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