != NONE),
2899      * the fog frag coord is thrown away. If the fog frag coord is used, but not written by
2900      * the shader, it is set to 0.0(fully fogged, since start = 1.0, end = 0.0)
2901      */
2902     if(args->super.fog_src == VS_FOG_Z) {
2903         shader_addline(buffer, "MOV result.fogcoord, TMP_OUT.z;\n");
2904     } else if (!reg_maps->fog) {
2905         /* posFixup.x is always 1.0, so we can savely use it */
2906         shader_addline(buffer, "ADD result.fogcoord, posFixup.x, -posFixup.x;\n");
2907     }
2908
2909     /* Write the final position.
2910      *
2911      * OpenGL coordinates specify the center of the pixel while d3d coords specify
2912      * the corner. The offsets are stored in z and w in posFixup. posFixup.y contains
2913      * 1.0 or -1.0 to turn the rendering upside down for offscreen rendering. PosFixup.x
2914      * contains 1.0 to allow a mad, but arb vs swizzles are too restricted for that.
2915      */
2916     shader_addline(buffer, "MUL TA, posFixup, TMP_OUT.w;\n");
2917     shader_addline(buffer, "ADD TMP_OUT.x, TMP_OUT.x, TA.z;\n");
2918     shader_addline(buffer, "MAD TMP_OUT.y, TMP_OUT.y, posFixup.y, TA.w;\n");
2919
2920     if(use_nv_clip(gl_info) && priv_ctx->target_version >= NV2)
2921     {
2922         for(i = 0; i < priv_ctx->vs_clipplanes; i++)
2923         {
2924             shader_addline(buffer, "DP4 result.clip[%u].x, TMP_OUT, state.clip[%u].plane;\n", i, i);
2925         }
2926     }
2927     else if(args->boolclip.clip_control[0])
2928     {
2929         unsigned int cur_clip = 0;
2930         char component[4] = {'x', 'y', 'z', 'w'};
2931
2932         for(i = 0; i < GL_LIMITS(clipplanes); i++)
2933         {
2934             if(args->boolclip.clip_control[1] & (1 << i))
2935             {
2936                 shader_addline(buffer, "DP4 TA.%c, TMP_OUT, state.clip[%u].plane;\n",
2937                                component[cur_clip++], i);
2938             }
2939         }
2940         switch(cur_clip)
2941         {
2942             case 0:
2943                 shader_addline(buffer, "MOV TA, -helper_const.w;\n");
2944                 break;
2945             case 1:
2946                 shader_addline(buffer, "MOV TA.yzw, -helper_const.w;\n");
2947                 break;
2948             case 2:
2949                 shader_addline(buffer, "MOV TA.zw, -helper_const.w;\n");
2950                 break;
2951             case 3:
2952                 shader_addline(buffer, "MOV TA.w, -helper_const.w;\n");
2953                 break;
2954         }
2955         shader_addline(buffer, "MOV result.texcoord[%u], TA;\n",
2956                        args->boolclip.clip_control[0] - 1);
2957     }
2958
2959     /* Z coord [0;1]->[-1;1] mapping, see comment in transform_projection in state.c
2960      * and the glsl equivalent
2961      */
2962     if(need_helper_const(gl_info)) {
2963         shader_addline(buffer, "MAD TMP_OUT.z, TMP_OUT.z, helper_const.x, -TMP_OUT.w;\n");
2964     } else {
2965         shader_addline(buffer, "ADD TMP_OUT.z, TMP_OUT.z, TMP_OUT.z;\n");
2966         shader_addline(buffer, "ADD TMP_OUT.z, TMP_OUT.z, -TMP_OUT.w;\n");
2967     }
2968
2969     shader_addline(buffer, "MOV result.position, TMP_OUT;\n");
2970
2971     priv_ctx->footer_written = TRUE;
2972 }
2973
2974 static void shader_hw_ret(const struct wined3d_shader_instruction *ins)
2975 {
2976     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2977     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2978     IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *) ins->ctx->shader;
2979     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2980
2981     if(priv->target_version == ARB) return;
2982
2983     if(vshader)
2984     {
2985         if(priv->in_main_func) vshader_add_footer((IWineD3DVertexShaderImpl *) shader, buffer, priv->cur_vs_args, priv);
2986     }
2987
2988     shader_addline(buffer, "RET;\n");
2989 }
2990
2991 static void shader_hw_call(const struct wined3d_shader_instruction *ins)
2992 {
2993     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2994     shader_addline(buffer, "CAL l%u;\n", ins->src[0].reg.idx);
2995 }
2996
2997 /* GL locking is done by the caller */
2998 static GLuint create_arb_blt_vertex_program(const struct wined3d_gl_info *gl_info)
2999 {
3000     GLuint program_id = 0;
3001     GLint pos;
3002
3003     const char *blt_vprogram =
3004         "!!ARBvp1.0\n"
3005         "PARAM c[1] = { { 1, 0.5 } };\n"
3006         "MOV result.position, vertex.position;\n"
3007         "MOV result.color, c[0].x;\n"
3008         "MOV result.texcoord[0], vertex.texcoord[0];\n"
3009         "END\n";
3010
3011     GL_EXTCALL(glGenProgramsARB(1, &program_id));
3012     GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, program_id));
3013     GL_EXTCALL(glProgramStringARB(GL_VERTEX_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
3014             strlen(blt_vprogram), blt_vprogram));
3015     checkGLcall("glProgramStringARB()");
3016
3017     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
3018     if (pos != -1)
3019     {
3020         FIXME("Vertex program error at position %d: %s\n", pos,
3021             debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
3022     }
3023     else
3024     {
3025         GLint native;
3026
3027         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
3028         checkGLcall("glGetProgramivARB()");
3029         if (!native) WARN("Program exceeds native resource limits.\n");
3030     }
3031
3032     return program_id;
3033 }
3034
3035 /* GL locking is done by the caller */
3036 static GLuint create_arb_blt_fragment_program(const struct wined3d_gl_info *gl_info, enum tex_types tex_type)
3037 {
3038     GLuint program_id = 0;
3039     GLint pos;
3040
3041     static const char * const blt_fprograms[tex_type_count] =
3042     {
3043         /* tex_1d */
3044         NULL,
3045         /* tex_2d */
3046         "!!ARBfp1.0\n"
3047         "TEMP R0;\n"
3048         "TEX R0.x, fragment.texcoord[0], texture[0], 2D;\n"
3049         "MOV result.depth.z, R0.x;\n"
3050         "END\n",
3051         /* tex_3d */
3052         NULL,
3053         /* tex_cube */
3054         "!!ARBfp1.0\n"
3055         "TEMP R0;\n"
3056         "TEX R0.x, fragment.texcoord[0], texture[0], CUBE;\n"
3057         "MOV result.depth.z, R0.x;\n"
3058         "END\n",
3059         /* tex_rect */
3060         "!!ARBfp1.0\n"
3061         "TEMP R0;\n"
3062         "TEX R0.x, fragment.texcoord[0], texture[0], RECT;\n"
3063         "MOV result.depth.z, R0.x;\n"
3064         "END\n",
3065     };
3066
3067     if (!blt_fprograms[tex_type])
3068     {
3069         FIXME("tex_type %#x not supported\n", tex_type);
3070         tex_type = tex_2d;
3071     }
3072
3073     GL_EXTCALL(glGenProgramsARB(1, &program_id));
3074     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, program_id));
3075     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
3076             strlen(blt_fprograms[tex_type]), blt_fprograms[tex_type]));
3077     checkGLcall("glProgramStringARB()");
3078
3079     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
3080     if (pos != -1)
3081     {
3082         FIXME("Fragment program error at position %d: %s\n", pos,
3083             debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
3084     }
3085     else
3086     {
3087         GLint native;
3088
3089         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
3090         checkGLcall("glGetProgramivARB()");
3091         if (!native) WARN("Program exceeds native resource limits.\n");
3092     }
3093
3094     return program_id;
3095 }
3096
3097 static void arbfp_add_sRGB_correction(struct wined3d_shader_buffer *buffer, const char *fragcolor,
3098         const char *tmp1, const char *tmp2, const char *tmp3, const char *tmp4, BOOL condcode)
3099 {
3100     /* Perform sRGB write correction. See GLX_EXT_framebuffer_sRGB */
3101
3102     if(condcode)
3103     {
3104         /* Sigh. MOVC CC doesn't work, so use one of the temps as dummy dest */
3105         shader_addline(buffer, "SUBC %s, %s.x, srgb_consts1.y;\n", tmp1, fragcolor);
3106         /* Calculate the > 0.0031308 case */
3107         shader_addline(buffer, "POW %s.x (GE), %s.x, srgb_consts1.z;\n", fragcolor, fragcolor);
3108         shader_addline(buffer, "POW %s.y (GE), %s.y, srgb_consts1.z;\n", fragcolor, fragcolor);
3109         shader_addline(buffer, "POW %s.z (GE), %s.z, srgb_consts1.z;\n", fragcolor, fragcolor);
3110         shader_addline(buffer, "MUL %s.xyz (GE), %s, srgb_consts1.w;\n", fragcolor, fragcolor);
3111         shader_addline(buffer, "SUB %s.xyz (GE), %s, srgb_consts2.x;\n", fragcolor, fragcolor);
3112         /* Calculate the < case */
3113         shader_addline(buffer, "MUL %s.xyz (LT), srgb_consts1.x, %s;\n", fragcolor, fragcolor);
3114     }
3115     else
3116     {
3117         /* Calculate the > 0.0031308 case */
3118         shader_addline(buffer, "POW %s.x, %s.x, srgb_consts1.z;\n", tmp1, fragcolor);
3119         shader_addline(buffer, "POW %s.y, %s.y, srgb_consts1.z;\n", tmp1, fragcolor);
3120         shader_addline(buffer, "POW %s.z, %s.z, srgb_consts1.z;\n", tmp1, fragcolor);
3121         shader_addline(buffer, "MUL %s, %s, srgb_consts1.w;\n", tmp1, tmp1);
3122         shader_addline(buffer, "SUB %s, %s, srgb_consts2.x;\n", tmp1, tmp1);
3123         /* Calculate the < case */
3124         shader_addline(buffer, "MUL %s, srgb_consts1.x, %s;\n", tmp2, fragcolor);
3125         /* Get 1.0 / 0.0 masks for > 0.0031308 and < 0.0031308 */
3126         shader_addline(buffer, "SLT %s, srgb_consts1.y, %s;\n", tmp3, fragcolor);
3127         shader_addline(buffer, "SGE %s, srgb_consts1.y, %s;\n", tmp4, fragcolor);
3128         /* Store the components > 0.0031308 in the destination */
3129         shader_addline(buffer, "MUL %s.xyz, %s, %s;\n", fragcolor, tmp1, tmp3);
3130         /* Add the components that are < 0.0031308 */
3131         shader_addline(buffer, "MAD %s.xyz, %s, %s, %s;\n", fragcolor, tmp2, tmp4, fragcolor);
3132         /* Move everything into result.color at once. Nvidia hardware cannot handle partial
3133         * result.color writes(.rgb first, then .a), or handle overwriting already written
3134         * components. The assembler uses a temporary register in this case, which is usually
3135         * not allocated from one of our registers that were used earlier.
3136         */
3137     }
3138     shader_addline(buffer, "MOV result.color, %s;\n", fragcolor);
3139     /* [0.0;1.0] clamping. Not needed, this is done implicitly */
3140 }
3141
3142 static const DWORD *find_loop_control_values(IWineD3DBaseShaderImpl *This, DWORD idx)
3143 {
3144     const local_constant *constant;
3145
3146     LIST_FOR_EACH_ENTRY(constant, &This->baseShader.constantsI, local_constant, entry)
3147     {
3148         if (constant->idx == idx)
3149         {
3150             return constant->value;
3151         }
3152     }
3153     return NULL;
3154 }
3155
3156 static void init_ps_input(const IWineD3DPixelShaderImpl *This, const struct arb_ps_compile_args *args,
3157                           struct shader_arb_ctx_priv *priv)
3158 {
3159     const char *texcoords[8] =
3160     {
3161         "fragment.texcoord[0]", "fragment.texcoord[1]", "fragment.texcoord[2]", "fragment.texcoord[3]",
3162         "fragment.texcoord[4]", "fragment.texcoord[5]", "fragment.texcoord[6]", "fragment.texcoord[7]"
3163     };
3164     unsigned int i;
3165     const struct wined3d_shader_signature_element *sig = This->input_signature;
3166     const char *semantic_name;
3167     DWORD semantic_idx;
3168
3169     switch(args->super.vp_mode)
3170     {
3171         case pretransformed:
3172         case fixedfunction:
3173             /* The pixelshader has to collect the varyings on its own. In any case properly load
3174              * color0 and color1. In the case of pretransformed vertices also load texcoords. Set
3175              * other attribs to 0.0.
3176              *
3177              * For fixedfunction this behavior is correct, according to the tests. For pretransformed
3178              * we'd either need a replacement shader that can load other attribs like BINORMAL, or
3179              * load the texcoord attrib pointers to match the pixel shader signature
3180              */
3181             for(i = 0; i < MAX_REG_INPUT; i++)
3182             {
3183                 semantic_name = sig[i].semantic_name;
3184                 semantic_idx = sig[i].semantic_idx;
3185                 if(semantic_name == NULL) continue;
3186
3187                 if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_COLOR))
3188                 {
3189                     if(semantic_idx == 0) priv->ps_input[i] = "fragment.color.primary";
3190                     else if(semantic_idx == 1) priv->ps_input[i] = "fragment.color.secondary";
3191                     else priv->ps_input[i] = "0.0";
3192                 }
3193                 else if(args->super.vp_mode == fixedfunction)
3194                 {
3195                     priv->ps_input[i] = "0.0";
3196                 }
3197                 else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_TEXCOORD))
3198                 {
3199                     if(semantic_idx < 8) priv->ps_input[i] = texcoords[semantic_idx];
3200                     else priv->ps_input[i] = "0.0";
3201                 }
3202                 else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_FOG))
3203                 {
3204                     if(semantic_idx == 0) priv->ps_input[i] = "fragment.fogcoord";
3205                     else priv->ps_input[i] = "0.0";
3206                 }
3207                 else
3208                 {
3209                     priv->ps_input[i] = "0.0";
3210                 }
3211
3212                 TRACE("v%u, semantic %s%u is %s\n", i, semantic_name, semantic_idx, priv->ps_input[i]);
3213             }
3214             break;
3215
3216         case vertexshader:
3217             /* That one is easy. The vertex shaders provide v0-v7 in fragment.texcoord and v8 and v9 in
3218              * fragment.color
3219              */
3220             for(i = 0; i < 8; i++)
3221             {
3222                 priv->ps_input[i] = texcoords[i];
3223             }
3224             priv->ps_input[8] = "fragment.color.primary";
3225             priv->ps_input[9] = "fragment.color.secondary";
3226             break;
3227     }
3228 }
3229
3230 /* GL locking is done by the caller */
3231 static GLuint shader_arb_generate_pshader(IWineD3DPixelShaderImpl *This, struct wined3d_shader_buffer *buffer,
3232         const struct arb_ps_compile_args *args, struct arb_ps_compiled_shader *compiled)
3233 {
3234     const shader_reg_maps* reg_maps = &This->baseShader.reg_maps;
3235     CONST DWORD *function = This->baseShader.function;
3236     const struct wined3d_gl_info *gl_info = &((IWineD3DDeviceImpl *)This->baseShader.device)->adapter->gl_info;
3237     const local_constant *lconst;
3238     GLuint retval;
3239     char fragcolor[16];
3240     DWORD *lconst_map = local_const_mapping((IWineD3DBaseShaderImpl *) This), next_local, cur;
3241     struct shader_arb_ctx_priv priv_ctx;
3242     BOOL dcl_tmp = args->super.srgb_correction, dcl_td = FALSE;
3243     BOOL want_nv_prog = FALSE;
3244     struct arb_pshader_private *shader_priv = This->backend_priv;
3245     GLint errPos;
3246     DWORD map;
3247
3248     char srgbtmp[4][4];
3249     unsigned int i, found = 0;
3250
3251     for (i = 0, map = reg_maps->temporary; map; map >>= 1, ++i)
3252     {
3253         if (!(map & 1)
3254                 || (This->color0_mov && i == This->color0_reg)
3255                 || (reg_maps->shader_version.major < 2 && i == 0))
3256             continue;
3257
3258         sprintf(srgbtmp[found], "R%u", i);
3259         ++found;
3260         if (found == 4) break;
3261     }
3262
3263     switch(found) {
3264         case 4: dcl_tmp = FALSE; break;
3265         case 0:
3266             sprintf(srgbtmp[0], "TA");
3267             sprintf(srgbtmp[1], "TB");
3268             sprintf(srgbtmp[2], "TC");
3269             sprintf(srgbtmp[3], "TD");
3270             dcl_td = TRUE;
3271             break;
3272         case 1:
3273             sprintf(srgbtmp[1], "TA");
3274             sprintf(srgbtmp[2], "TB");
3275             sprintf(srgbtmp[3], "TC");
3276             break;
3277         case 2:
3278             sprintf(srgbtmp[2], "TA");
3279             sprintf(srgbtmp[3], "TB");
3280             break;
3281         case 3:
3282             sprintf(srgbtmp[3], "TA");
3283             break;
3284     }
3285
3286     /*  Create the hw ARB shader */
3287     memset(&priv_ctx, 0, sizeof(priv_ctx));
3288     priv_ctx.cur_ps_args = args;
3289     priv_ctx.compiled_fprog = compiled;
3290     priv_ctx.cur_np2fixup_info = &compiled->np2fixup_info;
3291     init_ps_input(This, args, &priv_ctx);
3292     list_init(&priv_ctx.control_frames);
3293
3294     /* Avoid enabling NV_fragment_program* if we do not need it.
3295      *
3296      * Enabling GL_NV_fragment_program_option causes the driver to occupy a temporary register,
3297      * and it slows down the shader execution noticeably(about 5%). Usually our instruction emulation
3298      * is faster than what we gain from using higher native instructions. There are some things though
3299      * that cannot be emulated. In that case enable the extensions.
3300      * If the extension is enabled, instruction handlers that support both ways will use it.
3301      *
3302      * Testing shows no performance difference between OPTION NV_fragment_program2 and NV_fragment_program.
3303      * So enable the best we can get.
3304      */
3305     if(reg_maps->usesdsx || reg_maps->usesdsy || reg_maps->loop_depth > 0 || reg_maps->usestexldd ||
3306        reg_maps->usestexldl || reg_maps->usesfacing || reg_maps->usesifc || reg_maps->usescall)
3307     {
3308         want_nv_prog = TRUE;
3309     }
3310
3311     shader_addline(buffer, "!!ARBfp1.0\n");
3312     if(want_nv_prog && GL_SUPPORT(NV_FRAGMENT_PROGRAM2)) {
3313         shader_addline(buffer, "OPTION NV_fragment_program2;\n");
3314         priv_ctx.target_version = NV3;
3315     } else if(want_nv_prog && GL_SUPPORT(NV_FRAGMENT_PROGRAM_OPTION)) {
3316         shader_addline(buffer, "OPTION NV_fragment_program;\n");
3317         priv_ctx.target_version = NV2;
3318     } else {
3319         if(want_nv_prog)
3320         {
3321             /* This is an error - either we're advertising the wrong shader version, or aren't enforcing some
3322              * limits properly
3323              */
3324             ERR("The shader requires instructions that are not available in plain GL_ARB_fragment_program\n");
3325             ERR("Try GLSL\n");
3326         }
3327         priv_ctx.target_version = ARB;
3328     }
3329
3330     if(This->baseShader.reg_maps.highest_render_target > 0)
3331     {
3332         shader_addline(buffer, "OPTION ARB_draw_buffers;\n");
3333     }
3334
3335     if (reg_maps->shader_version.major < 3)
3336     {
3337         switch(args->super.fog) {
3338             case FOG_OFF:
3339                 break;
3340             case FOG_LINEAR:
3341                 shader_addline(buffer, "OPTION ARB_fog_linear;\n");
3342                 break;
3343             case FOG_EXP:
3344                 shader_addline(buffer, "OPTION ARB_fog_exp;\n");
3345                 break;
3346             case FOG_EXP2:
3347                 shader_addline(buffer, "OPTION ARB_fog_exp2;\n");
3348                 break;
3349         }
3350     }
3351
3352     /* For now always declare the temps. At least the Nvidia assembler optimizes completely
3353      * unused temps away(but occupies them for the whole shader if they're used once). Always
3354      * declaring them avoids tricky bookkeeping work
3355      */
3356     shader_addline(buffer, "TEMP TA;\n");      /* Used for modifiers */
3357     shader_addline(buffer, "TEMP TB;\n");      /* Used for modifiers */
3358     shader_addline(buffer, "TEMP TC;\n");      /* Used for modifiers */
3359     if(dcl_td) shader_addline(buffer, "TEMP TD;\n"); /* Used for sRGB writing */
3360     shader_addline(buffer, "PARAM coefdiv = { 0.5, 0.25, 0.125, 0.0625 };\n");
3361     shader_addline(buffer, "PARAM coefmul = { 2, 4, 8, 16 };\n");
3362     shader_addline(buffer, "PARAM one = { 1.0, 1.0, 1.0, 1.0 };\n");
3363
3364     if (reg_maps->shader_version.major < 2)
3365     {
3366         strcpy(fragcolor, "R0");
3367     } else {
3368         if(args->super.srgb_correction) {
3369             if(This->color0_mov) {
3370                 sprintf(fragcolor, "R%u", This->color0_reg);
3371             } else {
3372                 shader_addline(buffer, "TEMP TMP_COLOR;\n");
3373                 strcpy(fragcolor, "TMP_COLOR");
3374             }
3375         } else {
3376             strcpy(fragcolor, "result.color");
3377         }
3378     }
3379
3380     if(args->super.srgb_correction) {
3381         shader_addline(buffer, "PARAM srgb_consts1 = {%f, %f, %f, %f};\n",
3382                        srgb_mul_low, srgb_cmp, srgb_pow, srgb_mul_high);
3383         shader_addline(buffer, "PARAM srgb_consts2 = {%f, %f, %f, %f};\n",
3384                        srgb_sub_high, 0.0, 0.0, 0.0);
3385     }
3386
3387     /* Base Declarations */
3388     next_local = shader_generate_arb_declarations( (IWineD3DBaseShader*) This, reg_maps, buffer, &GLINFO_LOCATION,
3389             lconst_map, NULL, &priv_ctx);
3390
3391     for (i = 0, map = reg_maps->bumpmat; map; map >>= 1, ++i)
3392     {
3393         if (!(map & 1)) continue;
3394
3395         cur = compiled->numbumpenvmatconsts;
3396         compiled->bumpenvmatconst[cur].const_num = WINED3D_CONST_NUM_UNUSED;
3397         compiled->bumpenvmatconst[cur].texunit = i;
3398         compiled->luminanceconst[cur].const_num = WINED3D_CONST_NUM_UNUSED;
3399         compiled->luminanceconst[cur].texunit = i;
3400
3401         /* We can fit the constants into the constant limit for sure because texbem, texbeml, bem and beml are only supported
3402          * in 1.x shaders, and GL_ARB_fragment_program has a constant limit of 24 constants. So in the worst case we're loading
3403          * 8 shader constants, 8 bump matrices and 8 luminance parameters and are perfectly fine. (No NP2 fixup on bumpmapped
3404          * textures due to conditional NP2 restrictions)
3405          *
3406          * Use local constants to load the bump env parameters, not program.env. This avoids collisions with d3d constants of
3407          * shaders in newer shader models. Since the bump env parameters have to share their space with NP2 fixup constants,
3408          * their location is shader dependent anyway and they cannot be loaded globally.
