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

   1 /*
   2  * Pixel and vertex shaders implementation using ARB_vertex_program
   3  * and ARB_fragment_program GL extensions.
   4  *
   5  * Copyright 2002-2003 Jason Edmeades
   6  * Copyright 2002-2003 Raphael Junqueira
   7  * Copyright 2004 Christian Costa
   8  * Copyright 2005 Oliver Stieber
   9  * Copyright 2006 Ivan Gyurdiev
  10  * Copyright 2006 Jason Green
  11  * Copyright 2006 Henri Verbeet
  12  * Copyright 2007-2008 Stefan Dösinger for CodeWeavers
  13  * Copyright 2009 Henri Verbeet for CodeWeavers
  14  *
  15  * This library is free software; you can redistribute it and/or
  16  * modify it under the terms of the GNU Lesser General Public
  17  * License as published by the Free Software Foundation; either
  18  * version 2.1 of the License, or (at your option) any later version.
  19  *
  20  * This library is distributed in the hope that it will be useful,
  21  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  22  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  23  * Lesser General Public License for more details.
  24  *
  25  * You should have received a copy of the GNU Lesser General Public
  26  * License along with this library; if not, write to the Free Software
  27  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301, USA
  28  */
  29
  30 #include "config.h"
  31
  32 #include <math.h>
  33 #include <stdio.h>
  34
  35 #include "wined3d_private.h"
  36
  37 WINE_DEFAULT_DEBUG_CHANNEL(d3d_shader);
  38 WINE_DECLARE_DEBUG_CHANNEL(d3d_constants);
  39 WINE_DECLARE_DEBUG_CHANNEL(d3d_caps);
  40 WINE_DECLARE_DEBUG_CHANNEL(d3d);
  41
  42 #define GLINFO_LOCATION      (*gl_info)
  43
  44 /* GL locking for state handlers is done by the caller. */
  45 static BOOL need_mova_const(IWineD3DBaseShader *shader, const struct wined3d_gl_info *gl_info)
  46 {
  47     IWineD3DBaseShaderImpl *This = (IWineD3DBaseShaderImpl *) shader;
  48     if(!This->baseShader.reg_maps.usesmova) return FALSE;
  49     return !GL_SUPPORT(NV_VERTEX_PROGRAM2_OPTION);
  50 }
  51
  52 /* Returns TRUE if result.clip from GL_NV_vertex_program2 should be used and FALSE otherwise */
  53 static inline BOOL use_nv_clip(const struct wined3d_gl_info *gl_info)
  54 {
  55     return GL_SUPPORT(NV_VERTEX_PROGRAM2_OPTION);
  56 }
  57
  58 static BOOL need_helper_const(const struct wined3d_gl_info *gl_info)
  59 {
  60     if (!GL_SUPPORT(NV_VERTEX_PROGRAM) /* Need to init colors. */
  61         || gl_info->quirks & WINED3D_QUIRK_ARB_VS_OFFSET_LIMIT /* Load the immval offset. */
  62         || gl_info->quirks & WINED3D_QUIRK_SET_TEXCOORD_W /* Have to init texcoords. */
  63         || (!use_nv_clip(gl_info)) /* Init the clip texcoord */)
  64     {
  65         return TRUE;
  66     }
  67     return FALSE;
  68 }
  69
  70 static unsigned int reserved_vs_const(IWineD3DBaseShader *shader, const struct wined3d_gl_info *gl_info)
  71 {
  72     unsigned int ret = 1;
  73     /* We use one PARAM for the pos fixup, and in some cases one to load
  74      * some immediate values into the shader
  75      */
  76     if(need_helper_const(gl_info)) ret++;
  77     if(need_mova_const(shader, gl_info)) ret++;
  78     return ret;
  79 }
  80
  81 static inline BOOL ffp_clip_emul(IWineD3DStateBlockImpl *stateblock)
  82 {
  83     return stateblock->lowest_disabled_stage < 7;
  84 }
  85
  86 /* Internally used shader constants. Applications can use constants 0 to GL_LIMITS(vshader_constantsF) - 1,
  87  * so upload them above that
  88  */
  89 #define ARB_SHADER_PRIVCONST_BASE (GL_LIMITS(vshader_constantsF) - 1)
  90 #define ARB_SHADER_PRIVCONST_POS ARB_SHADER_PRIVCONST_BASE + 0
  91
  92 /* ARB_program_shader private data */
  93
  94 struct control_frame
  95 {
  96     struct                          list entry;
  97     enum
  98     {
  99         IF,
 100         IFC,
 101         LOOP,
 102         REP
 103     } type;
 104     BOOL                            muting;
 105     BOOL                            outer_loop;
 106     union
 107     {
 108         unsigned int                loop_no;
 109         unsigned int                ifc_no;
 110     };
 111     struct wined3d_shader_loop_control loop_control;
 112     BOOL                            had_else;
 113 };
 114
 115 struct arb_ps_np2fixup_info
 116 {
 117     struct ps_np2fixup_info         super;
 118     /* For ARB we need a offset value:
 119      * With both GLSL and ARB mode the NP2 fixup information (the texture dimensions) are stored in a
 120      * consecutive way (GLSL uses a uniform array). Since ARB doesn't know the notion of a "standalone"
 121      * array we need an offset to the index inside the program local parameter array. */
 122     UINT                            offset;
 123 };
 124
 125 struct arb_ps_compile_args
 126 {
 127     struct ps_compile_args          super;
 128     WORD                            bools;
 129     WORD                            clip;  /* only a boolean, use a WORD for alignment */
 130     unsigned char                   loop_ctrl[MAX_CONST_I][3];
 131 };
 132
 133 struct stb_const_desc
 134 {
 135     unsigned char           texunit;
 136     UINT                    const_num;
 137 };
 138
 139 struct arb_ps_compiled_shader
 140 {
 141     struct arb_ps_compile_args      args;
 142     struct arb_ps_np2fixup_info     np2fixup_info;
 143     struct stb_const_desc           bumpenvmatconst[MAX_TEXTURES];
 144     struct stb_const_desc           luminanceconst[MAX_TEXTURES];
 145     UINT                            int_consts[MAX_CONST_I];
 146     GLuint                          prgId;
 147     UINT                            ycorrection;
 148     unsigned char                   numbumpenvmatconsts;
 149     char                            num_int_consts;
 150 };
 151
 152 struct arb_vs_compile_args
 153 {
 154     struct vs_compile_args          super;
 155     union
 156     {
 157         struct
 158         {
 159             WORD                    bools;
 160             char                    clip_control[2];
 161         }                           boolclip;
 162         DWORD                       boolclip_compare;
 163     };
 164     DWORD                           ps_signature;
 165     union
 166     {
 167         unsigned char               vertex_samplers[4];
 168         DWORD                       vertex_samplers_compare;
 169     };
 170     unsigned char                   loop_ctrl[MAX_CONST_I][3];
 171 };
 172
 173 struct arb_vs_compiled_shader
 174 {
 175     struct arb_vs_compile_args      args;
 176     GLuint                          prgId;
 177     UINT                            int_consts[MAX_CONST_I];
 178     char                            num_int_consts;
 179     char                            need_color_unclamp;
 180     UINT                            pos_fixup;
 181 };
 182
 183 struct recorded_instruction
 184 {
 185     struct wined3d_shader_instruction ins;
 186     struct list entry;
 187 };
 188
 189 struct shader_arb_ctx_priv
 190 {
 191     char addr_reg[20];
 192     enum
 193     {
 194         /* plain GL_ARB_vertex_program or GL_ARB_fragment_program */
 195         ARB,
 196         /* GL_NV_vertex_progam2_option or GL_NV_fragment_program_option */
 197         NV2,
 198         /* GL_NV_vertex_program3 or GL_NV_fragment_program2 */
 199         NV3
 200     } target_version;
 201
 202     const struct arb_vs_compile_args    *cur_vs_args;
 203     const struct arb_ps_compile_args    *cur_ps_args;
 204     const struct arb_ps_compiled_shader *compiled_fprog;
 205     const struct arb_vs_compiled_shader *compiled_vprog;
 206     struct arb_ps_np2fixup_info         *cur_np2fixup_info;
 207     struct list                         control_frames;
 208     struct list                         record;
 209     BOOL                                recording;
 210     BOOL                                muted;
 211     unsigned int                        num_loops, loop_depth, num_ifcs;
 212     int                                 aL;
 213
 214     unsigned int                        vs_clipplanes;
 215     BOOL                                footer_written;
 216     BOOL                                in_main_func;
 217
 218     /* For 3.0 vertex shaders */
 219     const char                          *vs_output[MAX_REG_OUTPUT];
 220     /* For 2.x and earlier vertex shaders */
 221     const char                          *texcrd_output[8], *color_output[2], *fog_output;
 222
 223     /* 3.0 pshader input for compatibility with fixed function */
 224     const char                          *ps_input[MAX_REG_INPUT];
 225 };
 226
 227 struct ps_signature
 228 {
 229     struct wined3d_shader_signature_element *sig;
 230     DWORD                               idx;
 231     struct wine_rb_entry                entry;
 232 };
 233
 234 struct arb_pshader_private {
 235     struct arb_ps_compiled_shader   *gl_shaders;
 236     UINT                            num_gl_shaders, shader_array_size;
 237     BOOL                            has_signature_idx;
 238     DWORD                           input_signature_idx;
 239     DWORD                           clipplane_emulation;
 240     BOOL                            clamp_consts;
 241 };
 242
 243 struct arb_vshader_private {
 244     struct arb_vs_compiled_shader   *gl_shaders;
 245     UINT                            num_gl_shaders, shader_array_size;
 246 };
 247
 248 struct shader_arb_priv
 249 {
 250     GLuint                  current_vprogram_id;
 251     GLuint                  current_fprogram_id;
 252     const struct arb_ps_compiled_shader *compiled_fprog;
 253     const struct arb_vs_compiled_shader *compiled_vprog;
 254     GLuint                  depth_blt_vprogram_id;
 255     GLuint                  depth_blt_fprogram_id[tex_type_count];
 256     BOOL                    use_arbfp_fixed_func;
 257     struct wine_rb_tree     fragment_shaders;
 258     BOOL                    last_ps_const_clamped;
 259     BOOL                    last_vs_color_unclamp;
 260
 261     struct wine_rb_tree     signature_tree;
 262     DWORD ps_sig_number;
 263 };
 264
 265 /********************************************************
 266  * ARB_[vertex/fragment]_program helper functions follow
 267  ********************************************************/
 268
 269 /* Loads floating point constants into the currently set ARB_vertex/fragment_program.
 270  * When constant_list == NULL, it will load all the constants.
 271  *
 272  * @target_type should be either GL_VERTEX_PROGRAM_ARB (for vertex shaders)
 273  *  or GL_FRAGMENT_PROGRAM_ARB (for pixel shaders)
 274  */
 275 /* GL locking is done by the caller */
 276 static unsigned int shader_arb_load_constantsF(IWineD3DBaseShaderImpl *This, const struct wined3d_gl_info *gl_info,
 277         GLuint target_type, unsigned int max_constants, const float *constants, char *dirty_consts)
 278 {
 279     local_constant* lconst;
 280     DWORD i = 0, j;
 281     unsigned int ret;
 282
 283     if (TRACE_ON(d3d_shader)) {
 284         for(i = 0; i < max_constants; i++) {
 285             if(!dirty_consts[i]) continue;
 286             TRACE_(d3d_constants)("Loading constants %i: %f, %f, %f, %f\n", i,
 287                         constants[i * 4 + 0], constants[i * 4 + 1],
 288                         constants[i * 4 + 2], constants[i * 4 + 3]);
 289         }
 290     }
 291     /* In 1.X pixel shaders constants are implicitly clamped in the range [-1;1] */
 292     if (target_type == GL_FRAGMENT_PROGRAM_ARB && This->baseShader.reg_maps.shader_version.major == 1)
 293     {
 294         float lcl_const[4];
 295         /* ps 1.x supports only 8 constants, clamp only those. When switching between 1.x and higher
 296          * shaders, the first 8 constants are marked dirty for reload
 297          */
 298         for(; i < min(8, max_constants); i++) {
 299             if(!dirty_consts[i]) continue;
 300             dirty_consts[i] = 0;
 301
 302             j = 4 * i;
 303             if (constants[j + 0] > 1.0f) lcl_const[0] = 1.0f;
 304             else if (constants[j + 0] < -1.0f) lcl_const[0] = -1.0f;
 305             else lcl_const[0] = constants[j + 0];
 306
 307             if (constants[j + 1] > 1.0f) lcl_const[1] = 1.0f;
 308             else if (constants[j + 1] < -1.0f) lcl_const[1] = -1.0f;
 309             else lcl_const[1] = constants[j + 1];
 310
 311             if (constants[j + 2] > 1.0f) lcl_const[2] = 1.0f;
 312             else if (constants[j + 2] < -1.0f) lcl_const[2] = -1.0f;
 313             else lcl_const[2] = constants[j + 2];
 314
 315             if (constants[j + 3] > 1.0f) lcl_const[3] = 1.0f;
 316             else if (constants[j + 3] < -1.0f) lcl_const[3] = -1.0f;
 317             else lcl_const[3] = constants[j + 3];
 318
 319             GL_EXTCALL(glProgramEnvParameter4fvARB(target_type, i, lcl_const));
 320         }
 321
 322         /* If further constants are dirty, reload them without clamping.
 323          *
 324          * The alternative is not to touch them, but then we cannot reset the dirty constant count
 325          * to zero. That's bad for apps that only use PS 1.x shaders, because in that case the code
 326          * above would always re-check the first 8 constants since max_constant remains at the init
 327          * value
 328          */
 329     }
 330
 331     if(GL_SUPPORT(EXT_GPU_PROGRAM_PARAMETERS)) {
 332         /* TODO: Benchmark if we're better of with finding the dirty constants ourselves,
 333          * or just reloading *all* constants at once
 334          *
 335         GL_EXTCALL(glProgramEnvParameters4fvEXT(target_type, i, max_constants, constants + (i * 4)));
 336          */
 337         for(; i < max_constants; i++) {
 338             if(!dirty_consts[i]) continue;
 339
 340             /* Find the next block of dirty constants */
 341             dirty_consts[i] = 0;
 342             j = i;
 343             for(i++; (i < max_constants) && dirty_consts[i]; i++) {
 344                 dirty_consts[i] = 0;
 345             }
 346
 347             GL_EXTCALL(glProgramEnvParameters4fvEXT(target_type, j, i - j, constants + (j * 4)));
 348         }
 349     } else {
 350         for(; i < max_constants; i++) {
 351             if(dirty_consts[i]) {
 352                 dirty_consts[i] = 0;
 353                 GL_EXTCALL(glProgramEnvParameter4fvARB(target_type, i, constants + (i * 4)));
 354             }
 355         }
 356     }
 357     checkGLcall("glProgramEnvParameter4fvARB()");
 358
 359     /* Load immediate constants */
 360     if(This->baseShader.load_local_constsF) {
 361         if (TRACE_ON(d3d_shader)) {
 362             LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry) {
 363                 GLfloat* values = (GLfloat*)lconst->value;
 364                 TRACE_(d3d_constants)("Loading local constants %i: %f, %f, %f, %f\n", lconst->idx,
 365                         values[0], values[1], values[2], values[3]);
 366             }
 367         }
 368         /* Immediate constants are clamped for 1.X shaders at loading times */
 369         ret = 0;
 370         LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry) {
 371             dirty_consts[lconst->idx] = 1; /* Dirtify so the non-immediate constant overwrites it next time */
 372             ret = max(ret, lconst->idx + 1);
 373             GL_EXTCALL(glProgramEnvParameter4fvARB(target_type, lconst->idx, (GLfloat*)lconst->value));
 374         }
 375         checkGLcall("glProgramEnvParameter4fvARB()");
 376         return ret; /* The loaded immediate constants need reloading for the next shader */
 377     } else {
 378         return 0; /* No constants are dirty now */
 379     }
 380 }
 381
 382 /**
 383  * Loads the texture dimensions for NP2 fixup into the currently set ARB_[vertex/fragment]_programs.
 384  */
 385 static void shader_arb_load_np2fixup_constants(
 386     IWineD3DDevice* device,
 387     char usePixelShader,
 388     char useVertexShader) {
 389
 390     IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl *) device;
 391     const struct shader_arb_priv* const priv = (const struct shader_arb_priv *) deviceImpl->shader_priv;
 392     IWineD3DStateBlockImpl* stateBlock = deviceImpl->stateBlock;
 393     const struct wined3d_gl_info *gl_info = &deviceImpl->adapter->gl_info;
 394
 395     if (!usePixelShader) {
 396         /* NP2 texcoord fixup is (currently) only done for pixelshaders. */
 397         return;
 398     }
 399
 400     if (priv->compiled_fprog && priv->compiled_fprog->np2fixup_info.super.active) {
 401         const struct arb_ps_np2fixup_info* const fixup = &priv->compiled_fprog->np2fixup_info;
 402         UINT i;
 403         WORD active = fixup->super.active;
 404         GLfloat np2fixup_constants[4 * MAX_FRAGMENT_SAMPLERS];
 405
 406         for (i = 0; active; active >>= 1, ++i) {
 407             const unsigned char idx = fixup->super.idx[i];
 408             const IWineD3DTextureImpl* const tex = (const IWineD3DTextureImpl*) stateBlock->textures[i];
 409             GLfloat* tex_dim = &np2fixup_constants[(idx >> 1) * 4];
 410
 411             if (!(active & 1)) continue;
 412
 413             if (!tex) {
 414                 FIXME("Nonexistent texture is flagged for NP2 texcoord fixup\n");
 415                 continue;
 416             }
 417
 418             if (idx % 2) {
 419                 tex_dim[2] = tex->baseTexture.pow2Matrix[0]; tex_dim[3] = tex->baseTexture.pow2Matrix[5];
 420             } else {
 421                 tex_dim[0] = tex->baseTexture.pow2Matrix[0]; tex_dim[1] = tex->baseTexture.pow2Matrix[5];
 422             }
 423         }
 424
 425         for (i = 0; i < fixup->super.num_consts; ++i) {
 426             GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB,
 427                                                    fixup->offset + i, &np2fixup_constants[i * 4]));
 428         }
 429     }
 430 }
 431
 432 /* GL locking is done by the caller. */
 433 static inline void shader_arb_ps_local_constants(IWineD3DDeviceImpl* deviceImpl)
 434 {
 435     const struct wined3d_context *context = context_get_current();
 436     IWineD3DStateBlockImpl* stateBlock = deviceImpl->stateBlock;
 437     const struct wined3d_gl_info *gl_info = context->gl_info;
 438     unsigned char i;
 439     struct shader_arb_priv *priv = deviceImpl->shader_priv;
 440     const struct arb_ps_compiled_shader *gl_shader = priv->compiled_fprog;
 441
 442     for(i = 0; i < gl_shader->numbumpenvmatconsts; i++)
 443     {
 444         int texunit = gl_shader->bumpenvmatconst[i].texunit;
 445
 446         /* The state manager takes care that this function is always called if the bump env matrix changes */
 447         const float *data = (const float *)&stateBlock->textureState[texunit][WINED3DTSS_BUMPENVMAT00];
 448         GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, gl_shader->bumpenvmatconst[i].const_num, data));
 449
 450         if (gl_shader->luminanceconst[i].const_num != WINED3D_CONST_NUM_UNUSED)
 451         {
 452             /* WINED3DTSS_BUMPENVLSCALE and WINED3DTSS_BUMPENVLOFFSET are next to each other.
 453              * point gl to the scale, and load 4 floats. x = scale, y = offset, z and w are junk, we
 454              * don't care about them. The pointers are valid for sure because the stateblock is bigger.
 455              * (they're WINED3DTSS_TEXTURETRANSFORMFLAGS and WINED3DTSS_ADDRESSW, so most likely 0 or NaN
 456             */
 457             const float *scale = (const float *)&stateBlock->textureState[texunit][WINED3DTSS_BUMPENVLSCALE];
 458             GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, gl_shader->luminanceconst[i].const_num, scale));
 459         }
 460     }
 461     checkGLcall("Load bumpmap consts");
 462
 463     if(gl_shader->ycorrection != WINED3D_CONST_NUM_UNUSED)
 464     {
 465         /* ycorrection.x: Backbuffer height(onscreen) or 0(offscreen).
 466         * ycorrection.y: -1.0(onscreen), 1.0(offscreen)
 467         * ycorrection.z: 1.0
 468         * ycorrection.w: 0.0
 469         */
 470         float val[4];
 471         val[0] = context->render_offscreen ? 0.0f
 472                 : ((IWineD3DSurfaceImpl *) deviceImpl->render_targets[0])->currentDesc.Height;
 473         val[1] = context->render_offscreen ? 1.0f : -1.0f;
 474         val[2] = 1.0f;
 475         val[3] = 0.0f;
 476         GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, gl_shader->ycorrection, val));
 477         checkGLcall("y correction loading");
 478     }
 479
 480     if(gl_shader->num_int_consts == 0) return;
 481
 482     for(i = 0; i < MAX_CONST_I; i++)
 483     {
 484         if(gl_shader->int_consts[i] != WINED3D_CONST_NUM_UNUSED)
 485         {
 486             float val[4];
 487             val[0] = stateBlock->pixelShaderConstantI[4 * i];
 488             val[1] = stateBlock->pixelShaderConstantI[4 * i + 1];
 489             val[2] = stateBlock->pixelShaderConstantI[4 * i + 2];
 490             val[3] = -1.0f;
 491
 492             GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, gl_shader->int_consts[i], val));
 493         }
 494     }
 495     checkGLcall("Load ps int consts");
 496 }
 497
 498 /* GL locking is done by the caller. */
 499 static inline void shader_arb_vs_local_constants(IWineD3DDeviceImpl* deviceImpl)
 500 {
 501     IWineD3DStateBlockImpl* stateBlock;
 502     const struct wined3d_gl_info *gl_info = &deviceImpl->adapter->gl_info;
 503     unsigned char i;
 504     struct shader_arb_priv *priv = deviceImpl->shader_priv;
 505     const struct arb_vs_compiled_shader *gl_shader = priv->compiled_vprog;
 506
 507     /* Upload the position fixup */
 508     GL_EXTCALL(glProgramLocalParameter4fvARB(GL_VERTEX_PROGRAM_ARB, gl_shader->pos_fixup, deviceImpl->posFixup));
 509
 510     if(gl_shader->num_int_consts == 0) return;
 511
 512     stateBlock = deviceImpl->stateBlock;
 513
 514     for(i = 0; i < MAX_CONST_I; i++)
 515     {
 516         if(gl_shader->int_consts[i] != WINED3D_CONST_NUM_UNUSED)
 517         {
 518             float val[4];
 519             val[0] = stateBlock->vertexShaderConstantI[4 * i];
 520             val[1] = stateBlock->vertexShaderConstantI[4 * i + 1];
 521             val[2] = stateBlock->vertexShaderConstantI[4 * i + 2];
 522             val[3] = -1.0f;
 523
 524             GL_EXTCALL(glProgramLocalParameter4fvARB(GL_VERTEX_PROGRAM_ARB, gl_shader->int_consts[i], val));
 525         }
 526     }
 527     checkGLcall("Load vs int consts");
 528 }
 529
 530 /**
 531  * Loads the app-supplied constants into the currently set ARB_[vertex/fragment]_programs.
 532  *
 533  * We only support float constants in ARB at the moment, so don't
 534  * worry about the Integers or Booleans
 535  */
 536 /* GL locking is done by the caller (state handler) */
 537 static void shader_arb_load_constants(const struct wined3d_context *context, char usePixelShader, char useVertexShader)
 538 {
 539     IWineD3DDeviceImpl *device = ((IWineD3DSurfaceImpl *)context->surface)->resource.wineD3DDevice;
 540     IWineD3DStateBlockImpl* stateBlock = device->stateBlock;
 541     const struct wined3d_gl_info *gl_info = context->gl_info;
 542
 543     if (useVertexShader) {
 544         IWineD3DBaseShaderImpl* vshader = (IWineD3DBaseShaderImpl*) stateBlock->vertexShader;
 545
 546         /* Load DirectX 9 float constants for vertex shader */
 547         device->highest_dirty_vs_const = shader_arb_load_constantsF(vshader, gl_info, GL_VERTEX_PROGRAM_ARB,
 548                 device->highest_dirty_vs_const, stateBlock->vertexShaderConstantF, context->vshader_const_dirty);
 549         shader_arb_vs_local_constants(device);
 550     }
 551
 552     if (usePixelShader) {
 553         IWineD3DBaseShaderImpl* pshader = (IWineD3DBaseShaderImpl*) stateBlock->pixelShader;
 554
 555         /* Load DirectX 9 float constants for pixel shader */
 556         device->highest_dirty_ps_const = shader_arb_load_constantsF(pshader, gl_info, GL_FRAGMENT_PROGRAM_ARB,
 557                 device->highest_dirty_ps_const, stateBlock->pixelShaderConstantF, context->pshader_const_dirty);
 558         shader_arb_ps_local_constants(device);
 559     }
 560 }
 561
 562 static void shader_arb_update_float_vertex_constants(IWineD3DDevice *iface, UINT start, UINT count)
 563 {
 564     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
 565     struct wined3d_context *context = context_get_current();
 566
 567     /* We don't want shader constant dirtification to be an O(contexts), so just dirtify the active
 568      * context. On a context switch the old context will be fully dirtified */
 569     if (!context || ((IWineD3DSurfaceImpl *)context->surface)->resource.wineD3DDevice != This) return;
 570
 571     memset(context->vshader_const_dirty + start, 1, sizeof(*context->vshader_const_dirty) * count);
 572     This->highest_dirty_vs_const = max(This->highest_dirty_vs_const, start + count);
 573 }
 574
 575 static void shader_arb_update_float_pixel_constants(IWineD3DDevice *iface, UINT start, UINT count)
 576 {
 577     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
 578     struct wined3d_context *context = context_get_current();
 579
 580     /* We don't want shader constant dirtification to be an O(contexts), so just dirtify the active
 581      * context. On a context switch the old context will be fully dirtified */
 582     if (!context || ((IWineD3DSurfaceImpl *)context->surface)->resource.wineD3DDevice != This) return;
 583
 584     memset(context->pshader_const_dirty + start, 1, sizeof(*context->pshader_const_dirty) * count);
 585     This->highest_dirty_ps_const = max(This->highest_dirty_ps_const, start + count);
 586 }
 587
 588 static DWORD *local_const_mapping(IWineD3DBaseShaderImpl *This)
 589 {
 590     DWORD *ret;
 591     DWORD idx = 0;
 592     const local_constant *lconst;
 593
 594     if(This->baseShader.load_local_constsF || list_empty(&This->baseShader.constantsF)) return NULL;
 595
 596     ret = HeapAlloc(GetProcessHeap(), 0, sizeof(DWORD) * This->baseShader.limits.constant_float);
 597     if(!ret) {
 598         ERR("Out of memory\n");
 599         return NULL;
 600     }
 601
 602     LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry) {
 603         ret[lconst->idx] = idx++;
 604     }
 605     return ret;
 606 }
 607
 608 /* Generate the variable & register declarations for the ARB_vertex_program output target */
 609 static DWORD shader_generate_arb_declarations(IWineD3DBaseShader *iface, const shader_reg_maps *reg_maps,
 610         struct wined3d_shader_buffer *buffer, const struct wined3d_gl_info *gl_info, DWORD *lconst_map,
 611         DWORD *num_clipplanes, struct shader_arb_ctx_priv *ctx)
 612 {
 613     IWineD3DBaseShaderImpl* This = (IWineD3DBaseShaderImpl*) iface;
 614     DWORD i, next_local = 0;
 615     char pshader = shader_is_pshader_version(reg_maps->shader_version.type);
 616     unsigned max_constantsF;
 617     const local_constant *lconst;
 618     DWORD map;
 619
 620     /* In pixel shaders, all private constants are program local, we don't need anything
 621      * from program.env. Thus we can advertise the full set of constants in pixel shaders.
 622      * If we need a private constant the GL implementation will squeeze it in somewhere
 623      *
 624      * With vertex shaders we need the posFixup and on some GL implementations 4 helper
 625      * immediate values. The posFixup is loaded using program.env for now, so always
 626      * subtract one from the number of constants. If the shader uses indirect addressing,
 627      * account for the helper const too because we have to declare all availabke d3d constants
 628      * and don't know which are actually used.
 629      */
 630     if(pshader) {
 631         max_constantsF = GL_LIMITS(pshader_constantsF);
 632     } else {
 633         if(This->baseShader.reg_maps.usesrelconstF) {
 634             DWORD highest_constf = 0, clip_limit;
 635             max_constantsF = GL_LIMITS(vshader_constantsF) - reserved_vs_const(iface, gl_info);
 636             max_constantsF -= count_bits(This->baseShader.reg_maps.integer_constants);
 637
 638             for(i = 0; i < This->baseShader.limits.constant_float; i++)
 639             {
 640                 DWORD idx = i >> 5;
 641                 DWORD shift = i & 0x1f;
 642                 if(reg_maps->constf[idx] & (1 << shift)) highest_constf = i;
 643             }
 644
 645             clip_limit = GL_LIMITS(clipplanes);
 646             if(ctx->target_version == ARB) clip_limit = min(clip_limit, 4);
 647             *num_clipplanes = min(clip_limit, max_constantsF - highest_constf - 1);
 648             max_constantsF -= *num_clipplanes;
 649             if(*num_clipplanes < clip_limit)
 650             {
 651                 WARN("Only %u clipplanes out of %u enabled\n", *num_clipplanes, GL_LIMITS(clipplanes));
 652             }
 653         }
 654         else
 655         {
 656             if(ctx->target_version >= NV2) *num_clipplanes = GL_LIMITS(clipplanes);
 657             else *num_clipplanes = min(GL_LIMITS(clipplanes), 4);
 658             max_constantsF = GL_LIMITS(vshader_constantsF);
 659         }
 660     }
 661
 662     for (i = 0, map = reg_maps->temporary; map; map >>= 1, ++i)
 663     {
 664         if (map & 1) shader_addline(buffer, "TEMP R%u;\n", i);
 665     }
 666
 667     for (i = 0, map = reg_maps->address; map; map >>= 1, ++i)
 668     {
 669         if (map & 1) shader_addline(buffer, "ADDRESS A%u;\n", i);
 670     }
 671
 672     if (pshader && reg_maps->shader_version.major == 1 && reg_maps->shader_version.minor <= 3)
 673     {
 674         for (i = 0, map = reg_maps->texcoord; map; map >>= 1, ++i)
 675         {
 676             if (map & 1) shader_addline(buffer, "TEMP T%u;\n", i);
 677         }
 678     }
 679
 680     /* Load local constants using the program-local space,
 681      * this avoids reloading them each time the shader is used
 682      */
 683     if(lconst_map) {
 684         LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry) {
 685             shader_addline(buffer, "PARAM C%u = program.local[%u];\n", lconst->idx,
 686                            lconst_map[lconst->idx]);
 687             next_local = max(next_local, lconst_map[lconst->idx] + 1);
 688         }
 689     }
 690
 691     /* we use the array-based constants array if the local constants are marked for loading,
 692      * because then we use indirect addressing, or when the local constant list is empty,
 693      * because then we don't know if we're using indirect addressing or not. If we're hardcoding
 694      * local constants do not declare the loaded constants as an array because ARB compilers usually
 695      * do not optimize unused constants away
 696      */
 697     if(This->baseShader.reg_maps.usesrelconstF) {
 698         /* Need to PARAM the environment parameters (constants) so we can use relative addressing */
 699         shader_addline(buffer, "PARAM C[%d] = { program.env[0..%d] };\n",
 700                     max_constantsF, max_constantsF - 1);
 701     } else {
 702         for(i = 0; i < max_constantsF; i++) {
 703             DWORD idx, mask;
 704             idx = i >> 5;
 705             mask = 1 << (i & 0x1f);
 706             if(!shader_constant_is_local(This, i) && (This->baseShader.reg_maps.constf[idx] & mask)) {
 707                 shader_addline(buffer, "PARAM C%d = program.env[%d];\n",i, i);
 708             }
 709         }
 710     }
 711
 712     return next_local;
 713 }
 714
 715 static const char * const shift_tab[] = {
 716     "dummy",     /*  0 (none) */
 717     "coefmul.x", /*  1 (x2)   */
 718     "coefmul.y", /*  2 (x4)   */
 719     "coefmul.z", /*  3 (x8)   */
 720     "coefmul.w", /*  4 (x16)  */
 721     "dummy",     /*  5 (x32)  */
 722     "dummy",     /*  6 (x64)  */
 723     "dummy",     /*  7 (x128) */
 724     "dummy",     /*  8 (d256) */
 725     "dummy",     /*  9 (d128) */
 726     "dummy",     /* 10 (d64)  */
 727     "dummy",     /* 11 (d32)  */
 728     "coefdiv.w", /* 12 (d16)  */
 729     "coefdiv.z", /* 13 (d8)   */
 730     "coefdiv.y", /* 14 (d4)   */
 731     "coefdiv.x"  /* 15 (d2)   */
 732 };
 733
 734 static void shader_arb_get_write_mask(const struct wined3d_shader_instruction *ins,
 735         const struct wined3d_shader_dst_param *dst, char *write_mask)
 736 {
 737     char *ptr = write_mask;
 738
 739     if (dst->write_mask != WINED3DSP_WRITEMASK_ALL)
 740     {
 741         *ptr++ = '.';
 742         if (dst->write_mask & WINED3DSP_WRITEMASK_0) *ptr++ = 'x';
 743         if (dst->write_mask & WINED3DSP_WRITEMASK_1) *ptr++ = 'y';
 744         if (dst->write_mask & WINED3DSP_WRITEMASK_2) *ptr++ = 'z';
 745         if (dst->write_mask & WINED3DSP_WRITEMASK_3) *ptr++ = 'w';
 746     }
 747
 748     *ptr = '\0';
 749 }
 750
 751 static void shader_arb_get_swizzle(const struct wined3d_shader_src_param *param, BOOL fixup, char *swizzle_str)
 752 {
 753     /* For registers of type WINED3DDECLTYPE_D3DCOLOR, data is stored as "bgra",
 754      * but addressed as "rgba". To fix this we need to swap the register's x
 755      * and z components. */
 756     const char *swizzle_chars = fixup ? "zyxw" : "xyzw";
 757     char *ptr = swizzle_str;
 758
 759     /* swizzle bits fields: wwzzyyxx */
 760     DWORD swizzle = param->swizzle;
 761     DWORD swizzle_x = swizzle & 0x03;
 762     DWORD swizzle_y = (swizzle >> 2) & 0x03;
 763     DWORD swizzle_z = (swizzle >> 4) & 0x03;
 764     DWORD swizzle_w = (swizzle >> 6) & 0x03;
 765
 766     /* If the swizzle is the default swizzle (ie, "xyzw"), we don't need to
 767      * generate a swizzle string. Unless we need to our own swizzling. */
 768     if (swizzle != WINED3DSP_NOSWIZZLE || fixup)
 769     {
 770         *ptr++ = '.';
 771         if (swizzle_x == swizzle_y && swizzle_x == swizzle_z && swizzle_x == swizzle_w) {
 772             *ptr++ = swizzle_chars[swizzle_x];
 773         } else {
 774             *ptr++ = swizzle_chars[swizzle_x];
 775             *ptr++ = swizzle_chars[swizzle_y];
 776             *ptr++ = swizzle_chars[swizzle_z];
 777             *ptr++ = swizzle_chars[swizzle_w];
 778         }
 779     }
 780
 781     *ptr = '\0';
 782 }
 783
 784 static void shader_arb_request_a0(const struct wined3d_shader_instruction *ins, const char *src)
 785 {
 786     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
 787     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
 788
 789     if(strcmp(priv->addr_reg, src) == 0) return;
 790
 791     strcpy(priv->addr_reg, src);
 792     shader_addline(buffer, "ARL A0.x, %s;\n", src);
 793 }
 794
 795 static void shader_arb_get_src_param(const struct wined3d_shader_instruction *ins,
 796         const struct wined3d_shader_src_param *src, unsigned int tmpreg, char *outregstr);
 797
 798 static void shader_arb_get_register_name(const struct wined3d_shader_instruction *ins,
 799         const struct wined3d_shader_register *reg, char *register_name, BOOL *is_color)
 800 {
 801     /* oPos, oFog and oPts in D3D */
 802     static const char * const rastout_reg_names[] = {"TMP_OUT", "result.fogcoord", "result.pointsize"};
 803     IWineD3DBaseShaderImpl *This = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
 804     BOOL pshader = shader_is_pshader_version(This->baseShader.reg_maps.shader_version.type);
 805     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
 806
 807     *is_color = FALSE;
 808
 809     switch (reg->type)
 810     {
 811         case WINED3DSPR_TEMP:
 812             sprintf(register_name, "R%u", reg->idx);
 813             break;
 814
 815         case WINED3DSPR_INPUT:
 816             if (pshader)
 817             {
 818                 if(This->baseShader.reg_maps.shader_version.major < 3)
 819                 {
 820                     if (reg->idx == 0) strcpy(register_name, "fragment.color.primary");
 821                     else strcpy(register_name, "fragment.color.secondary");
 822                 }
 823                 else
 824                 {
 825                     if(reg->rel_addr)
 826                     {
 827                         char rel_reg[50];
 828                         shader_arb_get_src_param(ins, reg->rel_addr, 0, rel_reg);
 829
 830                         if(strcmp(rel_reg, "**aL_emul**") == 0)
 831                         {
 832                             DWORD idx = ctx->aL + reg->idx;
 833                             if(idx < MAX_REG_INPUT)
 834                             {
 835                                 strcpy(register_name, ctx->ps_input[idx]);
 836                             }
 837                             else
 838                             {
 839                                 ERR("Pixel shader input register out of bounds: %u\n", idx);
 840                                 sprintf(register_name, "out_of_bounds_%u", idx);
 841                             }
 842                         }
 843                         else if(This->baseShader.reg_maps.input_registers & 0x0300)
 844                         {
 845                             /* There are two ways basically:
 846                              *
 847                              * 1) Use the unrolling code that is used for loop emulation and unroll the loop.
 848                              *    That means trouble if the loop also contains a breakc or if the control values
 849                              *    aren't local constants.
