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_info->supported[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_info->supported[NV_VERTEX_PROGRAM2_OPTION]
  56             && !(gl_info->quirks & WINED3D_QUIRK_NV_CLIP_BROKEN);
  57 }
  58
  59 static BOOL need_helper_const(const struct wined3d_gl_info *gl_info)
  60 {
  61     if (!gl_info->supported[NV_VERTEX_PROGRAM] /* Need to init colors. */
  62             || gl_info->quirks & WINED3D_QUIRK_ARB_VS_OFFSET_LIMIT /* Load the immval offset. */
  63             || gl_info->quirks & WINED3D_QUIRK_SET_TEXCOORD_W /* Have to init texcoords. */
  64             || (!use_nv_clip(gl_info)) /* Init the clip texcoord */)
  65     {
  66         return TRUE;
  67     }
  68     return FALSE;
  69 }
  70
  71 static unsigned int reserved_vs_const(IWineD3DBaseShader *shader, const struct wined3d_gl_info *gl_info)
  72 {
  73     unsigned int ret = 1;
  74     /* We use one PARAM for the pos fixup, and in some cases one to load
  75      * some immediate values into the shader
  76      */
  77     if(need_helper_const(gl_info)) ret++;
  78     if(need_mova_const(shader, gl_info)) ret++;
  79     return ret;
  80 }
  81
  82 static inline BOOL ffp_clip_emul(IWineD3DStateBlockImpl *stateblock)
  83 {
  84     return stateblock->lowest_disabled_stage < 7;
  85 }
  86
  87 /* ARB_program_shader private data */
  88
  89 struct control_frame
  90 {
  91     struct                          list entry;
  92     enum
  93     {
  94         IF,
  95         IFC,
  96         LOOP,
  97         REP
  98     } type;
  99     BOOL                            muting;
 100     BOOL                            outer_loop;
 101     union
 102     {
 103         unsigned int                loop;
 104         unsigned int                ifc;
 105     } no;
 106     struct wined3d_shader_loop_control loop_control;
 107     BOOL                            had_else;
 108 };
 109
 110 struct arb_ps_np2fixup_info
 111 {
 112     struct ps_np2fixup_info         super;
 113     /* For ARB we need a offset value:
 114      * With both GLSL and ARB mode the NP2 fixup information (the texture dimensions) are stored in a
 115      * consecutive way (GLSL uses a uniform array). Since ARB doesn't know the notion of a "standalone"
 116      * array we need an offset to the index inside the program local parameter array. */
 117     UINT                            offset;
 118 };
 119
 120 struct arb_ps_compile_args
 121 {
 122     struct ps_compile_args          super;
 123     WORD                            bools;
 124     WORD                            clip;  /* only a boolean, use a WORD for alignment */
 125     unsigned char                   loop_ctrl[MAX_CONST_I][3];
 126 };
 127
 128 struct stb_const_desc
 129 {
 130     unsigned char           texunit;
 131     UINT                    const_num;
 132 };
 133
 134 struct arb_ps_compiled_shader
 135 {
 136     struct arb_ps_compile_args      args;
 137     struct arb_ps_np2fixup_info     np2fixup_info;
 138     struct stb_const_desc           bumpenvmatconst[MAX_TEXTURES];
 139     struct stb_const_desc           luminanceconst[MAX_TEXTURES];
 140     UINT                            int_consts[MAX_CONST_I];
 141     GLuint                          prgId;
 142     UINT                            ycorrection;
 143     unsigned char                   numbumpenvmatconsts;
 144     char                            num_int_consts;
 145 };
 146
 147 struct arb_vs_compile_args
 148 {
 149     struct vs_compile_args          super;
 150     union
 151     {
 152         struct
 153         {
 154             WORD                    bools;
 155             char                    clip_texcoord;
 156             char                    clipplane_mask;
 157         }                           boolclip;
 158         DWORD                       boolclip_compare;
 159     } clip;
 160     DWORD                           ps_signature;
 161     union
 162     {
 163         unsigned char               samplers[4];
 164         DWORD                       samplers_compare;
 165     } vertex;
 166     unsigned char                   loop_ctrl[MAX_CONST_I][3];
 167 };
 168
 169 struct arb_vs_compiled_shader
 170 {
 171     struct arb_vs_compile_args      args;
 172     GLuint                          prgId;
 173     UINT                            int_consts[MAX_CONST_I];
 174     char                            num_int_consts;
 175     char                            need_color_unclamp;
 176     UINT                            pos_fixup;
 177 };
 178
 179 struct recorded_instruction
 180 {
 181     struct wined3d_shader_instruction ins;
 182     struct list entry;
 183 };
 184
 185 struct shader_arb_ctx_priv
 186 {
 187     char addr_reg[20];
 188     enum
 189     {
 190         /* plain GL_ARB_vertex_program or GL_ARB_fragment_program */
 191         ARB,
 192         /* GL_NV_vertex_progam2_option or GL_NV_fragment_program_option */
 193         NV2,
 194         /* GL_NV_vertex_program3 or GL_NV_fragment_program2 */