3409          */
3410         compiled->bumpenvmatconst[cur].const_num = next_local++;
3411         shader_addline(buffer, "PARAM bumpenvmat%d = program.local[%d];\n",
3412                        i, compiled->bumpenvmatconst[cur].const_num);
3413         compiled->numbumpenvmatconsts = cur + 1;
3414
3415         if (!(reg_maps->luminanceparams & (1 << i))) continue;
3416
3417         compiled->luminanceconst[cur].const_num = next_local++;
3418         shader_addline(buffer, "PARAM luminance%d = program.local[%d];\n",
3419                        i, compiled->luminanceconst[cur].const_num);
3420     }
3421
3422     for(i = 0; i < MAX_CONST_I; i++)
3423     {
3424         compiled->int_consts[i] = WINED3D_CONST_NUM_UNUSED;
3425         if (reg_maps->integer_constants & (1 << i) && priv_ctx.target_version >= NV2)
3426         {
3427             const DWORD *control_values = find_loop_control_values((IWineD3DBaseShaderImpl *) This, i);
3428
3429             if(control_values)
3430             {
3431                 shader_addline(buffer, "PARAM I%u = {%u, %u, %u, -1};\n", i,
3432                                 control_values[0], control_values[1], control_values[2]);
3433             }
3434             else
3435             {
3436                 compiled->int_consts[i] = next_local;
3437                 compiled->num_int_consts++;
3438                 shader_addline(buffer, "PARAM I%u = program.local[%u];\n", i, next_local++);
3439             }
3440         }
3441     }
3442
3443     if(reg_maps->vpos || reg_maps->usesdsy)
3444     {
3445         compiled->ycorrection = next_local;
3446         shader_addline(buffer, "PARAM ycorrection = program.local[%u];\n", next_local++);
3447
3448         if(reg_maps->vpos)
3449         {
3450             shader_addline(buffer, "TEMP vpos;\n");
3451             /* ycorrection.x: Backbuffer height(onscreen) or 0(offscreen).
3452              * ycorrection.y: -1.0(onscreen), 1.0(offscreen)
3453              * ycorrection.z: 1.0
3454              * ycorrection.w: 0.0
3455              */
3456             shader_addline(buffer, "MAD vpos, fragment.position, ycorrection.zyww, ycorrection.wxww;\n");
3457             shader_addline(buffer, "FLR vpos.xy, vpos;\n");
3458         }
3459     }
3460     else
3461     {
3462         compiled->ycorrection = WINED3D_CONST_NUM_UNUSED;
3463     }
3464
3465     /* Load constants to fixup NP2 texcoords if there are still free constants left:
3466      * Constants (texture dimensions) for the NP2 fixup are loaded as local program parameters. This will consume
3467      * at most 8 (MAX_FRAGMENT_SAMPLERS / 2) parameters, which is highly unlikely, since the application had to
3468      * use 16 NP2 textures at the same time. In case that we run out of constants the fixup is simply not
3469      * applied / activated. This will probably result in wrong rendering of the texture, but will save us from
3470      * shader compilation errors and the subsequent errors when drawing with this shader. */
3471     if (priv_ctx.cur_ps_args->super.np2_fixup) {
3472
3473         struct arb_ps_np2fixup_info* const fixup = priv_ctx.cur_np2fixup_info;
3474         const WORD map = priv_ctx.cur_ps_args->super.np2_fixup;
3475         const UINT max_lconsts = gl_info->ps_arb_max_local_constants;
3476
3477         fixup->offset = next_local;
3478         fixup->super.active = 0;
3479
3480         cur = 0;
3481         for (i = 0; i < MAX_FRAGMENT_SAMPLERS; ++i) {
3482             if (!(map & (1 << i))) continue;
3483
3484             if (fixup->offset + (cur >> 1) < max_lconsts) {
3485                 fixup->super.active |= (1 << i);
3486                 fixup->super.idx[i] = cur++;
3487             } else {
3488                 FIXME("No free constant found to load NP2 fixup data into shader. "
3489                       "Sampling from this texture will probably look wrong.\n");
3490                 break;
3491             }
3492         }
3493
3494         fixup->super.num_consts = (cur + 1) >> 1;
3495         if (fixup->super.num_consts) {
3496             shader_addline(buffer, "PARAM np2fixup[%u] = { program.env[%u..%u] };\n",
3497                            fixup->super.num_consts, fixup->offset, fixup->super.num_consts + fixup->offset - 1);
3498         }
3499
3500         next_local += fixup->super.num_consts;
3501     }
3502
3503     if (shader_priv->clipplane_emulation != ~0U && args->clip)
3504     {
3505         shader_addline(buffer, "KIL fragment.texcoord[%u];\n", shader_priv->clipplane_emulation);
3506     }
3507
3508     /* Base Shader Body */
3509     shader_generate_main((IWineD3DBaseShader *)This, buffer, reg_maps, function, &priv_ctx);
3510
3511     if(args->super.srgb_correction) {
3512         arbfp_add_sRGB_correction(buffer, fragcolor, srgbtmp[0], srgbtmp[1], srgbtmp[2], srgbtmp[3],
3513                                   priv_ctx.target_version >= NV2);
3514     } else if(reg_maps->shader_version.major < 2) {
3515         shader_addline(buffer, "MOV result.color, %s;\n", fragcolor);
3516     }
3517     shader_addline(buffer, "END\n");
3518
3519     /* TODO: change to resource.glObjectHandle or something like that */
3520     GL_EXTCALL(glGenProgramsARB(1, &retval));
3521
3522     TRACE("Creating a hw pixel shader, prg=%d\n", retval);
3523     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, retval));
3524
3525     TRACE("Created hw pixel shader, prg=%d\n", retval);
3526     /* Create the program and check for errors */
3527     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
3528                buffer->bsize, buffer->buffer));
3529     checkGLcall("glProgramStringARB()");
3530
3531     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &errPos);
3532     if (errPos != -1)
3533     {
3534         FIXME("HW PixelShader Error at position %d: %s\n",
3535               errPos, debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
3536         retval = 0;
3537     }
3538     else
3539     {
3540         GLint native;
3541
3542         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
3543         checkGLcall("glGetProgramivARB()");
3544         if (!native) WARN("Program exceeds native resource limits.\n");
3545     }
3546
3547     /* Load immediate constants */
3548     if(lconst_map) {
3549         LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry) {
3550             const float *value = (const float *)lconst->value;
3551             GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, lconst_map[lconst->idx], value));
3552             checkGLcall("glProgramLocalParameter4fvARB");
3553         }
3554         HeapFree(GetProcessHeap(), 0, lconst_map);
3555     }
3556
3557     return retval;
3558 }
3559
3560 static int compare_sig(const struct wined3d_shader_signature_element *sig1, const struct wined3d_shader_signature_element *sig2)
3561 {
3562     unsigned int i;
3563     int ret;
3564
3565     for(i = 0; i < MAX_REG_INPUT; i++)
3566     {
3567         if(sig1[i].semantic_name == NULL || sig2[i].semantic_name == NULL)
3568         {
3569             /* Compare pointers, not contents. One string is NULL(element does not exist), the other one is not NULL */
3570             if(sig1[i].semantic_name != sig2[i].semantic_name) return sig1[i].semantic_name < sig2[i].semantic_name ? -1 : 1;
3571             continue;
3572         }
3573
3574         ret = strcmp(sig1[i].semantic_name, sig2[i].semantic_name);
3575         if(ret != 0) return ret;
3576         if(sig1[i].semantic_idx    != sig2[i].semantic_idx)    return sig1[i].semantic_idx    < sig2[i].semantic_idx    ? -1 : 1;
3577         if(sig1[i].sysval_semantic != sig2[i].sysval_semantic) return sig1[i].sysval_semantic < sig2[i].sysval_semantic ? -1 : 1;
3578         if(sig1[i].component_type  != sig2[i].component_type)  return sig1[i].sysval_semantic < sig2[i].component_type  ? -1 : 1;
3579         if(sig1[i].register_idx    != sig2[i].register_idx)    return sig1[i].register_idx    < sig2[i].register_idx    ? -1 : 1;
3580         if(sig1[i].mask            != sig2->mask)              return sig1[i].mask            < sig2[i].mask            ? -1 : 1;
3581     }
3582     return 0;
3583 }
3584
3585 static struct wined3d_shader_signature_element *clone_sig(const struct wined3d_shader_signature_element *sig)
3586 {
3587     struct wined3d_shader_signature_element *new;
3588     int i;
3589     char *name;
3590
3591     new = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*new) * MAX_REG_INPUT);
3592     for(i = 0; i < MAX_REG_INPUT; i++)
3593     {
3594         if(sig[i].semantic_name == NULL)
3595         {
3596             continue;
3597         }
3598
3599         new[i] = sig[i];
3600         /* Clone the semantic string */
3601         name = HeapAlloc(GetProcessHeap(), 0, strlen(sig[i].semantic_name) + 1);
3602         strcpy(name, sig[i].semantic_name);
3603         new[i].semantic_name = name;
3604     }
3605     return new;
3606 }
3607
3608 static DWORD find_input_signature(struct shader_arb_priv *priv, const struct wined3d_shader_signature_element *sig)
3609 {
3610     struct wine_rb_entry *entry = wine_rb_get(&priv->signature_tree, sig);
3611     struct ps_signature *found_sig;
3612
3613     if(entry != NULL)
3614     {
3615         found_sig = WINE_RB_ENTRY_VALUE(entry, struct ps_signature, entry);
3616         TRACE("Found existing signature %u\n", found_sig->idx);
3617         return found_sig->idx;
3618     }
3619     found_sig = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*sig));
3620     found_sig->sig = clone_sig(sig);
3621     found_sig->idx = priv->ps_sig_number++;
3622     TRACE("New signature stored and assigned number %u\n", found_sig->idx);
3623     if(wine_rb_put(&priv->signature_tree, sig, &found_sig->entry) == -1)
3624     {
3625         ERR("Failed to insert program entry.\n");
3626     }
3627     return found_sig->idx;
3628 }
3629
3630 static void init_output_registers(IWineD3DVertexShaderImpl *shader, DWORD sig_num, struct shader_arb_ctx_priv *priv_ctx,
3631                                   struct arb_vs_compiled_shader *compiled)
3632 {
3633     unsigned int i, j;
3634     static const char *texcoords[8] =
3635     {
3636         "result.texcoord[0]", "result.texcoord[1]", "result.texcoord[2]", "result.texcoord[3]",
3637         "result.texcoord[4]", "result.texcoord[5]", "result.texcoord[6]", "result.texcoord[7]"
3638     };
3639     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *) shader->baseShader.device;
3640     const struct wined3d_shader_signature_element *sig;
3641     const char *semantic_name;
3642     DWORD semantic_idx, reg_idx;
3643
3644     /* Write generic input varyings 0 to 7 to result.texcoord[], varying 8 to result.color.primary
3645      * and varying 9 to result.color.secondary
3646      */
3647     const char *decl_idx_to_string[MAX_REG_INPUT] =
3648     {
3649         texcoords[0], texcoords[1], texcoords[2], texcoords[3],
3650         texcoords[4], texcoords[5], texcoords[6], texcoords[7],
3651         "result.color.primary", "result.color.secondary"
3652     };
3653
3654     if(sig_num == ~0)
3655     {
3656         TRACE("Pixel shader uses builtin varyings\n");
3657         /* Map builtins to builtins */
3658         for(i = 0; i < 8; i++)
3659         {
3660             priv_ctx->texcrd_output[i] = texcoords[i];
3661         }
3662         priv_ctx->color_output[0] = "result.color.primary";
3663         priv_ctx->color_output[1] = "result.color.secondary";
3664         priv_ctx->fog_output = "result.fogcoord";
3665
3666         /* Map declared regs to builtins. Use "TA" to /dev/null unread output */
3667         for(i = 0; i < (sizeof(shader->output_signature) / sizeof(*shader->output_signature)); i++)
3668         {
3669             semantic_name = shader->output_signature[i].semantic_name;
3670             if(semantic_name == NULL) continue;
3671
3672             if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_POSITION))
3673             {
3674                 TRACE("o%u is TMP_OUT\n", i);
3675                 if(shader->output_signature[i].semantic_idx == 0) priv_ctx->vs_output[i] = "TMP_OUT";
3676                 else priv_ctx->vs_output[i] = "TA";
3677             }
3678             else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_PSIZE))
3679             {
3680                 TRACE("o%u is result.pointsize\n", i);
3681                 if(shader->output_signature[i].semantic_idx == 0) priv_ctx->vs_output[i] = "result.pointsize";
3682                 else priv_ctx->vs_output[i] = "TA";
3683             }
3684             else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_COLOR))
3685             {
3686                 TRACE("o%u is result.color.?, idx %u\n", i, shader->output_signature[i].semantic_idx);
3687                 if(shader->output_signature[i].semantic_idx == 0) priv_ctx->vs_output[i] = "result.color.primary";
3688                 else if(shader->output_signature[i].semantic_idx == 1) priv_ctx->vs_output[i] = "result.color.secondary";
3689                 else priv_ctx->vs_output[i] = "TA";
3690             }
3691             else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_TEXCOORD))
3692             {
3693                 TRACE("o%u is %s\n", i, texcoords[shader->output_signature[i].semantic_idx]);
3694                 if(shader->output_signature[i].semantic_idx >= 8) priv_ctx->vs_output[i] = "TA";
3695                 else priv_ctx->vs_output[i] = texcoords[shader->output_signature[i].semantic_idx];
3696             }
3697             else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_FOG))
3698             {
3699                 TRACE("o%u is result.fogcoord\n", i);
3700                 if(shader->output_signature[i].semantic_idx > 0) priv_ctx->vs_output[i] = "TA";
3701                 else priv_ctx->vs_output[i] = "result.fogcoord";
3702             }
3703             else
3704             {
3705                 priv_ctx->vs_output[i] = "TA";
3706             }
3707         }
3708         return;
3709     }
3710
3711     /* Instead of searching for the signature in the signature list, read the one from the current pixel shader.
3712      * Its maybe not the shader where the signature came from, but it is the same signature and faster to find
3713      */
3714     sig = ((IWineD3DPixelShaderImpl *)device->stateBlock->pixelShader)->input_signature;
3715     TRACE("Pixel shader uses declared varyings\n");
3716
3717     /* Map builtin to declared. /dev/null the results by default to the TA temp reg */
3718     for(i = 0; i < 8; i++)
3719     {
3720         priv_ctx->texcrd_output[i] = "TA";
3721     }
3722     priv_ctx->color_output[0] = "TA";
3723     priv_ctx->color_output[1] = "TA";
3724     priv_ctx->fog_output = "TA";
3725
3726     for(i = 0; i < MAX_REG_INPUT; i++)
3727     {
3728         semantic_name = sig[i].semantic_name;
3729         semantic_idx = sig[i].semantic_idx;
3730         reg_idx = sig[i].register_idx;
3731         if(semantic_name == NULL) continue;
3732
3733         /* If a declared input register is not written by builtin arguments, don't write to it.
3734          * GL_NV_vertex_program makes sure the input defaults to 0.0, which is correct with D3D
3735          *
3736          * Don't care about POSITION and PSIZE here - this is a builtin vertex shader, position goes
3737          * to TMP_OUT in any case
3738          */
3739         if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_TEXCOORD))
3740         {
3741             if(semantic_idx < 8) priv_ctx->texcrd_output[semantic_idx] = decl_idx_to_string[reg_idx];
3742         }
3743         else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_COLOR))
3744         {
3745             if(semantic_idx < 2) priv_ctx->color_output[semantic_idx] = decl_idx_to_string[reg_idx];
3746         }
3747         else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_FOG))
3748         {
3749             if(semantic_idx == 0) priv_ctx->fog_output = decl_idx_to_string[reg_idx];
3750         }
3751         else
3752         {
3753             continue;
3754         }
3755
3756         if(strcmp(decl_idx_to_string[reg_idx], "result.color.primary") == 0 ||
3757            strcmp(decl_idx_to_string[reg_idx], "result.color.secondary") == 0)
3758         {
3759             compiled->need_color_unclamp = TRUE;
3760         }
3761     }
3762
3763     /* Map declared to declared */
3764     for(i = 0; i < (sizeof(shader->output_signature) / sizeof(*shader->output_signature)); i++)
3765     {
3766         /* Write unread output to TA to throw them away */
3767         priv_ctx->vs_output[i] = "TA";
3768         semantic_name = shader->output_signature[i].semantic_name;
3769         if(semantic_name == NULL)
3770         {
3771             continue;
3772         }
3773
3774         if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_POSITION) &&
3775            shader->output_signature[i].semantic_idx == 0)
3776         {
3777             priv_ctx->vs_output[i] = "TMP_OUT";
3778             continue;
3779         }
3780         else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_PSIZE) &&
3781            shader->output_signature[i].semantic_idx == 0)
3782         {
3783             priv_ctx->vs_output[i] = "result.pointsize";
3784             continue;
3785         }
3786
3787         for(j = 0; j < MAX_REG_INPUT; j++)
3788         {
3789             if(sig[j].semantic_name == NULL)
3790             {
3791                 continue;
3792             }
3793
3794             if(strcmp(sig[j].semantic_name, semantic_name) == 0 &&
3795                sig[j].semantic_idx == shader->output_signature[i].semantic_idx)
3796             {
3797                 priv_ctx->vs_output[i] = decl_idx_to_string[sig[j].register_idx];
3798
3799                 if(strcmp(priv_ctx->vs_output[i], "result.color.primary") == 0 ||
3800                    strcmp(priv_ctx->vs_output[i], "result.color.secondary") == 0)
3801                 {
3802                     compiled->need_color_unclamp = TRUE;
3803                 }
3804             }
3805         }
3806     }
3807 }
3808
3809 /* GL locking is done by the caller */
3810 static GLuint shader_arb_generate_vshader(IWineD3DVertexShaderImpl *This, struct wined3d_shader_buffer *buffer,
3811         const struct arb_vs_compile_args *args, struct arb_vs_compiled_shader *compiled)
3812 {
3813     const shader_reg_maps *reg_maps = &This->baseShader.reg_maps;
3814     CONST DWORD *function = This->baseShader.function;
3815     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *)This->baseShader.device;
3816     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
3817     const local_constant *lconst;
3818     GLuint ret;
3819     DWORD next_local, *lconst_map = local_const_mapping((IWineD3DBaseShaderImpl *) This);
3820     struct shader_arb_ctx_priv priv_ctx;
3821     unsigned int i;
3822     GLint errPos;
3823
3824     memset(&priv_ctx, 0, sizeof(priv_ctx));
3825     priv_ctx.cur_vs_args = args;
3826     list_init(&priv_ctx.control_frames);
3827     init_output_registers(This, args->ps_signature, &priv_ctx, compiled);
3828
3829     /*  Create the hw ARB shader */
3830     shader_addline(buffer, "!!ARBvp1.0\n");
3831
3832     /* Always enable the NV extension if available. Unlike fragment shaders, there is no
3833      * mesurable performance penalty, and we can always make use of it for clipplanes.
3834      */
3835     if(GL_SUPPORT(NV_VERTEX_PROGRAM3)) {
3836         shader_addline(buffer, "OPTION NV_vertex_program3;\n");
3837         priv_ctx.target_version = NV3;
3838         shader_addline(buffer, "ADDRESS aL;\n");
3839     } else if(GL_SUPPORT(NV_VERTEX_PROGRAM2_OPTION)) {
3840         shader_addline(buffer, "OPTION NV_vertex_program2;\n");
3841         priv_ctx.target_version = NV2;
3842         shader_addline(buffer, "ADDRESS aL;\n");
3843     } else {
3844         priv_ctx.target_version = ARB;
3845     }
3846
3847     shader_addline(buffer, "TEMP TMP_OUT;\n");
3848     if(need_helper_const(gl_info)) {
3849         shader_addline(buffer, "PARAM helper_const = { 2.0, -1.0, %d.0, 0.0 };\n", This->rel_offset);
3850     }
3851     if(need_mova_const((IWineD3DBaseShader *) This, gl_info)) {
3852         shader_addline(buffer, "PARAM mova_const = { 0.5, 0.0, 2.0, 1.0 };\n");
3853         shader_addline(buffer, "TEMP A0_SHADOW;\n");
3854     }
3855
3856     shader_addline(buffer, "TEMP TA;\n");
3857
3858     /* Base Declarations */
3859     next_local = shader_generate_arb_declarations( (IWineD3DBaseShader*) This, reg_maps, buffer, &GLINFO_LOCATION,
3860             lconst_map, &priv_ctx.vs_clipplanes, &priv_ctx);
3861
3862     for(i = 0; i < MAX_CONST_I; i++)
3863     {
3864         compiled->int_consts[i] = WINED3D_CONST_NUM_UNUSED;
3865         if(reg_maps->integer_constants & (1 << i) && priv_ctx.target_version >= NV2)
3866         {
3867             const DWORD *control_values = find_loop_control_values((IWineD3DBaseShaderImpl *) This, i);
3868
3869             if(control_values)
3870             {
3871                 shader_addline(buffer, "PARAM I%u = {%u, %u, %u, -1};\n", i,
3872                                 control_values[0], control_values[1], control_values[2]);
3873             }
3874             else
3875             {
3876                 compiled->int_consts[i] = next_local;
3877                 compiled->num_int_consts++;
3878                 shader_addline(buffer, "PARAM I%u = program.local[%u];\n", i, next_local++);
3879             }
3880         }
3881     }
3882
3883     /* We need a constant to fixup the final position */
3884     shader_addline(buffer, "PARAM posFixup = program.local[%u];\n", next_local);
3885     compiled->pos_fixup = next_local++;
3886
3887     /* Initialize output parameters. GL_ARB_vertex_program does not require special initialization values
3888      * for output parameters. D3D in theory does not do that either, but some applications depend on a
3889      * proper initialization of the secondary color, and programs using the fixed function pipeline without
3890      * a replacement shader depend on the texcoord.w being set properly.