 850                              * 2) Generate an if block that checks if aL.y < 8, == 8 or == 9 and selects the
 851                              *    source dynamically. The trouble is that we cannot simply read aL.y because it
 852                              *    is an ADDRESS register. We could however push it, load .zw with a value and use
 853                              *    ADAC to load the condition code register and pop it again afterwards
 854                              */
 855                             FIXME("Relative input register addressing with more than 8 registers\n");
 856
 857                             /* This is better than nothing for now */
 858                             sprintf(register_name, "fragment.texcoord[%s + %u]", rel_reg, reg->idx);
 859                         }
 860                         else if(ctx->cur_ps_args->super.vp_mode != vertexshader)
 861                         {
 862                             /* This is problematic because we'd have to consult the ctx->ps_input strings
 863                              * for where to find the varying. Some may be "0.0", others can be texcoords or
 864                              * colors. This needs either a pipeline replacement to make the vertex shader feed
 865                              * proper varyings, or loop unrolling
 866                              *
 867                              * For now use the texcoords and hope for the best
 868                              */
 869                             FIXME("Non-vertex shader varying input with indirect addressing\n");
 870                             sprintf(register_name, "fragment.texcoord[%s + %u]", rel_reg, reg->idx);
 871                         }
 872                         else
 873                         {
 874                             /* D3D supports indirect addressing only with aL in loop registers. The loop instruction
 875                              * pulls GL_NV_fragment_program2 in
 876                              */
 877                             sprintf(register_name, "fragment.texcoord[%s + %u]", rel_reg, reg->idx);
 878                         }
 879                     }
 880                     else
 881                     {
 882                         if(reg->idx < MAX_REG_INPUT)
 883                         {
 884                             strcpy(register_name, ctx->ps_input[reg->idx]);
 885                         }
 886                         else
 887                         {
 888                             ERR("Pixel shader input register out of bounds: %u\n", reg->idx);
 889                             sprintf(register_name, "out_of_bounds_%u", reg->idx);
 890                         }
 891                     }
 892                 }
 893             }
 894             else
 895             {
 896                 if (ctx->cur_vs_args->super.swizzle_map & (1 << reg->idx)) *is_color = TRUE;
 897                 sprintf(register_name, "vertex.attrib[%u]", reg->idx);
 898             }
 899             break;
 900
 901         case WINED3DSPR_CONST:
 902             if (!pshader && reg->rel_addr)
 903             {
 904                 BOOL aL = FALSE;
 905                 char rel_reg[50];
 906                 UINT rel_offset = ((IWineD3DVertexShaderImpl *)This)->rel_offset;
 907                 if(This->baseShader.reg_maps.shader_version.major < 2) {
 908                     sprintf(rel_reg, "A0.x");
 909                 } else {
 910                     shader_arb_get_src_param(ins, reg->rel_addr, 0, rel_reg);
 911                     if(ctx->target_version == ARB) {
 912                         if(strcmp(rel_reg, "**aL_emul**") == 0) {
 913                             aL = TRUE;
 914                         } else {
 915                             shader_arb_request_a0(ins, rel_reg);
 916                             sprintf(rel_reg, "A0.x");
 917                         }
 918                     }
 919                 }
 920                 if(aL)
 921                     sprintf(register_name, "C[%u]", ctx->aL + reg->idx);
 922                 else if (reg->idx >= rel_offset)
 923                     sprintf(register_name, "C[%s + %u]", rel_reg, reg->idx - rel_offset);
 924                 else
 925                     sprintf(register_name, "C[%s - %u]", rel_reg, -reg->idx + rel_offset);
 926             }
 927             else
 928             {
 929                 if (This->baseShader.reg_maps.usesrelconstF)
 930                     sprintf(register_name, "C[%u]", reg->idx);
 931                 else
 932                     sprintf(register_name, "C%u", reg->idx);
 933             }
 934             break;
 935
 936         case WINED3DSPR_TEXTURE: /* case WINED3DSPR_ADDR: */
 937             if (pshader) {
 938                 if(This->baseShader.reg_maps.shader_version.major == 1 &&
 939                    This->baseShader.reg_maps.shader_version.minor <= 3) {
 940                     /* In ps <= 1.3, Tx is a temporary register as destination to all instructions,
 941                      * and as source to most instructions. For some instructions it is the texcoord
 942                      * input. Those instructions know about the special use
 943                      */
 944                     sprintf(register_name, "T%u", reg->idx);
 945                 } else {
 946                     /* in ps 1.4 and 2.x Tx is always a (read-only) varying */
 947                     sprintf(register_name, "fragment.texcoord[%u]", reg->idx);
 948                 }
 949             }
 950             else
 951             {
 952                 if(This->baseShader.reg_maps.shader_version.major == 1 || ctx->target_version >= NV2)
 953                 {
 954                     sprintf(register_name, "A%u", reg->idx);
 955                 }
 956                 else
 957                 {
 958                     sprintf(register_name, "A%u_SHADOW", reg->idx);
 959                 }
 960             }
 961             break;
 962
 963         case WINED3DSPR_COLOROUT:
 964             if(ctx->cur_ps_args->super.srgb_correction && reg->idx == 0)
 965             {
 966                 strcpy(register_name, "TMP_COLOR");
 967             }
 968             else
 969             {
 970                 if(ctx->cur_ps_args->super.srgb_correction) FIXME("sRGB correction on higher render targets\n");
 971                 if(This->baseShader.reg_maps.highest_render_target > 0)
 972                 {
 973                     sprintf(register_name, "result.color[%u]", reg->idx);
 974                 }
 975                 else
 976                 {
 977                     strcpy(register_name, "result.color");
 978                 }
 979             }
 980             break;
 981
 982         case WINED3DSPR_RASTOUT:
 983             if(reg->idx == 1) sprintf(register_name, "%s", ctx->fog_output);
 984             else sprintf(register_name, "%s", rastout_reg_names[reg->idx]);
 985             break;
 986
 987         case WINED3DSPR_DEPTHOUT:
 988             strcpy(register_name, "result.depth");
 989             break;
 990
 991         case WINED3DSPR_ATTROUT:
 992         /* case WINED3DSPR_OUTPUT: */
 993             if (pshader) sprintf(register_name, "oD[%u]", reg->idx);
 994             else strcpy(register_name, ctx->color_output[reg->idx]);
 995             break;
 996
 997         case WINED3DSPR_TEXCRDOUT:
 998             if (pshader)
 999             {
1000                 sprintf(register_name, "oT[%u]", reg->idx);
1001             }
1002             else
1003             {
1004                 if(This->baseShader.reg_maps.shader_version.major < 3)
1005                 {
1006                     strcpy(register_name, ctx->texcrd_output[reg->idx]);
1007                 }
1008                 else
1009                 {
1010                     strcpy(register_name, ctx->vs_output[reg->idx]);
1011                 }
1012             }
1013             break;
1014
1015         case WINED3DSPR_LOOP:
1016             if(ctx->target_version >= NV2)
1017             {
1018                 /* Pshader has an implicitly declared loop index counter A0.x that cannot be renamed */
1019                 if(pshader) sprintf(register_name, "A0.x");
1020                 else sprintf(register_name, "aL.y");
1021             }
1022             else
1023             {
1024                 /* Unfortunately this code cannot return the value of ctx->aL here. An immediate value
1025                  * would be valid, but if aL is used for indexing(its only use), there's likely an offset,
1026                  * thus the result would be something like C[15 + 30], which is not valid in the ARB program
1027                  * grammar. So return a marker for the emulated aL and intercept it in constant and varying
1028                  * indexing
1029                  */
1030                 sprintf(register_name, "**aL_emul**");
1031             }
1032
1033             break;
1034
1035         case WINED3DSPR_CONSTINT:
1036             sprintf(register_name, "I%u", reg->idx);
1037             break;
1038
1039         case WINED3DSPR_MISCTYPE:
1040             if(reg->idx == 0)
1041             {
1042                 sprintf(register_name, "vpos");
1043             }
1044             else if(reg->idx == 1)
1045             {
1046                 sprintf(register_name, "fragment.facing.x");
1047             }
1048             else
1049             {
1050                 FIXME("Unknown MISCTYPE register index %u\n", reg->idx);
1051             }
1052             break;
1053
1054         default:
1055             FIXME("Unhandled register type %#x[%u]\n", reg->type, reg->idx);
1056             sprintf(register_name, "unrecognized_register[%u]", reg->idx);
1057             break;
1058     }
1059 }
1060
1061 static void shader_arb_get_dst_param(const struct wined3d_shader_instruction *ins,
1062         const struct wined3d_shader_dst_param *wined3d_dst, char *str)
1063 {
1064     char register_name[255];
1065     char write_mask[6];
1066     BOOL is_color;
1067
1068     shader_arb_get_register_name(ins, &wined3d_dst->reg, register_name, &is_color);
1069     strcpy(str, register_name);
1070
1071     shader_arb_get_write_mask(ins, wined3d_dst, write_mask);
1072     strcat(str, write_mask);
1073 }
1074
1075 static const char *shader_arb_get_fixup_swizzle(enum fixup_channel_source channel_source)
1076 {
1077     switch(channel_source)
1078     {
1079         case CHANNEL_SOURCE_ZERO: return "0";
1080         case CHANNEL_SOURCE_ONE: return "1";
1081         case CHANNEL_SOURCE_X: return "x";
1082         case CHANNEL_SOURCE_Y: return "y";
1083         case CHANNEL_SOURCE_Z: return "z";
1084         case CHANNEL_SOURCE_W: return "w";
1085         default:
1086             FIXME("Unhandled channel source %#x\n", channel_source);
1087             return "undefined";
1088     }
1089 }
1090
1091 static void gen_color_correction(struct wined3d_shader_buffer *buffer, const char *reg,
1092         DWORD dst_mask, const char *one, const char *two, struct color_fixup_desc fixup)
1093 {
1094     DWORD mask;
1095
1096     if (is_yuv_fixup(fixup))
1097     {
1098         enum yuv_fixup yuv_fixup = get_yuv_fixup(fixup);
1099         FIXME("YUV fixup (%#x) not supported\n", yuv_fixup);
1100         return;
1101     }
1102
1103     mask = 0;
1104     if (fixup.x_source != CHANNEL_SOURCE_X) mask |= WINED3DSP_WRITEMASK_0;
1105     if (fixup.y_source != CHANNEL_SOURCE_Y) mask |= WINED3DSP_WRITEMASK_1;
1106     if (fixup.z_source != CHANNEL_SOURCE_Z) mask |= WINED3DSP_WRITEMASK_2;
1107     if (fixup.w_source != CHANNEL_SOURCE_W) mask |= WINED3DSP_WRITEMASK_3;
1108     mask &= dst_mask;
1109
1110     if (mask)
1111     {
1112         shader_addline(buffer, "SWZ %s, %s, %s, %s, %s, %s;\n", reg, reg,
1113                 shader_arb_get_fixup_swizzle(fixup.x_source), shader_arb_get_fixup_swizzle(fixup.y_source),
1114                 shader_arb_get_fixup_swizzle(fixup.z_source), shader_arb_get_fixup_swizzle(fixup.w_source));
1115     }
1116
1117     mask = 0;
1118     if (fixup.x_sign_fixup) mask |= WINED3DSP_WRITEMASK_0;
1119     if (fixup.y_sign_fixup) mask |= WINED3DSP_WRITEMASK_1;
1120     if (fixup.z_sign_fixup) mask |= WINED3DSP_WRITEMASK_2;
1121     if (fixup.w_sign_fixup) mask |= WINED3DSP_WRITEMASK_3;
1122     mask &= dst_mask;
1123
1124     if (mask)
1125     {
1126         char reg_mask[6];
1127         char *ptr = reg_mask;
1128
1129         if (mask != WINED3DSP_WRITEMASK_ALL)
1130         {
1131             *ptr++ = '.';
1132             if (mask & WINED3DSP_WRITEMASK_0) *ptr++ = 'x';
1133             if (mask & WINED3DSP_WRITEMASK_1) *ptr++ = 'y';
1134             if (mask & WINED3DSP_WRITEMASK_2) *ptr++ = 'z';
1135             if (mask & WINED3DSP_WRITEMASK_3) *ptr++ = 'w';
1136         }
1137         *ptr = '\0';
1138
1139         shader_addline(buffer, "MAD %s%s, %s, %s, -%s;\n", reg, reg_mask, reg, two, one);
1140     }
1141 }
1142
1143 static const char *shader_arb_get_modifier(const struct wined3d_shader_instruction *ins)
1144 {
1145     DWORD mod;
1146     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
1147     if (!ins->dst_count) return "";
1148
1149     mod = ins->dst[0].modifiers;
1150
1151     /* Silently ignore PARTIALPRECISION if its not supported */
1152     if(priv->target_version == ARB) mod &= ~WINED3DSPDM_PARTIALPRECISION;
1153
1154     if(mod & WINED3DSPDM_MSAMPCENTROID)
1155     {
1156         FIXME("Unhandled modifier WINED3DSPDM_MSAMPCENTROID\n");
1157         mod &= ~WINED3DSPDM_MSAMPCENTROID;
1158     }
1159
1160     switch(mod)
1161     {
1162         case WINED3DSPDM_SATURATE | WINED3DSPDM_PARTIALPRECISION:
1163             return "H_SAT";
1164
1165         case WINED3DSPDM_SATURATE:
1166             return "_SAT";
1167
1168         case WINED3DSPDM_PARTIALPRECISION:
1169             return "H";
1170
1171         case 0:
1172             return "";
1173
1174         default:
1175             FIXME("Unknown modifiers 0x%08x\n", mod);
1176             return "";
1177     }
1178 }
1179
1180 #define TEX_PROJ        0x1
1181 #define TEX_BIAS        0x2
1182 #define TEX_LOD         0x4
1183 #define TEX_DERIV       0x10
1184
1185 static void shader_hw_sample(const struct wined3d_shader_instruction *ins, DWORD sampler_idx,
1186         const char *dst_str, const char *coord_reg, WORD flags, const char *dsx, const char *dsy)
1187 {
1188     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1189     DWORD sampler_type = ins->ctx->reg_maps->sampler_type[sampler_idx];
1190     const char *tex_type;
1191     BOOL np2_fixup = FALSE;
1192     IWineD3DBaseShaderImpl *This = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
1193     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *) This->baseShader.device;
1194     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
1195     const char *mod;
1196     BOOL pshader = shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type);
1197
1198     /* D3D vertex shader sampler IDs are vertex samplers(0-3), not global d3d samplers */
1199     if(!pshader) sampler_idx += MAX_FRAGMENT_SAMPLERS;
1200
1201     switch(sampler_type) {
1202         case WINED3DSTT_1D:
1203             tex_type = "1D";
1204             break;
1205
1206         case WINED3DSTT_2D:
1207             if(device->stateBlock->textures[sampler_idx] &&
1208                IWineD3DBaseTexture_GetTextureDimensions(device->stateBlock->textures[sampler_idx]) == GL_TEXTURE_RECTANGLE_ARB) {
1209                 tex_type = "RECT";
1210             } else {
1211                 tex_type = "2D";
1212             }
1213             if (shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type))
1214             {
1215                 if (priv->cur_np2fixup_info->super.active & (1 << sampler_idx))
1216                 {
1217                     if (flags) FIXME("Only ordinary sampling from NP2 textures is supported.\n");
1218                     else np2_fixup = TRUE;
1219                 }
1220             }
1221             break;
1222
1223         case WINED3DSTT_VOLUME:
1224             tex_type = "3D";
1225             break;
1226
1227         case WINED3DSTT_CUBE:
1228             tex_type = "CUBE";
1229             break;
1230
1231         default:
1232             ERR("Unexpected texture type %d\n", sampler_type);
1233             tex_type = "";
1234     }
1235
1236     /* TEX, TXL, TXD and TXP do not support the "H" modifier,
1237      * so don't use shader_arb_get_modifier
1238      */
1239     if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE) mod = "_SAT";
1240     else mod = "";
1241
1242     /* Fragment samplers always have indentity mapping */
1243     if(sampler_idx >= MAX_FRAGMENT_SAMPLERS)
1244     {
1245         sampler_idx = priv->cur_vs_args->vertex_samplers[sampler_idx - MAX_FRAGMENT_SAMPLERS];
1246     }
1247
1248     if (flags & TEX_DERIV)
1249     {
1250         if(flags & TEX_PROJ) FIXME("Projected texture sampling with custom derivatives\n");
1251         if(flags & TEX_BIAS) FIXME("Biased texture sampling with custom derivatives\n");
1252         shader_addline(buffer, "TXD%s %s, %s, %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg,
1253                        dsx, dsy,sampler_idx, tex_type);
1254     }
1255     else if(flags & TEX_LOD)
1256     {
1257         if(flags & TEX_PROJ) FIXME("Projected texture sampling with explicit lod\n");
1258         if(flags & TEX_BIAS) FIXME("Biased texture sampling with explicit lod\n");
1259         shader_addline(buffer, "TXL%s %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg,
1260                        sampler_idx, tex_type);
1261     }
1262     else if (flags & TEX_BIAS)
1263     {
1264         /* Shouldn't be possible, but let's check for it */
1265         if(flags & TEX_PROJ) FIXME("Biased and Projected texture sampling\n");
1266         /* TXB takes the 4th component of the source vector automatically, as d3d. Nothing more to do */
1267         shader_addline(buffer, "TXB%s %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg, sampler_idx, tex_type);
1268     }
1269     else if (flags & TEX_PROJ)
1270     {
1271         shader_addline(buffer, "TXP%s %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg, sampler_idx, tex_type);
1272     }
1273     else
1274     {
1275         if (np2_fixup)
1276         {
1277             const unsigned char idx = priv->cur_np2fixup_info->super.idx[sampler_idx];
1278             shader_addline(buffer, "MUL TA, np2fixup[%u].%s, %s;\n", idx >> 1,
1279                            (idx % 2) ? "zwxy" : "xyzw", coord_reg);
1280
1281             shader_addline(buffer, "TEX%s %s, TA, texture[%u], %s;\n", mod, dst_str, sampler_idx, tex_type);
1282         }
1283         else
1284             shader_addline(buffer, "TEX%s %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg, sampler_idx, tex_type);
1285     }
1286
1287     if (pshader)
1288     {
1289         gen_color_correction(buffer, dst_str, ins->dst[0].write_mask,
1290                 "one", "coefmul.x", priv->cur_ps_args->super.color_fixup[sampler_idx]);
1291     }
1292 }
1293
1294 static void shader_arb_get_src_param(const struct wined3d_shader_instruction *ins,
1295         const struct wined3d_shader_src_param *src, unsigned int tmpreg, char *outregstr)
1296 {
1297     /* Generate a line that does the input modifier computation and return the input register to use */
1298     BOOL is_color = FALSE;
1299     char regstr[256];
1300     char swzstr[20];
1301     int insert_line;
1302     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1303     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
1304
1305     /* Assume a new line will be added */
1306     insert_line = 1;
1307
1308     /* Get register name */
1309     shader_arb_get_register_name(ins, &src->reg, regstr, &is_color);
1310     shader_arb_get_swizzle(src, is_color, swzstr);
1311
1312     switch (src->modifiers)
1313     {
1314     case WINED3DSPSM_NONE:
1315         sprintf(outregstr, "%s%s", regstr, swzstr);
1316         insert_line = 0;
1317         break;
1318     case WINED3DSPSM_NEG:
1319         sprintf(outregstr, "-%s%s", regstr, swzstr);
1320         insert_line = 0;
1321         break;
1322     case WINED3DSPSM_BIAS:
1323         shader_addline(buffer, "ADD T%c, %s, -coefdiv.x;\n", 'A' + tmpreg, regstr);
1324         break;
1325     case WINED3DSPSM_BIASNEG:
1326         shader_addline(buffer, "ADD T%c, -%s, coefdiv.x;\n", 'A' + tmpreg, regstr);
1327         break;
1328     case WINED3DSPSM_SIGN:
1329         shader_addline(buffer, "MAD T%c, %s, coefmul.x, -one.x;\n", 'A' + tmpreg, regstr);
1330         break;
1331     case WINED3DSPSM_SIGNNEG:
1332         shader_addline(buffer, "MAD T%c, %s, -coefmul.x, one.x;\n", 'A' + tmpreg, regstr);
1333         break;
1334     case WINED3DSPSM_COMP:
1335         shader_addline(buffer, "SUB T%c, one.x, %s;\n", 'A' + tmpreg, regstr);
1336         break;
1337     case WINED3DSPSM_X2:
1338         shader_addline(buffer, "ADD T%c, %s, %s;\n", 'A' + tmpreg, regstr, regstr);
1339         break;
1340     case WINED3DSPSM_X2NEG:
1341         shader_addline(buffer, "ADD T%c, -%s, -%s;\n", 'A' + tmpreg, regstr, regstr);
1342         break;
1343     case WINED3DSPSM_DZ:
1344         shader_addline(buffer, "RCP T%c, %s.z;\n", 'A' + tmpreg, regstr);
1345         shader_addline(buffer, "MUL T%c, %s, T%c;\n", 'A' + tmpreg, regstr, 'A' + tmpreg);
1346         break;
1347     case WINED3DSPSM_DW:
1348         shader_addline(buffer, "RCP T%c, %s.w;\n", 'A' + tmpreg, regstr);
1349         shader_addline(buffer, "MUL T%c, %s, T%c;\n", 'A' + tmpreg, regstr, 'A' + tmpreg);
1350         break;
1351     case WINED3DSPSM_ABS:
1352         if(ctx->target_version >= NV2) {
1353             sprintf(outregstr, "|%s%s|", regstr, swzstr);
1354             insert_line = 0;
1355         } else {
1356             shader_addline(buffer, "ABS T%c, %s;\n", 'A' + tmpreg, regstr);
1357         }
1358         break;
1359     case WINED3DSPSM_ABSNEG:
1360         if(ctx->target_version >= NV2) {
1361             sprintf(outregstr, "-|%s%s|", regstr, swzstr);
1362         } else {
1363             shader_addline(buffer, "ABS T%c, %s;\n", 'A' + tmpreg, regstr);
1364             sprintf(outregstr, "-T%c%s", 'A' + tmpreg, swzstr);
1365         }
1366         insert_line = 0;
1367         break;
1368     default:
1369         sprintf(outregstr, "%s%s", regstr, swzstr);
1370         insert_line = 0;
1371     }
1372
1373     /* Return modified or original register, with swizzle */
1374     if (insert_line)
1375         sprintf(outregstr, "T%c%s", 'A' + tmpreg, swzstr);
1376 }
1377
1378 static void pshader_hw_bem(const struct wined3d_shader_instruction *ins)
1379 {
1380     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1381     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1382     char dst_name[50];
1383     char src_name[2][50];
1384     DWORD sampler_code = dst->reg.idx;
1385
1386     shader_arb_get_dst_param(ins, dst, dst_name);
1387
1388     /* Sampling the perturbation map in Tsrc was done already, including the signedness correction if needed
1389      *
1390      * Keep in mind that src_name[1] can be "TB" and src_name[0] can be "TA" because modifiers like _x2 are valid
1391      * with bem. So delay loading the first parameter until after the perturbation calculation which needs two
1392      * temps is done.
1393      */
1394     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1395     shader_addline(buffer, "SWZ TA, bumpenvmat%d, x, z, 0, 0;\n", sampler_code);
1396     shader_addline(buffer, "DP3 TC.r, TA, %s;\n", src_name[1]);
1397     shader_addline(buffer, "SWZ TA, bumpenvmat%d, y, w, 0, 0;\n", sampler_code);
1398     shader_addline(buffer, "DP3 TC.g, TA, %s;\n", src_name[1]);
1399
1400     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
1401     shader_addline(buffer, "ADD %s, %s, TC;\n", dst_name, src_name[0]);
1402 }
1403
1404 static DWORD negate_modifiers(DWORD mod, char *extra_char)
1405 {
1406     *extra_char = ' ';
1407     switch(mod)
1408     {
1409         case WINED3DSPSM_NONE:      return WINED3DSPSM_NEG;
1410         case WINED3DSPSM_NEG:       return WINED3DSPSM_NONE;
1411         case WINED3DSPSM_BIAS:      return WINED3DSPSM_BIASNEG;
1412         case WINED3DSPSM_BIASNEG:   return WINED3DSPSM_BIAS;
1413         case WINED3DSPSM_SIGN:      return WINED3DSPSM_SIGNNEG;
1414         case WINED3DSPSM_SIGNNEG:   return WINED3DSPSM_SIGN;
1415         case WINED3DSPSM_COMP:      *extra_char = '-'; return WINED3DSPSM_COMP;
1416         case WINED3DSPSM_X2:        return WINED3DSPSM_X2NEG;
1417         case WINED3DSPSM_X2NEG:     return WINED3DSPSM_X2;
1418         case WINED3DSPSM_DZ:        *extra_char = '-'; return WINED3DSPSM_DZ;
1419         case WINED3DSPSM_DW:        *extra_char = '-'; return WINED3DSPSM_DW;
1420         case WINED3DSPSM_ABS:       return WINED3DSPSM_ABSNEG;
1421         case WINED3DSPSM_ABSNEG:    return WINED3DSPSM_ABS;
1422     }
1423     FIXME("Unknown modifier %u\n", mod);
1424     return mod;
1425 }
1426
1427 static void pshader_hw_cnd(const struct wined3d_shader_instruction *ins)
1428 {
1429     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1430     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1431     char dst_name[50];
1432     char src_name[3][50];
1433     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
1434             ins->ctx->reg_maps->shader_version.minor);
1435     BOOL is_color;
1436
1437     shader_arb_get_dst_param(ins, dst, dst_name);
1438     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1439
1440     /* The coissue flag changes the semantic of the cnd instruction in <= 1.3 shaders */
1441     if (shader_version <= WINED3D_SHADER_VERSION(1, 3) && ins->coissue)
1442     {
1443         shader_addline(buffer, "MOV%s %s, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name[1]);
1444     } else {
1445         struct wined3d_shader_src_param src0_copy = ins->src[0];
1446         char extra_neg;
1447
1448         /* src0 may have a negate srcmod set, so we can't blindly add "-" to the name */
1449         src0_copy.modifiers = negate_modifiers(src0_copy.modifiers, &extra_neg);
1450
1451         shader_arb_get_src_param(ins, &src0_copy, 0, src_name[0]);
1452         shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
1453         shader_addline(buffer, "ADD TA, %c%s, coefdiv.x;\n", extra_neg, src_name[0]);
1454         /* No modifiers supported on CMP */
1455         shader_addline(buffer, "CMP %s, TA, %s, %s;\n", dst_name, src_name[1], src_name[2]);
1456
1457         /* _SAT on CMP doesn't make much sense, but it is not a pure NOP */
1458         if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE)
1459         {
1460             shader_arb_get_register_name(ins, &dst->reg, src_name[0], &is_color);
1461             shader_addline(buffer, "MOV_SAT %s, %s;\n", dst_name, dst_name);
1462         }
1463     }
1464 }
1465
1466 static void pshader_hw_cmp(const struct wined3d_shader_instruction *ins)
1467 {
1468     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1469     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1470     char dst_name[50];
1471     char src_name[3][50];
1472     BOOL is_color;
1473
1474     shader_arb_get_dst_param(ins, dst, dst_name);
1475
1476     /* Generate input register names (with modifiers) */
1477     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
1478     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1479     shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
1480
1481     /* No modifiers are supported on CMP */
1482     shader_addline(buffer, "CMP %s, %s, %s, %s;\n", dst_name,
1483                    src_name[0], src_name[2], src_name[1]);
1484
1485     if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE)
1486     {
1487         shader_arb_get_register_name(ins, &dst->reg, src_name[0], &is_color);
1488         shader_addline(buffer, "MOV_SAT %s, %s;\n", dst_name, src_name[0]);
1489     }
1490 }
1491
1492 /** Process the WINED3DSIO_DP2ADD instruction in ARB.
1493  * dst = dot2(src0, src1) + src2 */
1494 static void pshader_hw_dp2add(const struct wined3d_shader_instruction *ins)
1495 {
1496     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1497     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1498     char dst_name[50];
1499     char src_name[3][50];
1500     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
1501
1502     shader_arb_get_dst_param(ins, dst, dst_name);
1503     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
1504     shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
1505
1506     if(ctx->target_version >= NV3)
1507     {
1508         /* GL_NV_fragment_program2 has a 1:1 matching instruction */
1509         shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1510         shader_addline(buffer, "DP2A%s %s, %s, %s, %s;\n", shader_arb_get_modifier(ins),
1511                        dst_name, src_name[0], src_name[1], src_name[2]);
1512     }
1513     else if(ctx->target_version >= NV2)
1514     {
1515         /* dst.x = src2.?, src0.x, src1.x + src0.y * src1.y
1516          * dst.y = src2.?, src0.x, src1.z + src0.y * src1.w
1517          * dst.z = src2.?, src0.x, src1.x + src0.y * src1.y
1518          * dst.z = src2.?, src0.x, src1.z + src0.y * src1.w
1519          *
1520          * Make sure that src1.zw = src1.xy, then we get a classic dp2add
1521          *
1522          * .xyxy and other swizzles that we could get with this are not valid in
1523          * plain ARBfp, but luckily the NV extension grammar lifts this limitation.
1524          */
1525         struct wined3d_shader_src_param tmp_param = ins->src[1];
1526         DWORD swizzle = tmp_param.swizzle & 0xf; /* Selects .xy */
1527         tmp_param.swizzle = swizzle | (swizzle << 4); /* Creates .xyxy */
1528
1529         shader_arb_get_src_param(ins, &tmp_param, 1, src_name[1]);
1530
1531         shader_addline(buffer, "X2D%s %s, %s, %s, %s;\n", shader_arb_get_modifier(ins),
1532                        dst_name, src_name[2], src_name[0], src_name[1]);
1533     }
1534     else
1535     {
1536         shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1537         /* 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]
1538         * src_name[0] can be TA, but TA is a private temp for modifiers, so it is save to overwrite
1539         */
1540         shader_addline(buffer, "MOV TA, %s;\n", src_name[0]);
1541         shader_addline(buffer, "MOV TA.z, 0.0;\n");
1542         shader_addline(buffer, "DP3 TA, TA, %s;\n", src_name[1]);
1543         shader_addline(buffer, "ADD%s %s, TA, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name[2]);
1544     }
1545 }
1546
1547 /* Map the opcode 1-to-1 to the GL code */
1548 static void shader_hw_map2gl(const struct wined3d_shader_instruction *ins)
1549 {
1550     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1551     const char *instruction;
1552     char arguments[256], dst_str[50];
1553     unsigned int i;
1554     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1555
1556     switch (ins->handler_idx)
1557     {
1558         case WINED3DSIH_ABS: instruction = "ABS"; break;
1559         case WINED3DSIH_ADD: instruction = "ADD"; break;
1560         case WINED3DSIH_CRS: instruction = "XPD"; break;
1561         case WINED3DSIH_DP3: instruction = "DP3"; break;
1562         case WINED3DSIH_DP4: instruction = "DP4"; break;
1563         case WINED3DSIH_DST: instruction = "DST"; break;
1564         case WINED3DSIH_FRC: instruction = "FRC"; break;
1565         case WINED3DSIH_LIT: instruction = "LIT"; break;
1566         case WINED3DSIH_LRP: instruction = "LRP"; break;
1567         case WINED3DSIH_MAD: instruction = "MAD"; break;
1568         case WINED3DSIH_MAX: instruction = "MAX"; break;
1569         case WINED3DSIH_MIN: instruction = "MIN"; break;
1570         case WINED3DSIH_MOV: instruction = "MOV"; break;
1571         case WINED3DSIH_MUL: instruction = "MUL"; break;
1572         case WINED3DSIH_SGE: instruction = "SGE"; break;
1573         case WINED3DSIH_SLT: instruction = "SLT"; break;
1574         case WINED3DSIH_SUB: instruction = "SUB"; break;
1575         case WINED3DSIH_MOVA:instruction = "ARR"; break;
1576         case WINED3DSIH_SGN: instruction = "SSG"; break;
1577         case WINED3DSIH_DSX: instruction = "DDX"; break;
1578         default: instruction = "";
1579             FIXME("Unhandled opcode %#x\n", ins->handler_idx);
1580             break;
1581     }
1582
1583     /* Note that shader_arb_add_dst_param() adds spaces. */
1584     arguments[0] = '\0';
1585     shader_arb_get_dst_param(ins, dst, dst_str);
1586     for (i = 0; i < ins->src_count; ++i)
1587     {
1588         char operand[100];
1589         strcat(arguments, ", ");
1590         shader_arb_get_src_param(ins, &ins->src[i], i, operand);
1591         strcat(arguments, operand);
1592     }
1593     shader_addline(buffer, "%s%s %s%s;\n", instruction, shader_arb_get_modifier(ins), dst_str, arguments);
1594 }
1595
1596 static void shader_hw_nop(const struct wined3d_shader_instruction *ins)
1597 {
1598     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1599     shader_addline(buffer, "NOP;\n");
1600 }
1601
1602 static void shader_hw_mov(const struct wined3d_shader_instruction *ins)
1603 {
1604     IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
1605     BOOL pshader = shader_is_pshader_version(shader->baseShader.reg_maps.shader_version.type);
1606     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
1607
1608     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1609     char src0_param[256];
1610
1611     if(ins->handler_idx == WINED3DSIH_MOVA) {
1612         char write_mask[6];
1613
1614         if(ctx->target_version >= NV2) {
1615             shader_hw_map2gl(ins);
1616             return;
1617         }
1618         shader_arb_get_src_param(ins, &ins->src[0], 0, src0_param);
1619         shader_arb_get_write_mask(ins, &ins->dst[0], write_mask);
1620
1621         /* This implements the mova formula used in GLSL. The first two instructions
1622          * prepare the sign() part. Note that it is fine to have my_sign(0.0) = 1.0
1623          * in this case:
1624          * mova A0.x, 0.0
1625          *
1626          * A0.x = arl(floor(abs(0.0) + 0.5) * 1.0) = floor(0.5) = 0.0 since arl does a floor
1627          *
1628          * The ARL is performed when A0 is used - the requested component is read from A0_SHADOW into
1629          * A0.x. We can use the overwritten component of A0_shadow as temporary storage for the sign.
1630          */
1631         shader_addline(buffer, "SGE A0_SHADOW%s, %s, mova_const.y;\n", write_mask, src0_param);
1632         shader_addline(buffer, "MAD A0_SHADOW%s, A0_SHADOW, mova_const.z, -mova_const.w;\n", write_mask);
1633
1634         shader_addline(buffer, "ABS TA%s, %s;\n", write_mask, src0_param);
1635         shader_addline(buffer, "ADD TA%s, TA, mova_const.x;\n", write_mask);
1636         shader_addline(buffer, "FLR TA%s, TA;\n", write_mask);
1637         if (((IWineD3DVertexShaderImpl *)shader)->rel_offset)
1638         {
1639             shader_addline(buffer, "ADD TA%s, TA, helper_const.z;\n", write_mask);
1640         }
1641         shader_addline(buffer, "MUL A0_SHADOW%s, TA, A0_SHADOW;\n", write_mask);
1642
1643         ((struct shader_arb_ctx_priv *)ins->ctx->backend_data)->addr_reg[0] = '\0';
1644     } else if (ins->ctx->reg_maps->shader_version.major == 1
1645           && !shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type)
1646           && ins->dst[0].reg.type == WINED3DSPR_ADDR)
1647     {
1648         src0_param[0] = '\0';
1649         if (((IWineD3DVertexShaderImpl *)shader)->rel_offset)
1650         {
1651             shader_arb_get_src_param(ins, &ins->src[0], 0, src0_param);
1652             shader_addline(buffer, "ADD TA.x, %s, helper_const.z;\n", src0_param);
1653             shader_addline(buffer, "ARL A0.x, TA.x;\n");
1654         }
1655         else
1656         {
1657             /* Apple's ARB_vertex_program implementation does not accept an ARL source argument
1658              * with more than one component. Thus replicate the first source argument over all
1659              * 4 components. For example, .xyzw -> .x (or better: .xxxx), .zwxy -> .z, etc) */
1660             struct wined3d_shader_src_param tmp_src = ins->src[0];
1661             tmp_src.swizzle = (tmp_src.swizzle & 0x3) * 0x55;
1662             shader_arb_get_src_param(ins, &tmp_src, 0, src0_param);
1663             shader_addline(buffer, "ARL A0.x, %s;\n", src0_param);
1664         }
1665     }
1666     else if(ins->dst[0].reg.type == WINED3DSPR_COLOROUT && ins->dst[0].reg.idx == 0 && pshader)
1667     {
1668         IWineD3DPixelShaderImpl *ps = (IWineD3DPixelShaderImpl *) shader;
1669         if(ctx->cur_ps_args->super.srgb_correction && ps->color0_mov)
1670         {
1671             shader_addline(buffer, "#mov handled in srgb write code\n");
1672             return;
1673         }
1674         shader_hw_map2gl(ins);
1675     }
1676     else
1677     {
1678         shader_hw_map2gl(ins);
1679     }
1680 }
1681
1682 static void pshader_hw_texkill(const struct wined3d_shader_instruction *ins)
1683 {
1684     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1685     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1686     char reg_dest[40];
1687
1688     /* No swizzles are allowed in d3d's texkill. PS 1.x ignores the 4th component as documented,
1689      * but >= 2.0 honors it(undocumented, but tested by the d3d9 testsuit)
1690      */
1691     shader_arb_get_dst_param(ins, dst, reg_dest);
1692
1693     if (ins->ctx->reg_maps->shader_version.major >= 2)
1694     {
1695         const char *kilsrc = "TA";
1696         BOOL is_color;
1697
1698         shader_arb_get_register_name(ins, &dst->reg, reg_dest, &is_color);
1699         if(dst->write_mask == WINED3DSP_WRITEMASK_ALL)
1700         {
1701             kilsrc = reg_dest;
1702         }
1703         else
1704         {
1705             /* Sigh. KIL doesn't support swizzles/writemasks. KIL passes a writemask, but ".xy" for example
1706              * is not valid as a swizzle in ARB (needs ".xyyy"). Use SWZ to load the register properly, and set
1707              * masked out components to 0(won't kill)
1708              */
1709             char x = '0', y = '0', z = '0', w = '0';
1710             if(dst->write_mask & WINED3DSP_WRITEMASK_0) x = 'x';
1711             if(dst->write_mask & WINED3DSP_WRITEMASK_1) y = 'y';
1712             if(dst->write_mask & WINED3DSP_WRITEMASK_2) z = 'z';
1713             if(dst->write_mask & WINED3DSP_WRITEMASK_3) w = 'w';
1714             shader_addline(buffer, "SWZ TA, %s, %c, %c, %c, %c;\n", reg_dest, x, y, z, w);
1715         }
1716         shader_addline(buffer, "KIL %s;\n", kilsrc);
1717     } else {
1718         /* ARB fp doesn't like swizzles on the parameter of the KIL instruction. To mask the 4th component,
1719          * copy the register into our general purpose TMP variable, overwrite .w and pass TMP to KIL
1720          *
1721          * ps_1_3 shaders use the texcoord incarnation of the Tx register. ps_1_4 shaders can use the same,
1722          * or pass in any temporary register(in shader phase 2)
1723          */
1724         if(ins->ctx->reg_maps->shader_version.minor <= 3) {
1725             sprintf(reg_dest, "fragment.texcoord[%u]", dst->reg.idx);
1726         } else {
1727             shader_arb_get_dst_param(ins, dst, reg_dest);
1728         }
1729         shader_addline(buffer, "SWZ TA, %s, x, y, z, 1;\n", reg_dest);
1730         shader_addline(buffer, "KIL TA;\n");
1731     }
1732 }
1733
1734 static void pshader_hw_tex(const struct wined3d_shader_instruction *ins)
1735 {
1736     IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)ins->ctx->shader;
1737     IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
1738     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1739     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
1740             ins->ctx->reg_maps->shader_version.minor);
1741     struct wined3d_shader_src_param src;
1742
1743     char reg_dest[40];
1744     char reg_coord[40];
1745     DWORD reg_sampler_code;
1746     DWORD myflags = 0;
1747
1748     /* All versions have a destination register */
1749     shader_arb_get_dst_param(ins, dst, reg_dest);
1750
1751     /* 1.0-1.4: Use destination register number as texture code.
1752        2.0+: Use provided sampler number as texure code. */
1753     if (shader_version < WINED3D_SHADER_VERSION(2,0))
1754         reg_sampler_code = dst->reg.idx;
1755     else
1756         reg_sampler_code = ins->src[1].reg.idx;
1757
1758     /* 1.0-1.3: Use the texcoord varying.
1759        1.4+: Use provided coordinate source register. */
1760     if (shader_version < WINED3D_SHADER_VERSION(1,4))
1761         sprintf(reg_coord, "fragment.texcoord[%u]", reg_sampler_code);
1762     else {
1763         /* TEX is the only instruction that can handle DW and DZ natively */
1764         src = ins->src[0];
1765         if(src.modifiers == WINED3DSPSM_DW) src.modifiers = WINED3DSPSM_NONE;
1766         if(src.modifiers == WINED3DSPSM_DZ) src.modifiers = WINED3DSPSM_NONE;
1767         shader_arb_get_src_param(ins, &src, 0, reg_coord);
1768     }
1769
1770     /* projection flag:
1771      * 1.1, 1.2, 1.3: Use WINED3DTSS_TEXTURETRANSFORMFLAGS
1772      * 1.4: Use WINED3DSPSM_DZ or WINED3DSPSM_DW on src[0]
1773      * 2.0+: Use WINED3DSI_TEXLD_PROJECT on the opcode
1774      */
1775     if (shader_version < WINED3D_SHADER_VERSION(1,4))
1776     {
1777         DWORD flags = 0;
1778         if(reg_sampler_code < MAX_TEXTURES) {
1779             flags = deviceImpl->stateBlock->textureState[reg_sampler_code][WINED3DTSS_TEXTURETRANSFORMFLAGS];
1780         }
1781         if (flags & WINED3DTTFF_PROJECTED) {
1782             myflags |= TEX_PROJ;
1783         }
1784     }
1785     else if (shader_version < WINED3D_SHADER_VERSION(2,0))
1786     {
1787         DWORD src_mod = ins->src[0].modifiers;
1788         if (src_mod == WINED3DSPSM_DZ) {
1789             /* TXP cannot handle DZ natively, so move the z coordinate to .w. reg_coord is a read-only
1790              * varying register, so we need a temp reg
1791              */
1792             shader_addline(ins->ctx->buffer, "SWZ TA, %s, x, y, z, z;\n", reg_coord);
1793             strcpy(reg_coord, "TA");
1794             myflags |= TEX_PROJ;
1795         } else if(src_mod == WINED3DSPSM_DW) {
1796             myflags |= TEX_PROJ;
1797         }
1798     } else {
1799         if (ins->flags & WINED3DSI_TEXLD_PROJECT) myflags |= TEX_PROJ;
1800         if (ins->flags & WINED3DSI_TEXLD_BIAS) myflags |= TEX_BIAS;
1801     }
1802     shader_hw_sample(ins, reg_sampler_code, reg_dest, reg_coord, myflags, NULL, NULL);
1803 }
1804
1805 static void pshader_hw_texcoord(const struct wined3d_shader_instruction *ins)
1806 {
1807     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1808     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1809     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
1810             ins->ctx->reg_maps->shader_version.minor);
1811     char dst_str[50];
1812
1813     if (shader_version < WINED3D_SHADER_VERSION(1,4))
1814     {
1815         DWORD reg = dst->reg.idx;
1816
1817         shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
1818         shader_addline(buffer, "MOV_SAT %s, fragment.texcoord[%u];\n", dst_str, reg);
1819     } else {
1820         char reg_src[40];
1821
1822         shader_arb_get_src_param(ins, &ins->src[0], 0, reg_src);
1823         shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
1824         shader_addline(buffer, "MOV %s, %s;\n", dst_str, reg_src);
1825    }
1826 }
1827
1828 static void pshader_hw_texreg2ar(const struct wined3d_shader_instruction *ins)
1829 {
1830      struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1831      IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)ins->ctx->shader;
1832      IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
1833      DWORD flags;
1834
1835      DWORD reg1 = ins->dst[0].reg.idx;
1836      char dst_str[50];
1837      char src_str[50];
1838
1839      /* Note that texreg2ar treats Tx as a temporary register, not as a varying */
1840      shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
1841      shader_arb_get_src_param(ins, &ins->src[0], 0, src_str);
1842      /* Move .x first in case src_str is "TA" */
1843      shader_addline(buffer, "MOV TA.y, %s.x;\n", src_str);
1844      shader_addline(buffer, "MOV TA.x, %s.w;\n", src_str);
1845      flags = reg1 < MAX_TEXTURES ? deviceImpl->stateBlock->textureState[reg1][WINED3DTSS_TEXTURETRANSFORMFLAGS] : 0;
1846      shader_hw_sample(ins, reg1, dst_str, "TA", flags & WINED3DTTFF_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
1847 }
1848
1849 static void pshader_hw_texreg2gb(const struct wined3d_shader_instruction *ins)
1850 {
1851      struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1852
1853      DWORD reg1 = ins->dst[0].reg.idx;
1854      char dst_str[50];
1855      char src_str[50];
1856
1857      /* Note that texreg2gb treats Tx as a temporary register, not as a varying */
1858      shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
1859      shader_arb_get_src_param(ins, &ins->src[0], 0, src_str);
1860      shader_addline(buffer, "MOV TA.x, %s.y;\n", src_str);
1861      shader_addline(buffer, "MOV TA.y, %s.z;\n", src_str);
1862      shader_hw_sample(ins, reg1, dst_str, "TA", 0, NULL, NULL);
1863 }
1864
1865 static void pshader_hw_texreg2rgb(const struct wined3d_shader_instruction *ins)
1866 {
1867     DWORD reg1 = ins->dst[0].reg.idx;
1868     char dst_str[50];
1869     char src_str[50];
1870
1871     /* Note that texreg2rg treats Tx as a temporary register, not as a varying */
1872     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
1873     shader_arb_get_src_param(ins, &ins->src[0], 0, src_str);
1874     shader_hw_sample(ins, reg1, dst_str, src_str, 0, NULL, NULL);
1875 }
1876
1877 static void pshader_hw_texbem(const struct wined3d_shader_instruction *ins)
1878 {
1879     IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)ins->ctx->shader;
1880     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1881     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1882     char reg_coord[40], dst_reg[50], src_reg[50];
1883     DWORD reg_dest_code;
1884
1885     /* All versions have a destination register. The Tx where the texture coordinates come
1886      * from is the varying incarnation of the texture register
1887      */
1888     reg_dest_code = dst->reg.idx;
1889     shader_arb_get_dst_param(ins, &ins->dst[0], dst_reg);
1890     shader_arb_get_src_param(ins, &ins->src[0], 0, src_reg);
1891     sprintf(reg_coord, "fragment.texcoord[%u]", reg_dest_code);
1892
1893     /* Sampling the perturbation map in Tsrc was done already, including the signedness correction if needed
1894      * The Tx in which the perturbation map is stored is the tempreg incarnation of the texture register
1895      *
1896      * GL_NV_fragment_program_option could handle this in one instruction via X2D:
1897      * X2D TA.xy, fragment.texcoord, T%u, bumpenvmat%u.xzyw
1898      *
1899      * However, the NV extensions are never enabled for <= 2.0 shaders because of the performance penalty that
1900      * comes with it, and texbem is an 1.x only instruction. No 1.x instruction forces us to enable the NV
1901      * extension.
1902      */
1903     shader_addline(buffer, "SWZ TB, bumpenvmat%d, x, z, 0, 0;\n", reg_dest_code);
1904     shader_addline(buffer, "DP3 TA.x, TB, %s;\n", src_reg);
1905     shader_addline(buffer, "SWZ TB, bumpenvmat%d, y, w, 0, 0;\n", reg_dest_code);
1906     shader_addline(buffer, "DP3 TA.y, TB, %s;\n", src_reg);
1907
1908     /* with projective textures, texbem only divides the static texture coord, not the displacement,
1909      * so we can't let the GL handle this.
1910      */
1911     if (((IWineD3DDeviceImpl*) This->baseShader.device)->stateBlock->textureState[reg_dest_code][WINED3DTSS_TEXTURETRANSFORMFLAGS]
1912             & WINED3DTTFF_PROJECTED) {
1913         shader_addline(buffer, "RCP TB.w, %s.w;\n", reg_coord);
1914         shader_addline(buffer, "MUL TB.xy, %s, TB.w;\n", reg_coord);
1915         shader_addline(buffer, "ADD TA.xy, TA, TB;\n");
1916     } else {
1917         shader_addline(buffer, "ADD TA.xy, TA, %s;\n", reg_coord);
1918     }
1919
1920     shader_hw_sample(ins, reg_dest_code, dst_reg, "TA", 0, NULL, NULL);
1921
1922     if (ins->handler_idx == WINED3DSIH_TEXBEML)
1923     {
1924         /* No src swizzles are allowed, so this is ok */
1925         shader_addline(buffer, "MAD TA, %s.z, luminance%d.x, luminance%d.y;\n",
1926                        src_reg, reg_dest_code, reg_dest_code);
1927         shader_addline(buffer, "MUL %s, %s, TA;\n", dst_reg, dst_reg);
1928     }
1929 }
1930
1931 static void pshader_hw_texm3x2pad(const struct wined3d_shader_instruction *ins)
1932 {
1933     DWORD reg = ins->dst[0].reg.idx;
1934     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1935     char src0_name[50], dst_name[50];
1936     BOOL is_color;
1937     struct wined3d_shader_register tmp_reg = ins->dst[0].reg;
1938
1939     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
1940     /* The next instruction will be a texm3x2tex or texm3x2depth that writes to the uninitialized
1941      * T<reg+1> register. Use this register to store the calculated vector
1942      */
1943     tmp_reg.idx = reg + 1;
1944     shader_arb_get_register_name(ins, &tmp_reg, dst_name, &is_color);
1945     shader_addline(buffer, "DP3 %s.x, fragment.texcoord[%u], %s;\n", dst_name, reg, src0_name);
1946 }
1947
1948 static void pshader_hw_texm3x2tex(const struct wined3d_shader_instruction *ins)
1949 {
1950     IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)ins->ctx->shader;
1951     IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
1952     DWORD flags;
1953     DWORD reg = ins->dst[0].reg.idx;
1954     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1955     char dst_str[50];
1956     char src0_name[50];
1957     char dst_reg[50];
1958     BOOL is_color;
1959
1960     /* We know that we're writing to the uninitialized T<reg> register, so use it for temporary storage */
1961     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_reg, &is_color);
1962
1963     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
1964     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
1965     shader_addline(buffer, "DP3 %s.y, fragment.texcoord[%u], %s;\n", dst_reg, reg, src0_name);
1966     flags = reg < MAX_TEXTURES ? deviceImpl->stateBlock->textureState[reg][WINED3DTSS_TEXTURETRANSFORMFLAGS] : 0;
1967     shader_hw_sample(ins, reg, dst_str, dst_reg, flags & WINED3DTTFF_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
1968 }
1969
1970 static void pshader_hw_texm3x3pad(const struct wined3d_shader_instruction *ins)
1971 {
1972     IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)ins->ctx->shader;
1973     DWORD reg = ins->dst[0].reg.idx;
1974     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1975     SHADER_PARSE_STATE* current_state = &This->baseShader.parse_state;
1976     char src0_name[50], dst_name[50];
1977     struct wined3d_shader_register tmp_reg = ins->dst[0].reg;
1978     BOOL is_color;
1979
1980     /* There are always 2 texm3x3pad instructions followed by one texm3x3[tex,vspec, ...] instruction, with
1981      * incrementing ins->dst[0].register_idx numbers. So the pad instruction already knows the final destination
1982      * register, and this register is uninitialized(otherwise the assembler complains that it is 'redeclared')
1983      */
1984     tmp_reg.idx = reg + 2 - current_state->current_row;
1985     shader_arb_get_register_name(ins, &tmp_reg, dst_name, &is_color);
1986
1987     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
1988     shader_addline(buffer, "DP3 %s.%c, fragment.texcoord[%u], %s;\n",
1989                    dst_name, 'x' + current_state->current_row, reg, src0_name);
1990     current_state->texcoord_w[current_state->current_row++] = reg;
1991 }
1992
1993 static void pshader_hw_texm3x3tex(const struct wined3d_shader_instruction *ins)
1994 {
1995     IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)ins->ctx->shader;
1996     IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
1997     DWORD flags;
1998     DWORD reg = ins->dst[0].reg.idx;
1999     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2000     SHADER_PARSE_STATE* current_state = &This->baseShader.parse_state;
2001     char dst_str[50];
2002     char src0_name[50], dst_name[50];
2003     BOOL is_color;
2004
2005     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2006     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2007     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_name, reg, src0_name);
2008
2009     /* Sample the texture using the calculated coordinates */
2010     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2011     flags = reg < MAX_TEXTURES ? deviceImpl->stateBlock->textureState[reg][WINED3DTSS_TEXTURETRANSFORMFLAGS] : 0;
2012     shader_hw_sample(ins, reg, dst_str, dst_name, flags & WINED3DTTFF_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2013     current_state->current_row = 0;
2014 }
2015
2016 static void pshader_hw_texm3x3vspec(const struct wined3d_shader_instruction *ins)
2017 {
2018     IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)ins->ctx->shader;
2019     IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
2020     DWORD flags;
2021     DWORD reg = ins->dst[0].reg.idx;
2022     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2023     SHADER_PARSE_STATE* current_state = &This->baseShader.parse_state;
2024     char dst_str[50];
2025     char src0_name[50];
2026     char dst_reg[50];
2027     BOOL is_color;
2028
2029     /* Get the dst reg without writemask strings. We know this register is uninitialized, so we can use all
2030      * components for temporary data storage
2031      */
2032     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_reg, &is_color);
2033     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2034     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_reg, reg, src0_name);
2035
2036     /* Construct the eye-ray vector from w coordinates */
2037     shader_addline(buffer, "MOV TB.x, fragment.texcoord[%u].w;\n", current_state->texcoord_w[0]);
2038     shader_addline(buffer, "MOV TB.y, fragment.texcoord[%u].w;\n", current_state->texcoord_w[1]);
2039     shader_addline(buffer, "MOV TB.z, fragment.texcoord[%u].w;\n", reg);
2040
2041     /* Calculate reflection vector
2042      */
2043     shader_addline(buffer, "DP3 %s.w, %s, TB;\n", dst_reg, dst_reg);
2044     /* The .w is ignored when sampling, so I can use TB.w to calculate dot(N, N) */
2045     shader_addline(buffer, "DP3 TB.w, %s, %s;\n", dst_reg, dst_reg);
2046     shader_addline(buffer, "RCP TB.w, TB.w;\n");
2047     shader_addline(buffer, "MUL %s.w, %s.w, TB.w;\n", dst_reg, dst_reg);
2048     shader_addline(buffer, "MUL %s, %s.w, %s;\n", dst_reg, dst_reg, dst_reg);
2049     shader_addline(buffer, "MAD %s, coefmul.x, %s, -TB;\n", dst_reg, dst_reg);
2050
2051     /* Sample the texture using the calculated coordinates */
2052     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2053     flags = reg < MAX_TEXTURES ? deviceImpl->stateBlock->textureState[reg][WINED3DTSS_TEXTURETRANSFORMFLAGS] : 0;
2054     shader_hw_sample(ins, reg, dst_str, dst_reg, flags & WINED3DTTFF_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2055     current_state->current_row = 0;
2056 }
2057
2058 static void pshader_hw_texm3x3spec(const struct wined3d_shader_instruction *ins)
2059 {
2060     IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)ins->ctx->shader;
2061     IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
2062     DWORD flags;
2063     DWORD reg = ins->dst[0].reg.idx;
2064     SHADER_PARSE_STATE* current_state = &This->baseShader.parse_state;
2065     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2066     char dst_str[50];
2067     char src0_name[50];
2068     char src1_name[50];
2069     char dst_reg[50];
2070     BOOL is_color;
2071
2072     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2073     shader_arb_get_src_param(ins, &ins->src[0], 1, src1_name);
2074     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_reg, &is_color);
2075     /* Note: dst_reg.xy is input here, generated by two texm3x3pad instructions */
2076     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_reg, reg, src0_name);
2077
2078     /* Calculate reflection vector.
2079      *
2080      *                   dot(N, E)
2081      * dst_reg.xyz = 2 * --------- * N - E
2082      *                   dot(N, N)
2083      *
2084      * Which normalizes the normal vector
2085      */
2086     shader_addline(buffer, "DP3 %s.w, %s, %s;\n", dst_reg, dst_reg, src1_name);
2087     shader_addline(buffer, "DP3 TC.w, %s, %s;\n", dst_reg, dst_reg);
2088     shader_addline(buffer, "RCP TC.w, TC.w;\n");
2089     shader_addline(buffer, "MUL %s.w, %s.w, TC.w;\n", dst_reg, dst_reg);
2090     shader_addline(buffer, "MUL %s, %s.w, %s;\n", dst_reg, dst_reg, dst_reg);
2091     shader_addline(buffer, "MAD %s, coefmul.x, %s, -%s;\n", dst_reg, dst_reg, src1_name);
2092
2093     /* Sample the texture using the calculated coordinates */
2094     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2095     flags = reg < MAX_TEXTURES ? deviceImpl->stateBlock->textureState[reg][WINED3DTSS_TEXTURETRANSFORMFLAGS] : 0;
2096     shader_hw_sample(ins, reg, dst_str, dst_reg, flags & WINED3DTTFF_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2097     current_state->current_row = 0;
2098 }
2099
2100 static void pshader_hw_texdepth(const struct wined3d_shader_instruction *ins)
2101 {
2102     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2103     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2104     char dst_name[50];
2105
2106     /* texdepth has an implicit destination, the fragment depth value. It's only parameter,
2107      * which is essentially an input, is the destination register because it is the first
2108      * parameter. According to the msdn, this must be register r5, but let's keep it more flexible
2109      * here(writemasks/swizzles are not valid on texdepth)
2110      */
2111     shader_arb_get_dst_param(ins, dst, dst_name);
2112
2113     /* According to the msdn, the source register(must be r5) is unusable after
2114      * the texdepth instruction, so we're free to modify it
2115      */
2116     shader_addline(buffer, "MIN %s.y, %s.y, one.y;\n", dst_name, dst_name);
2117
2118     /* How to deal with the special case dst_name.g == 0? if r != 0, then
2119      * the r * (1 / 0) will give infinity, which is clamped to 1.0, the correct
2120      * result. But if r = 0.0, then 0 * inf = 0, which is incorrect.
2121      */
2122     shader_addline(buffer, "RCP %s.y, %s.y;\n", dst_name, dst_name);
2123     shader_addline(buffer, "MUL TA.x, %s.x, %s.y;\n", dst_name, dst_name);
2124     shader_addline(buffer, "MIN TA.x, TA.x, one.x;\n");
2125     shader_addline(buffer, "MAX result.depth, TA.x, 0.0;\n");
2126 }
2127
2128 /** Process the WINED3DSIO_TEXDP3TEX instruction in ARB:
2129  * Take a 3-component dot product of the TexCoord[dstreg] and src,
2130  * then perform a 1D texture lookup from stage dstregnum, place into dst. */
2131 static void pshader_hw_texdp3tex(const struct wined3d_shader_instruction *ins)
2132 {
2133     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2134     DWORD sampler_idx = ins->dst[0].reg.idx;
2135     char src0[50];
2136     char dst_str[50];
2137
2138     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2139     shader_addline(buffer, "MOV TB, 0.0;\n");
2140     shader_addline(buffer, "DP3 TB.x, fragment.texcoord[%u], %s;\n", sampler_idx, src0);
2141
2142     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2143     shader_hw_sample(ins, sampler_idx, dst_str, "TB", 0 /* Only one coord, can't be projected */, NULL, NULL);
2144 }
2145
2146 /** Process the WINED3DSIO_TEXDP3 instruction in ARB:
2147  * Take a 3-component dot product of the TexCoord[dstreg] and src. */
2148 static void pshader_hw_texdp3(const struct wined3d_shader_instruction *ins)
2149 {
2150     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2151     char src0[50];
2152     char dst_str[50];
2153     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2154
2155     /* Handle output register */
2156     shader_arb_get_dst_param(ins, dst, dst_str);
2157     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2158     shader_addline(buffer, "DP3 %s, fragment.texcoord[%u], %s;\n", dst_str, dst->reg.idx, src0);
2159 }
2160
2161 /** Process the WINED3DSIO_TEXM3X3 instruction in ARB
2162  * Perform the 3rd row of a 3x3 matrix multiply */
2163 static void pshader_hw_texm3x3(const struct wined3d_shader_instruction *ins)
2164 {
2165     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2166     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2167     char dst_str[50], dst_name[50];
2168     char src0[50];
2169     BOOL is_color;
2170
2171     shader_arb_get_dst_param(ins, dst, dst_str);
2172     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2173     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2174     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_name, dst->reg.idx, src0);
2175     shader_addline(buffer, "MOV %s, %s;\n", dst_str, dst_name);
2176 }
2177
2178 /** Process the WINED3DSIO_TEXM3X2DEPTH instruction in ARB:
2179  * Last row of a 3x2 matrix multiply, use the result to calculate the depth:
2180  * Calculate tmp0.y = TexCoord[dstreg] . src.xyz;  (tmp0.x has already been calculated)
2181  * depth = (tmp0.y == 0.0) ? 1.0 : tmp0.x / tmp0.y
2182  */
2183 static void pshader_hw_texm3x2depth(const struct wined3d_shader_instruction *ins)
2184 {
2185     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2186     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2187     char src0[50], dst_name[50];
2188     BOOL is_color;
2189
2190     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2191     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2192     shader_addline(buffer, "DP3 %s.y, fragment.texcoord[%u], %s;\n", dst_name, dst->reg.idx, src0);
2193
2194     /* How to deal with the special case dst_name.g == 0? if r != 0, then
2195      * the r * (1 / 0) will give infinity, which is clamped to 1.0, the correct
2196      * result. But if r = 0.0, then 0 * inf = 0, which is incorrect.
2197      */
2198     shader_addline(buffer, "RCP %s.y, %s.y;\n", dst_name, dst_name);
2199     shader_addline(buffer, "MUL %s.x, %s.x, %s.y;\n", dst_name, dst_name, dst_name);
2200     shader_addline(buffer, "MIN %s.x, %s.x, one.x;\n", dst_name, dst_name);
2201     shader_addline(buffer, "MAX result.depth, %s.x, 0.0;\n", dst_name);
2202 }
2203
2204 /** Handles transforming all WINED3DSIO_M?x? opcodes for
2205     Vertex/Pixel shaders to ARB_vertex_program codes */
2206 static void shader_hw_mnxn(const struct wined3d_shader_instruction *ins)
2207 {
2208     int i;
2209     int nComponents = 0;
2210     struct wined3d_shader_dst_param tmp_dst = {{0}};
2211     struct wined3d_shader_src_param tmp_src[2] = {{{0}}};
2212     struct wined3d_shader_instruction tmp_ins;
2213
2214     memset(&tmp_ins, 0, sizeof(tmp_ins));
2215
2216     /* Set constants for the temporary argument */
2217     tmp_ins.ctx = ins->ctx;
2218     tmp_ins.dst_count = 1;
2219     tmp_ins.dst = &tmp_dst;
2220     tmp_ins.src_count = 2;
2221     tmp_ins.src = tmp_src;
2222
2223     switch(ins->handler_idx)
2224     {
2225         case WINED3DSIH_M4x4:
2226             nComponents = 4;
2227             tmp_ins.handler_idx = WINED3DSIH_DP4;
2228             break;
2229         case WINED3DSIH_M4x3:
2230             nComponents = 3;
2231             tmp_ins.handler_idx = WINED3DSIH_DP4;
2232             break;
2233         case WINED3DSIH_M3x4:
2234             nComponents = 4;
2235             tmp_ins.handler_idx = WINED3DSIH_DP3;
2236             break;
2237         case WINED3DSIH_M3x3:
2238             nComponents = 3;
2239             tmp_ins.handler_idx = WINED3DSIH_DP3;
2240             break;
2241         case WINED3DSIH_M3x2:
2242             nComponents = 2;
2243             tmp_ins.handler_idx = WINED3DSIH_DP3;
2244             break;
2245         default:
2246             FIXME("Unhandled opcode %#x\n", ins->handler_idx);
2247             break;
2248     }
2249
2250     tmp_dst = ins->dst[0];
2251     tmp_src[0] = ins->src[0];
2252     tmp_src[1] = ins->src[1];
2253     for (i = 0; i < nComponents; i++) {
2254         tmp_dst.write_mask = WINED3DSP_WRITEMASK_0 << i;
2255         shader_hw_map2gl(&tmp_ins);
2256         ++tmp_src[1].reg.idx;
2257     }
2258 }
2259
2260 static void shader_hw_scalar_op(const struct wined3d_shader_instruction *ins)
2261 {
2262     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2263     const char *instruction;
2264
2265     char dst[50];
2266     char src[50];
2267
2268     switch(ins->handler_idx)
2269     {
2270         case WINED3DSIH_RSQ:  instruction = "RSQ"; break;
2271         case WINED3DSIH_RCP:  instruction = "RCP"; break;
2272         case WINED3DSIH_EXP:  instruction = "EX2"; break;
2273         case WINED3DSIH_EXPP: instruction = "EXP"; break;
2274         default: instruction = "";
2275             FIXME("Unhandled opcode %#x\n", ins->handler_idx);
2276             break;
2277     }
2278
2279     shader_arb_get_dst_param(ins, &ins->dst[0], dst); /* Destination */
2280     shader_arb_get_src_param(ins, &ins->src[0], 0, src);
2281     if (ins->src[0].swizzle == WINED3DSP_NOSWIZZLE)
2282     {
2283         /* Dx sdk says .x is used if no swizzle is given, but our test shows that
2284          * .w is used
2285          */
2286         strcat(src, ".w");
2287     }
2288
2289     shader_addline(buffer, "%s%s %s, %s;\n", instruction, shader_arb_get_modifier(ins), dst, src);
2290 }
2291
2292 static void shader_hw_nrm(const struct wined3d_shader_instruction *ins)
2293 {
2294     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2295     char dst_name[50];
2296     char src_name[50];
2297     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2298     BOOL pshader = shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type);
2299
2300     shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2301     shader_arb_get_src_param(ins, &ins->src[0], 1 /* Use TB */, src_name);
2302
2303     if(pshader && priv->target_version >= NV3)
2304     {
2305         shader_addline(buffer, "NRM%s %s, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name);
2306     }
2307     else
2308     {
2309         shader_addline(buffer, "DP3 TA, %s, %s;\n", src_name, src_name);
2310         shader_addline(buffer, "RSQ TA, TA.x;\n");
2311         /* dst.w = src[0].w * 1 / (src.x^2 + src.y^2 + src.z^2)^(1/2) according to msdn*/
2312         shader_addline(buffer, "MUL%s %s, %s, TA;\n", shader_arb_get_modifier(ins), dst_name,
2313                     src_name);
2314     }
2315 }
2316
2317 static void shader_hw_lrp(const struct wined3d_shader_instruction *ins)
2318 {
2319     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2320     char dst_name[50];
2321     char src_name[3][50];
2322
2323     /* ARB_fragment_program has a convenient LRP instruction */
2324     if(shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type)) {
2325         shader_hw_map2gl(ins);
2326         return;
2327     }
2328
2329     shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2330     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
2331     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
2332     shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
2333
2334     shader_addline(buffer, "SUB TA, %s, %s;\n", src_name[1], src_name[2]);
2335     shader_addline(buffer, "MAD%s %s, %s, TA, %s;\n", shader_arb_get_modifier(ins),
2336                    dst_name, src_name[0], src_name[2]);
2337 }
2338
2339 static void shader_hw_sincos(const struct wined3d_shader_instruction *ins)
2340 {
2341     /* This instruction exists in ARB, but the d3d instruction takes two extra parameters which
2342      * must contain fixed constants. So we need a separate function to filter those constants and
2343      * can't use map2gl
2344      */
2345     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2346     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2347     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2348     char dst_name[50];
2349     char src_name0[50], src_name1[50], src_name2[50];
2350     BOOL is_color;
2351
2352     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name0);
2353     if(shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type)) {
2354         shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2355         /* No modifiers are supported on SCS */
2356         shader_addline(buffer, "SCS %s, %s;\n", dst_name, src_name0);
2357
2358         if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE)
2359         {
2360             shader_arb_get_register_name(ins, &dst->reg, src_name0, &is_color);
2361             shader_addline(buffer, "MOV_SAT %s, %s;\n", dst_name, src_name0);
2362         }
2363     } else if(priv->target_version >= NV2) {
2364         shader_arb_get_register_name(ins, &dst->reg, dst_name, &is_color);
2365
2366         /* Sincos writemask must be .x, .y or .xy */
2367         if(dst->write_mask & WINED3DSP_WRITEMASK_0)
2368             shader_addline(buffer, "COS%s %s.x, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name0);
2369         if(dst->write_mask & WINED3DSP_WRITEMASK_1)
2370             shader_addline(buffer, "SIN%s %s.y, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name0);
2371     } else {
2372         /* Approximate sine and cosine with a taylor series, as per math textbook. The application passes 8
2373          * helper constants(D3DSINCOSCONST1 and D3DSINCOSCONST2) in src1 and src2.
2374          *
2375          * sin(x) = x - x^3/3! + x^5/5! - x^7/7! + ...
2376          * cos(x) = 1 - x^2/2! + x^4/4! - x^6/6! + ...
2377          *
2378          * The constants we get are:
2379          *
2380          *  +1   +1,     -1     -1     +1      +1      -1       -1
2381          *      ---- ,  ---- , ---- , ----- , ----- , ----- , ------
2382          *      1!*2    2!*4   3!*8   4!*16   5!*32   6!*64   7!*128
2383          *
2384          * If used with x^2, x^3, x^4 etc they calculate sin(x/2) and cos(x/2):
2385          *
2386          * (x/2)^2 = x^2 / 4
2387          * (x/2)^3 = x^3 / 8
2388          * (x/2)^4 = x^4 / 16
2389          * (x/2)^5 = x^5 / 32
2390          * etc
2391          *
2392          * To get the final result:
2393          * sin(x) = 2 * sin(x/2) * cos(x/2)
2394          * cos(x) = cos(x/2)^2 - sin(x/2)^2
2395          * (from sin(x+y) and cos(x+y) rules)
2396          *
2397          * As per MSDN, dst.z is undefined after the operation, and so is
2398          * dst.x and dst.y if they're masked out by the writemask. Ie
2399          * sincos dst.y, src1, c0, c1
2400          * returns the sine in dst.y. dst.x and dst.z are undefined, dst.w is not touched. The assembler
2401          * vsa.exe also stops with an error if the dest register is the same register as the source
2402          * register. This means we can use dest.xyz as temporary storage. The assembler vsa.exe output also
2403          * indicates that sincos consumes 8 instruction slots in vs_2_0(and, strangely, in vs_3_0).
2404          */
2405         shader_arb_get_src_param(ins, &ins->src[1], 1, src_name1);
2406         shader_arb_get_src_param(ins, &ins->src[2], 2, src_name2);
2407         shader_arb_get_register_name(ins, &dst->reg, dst_name, &is_color);
2408
2409         shader_addline(buffer, "MUL %s.x, %s, %s;\n", dst_name, src_name0, src_name0);  /* x ^ 2 */
2410         shader_addline(buffer, "MUL TA.y, %s.x, %s;\n", dst_name, src_name0);           /* x ^ 3 */
2411         shader_addline(buffer, "MUL %s.y, TA.y, %s;\n", dst_name, src_name0);           /* x ^ 4 */
2412         shader_addline(buffer, "MUL TA.z, %s.y, %s;\n", dst_name, src_name0);           /* x ^ 5 */
2413         shader_addline(buffer, "MUL %s.z, TA.z, %s;\n", dst_name, src_name0);           /* x ^ 6 */
2414         shader_addline(buffer, "MUL TA.w, %s.z, %s;\n", dst_name, src_name0);           /* x ^ 7 */
2415
2416         /* sin(x/2)
2417          *
2418          * Unfortunately we don't get the constants in a DP4-capable form. Is there a way to
2419          * properly merge that with MULs in the code above?
2420          * The swizzles .yz and xw however fit into the .yzxw swizzle added to ps_2_0. Maybe
2421          * we can merge the sine and cosine MAD rows to calculate them together.
2422          */
2423         shader_addline(buffer, "MUL TA.x, %s, %s.w;\n", src_name0, src_name2); /* x^1, +1/(1!*2) */
2424         shader_addline(buffer, "MAD TA.x, TA.y, %s.x, TA.x;\n", src_name2); /* -1/(3!*8) */
2425         shader_addline(buffer, "MAD TA.x, TA.z, %s.w, TA.x;\n", src_name1); /* +1/(5!*32) */
2426         shader_addline(buffer, "MAD TA.x, TA.w, %s.x, TA.x;\n", src_name1); /* -1/(7!*128) */
2427
2428         /* cos(x/2) */
2429         shader_addline(buffer, "MAD TA.y, %s.x, %s.y, %s.z;\n", dst_name, src_name2, src_name2); /* -1/(2!*4), +1.0 */
2430         shader_addline(buffer, "MAD TA.y, %s.y, %s.z, TA.y;\n", dst_name, src_name1); /* +1/(4!*16) */
2431         shader_addline(buffer, "MAD TA.y, %s.z, %s.y, TA.y;\n", dst_name, src_name1); /* -1/(6!*64) */
2432
2433         if(dst->write_mask & WINED3DSP_WRITEMASK_0) {
2434             /* cos x */
2435             shader_addline(buffer, "MUL TA.z, TA.y, TA.y;\n");
2436             shader_addline(buffer, "MAD %s.x, -TA.x, TA.x, TA.z;\n", dst_name);
2437         }
2438         if(dst->write_mask & WINED3DSP_WRITEMASK_1) {
2439             /* sin x */
2440             shader_addline(buffer, "MUL %s.y, TA.x, TA.y;\n", dst_name);
2441             shader_addline(buffer, "ADD %s.y, %s.y, %s.y;\n", dst_name, dst_name, dst_name);
2442         }
2443     }
2444 }
2445
2446 static void shader_hw_sgn(const struct wined3d_shader_instruction *ins)
2447 {
2448     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2449     char dst_name[50];
2450     char src_name[50];
2451     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
2452
2453     /* SGN is only valid in vertex shaders */
2454     if(ctx->target_version == NV2) {
2455         shader_hw_map2gl(ins);
2456         return;
2457     }
2458     shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2459     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name);
2460
2461     /* If SRC > 0.0, -SRC < SRC = TRUE, otherwise false.
2462      * if SRC < 0.0,  SRC < -SRC = TRUE. If neither is true, src = 0.0
2463      */
2464     if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE) {
2465         shader_addline(buffer, "SLT %s, -%s, %s;\n", dst_name, src_name, src_name);
2466     } else {
2467         /* src contains TA? Write to the dest first. This won't overwrite our destination.