3891      *
3892      * GL_NV_vertex_program defines that all output values are initialized to {0.0, 0.0, 0.0, 1.0}. This
3893      * assertion is in effect even when using GL_ARB_vertex_program without any NV specific additions. So
3894      * skip this if NV_vertex_program is supported. Otherwise, initialize the secondary color. For the tex-
3895      * coords, we have a flag in the opengl caps. Many cards do not require the texcoord being set, and
3896      * this can eat a number of instructions, so skip it unless this cap is set as well
3897      */
3898     if(!GL_SUPPORT(NV_VERTEX_PROGRAM)) {
3899         shader_addline(buffer, "MOV result.color.secondary, -helper_const.wwwy;\n");
3900
3901         if ((GLINFO_LOCATION).quirks & WINED3D_QUIRK_SET_TEXCOORD_W && !device->frag_pipe->ffp_proj_control)
3902         {
3903             int i;
3904             for(i = 0; i < min(8, MAX_REG_TEXCRD); i++) {
3905                 if(This->baseShader.reg_maps.texcoord_mask[i] != 0 &&
3906                 This->baseShader.reg_maps.texcoord_mask[i] != WINED3DSP_WRITEMASK_ALL) {
3907                     shader_addline(buffer, "MOV result.texcoord[%u].w, -helper_const.y;\n", i);
3908                 }
3909             }
3910         }
3911     }
3912
3913     /* The shader starts with the main function */
3914     priv_ctx.in_main_func = TRUE;
3915     /* Base Shader Body */
3916     shader_generate_main((IWineD3DBaseShader *)This, buffer, reg_maps, function, &priv_ctx);
3917
3918     if(!priv_ctx.footer_written) vshader_add_footer(This, buffer, args, &priv_ctx);
3919
3920     shader_addline(buffer, "END\n");
3921
3922     /* TODO: change to resource.glObjectHandle or something like that */
3923     GL_EXTCALL(glGenProgramsARB(1, &ret));
3924
3925     TRACE("Creating a hw vertex shader, prg=%d\n", ret);
3926     GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, ret));
3927
3928     TRACE("Created hw vertex shader, prg=%d\n", ret);
3929     /* Create the program and check for errors */
3930     GL_EXTCALL(glProgramStringARB(GL_VERTEX_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
3931                buffer->bsize, buffer->buffer));
3932     checkGLcall("glProgramStringARB()");
3933
3934     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &errPos);
3935     if (errPos != -1)
3936     {
3937         FIXME("HW VertexShader Error at position %d: %s\n",
3938               errPos, debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
3939         ret = -1;
3940     }
3941     else
3942     {
3943         GLint native;
3944
3945         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
3946         checkGLcall("glGetProgramivARB()");
3947         if (!native) WARN("Program exceeds native resource limits.\n");
3948
3949         /* Load immediate constants */
3950         if(lconst_map) {
3951             LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry) {
3952                 const float *value = (const float *)lconst->value;
3953                 GL_EXTCALL(glProgramLocalParameter4fvARB(GL_VERTEX_PROGRAM_ARB, lconst_map[lconst->idx], value));
3954             }
3955         }
3956     }
3957     HeapFree(GetProcessHeap(), 0, lconst_map);
3958
3959     return ret;
3960 }
3961
3962 /* GL locking is done by the caller */
3963 static struct arb_ps_compiled_shader *find_arb_pshader(IWineD3DPixelShaderImpl *shader, const struct arb_ps_compile_args *args)
3964 {
3965     UINT i;
3966     DWORD new_size;
3967     struct arb_ps_compiled_shader *new_array;
3968     struct wined3d_shader_buffer buffer;
3969     struct arb_pshader_private *shader_data;
3970     GLuint ret;
3971
3972     if(!shader->backend_priv) {
3973         IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *) shader->baseShader.device;
3974         const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
3975         struct shader_arb_priv *priv = device->shader_priv;
3976
3977         shader->backend_priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data));
3978         shader_data = shader->backend_priv;
3979         shader_data->clamp_consts = shader->baseShader.reg_maps.shader_version.major == 1;
3980
3981         if(shader->baseShader.reg_maps.shader_version.major < 3) shader_data->input_signature_idx = ~0;
3982         else shader_data->input_signature_idx = find_input_signature(priv, shader->input_signature);
3983
3984         shader_data->has_signature_idx = TRUE;
3985         TRACE("Shader got assigned input signature index %u\n", shader_data->input_signature_idx);
3986
3987         if (!device->vs_clipping)
3988             shader_data->clipplane_emulation = shader_find_free_input_register(&shader->baseShader.reg_maps,
3989                     GL_LIMITS(texture_stages) - 1);
3990         else
3991             shader_data->clipplane_emulation = ~0U;
3992     }
3993     shader_data = shader->backend_priv;
3994
3995     /* Usually we have very few GL shaders for each d3d shader(just 1 or maybe 2),
3996      * so a linear search is more performant than a hashmap or a binary search
3997      * (cache coherency etc)
3998      */
3999     for(i = 0; i < shader_data->num_gl_shaders; i++) {
4000         if(memcmp(&shader_data->gl_shaders[i].args, args, sizeof(*args)) == 0) {
4001             return &shader_data->gl_shaders[i];
4002         }
4003     }
4004
4005     TRACE("No matching GL shader found, compiling a new shader\n");
4006     if(shader_data->shader_array_size == shader_data->num_gl_shaders) {
4007         if (shader_data->num_gl_shaders)
4008         {
4009             new_size = shader_data->shader_array_size + max(1, shader_data->shader_array_size / 2);
4010             new_array = HeapReAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, shader_data->gl_shaders,
4011                                     new_size * sizeof(*shader_data->gl_shaders));
4012         } else {
4013             new_array = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data->gl_shaders));
4014             new_size = 1;
4015         }
4016
4017         if(!new_array) {
4018             ERR("Out of memory\n");
4019             return 0;
4020         }
4021         shader_data->gl_shaders = new_array;
4022         shader_data->shader_array_size = new_size;
4023     }
4024
4025     shader_data->gl_shaders[shader_data->num_gl_shaders].args = *args;
4026
4027     pixelshader_update_samplers(&shader->baseShader.reg_maps,
4028             ((IWineD3DDeviceImpl *)shader->baseShader.device)->stateBlock->textures);
4029
4030     if (!shader_buffer_init(&buffer))
4031     {
4032         ERR("Failed to initialize shader buffer.\n");
4033         return 0;
4034     }
4035
4036     ret = shader_arb_generate_pshader(shader, &buffer, args,
4037                                       &shader_data->gl_shaders[shader_data->num_gl_shaders]);
4038     shader_buffer_free(&buffer);
4039     shader_data->gl_shaders[shader_data->num_gl_shaders].prgId = ret;
4040
4041     return &shader_data->gl_shaders[shader_data->num_gl_shaders++];
4042 }
4043
4044 static inline BOOL vs_args_equal(const struct arb_vs_compile_args *stored, const struct arb_vs_compile_args *new,
4045                                  const DWORD use_map, BOOL skip_int) {
4046     if((stored->super.swizzle_map & use_map) != new->super.swizzle_map) return FALSE;
4047     if(stored->super.fog_src != new->super.fog_src) return FALSE;
4048     if(stored->boolclip_compare != new->boolclip_compare) return FALSE;
4049     if(stored->ps_signature != new->ps_signature) return FALSE;
4050     if(stored->vertex_samplers_compare != new->vertex_samplers_compare) return FALSE;
4051     if(skip_int) return TRUE;
4052
4053     return memcmp(stored->loop_ctrl, new->loop_ctrl, sizeof(stored->loop_ctrl)) == 0;
4054 }
4055
4056 static struct arb_vs_compiled_shader *find_arb_vshader(IWineD3DVertexShaderImpl *shader, const struct arb_vs_compile_args *args)
4057 {
4058     UINT i;
4059     DWORD new_size;
4060     struct arb_vs_compiled_shader *new_array;
4061     DWORD use_map = ((IWineD3DDeviceImpl *)shader->baseShader.device)->strided_streams.use_map;
4062     struct wined3d_shader_buffer buffer;
4063     struct arb_vshader_private *shader_data;
4064     GLuint ret;
4065     const struct wined3d_gl_info *gl_info = &((IWineD3DDeviceImpl *)shader->baseShader.device)->adapter->gl_info;
4066
4067     if(!shader->backend_priv) {
4068         shader->backend_priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data));
4069     }
4070     shader_data = shader->backend_priv;
4071
4072     /* Usually we have very few GL shaders for each d3d shader(just 1 or maybe 2),
4073      * so a linear search is more performant than a hashmap or a binary search
4074      * (cache coherency etc)
4075      */
4076     for(i = 0; i < shader_data->num_gl_shaders; i++) {
4077         if(vs_args_equal(&shader_data->gl_shaders[i].args, args, use_map, GL_SUPPORT(NV_VERTEX_PROGRAM2_OPTION))) {
4078             return &shader_data->gl_shaders[i];
4079         }
4080     }
4081
4082     TRACE("No matching GL shader found, compiling a new shader\n");
4083
4084     if(shader_data->shader_array_size == shader_data->num_gl_shaders) {
4085         if (shader_data->num_gl_shaders)
4086         {
4087             new_size = shader_data->shader_array_size + max(1, shader_data->shader_array_size / 2);
4088             new_array = HeapReAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, shader_data->gl_shaders,
4089                                     new_size * sizeof(*shader_data->gl_shaders));
4090         } else {
4091             new_array = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data->gl_shaders));
4092             new_size = 1;
4093         }
4094
4095         if(!new_array) {
4096             ERR("Out of memory\n");
4097             return 0;
4098         }
4099         shader_data->gl_shaders = new_array;
4100         shader_data->shader_array_size = new_size;
4101     }
4102
4103     shader_data->gl_shaders[shader_data->num_gl_shaders].args = *args;
4104
4105     if (!shader_buffer_init(&buffer))
4106     {
4107         ERR("Failed to initialize shader buffer.\n");
4108         return 0;
4109     }
4110
4111     ret = shader_arb_generate_vshader(shader, &buffer, args,
4112             &shader_data->gl_shaders[shader_data->num_gl_shaders]);
4113     shader_buffer_free(&buffer);
4114     shader_data->gl_shaders[shader_data->num_gl_shaders].prgId = ret;
4115
4116     return &shader_data->gl_shaders[shader_data->num_gl_shaders++];
4117 }
4118
4119 static inline void find_arb_ps_compile_args(IWineD3DPixelShaderImpl *shader, IWineD3DStateBlockImpl *stateblock,
4120         struct arb_ps_compile_args *args)
4121 {
4122     int i;
4123     WORD int_skip;
4124     const struct wined3d_gl_info *gl_info = &((IWineD3DDeviceImpl *)shader->baseShader.device)->adapter->gl_info;
4125     find_ps_compile_args(shader, stateblock, &args->super);
4126
4127     /* This forces all local boolean constants to 1 to make them stateblock independent */
4128     args->bools = shader->baseShader.reg_maps.local_bool_consts;
4129
4130     for(i = 0; i < MAX_CONST_B; i++)
4131     {
4132         if(stateblock->pixelShaderConstantB[i]) args->bools |= ( 1 << i);
4133     }
4134
4135     /* Only enable the clip plane emulation KIL if at least one clipplane is enabled. The KIL instruction
4136      * is quite expensive because it forces the driver to disable early Z discards. It is cheaper to
4137      * duplicate the shader than have a no-op KIL instruction in every shader
4138      */
4139     if((!((IWineD3DDeviceImpl *) shader->baseShader.device)->vs_clipping) && use_vs(stateblock) &&
4140        stateblock->renderState[WINED3DRS_CLIPPING] && stateblock->renderState[WINED3DRS_CLIPPLANEENABLE])
4141     {
4142         args->clip = 1;
4143     }
4144     else
4145     {
4146         args->clip = 0;
4147     }
4148
4149     /* Skip if unused or local, or supported natively */
4150     int_skip = ~shader->baseShader.reg_maps.integer_constants | shader->baseShader.reg_maps.local_int_consts;
4151     if(int_skip == 0xffff || GL_SUPPORT(NV_FRAGMENT_PROGRAM_OPTION))
4152     {
4153         memset(&args->loop_ctrl, 0, sizeof(args->loop_ctrl));
4154         return;
4155     }
4156
4157     for(i = 0; i < MAX_CONST_I; i++)
4158     {
4159         if(int_skip & (1 << i))
4160         {
4161             args->loop_ctrl[i][0] = 0;
4162             args->loop_ctrl[i][1] = 0;
4163             args->loop_ctrl[i][2] = 0;
4164         }
4165         else
4166         {
4167             args->loop_ctrl[i][0] = stateblock->pixelShaderConstantI[i * 4];
4168             args->loop_ctrl[i][1] = stateblock->pixelShaderConstantI[i * 4 + 1];
4169             args->loop_ctrl[i][2] = stateblock->pixelShaderConstantI[i * 4 + 2];
4170         }
4171     }
4172 }
4173
4174 static inline void find_arb_vs_compile_args(IWineD3DVertexShaderImpl *shader, IWineD3DStateBlockImpl *stateblock,
4175         struct arb_vs_compile_args *args)
4176 {
4177     int i;
4178     WORD int_skip;
4179     IWineD3DDeviceImpl *dev = (IWineD3DDeviceImpl *)shader->baseShader.device;
4180     const struct wined3d_gl_info *gl_info = &dev->adapter->gl_info;
4181     find_vs_compile_args(shader, stateblock, &args->super);
4182
4183     args->boolclip_compare = 0;
4184     if(use_ps(stateblock))
4185     {
4186         IWineD3DPixelShaderImpl *ps = (IWineD3DPixelShaderImpl *) stateblock->pixelShader;
4187         struct arb_pshader_private *shader_priv = ps->backend_priv;
4188         args->ps_signature = shader_priv->input_signature_idx;
4189
4190         args->boolclip.clip_control[0] = shader_priv->clipplane_emulation + 1;
4191     }
4192     else
4193     {
4194         args->ps_signature = ~0;
4195         if(!dev->vs_clipping)
4196         {
4197             args->boolclip.clip_control[0] = ffp_clip_emul(stateblock) ? GL_LIMITS(texture_stages) : 0;
4198         }
4199         /* Otherwise: Setting boolclip_compare set clip_control[0] to 0 */
4200     }
4201
4202     if(args->boolclip.clip_control[0])
4203     {
4204         if(stateblock->renderState[WINED3DRS_CLIPPING])
4205         {
4206             args->boolclip.clip_control[1] = stateblock->renderState[WINED3DRS_CLIPPLANEENABLE];
4207         }
4208         /* clip_control[1] was set to 0 by setting boolclip_compare to 0 */
4209     }
4210
4211     /* This forces all local boolean constants to 1 to make them stateblock independent */
4212     args->boolclip.bools = shader->baseShader.reg_maps.local_bool_consts;
4213     /* TODO: Figure out if it would be better to store bool constants as bitmasks in the stateblock */
4214     for(i = 0; i < MAX_CONST_B; i++)
4215     {
4216         if(stateblock->vertexShaderConstantB[i]) args->boolclip.bools |= ( 1 << i);
4217     }
4218
4219     args->vertex_samplers[0] = dev->texUnitMap[MAX_FRAGMENT_SAMPLERS + 0];
4220     args->vertex_samplers[1] = dev->texUnitMap[MAX_FRAGMENT_SAMPLERS + 1];
4221     args->vertex_samplers[2] = dev->texUnitMap[MAX_FRAGMENT_SAMPLERS + 2];
4222     args->vertex_samplers[3] = 0;
4223
4224     /* Skip if unused or local */
4225     int_skip = ~shader->baseShader.reg_maps.integer_constants | shader->baseShader.reg_maps.local_int_consts;
4226     if(int_skip == 0xffff || GL_SUPPORT(NV_VERTEX_PROGRAM2_OPTION)) /* This is about flow control, not clipping */
4227     {
4228         memset(&args->loop_ctrl, 0, sizeof(args->loop_ctrl));
4229         return;
4230     }
4231
4232     for(i = 0; i < MAX_CONST_I; i++)
4233     {
4234         if(int_skip & (1 << i))
4235         {
4236             args->loop_ctrl[i][0] = 0;
4237             args->loop_ctrl[i][1] = 0;
4238             args->loop_ctrl[i][2] = 0;
4239         }
4240         else
4241         {
4242             args->loop_ctrl[i][0] = stateblock->vertexShaderConstantI[i * 4];
4243             args->loop_ctrl[i][1] = stateblock->vertexShaderConstantI[i * 4 + 1];
4244             args->loop_ctrl[i][2] = stateblock->vertexShaderConstantI[i * 4 + 2];
4245         }
4246     }
4247 }
4248
4249 /* GL locking is done by the caller */
4250 static void shader_arb_select(const struct wined3d_context *context, BOOL usePS, BOOL useVS)
4251 {
4252     IWineD3DDeviceImpl *This = ((IWineD3DSurfaceImpl *)context->surface)->resource.wineD3DDevice;
4253     struct shader_arb_priv *priv = This->shader_priv;
4254     const struct wined3d_gl_info *gl_info = context->gl_info;
4255     int i;
4256
4257     /* Deal with pixel shaders first so the vertex shader arg function has the input signature ready */
4258     if (usePS) {
4259         struct arb_ps_compile_args compile_args;
4260         struct arb_ps_compiled_shader *compiled;
4261         IWineD3DPixelShaderImpl *ps = (IWineD3DPixelShaderImpl *) This->stateBlock->pixelShader;
4262
4263         TRACE("Using pixel shader %p\n", This->stateBlock->pixelShader);
4264         find_arb_ps_compile_args(ps, This->stateBlock, &compile_args);
4265         compiled = find_arb_pshader(ps, &compile_args);
4266         priv->current_fprogram_id = compiled->prgId;
4267         priv->compiled_fprog = compiled;
4268
4269         /* Bind the fragment program */
4270         GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, priv->current_fprogram_id));
4271         checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, priv->current_fprogram_id);");
4272
4273         if(!priv->use_arbfp_fixed_func) {
4274             /* Enable OpenGL fragment programs */
4275             glEnable(GL_FRAGMENT_PROGRAM_ARB);
4276             checkGLcall("glEnable(GL_FRAGMENT_PROGRAM_ARB);");
4277         }
4278         TRACE("(%p) : Bound fragment program %u and enabled GL_FRAGMENT_PROGRAM_ARB\n", This, priv->current_fprogram_id);
4279
4280         /* Pixel Shader 1.x constants are clamped to [-1;1], Pixel Shader 2.0 constants are not. If switching between
4281          * a 1.x and newer shader, reload the first 8 constants
4282          */
4283         if(priv->last_ps_const_clamped != ((struct arb_pshader_private *) ps->backend_priv)->clamp_consts)
4284         {
4285             priv->last_ps_const_clamped = ((struct arb_pshader_private *) ps->backend_priv)->clamp_consts;
4286             This->highest_dirty_ps_const = max(This->highest_dirty_ps_const, 8);
4287             for(i = 0; i < 8; i++)
4288             {
4289                 context->pshader_const_dirty[i] = 1;
4290             }
4291             /* Also takes care of loading local constants */
4292             shader_arb_load_constants(context, TRUE, FALSE);
4293         }
4294         else
4295         {
4296             shader_arb_ps_local_constants(This);
4297         }
4298
4299         /* Force constant reloading for the NP2 fixup (see comment in shader_glsl_select for more info) */
4300         if (compiled->np2fixup_info.super.active)
4301             shader_arb_load_np2fixup_constants((IWineD3DDevice *)This, usePS, useVS);
4302     } else if(GL_SUPPORT(ARB_FRAGMENT_PROGRAM) && !priv->use_arbfp_fixed_func) {
4303         /* Disable only if we're not using arbfp fixed function fragment processing. If this is used,
4304         * keep GL_FRAGMENT_PROGRAM_ARB enabled, and the fixed function pipeline will bind the fixed function
4305         * replacement shader
4306         */
4307         glDisable(GL_FRAGMENT_PROGRAM_ARB);
4308         checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
4309         priv->current_fprogram_id = 0;
4310     }
4311
4312     if (useVS) {
4313         struct arb_vs_compile_args compile_args;
4314         struct arb_vs_compiled_shader *compiled;
4315         IWineD3DVertexShaderImpl *vs = (IWineD3DVertexShaderImpl *) This->stateBlock->vertexShader;
4316
4317         TRACE("Using vertex shader %p\n", This->stateBlock->vertexShader);
4318         find_arb_vs_compile_args(vs, This->stateBlock, &compile_args);
4319         compiled = find_arb_vshader(vs, &compile_args);
4320         priv->current_vprogram_id = compiled->prgId;
4321         priv->compiled_vprog = compiled;
4322
4323         /* Bind the vertex program */
4324         GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->current_vprogram_id));
4325         checkGLcall("glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->current_vprogram_id);");
4326
4327         /* Enable OpenGL vertex programs */
4328         glEnable(GL_VERTEX_PROGRAM_ARB);
4329         checkGLcall("glEnable(GL_VERTEX_PROGRAM_ARB);");
4330         TRACE("(%p) : Bound vertex program %u and enabled GL_VERTEX_PROGRAM_ARB\n", This, priv->current_vprogram_id);
4331         shader_arb_vs_local_constants(This);
4332
4333         if(priv->last_vs_color_unclamp != compiled->need_color_unclamp) {
4334             priv->last_vs_color_unclamp = compiled->need_color_unclamp;
4335
4336             if (GL_SUPPORT(ARB_COLOR_BUFFER_FLOAT)) {
4337                 GL_EXTCALL(glClampColorARB(GL_CLAMP_VERTEX_COLOR_ARB, !compiled->need_color_unclamp));
4338                 checkGLcall("glClampColorARB");
4339             } else {
4340                 FIXME("vertex color clamp needs to be changed, but extension not supported.\n");
4341             }
4342         }
4343     } else if(GL_SUPPORT(ARB_VERTEX_PROGRAM)) {
4344         priv->current_vprogram_id = 0;
4345         glDisable(GL_VERTEX_PROGRAM_ARB);
4346         checkGLcall("glDisable(GL_VERTEX_PROGRAM_ARB)");
4347     }
4348 }
4349
4350 /* GL locking is done by the caller */
4351 static void shader_arb_select_depth_blt(IWineD3DDevice *iface, enum tex_types tex_type) {
4352     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
4353     struct shader_arb_priv *priv = This->shader_priv;
4354     GLuint *blt_fprogram = &priv->depth_blt_fprogram_id[tex_type];
4355     const struct wined3d_gl_info *gl_info = &This->adapter->gl_info;
4356
4357     if (!priv->depth_blt_vprogram_id) priv->depth_blt_vprogram_id = create_arb_blt_vertex_program(gl_info);
4358     GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->depth_blt_vprogram_id));
4359     glEnable(GL_VERTEX_PROGRAM_ARB);
4360
4361     if (!*blt_fprogram) *blt_fprogram = create_arb_blt_fragment_program(gl_info, tex_type);
4362     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, *blt_fprogram));
4363     glEnable(GL_FRAGMENT_PROGRAM_ARB);
4364 }
4365
4366 /* GL locking is done by the caller */
4367 static void shader_arb_deselect_depth_blt(IWineD3DDevice *iface) {
4368     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
4369     struct shader_arb_priv *priv = This->shader_priv;
4370     const struct wined3d_gl_info *gl_info = &This->adapter->gl_info;
4371
4372     if (priv->current_vprogram_id) {
4373         GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->current_vprogram_id));
4374         checkGLcall("glBindProgramARB(GL_VERTEX_PROGRAM_ARB, vertexShader->prgId);");
4375