2468          * Then use TA, and calculate the final result
2469          *
2470          * Not reading from TA? Store the first result in TA to avoid overwriting the
2471          * destination if src reg = dst reg
2472          */
2473         if(strstr(src_name, "TA"))
2474         {
2475             shader_addline(buffer, "SLT %s,  %s, -%s;\n", dst_name, src_name, src_name);
2476             shader_addline(buffer, "SLT TA, -%s, %s;\n", src_name, src_name);
2477             shader_addline(buffer, "ADD %s, %s, -TA;\n", dst_name, dst_name);
2478         }
2479         else
2480         {
2481             shader_addline(buffer, "SLT TA, -%s, %s;\n", src_name, src_name);
2482             shader_addline(buffer, "SLT %s,  %s, -%s;\n", dst_name, src_name, src_name);
2483             shader_addline(buffer, "ADD %s, TA, -%s;\n", dst_name, dst_name);
2484         }
2485     }
2486 }
2487
2488 static void shader_hw_dsy(const struct wined3d_shader_instruction *ins)
2489 {
2490     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2491     char src[50];
2492     char dst[50];
2493     char dst_name[50];
2494     BOOL is_color;
2495
2496     shader_arb_get_dst_param(ins, &ins->dst[0], dst);
2497     shader_arb_get_src_param(ins, &ins->src[0], 0, src);
2498     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2499
2500     shader_addline(buffer, "DDY %s, %s;\n", dst, src);
2501     shader_addline(buffer, "MUL%s %s, %s, ycorrection.y;\n", shader_arb_get_modifier(ins), dst, dst_name);
2502 }
2503
2504 static DWORD abs_modifier(DWORD mod, BOOL *need_abs)
2505 {
2506     *need_abs = FALSE;
2507
2508     switch(mod)
2509     {
2510         case WINED3DSPSM_NONE:      return WINED3DSPSM_ABS;
2511         case WINED3DSPSM_NEG:       return WINED3DSPSM_ABS;
2512         case WINED3DSPSM_BIAS:      *need_abs = TRUE; return WINED3DSPSM_BIAS;
2513         case WINED3DSPSM_BIASNEG:   *need_abs = TRUE; return WINED3DSPSM_BIASNEG;
2514         case WINED3DSPSM_SIGN:      *need_abs = TRUE; return WINED3DSPSM_SIGN;
2515         case WINED3DSPSM_SIGNNEG:   *need_abs = TRUE; return WINED3DSPSM_SIGNNEG;
2516         case WINED3DSPSM_COMP:      *need_abs = TRUE; return WINED3DSPSM_COMP;
2517         case WINED3DSPSM_X2:        *need_abs = TRUE; return WINED3DSPSM_X2;
2518         case WINED3DSPSM_X2NEG:     *need_abs = TRUE; return WINED3DSPSM_X2NEG;
2519         case WINED3DSPSM_DZ:        *need_abs = TRUE; return WINED3DSPSM_DZ;
2520         case WINED3DSPSM_DW:        *need_abs = TRUE; return WINED3DSPSM_DW;
2521         case WINED3DSPSM_ABS:       return WINED3DSPSM_ABS;
2522         case WINED3DSPSM_ABSNEG:    return WINED3DSPSM_ABS;
2523     }
2524     FIXME("Unknown modifier %u\n", mod);
2525     return mod;
2526 }
2527
2528 static void shader_hw_log_pow(const struct wined3d_shader_instruction *ins)
2529 {
2530     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2531     char src0[50], src1[50], dst[50];
2532     struct wined3d_shader_src_param src0_copy = ins->src[0];
2533     BOOL need_abs = FALSE;
2534     const char *instr;
2535     BOOL arg2 = FALSE;
2536
2537     switch(ins->handler_idx)
2538     {
2539         case WINED3DSIH_LOG:  instr = "LG2"; break;
2540         case WINED3DSIH_LOGP: instr = "LOG"; break;
2541         case WINED3DSIH_POW:  instr = "POW"; arg2 = TRUE; break;
2542         default:
2543             ERR("Unexpected instruction %d\n", ins->handler_idx);
2544             return;
2545     }
2546
2547     /* LOG, LOGP and POW operate on the absolute value of the input */
2548     src0_copy.modifiers = abs_modifier(src0_copy.modifiers, &need_abs);
2549
2550     shader_arb_get_dst_param(ins, &ins->dst[0], dst);
2551     shader_arb_get_src_param(ins, &src0_copy, 0, src0);
2552     if(arg2) shader_arb_get_src_param(ins, &ins->src[1], 1, src1);
2553
2554     if(need_abs)
2555     {
2556         shader_addline(buffer, "ABS TA, %s;\n", src0);
2557         if(arg2)
2558         {
2559             shader_addline(buffer, "%s%s %s, TA, %s;\n", instr, shader_arb_get_modifier(ins), dst, src1);
2560         }
2561         else
2562         {
2563             shader_addline(buffer, "%s%s %s, TA;\n", instr, shader_arb_get_modifier(ins), dst);
2564         }
2565     }
2566     else if(arg2)
2567     {
2568         shader_addline(buffer, "%s%s %s, %s, %s;\n", instr, shader_arb_get_modifier(ins), dst, src0, src1);
2569     }
2570     else
2571     {
2572         shader_addline(buffer, "%s%s %s, %s;\n", instr, shader_arb_get_modifier(ins), dst, src0);
2573     }
2574 }
2575
2576 static void shader_hw_loop(const struct wined3d_shader_instruction *ins)
2577 {
2578     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2579     char src_name[50];
2580     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2581
2582     /* src0 is aL */
2583     shader_arb_get_src_param(ins, &ins->src[1], 0, src_name);
2584
2585     if(vshader)
2586     {
2587         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2588         struct list *e = list_head(&priv->control_frames);
2589         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2590
2591         if(priv->loop_depth > 1) shader_addline(buffer, "PUSHA aL;\n");
2592         /* The constant loader makes sure to load -1 into iX.w */
2593         shader_addline(buffer, "ARLC aL, %s.xywz;\n", src_name);
2594         shader_addline(buffer, "BRA loop_%u_end (LE.x);\n", control_frame->loop_no);
2595         shader_addline(buffer, "loop_%u_start:\n", control_frame->loop_no);
2596     }
2597     else
2598     {
2599         shader_addline(buffer, "LOOP %s;\n", src_name);
2600     }
2601 }
2602
2603 static void shader_hw_rep(const struct wined3d_shader_instruction *ins)
2604 {
2605     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2606     char src_name[50];
2607     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2608
2609     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name);
2610
2611     /* The constant loader makes sure to load -1 into iX.w */
2612     if(vshader)
2613     {
2614         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2615         struct list *e = list_head(&priv->control_frames);
2616         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2617
2618         if(priv->loop_depth > 1) shader_addline(buffer, "PUSHA aL;\n");
2619
2620         shader_addline(buffer, "ARLC aL, %s.xywz;\n", src_name);
2621         shader_addline(buffer, "BRA loop_%u_end (LE.x);\n", control_frame->loop_no);
2622         shader_addline(buffer, "loop_%u_start:\n", control_frame->loop_no);
2623     }
2624     else
2625     {
2626         shader_addline(buffer, "REP %s;\n", src_name);
2627     }
2628 }
2629
2630 static void shader_hw_endloop(const struct wined3d_shader_instruction *ins)
2631 {
2632     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2633     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2634
2635     if(vshader)
2636     {
2637         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2638         struct list *e = list_head(&priv->control_frames);
2639         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2640
2641         shader_addline(buffer, "ARAC aL.xy, aL;\n");
2642         shader_addline(buffer, "BRA loop_%u_start (GT.x);\n", control_frame->loop_no);
2643         shader_addline(buffer, "loop_%u_end:\n", control_frame->loop_no);
2644
2645         if(priv->loop_depth > 1) shader_addline(buffer, "POPA aL;\n");
2646     }
2647     else
2648     {
2649         shader_addline(buffer, "ENDLOOP;\n");
2650     }
2651 }
2652
2653 static void shader_hw_endrep(const struct wined3d_shader_instruction *ins)
2654 {
2655     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2656     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2657
2658     if(vshader)
2659     {
2660         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2661         struct list *e = list_head(&priv->control_frames);
2662         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2663
2664         shader_addline(buffer, "ARAC aL.xy, aL;\n");
2665         shader_addline(buffer, "BRA loop_%u_start (GT.x);\n", control_frame->loop_no);
2666         shader_addline(buffer, "loop_%u_end:\n", control_frame->loop_no);
2667
2668         if(priv->loop_depth > 1) shader_addline(buffer, "POPA aL;\n");
2669     }
2670     else
2671     {
2672         shader_addline(buffer, "ENDREP;\n");
2673     }
2674 }
2675
2676 static const struct control_frame *find_last_loop(const struct shader_arb_ctx_priv *priv)
2677 {
2678     struct control_frame *control_frame;
2679
2680     LIST_FOR_EACH_ENTRY(control_frame, &priv->control_frames, struct control_frame, entry)
2681     {
2682         if(control_frame->type == LOOP || control_frame->type == REP) return control_frame;
2683     }
2684     ERR("Could not find loop for break\n");
2685     return NULL;
2686 }
2687
2688 static void shader_hw_break(const struct wined3d_shader_instruction *ins)
2689 {
2690     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2691     const struct control_frame *control_frame = find_last_loop(ins->ctx->backend_data);
2692     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2693
2694     if(vshader)
2695     {
2696         shader_addline(buffer, "BRA loop_%u_end;\n", control_frame->loop_no);
2697     }
2698     else
2699     {
2700         shader_addline(buffer, "BRK;\n");
2701     }
2702 }
2703
2704 static const char *get_compare(COMPARISON_TYPE flags)
2705 {
2706     switch (flags)
2707     {
2708         case COMPARISON_GT: return "GT";
2709         case COMPARISON_EQ: return "EQ";
2710         case COMPARISON_GE: return "GE";
2711         case COMPARISON_LT: return "LT";
2712         case COMPARISON_NE: return "NE";
2713         case COMPARISON_LE: return "LE";
2714         default:
2715             FIXME("Unrecognized comparison value: %u\n", flags);
2716             return "(\?\?)";
2717     }
2718 }
2719
2720 static COMPARISON_TYPE invert_compare(COMPARISON_TYPE flags)
2721 {
2722     switch (flags)
2723     {
2724         case COMPARISON_GT: return COMPARISON_LE;
2725         case COMPARISON_EQ: return COMPARISON_NE;
2726         case COMPARISON_GE: return COMPARISON_LT;
2727         case COMPARISON_LT: return COMPARISON_GE;
2728         case COMPARISON_NE: return COMPARISON_EQ;
2729         case COMPARISON_LE: return COMPARISON_GT;
2730         default:
2731             FIXME("Unrecognized comparison value: %u\n", flags);
2732             return -1;
2733     }
2734 }
2735
2736 static void shader_hw_breakc(const struct wined3d_shader_instruction *ins)
2737 {
2738     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2739     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2740     const struct control_frame *control_frame = find_last_loop(ins->ctx->backend_data);
2741     char src_name0[50];
2742     char src_name1[50];
2743     const char *comp = get_compare(ins->flags);
2744
2745     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name0);
2746     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name1);
2747
2748     if(vshader)
2749     {
2750         /* SUBC CC, src0, src1" works only in pixel shaders, so use TA to throw
2751          * away the subtraction result
2752          */
2753         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
2754         shader_addline(buffer, "BRA loop_%u_end (%s.x);\n", control_frame->loop_no, comp);
2755     }
2756     else
2757     {
2758         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
2759         shader_addline(buffer, "BRK (%s.x);\n", comp);
2760     }
2761 }
2762
2763 static void shader_hw_ifc(const struct wined3d_shader_instruction *ins)
2764 {
2765     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2766     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2767     struct list *e = list_head(&priv->control_frames);
2768     struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2769     const char *comp;
2770     char src_name0[50];
2771     char src_name1[50];
2772     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2773
2774     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name0);
2775     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name1);
2776
2777     if(vshader)
2778     {
2779         /* Invert the flag. We jump to the else label if the condition is NOT true */
2780         comp = get_compare(invert_compare(ins->flags));
2781         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
2782         shader_addline(buffer, "BRA ifc_%u_else (%s.x);\n", control_frame->ifc_no, comp);
2783     }
2784     else
2785     {
2786         comp = get_compare(ins->flags);
2787         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
2788         shader_addline(buffer, "IF %s.x;\n", comp);
2789     }
2790 }
2791
2792 static void shader_hw_else(const struct wined3d_shader_instruction *ins)
2793 {
2794     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2795     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2796     struct list *e = list_head(&priv->control_frames);
2797     struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2798     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2799
2800     if(vshader)
2801     {
2802         shader_addline(buffer, "BRA ifc_%u_endif;\n", control_frame->ifc_no);
2803         shader_addline(buffer, "ifc_%u_else:\n", control_frame->ifc_no);
2804         control_frame->had_else = TRUE;
2805     }
2806     else
2807     {
2808         shader_addline(buffer, "ELSE;\n");
2809     }
2810 }
2811
2812 static void shader_hw_endif(const struct wined3d_shader_instruction *ins)
2813 {
2814     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2815     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2816     struct list *e = list_head(&priv->control_frames);
2817     struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2818     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2819
2820     if(vshader)
2821     {
2822         if(control_frame->had_else)
2823         {
2824             shader_addline(buffer, "ifc_%u_endif:\n", control_frame->ifc_no);
2825         }
2826         else
2827         {
2828             shader_addline(buffer, "#No else branch. else is endif\n");
2829             shader_addline(buffer, "ifc_%u_else:\n", control_frame->ifc_no);
2830         }
2831     }
2832     else
2833     {
2834         shader_addline(buffer, "ENDIF;\n");
2835     }
2836 }
2837
2838 static void shader_hw_texldd(const struct wined3d_shader_instruction *ins)
2839 {
2840     DWORD sampler_idx = ins->src[1].reg.idx;
2841     char reg_dest[40];
2842     char reg_src[3][40];
2843     DWORD flags = TEX_DERIV;
2844
2845     shader_arb_get_dst_param(ins, &ins->dst[0], reg_dest);
2846     shader_arb_get_src_param(ins, &ins->src[0], 0, reg_src[0]);
2847     shader_arb_get_src_param(ins, &ins->src[2], 1, reg_src[1]);
2848     shader_arb_get_src_param(ins, &ins->src[3], 2, reg_src[2]);
2849
2850     if (ins->flags & WINED3DSI_TEXLD_PROJECT) flags |= TEX_PROJ;
2851     if (ins->flags & WINED3DSI_TEXLD_BIAS) flags |= TEX_BIAS;
2852
2853     shader_hw_sample(ins, sampler_idx, reg_dest, reg_src[0], flags, reg_src[1], reg_src[2]);
2854 }
2855
2856 static void shader_hw_texldl(const struct wined3d_shader_instruction *ins)
2857 {
2858     DWORD sampler_idx = ins->src[1].reg.idx;
2859     char reg_dest[40];
2860     char reg_coord[40];
2861     DWORD flags = TEX_LOD;
2862
2863     shader_arb_get_dst_param(ins, &ins->dst[0], reg_dest);
2864     shader_arb_get_src_param(ins, &ins->src[0], 0, reg_coord);
2865
2866     if (ins->flags & WINED3DSI_TEXLD_PROJECT) flags |= TEX_PROJ;
2867     if (ins->flags & WINED3DSI_TEXLD_BIAS) flags |= TEX_BIAS;
2868
2869     shader_hw_sample(ins, sampler_idx, reg_dest, reg_coord, flags, NULL, NULL);
2870 }
2871
2872 static void shader_hw_label(const struct wined3d_shader_instruction *ins)
2873 {
2874     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2875     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2876
2877     priv->in_main_func = FALSE;
2878     /* Call instructions activate the NV extensions, not labels and rets. If there is an uncalled
2879      * subroutine, don't generate a label that will make GL complain
2880      */
2881     if(priv->target_version == ARB) return;
2882
2883     shader_addline(buffer, "l%u:\n", ins->src[0].reg.idx);
2884 }
2885
2886 static void vshader_add_footer(IWineD3DVertexShaderImpl *This, struct wined3d_shader_buffer *buffer,
2887         const struct arb_vs_compile_args *args, struct shader_arb_ctx_priv *priv_ctx)
2888 {
2889     const shader_reg_maps *reg_maps = &This->baseShader.reg_maps;
2890     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *)This->baseShader.device;
2891     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
2892     unsigned int i;
2893
2894     /* The D3DRS_FOGTABLEMODE render state defines if the shader-generated fog coord is used
2895      * or if the fragment depth is used. If the fragment depth is used(FOGTABLEMODE != NONE),
2896      * the fog frag coord is thrown away. If the fog frag coord is used, but not written by
2897      * the shader, it is set to 0.0(fully fogged, since start = 1.0, end = 0.0)
2898      */
2899     if(args->super.fog_src == VS_FOG_Z) {
2900         shader_addline(buffer, "MOV result.fogcoord, TMP_OUT.z;\n");
2901     } else if (!reg_maps->fog) {
2902         /* posFixup.x is always 1.0, so we can savely use it */
2903         shader_addline(buffer, "ADD result.fogcoord, posFixup.x, -posFixup.x;\n");
2904     }
2905
2906     /* Write the final position.
2907      *
2908      * OpenGL coordinates specify the center of the pixel while d3d coords specify
2909      * the corner. The offsets are stored in z and w in posFixup. posFixup.y contains
2910      * 1.0 or -1.0 to turn the rendering upside down for offscreen rendering. PosFixup.x
2911      * contains 1.0 to allow a mad, but arb vs swizzles are too restricted for that.
2912      */
2913     shader_addline(buffer, "MUL TA, posFixup, TMP_OUT.w;\n");
2914     shader_addline(buffer, "ADD TMP_OUT.x, TMP_OUT.x, TA.z;\n");
2915     shader_addline(buffer, "MAD TMP_OUT.y, TMP_OUT.y, posFixup.y, TA.w;\n");
2916
2917     if(use_nv_clip(gl_info) && priv_ctx->target_version >= NV2)
2918     {
2919         for(i = 0; i < priv_ctx->vs_clipplanes; i++)
2920         {
2921             shader_addline(buffer, "DP4 result.clip[%u].x, TMP_OUT, state.clip[%u].plane;\n", i, i);
2922         }
2923     }
2924     else if(args->boolclip.clip_control[0])
2925     {
2926         unsigned int cur_clip = 0;
2927         char component[4] = {'x', 'y', 'z', 'w'};
2928
2929         for(i = 0; i < GL_LIMITS(clipplanes); i++)
2930         {
2931             if(args->boolclip.clip_control[1] & (1 << i))
2932             {
2933                 shader_addline(buffer, "DP4 TA.%c, TMP_OUT, state.clip[%u].plane;\n",
2934                                component[cur_clip++], i);
2935             }
2936         }
2937         switch(cur_clip)
2938         {
2939             case 0:
2940                 shader_addline(buffer, "MOV TA, -helper_const.w;\n");
2941                 break;
2942             case 1:
2943                 shader_addline(buffer, "MOV TA.yzw, -helper_const.w;\n");
2944                 break;
2945             case 2:
2946                 shader_addline(buffer, "MOV TA.zw, -helper_const.w;\n");
2947                 break;
2948             case 3:
2949                 shader_addline(buffer, "MOV TA.w, -helper_const.w;\n");
2950                 break;
2951         }
2952         shader_addline(buffer, "MOV result.texcoord[%u], TA;\n",
2953                        args->boolclip.clip_control[0] - 1);
2954     }
2955
2956     /* Z coord [0;1]->[-1;1] mapping, see comment in transform_projection in state.c
2957      * and the glsl equivalent
2958      */
2959     if(need_helper_const(gl_info)) {
2960         shader_addline(buffer, "MAD TMP_OUT.z, TMP_OUT.z, helper_const.x, -TMP_OUT.w;\n");
2961     } else {
2962         shader_addline(buffer, "ADD TMP_OUT.z, TMP_OUT.z, TMP_OUT.z;\n");
2963         shader_addline(buffer, "ADD TMP_OUT.z, TMP_OUT.z, -TMP_OUT.w;\n");
2964     }
2965
2966     shader_addline(buffer, "MOV result.position, TMP_OUT;\n");
2967
2968     priv_ctx->footer_written = TRUE;
2969 }
2970
2971 static void shader_hw_ret(const struct wined3d_shader_instruction *ins)
2972 {
2973     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2974     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2975     IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *) ins->ctx->shader;
2976     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2977
2978     if(priv->target_version == ARB) return;
2979
2980     if(vshader)
2981     {
2982         if(priv->in_main_func) vshader_add_footer((IWineD3DVertexShaderImpl *) shader, buffer, priv->cur_vs_args, priv);
2983     }
2984
2985     shader_addline(buffer, "RET;\n");
2986 }
2987
2988 static void shader_hw_call(const struct wined3d_shader_instruction *ins)
2989 {
2990     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2991     shader_addline(buffer, "CAL l%u;\n", ins->src[0].reg.idx);
2992 }
2993
2994 /* GL locking is done by the caller */
2995 static GLuint create_arb_blt_vertex_program(const struct wined3d_gl_info *gl_info)
2996 {
2997     GLuint program_id = 0;
2998     GLint pos;
2999
3000     const char *blt_vprogram =
3001         "!!ARBvp1.0\n"
3002         "PARAM c[1] = { { 1, 0.5 } };\n"
3003         "MOV result.position, vertex.position;\n"
3004         "MOV result.color, c[0].x;\n"
3005         "MOV result.texcoord[0], vertex.texcoord[0];\n"
3006         "END\n";
3007
3008     GL_EXTCALL(glGenProgramsARB(1, &program_id));
3009     GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, program_id));
3010     GL_EXTCALL(glProgramStringARB(GL_VERTEX_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
3011             strlen(blt_vprogram), blt_vprogram));
3012     checkGLcall("glProgramStringARB()");
3013
3014     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
3015     if (pos != -1)
3016     {
3017         FIXME("Vertex program error at position %d: %s\n", pos,
3018             debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
3019     }
3020     else
3021     {
3022         GLint native;
3023
3024         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
3025         checkGLcall("glGetProgramivARB()");
3026         if (!native) WARN("Program exceeds native resource limits.\n");
3027     }
3028
3029     return program_id;
3030 }
3031
3032 /* GL locking is done by the caller */
3033 static GLuint create_arb_blt_fragment_program(const struct wined3d_gl_info *gl_info, enum tex_types tex_type)
3034 {
3035     GLuint program_id = 0;
3036     GLint pos;
3037
3038     static const char * const blt_fprograms[tex_type_count] =
3039     {
3040         /* tex_1d */
3041         NULL,
3042         /* tex_2d */
3043         "!!ARBfp1.0\n"
3044         "TEMP R0;\n"
3045         "TEX R0.x, fragment.texcoord[0], texture[0], 2D;\n"
3046         "MOV result.depth.z, R0.x;\n"
3047         "END\n",
3048         /* tex_3d */
3049         NULL,
3050         /* tex_cube */
3051         "!!ARBfp1.0\n"
3052         "TEMP R0;\n"
3053         "TEX R0.x, fragment.texcoord[0], texture[0], CUBE;\n"
3054         "MOV result.depth.z, R0.x;\n"
3055         "END\n",
3056         /* tex_rect */
3057         "!!ARBfp1.0\n"
3058         "TEMP R0;\n"
3059         "TEX R0.x, fragment.texcoord[0], texture[0], RECT;\n"
3060         "MOV result.depth.z, R0.x;\n"
3061         "END\n",
3062     };
3063
3064     if (!blt_fprograms[tex_type])
3065     {
3066         FIXME("tex_type %#x not supported\n", tex_type);
3067         tex_type = tex_2d;
3068     }
3069
3070     GL_EXTCALL(glGenProgramsARB(1, &program_id));
3071     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, program_id));
3072     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
3073             strlen(blt_fprograms[tex_type]), blt_fprograms[tex_type]));
3074     checkGLcall("glProgramStringARB()");
3075
3076     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
3077     if (pos != -1)
3078     {
3079         FIXME("Fragment program error at position %d: %s\n", pos,
3080             debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
3081     }
3082     else
3083     {
3084         GLint native;
3085
3086         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
3087         checkGLcall("glGetProgramivARB()");
3088         if (!native) WARN("Program exceeds native resource limits.\n");
3089     }
3090
3091     return program_id;
3092 }
3093
3094 static void arbfp_add_sRGB_correction(struct wined3d_shader_buffer *buffer, const char *fragcolor,
3095         const char *tmp1, const char *tmp2, const char *tmp3, const char *tmp4, BOOL condcode)
3096 {
3097     /* Perform sRGB write correction. See GLX_EXT_framebuffer_sRGB */
3098
3099     if(condcode)
3100     {
3101         /* Sigh. MOVC CC doesn't work, so use one of the temps as dummy dest */
3102         shader_addline(buffer, "SUBC %s, %s.x, srgb_consts1.y;\n", tmp1, fragcolor);
3103         /* Calculate the > 0.0031308 case */
3104         shader_addline(buffer, "POW %s.x (GE), %s.x, srgb_consts1.z;\n", fragcolor, fragcolor);
3105         shader_addline(buffer, "POW %s.y (GE), %s.y, srgb_consts1.z;\n", fragcolor, fragcolor);
3106         shader_addline(buffer, "POW %s.z (GE), %s.z, srgb_consts1.z;\n", fragcolor, fragcolor);
3107         shader_addline(buffer, "MUL %s.xyz (GE), %s, srgb_consts1.w;\n", fragcolor, fragcolor);
3108         shader_addline(buffer, "SUB %s.xyz (GE), %s, srgb_consts2.x;\n", fragcolor, fragcolor);
3109         /* Calculate the < case */
3110         shader_addline(buffer, "MUL %s.xyz (LT), srgb_consts1.x, %s;\n", fragcolor, fragcolor);
3111     }
3112     else
3113     {
3114         /* Calculate the > 0.0031308 case */
3115         shader_addline(buffer, "POW %s.x, %s.x, srgb_consts1.z;\n", tmp1, fragcolor);
3116         shader_addline(buffer, "POW %s.y, %s.y, srgb_consts1.z;\n", tmp1, fragcolor);
3117         shader_addline(buffer, "POW %s.z, %s.z, srgb_consts1.z;\n", tmp1, fragcolor);
3118         shader_addline(buffer, "MUL %s, %s, srgb_consts1.w;\n", tmp1, tmp1);
3119         shader_addline(buffer, "SUB %s, %s, srgb_consts2.x;\n", tmp1, tmp1);
3120         /* Calculate the < case */
3121         shader_addline(buffer, "MUL %s, srgb_consts1.x, %s;\n", tmp2, fragcolor);
3122         /* Get 1.0 / 0.0 masks for > 0.0031308 and < 0.0031308 */
3123         shader_addline(buffer, "SLT %s, srgb_consts1.y, %s;\n", tmp3, fragcolor);
3124         shader_addline(buffer, "SGE %s, srgb_consts1.y, %s;\n", tmp4, fragcolor);
3125         /* Store the components > 0.0031308 in the destination */
3126         shader_addline(buffer, "MUL %s.xyz, %s, %s;\n", fragcolor, tmp1, tmp3);
3127         /* Add the components that are < 0.0031308 */
3128         shader_addline(buffer, "MAD %s.xyz, %s, %s, %s;\n", fragcolor, tmp2, tmp4, fragcolor);
3129         /* Move everything into result.color at once. Nvidia hardware cannot handle partial
3130         * result.color writes(.rgb first, then .a), or handle overwriting already written
3131         * components. The assembler uses a temporary register in this case, which is usually
3132         * not allocated from one of our registers that were used earlier.
3133         */
3134     }
3135     shader_addline(buffer, "MOV result.color, %s;\n", fragcolor);
3136     /* [0.0;1.0] clamping. Not needed, this is done implicitly */
3137 }
3138
3139 static const DWORD *find_loop_control_values(IWineD3DBaseShaderImpl *This, DWORD idx)
3140 {
3141     const local_constant *constant;
3142
3143     LIST_FOR_EACH_ENTRY(constant, &This->baseShader.constantsI, local_constant, entry)
3144     {
3145         if (constant->idx == idx)
3146         {
3147             return constant->value;
3148         }
3149     }
3150     return NULL;
3151 }
3152
3153 static void init_ps_input(const IWineD3DPixelShaderImpl *This, const struct arb_ps_compile_args *args,
3154                           struct shader_arb_ctx_priv *priv)
3155 {
3156     const char *texcoords[8] =
3157     {
3158         "fragment.texcoord[0]", "fragment.texcoord[1]", "fragment.texcoord[2]", "fragment.texcoord[3]",
3159         "fragment.texcoord[4]", "fragment.texcoord[5]", "fragment.texcoord[6]", "fragment.texcoord[7]"
3160     };
3161     unsigned int i;
3162     const struct wined3d_shader_signature_element *sig = This->input_signature;
3163     const char *semantic_name;
3164     DWORD semantic_idx;
3165
3166     switch(args->super.vp_mode)
3167     {
3168         case pretransformed:
3169         case fixedfunction:
3170             /* The pixelshader has to collect the varyings on its own. In any case properly load
3171              * color0 and color1. In the case of pretransformed vertices also load texcoords. Set
3172              * other attribs to 0.0.
3173              *
3174              * For fixedfunction this behavior is correct, according to the tests. For pretransformed
3175              * we'd either need a replacement shader that can load other attribs like BINORMAL, or
3176              * load the texcoord attrib pointers to match the pixel shader signature
3177              */
3178             for(i = 0; i < MAX_REG_INPUT; i++)
3179             {
3180                 semantic_name = sig[i].semantic_name;
3181                 semantic_idx = sig[i].semantic_idx;
3182                 if(semantic_name == NULL) continue;
3183
3184                 if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_COLOR))
3185                 {
3186                     if(semantic_idx == 0) priv->ps_input[i] = "fragment.color.primary";
3187                     else if(semantic_idx == 1) priv->ps_input[i] = "fragment.color.secondary";
3188                     else priv->ps_input[i] = "0.0";
3189                 }
3190                 else if(args->super.vp_mode == fixedfunction)
3191                 {
3192                     priv->ps_input[i] = "0.0";
3193                 }
3194                 else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_TEXCOORD))
3195                 {
3196                     if(semantic_idx < 8) priv->ps_input[i] = texcoords[semantic_idx];
3197                     else priv->ps_input[i] = "0.0";
3198                 }
3199                 else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_FOG))
3200                 {
3201                     if(semantic_idx == 0) priv->ps_input[i] = "fragment.fogcoord";
3202                     else priv->ps_input[i] = "0.0";
3203                 }
3204                 else
3205                 {
3206                     priv->ps_input[i] = "0.0";
3207                 }
3208
3209                 TRACE("v%u, semantic %s%u is %s\n", i, semantic_name, semantic_idx, priv->ps_input[i]);
3210             }
3211             break;
3212
3213         case vertexshader:
3214             /* That one is easy. The vertex shaders provide v0-v7 in fragment.texcoord and v8 and v9 in
3215              * fragment.color
3216              */
3217             for(i = 0; i < 8; i++)
3218             {
3219                 priv->ps_input[i] = texcoords[i];
3220             }
3221             priv->ps_input[8] = "fragment.color.primary";
3222             priv->ps_input[9] = "fragment.color.secondary";
3223             break;
3224     }
3225 }
3226
3227 /* GL locking is done by the caller */
3228 static GLuint shader_arb_generate_pshader(IWineD3DPixelShaderImpl *This, struct wined3d_shader_buffer *buffer,
3229         const struct arb_ps_compile_args *args, struct arb_ps_compiled_shader *compiled)
3230 {
3231     const shader_reg_maps* reg_maps = &This->baseShader.reg_maps;
3232     CONST DWORD *function = This->baseShader.function;
3233     const struct wined3d_gl_info *gl_info = &((IWineD3DDeviceImpl *)This->baseShader.device)->adapter->gl_info;
3234     const local_constant *lconst;
3235     GLuint retval;
3236     char fragcolor[16];
3237     DWORD *lconst_map = local_const_mapping((IWineD3DBaseShaderImpl *) This), next_local, cur;
3238     struct shader_arb_ctx_priv priv_ctx;
3239     BOOL dcl_tmp = args->super.srgb_correction, dcl_td = FALSE;
3240     BOOL want_nv_prog = FALSE;
3241     struct arb_pshader_private *shader_priv = This->backend_priv;
3242     GLint errPos;
3243     DWORD map;
3244
3245     char srgbtmp[4][4];
3246     unsigned int i, found = 0;
3247
3248     for (i = 0, map = reg_maps->temporary; map; map >>= 1, ++i)
3249     {
3250         if (!(map & 1)
3251                 || (This->color0_mov && i == This->color0_reg)
3252                 || (reg_maps->shader_version.major < 2 && i == 0))
3253             continue;
3254
3255         sprintf(srgbtmp[found], "R%u", i);
3256         ++found;
3257         if (found == 4) break;
3258     }
3259
3260     switch(found) {
3261         case 4: dcl_tmp = FALSE; break;
3262         case 0:
3263             sprintf(srgbtmp[0], "TA");
3264             sprintf(srgbtmp[1], "TB");
3265             sprintf(srgbtmp[2], "TC");
3266             sprintf(srgbtmp[3], "TD");
3267             dcl_td = TRUE;
3268             break;
3269         case 1:
3270             sprintf(srgbtmp[1], "TA");
3271             sprintf(srgbtmp[2], "TB");
3272             sprintf(srgbtmp[3], "TC");
3273             break;
3274         case 2:
3275             sprintf(srgbtmp[2], "TA");
3276             sprintf(srgbtmp[3], "TB");
3277             break;
3278         case 3:
3279             sprintf(srgbtmp[3], "TA");
3280             break;
3281     }
3282
3283     /*  Create the hw ARB shader */
3284     memset(&priv_ctx, 0, sizeof(priv_ctx));
3285     priv_ctx.cur_ps_args = args;
3286     priv_ctx.compiled_fprog = compiled;
3287     priv_ctx.cur_np2fixup_info = &compiled->np2fixup_info;
3288     init_ps_input(This, args, &priv_ctx);
3289     list_init(&priv_ctx.control_frames);
3290
3291     /* Avoid enabling NV_fragment_program* if we do not need it.
3292      *
3293      * Enabling GL_NV_fragment_program_option causes the driver to occupy a temporary register,
3294      * and it slows down the shader execution noticeably(about 5%). Usually our instruction emulation
3295      * is faster than what we gain from using higher native instructions. There are some things though
3296      * that cannot be emulated. In that case enable the extensions.
3297      * If the extension is enabled, instruction handlers that support both ways will use it.
3298      *
3299      * Testing shows no performance difference between OPTION NV_fragment_program2 and NV_fragment_program.
3300      * So enable the best we can get.
3301      */
3302     if(reg_maps->usesdsx || reg_maps->usesdsy || reg_maps->loop_depth > 0 || reg_maps->usestexldd ||
3303        reg_maps->usestexldl || reg_maps->usesfacing || reg_maps->usesifc || reg_maps->usescall)
3304     {
3305         want_nv_prog = TRUE;
3306     }
3307
3308     shader_addline(buffer, "!!ARBfp1.0\n");
3309     if(want_nv_prog && GL_SUPPORT(NV_FRAGMENT_PROGRAM2)) {
3310         shader_addline(buffer, "OPTION NV_fragment_program2;\n");
3311         priv_ctx.target_version = NV3;
3312     } else if(want_nv_prog && GL_SUPPORT(NV_FRAGMENT_PROGRAM_OPTION)) {
3313         shader_addline(buffer, "OPTION NV_fragment_program;\n");
3314         priv_ctx.target_version = NV2;
3315     } else {
3316         if(want_nv_prog)
3317         {
3318             /* This is an error - either we're advertising the wrong shader version, or aren't enforcing some
3319              * limits properly
3320              */
3321             ERR("The shader requires instructions that are not available in plain GL_ARB_fragment_program\n");
3322             ERR("Try GLSL\n");
3323         }
3324         priv_ctx.target_version = ARB;
3325     }
3326
3327     if(This->baseShader.reg_maps.highest_render_target > 0)
3328     {
3329         shader_addline(buffer, "OPTION ARB_draw_buffers;\n");
3330     }
3331
3332     if (reg_maps->shader_version.major < 3)
3333     {
3334         switch(args->super.fog) {
3335             case FOG_OFF:
3336                 break;
3337             case FOG_LINEAR:
3338                 shader_addline(buffer, "OPTION ARB_fog_linear;\n");
3339                 break;
3340             case FOG_EXP:
3341                 shader_addline(buffer, "OPTION ARB_fog_exp;\n");
3342                 break;
3343             case FOG_EXP2:
3344                 shader_addline(buffer, "OPTION ARB_fog_exp2;\n");
3345                 break;
3346         }
3347     }
3348
3349     /* For now always declare the temps. At least the Nvidia assembler optimizes completely
3350      * unused temps away(but occupies them for the whole shader if they're used once). Always
3351      * declaring them avoids tricky bookkeeping work
3352      */
3353     shader_addline(buffer, "TEMP TA;\n");      /* Used for modifiers */
3354     shader_addline(buffer, "TEMP TB;\n");      /* Used for modifiers */
3355     shader_addline(buffer, "TEMP TC;\n");      /* Used for modifiers */
3356     if(dcl_td) shader_addline(buffer, "TEMP TD;\n"); /* Used for sRGB writing */
3357     shader_addline(buffer, "PARAM coefdiv = { 0.5, 0.25, 0.125, 0.0625 };\n");
3358     shader_addline(buffer, "PARAM coefmul = { 2, 4, 8, 16 };\n");
3359     shader_addline(buffer, "PARAM one = { 1.0, 1.0, 1.0, 1.0 };\n");
3360
3361     if (reg_maps->shader_version.major < 2)
3362     {
3363         strcpy(fragcolor, "R0");
3364     } else {
3365         if(args->super.srgb_correction) {
3366             if(This->color0_mov) {
3367                 sprintf(fragcolor, "R%u", This->color0_reg);
3368             } else {
3369                 shader_addline(buffer, "TEMP TMP_COLOR;\n");
3370                 strcpy(fragcolor, "TMP_COLOR");
3371             }
3372         } else {
3373             strcpy(fragcolor, "result.color");
3374         }
3375     }
3376
3377     if(args->super.srgb_correction) {
3378         shader_addline(buffer, "PARAM srgb_consts1 = {%f, %f, %f, %f};\n",
3379                        srgb_mul_low, srgb_cmp, srgb_pow, srgb_mul_high);
3380         shader_addline(buffer, "PARAM srgb_consts2 = {%f, %f, %f, %f};\n",
3381                        srgb_sub_high, 0.0, 0.0, 0.0);
3382     }
3383
3384     /* Base Declarations */
3385     next_local = shader_generate_arb_declarations( (IWineD3DBaseShader*) This, reg_maps, buffer, &GLINFO_LOCATION,
3386             lconst_map, NULL, &priv_ctx);
3387
3388     for (i = 0, map = reg_maps->bumpmat; map; map >>= 1, ++i)
3389     {
3390         if (!(map & 1)) continue;
3391
3392         cur = compiled->numbumpenvmatconsts;
3393         compiled->bumpenvmatconst[cur].const_num = WINED3D_CONST_NUM_UNUSED;
3394         compiled->bumpenvmatconst[cur].texunit = i;
3395         compiled->luminanceconst[cur].const_num = WINED3D_CONST_NUM_UNUSED;
3396         compiled->luminanceconst[cur].texunit = i;
3397
3398         /* We can fit the constants into the constant limit for sure because texbem, texbeml, bem and beml are only supported
3399          * in 1.x shaders, and GL_ARB_fragment_program has a constant limit of 24 constants. So in the worst case we're loading
3400          * 8 shader constants, 8 bump matrices and 8 luminance parameters and are perfectly fine. (No NP2 fixup on bumpmapped
3401          * textures due to conditional NP2 restrictions)
3402          *
3403          * Use local constants to load the bump env parameters, not program.env. This avoids collisions with d3d constants of
3404          * shaders in newer shader models. Since the bump env parameters have to share their space with NP2 fixup constants,
3405          * their location is shader dependent anyway and they cannot be loaded globally.
3406          */
3407         compiled->bumpenvmatconst[cur].const_num = next_local++;
3408         shader_addline(buffer, "PARAM bumpenvmat%d = program.local[%d];\n",
3409                        i, compiled->bumpenvmatconst[cur].const_num);
3410         compiled->numbumpenvmatconsts = cur + 1;
3411
3412         if (!(reg_maps->luminanceparams & (1 << i))) continue;
3413
3414         compiled->luminanceconst[cur].const_num = next_local++;
3415         shader_addline(buffer, "PARAM luminance%d = program.local[%d];\n",
3416                        i, compiled->luminanceconst[cur].const_num);
3417     }
3418
3419     for(i = 0; i < MAX_CONST_I; i++)
3420     {
3421         compiled->int_consts[i] = WINED3D_CONST_NUM_UNUSED;
3422         if (reg_maps->integer_constants & (1 << i) && priv_ctx.target_version >= NV2)
3423         {
3424             const DWORD *control_values = find_loop_control_values((IWineD3DBaseShaderImpl *) This, i);
3425
3426             if(control_values)
3427             {
3428                 shader_addline(buffer, "PARAM I%u = {%u, %u, %u, -1};\n", i,
3429                                 control_values[0], control_values[1], control_values[2]);
3430             }
3431             else
3432             {
3433                 compiled->int_consts[i] = next_local;
3434                 compiled->num_int_consts++;
3435                 shader_addline(buffer, "PARAM I%u = program.local[%u];\n", i, next_local++);
3436             }
3437         }
3438     }
3439
3440     if(reg_maps->vpos || reg_maps->usesdsy)
3441     {
3442         compiled->ycorrection = next_local;
3443         shader_addline(buffer, "PARAM ycorrection = program.local[%u];\n", next_local++);
3444
3445         if(reg_maps->vpos)
3446         {
3447             shader_addline(buffer, "TEMP vpos;\n");
3448             /* ycorrection.x: Backbuffer height(onscreen) or 0(offscreen).
3449              * ycorrection.y: -1.0(onscreen), 1.0(offscreen)
3450              * ycorrection.z: 1.0
3451              * ycorrection.w: 0.0
3452              */
3453             shader_addline(buffer, "MAD vpos, fragment.position, ycorrection.zyww, ycorrection.wxww;\n");
3454             shader_addline(buffer, "FLR vpos.xy, vpos;\n");
3455         }
3456     }
3457     else
3458     {
3459         compiled->ycorrection = WINED3D_CONST_NUM_UNUSED;
3460     }
3461
3462     /* Load constants to fixup NP2 texcoords if there are still free constants left:
3463      * Constants (texture dimensions) for the NP2 fixup are loaded as local program parameters. This will consume
3464      * at most 8 (MAX_FRAGMENT_SAMPLERS / 2) parameters, which is highly unlikely, since the application had to
3465      * use 16 NP2 textures at the same time. In case that we run out of constants the fixup is simply not
3466      * applied / activated. This will probably result in wrong rendering of the texture, but will save us from
3467      * shader compilation errors and the subsequent errors when drawing with this shader. */
3468     if (priv_ctx.cur_ps_args->super.np2_fixup) {
3469
3470         struct arb_ps_np2fixup_info* const fixup = priv_ctx.cur_np2fixup_info;
3471         const WORD map = priv_ctx.cur_ps_args->super.np2_fixup;
3472         const UINT max_lconsts = gl_info->ps_arb_max_local_constants;
3473
3474         fixup->offset = next_local;
3475         fixup->super.active = 0;
3476
3477         cur = 0;
3478         for (i = 0; i < MAX_FRAGMENT_SAMPLERS; ++i) {
3479             if (!(map & (1 << i))) continue;
3480
3481             if (fixup->offset + (cur >> 1) < max_lconsts) {
3482                 fixup->super.active |= (1 << i);
3483                 fixup->super.idx[i] = cur++;
3484             } else {
3485                 FIXME("No free constant found to load NP2 fixup data into shader. "
3486                       "Sampling from this texture will probably look wrong.\n");
3487                 break;
3488             }
3489         }
3490
3491         fixup->super.num_consts = (cur + 1) >> 1;
3492         if (fixup->super.num_consts) {
3493             shader_addline(buffer, "PARAM np2fixup[%u] = { program.env[%u..%u] };\n",
3494                            fixup->super.num_consts, fixup->offset, fixup->super.num_consts + fixup->offset - 1);
3495         }
3496
3497         next_local += fixup->super.num_consts;
3498     }
3499
3500     if (shader_priv->clipplane_emulation != ~0U && args->clip)
3501     {
3502         shader_addline(buffer, "KIL fragment.texcoord[%u];\n", shader_priv->clipplane_emulation);
3503     }
3504
3505     /* Base Shader Body */
3506     shader_generate_main((IWineD3DBaseShader *)This, buffer, reg_maps, function, &priv_ctx);
3507
3508     if(args->super.srgb_correction) {
3509         arbfp_add_sRGB_correction(buffer, fragcolor, srgbtmp[0], srgbtmp[1], srgbtmp[2], srgbtmp[3],
3510                                   priv_ctx.target_version >= NV2);
3511     } else if(reg_maps->shader_version.major < 2) {
3512         shader_addline(buffer, "MOV result.color, %s;\n", fragcolor);
3513     }
3514     shader_addline(buffer, "END\n");
3515
3516     /* TODO: change to resource.glObjectHandle or something like that */
3517     GL_EXTCALL(glGenProgramsARB(1, &retval));
3518
3519     TRACE("Creating a hw pixel shader, prg=%d\n", retval);
3520     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, retval));
3521
3522     TRACE("Created hw pixel shader, prg=%d\n", retval);
3523     /* Create the program and check for errors */
3524     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
3525                buffer->bsize, buffer->buffer));
3526     checkGLcall("glProgramStringARB()");
3527
3528     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &errPos);
3529     if (errPos != -1)
3530     {
3531         FIXME("HW PixelShader Error at position %d: %s\n",
3532               errPos, debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
3533         retval = 0;
3534     }
3535     else
3536     {
3537         GLint native;
3538
3539         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
3540         checkGLcall("glGetProgramivARB()");
3541         if (!native) WARN("Program exceeds native resource limits.\n");
3542     }
3543
3544     /* Load immediate constants */
3545     if(lconst_map) {
3546         LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry) {
3547             const float *value = (const float *)lconst->value;
3548             GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, lconst_map[lconst->idx], value));
3549             checkGLcall("glProgramLocalParameter4fvARB");
3550         }
3551         HeapFree(GetProcessHeap(), 0, lconst_map);
3552     }
3553
3554     return retval;
3555 }
3556
3557 static int compare_sig(const struct wined3d_shader_signature_element *sig1, const struct wined3d_shader_signature_element *sig2)
3558 {
3559     unsigned int i;
3560     int ret;
3561
3562     for(i = 0; i < MAX_REG_INPUT; i++)
3563     {
3564         if(sig1[i].semantic_name == NULL || sig2[i].semantic_name == NULL)
3565         {
3566             /* Compare pointers, not contents. One string is NULL(element does not exist), the other one is not NULL */
3567             if(sig1[i].semantic_name != sig2[i].semantic_name) return sig1[i].semantic_name < sig2[i].semantic_name ? -1 : 1;
3568             continue;
3569         }
3570
3571         ret = strcmp(sig1[i].semantic_name, sig2[i].semantic_name);
3572         if(ret != 0) return ret;
3573         if(sig1[i].semantic_idx    != sig2[i].semantic_idx)    return sig1[i].semantic_idx    < sig2[i].semantic_idx    ? -1 : 1;
3574         if(sig1[i].sysval_semantic != sig2[i].sysval_semantic) return sig1[i].sysval_semantic < sig2[i].sysval_semantic ? -1 : 1;
3575         if(sig1[i].component_type  != sig2[i].component_type)  return sig1[i].sysval_semantic < sig2[i].component_type  ? -1 : 1;
3576         if(sig1[i].register_idx    != sig2[i].register_idx)    return sig1[i].register_idx    < sig2[i].register_idx    ? -1 : 1;
3577         if(sig1[i].mask            != sig2->mask)              return sig1[i].mask            < sig2[i].mask            ? -1 : 1;
3578     }
3579     return 0;
3580 }
3581
3582 static struct wined3d_shader_signature_element *clone_sig(const struct wined3d_shader_signature_element *sig)
3583 {
3584     struct wined3d_shader_signature_element *new;
3585     int i;
3586     char *name;
3587
3588     new = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*new) * MAX_REG_INPUT);
3589     for(i = 0; i < MAX_REG_INPUT; i++)
3590     {
3591         if(sig[i].semantic_name == NULL)
3592         {
3593             continue;
3594         }
3595
3596         new[i] = sig[i];
3597         /* Clone the semantic string */
3598         name = HeapAlloc(GetProcessHeap(), 0, strlen(sig[i].semantic_name) + 1);
3599         strcpy(name, sig[i].semantic_name);
3600         new[i].semantic_name = name;
3601     }
3602     return new;
3603 }
3604
3605 static DWORD find_input_signature(struct shader_arb_priv *priv, const struct wined3d_shader_signature_element *sig)
3606 {
3607     struct wine_rb_entry *entry = wine_rb_get(&priv->signature_tree, sig);
3608     struct ps_signature *found_sig;
3609
3610     if(entry != NULL)
3611     {
3612         found_sig = WINE_RB_ENTRY_VALUE(entry, struct ps_signature, entry);
3613         TRACE("Found existing signature %u\n", found_sig->idx);
3614         return found_sig->idx;
3615     }
3616     found_sig = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*sig));
3617     found_sig->sig = clone_sig(sig);
3618     found_sig->idx = priv->ps_sig_number++;
3619     TRACE("New signature stored and assigned number %u\n", found_sig->idx);
3620     if(wine_rb_put(&priv->signature_tree, sig, &found_sig->entry) == -1)
3621     {
3622         ERR("Failed to insert program entry.\n");
3623     }
3624     return found_sig->idx;
3625 }
3626
3627 static void init_output_registers(IWineD3DVertexShaderImpl *shader, DWORD sig_num, struct shader_arb_ctx_priv *priv_ctx,
3628                                   struct arb_vs_compiled_shader *compiled)
3629 {
3630     unsigned int i, j;
3631     static const char *texcoords[8] =
3632     {
3633         "result.texcoord[0]", "result.texcoord[1]", "result.texcoord[2]", "result.texcoord[3]",
3634         "result.texcoord[4]", "result.texcoord[5]", "result.texcoord[6]", "result.texcoord[7]"
3635     };
3636     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *) shader->baseShader.device;
3637     const struct wined3d_shader_signature_element *sig;
3638     const char *semantic_name;
3639     DWORD semantic_idx, reg_idx;
3640
3641     /* Write generic input varyings 0 to 7 to result.texcoord[], varying 8 to result.color.primary
3642      * and varying 9 to result.color.secondary
3643      */
3644     const char *decl_idx_to_string[MAX_REG_INPUT] =
3645     {
3646         texcoords[0], texcoords[1], texcoords[2], texcoords[3],
3647         texcoords[4], texcoords[5], texcoords[6], texcoords[7],
3648         "result.color.primary", "result.color.secondary"
3649     };
3650
3651     if(sig_num == ~0)
3652     {
3653         TRACE("Pixel shader uses builtin varyings\n");
3654         /* Map builtins to builtins */
3655         for(i = 0; i < 8; i++)
3656         {
3657             priv_ctx->texcrd_output[i] = texcoords[i];
3658         }
3659         priv_ctx->color_output[0] = "result.color.primary";
3660         priv_ctx->color_output[1] = "result.color.secondary";
3661         priv_ctx->fog_output = "result.fogcoord";
3662
3663         /* Map declared regs to builtins. Use "TA" to /dev/null unread output */
3664         for(i = 0; i < (sizeof(shader->output_signature) / sizeof(*shader->output_signature)); i++)
3665         {
3666             semantic_name = shader->output_signature[i].semantic_name;
3667             if(semantic_name == NULL) continue;
3668
3669             if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_POSITION))
3670             {
3671                 TRACE("o%u is TMP_OUT\n", i);
3672                 if(shader->output_signature[i].semantic_idx == 0) priv_ctx->vs_output[i] = "TMP_OUT";
3673                 else priv_ctx->vs_output[i] = "TA";
3674             }
3675             else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_PSIZE))
3676             {
3677                 TRACE("o%u is result.pointsize\n", i);
3678                 if(shader->output_signature[i].semantic_idx == 0) priv_ctx->vs_output[i] = "result.pointsize";
3679                 else priv_ctx->vs_output[i] = "TA";
3680             }
3681             else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_COLOR))
3682             {
3683                 TRACE("o%u is result.color.?, idx %u\n", i, shader->output_signature[i].semantic_idx);
3684                 if(shader->output_signature[i].semantic_idx == 0) priv_ctx->vs_output[i] = "result.color.primary";
3685                 else if(shader->output_signature[i].semantic_idx == 1) priv_ctx->vs_output[i] = "result.color.secondary";
3686                 else priv_ctx->vs_output[i] = "TA";
3687             }
3688             else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_TEXCOORD))
3689             {
3690                 TRACE("o%u is %s\n", i, texcoords[shader->output_signature[i].semantic_idx]);
3691                 if(shader->output_signature[i].semantic_idx >= 8) priv_ctx->vs_output[i] = "TA";
3692                 else priv_ctx->vs_output[i] = texcoords[shader->output_signature[i].semantic_idx];
3693             }
3694             else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_FOG))
3695             {
3696                 TRACE("o%u is result.fogcoord\n", i);
3697                 if(shader->output_signature[i].semantic_idx > 0) priv_ctx->vs_output[i] = "TA";
3698                 else priv_ctx->vs_output[i] = "result.fogcoord";
3699             }
3700             else
3701             {
3702                 priv_ctx->vs_output[i] = "TA";
3703             }
3704         }
3705         return;
3706     }
3707
3708     /* Instead of searching for the signature in the signature list, read the one from the current pixel shader.
3709      * Its maybe not the shader where the signature came from, but it is the same signature and faster to find
3710      */
3711     sig = ((IWineD3DPixelShaderImpl *)device->stateBlock->pixelShader)->input_signature;
3712     TRACE("Pixel shader uses declared varyings\n");
3713
3714     /* Map builtin to declared. /dev/null the results by default to the TA temp reg */
3715     for(i = 0; i < 8; i++)
3716     {
3717         priv_ctx->texcrd_output[i] = "TA";
3718     }
3719     priv_ctx->color_output[0] = "TA";
3720     priv_ctx->color_output[1] = "TA";
3721     priv_ctx->fog_output = "TA";
3722
3723     for(i = 0; i < MAX_REG_INPUT; i++)
3724     {
3725         semantic_name = sig[i].semantic_name;
3726         semantic_idx = sig[i].semantic_idx;
3727         reg_idx = sig[i].register_idx;
3728         if(semantic_name == NULL) continue;
3729
3730         /* If a declared input register is not written by builtin arguments, don't write to it.
3731          * GL_NV_vertex_program makes sure the input defaults to 0.0, which is correct with D3D
3732          *
3733          * Don't care about POSITION and PSIZE here - this is a builtin vertex shader, position goes
3734          * to TMP_OUT in any case
3735          */
3736         if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_TEXCOORD))
3737         {
3738             if(semantic_idx < 8) priv_ctx->texcrd_output[semantic_idx] = decl_idx_to_string[reg_idx];
3739         }
3740         else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_COLOR))
3741         {
3742             if(semantic_idx < 2) priv_ctx->color_output[semantic_idx] = decl_idx_to_string[reg_idx];
3743         }
3744         else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_FOG))
3745         {
3746             if(semantic_idx == 0) priv_ctx->fog_output = decl_idx_to_string[reg_idx];
3747         }
3748         else
3749         {
3750             continue;
3751         }
3752
3753         if(strcmp(decl_idx_to_string[reg_idx], "result.color.primary") == 0 ||
3754            strcmp(decl_idx_to_string[reg_idx], "result.color.secondary") == 0)
3755         {
3756             compiled->need_color_unclamp = TRUE;
3757         }
3758     }
3759
3760     /* Map declared to declared */
3761     for(i = 0; i < (sizeof(shader->output_signature) / sizeof(*shader->output_signature)); i++)
3762     {
3763         /* Write unread output to TA to throw them away */
3764         priv_ctx->vs_output[i] = "TA";
3765         semantic_name = shader->output_signature[i].semantic_name;
3766         if(semantic_name == NULL)
3767         {
3768             continue;
3769         }
3770
3771         if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_POSITION) &&
3772            shader->output_signature[i].semantic_idx == 0)
3773         {
3774             priv_ctx->vs_output[i] = "TMP_OUT";
3775             continue;
3776         }
3777         else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_PSIZE) &&
3778            shader->output_signature[i].semantic_idx == 0)
3779         {
3780             priv_ctx->vs_output[i] = "result.pointsize";
3781             continue;
3782         }
3783
3784         for(j = 0; j < MAX_REG_INPUT; j++)
3785         {
3786             if(sig[j].semantic_name == NULL)
3787             {
3788                 continue;
3789             }
3790
3791             if(strcmp(sig[j].semantic_name, semantic_name) == 0 &&
3792                sig[j].semantic_idx == shader->output_signature[i].semantic_idx)
3793             {
3794                 priv_ctx->vs_output[i] = decl_idx_to_string[sig[j].register_idx];
3795
3796                 if(strcmp(priv_ctx->vs_output[i], "result.color.primary") == 0 ||
3797                    strcmp(priv_ctx->vs_output[i], "result.color.secondary") == 0)
3798                 {
3799                     compiled->need_color_unclamp = TRUE;
3800                 }
3801             }
3802         }
3803     }
3804 }
3805
3806 /* GL locking is done by the caller */
3807 static GLuint shader_arb_generate_vshader(IWineD3DVertexShaderImpl *This, struct wined3d_shader_buffer *buffer,
3808         const struct arb_vs_compile_args *args, struct arb_vs_compiled_shader *compiled)
3809 {
3810     const shader_reg_maps *reg_maps = &This->baseShader.reg_maps;
3811     CONST DWORD *function = This->baseShader.function;
3812     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *)This->baseShader.device;
3813     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
3814     const local_constant *lconst;
3815     GLuint ret;
3816     DWORD next_local, *lconst_map = local_const_mapping((IWineD3DBaseShaderImpl *) This);
3817     struct shader_arb_ctx_priv priv_ctx;
3818     unsigned int i;
3819     GLint errPos;
3820
3821     memset(&priv_ctx, 0, sizeof(priv_ctx));
3822     priv_ctx.cur_vs_args = args;
3823     list_init(&priv_ctx.control_frames);
3824     init_output_registers(This, args->ps_signature, &priv_ctx, compiled);
3825
3826     /*  Create the hw ARB shader */
3827     shader_addline(buffer, "!!ARBvp1.0\n");
3828
3829     /* Always enable the NV extension if available. Unlike fragment shaders, there is no
3830      * mesurable performance penalty, and we can always make use of it for clipplanes.
3831      */
3832     if(GL_SUPPORT(NV_VERTEX_PROGRAM3)) {
3833         shader_addline(buffer, "OPTION NV_vertex_program3;\n");
3834         priv_ctx.target_version = NV3;
3835         shader_addline(buffer, "ADDRESS aL;\n");
3836     } else if(GL_SUPPORT(NV_VERTEX_PROGRAM2_OPTION)) {
3837         shader_addline(buffer, "OPTION NV_vertex_program2;\n");
3838         priv_ctx.target_version = NV2;
3839         shader_addline(buffer, "ADDRESS aL;\n");
3840     } else {
3841         priv_ctx.target_version = ARB;
3842     }
3843
3844     shader_addline(buffer, "TEMP TMP_OUT;\n");
3845     if(need_helper_const(gl_info)) {
3846         shader_addline(buffer, "PARAM helper_const = { 2.0, -1.0, %d.0, 0.0 };\n", This->rel_offset);
3847     }
3848     if(need_mova_const((IWineD3DBaseShader *) This, gl_info)) {
3849         shader_addline(buffer, "PARAM mova_const = { 0.5, 0.0, 2.0, 1.0 };\n");
3850         shader_addline(buffer, "TEMP A0_SHADOW;\n");
3851     }
3852
3853     shader_addline(buffer, "TEMP TA;\n");
3854
3855     /* Base Declarations */
3856     next_local = shader_generate_arb_declarations( (IWineD3DBaseShader*) This, reg_maps, buffer, &GLINFO_LOCATION,
3857             lconst_map, &priv_ctx.vs_clipplanes, &priv_ctx);
3858
3859     for(i = 0; i < MAX_CONST_I; i++)
3860     {
3861         compiled->int_consts[i] = WINED3D_CONST_NUM_UNUSED;
3862         if(reg_maps->integer_constants & (1 << i) && priv_ctx.target_version >= NV2)
3863         {
3864             const DWORD *control_values = find_loop_control_values((IWineD3DBaseShaderImpl *) This, i);
3865
3866             if(control_values)
3867             {
3868                 shader_addline(buffer, "PARAM I%u = {%u, %u, %u, -1};\n", i,
3869                                 control_values[0], control_values[1], control_values[2]);
3870             }
3871             else
3872             {
3873                 compiled->int_consts[i] = next_local;
3874                 compiled->num_int_consts++;
3875                 shader_addline(buffer, "PARAM I%u = program.local[%u];\n", i, next_local++);
3876             }
3877         }
3878     }
3879
3880     /* We need a constant to fixup the final position */
3881     shader_addline(buffer, "PARAM posFixup = program.local[%u];\n", next_local);
3882     compiled->pos_fixup = next_local++;
3883
3884     /* Initialize output parameters. GL_ARB_vertex_program does not require special initialization values
3885      * for output parameters. D3D in theory does not do that either, but some applications depend on a
3886      * proper initialization of the secondary color, and programs using the fixed function pipeline without
3887      * a replacement shader depend on the texcoord.w being set properly.
3888      *
3889      * GL_NV_vertex_program defines that all output values are initialized to {0.0, 0.0, 0.0, 1.0}. This
3890      * assertion is in effect even when using GL_ARB_vertex_program without any NV specific additions. So
3891      * skip this if NV_vertex_program is supported. Otherwise, initialize the secondary color. For the tex-
3892      * coords, we have a flag in the opengl caps. Many cards do not require the texcoord being set, and
3893      * this can eat a number of instructions, so skip it unless this cap is set as well
3894      */
3895     if(!GL_SUPPORT(NV_VERTEX_PROGRAM)) {
3896         shader_addline(buffer, "MOV result.color.secondary, -helper_const.wwwy;\n");
3897
3898         if ((GLINFO_LOCATION).quirks & WINED3D_QUIRK_SET_TEXCOORD_W && !device->frag_pipe->ffp_proj_control)
3899         {
3900             int i;
3901             for(i = 0; i < min(8, MAX_REG_TEXCRD); i++) {
3902                 if(This->baseShader.reg_maps.texcoord_mask[i] != 0 &&
3903                 This->baseShader.reg_maps.texcoord_mask[i] != WINED3DSP_WRITEMASK_ALL) {
3904                     shader_addline(buffer, "MOV result.texcoord[%u].w, -helper_const.y;\n", i);
3905                 }
3906             }
3907         }
3908     }
3909
3910     /* The shader starts with the main function */
3911     priv_ctx.in_main_func = TRUE;
3912     /* Base Shader Body */
3913     shader_generate_main((IWineD3DBaseShader *)This, buffer, reg_maps, function, &priv_ctx);
3914
3915     if(!priv_ctx.footer_written) vshader_add_footer(This, buffer, args, &priv_ctx);
3916
3917     shader_addline(buffer, "END\n");
3918
3919     /* TODO: change to resource.glObjectHandle or something like that */
3920     GL_EXTCALL(glGenProgramsARB(1, &ret));
3921
3922     TRACE("Creating a hw vertex shader, prg=%d\n", ret);
3923     GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, ret));
3924
3925     TRACE("Created hw vertex shader, prg=%d\n", ret);
3926     /* Create the program and check for errors */
3927     GL_EXTCALL(glProgramStringARB(GL_VERTEX_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
3928                buffer->bsize, buffer->buffer));
3929     checkGLcall("glProgramStringARB()");
3930
3931     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &errPos);
3932     if (errPos != -1)
3933     {
3934         FIXME("HW VertexShader Error at position %d: %s\n",
3935               errPos, debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
3936         ret = -1;
3937     }
3938     else
3939     {
3940         GLint native;
3941
3942         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
3943         checkGLcall("glGetProgramivARB()");
3944         if (!native) WARN("Program exceeds native resource limits.\n");
3945
3946         /* Load immediate constants */
3947         if(lconst_map) {
3948             LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry) {
3949                 const float *value = (const float *)lconst->value;
3950                 GL_EXTCALL(glProgramLocalParameter4fvARB(GL_VERTEX_PROGRAM_ARB, lconst_map[lconst->idx], value));
3951             }
3952         }
3953     }
3954     HeapFree(GetProcessHeap(), 0, lconst_map);
3955
3956     return ret;
3957 }
3958
3959 /* GL locking is done by the caller */
3960 static struct arb_ps_compiled_shader *find_arb_pshader(IWineD3DPixelShaderImpl *shader, const struct arb_ps_compile_args *args)
3961 {
3962     UINT i;
3963     DWORD new_size;
3964     struct arb_ps_compiled_shader *new_array;
3965     struct wined3d_shader_buffer buffer;
3966     struct arb_pshader_private *shader_data;
3967     GLuint ret;
3968
3969     if(!shader->backend_priv) {
3970         IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *) shader->baseShader.device;
3971         const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
3972         struct shader_arb_priv *priv = device->shader_priv;
3973
3974         shader->backend_priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data));
3975         shader_data = shader->backend_priv;
3976         shader_data->clamp_consts = shader->baseShader.reg_maps.shader_version.major == 1;
3977
3978         if(shader->baseShader.reg_maps.shader_version.major < 3) shader_data->input_signature_idx = ~0;
3979         else shader_data->input_signature_idx = find_input_signature(priv, shader->input_signature);
3980
3981         shader_data->has_signature_idx = TRUE;
3982         TRACE("Shader got assigned input signature index %u\n", shader_data->input_signature_idx);
3983
3984         if (!device->vs_clipping)
3985             shader_data->clipplane_emulation = shader_find_free_input_register(&shader->baseShader.reg_maps,
3986                     GL_LIMITS(texture_stages) - 1);
3987         else
3988             shader_data->clipplane_emulation = ~0U;
3989     }
3990     shader_data = shader->backend_priv;
3991
3992     /* Usually we have very few GL shaders for each d3d shader(just 1 or maybe 2),
3993      * so a linear search is more performant than a hashmap or a binary search
3994      * (cache coherency etc)
3995      */
3996     for(i = 0; i < shader_data->num_gl_shaders; i++) {
3997         if(memcmp(&shader_data->gl_shaders[i].args, args, sizeof(*args)) == 0) {
3998             return &shader_data->gl_shaders[i];
3999         }
4000     }
4001
4002     TRACE("No matching GL shader found, compiling a new shader\n");
4003     if(shader_data->shader_array_size == shader_data->num_gl_shaders) {
4004         if (shader_data->num_gl_shaders)
4005         {
4006             new_size = shader_data->shader_array_size + max(1, shader_data->shader_array_size / 2);
4007             new_array = HeapReAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, shader_data->gl_shaders,
4008                                     new_size * sizeof(*shader_data->gl_shaders));
4009         } else {
4010             new_array = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data->gl_shaders));
4011             new_size = 1;
4012         }
4013
4014         if(!new_array) {
4015             ERR("Out of memory\n");
4016             return 0;
4017         }
4018         shader_data->gl_shaders = new_array;
4019         shader_data->shader_array_size = new_size;
4020     }
4021
4022     shader_data->gl_shaders[shader_data->num_gl_shaders].args = *args;
4023
4024     pixelshader_update_samplers(&shader->baseShader.reg_maps,
4025             ((IWineD3DDeviceImpl *)shader->baseShader.device)->stateBlock->textures);
4026
4027     if (!shader_buffer_init(&buffer))
4028     {
4029         ERR("Failed to initialize shader buffer.\n");
4030         return 0;
4031     }
4032
4033     ret = shader_arb_generate_pshader(shader, &buffer, args,
4034                                       &shader_data->gl_shaders[shader_data->num_gl_shaders]);
4035     shader_buffer_free(&buffer);
4036     shader_data->gl_shaders[shader_data->num_gl_shaders].prgId = ret;
4037
4038     return &shader_data->gl_shaders[shader_data->num_gl_shaders++];
4039 }
4040
4041 static inline BOOL vs_args_equal(const struct arb_vs_compile_args *stored, const struct arb_vs_compile_args *new,
4042                                  const DWORD use_map, BOOL skip_int) {
4043     if((stored->super.swizzle_map & use_map) != new->super.swizzle_map) return FALSE;
4044     if(stored->super.fog_src != new->super.fog_src) return FALSE;
4045     if(stored->boolclip_compare != new->boolclip_compare) return FALSE;
4046     if(stored->ps_signature != new->ps_signature) return FALSE;
4047     if(stored->vertex_samplers_compare != new->vertex_samplers_compare) return FALSE;
4048     if(skip_int) return TRUE;
4049
4050     return memcmp(stored->loop_ctrl, new->loop_ctrl, sizeof(stored->loop_ctrl)) == 0;
4051 }
4052
4053 static struct arb_vs_compiled_shader *find_arb_vshader(IWineD3DVertexShaderImpl *shader, const struct arb_vs_compile_args *args)
4054 {
4055     UINT i;
4056     DWORD new_size;
4057     struct arb_vs_compiled_shader *new_array;
4058     DWORD use_map = ((IWineD3DDeviceImpl *)shader->baseShader.device)->strided_streams.use_map;
4059     struct wined3d_shader_buffer buffer;
4060     struct arb_vshader_private *shader_data;
4061     GLuint ret;
4062     const struct wined3d_gl_info *gl_info = &((IWineD3DDeviceImpl *)shader->baseShader.device)->adapter->gl_info;
4063
4064     if(!shader->backend_priv) {
4065         shader->backend_priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data));
4066     }
4067     shader_data = shader->backend_priv;
4068
4069     /* Usually we have very few GL shaders for each d3d shader(just 1 or maybe 2),
4070      * so a linear search is more performant than a hashmap or a binary search
4071      * (cache coherency etc)
4072      */
4073     for(i = 0; i < shader_data->num_gl_shaders; i++) {
4074         if(vs_args_equal(&shader_data->gl_shaders[i].args, args, use_map, GL_SUPPORT(NV_VERTEX_PROGRAM2_OPTION))) {
4075             return &shader_data->gl_shaders[i];
4076         }
4077     }
4078
4079     TRACE("No matching GL shader found, compiling a new shader\n");
4080
4081     if(shader_data->shader_array_size == shader_data->num_gl_shaders) {
4082         if (shader_data->num_gl_shaders)
4083         {
4084             new_size = shader_data->shader_array_size + max(1, shader_data->shader_array_size / 2);
4085             new_array = HeapReAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, shader_data->gl_shaders,
4086                                     new_size * sizeof(*shader_data->gl_shaders));
4087         } else {
4088             new_array = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data->gl_shaders));
4089             new_size = 1;
4090         }
4091
4092         if(!new_array) {
4093             ERR("Out of memory\n");
4094             return 0;
4095         }
4096         shader_data->gl_shaders = new_array;
4097         shader_data->shader_array_size = new_size;
4098     }
4099
4100     shader_data->gl_shaders[shader_data->num_gl_shaders].args = *args;
4101
4102     if (!shader_buffer_init(&buffer))
4103     {
4104         ERR("Failed to initialize shader buffer.\n");
4105         return 0;
4106     }
4107
4108     ret = shader_arb_generate_vshader(shader, &buffer, args,
4109             &shader_data->gl_shaders[shader_data->num_gl_shaders]);
4110     shader_buffer_free(&buffer);
4111     shader_data->gl_shaders[shader_data->num_gl_shaders].prgId = ret;
4112
4113     return &shader_data->gl_shaders[shader_data->num_gl_shaders++];
4114 }
4115
4116 static inline void find_arb_ps_compile_args(IWineD3DPixelShaderImpl *shader, IWineD3DStateBlockImpl *stateblock,
4117         struct arb_ps_compile_args *args)
4118 {
4119     int i;
4120     WORD int_skip;
4121     const struct wined3d_gl_info *gl_info = &((IWineD3DDeviceImpl *)shader->baseShader.device)->adapter->gl_info;
4122     find_ps_compile_args(shader, stateblock, &args->super);
4123
4124     /* This forces all local boolean constants to 1 to make them stateblock independent */
4125     args->bools = shader->baseShader.reg_maps.local_bool_consts;
4126
4127     for(i = 0; i < MAX_CONST_B; i++)
4128     {
4129         if(stateblock->pixelShaderConstantB[i]) args->bools |= ( 1 << i);
4130     }
4131
4132     /* Only enable the clip plane emulation KIL if at least one clipplane is enabled. The KIL instruction
4133      * is quite expensive because it forces the driver to disable early Z discards. It is cheaper to
4134      * duplicate the shader than have a no-op KIL instruction in every shader
4135      */
4136     if((!((IWineD3DDeviceImpl *) shader->baseShader.device)->vs_clipping) && use_vs(stateblock) &&
4137        stateblock->renderState[WINED3DRS_CLIPPING] && stateblock->renderState[WINED3DRS_CLIPPLANEENABLE])
4138     {
4139         args->clip = 1;
4140     }
4141     else
4142     {
4143         args->clip = 0;
4144     }
4145
4146     /* Skip if unused or local, or supported natively */
4147     int_skip = ~shader->baseShader.reg_maps.integer_constants | shader->baseShader.reg_maps.local_int_consts;
4148     if(int_skip == 0xffff || GL_SUPPORT(NV_FRAGMENT_PROGRAM_OPTION))
4149     {
4150         memset(&args->loop_ctrl, 0, sizeof(args->loop_ctrl));
4151         return;
4152     }
4153
4154     for(i = 0; i < MAX_CONST_I; i++)
4155     {
4156         if(int_skip & (1 << i))
4157         {
4158             args->loop_ctrl[i][0] = 0;
4159             args->loop_ctrl[i][1] = 0;
4160             args->loop_ctrl[i][2] = 0;
4161         }
4162         else
4163         {
4164             args->loop_ctrl[i][0] = stateblock->pixelShaderConstantI[i * 4];
4165             args->loop_ctrl[i][1] = stateblock->pixelShaderConstantI[i * 4 + 1];
4166             args->loop_ctrl[i][2] = stateblock->pixelShaderConstantI[i * 4 + 2];
4167         }
4168     }
4169 }
4170
4171 static inline void find_arb_vs_compile_args(IWineD3DVertexShaderImpl *shader, IWineD3DStateBlockImpl *stateblock,
4172         struct arb_vs_compile_args *args)
4173 {
4174     int i;
4175     WORD int_skip;
4176     IWineD3DDeviceImpl *dev = (IWineD3DDeviceImpl *)shader->baseShader.device;
4177     const struct wined3d_gl_info *gl_info = &dev->adapter->gl_info;
4178     find_vs_compile_args(shader, stateblock, &args->super);
4179
4180     args->boolclip_compare = 0;
4181     if(use_ps(stateblock))
4182     {
4183         IWineD3DPixelShaderImpl *ps = (IWineD3DPixelShaderImpl *) stateblock->pixelShader;
4184         struct arb_pshader_private *shader_priv = ps->backend_priv;
4185         args->ps_signature = shader_priv->input_signature_idx;
4186
4187         args->boolclip.clip_control[0] = shader_priv->clipplane_emulation + 1;
4188     }
4189     else
4190     {
4191         args->ps_signature = ~0;
4192         if(!dev->vs_clipping)
4193         {
4194             args->boolclip.clip_control[0] = ffp_clip_emul(stateblock) ? GL_LIMITS(texture_stages) : 0;
4195         }
4196         /* Otherwise: Setting boolclip_compare set clip_control[0] to 0 */
4197     }
4198
4199     if(args->boolclip.clip_control[0])
4200     {
4201         if(stateblock->renderState[WINED3DRS_CLIPPING])
4202         {
4203             args->boolclip.clip_control[1] = stateblock->renderState[WINED3DRS_CLIPPLANEENABLE];
4204         }
4205         /* clip_control[1] was set to 0 by setting boolclip_compare to 0 */
4206     }
4207
4208     /* This forces all local boolean constants to 1 to make them stateblock independent */
4209     args->boolclip.bools = shader->baseShader.reg_maps.local_bool_consts;
4210     /* TODO: Figure out if it would be better to store bool constants as bitmasks in the stateblock */
4211     for(i = 0; i < MAX_CONST_B; i++)
4212     {
4213         if(stateblock->vertexShaderConstantB[i]) args->boolclip.bools |= ( 1 << i);
4214     }
4215
4216     args->vertex_samplers[0] = dev->texUnitMap[MAX_FRAGMENT_SAMPLERS + 0];
4217     args->vertex_samplers[1] = dev->texUnitMap[MAX_FRAGMENT_SAMPLERS + 1];
4218     args->vertex_samplers[2] = dev->texUnitMap[MAX_FRAGMENT_SAMPLERS + 2];
4219     args->vertex_samplers[3] = 0;
4220
4221     /* Skip if unused or local */
4222     int_skip = ~shader->baseShader.reg_maps.integer_constants | shader->baseShader.reg_maps.local_int_consts;
4223     if(int_skip == 0xffff || GL_SUPPORT(NV_VERTEX_PROGRAM2_OPTION)) /* This is about flow control, not clipping */
4224     {
4225         memset(&args->loop_ctrl, 0, sizeof(args->loop_ctrl));
4226         return;
4227     }
4228
4229     for(i = 0; i < MAX_CONST_I; i++)
4230     {
4231         if(int_skip & (1 << i))
4232         {
4233             args->loop_ctrl[i][0] = 0;
4234             args->loop_ctrl[i][1] = 0;
4235             args->loop_ctrl[i][2] = 0;
4236         }
4237         else
4238         {
4239             args->loop_ctrl[i][0] = stateblock->vertexShaderConstantI[i * 4];
4240             args->loop_ctrl[i][1] = stateblock->vertexShaderConstantI[i * 4 + 1];
4241             args->loop_ctrl[i][2] = stateblock->vertexShaderConstantI[i * 4 + 2];
4242         }
4243     }
4244 }
4245
4246 /* GL locking is done by the caller */
4247 static void shader_arb_select(const struct wined3d_context *context, BOOL usePS, BOOL useVS)
4248 {
4249     IWineD3DDeviceImpl *This = ((IWineD3DSurfaceImpl *)context->surface)->resource.wineD3DDevice;
4250     struct shader_arb_priv *priv = This->shader_priv;
4251     const struct wined3d_gl_info *gl_info = context->gl_info;
4252     int i;
4253
4254     /* Deal with pixel shaders first so the vertex shader arg function has the input signature ready */
4255     if (usePS) {
4256         struct arb_ps_compile_args compile_args;
4257         struct arb_ps_compiled_shader *compiled;
4258         IWineD3DPixelShaderImpl *ps = (IWineD3DPixelShaderImpl *) This->stateBlock->pixelShader;
4259
4260         TRACE("Using pixel shader %p\n", This->stateBlock->pixelShader);
4261         find_arb_ps_compile_args(ps, This->stateBlock, &compile_args);
4262         compiled = find_arb_pshader(ps, &compile_args);
4263         priv->current_fprogram_id = compiled->prgId;
4264         priv->compiled_fprog = compiled;
4265
4266         /* Bind the fragment program */
4267         GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, priv->current_fprogram_id));
4268         checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, priv->current_fprogram_id);");