4376         TRACE("(%p) : Bound vertex program %u and enabled GL_VERTEX_PROGRAM_ARB\n", This, priv->current_vprogram_id);
4377     } else {
4378         glDisable(GL_VERTEX_PROGRAM_ARB);
4379         checkGLcall("glDisable(GL_VERTEX_PROGRAM_ARB)");
4380     }
4381
4382     if (priv->current_fprogram_id) {
4383         GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, priv->current_fprogram_id));
4384         checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, pixelShader->prgId);");
4385
4386         TRACE("(%p) : Bound fragment program %u and enabled GL_FRAGMENT_PROGRAM_ARB\n", This, priv->current_fprogram_id);
4387     } else if(!priv->use_arbfp_fixed_func) {
4388         glDisable(GL_FRAGMENT_PROGRAM_ARB);
4389         checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
4390     }
4391 }
4392
4393 static void shader_arb_destroy(IWineD3DBaseShader *iface) {
4394     IWineD3DBaseShaderImpl *baseShader = (IWineD3DBaseShaderImpl *) iface;
4395     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *)baseShader->baseShader.device;
4396     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4397
4398     if (shader_is_pshader_version(baseShader->baseShader.reg_maps.shader_version.type))
4399     {
4400         IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *) iface;
4401         struct arb_pshader_private *shader_data = This->backend_priv;
4402         UINT i;
4403
4404         if(!shader_data) return; /* This can happen if a shader was never compiled */
4405         ENTER_GL();
4406
4407         if(shader_data->num_gl_shaders) ActivateContext(device, NULL, CTXUSAGE_RESOURCELOAD);
4408
4409         for(i = 0; i < shader_data->num_gl_shaders; i++) {
4410             GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId));
4411             checkGLcall("GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId))");
4412         }
4413         LEAVE_GL();
4414         HeapFree(GetProcessHeap(), 0, shader_data->gl_shaders);
4415         HeapFree(GetProcessHeap(), 0, shader_data);
4416         This->backend_priv = NULL;
4417     } else {
4418         IWineD3DVertexShaderImpl *This = (IWineD3DVertexShaderImpl *) iface;
4419         struct arb_vshader_private *shader_data = This->backend_priv;
4420         UINT i;
4421
4422         if(!shader_data) return; /* This can happen if a shader was never compiled */
4423         ENTER_GL();
4424
4425         if(shader_data->num_gl_shaders) ActivateContext(device, NULL, CTXUSAGE_RESOURCELOAD);
4426
4427         for(i = 0; i < shader_data->num_gl_shaders; i++) {
4428             GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId));
4429             checkGLcall("GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId))");
4430         }
4431         LEAVE_GL();
4432         HeapFree(GetProcessHeap(), 0, shader_data->gl_shaders);
4433         HeapFree(GetProcessHeap(), 0, shader_data);
4434         This->backend_priv = NULL;
4435     }
4436 }
4437
4438 static int sig_tree_compare(const void *key, const struct wine_rb_entry *entry)
4439 {
4440     struct ps_signature *e = WINE_RB_ENTRY_VALUE(entry, struct ps_signature, entry);
4441     return compare_sig(key, e->sig);
4442 }
4443
4444 struct wine_rb_functions sig_tree_functions =
4445 {
4446     wined3d_rb_alloc,
4447     wined3d_rb_realloc,
4448     wined3d_rb_free,
4449     sig_tree_compare
4450 };
4451
4452 static HRESULT shader_arb_alloc(IWineD3DDevice *iface) {
4453     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
4454     struct shader_arb_priv *priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*priv));
4455     if(wine_rb_init(&priv->signature_tree, &sig_tree_functions) == -1)
4456     {
4457         ERR("RB tree init failed\n");
4458         HeapFree(GetProcessHeap(), 0, priv);
4459         return E_OUTOFMEMORY;
4460     }
4461     This->shader_priv = priv;
4462     return WINED3D_OK;
4463 }
4464
4465 static void release_signature(struct wine_rb_entry *entry, void *context)
4466 {
4467     struct ps_signature *sig = WINE_RB_ENTRY_VALUE(entry, struct ps_signature, entry);
4468     int i;
4469     for(i = 0; i < MAX_REG_INPUT; i++)
4470     {
4471         HeapFree(GetProcessHeap(), 0, (char *) sig->sig[i].semantic_name);
4472     }
4473     HeapFree(GetProcessHeap(), 0, sig->sig);
4474     HeapFree(GetProcessHeap(), 0, sig);
4475 }
4476
4477 /* Context activation is done by the caller. */
4478 static void shader_arb_free(IWineD3DDevice *iface) {
4479     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
4480     const struct wined3d_gl_info *gl_info = &This->adapter->gl_info;
4481     struct shader_arb_priv *priv = This->shader_priv;
4482     int i;
4483
4484     ENTER_GL();
4485     if(priv->depth_blt_vprogram_id) {
4486         GL_EXTCALL(glDeleteProgramsARB(1, &priv->depth_blt_vprogram_id));
4487     }
4488     for (i = 0; i < tex_type_count; ++i) {
4489         if (priv->depth_blt_fprogram_id[i]) {
4490             GL_EXTCALL(glDeleteProgramsARB(1, &priv->depth_blt_fprogram_id[i]));
4491         }
4492     }
4493     LEAVE_GL();
4494
4495     wine_rb_destroy(&priv->signature_tree, release_signature, NULL);
4496     HeapFree(GetProcessHeap(), 0, This->shader_priv);
4497 }
4498
4499 static BOOL shader_arb_dirty_const(IWineD3DDevice *iface) {
4500     return TRUE;
4501 }
4502
4503 static void shader_arb_get_caps(WINED3DDEVTYPE devtype, const struct wined3d_gl_info *gl_info,
4504         struct shader_caps *pCaps)
4505 {
4506     /* We don't have an ARB fixed function pipeline yet, so let the none backend set its caps,
4507      * then overwrite the shader specific ones
4508      */
4509     none_shader_backend.shader_get_caps(devtype, gl_info, pCaps);
4510
4511     if(GL_SUPPORT(ARB_VERTEX_PROGRAM)) {
4512         if(GL_SUPPORT(NV_VERTEX_PROGRAM3))
4513         {
4514             pCaps->VertexShaderVersion = WINED3DVS_VERSION(3,0);
4515             TRACE_(d3d_caps)("Hardware vertex shader version 3.0 enabled (NV_VERTEX_PROGRAM3)\n");
4516         }
4517         else if(GL_LIMITS(vshader_constantsF) >= 256)
4518         {
4519             /* Shader Model 2.0 requires at least 256 vertex shader constants */
4520             pCaps->VertexShaderVersion = WINED3DVS_VERSION(2,0);
4521             TRACE_(d3d_caps)("Hardware vertex shader version 2.0 enabled (ARB_PROGRAM)\n");
4522         }
4523         else
4524         {
4525             pCaps->VertexShaderVersion = WINED3DVS_VERSION(1,1);
4526             TRACE_(d3d_caps)("Hardware vertex shader version 1.1 enabled (ARB_PROGRAM)\n");
4527         }
4528         pCaps->MaxVertexShaderConst = GL_LIMITS(vshader_constantsF);
4529     }
4530
4531     if(GL_SUPPORT(ARB_FRAGMENT_PROGRAM)) {
4532         if(GL_SUPPORT(NV_FRAGMENT_PROGRAM2))
4533         {
4534             pCaps->PixelShaderVersion    = WINED3DPS_VERSION(3,0);
4535             TRACE_(d3d_caps)("Hardware pixel shader version 3.0 enabled (NV_FRAGMENT_PROGRAM2)\n");
4536         }
4537         else if(GL_LIMITS(pshader_constantsF) >= 32)
4538         {
4539             /* Shader Model 2.0 requires at least 32 pixel shader constants */
4540             pCaps->PixelShaderVersion    = WINED3DPS_VERSION(2,0);
4541             TRACE_(d3d_caps)("Hardware pixel shader version 2.0 enabled (ARB_PROGRAM)\n");
4542         }
4543         else
4544         {
4545             pCaps->PixelShaderVersion    = WINED3DPS_VERSION(1,4);
4546             TRACE_(d3d_caps)("Hardware pixel shader version 1.4 enabled (ARB_PROGRAM)\n");
4547         }
4548         pCaps->PixelShader1xMaxValue = 8.0f;
4549         pCaps->MaxPixelShaderConst = GL_LIMITS(pshader_constantsF);
4550     }
4551
4552     pCaps->VSClipping = use_nv_clip(gl_info);
4553 }
4554
4555 static BOOL shader_arb_color_fixup_supported(struct color_fixup_desc fixup)
4556 {
4557     if (TRACE_ON(d3d_shader) && TRACE_ON(d3d))
4558     {
4559         TRACE("Checking support for color_fixup:\n");
4560         dump_color_fixup_desc(fixup);
4561     }
4562
4563     /* We support everything except YUV conversions. */
4564     if (!is_yuv_fixup(fixup))
4565     {
4566         TRACE("[OK]\n");
4567         return TRUE;
4568     }
4569
4570     TRACE("[FAILED]\n");
4571     return FALSE;
4572 }
4573
4574 static void shader_arb_add_instruction_modifiers(const struct wined3d_shader_instruction *ins) {
4575     DWORD shift;
4576     char write_mask[20], regstr[50];
4577     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
4578     BOOL is_color = FALSE;
4579     const struct wined3d_shader_dst_param *dst;
4580
4581     if (!ins->dst_count) return;
4582
4583     dst = &ins->dst[0];
4584     shift = dst->shift;
4585     if(shift == 0) return; /* Saturate alone is handled by the instructions */
4586
4587     shader_arb_get_write_mask(ins, dst, write_mask);
4588     shader_arb_get_register_name(ins, &dst->reg, regstr, &is_color);
4589
4590     /* Generate a line that does the output modifier computation
4591      * FIXME: _SAT vs shift? _SAT alone is already handled in the instructions, if this
4592      * maps problems in e.g. _d4_sat modify shader_arb_get_modifier
4593      */
4594     shader_addline(buffer, "MUL%s %s%s, %s, %s;\n", shader_arb_get_modifier(ins),
4595                    regstr, write_mask, regstr, shift_tab[shift]);
4596 }
4597
4598 static const SHADER_HANDLER shader_arb_instruction_handler_table[WINED3DSIH_TABLE_SIZE] =
4599 {
4600     /* WINED3DSIH_ABS           */ shader_hw_map2gl,
4601     /* WINED3DSIH_ADD           */ shader_hw_map2gl,
4602     /* WINED3DSIH_BEM           */ pshader_hw_bem,
4603     /* WINED3DSIH_BREAK         */ shader_hw_break,
4604     /* WINED3DSIH_BREAKC        */ shader_hw_breakc,
4605     /* WINED3DSIH_BREAKP        */ NULL,
4606     /* WINED3DSIH_CALL          */ shader_hw_call,
4607     /* WINED3DSIH_CALLNZ        */ NULL,
4608     /* WINED3DSIH_CMP           */ pshader_hw_cmp,
4609     /* WINED3DSIH_CND           */ pshader_hw_cnd,
4610     /* WINED3DSIH_CRS           */ shader_hw_map2gl,
4611     /* WINED3DSIH_DCL           */ NULL,
4612     /* WINED3DSIH_DEF           */ NULL,
4613     /* WINED3DSIH_DEFB          */ NULL,
4614     /* WINED3DSIH_DEFI          */ NULL,
4615     /* WINED3DSIH_DP2ADD        */ pshader_hw_dp2add,
4616     /* WINED3DSIH_DP3           */ shader_hw_map2gl,
4617     /* WINED3DSIH_DP4           */ shader_hw_map2gl,
4618     /* WINED3DSIH_DST           */ shader_hw_map2gl,
4619     /* WINED3DSIH_DSX           */ shader_hw_map2gl,
4620     /* WINED3DSIH_DSY           */ shader_hw_dsy,
4621     /* WINED3DSIH_ELSE          */ shader_hw_else,
4622     /* WINED3DSIH_ENDIF         */ shader_hw_endif,
4623     /* WINED3DSIH_ENDLOOP       */ shader_hw_endloop,
4624     /* WINED3DSIH_ENDREP        */ shader_hw_endrep,
4625     /* WINED3DSIH_EXP           */ shader_hw_scalar_op,
4626     /* WINED3DSIH_EXPP          */ shader_hw_scalar_op,
4627     /* WINED3DSIH_FRC           */ shader_hw_map2gl,
4628     /* WINED3DSIH_IF            */ NULL /* Hardcoded into the shader */,
4629     /* WINED3DSIH_IFC           */ shader_hw_ifc,
4630     /* WINED3DSIH_LABEL         */ shader_hw_label,
4631     /* WINED3DSIH_LIT           */ shader_hw_map2gl,
4632     /* WINED3DSIH_LOG           */ shader_hw_log_pow,
4633     /* WINED3DSIH_LOGP          */ shader_hw_log_pow,
4634     /* WINED3DSIH_LOOP          */ shader_hw_loop,
4635     /* WINED3DSIH_LRP           */ shader_hw_lrp,
4636     /* WINED3DSIH_M3x2          */ shader_hw_mnxn,
4637     /* WINED3DSIH_M3x3          */ shader_hw_mnxn,
4638     /* WINED3DSIH_M3x4          */ shader_hw_mnxn,
4639     /* WINED3DSIH_M4x3          */ shader_hw_mnxn,
4640     /* WINED3DSIH_M4x4          */ shader_hw_mnxn,
4641     /* WINED3DSIH_MAD           */ shader_hw_map2gl,
4642     /* WINED3DSIH_MAX           */ shader_hw_map2gl,
4643     /* WINED3DSIH_MIN           */ shader_hw_map2gl,
4644     /* WINED3DSIH_MOV           */ shader_hw_mov,
4645     /* WINED3DSIH_MOVA          */ shader_hw_mov,
4646     /* WINED3DSIH_MUL           */ shader_hw_map2gl,
4647     /* WINED3DSIH_NOP           */ shader_hw_nop,
4648     /* WINED3DSIH_NRM           */ shader_hw_nrm,
4649     /* WINED3DSIH_PHASE         */ NULL,
4650     /* WINED3DSIH_POW           */ shader_hw_log_pow,
4651     /* WINED3DSIH_RCP           */ shader_hw_scalar_op,
4652     /* WINED3DSIH_REP           */ shader_hw_rep,
4653     /* WINED3DSIH_RET           */ shader_hw_ret,
4654     /* WINED3DSIH_RSQ           */ shader_hw_scalar_op,
4655     /* WINED3DSIH_SETP          */ NULL,
4656     /* WINED3DSIH_SGE           */ shader_hw_map2gl,
4657     /* WINED3DSIH_SGN           */ shader_hw_sgn,
4658     /* WINED3DSIH_SINCOS        */ shader_hw_sincos,
4659     /* WINED3DSIH_SLT           */ shader_hw_map2gl,
4660     /* WINED3DSIH_SUB           */ shader_hw_map2gl,
4661     /* WINED3DSIH_TEX           */ pshader_hw_tex,
4662     /* WINED3DSIH_TEXBEM        */ pshader_hw_texbem,
4663     /* WINED3DSIH_TEXBEML       */ pshader_hw_texbem,
4664     /* WINED3DSIH_TEXCOORD      */ pshader_hw_texcoord,
4665     /* WINED3DSIH_TEXDEPTH      */ pshader_hw_texdepth,
4666     /* WINED3DSIH_TEXDP3        */ pshader_hw_texdp3,
4667     /* WINED3DSIH_TEXDP3TEX     */ pshader_hw_texdp3tex,
4668     /* WINED3DSIH_TEXKILL       */ pshader_hw_texkill,
4669     /* WINED3DSIH_TEXLDD        */ shader_hw_texldd,
4670     /* WINED3DSIH_TEXLDL        */ shader_hw_texldl,
4671     /* WINED3DSIH_TEXM3x2DEPTH  */ pshader_hw_texm3x2depth,
4672     /* WINED3DSIH_TEXM3x2PAD    */ pshader_hw_texm3x2pad,
4673     /* WINED3DSIH_TEXM3x2TEX    */ pshader_hw_texm3x2tex,
4674     /* WINED3DSIH_TEXM3x3       */ pshader_hw_texm3x3,
4675     /* WINED3DSIH_TEXM3x3DIFF   */ NULL,
4676     /* WINED3DSIH_TEXM3x3PAD    */ pshader_hw_texm3x3pad,
4677     /* WINED3DSIH_TEXM3x3SPEC   */ pshader_hw_texm3x3spec,
4678     /* WINED3DSIH_TEXM3x3TEX    */ pshader_hw_texm3x3tex,
4679     /* WINED3DSIH_TEXM3x3VSPEC  */ pshader_hw_texm3x3vspec,
4680     /* WINED3DSIH_TEXREG2AR     */ pshader_hw_texreg2ar,
4681     /* WINED3DSIH_TEXREG2GB     */ pshader_hw_texreg2gb,
4682     /* WINED3DSIH_TEXREG2RGB    */ pshader_hw_texreg2rgb,
4683 };
4684
4685 static inline BOOL get_bool_const(const struct wined3d_shader_instruction *ins, IWineD3DBaseShaderImpl *This, DWORD idx)
4686 {
4687     BOOL vshader = shader_is_vshader_version(This->baseShader.reg_maps.shader_version.type);
4688     WORD bools = 0;
4689     WORD flag = (1 << idx);
4690     const local_constant *constant;
4691     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
4692
4693     if(This->baseShader.reg_maps.local_bool_consts & flag)
4694     {
4695         /* What good is a if(bool) with a hardcoded local constant? I don't know, but handle it */
4696         LIST_FOR_EACH_ENTRY(constant, &This->baseShader.constantsB, local_constant, entry)
4697         {
4698             if (constant->idx == idx)
4699             {
4700                 return constant->value[0];
4701             }
4702         }
4703         ERR("Local constant not found\n");
4704         return FALSE;
4705     }
4706     else
4707     {
4708         if(vshader) bools = priv->cur_vs_args->boolclip.bools;
4709         else bools = priv->cur_ps_args->bools;
4710         return bools & flag;
4711     }
4712 }
4713
4714 static void get_loop_control_const(const struct wined3d_shader_instruction *ins,
4715         IWineD3DBaseShaderImpl *This, UINT idx, struct wined3d_shader_loop_control *loop_control)
4716 {
4717     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
4718
4719     /* Integer constants can either be a local constant, or they can be stored in the shader
4720      * type specific compile args. */
4721     if (This->baseShader.reg_maps.local_int_consts & (1 << idx))
4722     {
4723         const local_constant *constant;
4724
4725         LIST_FOR_EACH_ENTRY(constant, &This->baseShader.constantsI, local_constant, entry)
4726         {
4727             if (constant->idx == idx)
4728             {
4729                 loop_control->count = constant->value[0];
4730                 loop_control->start = constant->value[1];
4731                 /* Step is signed. */
4732                 loop_control->step = (int)constant->value[2];
4733                 return;
4734             }
4735         }
4736         /* If this happens the flag was set incorrectly */
4737         ERR("Local constant not found\n");
4738         loop_control->count = 0;
4739         loop_control->start = 0;
4740         loop_control->step = 0;
4741         return;
4742     }
4743
4744     switch (This->baseShader.reg_maps.shader_version.type)
4745     {
4746         case WINED3D_SHADER_TYPE_VERTEX:
4747             /* Count and aL start value are unsigned */
4748             loop_control->count = priv->cur_vs_args->loop_ctrl[idx][0];
4749             loop_control->start = priv->cur_vs_args->loop_ctrl[idx][1];
4750             /* Step is signed. */
4751             loop_control->step = ((char)priv->cur_vs_args->loop_ctrl[idx][2]);
4752             break;
4753
4754         case WINED3D_SHADER_TYPE_PIXEL:
4755             loop_control->count = priv->cur_ps_args->loop_ctrl[idx][0];
4756             loop_control->start = priv->cur_ps_args->loop_ctrl[idx][1];
4757             loop_control->step = ((char)priv->cur_ps_args->loop_ctrl[idx][2]);
4758             break;
4759
4760         default:
4761             FIXME("Unhandled shader type %#x.\n", This->baseShader.reg_maps.shader_version.type);
4762             break;
4763     }
4764 }
4765
4766 static void record_instruction(struct list *list, const struct wined3d_shader_instruction *ins)
4767 {
4768     unsigned int i;
4769     struct wined3d_shader_dst_param *dst_param = NULL;
4770     struct wined3d_shader_src_param *src_param = NULL, *rel_addr = NULL;
4771     struct recorded_instruction *rec = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*rec));
4772     if(!rec)
4773     {
4774         ERR("Out of memory\n");
4775         return;
4776     }
4777
4778     rec->ins = *ins;
4779     dst_param = HeapAlloc(GetProcessHeap(), 0, sizeof(*dst_param));
4780     if(!dst_param) goto free;
4781     *dst_param = *ins->dst;
4782     if(ins->dst->reg.rel_addr)
4783     {
4784         rel_addr = HeapAlloc(GetProcessHeap(), 0, sizeof(*dst_param->reg.rel_addr));
4785         if(!rel_addr) goto free;
4786         *rel_addr = *ins->dst->reg.rel_addr;
4787         dst_param->reg.rel_addr = rel_addr;
4788     }
4789     rec->ins.dst = dst_param;
4790
4791     src_param = HeapAlloc(GetProcessHeap(), 0, sizeof(*src_param) * ins->src_count);
4792     if(!src_param) goto free;
4793     for(i = 0; i < ins->src_count; i++)
4794     {
4795         src_param[i] = ins->src[i];
4796         if(ins->src[i].reg.rel_addr)
4797         {
4798             rel_addr = HeapAlloc(GetProcessHeap(), 0, sizeof(*rel_addr));
4799             if(!rel_addr) goto free;
4800             *rel_addr = *ins->src[i].reg.rel_addr;
4801             src_param[i].reg.rel_addr = rel_addr;
4802         }
4803     }
4804     rec->ins.src = src_param;
4805     list_add_tail(list, &rec->entry);
4806     return;
4807
4808 free:
4809     ERR("Out of memory\n");
4810     if(dst_param)
4811     {
4812         HeapFree(GetProcessHeap(), 0, (void *) dst_param->reg.rel_addr);
4813         HeapFree(GetProcessHeap(), 0, dst_param);
4814     }
4815     if(src_param)
4816     {
4817         for(i = 0; i < ins->src_count; i++)
4818         {
4819             HeapFree(GetProcessHeap(), 0, (void *) src_param[i].reg.rel_addr);
4820         }
4821         HeapFree(GetProcessHeap(), 0, src_param);
4822     }
4823     HeapFree(GetProcessHeap(), 0, rec);
4824 }
4825
4826 static void free_recorded_instruction(struct list *list)
4827 {
4828     struct recorded_instruction *rec_ins, *entry2;
4829     unsigned int i;
4830
4831     LIST_FOR_EACH_ENTRY_SAFE(rec_ins, entry2, list, struct recorded_instruction, entry)
4832     {
4833         list_remove(&rec_ins->entry);
4834         if(rec_ins->ins.dst)
4835         {
4836             HeapFree(GetProcessHeap(), 0, (void *) rec_ins->ins.dst->reg.rel_addr);
4837             HeapFree(GetProcessHeap(), 0, (void *) rec_ins->ins.dst);
4838         }
4839         if(rec_ins->ins.src)
4840         {
4841             for(i = 0; i < rec_ins->ins.src_count; i++)
4842             {
4843                 HeapFree(GetProcessHeap(), 0, (void *) rec_ins->ins.src[i].reg.rel_addr);
4844             }
4845             HeapFree(GetProcessHeap(), 0, (void *) rec_ins->ins.src);
4846         }
4847         HeapFree(GetProcessHeap(), 0, rec_ins);
4848     }
4849 }
4850
4851 static void shader_arb_handle_instruction(const struct wined3d_shader_instruction *ins) {
4852     SHADER_HANDLER hw_fct;
4853     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
4854     IWineD3DBaseShaderImpl *This = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
4855     struct control_frame *control_frame;
4856     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
4857     BOOL bool_const;
4858
4859     if(ins->handler_idx == WINED3DSIH_LOOP || ins->handler_idx == WINED3DSIH_REP)
4860     {
4861         control_frame = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*control_frame));
4862         list_add_head(&priv->control_frames, &control_frame->entry);
4863
4864         if(ins->handler_idx == WINED3DSIH_LOOP) control_frame->type = LOOP;
4865         if(ins->handler_idx == WINED3DSIH_REP) control_frame->type = REP;
4866
4867         if(priv->target_version >= NV2)
4868         {
4869             control_frame->loop_no = priv->num_loops++;
4870             priv->loop_depth++;
4871         }
4872         else
4873         {
4874             /* Don't bother recording when we're in a not used if branch */
4875             if(priv->muted)
4876             {
4877                 return;
4878             }
4879
4880             if(!priv->recording)
4881             {
4882                 list_init(&priv->record);
4883                 priv->recording = TRUE;
4884                 control_frame->outer_loop = TRUE;
4885                 get_loop_control_const(ins, This, ins->src[0].reg.idx, &control_frame->loop_control);
4886                 return; /* Instruction is handled */
4887             }
4888             /* Record this loop in the outer loop's recording */
4889         }
4890     }
4891     else if(ins->handler_idx == WINED3DSIH_ENDLOOP || ins->handler_idx == WINED3DSIH_ENDREP)
4892     {
4893         if(priv->target_version >= NV2)
4894         {
4895             /* Nothing to do. The control frame is popped after the HW instr handler */
4896         }
4897         else
4898         {
4899             struct list *e = list_head(&priv->control_frames);
4900             control_frame = LIST_ENTRY(e, struct control_frame, entry);
4901             list_remove(&control_frame->entry);
4902
4903             if(control_frame->outer_loop)