4269
4270         if(!priv->use_arbfp_fixed_func) {
4271             /* Enable OpenGL fragment programs */
4272             glEnable(GL_FRAGMENT_PROGRAM_ARB);
4273             checkGLcall("glEnable(GL_FRAGMENT_PROGRAM_ARB);");
4274         }
4275         TRACE("(%p) : Bound fragment program %u and enabled GL_FRAGMENT_PROGRAM_ARB\n", This, priv->current_fprogram_id);
4276
4277         /* Pixel Shader 1.x constants are clamped to [-1;1], Pixel Shader 2.0 constants are not. If switching between
4278          * a 1.x and newer shader, reload the first 8 constants
4279          */
4280         if(priv->last_ps_const_clamped != ((struct arb_pshader_private *) ps->backend_priv)->clamp_consts)
4281         {
4282             priv->last_ps_const_clamped = ((struct arb_pshader_private *) ps->backend_priv)->clamp_consts;
4283             This->highest_dirty_ps_const = max(This->highest_dirty_ps_const, 8);
4284             for(i = 0; i < 8; i++)
4285             {
4286                 context->pshader_const_dirty[i] = 1;
4287             }
4288             /* Also takes care of loading local constants */
4289             shader_arb_load_constants(context, TRUE, FALSE);
4290         }
4291         else
4292         {
4293             shader_arb_ps_local_constants(This);
4294         }
4295
4296         /* Force constant reloading for the NP2 fixup (see comment in shader_glsl_select for more info) */
4297         if (compiled->np2fixup_info.super.active)
4298             shader_arb_load_np2fixup_constants((IWineD3DDevice *)This, usePS, useVS);
4299     } else if(GL_SUPPORT(ARB_FRAGMENT_PROGRAM) && !priv->use_arbfp_fixed_func) {
4300         /* Disable only if we're not using arbfp fixed function fragment processing. If this is used,
4301         * keep GL_FRAGMENT_PROGRAM_ARB enabled, and the fixed function pipeline will bind the fixed function
4302         * replacement shader
4303         */
4304         glDisable(GL_FRAGMENT_PROGRAM_ARB);
4305         checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
4306         priv->current_fprogram_id = 0;
4307     }
4308
4309     if (useVS) {
4310         struct arb_vs_compile_args compile_args;
4311         struct arb_vs_compiled_shader *compiled;
4312         IWineD3DVertexShaderImpl *vs = (IWineD3DVertexShaderImpl *) This->stateBlock->vertexShader;
4313
4314         TRACE("Using vertex shader %p\n", This->stateBlock->vertexShader);
4315         find_arb_vs_compile_args(vs, This->stateBlock, &compile_args);
4316         compiled = find_arb_vshader(vs, &compile_args);
4317         priv->current_vprogram_id = compiled->prgId;
4318         priv->compiled_vprog = compiled;
4319
4320         /* Bind the vertex program */
4321         GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->current_vprogram_id));
4322         checkGLcall("glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->current_vprogram_id);");
4323
4324         /* Enable OpenGL vertex programs */
4325         glEnable(GL_VERTEX_PROGRAM_ARB);
4326         checkGLcall("glEnable(GL_VERTEX_PROGRAM_ARB);");
4327         TRACE("(%p) : Bound vertex program %u and enabled GL_VERTEX_PROGRAM_ARB\n", This, priv->current_vprogram_id);
4328         shader_arb_vs_local_constants(This);
4329
4330         if(priv->last_vs_color_unclamp != compiled->need_color_unclamp) {
4331             priv->last_vs_color_unclamp = compiled->need_color_unclamp;
4332
4333             if (GL_SUPPORT(ARB_COLOR_BUFFER_FLOAT)) {
4334                 GL_EXTCALL(glClampColorARB(GL_CLAMP_VERTEX_COLOR_ARB, !compiled->need_color_unclamp));
4335                 checkGLcall("glClampColorARB");
4336             } else {
4337                 FIXME("vertex color clamp needs to be changed, but extension not supported.\n");
4338             }
4339         }
4340     } else if(GL_SUPPORT(ARB_VERTEX_PROGRAM)) {
4341         priv->current_vprogram_id = 0;
4342         glDisable(GL_VERTEX_PROGRAM_ARB);
4343         checkGLcall("glDisable(GL_VERTEX_PROGRAM_ARB)");
4344     }
4345 }
4346
4347 /* GL locking is done by the caller */
4348 static void shader_arb_select_depth_blt(IWineD3DDevice *iface, enum tex_types tex_type) {
4349     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
4350     struct shader_arb_priv *priv = This->shader_priv;
4351     GLuint *blt_fprogram = &priv->depth_blt_fprogram_id[tex_type];
4352     const struct wined3d_gl_info *gl_info = &This->adapter->gl_info;
4353
4354     if (!priv->depth_blt_vprogram_id) priv->depth_blt_vprogram_id = create_arb_blt_vertex_program(gl_info);
4355     GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->depth_blt_vprogram_id));
4356     glEnable(GL_VERTEX_PROGRAM_ARB);
4357
4358     if (!*blt_fprogram) *blt_fprogram = create_arb_blt_fragment_program(gl_info, tex_type);
4359     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, *blt_fprogram));
4360     glEnable(GL_FRAGMENT_PROGRAM_ARB);
4361 }
4362
4363 /* GL locking is done by the caller */
4364 static void shader_arb_deselect_depth_blt(IWineD3DDevice *iface) {
4365     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
4366     struct shader_arb_priv *priv = This->shader_priv;
4367     const struct wined3d_gl_info *gl_info = &This->adapter->gl_info;
4368
4369     if (priv->current_vprogram_id) {
4370         GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->current_vprogram_id));
4371         checkGLcall("glBindProgramARB(GL_VERTEX_PROGRAM_ARB, vertexShader->prgId);");
4372
4373         TRACE("(%p) : Bound vertex program %u and enabled GL_VERTEX_PROGRAM_ARB\n", This, priv->current_vprogram_id);
4374     } else {
4375         glDisable(GL_VERTEX_PROGRAM_ARB);
4376         checkGLcall("glDisable(GL_VERTEX_PROGRAM_ARB)");
4377     }
4378
4379     if (priv->current_fprogram_id) {
4380         GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, priv->current_fprogram_id));
4381         checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, pixelShader->prgId);");
4382
4383         TRACE("(%p) : Bound fragment program %u and enabled GL_FRAGMENT_PROGRAM_ARB\n", This, priv->current_fprogram_id);
4384     } else if(!priv->use_arbfp_fixed_func) {
4385         glDisable(GL_FRAGMENT_PROGRAM_ARB);
4386         checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
4387     }
4388 }
4389
4390 static void shader_arb_destroy(IWineD3DBaseShader *iface) {
4391     IWineD3DBaseShaderImpl *baseShader = (IWineD3DBaseShaderImpl *) iface;
4392     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *)baseShader->baseShader.device;
4393     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4394
4395     if (shader_is_pshader_version(baseShader->baseShader.reg_maps.shader_version.type))
4396     {
4397         IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *) iface;
4398         struct arb_pshader_private *shader_data = This->backend_priv;
4399         UINT i;
4400
4401         if(!shader_data) return; /* This can happen if a shader was never compiled */
4402         ENTER_GL();
4403
4404         if(shader_data->num_gl_shaders) ActivateContext(device, NULL, CTXUSAGE_RESOURCELOAD);
4405
4406         for(i = 0; i < shader_data->num_gl_shaders; i++) {
4407             GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId));
4408             checkGLcall("GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId))");
4409         }
4410         LEAVE_GL();
4411         HeapFree(GetProcessHeap(), 0, shader_data->gl_shaders);
4412         HeapFree(GetProcessHeap(), 0, shader_data);
4413         This->backend_priv = NULL;
4414     } else {
4415         IWineD3DVertexShaderImpl *This = (IWineD3DVertexShaderImpl *) iface;
4416         struct arb_vshader_private *shader_data = This->backend_priv;
4417         UINT i;
4418
4419         if(!shader_data) return; /* This can happen if a shader was never compiled */
4420         ENTER_GL();
4421
4422         if(shader_data->num_gl_shaders) ActivateContext(device, NULL, CTXUSAGE_RESOURCELOAD);
4423
4424         for(i = 0; i < shader_data->num_gl_shaders; i++) {
4425             GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId));
4426             checkGLcall("GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId))");
4427         }
4428         LEAVE_GL();
4429         HeapFree(GetProcessHeap(), 0, shader_data->gl_shaders);
4430         HeapFree(GetProcessHeap(), 0, shader_data);
4431         This->backend_priv = NULL;
4432     }
4433 }
4434
4435 static int sig_tree_compare(const void *key, const struct wine_rb_entry *entry)
4436 {
4437     struct ps_signature *e = WINE_RB_ENTRY_VALUE(entry, struct ps_signature, entry);
4438     return compare_sig(key, e->sig);
4439 }
4440
4441 struct wine_rb_functions sig_tree_functions =
4442 {
4443     wined3d_rb_alloc,
4444     wined3d_rb_realloc,
4445     wined3d_rb_free,
4446     sig_tree_compare
4447 };
4448
4449 static HRESULT shader_arb_alloc(IWineD3DDevice *iface) {
4450     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
4451     struct shader_arb_priv *priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*priv));
4452     if(wine_rb_init(&priv->signature_tree, &sig_tree_functions) == -1)
4453     {
4454         ERR("RB tree init failed\n");
4455         HeapFree(GetProcessHeap(), 0, priv);
4456         return E_OUTOFMEMORY;
4457     }
4458     This->shader_priv = priv;
4459     return WINED3D_OK;
4460 }
4461
4462 static void release_signature(struct wine_rb_entry *entry, void *context)
4463 {
4464     struct ps_signature *sig = WINE_RB_ENTRY_VALUE(entry, struct ps_signature, entry);
4465     int i;
4466     for(i = 0; i < MAX_REG_INPUT; i++)
4467     {
4468         HeapFree(GetProcessHeap(), 0, (char *) sig->sig[i].semantic_name);
4469     }
4470     HeapFree(GetProcessHeap(), 0, sig->sig);
4471     HeapFree(GetProcessHeap(), 0, sig);
4472 }
4473
4474 /* Context activation is done by the caller. */
4475 static void shader_arb_free(IWineD3DDevice *iface) {
4476     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
4477     const struct wined3d_gl_info *gl_info = &This->adapter->gl_info;
4478     struct shader_arb_priv *priv = This->shader_priv;
4479     int i;
4480
4481     ENTER_GL();
4482     if(priv->depth_blt_vprogram_id) {
4483         GL_EXTCALL(glDeleteProgramsARB(1, &priv->depth_blt_vprogram_id));
4484     }
4485     for (i = 0; i < tex_type_count; ++i) {
4486         if (priv->depth_blt_fprogram_id[i]) {
4487             GL_EXTCALL(glDeleteProgramsARB(1, &priv->depth_blt_fprogram_id[i]));
4488         }
4489     }
4490     LEAVE_GL();
4491
4492     wine_rb_destroy(&priv->signature_tree, release_signature, NULL);
4493     HeapFree(GetProcessHeap(), 0, This->shader_priv);
4494 }
4495
4496 static BOOL shader_arb_dirty_const(IWineD3DDevice *iface) {
4497     return TRUE;
4498 }
4499
4500 static void shader_arb_get_caps(WINED3DDEVTYPE devtype, const struct wined3d_gl_info *gl_info,
4501         struct shader_caps *pCaps)
4502 {
4503     /* We don't have an ARB fixed function pipeline yet, so let the none backend set its caps,
4504      * then overwrite the shader specific ones
4505      */
4506     none_shader_backend.shader_get_caps(devtype, gl_info, pCaps);
4507
4508     if(GL_SUPPORT(ARB_VERTEX_PROGRAM)) {
4509         if(GL_SUPPORT(NV_VERTEX_PROGRAM3))
4510         {
4511             pCaps->VertexShaderVersion = WINED3DVS_VERSION(3,0);
4512             TRACE_(d3d_caps)("Hardware vertex shader version 3.0 enabled (NV_VERTEX_PROGRAM3)\n");
4513         }
4514         else if(GL_LIMITS(vshader_constantsF) >= 256)
4515         {
4516             /* Shader Model 2.0 requires at least 256 vertex shader constants */
4517             pCaps->VertexShaderVersion = WINED3DVS_VERSION(2,0);
4518             TRACE_(d3d_caps)("Hardware vertex shader version 2.0 enabled (ARB_PROGRAM)\n");
4519         }
4520         else
4521         {
4522             pCaps->VertexShaderVersion = WINED3DVS_VERSION(1,1);
4523             TRACE_(d3d_caps)("Hardware vertex shader version 1.1 enabled (ARB_PROGRAM)\n");
4524         }
4525         pCaps->MaxVertexShaderConst = GL_LIMITS(vshader_constantsF);
4526     }
4527
4528     if(GL_SUPPORT(ARB_FRAGMENT_PROGRAM)) {
4529         if(GL_SUPPORT(NV_FRAGMENT_PROGRAM2))
4530         {
4531             pCaps->PixelShaderVersion    = WINED3DPS_VERSION(3,0);
4532             TRACE_(d3d_caps)("Hardware pixel shader version 3.0 enabled (NV_FRAGMENT_PROGRAM2)\n");
4533         }
4534         else if(GL_LIMITS(pshader_constantsF) >= 32)
4535         {
4536             /* Shader Model 2.0 requires at least 32 pixel shader constants */
4537             pCaps->PixelShaderVersion    = WINED3DPS_VERSION(2,0);
4538             TRACE_(d3d_caps)("Hardware pixel shader version 2.0 enabled (ARB_PROGRAM)\n");
4539         }
4540         else
4541         {
4542             pCaps->PixelShaderVersion    = WINED3DPS_VERSION(1,4);
4543             TRACE_(d3d_caps)("Hardware pixel shader version 1.4 enabled (ARB_PROGRAM)\n");
4544         }
4545         pCaps->PixelShader1xMaxValue = 8.0f;
4546         pCaps->MaxPixelShaderConst = GL_LIMITS(pshader_constantsF);
4547     }
4548
4549     pCaps->VSClipping = use_nv_clip(gl_info);
4550 }
4551
4552 static BOOL shader_arb_color_fixup_supported(struct color_fixup_desc fixup)
4553 {
4554     if (TRACE_ON(d3d_shader) && TRACE_ON(d3d))
4555     {
4556         TRACE("Checking support for color_fixup:\n");
4557         dump_color_fixup_desc(fixup);
4558     }
4559
4560     /* We support everything except YUV conversions. */
4561     if (!is_yuv_fixup(fixup))
4562     {
4563         TRACE("[OK]\n");
4564         return TRUE;
4565     }
4566
4567     TRACE("[FAILED]\n");
4568     return FALSE;
4569 }
4570
4571 static void shader_arb_add_instruction_modifiers(const struct wined3d_shader_instruction *ins) {
4572     DWORD shift;
4573     char write_mask[20], regstr[50];
4574     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
4575     BOOL is_color = FALSE;
4576     const struct wined3d_shader_dst_param *dst;
4577
4578     if (!ins->dst_count) return;
4579
4580     dst = &ins->dst[0];
4581     shift = dst->shift;
4582     if(shift == 0) return; /* Saturate alone is handled by the instructions */
4583
4584     shader_arb_get_write_mask(ins, dst, write_mask);
4585     shader_arb_get_register_name(ins, &dst->reg, regstr, &is_color);
4586
4587     /* Generate a line that does the output modifier computation
4588      * FIXME: _SAT vs shift? _SAT alone is already handled in the instructions, if this
4589      * maps problems in e.g. _d4_sat modify shader_arb_get_modifier
4590      */
4591     shader_addline(buffer, "MUL%s %s%s, %s, %s;\n", shader_arb_get_modifier(ins),
4592                    regstr, write_mask, regstr, shift_tab[shift]);
4593 }
4594
4595 static const SHADER_HANDLER shader_arb_instruction_handler_table[WINED3DSIH_TABLE_SIZE] =
4596 {
4597     /* WINED3DSIH_ABS           */ shader_hw_map2gl,
4598     /* WINED3DSIH_ADD           */ shader_hw_map2gl,
4599     /* WINED3DSIH_BEM           */ pshader_hw_bem,
4600     /* WINED3DSIH_BREAK         */ shader_hw_break,
4601     /* WINED3DSIH_BREAKC        */ shader_hw_breakc,
4602     /* WINED3DSIH_BREAKP        */ NULL,
4603     /* WINED3DSIH_CALL          */ shader_hw_call,
4604     /* WINED3DSIH_CALLNZ        */ NULL,
4605     /* WINED3DSIH_CMP           */ pshader_hw_cmp,
4606     /* WINED3DSIH_CND           */ pshader_hw_cnd,
4607     /* WINED3DSIH_CRS           */ shader_hw_map2gl,
4608     /* WINED3DSIH_DCL           */ NULL,
4609     /* WINED3DSIH_DEF           */ NULL,
4610     /* WINED3DSIH_DEFB          */ NULL,
4611     /* WINED3DSIH_DEFI          */ NULL,
4612     /* WINED3DSIH_DP2ADD        */ pshader_hw_dp2add,
4613     /* WINED3DSIH_DP3           */ shader_hw_map2gl,
4614     /* WINED3DSIH_DP4           */ shader_hw_map2gl,
4615     /* WINED3DSIH_DST           */ shader_hw_map2gl,
4616     /* WINED3DSIH_DSX           */ shader_hw_map2gl,
4617     /* WINED3DSIH_DSY           */ shader_hw_dsy,
4618     /* WINED3DSIH_ELSE          */ shader_hw_else,
4619     /* WINED3DSIH_ENDIF         */ shader_hw_endif,
4620     /* WINED3DSIH_ENDLOOP       */ shader_hw_endloop,
4621     /* WINED3DSIH_ENDREP        */ shader_hw_endrep,
4622     /* WINED3DSIH_EXP           */ shader_hw_scalar_op,
4623     /* WINED3DSIH_EXPP          */ shader_hw_scalar_op,
4624     /* WINED3DSIH_FRC           */ shader_hw_map2gl,
4625     /* WINED3DSIH_IF            */ NULL /* Hardcoded into the shader */,
4626     /* WINED3DSIH_IFC           */ shader_hw_ifc,
4627     /* WINED3DSIH_LABEL         */ shader_hw_label,
4628     /* WINED3DSIH_LIT           */ shader_hw_map2gl,
4629     /* WINED3DSIH_LOG           */ shader_hw_log_pow,
4630     /* WINED3DSIH_LOGP          */ shader_hw_log_pow,
4631     /* WINED3DSIH_LOOP          */ shader_hw_loop,
4632     /* WINED3DSIH_LRP           */ shader_hw_lrp,
4633     /* WINED3DSIH_M3x2          */ shader_hw_mnxn,
4634     /* WINED3DSIH_M3x3          */ shader_hw_mnxn,
4635     /* WINED3DSIH_M3x4          */ shader_hw_mnxn,
4636     /* WINED3DSIH_M4x3          */ shader_hw_mnxn,
4637     /* WINED3DSIH_M4x4          */ shader_hw_mnxn,
4638     /* WINED3DSIH_MAD           */ shader_hw_map2gl,
4639     /* WINED3DSIH_MAX           */ shader_hw_map2gl,
4640     /* WINED3DSIH_MIN           */ shader_hw_map2gl,
4641     /* WINED3DSIH_MOV           */ shader_hw_mov,
4642     /* WINED3DSIH_MOVA          */ shader_hw_mov,
4643     /* WINED3DSIH_MUL           */ shader_hw_map2gl,
4644     /* WINED3DSIH_NOP           */ shader_hw_nop,
4645     /* WINED3DSIH_NRM           */ shader_hw_nrm,
4646     /* WINED3DSIH_PHASE         */ NULL,
4647     /* WINED3DSIH_POW           */ shader_hw_log_pow,
4648     /* WINED3DSIH_RCP           */ shader_hw_scalar_op,
4649     /* WINED3DSIH_REP           */ shader_hw_rep,
4650     /* WINED3DSIH_RET           */ shader_hw_ret,
4651     /* WINED3DSIH_RSQ           */ shader_hw_scalar_op,
4652     /* WINED3DSIH_SETP          */ NULL,
4653     /* WINED3DSIH_SGE           */ shader_hw_map2gl,
4654     /* WINED3DSIH_SGN           */ shader_hw_sgn,
4655     /* WINED3DSIH_SINCOS        */ shader_hw_sincos,
4656     /* WINED3DSIH_SLT           */ shader_hw_map2gl,
4657     /* WINED3DSIH_SUB           */ shader_hw_map2gl,
4658     /* WINED3DSIH_TEX           */ pshader_hw_tex,
4659     /* WINED3DSIH_TEXBEM        */ pshader_hw_texbem,
4660     /* WINED3DSIH_TEXBEML       */ pshader_hw_texbem,
4661     /* WINED3DSIH_TEXCOORD      */ pshader_hw_texcoord,
4662     /* WINED3DSIH_TEXDEPTH      */ pshader_hw_texdepth,
4663     /* WINED3DSIH_TEXDP3        */ pshader_hw_texdp3,
4664     /* WINED3DSIH_TEXDP3TEX     */ pshader_hw_texdp3tex,
4665     /* WINED3DSIH_TEXKILL       */ pshader_hw_texkill,
4666     /* WINED3DSIH_TEXLDD        */ shader_hw_texldd,
4667     /* WINED3DSIH_TEXLDL        */ shader_hw_texldl,
4668     /* WINED3DSIH_TEXM3x2DEPTH  */ pshader_hw_texm3x2depth,
4669     /* WINED3DSIH_TEXM3x2PAD    */ pshader_hw_texm3x2pad,
4670     /* WINED3DSIH_TEXM3x2TEX    */ pshader_hw_texm3x2tex,
4671     /* WINED3DSIH_TEXM3x3       */ pshader_hw_texm3x3,
4672     /* WINED3DSIH_TEXM3x3DIFF   */ NULL,
4673     /* WINED3DSIH_TEXM3x3PAD    */ pshader_hw_texm3x3pad,
4674     /* WINED3DSIH_TEXM3x3SPEC   */ pshader_hw_texm3x3spec,
4675     /* WINED3DSIH_TEXM3x3TEX    */ pshader_hw_texm3x3tex,
4676     /* WINED3DSIH_TEXM3x3VSPEC  */ pshader_hw_texm3x3vspec,
4677     /* WINED3DSIH_TEXREG2AR     */ pshader_hw_texreg2ar,
4678     /* WINED3DSIH_TEXREG2GB     */ pshader_hw_texreg2gb,
4679     /* WINED3DSIH_TEXREG2RGB    */ pshader_hw_texreg2rgb,
4680 };
4681
4682 static inline BOOL get_bool_const(const struct wined3d_shader_instruction *ins, IWineD3DBaseShaderImpl *This, DWORD idx)
4683 {
4684     BOOL vshader = shader_is_vshader_version(This->baseShader.reg_maps.shader_version.type);
4685     WORD bools = 0;
4686     WORD flag = (1 << idx);
4687     const local_constant *constant;
4688     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
4689
4690     if(This->baseShader.reg_maps.local_bool_consts & flag)
4691     {
4692         /* What good is a if(bool) with a hardcoded local constant? I don't know, but handle it */
4693         LIST_FOR_EACH_ENTRY(constant, &This->baseShader.constantsB, local_constant, entry)
4694         {
4695             if (constant->idx == idx)
4696             {
4697                 return constant->value[0];
4698             }
4699         }
4700         ERR("Local constant not found\n");
4701         return FALSE;
4702     }
4703     else
4704     {
4705         if(vshader) bools = priv->cur_vs_args->boolclip.bools;
4706         else bools = priv->cur_ps_args->bools;
4707         return bools & flag;
4708     }
4709 }
4710
4711 static void get_loop_control_const(const struct wined3d_shader_instruction *ins,
4712         IWineD3DBaseShaderImpl *This, UINT idx, struct wined3d_shader_loop_control *loop_control)
4713 {
4714     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
4715
4716     /* Integer constants can either be a local constant, or they can be stored in the shader
4717      * type specific compile args. */
4718     if (This->baseShader.reg_maps.local_int_consts & (1 << idx))
4719     {
4720         const local_constant *constant;
4721
4722         LIST_FOR_EACH_ENTRY(constant, &This->baseShader.constantsI, local_constant, entry)
4723         {
4724             if (constant->idx == idx)
4725             {
4726                 loop_control->count = constant->value[0];
4727                 loop_control->start = constant->value[1];
4728                 /* Step is signed. */
4729                 loop_control->step = (int)constant->value[2];
4730                 return;
4731             }
4732         }
4733         /* If this happens the flag was set incorrectly */
4734         ERR("Local constant not found\n");
4735         loop_control->count = 0;
4736         loop_control->start = 0;
4737         loop_control->step = 0;
4738         return;
4739     }
4740
4741     switch (This->baseShader.reg_maps.shader_version.type)
4742     {
4743         case WINED3D_SHADER_TYPE_VERTEX:
4744             /* Count and aL start value are unsigned */
4745             loop_control->count = priv->cur_vs_args->loop_ctrl[idx][0];
4746             loop_control->start = priv->cur_vs_args->loop_ctrl[idx][1];
4747             /* Step is signed. */
4748             loop_control->step = ((char)priv->cur_vs_args->loop_ctrl[idx][2]);
4749             break;
4750
4751         case WINED3D_SHADER_TYPE_PIXEL:
4752             loop_control->count = priv->cur_ps_args->loop_ctrl[idx][0];
4753             loop_control->start = priv->cur_ps_args->loop_ctrl[idx][1];
4754             loop_control->step = ((char)priv->cur_ps_args->loop_ctrl[idx][2]);
4755             break;
4756
4757         default:
4758             FIXME("Unhandled shader type %#x.\n", This->baseShader.reg_maps.shader_version.type);
4759             break;
4760     }
4761 }
4762
4763 static void record_instruction(struct list *list, const struct wined3d_shader_instruction *ins)
4764 {
4765     unsigned int i;
4766     struct wined3d_shader_dst_param *dst_param = NULL;
4767     struct wined3d_shader_src_param *src_param = NULL, *rel_addr = NULL;
4768     struct recorded_instruction *rec = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*rec));
4769     if(!rec)
4770     {
4771         ERR("Out of memory\n");
4772         return;
4773     }
4774
4775     rec->ins = *ins;
4776     dst_param = HeapAlloc(GetProcessHeap(), 0, sizeof(*dst_param));
4777     if(!dst_param) goto free;
4778     *dst_param = *ins->dst;
4779     if(ins->dst->reg.rel_addr)
4780     {
4781         rel_addr = HeapAlloc(GetProcessHeap(), 0, sizeof(*dst_param->reg.rel_addr));
4782         if(!rel_addr) goto free;
4783         *rel_addr = *ins->dst->reg.rel_addr;
4784         dst_param->reg.rel_addr = rel_addr;
4785     }
4786     rec->ins.dst = dst_param;
4787
4788     src_param = HeapAlloc(GetProcessHeap(), 0, sizeof(*src_param) * ins->src_count);
4789     if(!src_param) goto free;
4790     for(i = 0; i < ins->src_count; i++)
4791     {
4792         src_param[i] = ins->src[i];
4793         if(ins->src[i].reg.rel_addr)
4794         {
4795             rel_addr = HeapAlloc(GetProcessHeap(), 0, sizeof(*rel_addr));
4796             if(!rel_addr) goto free;
4797             *rel_addr = *ins->src[i].reg.rel_addr;
4798             src_param[i].reg.rel_addr = rel_addr;
4799         }
4800     }
4801     rec->ins.src = src_param;
4802     list_add_tail(list, &rec->entry);
4803     return;
4804
4805 free:
4806     ERR("Out of memory\n");
4807     if(dst_param)
4808     {
4809         HeapFree(GetProcessHeap(), 0, (void *) dst_param->reg.rel_addr);
4810         HeapFree(GetProcessHeap(), 0, dst_param);
4811     }
4812     if(src_param)
4813     {
4814         for(i = 0; i < ins->src_count; i++)
4815         {
4816             HeapFree(GetProcessHeap(), 0, (void *) src_param[i].reg.rel_addr);
4817         }
4818         HeapFree(GetProcessHeap(), 0, src_param);
4819     }
4820     HeapFree(GetProcessHeap(), 0, rec);
4821 }
4822
4823 static void free_recorded_instruction(struct list *list)
4824 {
4825     struct recorded_instruction *rec_ins, *entry2;
4826     unsigned int i;
4827
4828     LIST_FOR_EACH_ENTRY_SAFE(rec_ins, entry2, list, struct recorded_instruction, entry)
4829     {
4830         list_remove(&rec_ins->entry);
4831         if(rec_ins->ins.dst)
4832         {
4833             HeapFree(GetProcessHeap(), 0, (void *) rec_ins->ins.dst->reg.rel_addr);
4834             HeapFree(GetProcessHeap(), 0, (void *) rec_ins->ins.dst);
4835         }
4836         if(rec_ins->ins.src)
4837         {
4838             for(i = 0; i < rec_ins->ins.src_count; i++)
4839             {
4840                 HeapFree(GetProcessHeap(), 0, (void *) rec_ins->ins.src[i].reg.rel_addr);
4841             }
4842             HeapFree(GetProcessHeap(), 0, (void *) rec_ins->ins.src);
4843         }
4844         HeapFree(GetProcessHeap(), 0, rec_ins);
4845     }
4846 }
4847
4848 static void shader_arb_handle_instruction(const struct wined3d_shader_instruction *ins) {
4849     SHADER_HANDLER hw_fct;
4850     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
4851     IWineD3DBaseShaderImpl *This = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
4852     struct control_frame *control_frame;
4853     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
4854     BOOL bool_const;
4855
4856     if(ins->handler_idx == WINED3DSIH_LOOP || ins->handler_idx == WINED3DSIH_REP)
4857     {
4858         control_frame = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*control_frame));
4859         list_add_head(&priv->control_frames, &control_frame->entry);
4860
4861         if(ins->handler_idx == WINED3DSIH_LOOP) control_frame->type = LOOP;
4862         if(ins->handler_idx == WINED3DSIH_REP) control_frame->type = REP;
4863
4864         if(priv->target_version >= NV2)
4865         {
4866             control_frame->loop_no = priv->num_loops++;
4867             priv->loop_depth++;
4868         }
4869         else
4870         {
4871             /* Don't bother recording when we're in a not used if branch */
4872             if(priv->muted)
4873             {
4874                 return;
4875             }
4876
4877             if(!priv->recording)
4878             {
4879                 list_init(&priv->record);
4880                 priv->recording = TRUE;
4881                 control_frame->outer_loop = TRUE;
4882                 get_loop_control_const(ins, This, ins->src[0].reg.idx, &control_frame->loop_control);
4883                 return; /* Instruction is handled */
4884             }
4885             /* Record this loop in the outer loop's recording */
4886         }
4887     }
4888     else if(ins->handler_idx == WINED3DSIH_ENDLOOP || ins->handler_idx == WINED3DSIH_ENDREP)
4889     {
4890         if(priv->target_version >= NV2)
4891         {
4892             /* Nothing to do. The control frame is popped after the HW instr handler */
4893         }
4894         else
4895         {
4896             struct list *e = list_head(&priv->control_frames);
4897             control_frame = LIST_ENTRY(e, struct control_frame, entry);
4898             list_remove(&control_frame->entry);
4899
4900             if(control_frame->outer_loop)
4901             {
4902                 int iteration, aL = 0;
4903                 struct list copy;
4904
4905                 /* Turn off recording before playback */
4906                 priv->recording = FALSE;
4907
4908                 /* Move the recorded instructions to a separate list and get them out of the private data
4909                  * structure. If there are nested loops, the shader_arb_handle_instruction below will
4910                  * be recorded again, thus priv->record might be overwritten
4911                  */
4912                 list_init(&copy);
4913                 list_move_tail(&copy, &priv->record);
4914                 list_init(&priv->record);
4915
4916                 if(ins->handler_idx == WINED3DSIH_ENDLOOP)
4917                 {
4918                     shader_addline(buffer, "#unrolling loop: %u iterations, aL=%u, inc %d\n",
4919                                    control_frame->loop_control.count, control_frame->loop_control.start,
4920                                    control_frame->loop_control.step);
4921                     aL = control_frame->loop_control.start;
4922                 }
4923                 else
4924                 {
4925                     shader_addline(buffer, "#unrolling rep: %u iterations\n", control_frame->loop_control.count);
4926                 }
4927
4928                 for (iteration = 0; iteration < control_frame->loop_control.count; ++iteration)
4929                 {
4930                     struct recorded_instruction *rec_ins;
4931                     if(ins->handler_idx == WINED3DSIH_ENDLOOP)
4932                     {
4933                         priv->aL = aL;
4934                         shader_addline(buffer, "#Iteration %d, aL=%d\n", iteration, aL);
4935                     }
4936                     else
4937                     {
4938                         shader_addline(buffer, "#Iteration %d\n", iteration);
4939                     }
4940
4941                     LIST_FOR_EACH_ENTRY(rec_ins, &copy, struct recorded_instruction, entry)
4942                     {
4943                         shader_arb_handle_instruction(&rec_ins->ins);
4944                     }
4945
4946                     if(ins->handler_idx == WINED3DSIH_ENDLOOP)
4947                     {
4948                         aL += control_frame->loop_control.step;
4949                     }
4950                 }
4951                 shader_addline(buffer, "#end loop/rep\n");
4952
4953                 free_recorded_instruction(&copy);
4954                 HeapFree(GetProcessHeap(), 0, control_frame);
4955                 return; /* Instruction is handled */
4956             }
4957             else
4958             {
4959                 /* This is a nested loop. Proceed to the normal recording function */
4960                 HeapFree(GetProcessHeap(), 0, control_frame);
4961             }
4962         }
4963     }
4964
4965     if(priv->recording)
4966     {
4967         record_instruction(&priv->record, ins);
4968         return;
4969     }
4970
4971     /* boolean if */
4972     if(ins->handler_idx == WINED3DSIH_IF)
4973     {
4974         control_frame = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*control_frame));
4975         list_add_head(&priv->control_frames, &control_frame->entry);
4976         control_frame->type = IF;
4977
4978         bool_const = get_bool_const(ins, This, ins->src[0].reg.idx);
4979         if(ins->src[0].modifiers == WINED3DSPSM_NOT) bool_const = !bool_const;
4980         if(!priv->muted && bool_const == FALSE)
4981         {
4982             shader_addline(buffer, "#if(FALSE){\n");
4983             priv->muted = TRUE;
4984             control_frame->muting = TRUE;
4985         }
4986         else shader_addline(buffer, "#if(TRUE) {\n");
4987
4988         return; /* Instruction is handled */
4989     }
4990     else if(ins->handler_idx == WINED3DSIH_IFC)
4991     {
4992         /* IF(bool) and if_cond(a, b) use the same ELSE and ENDIF tokens */
4993         control_frame = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*control_frame));
4994         control_frame->type = IFC;
4995         control_frame->ifc_no = priv->num_ifcs++;
4996         list_add_head(&priv->control_frames, &control_frame->entry);
4997     }
4998     else if(ins->handler_idx == WINED3DSIH_ELSE)
4999     {
5000         struct list *e = list_head(&priv->control_frames);
5001         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5002
5003         if(control_frame->type == IF)
5004         {
5005             shader_addline(buffer, "#} else {\n");
5006             if(!priv->muted && !control_frame->muting)
5007             {
5008                 priv->muted = TRUE;
5009                 control_frame->muting = TRUE;
5010             }
5011             else if(control_frame->muting) priv->muted = FALSE;
5012             return; /* Instruction is handled. */
5013         }
5014         /* In case of an ifc, generate a HW shader instruction */
5015     }
5016     else if(ins->handler_idx == WINED3DSIH_ENDIF)
5017     {
5018         struct list *e = list_head(&priv->control_frames);
5019         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5020
5021         if(control_frame->type == IF)
5022         {
5023             shader_addline(buffer, "#} endif\n");
5024             if(control_frame->muting) priv->muted = FALSE;
5025             list_remove(&control_frame->entry);
5026             HeapFree(GetProcessHeap(), 0, control_frame);
5027             return; /* Instruction is handled */
5028         }
5029     }
5030
5031     if(priv->muted) return;
5032
5033     /* Select handler */
5034     hw_fct = shader_arb_instruction_handler_table[ins->handler_idx];
5035
5036     /* Unhandled opcode */
5037     if (!hw_fct)
5038     {
5039         FIXME("Backend can't handle opcode %#x\n", ins->handler_idx);
5040         return;
5041     }
5042     hw_fct(ins);
5043
5044     if(ins->handler_idx == WINED3DSIH_ENDLOOP || ins->handler_idx == WINED3DSIH_ENDREP)
5045     {
5046         struct list *e = list_head(&priv->control_frames);
5047         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5048         list_remove(&control_frame->entry);
5049         HeapFree(GetProcessHeap(), 0, control_frame);
5050         priv->loop_depth--;
5051     }
5052     else if(ins->handler_idx == WINED3DSIH_ENDIF)
5053     {
5054         /* Non-ifc ENDIFs don't reach that place because of the return in the if block above */
5055         struct list *e = list_head(&priv->control_frames);
5056         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5057         list_remove(&control_frame->entry);
5058         HeapFree(GetProcessHeap(), 0, control_frame);
5059     }
5060
5061
5062     shader_arb_add_instruction_modifiers(ins);
5063 }
5064
5065 const shader_backend_t arb_program_shader_backend = {
5066     shader_arb_handle_instruction,
5067     shader_arb_select,
5068     shader_arb_select_depth_blt,
5069     shader_arb_deselect_depth_blt,
5070     shader_arb_update_float_vertex_constants,
5071     shader_arb_update_float_pixel_constants,
5072     shader_arb_load_constants,
5073     shader_arb_load_np2fixup_constants,
5074     shader_arb_destroy,
5075     shader_arb_alloc,
5076     shader_arb_free,
5077     shader_arb_dirty_const,
5078     shader_arb_get_caps,
5079     shader_arb_color_fixup_supported,
5080 };
5081
5082 /* ARB_fragment_program fixed function pipeline replacement definitions */
5083 #define ARB_FFP_CONST_TFACTOR           0
5084 #define ARB_FFP_CONST_SPECULAR_ENABLE   ((ARB_FFP_CONST_TFACTOR) + 1)
5085 #define ARB_FFP_CONST_CONSTANT(i)       ((ARB_FFP_CONST_SPECULAR_ENABLE) + 1 + i)
5086 #define ARB_FFP_CONST_BUMPMAT(i)        ((ARB_FFP_CONST_CONSTANT(7)) + 1 + i)
5087 #define ARB_FFP_CONST_LUMINANCE(i)      ((ARB_FFP_CONST_BUMPMAT(7)) + 1 + i)
5088
5089 struct arbfp_ffp_desc
5090 {
5091     struct ffp_frag_desc parent;
5092     GLuint shader;
5093     unsigned int num_textures_used;
5094 };
5095
5096 /* Context activation is done by the caller. */
5097 static void arbfp_enable(IWineD3DDevice *iface, BOOL enable) {
5098     ENTER_GL();
5099     if(enable) {
5100         glEnable(GL_FRAGMENT_PROGRAM_ARB);
5101         checkGLcall("glEnable(GL_FRAGMENT_PROGRAM_ARB)");
5102     } else {
5103         glDisable(GL_FRAGMENT_PROGRAM_ARB);
5104         checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
5105     }
5106     LEAVE_GL();
5107 }
5108
5109 static HRESULT arbfp_alloc(IWineD3DDevice *iface) {
5110     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *) iface;
5111     struct shader_arb_priv *priv;
5112     /* Share private data between the shader backend and the pipeline replacement, if both
5113      * are the arb implementation. This is needed to figure out whether ARBfp should be disabled
5114      * if no pixel shader is bound or not
5115      */
5116     if(This->shader_backend == &arb_program_shader_backend) {
5117         This->fragment_priv = This->shader_priv;
5118     } else {
5119         This->fragment_priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(struct shader_arb_priv));
5120         if(!This->fragment_priv) return E_OUTOFMEMORY;
5121     }
5122     priv = This->fragment_priv;
5123     if (wine_rb_init(&priv->fragment_shaders, &wined3d_ffp_frag_program_rb_functions) == -1)
5124     {
5125         ERR("Failed to initialize rbtree.\n");
5126         HeapFree(GetProcessHeap(), 0, This->fragment_priv);
5127         return E_OUTOFMEMORY;
5128     }
5129     priv->use_arbfp_fixed_func = TRUE;
5130     return WINED3D_OK;
5131 }
5132
5133 /* Context activation is done by the caller. */
5134 static void arbfp_free_ffpshader(struct wine_rb_entry *entry, void *context)
5135 {
5136     const struct wined3d_gl_info *gl_info = context;
5137     struct arbfp_ffp_desc *entry_arb = WINE_RB_ENTRY_VALUE(entry, struct arbfp_ffp_desc, parent.entry);
5138
5139     ENTER_GL();
5140     GL_EXTCALL(glDeleteProgramsARB(1, &entry_arb->shader));
5141     checkGLcall("glDeleteProgramsARB(1, &entry_arb->shader)");
5142     HeapFree(GetProcessHeap(), 0, entry_arb);
5143     LEAVE_GL();
5144 }
5145
5146 /* Context activation is done by the caller. */
5147 static void arbfp_free(IWineD3DDevice *iface) {
5148     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *) iface;
5149     struct shader_arb_priv *priv = This->fragment_priv;
5150
5151     wine_rb_destroy(&priv->fragment_shaders, arbfp_free_ffpshader, &This->adapter->gl_info);
5152     priv->use_arbfp_fixed_func = FALSE;
5153
5154     if(This->shader_backend != &arb_program_shader_backend) {
5155         HeapFree(GetProcessHeap(), 0, This->fragment_priv);
5156     }
5157 }
5158
5159 static void arbfp_get_caps(WINED3DDEVTYPE devtype, const struct wined3d_gl_info *gl_info, struct fragment_caps *caps)
5160 {
5161     caps->TextureOpCaps =  WINED3DTEXOPCAPS_DISABLE                     |
5162                            WINED3DTEXOPCAPS_SELECTARG1                  |
5163                            WINED3DTEXOPCAPS_SELECTARG2                  |
5164                            WINED3DTEXOPCAPS_MODULATE4X                  |
5165                            WINED3DTEXOPCAPS_MODULATE2X                  |
5166                            WINED3DTEXOPCAPS_MODULATE                    |
5167                            WINED3DTEXOPCAPS_ADDSIGNED2X                 |
5168                            WINED3DTEXOPCAPS_ADDSIGNED                   |
5169                            WINED3DTEXOPCAPS_ADD                         |
5170                            WINED3DTEXOPCAPS_SUBTRACT                    |
5171                            WINED3DTEXOPCAPS_ADDSMOOTH                   |
5172                            WINED3DTEXOPCAPS_BLENDCURRENTALPHA           |
5173                            WINED3DTEXOPCAPS_BLENDFACTORALPHA            |
5174                            WINED3DTEXOPCAPS_BLENDTEXTUREALPHA           |
5175                            WINED3DTEXOPCAPS_BLENDDIFFUSEALPHA           |
5176                            WINED3DTEXOPCAPS_BLENDTEXTUREALPHAPM         |
5177                            WINED3DTEXOPCAPS_MODULATEALPHA_ADDCOLOR      |
5178                            WINED3DTEXOPCAPS_MODULATECOLOR_ADDALPHA      |
5179                            WINED3DTEXOPCAPS_MODULATEINVCOLOR_ADDALPHA   |
5180                            WINED3DTEXOPCAPS_MODULATEINVALPHA_ADDCOLOR   |
5181                            WINED3DTEXOPCAPS_DOTPRODUCT3                 |
5182                            WINED3DTEXOPCAPS_MULTIPLYADD                 |
5183                            WINED3DTEXOPCAPS_LERP                        |
5184                            WINED3DTEXOPCAPS_BUMPENVMAP                  |
5185                            WINED3DTEXOPCAPS_BUMPENVMAPLUMINANCE;
5186
5187     /* TODO: Implement WINED3DTEXOPCAPS_PREMODULATE */
5188
5189     caps->MaxTextureBlendStages   = 8;
5190     caps->MaxSimultaneousTextures = min(GL_LIMITS(fragment_samplers), 8);
5191
5192     caps->PrimitiveMiscCaps |= WINED3DPMISCCAPS_TSSARGTEMP;
5193 }
5194 #undef GLINFO_LOCATION
5195
5196 #define GLINFO_LOCATION stateblock->wineD3DDevice->adapter->gl_info
5197 static void state_texfactor_arbfp(DWORD state, IWineD3DStateBlockImpl *stateblock, struct wined3d_context *context)
5198 {
5199     float col[4];
5200     IWineD3DDeviceImpl *device = stateblock->wineD3DDevice;
5201
5202     /* Don't load the parameter if we're using an arbfp pixel shader, otherwise we'll overwrite
5203      * application provided constants
5204      */
5205     if(device->shader_backend == &arb_program_shader_backend) {
5206         if (use_ps(stateblock)) return;
5207
5208         device = stateblock->wineD3DDevice;
5209         context->pshader_const_dirty[ARB_FFP_CONST_TFACTOR] = 1;
5210         device->highest_dirty_ps_const = max(device->highest_dirty_ps_const, ARB_FFP_CONST_TFACTOR + 1);
5211     }
5212
5213     D3DCOLORTOGLFLOAT4(stateblock->renderState[WINED3DRS_TEXTUREFACTOR], col);
5214     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_TFACTOR, col));
5215     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_TFACTOR, col)");
5216
5217 }
5218
5219 static void state_arb_specularenable(DWORD state, IWineD3DStateBlockImpl *stateblock, struct wined3d_context *context)
5220 {
5221     float col[4];
5222     IWineD3DDeviceImpl *device = stateblock->wineD3DDevice;
5223
5224     /* Don't load the parameter if we're using an arbfp pixel shader, otherwise we'll overwrite
5225      * application provided constants
5226      */
5227     if(device->shader_backend == &arb_program_shader_backend) {
5228         if (use_ps(stateblock)) return;
5229
5230         device = stateblock->wineD3DDevice;
5231         context->pshader_const_dirty[ARB_FFP_CONST_SPECULAR_ENABLE] = 1;
5232         device->highest_dirty_ps_const = max(device->highest_dirty_ps_const, ARB_FFP_CONST_SPECULAR_ENABLE + 1);
5233     }
5234
5235     if(stateblock->renderState[WINED3DRS_SPECULARENABLE]) {
5236         /* The specular color has no alpha */
5237         col[0] = 1.0f; col[1] = 1.0f;
5238         col[2] = 1.0f; col[3] = 0.0f;
5239     } else {
5240         col[0] = 0.0f; col[1] = 0.0f;
5241         col[2] = 0.0f; col[3] = 0.0f;
5242     }
5243     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_SPECULAR_ENABLE, col));
5244     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_SPECULAR_ENABLE, col)");
5245 }
5246
5247 static void set_bumpmat_arbfp(DWORD state, IWineD3DStateBlockImpl *stateblock, struct wined3d_context *context)
5248 {
5249     DWORD stage = (state - STATE_TEXTURESTAGE(0, 0)) / (WINED3D_HIGHEST_TEXTURE_STATE + 1);
5250     IWineD3DDeviceImpl *device = stateblock->wineD3DDevice;
5251     float mat[2][2];
5252
5253     if (use_ps(stateblock))
5254     {
5255         if (stage != 0
5256                 && (((IWineD3DPixelShaderImpl *)stateblock->pixelShader)->baseShader.reg_maps.bumpmat & (1 << stage)))
5257         {
5258             /* The pixel shader has to know the bump env matrix. Do a constants update if it isn't scheduled
5259              * anyway
5260              */
5261             if(!isStateDirty(context, STATE_PIXELSHADERCONSTANT)) {
5262                 device->StateTable[STATE_PIXELSHADERCONSTANT].apply(STATE_PIXELSHADERCONSTANT, stateblock, context);
5263             }
5264         }
5265
5266         if(device->shader_backend == &arb_program_shader_backend) {
5267             /* Exit now, don't set the bumpmat below, otherwise we may overwrite pixel shader constants */
5268             return;
5269         }
5270     } else if(device->shader_backend == &arb_program_shader_backend) {
5271         context->pshader_const_dirty[ARB_FFP_CONST_BUMPMAT(stage)] = 1;
5272         device->highest_dirty_ps_const = max(device->highest_dirty_ps_const, ARB_FFP_CONST_BUMPMAT(stage) + 1);
5273     }
5274
5275     mat[0][0] = *((float *) &stateblock->textureState[stage][WINED3DTSS_BUMPENVMAT00]);
5276     mat[0][1] = *((float *) &stateblock->textureState[stage][WINED3DTSS_BUMPENVMAT01]);
5277     mat[1][0] = *((float *) &stateblock->textureState[stage][WINED3DTSS_BUMPENVMAT10]);
5278     mat[1][1] = *((float *) &stateblock->textureState[stage][WINED3DTSS_BUMPENVMAT11]);
5279
5280     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_BUMPMAT(stage), &mat[0][0]));
5281     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_BUMPMAT(stage), &mat[0][0])");
5282 }
5283
5284 static void tex_bumpenvlum_arbfp(DWORD state, IWineD3DStateBlockImpl *stateblock, struct wined3d_context *context)
5285 {
5286     DWORD stage = (state - STATE_TEXTURESTAGE(0, 0)) / (WINED3D_HIGHEST_TEXTURE_STATE + 1);
5287     IWineD3DDeviceImpl *device = stateblock->wineD3DDevice;
5288     float param[4];
5289
5290     if (use_ps(stateblock))
5291     {
5292         if (stage != 0
5293                 && (((IWineD3DPixelShaderImpl *)stateblock->pixelShader)->baseShader.reg_maps.luminanceparams & (1 << stage)))
5294         {
5295             /* The pixel shader has to know the luminance offset. Do a constants update if it
5296              * isn't scheduled anyway
5297              */
5298             if(!isStateDirty(context, STATE_PIXELSHADERCONSTANT)) {
5299                 device->StateTable[STATE_PIXELSHADERCONSTANT].apply(STATE_PIXELSHADERCONSTANT, stateblock, context);
5300             }
5301         }
5302
5303         if(device->shader_backend == &arb_program_shader_backend) {
5304             /* Exit now, don't set the bumpmat below, otherwise we may overwrite pixel shader constants */
5305             return;
5306         }
5307     } else if(device->shader_backend == &arb_program_shader_backend) {
5308         context->pshader_const_dirty[ARB_FFP_CONST_LUMINANCE(stage)] = 1;
5309         device->highest_dirty_ps_const = max(device->highest_dirty_ps_const, ARB_FFP_CONST_LUMINANCE(stage) + 1);
5310     }
5311
5312     param[0] = *((float *) &stateblock->textureState[stage][WINED3DTSS_BUMPENVLSCALE]);
5313     param[1] = *((float *) &stateblock->textureState[stage][WINED3DTSS_BUMPENVLOFFSET]);
5314     param[2] = 0.0f;
5315     param[3] = 0.0f;
5316
5317     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_LUMINANCE(stage), param));
5318     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_LUMINANCE(stage), param)");
5319 }
5320
5321 static const char *get_argreg(struct wined3d_shader_buffer *buffer, DWORD argnum, unsigned int stage, DWORD arg)
5322 {
5323     const char *ret;
5324
5325     if(arg == ARG_UNUSED) return "unused"; /* This is the marker for unused registers */
5326
5327     switch(arg & WINED3DTA_SELECTMASK) {
5328         case WINED3DTA_DIFFUSE:
5329             ret = "fragment.color.primary"; break;
5330
5331         case WINED3DTA_CURRENT:
5332             if(stage == 0) ret = "fragment.color.primary";
5333             else ret = "ret";
5334             break;
5335
5336         case WINED3DTA_TEXTURE:
5337             switch(stage) {
5338                 case 0: ret = "tex0"; break;
5339                 case 1: ret = "tex1"; break;
5340                 case 2: ret = "tex2"; break;
5341                 case 3: ret = "tex3"; break;
5342                 case 4: ret = "tex4"; break;
5343                 case 5: ret = "tex5"; break;
5344                 case 6: ret = "tex6"; break;
5345                 case 7: ret = "tex7"; break;
5346                 default: ret = "unknown texture";
5347             }
5348             break;
5349
5350         case WINED3DTA_TFACTOR:
5351             ret = "tfactor"; break;
5352
5353         case WINED3DTA_SPECULAR:
5354             ret = "fragment.color.secondary"; break;
5355
5356         case WINED3DTA_TEMP:
5357             ret = "tempreg"; break;
5358
5359         case WINED3DTA_CONSTANT:
5360             FIXME("Implement perstage constants\n");
5361             switch(stage) {
5362                 case 0: ret = "const0"; break;
5363                 case 1: ret = "const1"; break;
5364                 case 2: ret = "const2"; break;
5365                 case 3: ret = "const3"; break;
5366                 case 4: ret = "const4"; break;
5367                 case 5: ret = "const5"; break;
5368                 case 6: ret = "const6"; break;
5369                 case 7: ret = "const7"; break;
5370                 default: ret = "unknown constant";
5371             }
5372             break;
5373
5374         default:
5375             return "unknown";
5376     }
5377
5378     if(arg & WINED3DTA_COMPLEMENT) {
5379         shader_addline(buffer, "SUB arg%u, const.x, %s;\n", argnum, ret);
5380         if(argnum == 0) ret = "arg0";
5381         if(argnum == 1) ret = "arg1";
5382         if(argnum == 2) ret = "arg2";
5383     }
5384     if(arg & WINED3DTA_ALPHAREPLICATE) {
5385         shader_addline(buffer, "MOV arg%u, %s.w;\n", argnum, ret);
5386         if(argnum == 0) ret = "arg0";
5387         if(argnum == 1) ret = "arg1";
5388         if(argnum == 2) ret = "arg2";
5389     }
5390     return ret;
5391 }
5392
5393 static void gen_ffp_instr(struct wined3d_shader_buffer *buffer, unsigned int stage, BOOL color,
5394         BOOL alpha, DWORD dst, DWORD op, DWORD dw_arg0, DWORD dw_arg1, DWORD dw_arg2)
5395 {
5396     const char *dstmask, *dstreg, *arg0, *arg1, *arg2;
5397     unsigned int mul = 1;
5398     BOOL mul_final_dest = FALSE;
5399
5400     if(color && alpha) dstmask = "";
5401     else if(color) dstmask = ".xyz";
5402     else dstmask = ".w";
5403
5404     if(dst == tempreg) dstreg = "tempreg";
5405     else dstreg = "ret";
5406
5407     arg0 = get_argreg(buffer, 0, stage, dw_arg0);
5408     arg1 = get_argreg(buffer, 1, stage, dw_arg1);
5409     arg2 = get_argreg(buffer, 2, stage, dw_arg2);
5410
5411     switch(op) {
5412         case WINED3DTOP_DISABLE:
5413             if(stage == 0) shader_addline(buffer, "MOV %s%s, fragment.color.primary;\n", dstreg, dstmask);
5414             break;
5415
5416         case WINED3DTOP_SELECTARG2:
5417             arg1 = arg2;
5418         case WINED3DTOP_SELECTARG1:
5419             shader_addline(buffer, "MOV %s%s, %s;\n", dstreg, dstmask, arg1);
5420             break;
5421
5422         case WINED3DTOP_MODULATE4X:
5423             mul = 2;
5424         case WINED3DTOP_MODULATE2X:
5425             mul *= 2;
5426             if(strcmp(dstreg, "result.color") == 0) {
5427                 dstreg = "ret";
5428                 mul_final_dest = TRUE;
5429             }
5430         case WINED3DTOP_MODULATE:
5431             shader_addline(buffer, "MUL %s%s, %s, %s;\n", dstreg, dstmask, arg1, arg2);
5432             break;
5433
5434         case WINED3DTOP_ADDSIGNED2X:
5435             mul = 2;
5436             if(strcmp(dstreg, "result.color") == 0) {
5437                 dstreg = "ret";
5438                 mul_final_dest = TRUE;
5439             }
5440         case WINED3DTOP_ADDSIGNED:
5441             shader_addline(buffer, "SUB arg2, %s, const.w;\n", arg2);
5442             arg2 = "arg2";
5443         case WINED3DTOP_ADD:
5444             shader_addline(buffer, "ADD_SAT %s%s, %s, %s;\n", dstreg, dstmask, arg1, arg2);
5445             break;
5446
5447         case WINED3DTOP_SUBTRACT:
5448             shader_addline(buffer, "SUB_SAT %s%s, %s, %s;\n", dstreg, dstmask, arg1, arg2);
5449             break;
5450
5451         case WINED3DTOP_ADDSMOOTH:
5452             shader_addline(buffer, "SUB arg1, const.x, %s;\n", arg1);
5453             shader_addline(buffer, "MAD_SAT %s%s, arg1, %s, %s;\n", dstreg, dstmask, arg2, arg1);
5454             break;
5455
5456         case WINED3DTOP_BLENDCURRENTALPHA:
5457             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_CURRENT);
5458             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
5459             break;
5460         case WINED3DTOP_BLENDFACTORALPHA:
5461             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_TFACTOR);
5462             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
5463             break;
5464         case WINED3DTOP_BLENDTEXTUREALPHA:
5465             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_TEXTURE);
5466             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
5467             break;
5468         case WINED3DTOP_BLENDDIFFUSEALPHA:
5469             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_DIFFUSE);
5470             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
5471             break;
5472
5473         case WINED3DTOP_BLENDTEXTUREALPHAPM:
5474             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_TEXTURE);
5475             shader_addline(buffer, "SUB arg0.w, const.x, %s.w;\n", arg0);
5476             shader_addline(buffer, "MAD_SAT %s%s, %s, arg0.w, %s;\n", dstreg, dstmask, arg2, arg1);
5477             break;
5478
5479         /* D3DTOP_PREMODULATE ???? */
5480
5481         case WINED3DTOP_MODULATEINVALPHA_ADDCOLOR:
5482             shader_addline(buffer, "SUB arg0.w, const.x, %s;\n", arg1);
5483             shader_addline(buffer, "MAD_SAT %s%s, arg0.w, %s, %s;\n", dstreg, dstmask, arg2, arg1);
5484             break;
5485         case WINED3DTOP_MODULATEALPHA_ADDCOLOR:
5486             shader_addline(buffer, "MAD_SAT %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg1, arg2, arg1);
5487             break;
5488         case WINED3DTOP_MODULATEINVCOLOR_ADDALPHA:
5489             shader_addline(buffer, "SUB arg0, const.x, %s;\n", arg1);
5490             shader_addline(buffer, "MAD_SAT %s%s, arg0, %s, %s.w;\n", dstreg, dstmask, arg2, arg1);
5491             break;
5492         case WINED3DTOP_MODULATECOLOR_ADDALPHA:
5493             shader_addline(buffer, "MAD_SAT %s%s, %s, %s, %s.w;\n", dstreg, dstmask, arg1, arg2, arg1);
5494             break;
5495
5496         case WINED3DTOP_DOTPRODUCT3:
5497             mul = 4;
5498             if(strcmp(dstreg, "result.color") == 0) {
5499                 dstreg = "ret";
5500                 mul_final_dest = TRUE;
5501             }
5502             shader_addline(buffer, "SUB arg1, %s, const.w;\n", arg1);
5503             shader_addline(buffer, "SUB arg2, %s, const.w;\n", arg2);
5504             shader_addline(buffer, "DP3_SAT %s%s, arg1, arg2;\n", dstreg, dstmask);
5505             break;
5506
5507         case WINED3DTOP_MULTIPLYADD:
5508             shader_addline(buffer, "MAD_SAT %s%s, %s, %s, %s;\n", dstreg, dstmask, arg1, arg2, arg0);
5509             break;
5510
5511         case WINED3DTOP_LERP:
5512             /* The msdn is not quite right here */
5513             shader_addline(buffer, "LRP %s%s, %s, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
5514             break;
5515
5516         case WINED3DTOP_BUMPENVMAP:
5517         case WINED3DTOP_BUMPENVMAPLUMINANCE:
5518             /* Those are handled in the first pass of the shader(generation pass 1 and 2) already */
5519             break;
5520
5521         default:
5522             FIXME("Unhandled texture op %08x\n", op);
5523     }
5524
5525     if(mul == 2) {
5526         shader_addline(buffer, "MUL_SAT %s%s, %s, const.y;\n", mul_final_dest ? "result.color" : dstreg, dstmask, dstreg);
5527     } else if(mul == 4) {
5528         shader_addline(buffer, "MUL_SAT %s%s, %s, const.z;\n", mul_final_dest ? "result.color" : dstreg, dstmask, dstreg);
5529     }
5530 }
5531
5532 /* The stateblock is passed for GLINFO_LOCATION */
5533 static GLuint gen_arbfp_ffp_shader(const struct ffp_frag_settings *settings, IWineD3DStateBlockImpl *stateblock)
5534 {
5535     unsigned int stage;
5536     struct wined3d_shader_buffer buffer;
5537     BOOL tex_read[MAX_TEXTURES] = {FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE};
5538     BOOL bump_used[MAX_TEXTURES] = {FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE};
5539     BOOL luminance_used[MAX_TEXTURES] = {FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE};
5540     const char *textype;
5541     const char *instr, *sat;
5542     char colorcor_dst[8];
5543     GLuint ret;
5544     DWORD arg0, arg1, arg2;
5545     BOOL tempreg_used = FALSE, tfactor_used = FALSE;
5546     BOOL op_equal;
5547     const char *final_combiner_src = "ret";
5548     GLint pos;
5549
5550     /* Find out which textures are read */
5551     for(stage = 0; stage < MAX_TEXTURES; stage++) {
5552         if(settings->op[stage].cop == WINED3DTOP_DISABLE) break;
5553         arg0 = settings->op[stage].carg0 & WINED3DTA_SELECTMASK;
5554         arg1 = settings->op[stage].carg1 & WINED3DTA_SELECTMASK;
5555         arg2 = settings->op[stage].carg2 & WINED3DTA_SELECTMASK;
5556         if(arg0 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
5557         if(arg1 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
5558         if(arg2 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
5559
5560         if(settings->op[stage].cop == WINED3DTOP_BLENDTEXTUREALPHA) tex_read[stage] = TRUE;
5561         if(settings->op[stage].cop == WINED3DTOP_BLENDTEXTUREALPHAPM) tex_read[stage] = TRUE;
5562         if(settings->op[stage].cop == WINED3DTOP_BUMPENVMAP) {
5563             bump_used[stage] = TRUE;
5564             tex_read[stage] = TRUE;
5565         }
5566         if(settings->op[stage].cop == WINED3DTOP_BUMPENVMAPLUMINANCE) {
5567             bump_used[stage] = TRUE;
5568             tex_read[stage] = TRUE;
5569             luminance_used[stage] = TRUE;
5570         } else if(settings->op[stage].cop == WINED3DTOP_BLENDFACTORALPHA) {
5571             tfactor_used = TRUE;
5572         }
5573
5574         if(arg0 == WINED3DTA_TFACTOR || arg1 == WINED3DTA_TFACTOR || arg2 == WINED3DTA_TFACTOR) {
5575             tfactor_used = TRUE;
5576         }
5577
5578         if(settings->op[stage].dst == tempreg) tempreg_used = TRUE;
5579         if(arg0 == WINED3DTA_TEMP || arg1 == WINED3DTA_TEMP || arg2 == WINED3DTA_TEMP) {
5580             tempreg_used = TRUE;
5581         }
5582
5583         if(settings->op[stage].aop == WINED3DTOP_DISABLE) continue;
5584         arg0 = settings->op[stage].aarg0 & WINED3DTA_SELECTMASK;
5585         arg1 = settings->op[stage].aarg1 & WINED3DTA_SELECTMASK;
5586         arg2 = settings->op[stage].aarg2 & WINED3DTA_SELECTMASK;
5587         if(arg0 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
5588         if(arg1 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
5589         if(arg2 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
5590
5591         if(arg0 == WINED3DTA_TEMP || arg1 == WINED3DTA_TEMP || arg2 == WINED3DTA_TEMP) {
5592             tempreg_used = TRUE;
5593         }
5594         if(arg0 == WINED3DTA_TFACTOR || arg1 == WINED3DTA_TFACTOR || arg2 == WINED3DTA_TFACTOR) {
5595             tfactor_used = TRUE;
5596         }
5597     }
5598
5599     /* Shader header */
5600     if (!shader_buffer_init(&buffer))
5601     {
5602         ERR("Failed to initialize shader buffer.\n");
5603         return 0;
5604     }
5605
5606     shader_addline(&buffer, "!!ARBfp1.0\n");
5607
5608     switch(settings->fog) {
5609         case FOG_OFF:                                                         break;
5610         case FOG_LINEAR: shader_addline(&buffer, "OPTION ARB_fog_linear;\n"); break;
5611         case FOG_EXP:    shader_addline(&buffer, "OPTION ARB_fog_exp;\n");    break;
5612         case FOG_EXP2:   shader_addline(&buffer, "OPTION ARB_fog_exp2;\n");   break;
5613         default: FIXME("Unexpected fog setting %d\n", settings->fog);
5614     }
5615
5616     shader_addline(&buffer, "PARAM const = {1, 2, 4, 0.5};\n");
5617     shader_addline(&buffer, "TEMP TMP;\n");
5618     shader_addline(&buffer, "TEMP ret;\n");
5619     if(tempreg_used || settings->sRGB_write) shader_addline(&buffer, "TEMP tempreg;\n");
5620     shader_addline(&buffer, "TEMP arg0;\n");
5621     shader_addline(&buffer, "TEMP arg1;\n");
5622     shader_addline(&buffer, "TEMP arg2;\n");
5623     for(stage = 0; stage < MAX_TEXTURES; stage++) {
5624         if(!tex_read[stage]) continue;
5625         shader_addline(&buffer, "TEMP tex%u;\n", stage);
5626         if(!bump_used[stage]) continue;
5627         shader_addline(&buffer, "PARAM bumpmat%u = program.env[%u];\n", stage, ARB_FFP_CONST_BUMPMAT(stage));
5628         if(!luminance_used[stage]) continue;
5629         shader_addline(&buffer, "PARAM luminance%u = program.env[%u];\n", stage, ARB_FFP_CONST_LUMINANCE(stage));
5630     }
5631     if(tfactor_used) {
5632         shader_addline(&buffer, "PARAM tfactor = program.env[%u];\n", ARB_FFP_CONST_TFACTOR);
5633     }
5634         shader_addline(&buffer, "PARAM specular_enable = program.env[%u];\n", ARB_FFP_CONST_SPECULAR_ENABLE);
5635
5636     if(settings->sRGB_write) {
5637         shader_addline(&buffer, "PARAM srgb_consts1 = {%f, %f, %f, %f};\n",
5638                        srgb_mul_low, srgb_cmp, srgb_pow, srgb_mul_high);
5639         shader_addline(&buffer, "PARAM srgb_consts2 = {%f, %f, %f, %f};\n",
5640                        srgb_sub_high, 0.0, 0.0, 0.0);
5641     }
5642
5643     if(ffp_clip_emul(stateblock) && settings->emul_clipplanes) shader_addline(&buffer, "KIL fragment.texcoord[7];\n");
5644
5645     /* Generate texture sampling instructions) */
5646     for(stage = 0; stage < MAX_TEXTURES && settings->op[stage].cop != WINED3DTOP_DISABLE; stage++) {
5647         if(!tex_read[stage]) continue;
5648
5649         switch(settings->op[stage].tex_type) {
5650             case tex_1d:                    textype = "1D";     break;
5651             case tex_2d:                    textype = "2D";     break;
5652             case tex_3d:                    textype = "3D";     break;
5653             case tex_cube:                  textype = "CUBE";   break;
5654             case tex_rect:                  textype = "RECT";   break;
5655             default: textype = "unexpected_textype";   break;
5656         }
5657
5658         if(settings->op[stage].cop == WINED3DTOP_BUMPENVMAP ||
5659            settings->op[stage].cop == WINED3DTOP_BUMPENVMAPLUMINANCE) {
5660             sat = "";
5661         } else {
5662             sat = "_SAT";
5663         }
5664
5665         if(settings->op[stage].projected == proj_none) {
5666             instr = "TEX";
5667         } else if(settings->op[stage].projected == proj_count4 ||
5668                   settings->op[stage].projected == proj_count3) {
5669             instr = "TXP";
5670         } else {
5671             FIXME("Unexpected projection mode %d\n", settings->op[stage].projected);
5672             instr = "TXP";
5673         }
5674
5675         if(stage > 0 &&
5676            (settings->op[stage - 1].cop == WINED3DTOP_BUMPENVMAP ||
5677             settings->op[stage - 1].cop == WINED3DTOP_BUMPENVMAPLUMINANCE)) {
5678             shader_addline(&buffer, "SWZ arg1, bumpmat%u, x, z, 0, 0;\n", stage - 1);
5679             shader_addline(&buffer, "DP3 ret.x, arg1, tex%u;\n", stage - 1);
5680             shader_addline(&buffer, "SWZ arg1, bumpmat%u, y, w, 0, 0;\n", stage - 1);
5681             shader_addline(&buffer, "DP3 ret.y, arg1, tex%u;\n", stage - 1);
5682
5683             /* with projective textures, texbem only divides the static texture coord, not the displacement,
5684              * so multiply the displacement with the dividing parameter before passing it to TXP
5685              */
5686             if (settings->op[stage].projected != proj_none) {
5687                 if(settings->op[stage].projected == proj_count4) {
5688                     shader_addline(&buffer, "MOV ret.w, fragment.texcoord[%u].w;\n", stage);
5689                     shader_addline(&buffer, "MUL ret.xyz, ret, fragment.texcoord[%u].w, fragment.texcoord[%u];\n", stage, stage);
5690                 } else {
5691                     shader_addline(&buffer, "MOV ret.w, fragment.texcoord[%u].z;\n", stage);
5692                     shader_addline(&buffer, "MAD ret.xyz, ret, fragment.texcoord[%u].z, fragment.texcoord[%u];\n", stage, stage);
5693                 }
5694             } else {
5695                 shader_addline(&buffer, "ADD ret, ret, fragment.texcoord[%u];\n", stage);
5696             }
5697
5698             shader_addline(&buffer, "%s%s tex%u, ret, texture[%u], %s;\n",
5699                            instr, sat, stage, stage, textype);
5700             if(settings->op[stage - 1].cop == WINED3DTOP_BUMPENVMAPLUMINANCE) {
5701                 shader_addline(&buffer, "MAD_SAT ret.x, tex%u.z, luminance%u.x, luminance%u.y;\n",
5702                                stage - 1, stage - 1, stage - 1);
5703                 shader_addline(&buffer, "MUL tex%u, tex%u, ret.x;\n", stage, stage);
5704             }
5705         } else if(settings->op[stage].projected == proj_count3) {
5706             shader_addline(&buffer, "MOV ret, fragment.texcoord[%u];\n", stage);
5707             shader_addline(&buffer, "MOV ret.w, ret.z;\n");
5708             shader_addline(&buffer, "%s%s tex%u, ret, texture[%u], %s;\n",
5709                             instr, sat, stage, stage, textype);
5710         } else {
5711             shader_addline(&buffer, "%s%s tex%u, fragment.texcoord[%u], texture[%u], %s;\n",
5712                             instr, sat, stage, stage, stage, textype);
5713         }
5714
5715         sprintf(colorcor_dst, "tex%u", stage);
5716         gen_color_correction(&buffer, colorcor_dst, WINED3DSP_WRITEMASK_ALL, "const.x", "const.y",
5717                 settings->op[stage].color_fixup);
5718     }
5719
5720     /* Generate the main shader */
5721     for(stage = 0; stage < MAX_TEXTURES; stage++) {
5722         if(settings->op[stage].cop == WINED3DTOP_DISABLE) {
5723             if(stage == 0) {
5724                 final_combiner_src = "fragment.color.primary";
5725             }
5726             break;
5727         }
5728
5729         if(settings->op[stage].cop == WINED3DTOP_SELECTARG1 &&
5730            settings->op[stage].aop == WINED3DTOP_SELECTARG1) {
5731             op_equal = settings->op[stage].carg1 == settings->op[stage].aarg1;
5732         } else if(settings->op[stage].cop == WINED3DTOP_SELECTARG1 &&
5733                   settings->op[stage].aop == WINED3DTOP_SELECTARG2) {
5734             op_equal = settings->op[stage].carg1 == settings->op[stage].aarg2;
5735         } else if(settings->op[stage].cop == WINED3DTOP_SELECTARG2 &&
5736                   settings->op[stage].aop == WINED3DTOP_SELECTARG1) {
5737             op_equal = settings->op[stage].carg2 == settings->op[stage].aarg1;
5738         } else if(settings->op[stage].cop == WINED3DTOP_SELECTARG2 &&
5739                   settings->op[stage].aop == WINED3DTOP_SELECTARG2) {
5740             op_equal = settings->op[stage].carg2 == settings->op[stage].aarg2;
5741         } else {
5742             op_equal = settings->op[stage].aop   == settings->op[stage].cop &&
5743                        settings->op[stage].carg0 == settings->op[stage].aarg0 &&
5744                        settings->op[stage].carg1 == settings->op[stage].aarg1 &&
5745                        settings->op[stage].carg2 == settings->op[stage].aarg2;
5746         }
5747
5748         if(settings->op[stage].aop == WINED3DTOP_DISABLE) {
5749             gen_ffp_instr(&buffer, stage, TRUE, FALSE, settings->op[stage].dst,
5750                           settings->op[stage].cop, settings->op[stage].carg0,
5751                           settings->op[stage].carg1, settings->op[stage].carg2);
5752             if(stage == 0) {
5753                 shader_addline(&buffer, "MOV ret.w, fragment.color.primary.w;\n");
5754             }
5755         } else if(op_equal) {
5756             gen_ffp_instr(&buffer, stage, TRUE, TRUE, settings->op[stage].dst,
5757                           settings->op[stage].cop, settings->op[stage].carg0,
5758                           settings->op[stage].carg1, settings->op[stage].carg2);
5759         } else {
5760             gen_ffp_instr(&buffer, stage, TRUE, FALSE, settings->op[stage].dst,
5761                           settings->op[stage].cop, settings->op[stage].carg0,
5762                           settings->op[stage].carg1, settings->op[stage].carg2);
5763             gen_ffp_instr(&buffer, stage, FALSE, TRUE, settings->op[stage].dst,
5764                           settings->op[stage].aop, settings->op[stage].aarg0,
5765                           settings->op[stage].aarg1, settings->op[stage].aarg2);
5766         }
5767     }
5768
5769     if(settings->sRGB_write) {
5770         shader_addline(&buffer, "MAD ret, fragment.color.secondary, specular_enable, %s;\n", final_combiner_src);
5771         arbfp_add_sRGB_correction(&buffer, "ret", "arg0", "arg1", "arg2", "tempreg", FALSE);
5772     } else {
5773         shader_addline(&buffer, "MAD result.color, fragment.color.secondary, specular_enable, %s;\n", final_combiner_src);
5774     }
5775
5776     /* Footer */
5777     shader_addline(&buffer, "END\n");
5778
5779     /* Generate the shader */
5780     GL_EXTCALL(glGenProgramsARB(1, &ret));
5781     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, ret));
5782     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
5783             strlen(buffer.buffer), buffer.buffer));
5784     checkGLcall("glProgramStringARB()");
5785
5786     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
5787     if (pos != -1)
5788     {
5789         FIXME("Fragment program error at position %d: %s\n", pos,
5790               debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
5791     }
5792     else
5793     {
5794         GLint native;
5795
5796         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
5797         checkGLcall("glGetProgramivARB()");
5798         if (!native) WARN("Program exceeds native resource limits.\n");
5799     }
5800
5801     shader_buffer_free(&buffer);
5802     return ret;
5803 }
5804
5805 static void fragment_prog_arbfp(DWORD state, IWineD3DStateBlockImpl *stateblock, struct wined3d_context *context)
5806 {
5807     IWineD3DDeviceImpl *device = stateblock->wineD3DDevice;
5808     struct shader_arb_priv *priv = device->fragment_priv;
5809     BOOL use_pshader = use_ps(stateblock);
5810     BOOL use_vshader = use_vs(stateblock);
5811     struct ffp_frag_settings settings;
5812     const struct arbfp_ffp_desc *desc;
5813     unsigned int i;
5814
5815     TRACE("state %#x, stateblock %p, context %p\n", state, stateblock, context);
5816
5817     if(isStateDirty(context, STATE_RENDER(WINED3DRS_FOGENABLE))) {
5818         if(!use_pshader && device->shader_backend == &arb_program_shader_backend && context->last_was_pshader) {
5819             /* Reload fixed function constants since they collide with the pixel shader constants */
5820             for(i = 0; i < MAX_TEXTURES; i++) {
5821                 set_bumpmat_arbfp(STATE_TEXTURESTAGE(i, WINED3DTSS_BUMPENVMAT00), stateblock, context);
5822             }
5823             state_texfactor_arbfp(STATE_RENDER(WINED3DRS_TEXTUREFACTOR), stateblock, context);
5824             state_arb_specularenable(STATE_RENDER(WINED3DRS_SPECULARENABLE), stateblock, context);
5825         } else if(use_pshader && !isStateDirty(context, device->StateTable[STATE_VSHADER].representative)) {
5826             device->shader_backend->shader_select(context, use_pshader, use_vshader);
5827         }
5828         return;
5829     }
5830
5831     if(!use_pshader) {
5832         /* Find or create a shader implementing the fixed function pipeline settings, then activate it */
5833         gen_ffp_frag_op(stateblock, &settings, FALSE);
5834         desc = (const struct arbfp_ffp_desc *)find_ffp_frag_shader(&priv->fragment_shaders, &settings);
5835         if(!desc) {
5836             struct arbfp_ffp_desc *new_desc = HeapAlloc(GetProcessHeap(), 0, sizeof(*new_desc));
5837             if (!new_desc)
5838             {
5839                 ERR("Out of memory\n");
5840                 return;
5841             }
5842             new_desc->num_textures_used = 0;
5843             for(i = 0; i < GL_LIMITS(texture_stages); i++) {
5844                 if(settings.op[i].cop == WINED3DTOP_DISABLE) break;
5845                 new_desc->num_textures_used = i;
5846             }
5847
5848             memcpy(&new_desc->parent.settings, &settings, sizeof(settings));
5849             new_desc->shader = gen_arbfp_ffp_shader(&settings, stateblock);
5850             add_ffp_frag_shader(&priv->fragment_shaders, &new_desc->parent);
5851             TRACE("Allocated fixed function replacement shader descriptor %p\n", new_desc);
5852             desc = new_desc;
5853         }
5854
5855         /* Now activate the replacement program. GL_FRAGMENT_PROGRAM_ARB is already active(however, note the
5856          * comment above the shader_select call below). If e.g. GLSL is active, the shader_select call will
5857          * deactivate it.
5858          */
5859         GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, desc->shader));
5860         checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, desc->shader)");
5861         priv->current_fprogram_id = desc->shader;
5862
5863         if(device->shader_backend == &arb_program_shader_backend && context->last_was_pshader) {
5864             /* Reload fixed function constants since they collide with the pixel shader constants */
5865             for(i = 0; i < MAX_TEXTURES; i++) {
5866                 set_bumpmat_arbfp(STATE_TEXTURESTAGE(i, WINED3DTSS_BUMPENVMAT00), stateblock, context);
5867             }
5868             state_texfactor_arbfp(STATE_RENDER(WINED3DRS_TEXTUREFACTOR), stateblock, context);
5869             state_arb_specularenable(STATE_RENDER(WINED3DRS_SPECULARENABLE), stateblock, context);
5870         }
5871         context->last_was_pshader = FALSE;
5872     } else {
5873         context->last_was_pshader = TRUE;
5874     }
5875
5876     /* Finally, select the shader. If a pixel shader is used, it will be set and enabled by the shader backend.