4904             {
4905                 int iteration, aL = 0;
4906                 struct list copy;
4907
4908                 /* Turn off recording before playback */
4909                 priv->recording = FALSE;
4910
4911                 /* Move the recorded instructions to a separate list and get them out of the private data
4912                  * structure. If there are nested loops, the shader_arb_handle_instruction below will
4913                  * be recorded again, thus priv->record might be overwritten
4914                  */
4915                 list_init(&copy);
4916                 list_move_tail(&copy, &priv->record);
4917                 list_init(&priv->record);
4918
4919                 if(ins->handler_idx == WINED3DSIH_ENDLOOP)
4920                 {
4921                     shader_addline(buffer, "#unrolling loop: %u iterations, aL=%u, inc %d\n",
4922                                    control_frame->loop_control.count, control_frame->loop_control.start,
4923                                    control_frame->loop_control.step);
4924                     aL = control_frame->loop_control.start;
4925                 }
4926                 else
4927                 {
4928                     shader_addline(buffer, "#unrolling rep: %u iterations\n", control_frame->loop_control.count);
4929                 }
4930
4931                 for (iteration = 0; iteration < control_frame->loop_control.count; ++iteration)
4932                 {
4933                     struct recorded_instruction *rec_ins;
4934                     if(ins->handler_idx == WINED3DSIH_ENDLOOP)
4935                     {
4936                         priv->aL = aL;
4937                         shader_addline(buffer, "#Iteration %d, aL=%d\n", iteration, aL);
4938                     }
4939                     else
4940                     {
4941                         shader_addline(buffer, "#Iteration %d\n", iteration);
4942                     }
4943
4944                     LIST_FOR_EACH_ENTRY(rec_ins, &copy, struct recorded_instruction, entry)
4945                     {
4946                         shader_arb_handle_instruction(&rec_ins->ins);
4947                     }
4948
4949                     if(ins->handler_idx == WINED3DSIH_ENDLOOP)
4950                     {
4951                         aL += control_frame->loop_control.step;
4952                     }
4953                 }
4954                 shader_addline(buffer, "#end loop/rep\n");
4955
4956                 free_recorded_instruction(&copy);
4957                 HeapFree(GetProcessHeap(), 0, control_frame);
4958                 return; /* Instruction is handled */
4959             }
4960             else
4961             {
4962                 /* This is a nested loop. Proceed to the normal recording function */
4963                 HeapFree(GetProcessHeap(), 0, control_frame);
4964             }
4965         }
4966     }
4967
4968     if(priv->recording)
4969     {
4970         record_instruction(&priv->record, ins);
4971         return;
4972     }
4973
4974     /* boolean if */
4975     if(ins->handler_idx == WINED3DSIH_IF)
4976     {
4977         control_frame = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*control_frame));
4978         list_add_head(&priv->control_frames, &control_frame->entry);
4979         control_frame->type = IF;
4980
4981         bool_const = get_bool_const(ins, This, ins->src[0].reg.idx);
4982         if(ins->src[0].modifiers == WINED3DSPSM_NOT) bool_const = !bool_const;
4983         if(!priv->muted && bool_const == FALSE)
4984         {
4985             shader_addline(buffer, "#if(FALSE){\n");
4986             priv->muted = TRUE;
4987             control_frame->muting = TRUE;
4988         }
4989         else shader_addline(buffer, "#if(TRUE) {\n");
4990
4991         return; /* Instruction is handled */
4992     }
4993     else if(ins->handler_idx == WINED3DSIH_IFC)
4994     {
4995         /* IF(bool) and if_cond(a, b) use the same ELSE and ENDIF tokens */
4996         control_frame = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*control_frame));
4997         control_frame->type = IFC;
4998         control_frame->ifc_no = priv->num_ifcs++;
4999         list_add_head(&priv->control_frames, &control_frame->entry);
5000     }
5001     else if(ins->handler_idx == WINED3DSIH_ELSE)
5002     {
5003         struct list *e = list_head(&priv->control_frames);
5004         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5005
5006         if(control_frame->type == IF)
5007         {
5008             shader_addline(buffer, "#} else {\n");
5009             if(!priv->muted && !control_frame->muting)
5010             {
5011                 priv->muted = TRUE;
5012                 control_frame->muting = TRUE;
5013             }
5014             else if(control_frame->muting) priv->muted = FALSE;
5015             return; /* Instruction is handled. */
5016         }
5017         /* In case of an ifc, generate a HW shader instruction */
5018     }
5019     else if(ins->handler_idx == WINED3DSIH_ENDIF)
5020     {
5021         struct list *e = list_head(&priv->control_frames);
5022         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5023
5024         if(control_frame->type == IF)
5025         {
5026             shader_addline(buffer, "#} endif\n");
5027             if(control_frame->muting) priv->muted = FALSE;
5028             list_remove(&control_frame->entry);
5029             HeapFree(GetProcessHeap(), 0, control_frame);
5030             return; /* Instruction is handled */
5031         }
5032     }
5033
5034     if(priv->muted) return;
5035
5036     /* Select handler */
5037     hw_fct = shader_arb_instruction_handler_table[ins->handler_idx];
5038
5039     /* Unhandled opcode */
5040     if (!hw_fct)
5041     {
5042         FIXME("Backend can't handle opcode %#x\n", ins->handler_idx);
5043         return;
5044     }
5045     hw_fct(ins);
5046
5047     if(ins->handler_idx == WINED3DSIH_ENDLOOP || ins->handler_idx == WINED3DSIH_ENDREP)
5048     {
5049         struct list *e = list_head(&priv->control_frames);
5050         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5051         list_remove(&control_frame->entry);
5052         HeapFree(GetProcessHeap(), 0, control_frame);
5053         priv->loop_depth--;
5054     }
5055     else if(ins->handler_idx == WINED3DSIH_ENDIF)
5056     {
5057         /* Non-ifc ENDIFs don't reach that place because of the return in the if block above */
5058         struct list *e = list_head(&priv->control_frames);
5059         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5060         list_remove(&control_frame->entry);
5061         HeapFree(GetProcessHeap(), 0, control_frame);
5062     }
5063
5064
5065     shader_arb_add_instruction_modifiers(ins);
5066 }
5067
5068 const shader_backend_t arb_program_shader_backend = {
5069     shader_arb_handle_instruction,
5070     shader_arb_select,
5071     shader_arb_select_depth_blt,
5072     shader_arb_deselect_depth_blt,
5073     shader_arb_update_float_vertex_constants,
5074     shader_arb_update_float_pixel_constants,
5075     shader_arb_load_constants,
5076     shader_arb_load_np2fixup_constants,
5077     shader_arb_destroy,
5078     shader_arb_alloc,
5079     shader_arb_free,
5080     shader_arb_dirty_const,
5081     shader_arb_get_caps,
5082     shader_arb_color_fixup_supported,
5083 };
5084
5085 /* ARB_fragment_program fixed function pipeline replacement definitions */
5086 #define ARB_FFP_CONST_TFACTOR           0
5087 #define ARB_FFP_CONST_SPECULAR_ENABLE   ((ARB_FFP_CONST_TFACTOR) + 1)
5088 #define ARB_FFP_CONST_CONSTANT(i)       ((ARB_FFP_CONST_SPECULAR_ENABLE) + 1 + i)
5089 #define ARB_FFP_CONST_BUMPMAT(i)        ((ARB_FFP_CONST_CONSTANT(7)) + 1 + i)
5090 #define ARB_FFP_CONST_LUMINANCE(i)      ((ARB_FFP_CONST_BUMPMAT(7)) + 1 + i)
5091
5092 struct arbfp_ffp_desc
5093 {
5094     struct ffp_frag_desc parent;
5095     GLuint shader;
5096     unsigned int num_textures_used;
5097 };
5098
5099 /* Context activation is done by the caller. */
5100 static void arbfp_enable(IWineD3DDevice *iface, BOOL enable) {
5101     ENTER_GL();
5102     if(enable) {
5103         glEnable(GL_FRAGMENT_PROGRAM_ARB);
5104         checkGLcall("glEnable(GL_FRAGMENT_PROGRAM_ARB)");
5105     } else {
5106         glDisable(GL_FRAGMENT_PROGRAM_ARB);
5107         checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
5108     }
5109     LEAVE_GL();
5110 }
5111
5112 static HRESULT arbfp_alloc(IWineD3DDevice *iface) {
5113     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *) iface;
5114     struct shader_arb_priv *priv;
5115     /* Share private data between the shader backend and the pipeline replacement, if both
5116      * are the arb implementation. This is needed to figure out whether ARBfp should be disabled
5117      * if no pixel shader is bound or not
5118      */
5119     if(This->shader_backend == &arb_program_shader_backend) {
5120         This->fragment_priv = This->shader_priv;
5121     } else {
5122         This->fragment_priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(struct shader_arb_priv));
5123         if(!This->fragment_priv) return E_OUTOFMEMORY;
5124     }
5125     priv = This->fragment_priv;
5126     if (wine_rb_init(&priv->fragment_shaders, &wined3d_ffp_frag_program_rb_functions) == -1)
5127     {
5128         ERR("Failed to initialize rbtree.\n");
5129         HeapFree(GetProcessHeap(), 0, This->fragment_priv);
5130         return E_OUTOFMEMORY;
5131     }
5132     priv->use_arbfp_fixed_func = TRUE;
5133     return WINED3D_OK;
5134 }
5135
5136 /* Context activation is done by the caller. */
5137 static void arbfp_free_ffpshader(struct wine_rb_entry *entry, void *context)
5138 {
5139     const struct wined3d_gl_info *gl_info = context;
5140     struct arbfp_ffp_desc *entry_arb = WINE_RB_ENTRY_VALUE(entry, struct arbfp_ffp_desc, parent.entry);
5141
5142     ENTER_GL();
5143     GL_EXTCALL(glDeleteProgramsARB(1, &entry_arb->shader));
5144     checkGLcall("glDeleteProgramsARB(1, &entry_arb->shader)");
5145     HeapFree(GetProcessHeap(), 0, entry_arb);
5146     LEAVE_GL();
5147 }
5148
5149 /* Context activation is done by the caller. */
5150 static void arbfp_free(IWineD3DDevice *iface) {
5151     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *) iface;
5152     struct shader_arb_priv *priv = This->fragment_priv;
5153
5154     wine_rb_destroy(&priv->fragment_shaders, arbfp_free_ffpshader, &This->adapter->gl_info);
5155     priv->use_arbfp_fixed_func = FALSE;
5156
5157     if(This->shader_backend != &arb_program_shader_backend) {
5158         HeapFree(GetProcessHeap(), 0, This->fragment_priv);
5159     }
5160 }
5161
5162 static void arbfp_get_caps(WINED3DDEVTYPE devtype, const struct wined3d_gl_info *gl_info, struct fragment_caps *caps)
5163 {
5164     caps->TextureOpCaps =  WINED3DTEXOPCAPS_DISABLE                     |
5165                            WINED3DTEXOPCAPS_SELECTARG1                  |
5166                            WINED3DTEXOPCAPS_SELECTARG2                  |
5167                            WINED3DTEXOPCAPS_MODULATE4X                  |
5168                            WINED3DTEXOPCAPS_MODULATE2X                  |
5169                            WINED3DTEXOPCAPS_MODULATE                    |
5170                            WINED3DTEXOPCAPS_ADDSIGNED2X                 |
5171                            WINED3DTEXOPCAPS_ADDSIGNED                   |
5172                            WINED3DTEXOPCAPS_ADD                         |
5173                            WINED3DTEXOPCAPS_SUBTRACT                    |
5174                            WINED3DTEXOPCAPS_ADDSMOOTH                   |
5175                            WINED3DTEXOPCAPS_BLENDCURRENTALPHA           |
5176                            WINED3DTEXOPCAPS_BLENDFACTORALPHA            |
5177                            WINED3DTEXOPCAPS_BLENDTEXTUREALPHA           |
5178                            WINED3DTEXOPCAPS_BLENDDIFFUSEALPHA           |
5179                            WINED3DTEXOPCAPS_BLENDTEXTUREALPHAPM         |
5180                            WINED3DTEXOPCAPS_MODULATEALPHA_ADDCOLOR      |
5181                            WINED3DTEXOPCAPS_MODULATECOLOR_ADDALPHA      |
5182                            WINED3DTEXOPCAPS_MODULATEINVCOLOR_ADDALPHA   |
5183                            WINED3DTEXOPCAPS_MODULATEINVALPHA_ADDCOLOR   |
5184                            WINED3DTEXOPCAPS_DOTPRODUCT3                 |
5185                            WINED3DTEXOPCAPS_MULTIPLYADD                 |
5186                            WINED3DTEXOPCAPS_LERP                        |
5187                            WINED3DTEXOPCAPS_BUMPENVMAP                  |
5188                            WINED3DTEXOPCAPS_BUMPENVMAPLUMINANCE;
5189
5190     /* TODO: Implement WINED3DTEXOPCAPS_PREMODULATE */
5191
5192     caps->MaxTextureBlendStages   = 8;
5193     caps->MaxSimultaneousTextures = min(GL_LIMITS(fragment_samplers), 8);
5194
5195     caps->PrimitiveMiscCaps |= WINED3DPMISCCAPS_TSSARGTEMP;
5196 }
5197 #undef GLINFO_LOCATION
5198
5199 #define GLINFO_LOCATION stateblock->wineD3DDevice->adapter->gl_info
5200 static void state_texfactor_arbfp(DWORD state, IWineD3DStateBlockImpl *stateblock, struct wined3d_context *context)
5201 {
5202     float col[4];
5203     IWineD3DDeviceImpl *device = stateblock->wineD3DDevice;
5204
5205     /* Don't load the parameter if we're using an arbfp pixel shader, otherwise we'll overwrite
5206      * application provided constants
5207      */
5208     if(device->shader_backend == &arb_program_shader_backend) {
5209         if (use_ps(stateblock)) return;
5210
5211         device = stateblock->wineD3DDevice;
5212         context->pshader_const_dirty[ARB_FFP_CONST_TFACTOR] = 1;
5213         device->highest_dirty_ps_const = max(device->highest_dirty_ps_const, ARB_FFP_CONST_TFACTOR + 1);
5214     }
5215
5216     D3DCOLORTOGLFLOAT4(stateblock->renderState[WINED3DRS_TEXTUREFACTOR], col);
5217     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_TFACTOR, col));
5218     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_TFACTOR, col)");
5219
5220 }
5221
5222 static void state_arb_specularenable(DWORD state, IWineD3DStateBlockImpl *stateblock, struct wined3d_context *context)
5223 {
5224     float col[4];
5225     IWineD3DDeviceImpl *device = stateblock->wineD3DDevice;
5226
5227     /* Don't load the parameter if we're using an arbfp pixel shader, otherwise we'll overwrite
5228      * application provided constants
5229      */
5230     if(device->shader_backend == &arb_program_shader_backend) {
5231         if (use_ps(stateblock)) return;
5232
5233         device = stateblock->wineD3DDevice;
5234         context->pshader_const_dirty[ARB_FFP_CONST_SPECULAR_ENABLE] = 1;
5235         device->highest_dirty_ps_const = max(device->highest_dirty_ps_const, ARB_FFP_CONST_SPECULAR_ENABLE + 1);
5236     }
5237
5238     if(stateblock->renderState[WINED3DRS_SPECULARENABLE]) {
5239         /* The specular color has no alpha */
5240         col[0] = 1.0f; col[1] = 1.0f;
5241         col[2] = 1.0f; col[3] = 0.0f;
5242     } else {
5243         col[0] = 0.0f; col[1] = 0.0f;
5244         col[2] = 0.0f; col[3] = 0.0f;
5245     }
5246     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_SPECULAR_ENABLE, col));
5247     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_SPECULAR_ENABLE, col)");
5248 }
5249
5250 static void set_bumpmat_arbfp(DWORD state, IWineD3DStateBlockImpl *stateblock, struct wined3d_context *context)
5251 {
5252     DWORD stage = (state - STATE_TEXTURESTAGE(0, 0)) / (WINED3D_HIGHEST_TEXTURE_STATE + 1);
5253     IWineD3DDeviceImpl *device = stateblock->wineD3DDevice;
5254     float mat[2][2];
5255
5256     if (use_ps(stateblock))
5257     {
5258         if (stage != 0
5259                 && (((IWineD3DPixelShaderImpl *)stateblock->pixelShader)->baseShader.reg_maps.bumpmat & (1 << stage)))
5260         {
5261             /* The pixel shader has to know the bump env matrix. Do a constants update if it isn't scheduled
5262              * anyway
5263              */
5264             if(!isStateDirty(context, STATE_PIXELSHADERCONSTANT)) {
5265                 device->StateTable[STATE_PIXELSHADERCONSTANT].apply(STATE_PIXELSHADERCONSTANT, stateblock, context);
5266             }
5267         }
5268
5269         if(device->shader_backend == &arb_program_shader_backend) {
5270             /* Exit now, don't set the bumpmat below, otherwise we may overwrite pixel shader constants */
5271             return;
5272         }
5273     } else if(device->shader_backend == &arb_program_shader_backend) {
5274         context->pshader_const_dirty[ARB_FFP_CONST_BUMPMAT(stage)] = 1;
5275         device->highest_dirty_ps_const = max(device->highest_dirty_ps_const, ARB_FFP_CONST_BUMPMAT(stage) + 1);
5276     }
5277
5278     mat[0][0] = *((float *) &stateblock->textureState[stage][WINED3DTSS_BUMPENVMAT00]);
5279     mat[0][1] = *((float *) &stateblock->textureState[stage][WINED3DTSS_BUMPENVMAT01]);
5280     mat[1][0] = *((float *) &stateblock->textureState[stage][WINED3DTSS_BUMPENVMAT10]);
5281     mat[1][1] = *((float *) &stateblock->textureState[stage][WINED3DTSS_BUMPENVMAT11]);
5282
5283     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_BUMPMAT(stage), &mat[0][0]));
5284     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_BUMPMAT(stage), &mat[0][0])");
5285 }
5286
5287 static void tex_bumpenvlum_arbfp(DWORD state, IWineD3DStateBlockImpl *stateblock, struct wined3d_context *context)
5288 {
5289     DWORD stage = (state - STATE_TEXTURESTAGE(0, 0)) / (WINED3D_HIGHEST_TEXTURE_STATE + 1);
5290     IWineD3DDeviceImpl *device = stateblock->wineD3DDevice;
5291     float param[4];
5292
5293     if (use_ps(stateblock))
5294     {
5295         if (stage != 0
5296                 && (((IWineD3DPixelShaderImpl *)stateblock->pixelShader)->baseShader.reg_maps.luminanceparams & (1 << stage)))
5297         {
5298             /* The pixel shader has to know the luminance offset. Do a constants update if it
5299              * isn't scheduled anyway
5300              */
5301             if(!isStateDirty(context, STATE_PIXELSHADERCONSTANT)) {
5302                 device->StateTable[STATE_PIXELSHADERCONSTANT].apply(STATE_PIXELSHADERCONSTANT, stateblock, context);
5303             }
5304         }
5305
5306         if(device->shader_backend == &arb_program_shader_backend) {
5307             /* Exit now, don't set the bumpmat below, otherwise we may overwrite pixel shader constants */
5308             return;
5309         }
5310     } else if(device->shader_backend == &arb_program_shader_backend) {
5311         context->pshader_const_dirty[ARB_FFP_CONST_LUMINANCE(stage)] = 1;
5312         device->highest_dirty_ps_const = max(device->highest_dirty_ps_const, ARB_FFP_CONST_LUMINANCE(stage) + 1);
5313     }
5314
5315     param[0] = *((float *) &stateblock->textureState[stage][WINED3DTSS_BUMPENVLSCALE]);
5316     param[1] = *((float *) &stateblock->textureState[stage][WINED3DTSS_BUMPENVLOFFSET]);
5317     param[2] = 0.0f;
5318     param[3] = 0.0f;
5319
5320     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_LUMINANCE(stage), param));
5321     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_LUMINANCE(stage), param)");
5322 }
5323
5324 static const char *get_argreg(struct wined3d_shader_buffer *buffer, DWORD argnum, unsigned int stage, DWORD arg)
5325 {
5326     const char *ret;
5327
5328     if(arg == ARG_UNUSED) return "unused"; /* This is the marker for unused registers */
5329
5330     switch(arg & WINED3DTA_SELECTMASK) {
5331         case WINED3DTA_DIFFUSE:
5332             ret = "fragment.color.primary"; break;
5333
5334         case WINED3DTA_CURRENT:
5335             if(stage == 0) ret = "fragment.color.primary";
5336             else ret = "ret";
5337             break;
5338
5339         case WINED3DTA_TEXTURE:
5340             switch(stage) {
5341                 case 0: ret = "tex0"; break;
5342                 case 1: ret = "tex1"; break;
5343                 case 2: ret = "tex2"; break;
5344                 case 3: ret = "tex3"; break;
5345                 case 4: ret = "tex4"; break;
5346                 case 5: ret = "tex5"; break;
5347                 case 6: ret = "tex6"; break;
5348                 case 7: ret = "tex7"; break;
5349                 default: ret = "unknown texture";
5350             }
5351             break;
5352
5353         case WINED3DTA_TFACTOR:
5354             ret = "tfactor"; break;
5355
5356         case WINED3DTA_SPECULAR:
5357             ret = "fragment.color.secondary"; break;
5358
5359         case WINED3DTA_TEMP:
5360             ret = "tempreg"; break;
5361
5362         case WINED3DTA_CONSTANT:
5363             FIXME("Implement perstage constants\n");
5364             switch(stage) {
5365                 case 0: ret = "const0"; break;
5366                 case 1: ret = "const1"; break;
5367                 case 2: ret = "const2"; break;
5368                 case 3: ret = "const3"; break;
5369                 case 4: ret = "const4"; break;
5370                 case 5: ret = "const5"; break;
5371                 case 6: ret = "const6"; break;
5372                 case 7: ret = "const7"; break;
5373                 default: ret = "unknown constant";
5374             }
5375             break;
5376
5377         default:
5378             return "unknown";
5379     }
5380
5381     if(arg & WINED3DTA_COMPLEMENT) {
5382         shader_addline(buffer, "SUB arg%u, const.x, %s;\n", argnum, ret);
5383         if(argnum == 0) ret = "arg0";
5384         if(argnum == 1) ret = "arg1";
5385         if(argnum == 2) ret = "arg2";
5386     }
5387     if(arg & WINED3DTA_ALPHAREPLICATE) {
5388         shader_addline(buffer, "MOV arg%u, %s.w;\n", argnum, ret);
5389         if(argnum == 0) ret = "arg0";
5390         if(argnum == 1) ret = "arg1";
5391         if(argnum == 2) ret = "arg2";
5392     }
5393     return ret;
5394 }
5395
5396 static void gen_ffp_instr(struct wined3d_shader_buffer *buffer, unsigned int stage, BOOL color,
5397         BOOL alpha, DWORD dst, DWORD op, DWORD dw_arg0, DWORD dw_arg1, DWORD dw_arg2)
5398 {
5399     const char *dstmask, *dstreg, *arg0, *arg1, *arg2;
5400     unsigned int mul = 1;
5401     BOOL mul_final_dest = FALSE;
5402
5403     if(color && alpha) dstmask = "";
5404     else if(color) dstmask = ".xyz";
5405     else dstmask = ".w";
5406
5407     if(dst == tempreg) dstreg = "tempreg";
5408     else dstreg = "ret";
5409
5410     arg0 = get_argreg(buffer, 0, stage, dw_arg0);
5411     arg1 = get_argreg(buffer, 1, stage, dw_arg1);
5412     arg2 = get_argreg(buffer, 2, stage, dw_arg2);
5413
5414     switch(op) {
5415         case WINED3DTOP_DISABLE:
5416             if(stage == 0) shader_addline(buffer, "MOV %s%s, fragment.color.primary;\n", dstreg, dstmask);
5417             break;
5418
5419         case WINED3DTOP_SELECTARG2:
5420             arg1 = arg2;
5421         case WINED3DTOP_SELECTARG1:
5422             shader_addline(buffer, "MOV %s%s, %s;\n", dstreg, dstmask, arg1);
5423             break;
5424
5425         case WINED3DTOP_MODULATE4X:
5426             mul = 2;
5427         case WINED3DTOP_MODULATE2X:
5428             mul *= 2;
5429             if(strcmp(dstreg, "result.color") == 0) {
5430                 dstreg = "ret";
5431                 mul_final_dest = TRUE;
5432             }
5433         case WINED3DTOP_MODULATE:
5434             shader_addline(buffer, "MUL %s%s, %s, %s;\n", dstreg, dstmask, arg1, arg2);
5435             break;
5436
5437         case WINED3DTOP_ADDSIGNED2X:
5438             mul = 2;
5439             if(strcmp(dstreg, "result.color") == 0) {
5440                 dstreg = "ret";
5441                 mul_final_dest = TRUE;
5442             }
5443         case WINED3DTOP_ADDSIGNED:
5444             shader_addline(buffer, "SUB arg2, %s, const.w;\n", arg2);
5445             arg2 = "arg2";
5446         case WINED3DTOP_ADD:
5447             shader_addline(buffer, "ADD_SAT %s%s, %s, %s;\n", dstreg, dstmask, arg1, arg2);
5448             break;
5449
5450         case WINED3DTOP_SUBTRACT:
5451             shader_addline(buffer, "SUB_SAT %s%s, %s, %s;\n", dstreg, dstmask, arg1, arg2);
5452             break;
5453
5454         case WINED3DTOP_ADDSMOOTH:
5455             shader_addline(buffer, "SUB arg1, const.x, %s;\n", arg1);
5456             shader_addline(buffer, "MAD_SAT %s%s, arg1, %s, %s;\n", dstreg, dstmask, arg2, arg1);
5457             break;
5458
5459         case WINED3DTOP_BLENDCURRENTALPHA:
5460             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_CURRENT);
5461             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
5462             break;
5463         case WINED3DTOP_BLENDFACTORALPHA:
5464             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_TFACTOR);
5465             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
5466             break;
5467         case WINED3DTOP_BLENDTEXTUREALPHA:
5468             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_TEXTURE);
5469             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
5470             break;
5471         case WINED3DTOP_BLENDDIFFUSEALPHA:
5472             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_DIFFUSE);
5473             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
5474             break;
5475
5476         case WINED3DTOP_BLENDTEXTUREALPHAPM:
5477             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_TEXTURE);
5478             shader_addline(buffer, "SUB arg0.w, const.x, %s.w;\n", arg0);
5479             shader_addline(buffer, "MAD_SAT %s%s, %s, arg0.w, %s;\n", dstreg, dstmask, arg2, arg1);
5480             break;
5481
5482         /* D3DTOP_PREMODULATE ???? */
5483
5484         case WINED3DTOP_MODULATEINVALPHA_ADDCOLOR:
5485             shader_addline(buffer, "SUB arg0.w, const.x, %s;\n", arg1);
5486             shader_addline(buffer, "MAD_SAT %s%s, arg0.w, %s, %s;\n", dstreg, dstmask, arg2, arg1);
5487             break;
5488         case WINED3DTOP_MODULATEALPHA_ADDCOLOR:
5489             shader_addline(buffer, "MAD_SAT %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg1, arg2, arg1);
5490             break;
5491         case WINED3DTOP_MODULATEINVCOLOR_ADDALPHA:
5492             shader_addline(buffer, "SUB arg0, const.x, %s;\n", arg1);
5493             shader_addline(buffer, "MAD_SAT %s%s, arg0, %s, %s.w;\n", dstreg, dstmask, arg2, arg1);
5494             break;
5495         case WINED3DTOP_MODULATECOLOR_ADDALPHA:
5496             shader_addline(buffer, "MAD_SAT %s%s, %s, %s, %s.w;\n", dstreg, dstmask, arg1, arg2, arg1);
5497             break;
5498
5499         case WINED3DTOP_DOTPRODUCT3:
5500             mul = 4;
5501             if(strcmp(dstreg, "result.color") == 0) {
5502                 dstreg = "ret";
5503                 mul_final_dest = TRUE;
5504             }
5505             shader_addline(buffer, "SUB arg1, %s, const.w;\n", arg1);
5506             shader_addline(buffer, "SUB arg2, %s, const.w;\n", arg2);
5507             shader_addline(buffer, "DP3_SAT %s%s, arg1, arg2;\n", dstreg, dstmask);
5508             break;
5509
5510         case WINED3DTOP_MULTIPLYADD:
5511             shader_addline(buffer, "MAD_SAT %s%s, %s, %s, %s;\n", dstreg, dstmask, arg1, arg2, arg0);
5512             break;
5513
5514         case WINED3DTOP_LERP:
5515             /* The msdn is not quite right here */
5516             shader_addline(buffer, "LRP %s%s, %s, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
5517             break;
5518
5519         case WINED3DTOP_BUMPENVMAP:
5520         case WINED3DTOP_BUMPENVMAPLUMINANCE:
5521             /* Those are handled in the first pass of the shader(generation pass 1 and 2) already */
5522             break;
5523
5524         default:
5525             FIXME("Unhandled texture op %08x\n", op);
5526     }
5527
5528     if(mul == 2) {
5529         shader_addline(buffer, "MUL_SAT %s%s, %s, const.y;\n", mul_final_dest ? "result.color" : dstreg, dstmask, dstreg);
5530     } else if(mul == 4) {
5531         shader_addline(buffer, "MUL_SAT %s%s, %s, const.z;\n", mul_final_dest ? "result.color" : dstreg, dstmask, dstreg);
5532     }
5533 }
5534
5535 /* The stateblock is passed for GLINFO_LOCATION */
5536 static GLuint gen_arbfp_ffp_shader(const struct ffp_frag_settings *settings, IWineD3DStateBlockImpl *stateblock)
5537 {
5538     unsigned int stage;
5539     struct wined3d_shader_buffer buffer;
5540     BOOL tex_read[MAX_TEXTURES] = {FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE};
5541     BOOL bump_used[MAX_TEXTURES] = {FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE};
5542     BOOL luminance_used[MAX_TEXTURES] = {FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE};
5543     const char *textype;
5544     const char *instr, *sat;
5545     char colorcor_dst[8];
5546     GLuint ret;
5547     DWORD arg0, arg1, arg2;
5548     BOOL tempreg_used = FALSE, tfactor_used = FALSE;
5549     BOOL op_equal;
5550     const char *final_combiner_src = "ret";
5551     GLint pos;
5552
5553     /* Find out which textures are read */
5554     for(stage = 0; stage < MAX_TEXTURES; stage++) {
5555         if(settings->op[stage].cop == WINED3DTOP_DISABLE) break;
5556         arg0 = settings->op[stage].carg0 & WINED3DTA_SELECTMASK;
5557         arg1 = settings->op[stage].carg1 & WINED3DTA_SELECTMASK;
5558         arg2 = settings->op[stage].carg2 & WINED3DTA_SELECTMASK;
5559         if(arg0 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
5560         if(arg1 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
5561         if(arg2 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
5562
5563         if(settings->op[stage].cop == WINED3DTOP_BLENDTEXTUREALPHA) tex_read[stage] = TRUE;
5564         if(settings->op[stage].cop == WINED3DTOP_BLENDTEXTUREALPHAPM) tex_read[stage] = TRUE;
5565         if(settings->op[stage].cop == WINED3DTOP_BUMPENVMAP) {
5566             bump_used[stage] = TRUE;
5567             tex_read[stage] = TRUE;
5568         }
5569         if(settings->op[stage].cop == WINED3DTOP_BUMPENVMAPLUMINANCE) {
5570             bump_used[stage] = TRUE;
5571             tex_read[stage] = TRUE;
5572             luminance_used[stage] = TRUE;
5573         } else if(settings->op[stage].cop == WINED3DTOP_BLENDFACTORALPHA) {
5574             tfactor_used = TRUE;
5575         }
5576
5577         if(arg0 == WINED3DTA_TFACTOR || arg1 == WINED3DTA_TFACTOR || arg2 == WINED3DTA_TFACTOR) {
5578             tfactor_used = TRUE;
5579         }
5580
5581         if(settings->op[stage].dst == tempreg) tempreg_used = TRUE;
5582         if(arg0 == WINED3DTA_TEMP || arg1 == WINED3DTA_TEMP || arg2 == WINED3DTA_TEMP) {
5583             tempreg_used = TRUE;
5584         }
5585
5586         if(settings->op[stage].aop == WINED3DTOP_DISABLE) continue;
5587         arg0 = settings->op[stage].aarg0 & WINED3DTA_SELECTMASK;
5588         arg1 = settings->op[stage].aarg1 & WINED3DTA_SELECTMASK;
5589         arg2 = settings->op[stage].aarg2 & WINED3DTA_SELECTMASK;
5590         if(arg0 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
5591         if(arg1 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
5592         if(arg2 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
5593
5594         if(arg0 == WINED3DTA_TEMP || arg1 == WINED3DTA_TEMP || arg2 == WINED3DTA_TEMP) {
5595             tempreg_used = TRUE;
5596         }
5597         if(arg0 == WINED3DTA_TFACTOR || arg1 == WINED3DTA_TFACTOR || arg2 == WINED3DTA_TFACTOR) {
5598             tfactor_used = TRUE;
5599         }
5600     }
5601
5602     /* Shader header */
5603     if (!shader_buffer_init(&buffer))
5604     {
5605         ERR("Failed to initialize shader buffer.\n");
5606         return 0;
5607     }
5608
5609     shader_addline(&buffer, "!!ARBfp1.0\n");
5610
5611     switch(settings->fog) {
5612         case FOG_OFF:                                                         break;
5613         case FOG_LINEAR: shader_addline(&buffer, "OPTION ARB_fog_linear;\n"); break;
5614         case FOG_EXP:    shader_addline(&buffer, "OPTION ARB_fog_exp;\n");    break;
5615         case FOG_EXP2:   shader_addline(&buffer, "OPTION ARB_fog_exp2;\n");   break;
5616         default: FIXME("Unexpected fog setting %d\n", settings->fog);
5617     }
5618
5619     shader_addline(&buffer, "PARAM const = {1, 2, 4, 0.5};\n");
5620     shader_addline(&buffer, "TEMP TMP;\n");
5621     shader_addline(&buffer, "TEMP ret;\n");
5622     if(tempreg_used || settings->sRGB_write) shader_addline(&buffer, "TEMP tempreg;\n");
5623     shader_addline(&buffer, "TEMP arg0;\n");
5624     shader_addline(&buffer, "TEMP arg1;\n");
5625     shader_addline(&buffer, "TEMP arg2;\n");
5626     for(stage = 0; stage < MAX_TEXTURES; stage++) {
5627         if(!tex_read[stage]) continue;
5628         shader_addline(&buffer, "TEMP tex%u;\n", stage);
5629         if(!bump_used[stage]) continue;
5630         shader_addline(&buffer, "PARAM bumpmat%u = program.env[%u];\n", stage, ARB_FFP_CONST_BUMPMAT(stage));
5631         if(!luminance_used[stage]) continue;
5632         shader_addline(&buffer, "PARAM luminance%u = program.env[%u];\n", stage, ARB_FFP_CONST_LUMINANCE(stage));
5633     }
5634     if(tfactor_used) {
5635         shader_addline(&buffer, "PARAM tfactor = program.env[%u];\n", ARB_FFP_CONST_TFACTOR);
5636     }
5637         shader_addline(&buffer, "PARAM specular_enable = program.env[%u];\n", ARB_FFP_CONST_SPECULAR_ENABLE);
5638
5639     if(settings->sRGB_write) {
5640         shader_addline(&buffer, "PARAM srgb_consts1 = {%f, %f, %f, %f};\n",
5641                        srgb_mul_low, srgb_cmp, srgb_pow, srgb_mul_high);
5642         shader_addline(&buffer, "PARAM srgb_consts2 = {%f, %f, %f, %f};\n",
5643                        srgb_sub_high, 0.0, 0.0, 0.0);
5644     }
5645
5646     if(ffp_clip_emul(stateblock) && settings->emul_clipplanes) shader_addline(&buffer, "KIL fragment.texcoord[7];\n");
5647
5648     /* Generate texture sampling instructions) */
5649     for(stage = 0; stage < MAX_TEXTURES && settings->op[stage].cop != WINED3DTOP_DISABLE; stage++) {
5650         if(!tex_read[stage]) continue;
5651
5652         switch(settings->op[stage].tex_type) {
5653             case tex_1d:                    textype = "1D";     break;
5654             case tex_2d:                    textype = "2D";     break;
5655             case tex_3d:                    textype = "3D";     break;
5656             case tex_cube:                  textype = "CUBE";   break;
5657             case tex_rect:                  textype = "RECT";   break;
5658             default: textype = "unexpected_textype";   break;
5659         }
5660
5661         if(settings->op[stage].cop == WINED3DTOP_BUMPENVMAP ||
5662            settings->op[stage].cop == WINED3DTOP_BUMPENVMAPLUMINANCE) {
5663             sat = "";
5664         } else {
5665             sat = "_SAT";
5666         }
5667
5668         if(settings->op[stage].projected == proj_none) {
5669             instr = "TEX";
5670         } else if(settings->op[stage].projected == proj_count4 ||
5671                   settings->op[stage].projected == proj_count3) {
5672             instr = "TXP";
5673         } else {
5674             FIXME("Unexpected projection mode %d\n", settings->op[stage].projected);
5675             instr = "TXP";
5676         }
5677
5678         if(stage > 0 &&
5679            (settings->op[stage - 1].cop == WINED3DTOP_BUMPENVMAP ||
5680             settings->op[stage - 1].cop == WINED3DTOP_BUMPENVMAPLUMINANCE)) {
5681             shader_addline(&buffer, "SWZ arg1, bumpmat%u, x, z, 0, 0;\n", stage - 1);
5682             shader_addline(&buffer, "DP3 ret.x, arg1, tex%u;\n", stage - 1);
5683             shader_addline(&buffer, "SWZ arg1, bumpmat%u, y, w, 0, 0;\n", stage - 1);
5684             shader_addline(&buffer, "DP3 ret.y, arg1, tex%u;\n", stage - 1);
5685
5686             /* with projective textures, texbem only divides the static texture coord, not the displacement,
5687              * so multiply the displacement with the dividing parameter before passing it to TXP
5688              */
5689             if (settings->op[stage].projected != proj_none) {
5690                 if(settings->op[stage].projected == proj_count4) {
5691                     shader_addline(&buffer, "MOV ret.w, fragment.texcoord[%u].w;\n", stage);
5692                     shader_addline(&buffer, "MUL ret.xyz, ret, fragment.texcoord[%u].w, fragment.texcoord[%u];\n", stage, stage);
5693                 } else {
5694                     shader_addline(&buffer, "MOV ret.w, fragment.texcoord[%u].z;\n", stage);
5695                     shader_addline(&buffer, "MAD ret.xyz, ret, fragment.texcoord[%u].z, fragment.texcoord[%u];\n", stage, stage);
5696                 }
5697             } else {
5698                 shader_addline(&buffer, "ADD ret, ret, fragment.texcoord[%u];\n", stage);
5699             }
5700
5701             shader_addline(&buffer, "%s%s tex%u, ret, texture[%u], %s;\n",
5702                            instr, sat, stage, stage, textype);
5703             if(settings->op[stage - 1].cop == WINED3DTOP_BUMPENVMAPLUMINANCE) {
5704                 shader_addline(&buffer, "MAD_SAT ret.x, tex%u.z, luminance%u.x, luminance%u.y;\n",
5705                                stage - 1, stage - 1, stage - 1);
5706                 shader_addline(&buffer, "MUL tex%u, tex%u, ret.x;\n", stage, stage);
5707             }
5708         } else if(settings->op[stage].projected == proj_count3) {
5709             shader_addline(&buffer, "MOV ret, fragment.texcoord[%u];\n", stage);
5710             shader_addline(&buffer, "MOV ret.w, ret.z;\n");
5711             shader_addline(&buffer, "%s%s tex%u, ret, texture[%u], %s;\n",
5712                             instr, sat, stage, stage, textype);
5713         } else {
5714             shader_addline(&buffer, "%s%s tex%u, fragment.texcoord[%u], texture[%u], %s;\n",
5715                             instr, sat, stage, stage, stage, textype);
5716         }
5717
5718         sprintf(colorcor_dst, "tex%u", stage);
5719         gen_color_correction(&buffer, colorcor_dst, WINED3DSP_WRITEMASK_ALL, "const.x", "const.y",
5720                 settings->op[stage].color_fixup);
5721     }
5722
5723     /* Generate the main shader */
5724     for(stage = 0; stage < MAX_TEXTURES; stage++) {
5725         if(settings->op[stage].cop == WINED3DTOP_DISABLE) {
5726             if(stage == 0) {
5727                 final_combiner_src = "fragment.color.primary";
5728             }
5729             break;
5730         }
5731
5732         if(settings->op[stage].cop == WINED3DTOP_SELECTARG1 &&
5733            settings->op[stage].aop == WINED3DTOP_SELECTARG1) {
5734             op_equal = settings->op[stage].carg1 == settings->op[stage].aarg1;
5735         } else if(settings->op[stage].cop == WINED3DTOP_SELECTARG1 &&
5736                   settings->op[stage].aop == WINED3DTOP_SELECTARG2) {
5737             op_equal = settings->op[stage].carg1 == settings->op[stage].aarg2;
5738         } else if(settings->op[stage].cop == WINED3DTOP_SELECTARG2 &&
5739                   settings->op[stage].aop == WINED3DTOP_SELECTARG1) {
5740             op_equal = settings->op[stage].carg2 == settings->op[stage].aarg1;
5741         } else if(settings->op[stage].cop == WINED3DTOP_SELECTARG2 &&
5742                   settings->op[stage].aop == WINED3DTOP_SELECTARG2) {
5743             op_equal = settings->op[stage].carg2 == settings->op[stage].aarg2;
5744         } else {
5745             op_equal = settings->op[stage].aop   == settings->op[stage].cop &&
5746                        settings->op[stage].carg0 == settings->op[stage].aarg0 &&
5747                        settings->op[stage].carg1 == settings->op[stage].aarg1 &&
5748                        settings->op[stage].carg2 == settings->op[stage].aarg2;
5749         }
5750
5751         if(settings->op[stage].aop == WINED3DTOP_DISABLE) {
5752             gen_ffp_instr(&buffer, stage, TRUE, FALSE, settings->op[stage].dst,
5753                           settings->op[stage].cop, settings->op[stage].carg0,
5754                           settings->op[stage].carg1, settings->op[stage].carg2);
5755             if(stage == 0) {
5756                 shader_addline(&buffer, "MOV ret.w, fragment.color.primary.w;\n");
5757             }
5758         } else if(op_equal) {
5759             gen_ffp_instr(&buffer, stage, TRUE, TRUE, settings->op[stage].dst,
5760                           settings->op[stage].cop, settings->op[stage].carg0,
5761                           settings->op[stage].carg1, settings->op[stage].carg2);
5762         } else {
5763             gen_ffp_instr(&buffer, stage, TRUE, FALSE, settings->op[stage].dst,
5764                           settings->op[stage].cop, settings->op[stage].carg0,
5765                           settings->op[stage].carg1, settings->op[stage].carg2);
5766             gen_ffp_instr(&buffer, stage, FALSE, TRUE, settings->op[stage].dst,
5767                           settings->op[stage].aop, settings->op[stage].aarg0,
5768                           settings->op[stage].aarg1, settings->op[stage].aarg2);
5769         }
5770     }
5771
5772     if(settings->sRGB_write) {
5773         shader_addline(&buffer, "MAD ret, fragment.color.secondary, specular_enable, %s;\n", final_combiner_src);
5774         arbfp_add_sRGB_correction(&buffer, "ret", "arg0", "arg1", "arg2", "tempreg", FALSE);
5775     } else {
5776         shader_addline(&buffer, "MAD result.color, fragment.color.secondary, specular_enable, %s;\n", final_combiner_src);
5777     }
5778
5779     /* Footer */
5780     shader_addline(&buffer, "END\n");
5781
5782     /* Generate the shader */
5783     GL_EXTCALL(glGenProgramsARB(1, &ret));
5784     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, ret));
5785     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
5786             strlen(buffer.buffer), buffer.buffer));
5787     checkGLcall("glProgramStringARB()");
5788
5789     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
5790     if (pos != -1)
5791     {
5792         FIXME("Fragment program error at position %d: %s\n", pos,
5793               debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
5794     }
5795     else
5796     {
5797         GLint native;
5798
5799         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
5800         checkGLcall("glGetProgramivARB()");
5801         if (!native) WARN("Program exceeds native resource limits.\n");
5802     }
5803
5804     shader_buffer_free(&buffer);
5805     return ret;
5806 }
5807
5808 static void fragment_prog_arbfp(DWORD state, IWineD3DStateBlockImpl *stateblock, struct wined3d_context *context)
5809 {
5810     IWineD3DDeviceImpl *device = stateblock->wineD3DDevice;
5811     struct shader_arb_priv *priv = device->fragment_priv;
5812     BOOL use_pshader = use_ps(stateblock);
5813     BOOL use_vshader = use_vs(stateblock);
5814     struct ffp_frag_settings settings;
5815     const struct arbfp_ffp_desc *desc;
5816     unsigned int i;
5817
5818     TRACE("state %#x, stateblock %p, context %p\n", state, stateblock, context);
5819
5820     if(isStateDirty(context, STATE_RENDER(WINED3DRS_FOGENABLE))) {
5821         if(!use_pshader && device->shader_backend == &arb_program_shader_backend && context->last_was_pshader) {
5822             /* Reload fixed function constants since they collide with the pixel shader constants */
5823             for(i = 0; i < MAX_TEXTURES; i++) {
5824                 set_bumpmat_arbfp(STATE_TEXTURESTAGE(i, WINED3DTSS_BUMPENVMAT00), stateblock, context);
5825             }
5826             state_texfactor_arbfp(STATE_RENDER(WINED3DRS_TEXTUREFACTOR), stateblock, context);
5827             state_arb_specularenable(STATE_RENDER(WINED3DRS_SPECULARENABLE), stateblock, context);
5828         } else if(use_pshader && !isStateDirty(context, device->StateTable[STATE_VSHADER].representative)) {
5829             device->shader_backend->shader_select(context, use_pshader, use_vshader);
5830         }
5831         return;
5832     }
5833
5834     if(!use_pshader) {
5835         /* Find or create a shader implementing the fixed function pipeline settings, then activate it */
5836         gen_ffp_frag_op(stateblock, &settings, FALSE);
5837         desc = (const struct arbfp_ffp_desc *)find_ffp_frag_shader(&priv->fragment_shaders, &settings);
5838         if(!desc) {
5839             struct arbfp_ffp_desc *new_desc = HeapAlloc(GetProcessHeap(), 0, sizeof(*new_desc));
5840             if (!new_desc)
5841             {
5842                 ERR("Out of memory\n");
5843                 return;
5844             }
5845             new_desc->num_textures_used = 0;
5846             for(i = 0; i < GL_LIMITS(texture_stages); i++) {
5847                 if(settings.op[i].cop == WINED3DTOP_DISABLE) break;
5848                 new_desc->num_textures_used = i;
5849             }
5850
5851             memcpy(&new_desc->parent.settings, &settings, sizeof(settings));
5852             new_desc->shader = gen_arbfp_ffp_shader(&settings, stateblock);
5853             add_ffp_frag_shader(&priv->fragment_shaders, &new_desc->parent);
5854             TRACE("Allocated fixed function replacement shader descriptor %p\n", new_desc);
5855             desc = new_desc;
5856         }
5857
5858         /* Now activate the replacement program. GL_FRAGMENT_PROGRAM_ARB is already active(however, note the
5859          * comment above the shader_select call below). If e.g. GLSL is active, the shader_select call will
5860          * deactivate it.