5877      * If this shader backend is arbfp(most likely), then it will simply overwrite the last fixed function replace-
5878      * ment shader. If the shader backend is not ARB, it currently is important that the opengl implementation
5879      * type overwrites GL_ARB_fragment_program. This is currently the case with GLSL. If we really want to use
5880      * atifs or nvrc pixel shaders with arb fragment programs we'd have to disable GL_FRAGMENT_PROGRAM_ARB here
5881      *
5882      * Don't call shader_select if the vertex shader is dirty, because it will be called later on by the vertex
5883      * shader handler
5884      */
5885     if(!isStateDirty(context, device->StateTable[STATE_VSHADER].representative)) {
5886         device->shader_backend->shader_select(context, use_pshader, use_vshader);
5887
5888         if (!isStateDirty(context, STATE_VERTEXSHADERCONSTANT) && (use_vshader || use_pshader)) {
5889             device->StateTable[STATE_VERTEXSHADERCONSTANT].apply(STATE_VERTEXSHADERCONSTANT, stateblock, context);
5890         }
5891     }
5892     if(use_pshader) {
5893         device->StateTable[STATE_PIXELSHADERCONSTANT].apply(STATE_PIXELSHADERCONSTANT, stateblock, context);
5894     }
5895 }
5896
5897 /* We can't link the fog states to the fragment state directly since the vertex pipeline links them
5898  * to FOGENABLE. A different linking in different pipeline parts can't be expressed in the combined
5899  * state table, so we need to handle that with a forwarding function. The other invisible side effect
5900  * is that changing the fog start and fog end(which links to FOGENABLE in vertex) results in the
5901  * fragment_prog_arbfp function being called because FOGENABLE is dirty, which calls this function here
5902  */
5903 static void state_arbfp_fog(DWORD state, IWineD3DStateBlockImpl *stateblock, struct wined3d_context *context)
5904 {
5905     enum fogsource new_source;
5906
5907     TRACE("state %#x, stateblock %p, context %p\n", state, stateblock, context);
5908
5909     if(!isStateDirty(context, STATE_PIXELSHADER)) {
5910         fragment_prog_arbfp(state, stateblock, context);
5911     }
5912
5913     if(!stateblock->renderState[WINED3DRS_FOGENABLE]) return;
5914
5915     if(stateblock->renderState[WINED3DRS_FOGTABLEMODE] == WINED3DFOG_NONE) {
5916         if(use_vs(stateblock)) {
5917             new_source = FOGSOURCE_VS;
5918         } else {
5919             if(stateblock->renderState[WINED3DRS_FOGVERTEXMODE] == WINED3DFOG_NONE || context->last_was_rhw) {
5920                 new_source = FOGSOURCE_COORD;
5921             } else {
5922                 new_source = FOGSOURCE_FFP;
5923             }
5924         }
5925     } else {
5926         new_source = FOGSOURCE_FFP;
5927     }
5928     if(new_source != context->fog_source) {
5929         context->fog_source = new_source;
5930         state_fogstartend(STATE_RENDER(WINED3DRS_FOGSTART), stateblock, context);
5931     }
5932 }
5933
5934 static void textransform(DWORD state, IWineD3DStateBlockImpl *stateblock, struct wined3d_context *context)
5935 {
5936     if(!isStateDirty(context, STATE_PIXELSHADER)) {
5937         fragment_prog_arbfp(state, stateblock, context);
5938     }
5939 }
5940
5941 #undef GLINFO_LOCATION
5942
5943 static const struct StateEntryTemplate arbfp_fragmentstate_template[] = {
5944     {STATE_RENDER(WINED3DRS_TEXTUREFACTOR),               { STATE_RENDER(WINED3DRS_TEXTUREFACTOR),              state_texfactor_arbfp   }, WINED3D_GL_EXT_NONE             },
5945     {STATE_TEXTURESTAGE(0, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5946     {STATE_TEXTURESTAGE(0, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5947     {STATE_TEXTURESTAGE(0, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5948     {STATE_TEXTURESTAGE(0, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5949     {STATE_TEXTURESTAGE(0, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5950     {STATE_TEXTURESTAGE(0, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5951     {STATE_TEXTURESTAGE(0, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5952     {STATE_TEXTURESTAGE(0, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5953     {STATE_TEXTURESTAGE(0, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5954     {STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5955     {STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5956     {STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5957     {STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5958     {STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
5959     {STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
5960     {STATE_TEXTURESTAGE(1, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5961     {STATE_TEXTURESTAGE(1, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5962     {STATE_TEXTURESTAGE(1, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5963     {STATE_TEXTURESTAGE(1, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5964     {STATE_TEXTURESTAGE(1, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5965     {STATE_TEXTURESTAGE(1, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5966     {STATE_TEXTURESTAGE(1, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5967     {STATE_TEXTURESTAGE(1, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5968     {STATE_TEXTURESTAGE(1, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5969     {STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5970     {STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5971     {STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5972     {STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5973     {STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
5974     {STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
5975     {STATE_TEXTURESTAGE(2, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5976     {STATE_TEXTURESTAGE(2, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5977     {STATE_TEXTURESTAGE(2, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5978     {STATE_TEXTURESTAGE(2, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5979     {STATE_TEXTURESTAGE(2, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5980     {STATE_TEXTURESTAGE(2, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5981     {STATE_TEXTURESTAGE(2, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5982     {STATE_TEXTURESTAGE(2, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5983     {STATE_TEXTURESTAGE(2, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5984     {STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5985     {STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5986     {STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5987     {STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
5988     {STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
5989     {STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
5990     {STATE_TEXTURESTAGE(3, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5991     {STATE_TEXTURESTAGE(3, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5992     {STATE_TEXTURESTAGE(3, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5993     {STATE_TEXTURESTAGE(3, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5994     {STATE_TEXTURESTAGE(3, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5995     {STATE_TEXTURESTAGE(3, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5996     {STATE_TEXTURESTAGE(3, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5997     {STATE_TEXTURESTAGE(3, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5998     {STATE_TEXTURESTAGE(3, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
5999     {STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6000     {STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6001     {STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6002     {STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6003     {STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6004     {STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6005     {STATE_TEXTURESTAGE(4, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6006     {STATE_TEXTURESTAGE(4, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6007     {STATE_TEXTURESTAGE(4, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6008     {STATE_TEXTURESTAGE(4, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6009     {STATE_TEXTURESTAGE(4, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6010     {STATE_TEXTURESTAGE(4, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6011     {STATE_TEXTURESTAGE(4, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6012     {STATE_TEXTURESTAGE(4, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6013     {STATE_TEXTURESTAGE(4, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6014     {STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6015     {STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6016     {STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6017     {STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6018     {STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6019     {STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6020     {STATE_TEXTURESTAGE(5, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6021     {STATE_TEXTURESTAGE(5, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6022     {STATE_TEXTURESTAGE(5, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6023     {STATE_TEXTURESTAGE(5, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6024     {STATE_TEXTURESTAGE(5, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6025     {STATE_TEXTURESTAGE(5, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6026     {STATE_TEXTURESTAGE(5, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6027     {STATE_TEXTURESTAGE(5, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6028     {STATE_TEXTURESTAGE(5, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6029     {STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6030     {STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6031     {STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6032     {STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6033     {STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6034     {STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6035     {STATE_TEXTURESTAGE(6, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6036     {STATE_TEXTURESTAGE(6, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6037     {STATE_TEXTURESTAGE(6, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6038     {STATE_TEXTURESTAGE(6, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6039     {STATE_TEXTURESTAGE(6, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6040     {STATE_TEXTURESTAGE(6, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6041     {STATE_TEXTURESTAGE(6, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6042     {STATE_TEXTURESTAGE(6, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6043     {STATE_TEXTURESTAGE(6, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6044     {STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6045     {STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6046     {STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6047     {STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6048     {STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6049     {STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6050     {STATE_TEXTURESTAGE(7, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6051     {STATE_TEXTURESTAGE(7, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6052     {STATE_TEXTURESTAGE(7, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6053     {STATE_TEXTURESTAGE(7, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6054     {STATE_TEXTURESTAGE(7, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6055     {STATE_TEXTURESTAGE(7, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6056     {STATE_TEXTURESTAGE(7, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6057     {STATE_TEXTURESTAGE(7, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6058     {STATE_TEXTURESTAGE(7, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6059     {STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6060     {STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6061     {STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6062     {STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6063     {STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6064     {STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6065     {STATE_SAMPLER(0),                                    { STATE_SAMPLER(0),                                   sampler_texdim          }, WINED3D_GL_EXT_NONE             },
6066     {STATE_SAMPLER(1),                                    { STATE_SAMPLER(1),                                   sampler_texdim          }, WINED3D_GL_EXT_NONE             },
6067     {STATE_SAMPLER(2),                                    { STATE_SAMPLER(2),                                   sampler_texdim          }, WINED3D_GL_EXT_NONE             },
6068     {STATE_SAMPLER(3),                                    { STATE_SAMPLER(3),                                   sampler_texdim          }, WINED3D_GL_EXT_NONE             },
6069     {STATE_SAMPLER(4),                                    { STATE_SAMPLER(4),                                   sampler_texdim          }, WINED3D_GL_EXT_NONE             },
6070     {STATE_SAMPLER(5),                                    { STATE_SAMPLER(5),                                   sampler_texdim          }, WINED3D_GL_EXT_NONE             },
6071     {STATE_SAMPLER(6),                                    { STATE_SAMPLER(6),                                   sampler_texdim          }, WINED3D_GL_EXT_NONE             },
6072     {STATE_SAMPLER(7),                                    { STATE_SAMPLER(7),                                   sampler_texdim          }, WINED3D_GL_EXT_NONE             },
6073     {STATE_PIXELSHADER,                                   { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6074     {STATE_RENDER(WINED3DRS_FOGENABLE),                   { STATE_RENDER(WINED3DRS_FOGENABLE),                  state_arbfp_fog         }, WINED3D_GL_EXT_NONE             },
6075     {STATE_RENDER(WINED3DRS_FOGTABLEMODE),                { STATE_RENDER(WINED3DRS_FOGENABLE),                  state_arbfp_fog         }, WINED3D_GL_EXT_NONE             },
6076     {STATE_RENDER(WINED3DRS_FOGVERTEXMODE),               { STATE_RENDER(WINED3DRS_FOGENABLE),                  state_arbfp_fog         }, WINED3D_GL_EXT_NONE             },
6077     {STATE_RENDER(WINED3DRS_FOGSTART),                    { STATE_RENDER(WINED3DRS_FOGSTART),                   state_fogstartend       }, WINED3D_GL_EXT_NONE             },
6078     {STATE_RENDER(WINED3DRS_FOGEND),                      { STATE_RENDER(WINED3DRS_FOGSTART),                   state_fogstartend       }, WINED3D_GL_EXT_NONE             },
6079     {STATE_RENDER(WINED3DRS_SRGBWRITEENABLE),             { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6080     {STATE_RENDER(WINED3DRS_FOGCOLOR),                    { STATE_RENDER(WINED3DRS_FOGCOLOR),                   state_fogcolor          }, WINED3D_GL_EXT_NONE             },
6081     {STATE_RENDER(WINED3DRS_FOGDENSITY),                  { STATE_RENDER(WINED3DRS_FOGDENSITY),                 state_fogdensity        }, WINED3D_GL_EXT_NONE             },
6082     {STATE_TEXTURESTAGE(0,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(0, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6083     {STATE_TEXTURESTAGE(1,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(1, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6084     {STATE_TEXTURESTAGE(2,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(2, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6085     {STATE_TEXTURESTAGE(3,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(3, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6086     {STATE_TEXTURESTAGE(4,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(4, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6087     {STATE_TEXTURESTAGE(5,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(5, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6088     {STATE_TEXTURESTAGE(6,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(6, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6089     {STATE_TEXTURESTAGE(7,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(7, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6090     {STATE_RENDER(WINED3DRS_SPECULARENABLE),              { STATE_RENDER(WINED3DRS_SPECULARENABLE),             state_arb_specularenable}, WINED3D_GL_EXT_NONE             },
6091     {0 /* Terminate */,                                   { 0,                                                  0                       }, WINED3D_GL_EXT_NONE             },
6092 };
6093
6094 const struct fragment_pipeline arbfp_fragment_pipeline = {
6095     arbfp_enable,
6096     arbfp_get_caps,
6097     arbfp_alloc,
6098     arbfp_free,
6099     shader_arb_color_fixup_supported,
6100     arbfp_fragmentstate_template,
6101     TRUE /* We can disable projected textures */
6102 };
6103
6104 #define GLINFO_LOCATION device->adapter->gl_info
6105
6106 struct arbfp_blit_priv {
6107     GLenum yuy2_rect_shader, yuy2_2d_shader;
6108     GLenum uyvy_rect_shader, uyvy_2d_shader;
6109     GLenum yv12_rect_shader, yv12_2d_shader;
6110 };
6111
6112 static HRESULT arbfp_blit_alloc(IWineD3DDevice *iface) {
6113     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *) iface;
6114     device->blit_priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(struct arbfp_blit_priv));
6115     if(!device->blit_priv) {
6116         ERR("Out of memory\n");
6117         return E_OUTOFMEMORY;
6118     }
6119     return WINED3D_OK;
6120 }
6121
6122 /* Context activation is done by the caller. */
6123 static void arbfp_blit_free(IWineD3DDevice *iface) {
6124     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *) iface;
6125     struct arbfp_blit_priv *priv = device->blit_priv;
6126
6127     ENTER_GL();
6128     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yuy2_rect_shader));
6129     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yuy2_2d_shader));
6130     GL_EXTCALL(glDeleteProgramsARB(1, &priv->uyvy_rect_shader));
6131     GL_EXTCALL(glDeleteProgramsARB(1, &priv->uyvy_2d_shader));
6132     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yv12_rect_shader));
6133     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yv12_2d_shader));
6134     checkGLcall("Delete yuv programs");
6135     LEAVE_GL();
6136 }
6137
6138 static BOOL gen_planar_yuv_read(struct wined3d_shader_buffer *buffer, enum yuv_fixup yuv_fixup,
6139         GLenum textype, char *luminance)
6140 {
6141     char chroma;
6142     const char *tex, *texinstr;
6143
6144     if (yuv_fixup == YUV_FIXUP_UYVY) {
6145         chroma = 'x';
6146         *luminance = 'w';
6147     } else {
6148         chroma = 'w';
6149         *luminance = 'x';
6150     }
6151     switch(textype) {
6152         case GL_TEXTURE_2D:             tex = "2D";     texinstr = "TXP"; break;
6153         case GL_TEXTURE_RECTANGLE_ARB:  tex = "RECT";   texinstr = "TEX"; break;
6154         default:
6155             /* This is more tricky than just replacing the texture type - we have to navigate
6156              * properly in the texture to find the correct chroma values
6157              */
6158             FIXME("Implement yuv correction for non-2d, non-rect textures\n");
6159             return FALSE;
6160     }
6161
6162     /* First we have to read the chroma values. This means we need at least two pixels(no filtering),
6163      * or 4 pixels(with filtering). To get the unmodified chromas, we have to rid ourselves of the
6164      * filtering when we sample the texture.
6165      *
6166      * These are the rules for reading the chroma:
6167      *
6168      * Even pixel: Cr
6169      * Even pixel: U
6170      * Odd pixel: V
6171      *
6172      * So we have to get the sampling x position in non-normalized coordinates in integers
6173      */
6174     if(textype != GL_TEXTURE_RECTANGLE_ARB) {
6175         shader_addline(buffer, "MUL texcrd.xy, fragment.texcoord[0], size.x;\n");
6176         shader_addline(buffer, "MOV texcrd.w, size.x;\n");
6177     } else {
6178         shader_addline(buffer, "MOV texcrd, fragment.texcoord[0];\n");
6179     }
6180     /* We must not allow filtering between pixel x and x+1, this would mix U and V
6181      * Vertical filtering is ok. However, bear in mind that the pixel center is at
6182      * 0.5, so add 0.5.
6183      */
6184     shader_addline(buffer, "FLR texcrd.x, texcrd.x;\n");
6185     shader_addline(buffer, "ADD texcrd.x, texcrd.x, coef.y;\n");
6186
6187     /* Divide the x coordinate by 0.5 and get the fraction. This gives 0.25 and 0.75 for the
6188      * even and odd pixels respectively
6189      */
6190     shader_addline(buffer, "MUL texcrd2, texcrd, coef.y;\n");
6191     shader_addline(buffer, "FRC texcrd2, texcrd2;\n");
6192
6193     /* Sample Pixel 1 */
6194     shader_addline(buffer, "%s luminance, texcrd, texture[0], %s;\n", texinstr, tex);
6195
6196     /* Put the value into either of the chroma values */
6197     shader_addline(buffer, "SGE temp.x, texcrd2.x, coef.y;\n");
6198     shader_addline(buffer, "MUL chroma.x, luminance.%c, temp.x;\n", chroma);
6199     shader_addline(buffer, "SLT temp.x, texcrd2.x, coef.y;\n");
6200     shader_addline(buffer, "MUL chroma.y, luminance.%c, temp.x;\n", chroma);
6201
6202     /* Sample pixel 2. If we read an even pixel(SLT above returned 1), sample
6203      * the pixel right to the current one. Otherwise, sample the left pixel.
6204      * Bias and scale the SLT result to -1;1 and add it to the texcrd.x.
6205      */
6206     shader_addline(buffer, "MAD temp.x, temp.x, coef.z, -coef.x;\n");
6207     shader_addline(buffer, "ADD texcrd.x, texcrd, temp.x;\n");
6208     shader_addline(buffer, "%s luminance, texcrd, texture[0], %s;\n", texinstr, tex);
6209
6210     /* Put the value into the other chroma */
6211     shader_addline(buffer, "SGE temp.x, texcrd2.x, coef.y;\n");
6212     shader_addline(buffer, "MAD chroma.y, luminance.%c, temp.x, chroma.y;\n", chroma);
6213     shader_addline(buffer, "SLT temp.x, texcrd2.x, coef.y;\n");
6214     shader_addline(buffer, "MAD chroma.x, luminance.%c, temp.x, chroma.x;\n", chroma);
6215
6216     /* TODO: If filtering is enabled, sample a 2nd pair of pixels left or right of
6217      * the current one and lerp the two U and V values
6218      */
6219
6220     /* This gives the correctly filtered luminance value */
6221     shader_addline(buffer, "TEX luminance, fragment.texcoord[0], texture[0], %s;\n", tex);
6222
6223     return TRUE;
6224 }
6225
6226 static BOOL gen_yv12_read(struct wined3d_shader_buffer *buffer, GLenum textype, char *luminance)
6227 {
6228     const char *tex;
6229
6230     switch(textype) {
6231         case GL_TEXTURE_2D:             tex = "2D";     break;
6232         case GL_TEXTURE_RECTANGLE_ARB:  tex = "RECT";   break;
6233         default:
6234             FIXME("Implement yv12 correction for non-2d, non-rect textures\n");
6235             return FALSE;
6236     }
6237
6238     /* YV12 surfaces contain a WxH sized luminance plane, followed by a (W/2)x(H/2)
6239      * V and a (W/2)x(H/2) U plane, each with 8 bit per pixel. So the effective
6240      * bitdepth is 12 bits per pixel. Since the U and V planes have only half the
6241      * pitch of the luminance plane, the packing into the gl texture is a bit
6242      * unfortunate. If the whole texture is interpreted as luminance data it looks
6243      * approximately like this:
6244      *
6245      *        +----------------------------------+----
6246      *        |                                  |
6247      *        |                                  |
6248      *        |                                  |
6249      *        |                                  |
6250      *        |                                  |   2
6251      *        |            LUMINANCE             |   -
6252      *        |                                  |   3
6253      *        |                                  |
6254      *        |                                  |
6255      *        |                                  |
6256      *        |                                  |
6257      *        +----------------+-----------------+----
6258      *        |                |                 |
6259      *        |  U even rows   |  U odd rows     |
6260      *        |                |                 |   1
6261      *        +----------------+------------------   -
6262      *        |                |                 |   3
6263      *        |  V even rows   |  V odd rows     |
6264      *        |                |                 |
6265      *        +----------------+-----------------+----
6266      *        |                |                 |
6267      *        |     0.5        |       0.5       |
6268      *
6269      * So it appears as if there are 4 chroma images, but in fact the odd rows
6270      * in the chroma images are in the same row as the even ones. So its is
6271      * kinda tricky to read
6272      *
6273      * When reading from rectangle textures, keep in mind that the input y coordinates
6274      * go from 0 to d3d_height, whereas the opengl texture height is 1.5 * d3d_height
6275      */
6276     shader_addline(buffer, "PARAM yv12_coef = {%f, %f, %f, %f};\n",
6277             2.0f / 3.0f, 1.0f / 6.0f, (2.0f / 3.0f) + (1.0f / 6.0f), 1.0f / 3.0f);
6278
6279     shader_addline(buffer, "MOV texcrd, fragment.texcoord[0];\n");
6280     /* the chroma planes have only half the width */
6281     shader_addline(buffer, "MUL texcrd.x, texcrd.x, coef.y;\n");
6282
6283     /* The first value is between 2/3 and 5/6th of the texture's height, so scale+bias
6284      * the coordinate. Also read the right side of the image when reading odd lines
6285      *
6286      * Don't forget to clamp the y values in into the range, otherwise we'll get filtering
6287      * bleeding
6288      */
6289     if(textype == GL_TEXTURE_2D) {
6290
6291         shader_addline(buffer, "RCP chroma.w, size.y;\n");
6292
6293         shader_addline(buffer, "MUL texcrd2.y, texcrd.y, size.y;\n");
6294
6295         shader_addline(buffer, "FLR texcrd2.y, texcrd2.y;\n");
6296         shader_addline(buffer, "MAD texcrd.y, texcrd.y, yv12_coef.y, yv12_coef.x;\n");
6297
6298         /* Read odd lines from the right side(add size * 0.5 to the x coordinate */
6299         shader_addline(buffer, "ADD texcrd2.x, texcrd2.y, yv12_coef.y;\n"); /* To avoid 0.5 == 0.5 comparisons */
6300         shader_addline(buffer, "FRC texcrd2.x, texcrd2.x;\n");
6301         shader_addline(buffer, "SGE texcrd2.x, texcrd2.x, coef.y;\n");
6302         shader_addline(buffer, "MAD texcrd.x, texcrd2.x, coef.y, texcrd.x;\n");
6303
6304         /* clamp, keep the half pixel origin in mind */
6305         shader_addline(buffer, "MAD temp.y, coef.y, chroma.w, yv12_coef.x;\n");
6306         shader_addline(buffer, "MAX texcrd.y, temp.y, texcrd.y;\n");
6307         shader_addline(buffer, "MAD temp.y, -coef.y, chroma.w, yv12_coef.z;\n");
6308         shader_addline(buffer, "MIN texcrd.y, temp.y, texcrd.y;\n");
6309     } else {
6310         /* Read from [size - size+size/4] */
6311         shader_addline(buffer, "FLR texcrd.y, texcrd.y;\n");
6312         shader_addline(buffer, "MAD texcrd.y, texcrd.y, coef.w, size.y;\n");
6313
6314         /* Read odd lines from the right side(add size * 0.5 to the x coordinate */
6315         shader_addline(buffer, "ADD texcrd2.x, texcrd.y, yv12_coef.y;\n"); /* To avoid 0.5 == 0.5 comparisons */
6316         shader_addline(buffer, "FRC texcrd2.x, texcrd2.x;\n");
6317         shader_addline(buffer, "SGE texcrd2.x, texcrd2.x, coef.y;\n");
6318         shader_addline(buffer, "MUL texcrd2.x, texcrd2.x, size.x;\n");
6319         shader_addline(buffer, "MAD texcrd.x, texcrd2.x, coef.y, texcrd.x;\n");
6320
6321         /* Make sure to read exactly from the pixel center */
6322         shader_addline(buffer, "FLR texcrd.y, texcrd.y;\n");
6323         shader_addline(buffer, "ADD texcrd.y, texcrd.y, coef.y;\n");
6324
6325         /* Clamp */
6326         shader_addline(buffer, "MAD temp.y, size.y, coef.w, size.y;\n");
6327         shader_addline(buffer, "ADD temp.y, temp.y, -coef.y;\n");
6328         shader_addline(buffer, "MIN texcrd.y, temp.y, texcrd.y;\n");
6329         shader_addline(buffer, "ADD temp.y, size.y, -coef.y;\n");
6330         shader_addline(buffer, "MAX texcrd.y, temp.y, texcrd.y;\n");
6331     }
6332     /* Read the texture, put the result into the output register */
6333     shader_addline(buffer, "TEX temp, texcrd, texture[0], %s;\n", tex);
6334     shader_addline(buffer, "MOV chroma.x, temp.w;\n");
6335
6336     /* The other chroma value is 1/6th of the texture lower, from 5/6th to 6/6th
6337      * No need to clamp because we're just reusing the already clamped value from above
6338      */
6339     if(textype == GL_TEXTURE_2D) {
6340         shader_addline(buffer, "ADD texcrd.y, texcrd.y, yv12_coef.y;\n");
6341     } else {
6342         shader_addline(buffer, "MAD texcrd.y, size.y, coef.w, texcrd.y;\n");
6343     }
6344     shader_addline(buffer, "TEX temp, texcrd, texture[0], %s;\n", tex);
6345     shader_addline(buffer, "MOV chroma.y, temp.w;\n");
6346
6347     /* Sample the luminance value. It is in the top 2/3rd of the texture, so scale the y coordinate.
6348      * Clamp the y coordinate to prevent the chroma values from bleeding into the sampled luminance
6349      * values due to filtering
6350      */
6351     shader_addline(buffer, "MOV texcrd, fragment.texcoord[0];\n");
6352     if(textype == GL_TEXTURE_2D) {
6353         /* Multiply the y coordinate by 2/3 and clamp it */
6354         shader_addline(buffer, "MUL texcrd.y, texcrd.y, yv12_coef.x;\n");
6355         shader_addline(buffer, "MAD temp.y, -coef.y, chroma.w, yv12_coef.x;\n");
6356         shader_addline(buffer, "MIN texcrd.y, temp.y, texcrd.y;\n");
6357         shader_addline(buffer, "TEX luminance, texcrd, texture[0], %s;\n", tex);
6358     } else {
6359         /* Reading from texture_rectangles is pretty straightforward, just use the unmodified
6360          * texture coordinate. It is still a good idea to clamp it though, since the opengl texture
6361          * is bigger
6362          */
6363         shader_addline(buffer, "ADD temp.x, size.y, -coef.y;\n");
6364         shader_addline(buffer, "MIN texcrd.y, texcrd.y, size.x;\n");
6365         shader_addline(buffer, "TEX luminance, texcrd, texture[0], %s;\n", tex);
6366     }
6367     *luminance = 'a';
6368
6369     return TRUE;
6370 }
6371
6372 /* Context activation is done by the caller. */
6373 static GLuint gen_yuv_shader(IWineD3DDeviceImpl *device, enum yuv_fixup yuv_fixup, GLenum textype)
6374 {
6375     GLenum shader;
6376     struct wined3d_shader_buffer buffer;
6377     char luminance_component;
6378     struct arbfp_blit_priv *priv = device->blit_priv;
6379     GLint pos;
6380
6381     /* Shader header */
6382     if (!shader_buffer_init(&buffer))
6383     {
6384         ERR("Failed to initialize shader buffer.\n");
6385         return 0;
6386     }
6387
6388     ENTER_GL();
6389     GL_EXTCALL(glGenProgramsARB(1, &shader));
6390     checkGLcall("GL_EXTCALL(glGenProgramsARB(1, &shader))");
6391     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader));
6392     checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader)");
6393     LEAVE_GL();
6394     if(!shader) {
6395         shader_buffer_free(&buffer);
6396         return 0;
6397     }
6398
6399     /* The YUY2 and UYVY formats contain two pixels packed into a 32 bit macropixel,
6400      * giving effectively 16 bit per pixel. The color consists of a luminance(Y) and
6401      * two chroma(U and V) values. Each macropixel has two luminance values, one for
6402      * each single pixel it contains, and one U and one V value shared between both
6403      * pixels.
6404      *
6405      * The data is loaded into an A8L8 texture. With YUY2, the luminance component
6406      * contains the luminance and alpha the chroma. With UYVY it is vice versa. Thus
6407      * take the format into account when generating the read swizzles
6408      *
6409      * Reading the Y value is straightforward - just sample the texture. The hardware
6410      * takes care of filtering in the horizontal and vertical direction.
6411      *
6412      * Reading the U and V values is harder. We have to avoid filtering horizontally,
6413      * because that would mix the U and V values of one pixel or two adjacent pixels.
6414      * Thus floor the texture coordinate and add 0.5 to get an unfiltered read,
6415      * regardless of the filtering setting. Vertical filtering works automatically
6416      * though - the U and V values of two rows are mixed nicely.
6417      *
6418      * Appart of avoiding filtering issues, the code has to know which value it just
6419      * read, and where it can find the other one. To determine this, it checks if
6420      * it sampled an even or odd pixel, and shifts the 2nd read accordingly.
6421      *
6422      * Handling horizontal filtering of U and V values requires reading a 2nd pair
6423      * of pixels, extracting U and V and mixing them. This is not implemented yet.
6424      *
6425      * An alternative implementation idea is to load the texture as A8R8G8B8 texture,
6426      * with width / 2. This way one read gives all 3 values, finding U and V is easy
6427      * in an unfiltered situation. Finding the luminance on the other hand requires
6428      * finding out if it is an odd or even pixel. The real drawback of this approach
6429      * is filtering. This would have to be emulated completely in the shader, reading
6430      * up two 2 packed pixels in up to 2 rows and interpolating both horizontally and
6431      * vertically. Beyond that it would require adjustments to the texture handling
6432      * code to deal with the width scaling
6433      */
6434     shader_addline(&buffer, "!!ARBfp1.0\n");
6435     shader_addline(&buffer, "TEMP luminance;\n");
6436     shader_addline(&buffer, "TEMP temp;\n");
6437     shader_addline(&buffer, "TEMP chroma;\n");
6438     shader_addline(&buffer, "TEMP texcrd;\n");
6439     shader_addline(&buffer, "TEMP texcrd2;\n");
6440     shader_addline(&buffer, "PARAM coef = {1.0, 0.5, 2.0, 0.25};\n");
6441     shader_addline(&buffer, "PARAM yuv_coef = {1.403, 0.344, 0.714, 1.770};\n");
6442     shader_addline(&buffer, "PARAM size = program.local[0];\n");
6443
6444     switch (yuv_fixup)
6445     {
6446         case YUV_FIXUP_UYVY:
6447         case YUV_FIXUP_YUY2:
6448             if (!gen_planar_yuv_read(&buffer, yuv_fixup, textype, &luminance_component))
6449             {
6450                 shader_buffer_free(&buffer);
6451                 return 0;
6452             }
6453             break;
6454
6455         case YUV_FIXUP_YV12:
6456             if (!gen_yv12_read(&buffer, textype, &luminance_component))
6457             {
6458                 shader_buffer_free(&buffer);
6459                 return 0;
6460             }
6461             break;
6462
6463         default:
6464             FIXME("Unsupported YUV fixup %#x\n", yuv_fixup);
6465             shader_buffer_free(&buffer);
6466             return 0;
6467     }
6468
6469     /* Calculate the final result. Formula is taken from
6470      * http://www.fourcc.org/fccyvrgb.php. Note that the chroma
6471      * ranges from -0.5 to 0.5
6472      */
6473     shader_addline(&buffer, "SUB chroma.xy, chroma, coef.y;\n");
6474
6475     shader_addline(&buffer, "MAD result.color.x, chroma.x, yuv_coef.x, luminance.%c;\n", luminance_component);
6476     shader_addline(&buffer, "MAD temp.x, -chroma.y, yuv_coef.y, luminance.%c;\n", luminance_component);
6477     shader_addline(&buffer, "MAD result.color.y, -chroma.x, yuv_coef.z, temp.x;\n");
6478     shader_addline(&buffer, "MAD result.color.z, chroma.y, yuv_coef.w, luminance.%c;\n", luminance_component);
6479     shader_addline(&buffer, "END\n");
6480
6481     ENTER_GL();
6482     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
6483             strlen(buffer.buffer), buffer.buffer));
6484     checkGLcall("glProgramStringARB()");
6485
6486     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
6487     if (pos != -1)
6488     {
6489         FIXME("Fragment program error at position %d: %s\n", pos,
6490               debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
6491     }
6492     else
6493     {
6494         GLint native;
6495
6496         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
6497         checkGLcall("glGetProgramivARB()");
6498         if (!native) WARN("Program exceeds native resource limits.\n");
6499     }
6500
6501     shader_buffer_free(&buffer);
6502     LEAVE_GL();
6503
6504     switch (yuv_fixup)
6505     {
6506         case YUV_FIXUP_YUY2:
6507             if (textype == GL_TEXTURE_RECTANGLE_ARB) priv->yuy2_rect_shader = shader;
6508             else priv->yuy2_2d_shader = shader;
6509             break;
6510
6511         case YUV_FIXUP_UYVY:
6512             if (textype == GL_TEXTURE_RECTANGLE_ARB) priv->uyvy_rect_shader = shader;
6513             else priv->uyvy_2d_shader = shader;
6514             break;
6515
6516         case YUV_FIXUP_YV12:
6517             if (textype == GL_TEXTURE_RECTANGLE_ARB) priv->yv12_rect_shader = shader;
6518             else priv->yv12_2d_shader = shader;
6519             break;
6520     }
6521
6522     return shader;
6523 }
6524
6525 /* Context activation is done by the caller. */
6526 static HRESULT arbfp_blit_set(IWineD3DDevice *iface, const struct GlPixelFormatDesc *format_desc,
6527         GLenum textype, UINT width, UINT height)
6528 {
6529     GLenum shader;
6530     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *) iface;
6531     float size[4] = {width, height, 1, 1};
6532     struct arbfp_blit_priv *priv = device->blit_priv;
6533     enum yuv_fixup yuv_fixup;
6534
6535     if (!is_yuv_fixup(format_desc->color_fixup))
6536     {
6537         TRACE("Fixup:\n");
6538         dump_color_fixup_desc(format_desc->color_fixup);
6539         /* Don't bother setting up a shader for unconverted formats */
6540         ENTER_GL();
6541         glEnable(textype);
6542         checkGLcall("glEnable(textype)");
6543         LEAVE_GL();
6544         return WINED3D_OK;
6545     }
6546
6547     yuv_fixup = get_yuv_fixup(format_desc->color_fixup);
6548
6549     switch(yuv_fixup)
6550     {
6551         case YUV_FIXUP_YUY2:
6552             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->yuy2_rect_shader : priv->yuy2_2d_shader;
6553             break;
6554
6555         case YUV_FIXUP_UYVY:
6556             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->uyvy_rect_shader : priv->uyvy_2d_shader;
6557             break;
6558
6559         case YUV_FIXUP_YV12:
6560             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->yv12_rect_shader : priv->yv12_2d_shader;
6561             break;
6562
6563         default:
6564             FIXME("Unsupported YUV fixup %#x, not setting a shader\n", yuv_fixup);
6565             ENTER_GL();
6566             glEnable(textype);
6567             checkGLcall("glEnable(textype)");
6568             LEAVE_GL();
6569             return E_NOTIMPL;
6570     }
6571
6572     if (!shader) shader = gen_yuv_shader(device, yuv_fixup, textype);
6573
6574     ENTER_GL();
6575     glEnable(GL_FRAGMENT_PROGRAM_ARB);
6576     checkGLcall("glEnable(GL_FRAGMENT_PROGRAM_ARB)");
6577     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader));
6578     checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader)");
6579     GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, 0, size));
6580     checkGLcall("glProgramLocalParameter4fvARB");
6581     LEAVE_GL();
6582
6583     return WINED3D_OK;
6584 }
6585
6586 /* Context activation is done by the caller. */
6587 static void arbfp_blit_unset(IWineD3DDevice *iface) {
6588     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *) iface;
6589
6590     ENTER_GL();
6591     glDisable(GL_FRAGMENT_PROGRAM_ARB);
6592     checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
6593     glDisable(GL_TEXTURE_2D);
6594     checkGLcall("glDisable(GL_TEXTURE_2D)");
6595     if(GL_SUPPORT(ARB_TEXTURE_CUBE_MAP)) {
6596         glDisable(GL_TEXTURE_CUBE_MAP_ARB);
6597         checkGLcall("glDisable(GL_TEXTURE_CUBE_MAP_ARB)");
6598     }
6599     if(GL_SUPPORT(ARB_TEXTURE_RECTANGLE)) {
6600         glDisable(GL_TEXTURE_RECTANGLE_ARB);
6601         checkGLcall("glDisable(GL_TEXTURE_RECTANGLE_ARB)");
6602     }
6603     LEAVE_GL();
6604 }
6605
6606 static BOOL arbfp_blit_color_fixup_supported(struct color_fixup_desc fixup)
6607 {
6608     enum yuv_fixup yuv_fixup;
6609
6610     if (TRACE_ON(d3d_shader) && TRACE_ON(d3d))
6611     {
6612         TRACE("Checking support for fixup:\n");
6613         dump_color_fixup_desc(fixup);
6614     }
6615
6616     if (is_identity_fixup(fixup))
6617     {
6618         TRACE("[OK]\n");
6619         return TRUE;
6620     }
6621
6622     /* We only support YUV conversions. */
6623     if (!is_yuv_fixup(fixup))
6624     {
6625         TRACE("[FAILED]\n");
6626         return FALSE;
6627     }
6628
6629     yuv_fixup = get_yuv_fixup(fixup);
6630     switch(yuv_fixup)
6631     {
6632         case YUV_FIXUP_YUY2:
6633         case YUV_FIXUP_UYVY:
6634         case YUV_FIXUP_YV12:
6635             TRACE("[OK]\n");
6636             return TRUE;
6637
6638         default:
6639             FIXME("Unsupported YUV fixup %#x\n", yuv_fixup);
6640             TRACE("[FAILED]\n");
6641             return FALSE;
6642     }
6643 }
6644
6645 const struct blit_shader arbfp_blit = {
6646     arbfp_blit_alloc,
6647     arbfp_blit_free,
6648     arbfp_blit_set,
6649     arbfp_blit_unset,
6650     arbfp_blit_color_fixup_supported,
6651 };
6652
6653 #undef GLINFO_LOCATION