5861          */
5862         GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, desc->shader));
5863         checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, desc->shader)");
5864         priv->current_fprogram_id = desc->shader;
5865
5866         if(device->shader_backend == &arb_program_shader_backend && context->last_was_pshader) {
5867             /* Reload fixed function constants since they collide with the pixel shader constants */
5868             for(i = 0; i < MAX_TEXTURES; i++) {
5869                 set_bumpmat_arbfp(STATE_TEXTURESTAGE(i, WINED3DTSS_BUMPENVMAT00), stateblock, context);
5870             }
5871             state_texfactor_arbfp(STATE_RENDER(WINED3DRS_TEXTUREFACTOR), stateblock, context);
5872             state_arb_specularenable(STATE_RENDER(WINED3DRS_SPECULARENABLE), stateblock, context);
5873         }
5874         context->last_was_pshader = FALSE;
5875     } else {
5876         context->last_was_pshader = TRUE;
5877     }
5878
5879     /* Finally, select the shader. If a pixel shader is used, it will be set and enabled by the shader backend.
5880      * If this shader backend is arbfp(most likely), then it will simply overwrite the last fixed function replace-
5881      * ment shader. If the shader backend is not ARB, it currently is important that the opengl implementation
5882      * type overwrites GL_ARB_fragment_program. This is currently the case with GLSL. If we really want to use
5883      * atifs or nvrc pixel shaders with arb fragment programs we'd have to disable GL_FRAGMENT_PROGRAM_ARB here
5884      *
5885      * Don't call shader_select if the vertex shader is dirty, because it will be called later on by the vertex
5886      * shader handler
5887      */
5888     if(!isStateDirty(context, device->StateTable[STATE_VSHADER].representative)) {
5889         device->shader_backend->shader_select(context, use_pshader, use_vshader);
5890
5891         if (!isStateDirty(context, STATE_VERTEXSHADERCONSTANT) && (use_vshader || use_pshader)) {
5892             device->StateTable[STATE_VERTEXSHADERCONSTANT].apply(STATE_VERTEXSHADERCONSTANT, stateblock, context);
5893         }
5894     }
5895     if(use_pshader) {
5896         device->StateTable[STATE_PIXELSHADERCONSTANT].apply(STATE_PIXELSHADERCONSTANT, stateblock, context);
5897     }
5898 }
5899
5900 /* We can't link the fog states to the fragment state directly since the vertex pipeline links them
5901  * to FOGENABLE. A different linking in different pipeline parts can't be expressed in the combined
5902  * state table, so we need to handle that with a forwarding function. The other invisible side effect
5903  * is that changing the fog start and fog end(which links to FOGENABLE in vertex) results in the
5904  * fragment_prog_arbfp function being called because FOGENABLE is dirty, which calls this function here
5905  */
5906 static void state_arbfp_fog(DWORD state, IWineD3DStateBlockImpl *stateblock, struct wined3d_context *context)
5907 {
5908     enum fogsource new_source;
5909
5910     TRACE("state %#x, stateblock %p, context %p\n", state, stateblock, context);
5911
5912     if(!isStateDirty(context, STATE_PIXELSHADER)) {
5913         fragment_prog_arbfp(state, stateblock, context);
5914     }
5915
5916     if(!stateblock->renderState[WINED3DRS_FOGENABLE]) return;
5917
5918     if(stateblock->renderState[WINED3DRS_FOGTABLEMODE] == WINED3DFOG_NONE) {
5919         if(use_vs(stateblock)) {
5920             new_source = FOGSOURCE_VS;
5921         } else {
5922             if(stateblock->renderState[WINED3DRS_FOGVERTEXMODE] == WINED3DFOG_NONE || context->last_was_rhw) {
5923                 new_source = FOGSOURCE_COORD;
5924             } else {
5925                 new_source = FOGSOURCE_FFP;
5926             }
5927         }
5928     } else {
5929         new_source = FOGSOURCE_FFP;
5930     }
5931     if(new_source != context->fog_source) {
5932         context->fog_source = new_source;
5933         state_fogstartend(STATE_RENDER(WINED3DRS_FOGSTART), stateblock, context);
5934     }
5935 }
5936
5937 static void textransform(DWORD state, IWineD3DStateBlockImpl *stateblock, struct wined3d_context *context)
5938 {
5939     if(!isStateDirty(context, STATE_PIXELSHADER)) {
5940         fragment_prog_arbfp(state, stateblock, context);
5941     }
5942 }
5943
5944 #undef GLINFO_LOCATION
5945
5946 static const struct StateEntryTemplate arbfp_fragmentstate_template[] = {
5947     {STATE_RENDER(WINED3DRS_TEXTUREFACTOR),               { STATE_RENDER(WINED3DRS_TEXTUREFACTOR),              state_texfactor_arbfp   }, WINED3D_GL_EXT_NONE             },
5948     {STATE_TEXTURESTAGE(0, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5949     {STATE_TEXTURESTAGE(0, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5950     {STATE_TEXTURESTAGE(0, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5951     {STATE_TEXTURESTAGE(0, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5952     {STATE_TEXTURESTAGE(0, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5953     {STATE_TEXTURESTAGE(0, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5954     {STATE_TEXTURESTAGE(0, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5955     {STATE_TEXTURESTAGE(0, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5956     {STATE_TEXTURESTAGE(0, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5957     {STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5958     {STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5959     {STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5960     {STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5961     {STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
5962     {STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
5963     {STATE_TEXTURESTAGE(1, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5964     {STATE_TEXTURESTAGE(1, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5965     {STATE_TEXTURESTAGE(1, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5966     {STATE_TEXTURESTAGE(1, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5967     {STATE_TEXTURESTAGE(1, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5968     {STATE_TEXTURESTAGE(1, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5969     {STATE_TEXTURESTAGE(1, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5970     {STATE_TEXTURESTAGE(1, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5971     {STATE_TEXTURESTAGE(1, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5972     {STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5973     {STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5974     {STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5975     {STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5976     {STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
5977     {STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
5978     {STATE_TEXTURESTAGE(2, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5979     {STATE_TEXTURESTAGE(2, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5980     {STATE_TEXTURESTAGE(2, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5981     {STATE_TEXTURESTAGE(2, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5982     {STATE_TEXTURESTAGE(2, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5983     {STATE_TEXTURESTAGE(2, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5984     {STATE_TEXTURESTAGE(2, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5985     {STATE_TEXTURESTAGE(2, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5986     {STATE_TEXTURESTAGE(2, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5987     {STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5988     {STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5989     {STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5990     {STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5991     {STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
5992     {STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
5993     {STATE_TEXTURESTAGE(3, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5994     {STATE_TEXTURESTAGE(3, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5995     {STATE_TEXTURESTAGE(3, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5996     {STATE_TEXTURESTAGE(3, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5997     {STATE_TEXTURESTAGE(3, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5998     {STATE_TEXTURESTAGE(3, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5999     {STATE_TEXTURESTAGE(3, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6000     {STATE_TEXTURESTAGE(3, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6001     {STATE_TEXTURESTAGE(3, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6002     {STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6003     {STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6004     {STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6005     {STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6006     {STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6007     {STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6008     {STATE_TEXTURESTAGE(4, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6009     {STATE_TEXTURESTAGE(4, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6010     {STATE_TEXTURESTAGE(4, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6011     {STATE_TEXTURESTAGE(4, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6012     {STATE_TEXTURESTAGE(4, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6013     {STATE_TEXTURESTAGE(4, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6014     {STATE_TEXTURESTAGE(4, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6015     {STATE_TEXTURESTAGE(4, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6016     {STATE_TEXTURESTAGE(4, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6017     {STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6018     {STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6019     {STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6020     {STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6021     {STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6022     {STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6023     {STATE_TEXTURESTAGE(5, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6024     {STATE_TEXTURESTAGE(5, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6025     {STATE_TEXTURESTAGE(5, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6026     {STATE_TEXTURESTAGE(5, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6027     {STATE_TEXTURESTAGE(5, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6028     {STATE_TEXTURESTAGE(5, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6029     {STATE_TEXTURESTAGE(5, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6030     {STATE_TEXTURESTAGE(5, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6031     {STATE_TEXTURESTAGE(5, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6032     {STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6033     {STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6034     {STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6035     {STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6036     {STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6037     {STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6038     {STATE_TEXTURESTAGE(6, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6039     {STATE_TEXTURESTAGE(6, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6040     {STATE_TEXTURESTAGE(6, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6041     {STATE_TEXTURESTAGE(6, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6042     {STATE_TEXTURESTAGE(6, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6043     {STATE_TEXTURESTAGE(6, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6044     {STATE_TEXTURESTAGE(6, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6045     {STATE_TEXTURESTAGE(6, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6046     {STATE_TEXTURESTAGE(6, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6047     {STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6048     {STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6049     {STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6050     {STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6051     {STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6052     {STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6053     {STATE_TEXTURESTAGE(7, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6054     {STATE_TEXTURESTAGE(7, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6055     {STATE_TEXTURESTAGE(7, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6056     {STATE_TEXTURESTAGE(7, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6057     {STATE_TEXTURESTAGE(7, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6058     {STATE_TEXTURESTAGE(7, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6059     {STATE_TEXTURESTAGE(7, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6060     {STATE_TEXTURESTAGE(7, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6061     {STATE_TEXTURESTAGE(7, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6062     {STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6063     {STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6064     {STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6065     {STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6066     {STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6067     {STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6068     {STATE_SAMPLER(0),                                    { STATE_SAMPLER(0),                                   sampler_texdim          }, WINED3D_GL_EXT_NONE             },
6069     {STATE_SAMPLER(1),                                    { STATE_SAMPLER(1),                                   sampler_texdim          }, WINED3D_GL_EXT_NONE             },
6070     {STATE_SAMPLER(2),                                    { STATE_SAMPLER(2),                                   sampler_texdim          }, WINED3D_GL_EXT_NONE             },
6071     {STATE_SAMPLER(3),                                    { STATE_SAMPLER(3),                                   sampler_texdim          }, WINED3D_GL_EXT_NONE             },
6072     {STATE_SAMPLER(4),                                    { STATE_SAMPLER(4),                                   sampler_texdim          }, WINED3D_GL_EXT_NONE             },
6073     {STATE_SAMPLER(5),                                    { STATE_SAMPLER(5),                                   sampler_texdim          }, WINED3D_GL_EXT_NONE             },
6074     {STATE_SAMPLER(6),                                    { STATE_SAMPLER(6),                                   sampler_texdim          }, WINED3D_GL_EXT_NONE             },
6075     {STATE_SAMPLER(7),                                    { STATE_SAMPLER(7),                                   sampler_texdim          }, WINED3D_GL_EXT_NONE             },
6076     {STATE_PIXELSHADER,                                   { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6077     {STATE_RENDER(WINED3DRS_FOGENABLE),                   { STATE_RENDER(WINED3DRS_FOGENABLE),                  state_arbfp_fog         }, WINED3D_GL_EXT_NONE             },
6078     {STATE_RENDER(WINED3DRS_FOGTABLEMODE),                { STATE_RENDER(WINED3DRS_FOGENABLE),                  state_arbfp_fog         }, WINED3D_GL_EXT_NONE             },
6079     {STATE_RENDER(WINED3DRS_FOGVERTEXMODE),               { STATE_RENDER(WINED3DRS_FOGENABLE),                  state_arbfp_fog         }, WINED3D_GL_EXT_NONE             },
6080     {STATE_RENDER(WINED3DRS_FOGSTART),                    { STATE_RENDER(WINED3DRS_FOGSTART),                   state_fogstartend       }, WINED3D_GL_EXT_NONE             },
6081     {STATE_RENDER(WINED3DRS_FOGEND),                      { STATE_RENDER(WINED3DRS_FOGSTART),                   state_fogstartend       }, WINED3D_GL_EXT_NONE             },
6082     {STATE_RENDER(WINED3DRS_SRGBWRITEENABLE),             { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6083     {STATE_RENDER(WINED3DRS_FOGCOLOR),                    { STATE_RENDER(WINED3DRS_FOGCOLOR),                   state_fogcolor          }, WINED3D_GL_EXT_NONE             },
6084     {STATE_RENDER(WINED3DRS_FOGDENSITY),                  { STATE_RENDER(WINED3DRS_FOGDENSITY),                 state_fogdensity        }, WINED3D_GL_EXT_NONE             },
6085     {STATE_TEXTURESTAGE(0,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(0, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6086     {STATE_TEXTURESTAGE(1,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(1, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6087     {STATE_TEXTURESTAGE(2,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(2, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6088     {STATE_TEXTURESTAGE(3,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(3, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6089     {STATE_TEXTURESTAGE(4,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(4, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6090     {STATE_TEXTURESTAGE(5,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(5, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6091     {STATE_TEXTURESTAGE(6,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(6, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6092     {STATE_TEXTURESTAGE(7,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(7, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6093     {STATE_RENDER(WINED3DRS_SPECULARENABLE),              { STATE_RENDER(WINED3DRS_SPECULARENABLE),             state_arb_specularenable}, WINED3D_GL_EXT_NONE             },
6094     {0 /* Terminate */,                                   { 0,                                                  0                       }, WINED3D_GL_EXT_NONE             },
6095 };
6096
6097 const struct fragment_pipeline arbfp_fragment_pipeline = {
6098     arbfp_enable,
6099     arbfp_get_caps,
6100     arbfp_alloc,
6101     arbfp_free,
6102     shader_arb_color_fixup_supported,
6103     arbfp_fragmentstate_template,
6104     TRUE /* We can disable projected textures */
6105 };
6106
6107 #define GLINFO_LOCATION device->adapter->gl_info
6108
6109 struct arbfp_blit_priv {
6110     GLenum yuy2_rect_shader, yuy2_2d_shader;
6111     GLenum uyvy_rect_shader, uyvy_2d_shader;
6112     GLenum yv12_rect_shader, yv12_2d_shader;
6113 };
6114
6115 static HRESULT arbfp_blit_alloc(IWineD3DDevice *iface) {
6116     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *) iface;
6117     device->blit_priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(struct arbfp_blit_priv));
6118     if(!device->blit_priv) {
6119         ERR("Out of memory\n");
6120         return E_OUTOFMEMORY;
6121     }
6122     return WINED3D_OK;
6123 }
6124
6125 /* Context activation is done by the caller. */
6126 static void arbfp_blit_free(IWineD3DDevice *iface) {
6127     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *) iface;
6128     struct arbfp_blit_priv *priv = device->blit_priv;
6129
6130     ENTER_GL();
6131     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yuy2_rect_shader));
6132     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yuy2_2d_shader));
6133     GL_EXTCALL(glDeleteProgramsARB(1, &priv->uyvy_rect_shader));
6134     GL_EXTCALL(glDeleteProgramsARB(1, &priv->uyvy_2d_shader));
6135     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yv12_rect_shader));
6136     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yv12_2d_shader));
6137     checkGLcall("Delete yuv programs");
6138     LEAVE_GL();
6139 }
6140
6141 static BOOL gen_planar_yuv_read(struct wined3d_shader_buffer *buffer, enum yuv_fixup yuv_fixup,
6142         GLenum textype, char *luminance)
6143 {
6144     char chroma;
6145     const char *tex, *texinstr;
6146
6147     if (yuv_fixup == YUV_FIXUP_UYVY) {
6148         chroma = 'x';
6149         *luminance = 'w';
6150     } else {
6151         chroma = 'w';
6152         *luminance = 'x';
6153     }
6154     switch(textype) {
6155         case GL_TEXTURE_2D:             tex = "2D";     texinstr = "TXP"; break;
6156         case GL_TEXTURE_RECTANGLE_ARB:  tex = "RECT";   texinstr = "TEX"; break;
6157         default:
6158             /* This is more tricky than just replacing the texture type - we have to navigate
6159              * properly in the texture to find the correct chroma values
6160              */
6161             FIXME("Implement yuv correction for non-2d, non-rect textures\n");
6162             return FALSE;
6163     }
6164
6165     /* First we have to read the chroma values. This means we need at least two pixels(no filtering),
6166      * or 4 pixels(with filtering). To get the unmodified chromas, we have to rid ourselves of the
6167      * filtering when we sample the texture.
6168      *
6169      * These are the rules for reading the chroma:
6170      *
6171      * Even pixel: Cr
6172      * Even pixel: U
6173      * Odd pixel: V
6174      *
6175      * So we have to get the sampling x position in non-normalized coordinates in integers
6176      */
6177     if(textype != GL_TEXTURE_RECTANGLE_ARB) {
6178         shader_addline(buffer, "MUL texcrd.xy, fragment.texcoord[0], size.x;\n");
6179         shader_addline(buffer, "MOV texcrd.w, size.x;\n");
6180     } else {
6181         shader_addline(buffer, "MOV texcrd, fragment.texcoord[0];\n");
6182     }
6183     /* We must not allow filtering between pixel x and x+1, this would mix U and V
6184      * Vertical filtering is ok. However, bear in mind that the pixel center is at
6185      * 0.5, so add 0.5.
6186      */
6187     shader_addline(buffer, "FLR texcrd.x, texcrd.x;\n");
6188     shader_addline(buffer, "ADD texcrd.x, texcrd.x, coef.y;\n");
6189
6190     /* Divide the x coordinate by 0.5 and get the fraction. This gives 0.25 and 0.75 for the
6191      * even and odd pixels respectively
6192      */
6193     shader_addline(buffer, "MUL texcrd2, texcrd, coef.y;\n");
6194     shader_addline(buffer, "FRC texcrd2, texcrd2;\n");
6195
6196     /* Sample Pixel 1 */
6197     shader_addline(buffer, "%s luminance, texcrd, texture[0], %s;\n", texinstr, tex);
6198
6199     /* Put the value into either of the chroma values */
6200     shader_addline(buffer, "SGE temp.x, texcrd2.x, coef.y;\n");
6201     shader_addline(buffer, "MUL chroma.x, luminance.%c, temp.x;\n", chroma);
6202     shader_addline(buffer, "SLT temp.x, texcrd2.x, coef.y;\n");
6203     shader_addline(buffer, "MUL chroma.y, luminance.%c, temp.x;\n", chroma);
6204
6205     /* Sample pixel 2. If we read an even pixel(SLT above returned 1), sample
6206      * the pixel right to the current one. Otherwise, sample the left pixel.
6207      * Bias and scale the SLT result to -1;1 and add it to the texcrd.x.
6208      */
6209     shader_addline(buffer, "MAD temp.x, temp.x, coef.z, -coef.x;\n");
6210     shader_addline(buffer, "ADD texcrd.x, texcrd, temp.x;\n");
6211     shader_addline(buffer, "%s luminance, texcrd, texture[0], %s;\n", texinstr, tex);
6212
6213     /* Put the value into the other chroma */
6214     shader_addline(buffer, "SGE temp.x, texcrd2.x, coef.y;\n");
6215     shader_addline(buffer, "MAD chroma.y, luminance.%c, temp.x, chroma.y;\n", chroma);
6216     shader_addline(buffer, "SLT temp.x, texcrd2.x, coef.y;\n");
6217     shader_addline(buffer, "MAD chroma.x, luminance.%c, temp.x, chroma.x;\n", chroma);
6218
6219     /* TODO: If filtering is enabled, sample a 2nd pair of pixels left or right of
6220      * the current one and lerp the two U and V values
6221      */
6222
6223     /* This gives the correctly filtered luminance value */
6224     shader_addline(buffer, "TEX luminance, fragment.texcoord[0], texture[0], %s;\n", tex);
6225
6226     return TRUE;
6227 }
6228
6229 static BOOL gen_yv12_read(struct wined3d_shader_buffer *buffer, GLenum textype, char *luminance)
6230 {
6231     const char *tex;
6232
6233     switch(textype) {
6234         case GL_TEXTURE_2D:             tex = "2D";     break;
6235         case GL_TEXTURE_RECTANGLE_ARB:  tex = "RECT";   break;
6236         default:
6237             FIXME("Implement yv12 correction for non-2d, non-rect textures\n");
6238             return FALSE;
6239     }
6240
6241     /* YV12 surfaces contain a WxH sized luminance plane, followed by a (W/2)x(H/2)
6242      * V and a (W/2)x(H/2) U plane, each with 8 bit per pixel. So the effective
6243      * bitdepth is 12 bits per pixel. Since the U and V planes have only half the
6244      * pitch of the luminance plane, the packing into the gl texture is a bit
6245      * unfortunate. If the whole texture is interpreted as luminance data it looks
6246      * approximately like this:
6247      *
6248      *        +----------------------------------+----
6249      *        |                                  |
6250      *        |                                  |
6251      *        |                                  |
6252      *        |                                  |
6253      *        |                                  |   2
6254      *        |            LUMINANCE             |   -
6255      *        |                                  |   3
6256      *        |                                  |
6257      *        |                                  |
6258      *        |                                  |
6259      *        |                                  |
6260      *        +----------------+-----------------+----
6261      *        |                |                 |
6262      *        |  U even rows   |  U odd rows     |
6263      *        |                |                 |   1
6264      *        +----------------+------------------   -
6265      *        |                |                 |   3
6266      *        |  V even rows   |  V odd rows     |
6267      *        |                |                 |
6268      *        +----------------+-----------------+----
6269      *        |                |                 |
6270      *        |     0.5        |       0.5       |
6271      *
6272      * So it appears as if there are 4 chroma images, but in fact the odd rows
6273      * in the chroma images are in the same row as the even ones. So its is
6274      * kinda tricky to read
6275      *
6276      * When reading from rectangle textures, keep in mind that the input y coordinates
6277      * go from 0 to d3d_height, whereas the opengl texture height is 1.5 * d3d_height
6278      */
6279     shader_addline(buffer, "PARAM yv12_coef = {%f, %f, %f, %f};\n",
6280             2.0f / 3.0f, 1.0f / 6.0f, (2.0f / 3.0f) + (1.0f / 6.0f), 1.0f / 3.0f);
6281
6282     shader_addline(buffer, "MOV texcrd, fragment.texcoord[0];\n");
6283     /* the chroma planes have only half the width */
6284     shader_addline(buffer, "MUL texcrd.x, texcrd.x, coef.y;\n");
6285
6286     /* The first value is between 2/3 and 5/6th of the texture's height, so scale+bias
6287      * the coordinate. Also read the right side of the image when reading odd lines
6288      *
6289      * Don't forget to clamp the y values in into the range, otherwise we'll get filtering
6290      * bleeding
6291      */
6292     if(textype == GL_TEXTURE_2D) {
6293
6294         shader_addline(buffer, "RCP chroma.w, size.y;\n");
6295
6296         shader_addline(buffer, "MUL texcrd2.y, texcrd.y, size.y;\n");
6297
6298         shader_addline(buffer, "FLR texcrd2.y, texcrd2.y;\n");
6299         shader_addline(buffer, "MAD texcrd.y, texcrd.y, yv12_coef.y, yv12_coef.x;\n");
6300
6301         /* Read odd lines from the right side(add size * 0.5 to the x coordinate */
6302         shader_addline(buffer, "ADD texcrd2.x, texcrd2.y, yv12_coef.y;\n"); /* To avoid 0.5 == 0.5 comparisons */
6303         shader_addline(buffer, "FRC texcrd2.x, texcrd2.x;\n");
6304         shader_addline(buffer, "SGE texcrd2.x, texcrd2.x, coef.y;\n");
6305         shader_addline(buffer, "MAD texcrd.x, texcrd2.x, coef.y, texcrd.x;\n");
6306
6307         /* clamp, keep the half pixel origin in mind */
6308         shader_addline(buffer, "MAD temp.y, coef.y, chroma.w, yv12_coef.x;\n");
6309         shader_addline(buffer, "MAX texcrd.y, temp.y, texcrd.y;\n");
6310         shader_addline(buffer, "MAD temp.y, -coef.y, chroma.w, yv12_coef.z;\n");
6311         shader_addline(buffer, "MIN texcrd.y, temp.y, texcrd.y;\n");
6312     } else {
6313         /* Read from [size - size+size/4] */
6314         shader_addline(buffer, "FLR texcrd.y, texcrd.y;\n");
6315         shader_addline(buffer, "MAD texcrd.y, texcrd.y, coef.w, size.y;\n");
6316
6317         /* Read odd lines from the right side(add size * 0.5 to the x coordinate */
6318         shader_addline(buffer, "ADD texcrd2.x, texcrd.y, yv12_coef.y;\n"); /* To avoid 0.5 == 0.5 comparisons */
6319         shader_addline(buffer, "FRC texcrd2.x, texcrd2.x;\n");
6320         shader_addline(buffer, "SGE texcrd2.x, texcrd2.x, coef.y;\n");
6321         shader_addline(buffer, "MUL texcrd2.x, texcrd2.x, size.x;\n");
6322         shader_addline(buffer, "MAD texcrd.x, texcrd2.x, coef.y, texcrd.x;\n");
6323
6324         /* Make sure to read exactly from the pixel center */
6325         shader_addline(buffer, "FLR texcrd.y, texcrd.y;\n");
6326         shader_addline(buffer, "ADD texcrd.y, texcrd.y, coef.y;\n");
6327
6328         /* Clamp */
6329         shader_addline(buffer, "MAD temp.y, size.y, coef.w, size.y;\n");
6330         shader_addline(buffer, "ADD temp.y, temp.y, -coef.y;\n");
6331         shader_addline(buffer, "MIN texcrd.y, temp.y, texcrd.y;\n");
6332         shader_addline(buffer, "ADD temp.y, size.y, -coef.y;\n");
6333         shader_addline(buffer, "MAX texcrd.y, temp.y, texcrd.y;\n");
6334     }
6335     /* Read the texture, put the result into the output register */
6336     shader_addline(buffer, "TEX temp, texcrd, texture[0], %s;\n", tex);
6337     shader_addline(buffer, "MOV chroma.x, temp.w;\n");
6338
6339     /* The other chroma value is 1/6th of the texture lower, from 5/6th to 6/6th
6340      * No need to clamp because we're just reusing the already clamped value from above
6341      */
6342     if(textype == GL_TEXTURE_2D) {
6343         shader_addline(buffer, "ADD texcrd.y, texcrd.y, yv12_coef.y;\n");
6344     } else {
6345         shader_addline(buffer, "MAD texcrd.y, size.y, coef.w, texcrd.y;\n");
6346     }
6347     shader_addline(buffer, "TEX temp, texcrd, texture[0], %s;\n", tex);
6348     shader_addline(buffer, "MOV chroma.y, temp.w;\n");
6349
6350     /* Sample the luminance value. It is in the top 2/3rd of the texture, so scale the y coordinate.
6351      * Clamp the y coordinate to prevent the chroma values from bleeding into the sampled luminance
6352      * values due to filtering
6353      */
6354     shader_addline(buffer, "MOV texcrd, fragment.texcoord[0];\n");
6355     if(textype == GL_TEXTURE_2D) {
6356         /* Multiply the y coordinate by 2/3 and clamp it */
6357         shader_addline(buffer, "MUL texcrd.y, texcrd.y, yv12_coef.x;\n");
6358         shader_addline(buffer, "MAD temp.y, -coef.y, chroma.w, yv12_coef.x;\n");
6359         shader_addline(buffer, "MIN texcrd.y, temp.y, texcrd.y;\n");
6360         shader_addline(buffer, "TEX luminance, texcrd, texture[0], %s;\n", tex);
6361     } else {
6362         /* Reading from texture_rectangles is pretty straightforward, just use the unmodified
6363          * texture coordinate. It is still a good idea to clamp it though, since the opengl texture
6364          * is bigger
6365          */
6366         shader_addline(buffer, "ADD temp.x, size.y, -coef.y;\n");
6367         shader_addline(buffer, "MIN texcrd.y, texcrd.y, size.x;\n");
6368         shader_addline(buffer, "TEX luminance, texcrd, texture[0], %s;\n", tex);
6369     }
6370     *luminance = 'a';
6371
6372     return TRUE;
6373 }
6374
6375 /* Context activation is done by the caller. */
6376 static GLuint gen_yuv_shader(IWineD3DDeviceImpl *device, enum yuv_fixup yuv_fixup, GLenum textype)
6377 {
6378     GLenum shader;
6379     struct wined3d_shader_buffer buffer;
6380     char luminance_component;
6381     struct arbfp_blit_priv *priv = device->blit_priv;
6382     GLint pos;
6383
6384     /* Shader header */
6385     if (!shader_buffer_init(&buffer))
6386     {
6387         ERR("Failed to initialize shader buffer.\n");
6388         return 0;
6389     }
6390
6391     ENTER_GL();
6392     GL_EXTCALL(glGenProgramsARB(1, &shader));
6393     checkGLcall("GL_EXTCALL(glGenProgramsARB(1, &shader))");
6394     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader));
6395     checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader)");
6396     LEAVE_GL();
6397     if(!shader) {
6398         shader_buffer_free(&buffer);
6399         return 0;
6400     }
6401
6402     /* The YUY2 and UYVY formats contain two pixels packed into a 32 bit macropixel,
6403      * giving effectively 16 bit per pixel. The color consists of a luminance(Y) and
6404      * two chroma(U and V) values. Each macropixel has two luminance values, one for
6405      * each single pixel it contains, and one U and one V value shared between both
6406      * pixels.
6407      *
6408      * The data is loaded into an A8L8 texture. With YUY2, the luminance component
6409      * contains the luminance and alpha the chroma. With UYVY it is vice versa. Thus
6410      * take the format into account when generating the read swizzles
6411      *
6412      * Reading the Y value is straightforward - just sample the texture. The hardware
6413      * takes care of filtering in the horizontal and vertical direction.
6414      *
6415      * Reading the U and V values is harder. We have to avoid filtering horizontally,
6416      * because that would mix the U and V values of one pixel or two adjacent pixels.
6417      * Thus floor the texture coordinate and add 0.5 to get an unfiltered read,
6418      * regardless of the filtering setting. Vertical filtering works automatically
6419      * though - the U and V values of two rows are mixed nicely.
6420      *
6421      * Appart of avoiding filtering issues, the code has to know which value it just
6422      * read, and where it can find the other one. To determine this, it checks if
6423      * it sampled an even or odd pixel, and shifts the 2nd read accordingly.
6424      *
6425      * Handling horizontal filtering of U and V values requires reading a 2nd pair
6426      * of pixels, extracting U and V and mixing them. This is not implemented yet.
6427      *
6428      * An alternative implementation idea is to load the texture as A8R8G8B8 texture,
6429      * with width / 2. This way one read gives all 3 values, finding U and V is easy
6430      * in an unfiltered situation. Finding the luminance on the other hand requires
6431      * finding out if it is an odd or even pixel. The real drawback of this approach
6432      * is filtering. This would have to be emulated completely in the shader, reading
6433      * up two 2 packed pixels in up to 2 rows and interpolating both horizontally and
6434      * vertically. Beyond that it would require adjustments to the texture handling
6435      * code to deal with the width scaling
6436      */
6437     shader_addline(&buffer, "!!ARBfp1.0\n");
6438     shader_addline(&buffer, "TEMP luminance;\n");
6439     shader_addline(&buffer, "TEMP temp;\n");
6440     shader_addline(&buffer, "TEMP chroma;\n");
6441     shader_addline(&buffer, "TEMP texcrd;\n");
6442     shader_addline(&buffer, "TEMP texcrd2;\n");
6443     shader_addline(&buffer, "PARAM coef = {1.0, 0.5, 2.0, 0.25};\n");
6444     shader_addline(&buffer, "PARAM yuv_coef = {1.403, 0.344, 0.714, 1.770};\n");
6445     shader_addline(&buffer, "PARAM size = program.local[0];\n");
6446
6447     switch (yuv_fixup)
6448     {
6449         case YUV_FIXUP_UYVY:
6450         case YUV_FIXUP_YUY2:
6451             if (!gen_planar_yuv_read(&buffer, yuv_fixup, textype, &luminance_component))
6452             {
6453                 shader_buffer_free(&buffer);
6454                 return 0;
6455             }
6456             break;
6457
6458         case YUV_FIXUP_YV12:
6459             if (!gen_yv12_read(&buffer, textype, &luminance_component))
6460             {
6461                 shader_buffer_free(&buffer);
6462                 return 0;
6463             }
6464             break;
6465
6466         default:
6467             FIXME("Unsupported YUV fixup %#x\n", yuv_fixup);
6468             shader_buffer_free(&buffer);
6469             return 0;
6470     }
6471
6472     /* Calculate the final result. Formula is taken from
6473      * http://www.fourcc.org/fccyvrgb.php. Note that the chroma
6474      * ranges from -0.5 to 0.5
6475      */
6476     shader_addline(&buffer, "SUB chroma.xy, chroma, coef.y;\n");
6477
6478     shader_addline(&buffer, "MAD result.color.x, chroma.x, yuv_coef.x, luminance.%c;\n", luminance_component);
6479     shader_addline(&buffer, "MAD temp.x, -chroma.y, yuv_coef.y, luminance.%c;\n", luminance_component);
6480     shader_addline(&buffer, "MAD result.color.y, -chroma.x, yuv_coef.z, temp.x;\n");
6481     shader_addline(&buffer, "MAD result.color.z, chroma.y, yuv_coef.w, luminance.%c;\n", luminance_component);
6482     shader_addline(&buffer, "END\n");
6483
6484     ENTER_GL();
6485     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
6486             strlen(buffer.buffer), buffer.buffer));
6487     checkGLcall("glProgramStringARB()");
6488
6489     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
6490     if (pos != -1)
6491     {
6492         FIXME("Fragment program error at position %d: %s\n", pos,
6493               debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
6494     }
6495     else
6496     {
6497         GLint native;
6498
6499         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
6500         checkGLcall("glGetProgramivARB()");
6501         if (!native) WARN("Program exceeds native resource limits.\n");
6502     }
6503
6504     shader_buffer_free(&buffer);
6505     LEAVE_GL();
6506
6507     switch (yuv_fixup)
6508     {
6509         case YUV_FIXUP_YUY2:
6510             if (textype == GL_TEXTURE_RECTANGLE_ARB) priv->yuy2_rect_shader = shader;
6511             else priv->yuy2_2d_shader = shader;
6512             break;
6513
6514         case YUV_FIXUP_UYVY:
6515             if (textype == GL_TEXTURE_RECTANGLE_ARB) priv->uyvy_rect_shader = shader;
6516             else priv->uyvy_2d_shader = shader;
6517             break;
6518
6519         case YUV_FIXUP_YV12:
6520             if (textype == GL_TEXTURE_RECTANGLE_ARB) priv->yv12_rect_shader = shader;
6521             else priv->yv12_2d_shader = shader;
6522             break;
6523     }
6524
6525     return shader;
6526 }
6527
6528 /* Context activation is done by the caller. */
6529 static HRESULT arbfp_blit_set(IWineD3DDevice *iface, const struct GlPixelFormatDesc *format_desc,
6530         GLenum textype, UINT width, UINT height)
6531 {
6532     GLenum shader;
6533     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *) iface;
6534     float size[4] = {width, height, 1, 1};
6535     struct arbfp_blit_priv *priv = device->blit_priv;
6536     enum yuv_fixup yuv_fixup;
6537
6538     if (!is_yuv_fixup(format_desc->color_fixup))
6539     {
6540         TRACE("Fixup:\n");
6541         dump_color_fixup_desc(format_desc->color_fixup);
6542         /* Don't bother setting up a shader for unconverted formats */
6543         ENTER_GL();
6544         glEnable(textype);
6545         checkGLcall("glEnable(textype)");
6546         LEAVE_GL();
6547         return WINED3D_OK;
6548     }
6549
6550     yuv_fixup = get_yuv_fixup(format_desc->color_fixup);
6551
6552     switch(yuv_fixup)
6553     {
6554         case YUV_FIXUP_YUY2:
6555             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->yuy2_rect_shader : priv->yuy2_2d_shader;
6556             break;
6557
6558         case YUV_FIXUP_UYVY:
6559             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->uyvy_rect_shader : priv->uyvy_2d_shader;
6560             break;
6561
6562         case YUV_FIXUP_YV12:
6563             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->yv12_rect_shader : priv->yv12_2d_shader;
6564             break;
6565
6566         default:
6567             FIXME("Unsupported YUV fixup %#x, not setting a shader\n", yuv_fixup);
6568             ENTER_GL();
6569             glEnable(textype);
6570             checkGLcall("glEnable(textype)");
6571             LEAVE_GL();
6572             return E_NOTIMPL;
6573     }
6574
6575     if (!shader) shader = gen_yuv_shader(device, yuv_fixup, textype);
6576
6577     ENTER_GL();
6578     glEnable(GL_FRAGMENT_PROGRAM_ARB);
6579     checkGLcall("glEnable(GL_FRAGMENT_PROGRAM_ARB)");
6580     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader));
6581     checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader)");
6582     GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, 0, size));
6583     checkGLcall("glProgramLocalParameter4fvARB");
6584     LEAVE_GL();
6585
6586     return WINED3D_OK;
6587 }
6588
6589 /* Context activation is done by the caller. */
6590 static void arbfp_blit_unset(IWineD3DDevice *iface) {
6591     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *) iface;
6592
6593     ENTER_GL();
6594     glDisable(GL_FRAGMENT_PROGRAM_ARB);
6595     checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
6596     glDisable(GL_TEXTURE_2D);
6597     checkGLcall("glDisable(GL_TEXTURE_2D)");
6598     if(GL_SUPPORT(ARB_TEXTURE_CUBE_MAP)) {
6599         glDisable(GL_TEXTURE_CUBE_MAP_ARB);
6600         checkGLcall("glDisable(GL_TEXTURE_CUBE_MAP_ARB)");
6601     }
6602     if(GL_SUPPORT(ARB_TEXTURE_RECTANGLE)) {
6603         glDisable(GL_TEXTURE_RECTANGLE_ARB);
6604         checkGLcall("glDisable(GL_TEXTURE_RECTANGLE_ARB)");
6605     }
6606     LEAVE_GL();
6607 }
6608
6609 static BOOL arbfp_blit_color_fixup_supported(struct color_fixup_desc fixup)
6610 {
6611     enum yuv_fixup yuv_fixup;
6612
6613     if (TRACE_ON(d3d_shader) && TRACE_ON(d3d))
6614     {
6615         TRACE("Checking support for fixup:\n");
6616         dump_color_fixup_desc(fixup);
6617     }
6618
6619     if (is_identity_fixup(fixup))
6620     {
6621         TRACE("[OK]\n");
6622         return TRUE;
6623     }
6624
6625     /* We only support YUV conversions. */
6626     if (!is_yuv_fixup(fixup))
6627     {
6628         TRACE("[FAILED]\n");
6629         return FALSE;
6630     }
6631
6632     yuv_fixup = get_yuv_fixup(fixup);
6633     switch(yuv_fixup)
6634     {
6635         case YUV_FIXUP_YUY2:
6636         case YUV_FIXUP_UYVY:
6637         case YUV_FIXUP_YV12:
6638             TRACE("[OK]\n");
6639             return TRUE;
6640
6641         default:
6642             FIXME("Unsupported YUV fixup %#x\n", yuv_fixup);
6643             TRACE("[FAILED]\n");
6644             return FALSE;
6645     }
6646 }
6647
6648 const struct blit_shader arbfp_blit = {
6649     arbfp_blit_alloc,
6650     arbfp_blit_free,
6651     arbfp_blit_set,
6652     arbfp_blit_unset,
6653     arbfp_blit_color_fixup_supported,
6654 };
6655
6656 #undef GLINFO_LOCATION