 195         NV3
 196     } target_version;
 197
 198     const struct arb_vs_compile_args    *cur_vs_args;
 199     const struct arb_ps_compile_args    *cur_ps_args;
 200     const struct arb_ps_compiled_shader *compiled_fprog;
 201     const struct arb_vs_compiled_shader *compiled_vprog;
 202     struct arb_ps_np2fixup_info         *cur_np2fixup_info;
 203     struct list                         control_frames;
 204     struct list                         record;
 205     BOOL                                recording;
 206     BOOL                                muted;
 207     unsigned int                        num_loops, loop_depth, num_ifcs;
 208     int                                 aL;
 209
 210     unsigned int                        vs_clipplanes;
 211     BOOL                                footer_written;
 212     BOOL                                in_main_func;
 213
 214     /* For 3.0 vertex shaders */
 215     const char                          *vs_output[MAX_REG_OUTPUT];
 216     /* For 2.x and earlier vertex shaders */
 217     const char                          *texcrd_output[8], *color_output[2], *fog_output;
 218
 219     /* 3.0 pshader input for compatibility with fixed function */
 220     const char                          *ps_input[MAX_REG_INPUT];
 221 };
 222
 223 struct ps_signature
 224 {
 225     struct wined3d_shader_signature_element *sig;
 226     DWORD                               idx;
 227     struct wine_rb_entry                entry;
 228 };
 229
 230 struct arb_pshader_private {
 231     struct arb_ps_compiled_shader   *gl_shaders;
 232     UINT                            num_gl_shaders, shader_array_size;
 233     BOOL                            has_signature_idx;
 234     DWORD                           input_signature_idx;
 235     DWORD                           clipplane_emulation;
 236     BOOL                            clamp_consts;
 237 };
 238
 239 struct arb_vshader_private {
 240     struct arb_vs_compiled_shader   *gl_shaders;
 241     UINT                            num_gl_shaders, shader_array_size;
 242 };
 243
 244 struct shader_arb_priv
 245 {
 246     GLuint                  current_vprogram_id;
 247     GLuint                  current_fprogram_id;
 248     const struct arb_ps_compiled_shader *compiled_fprog;
 249     const struct arb_vs_compiled_shader *compiled_vprog;
 250     GLuint                  depth_blt_vprogram_id;
 251     GLuint                  depth_blt_fprogram_id[tex_type_count];
 252     BOOL                    use_arbfp_fixed_func;
 253     struct wine_rb_tree     fragment_shaders;
 254     BOOL                    last_ps_const_clamped;
 255     BOOL                    last_vs_color_unclamp;
 256
 257     struct wine_rb_tree     signature_tree;
 258     DWORD ps_sig_number;
 259 };
 260
 261 /********************************************************
 262  * ARB_[vertex/fragment]_program helper functions follow
 263  ********************************************************/
 264
 265 /* Loads floating point constants into the currently set ARB_vertex/fragment_program.
 266  * When constant_list == NULL, it will load all the constants.
 267  *
 268  * @target_type should be either GL_VERTEX_PROGRAM_ARB (for vertex shaders)
 269  *  or GL_FRAGMENT_PROGRAM_ARB (for pixel shaders)
 270  */
 271 /* GL locking is done by the caller */
 272 static unsigned int shader_arb_load_constantsF(IWineD3DBaseShaderImpl *This, const struct wined3d_gl_info *gl_info,
 273         GLuint target_type, unsigned int max_constants, const float *constants, char *dirty_consts)
 274 {
 275     local_constant* lconst;
 276     DWORD i, j;
 277     unsigned int ret;
 278
 279     if (TRACE_ON(d3d_constants))
 280     {
 281         for(i = 0; i < max_constants; i++) {
 282             if(!dirty_consts[i]) continue;
 283             TRACE_(d3d_constants)("Loading constants %i: %f, %f, %f, %f\n", i,
 284                         constants[i * 4 + 0], constants[i * 4 + 1],
 285                         constants[i * 4 + 2], constants[i * 4 + 3]);
 286         }
 287     }
 288
 289     i = 0;
 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_info->supported[EXT_GPU_PROGRAM_PARAMETERS])
 332     {
 333         /* TODO: Benchmark if we're better of with finding the dirty constants ourselves,
 334          * or just reloading *all* constants at once
 335          *
 336         GL_EXTCALL(glProgramEnvParameters4fvEXT(target_type, i, max_constants, constants + (i * 4)));
 337          */
 338         for(; i < max_constants; i++) {
 339             if(!dirty_consts[i]) continue;
 340
 341             /* Find the next block of dirty constants */
 342             dirty_consts[i] = 0;
 343             j = i;
 344             for(i++; (i < max_constants) && dirty_consts[i]; i++) {
 345                 dirty_consts[i] = 0;
 346             }
 347
 348             GL_EXTCALL(glProgramEnvParameters4fvEXT(target_type, j, i - j, constants + (j * 4)));
 349         }
 350     } else {
 351         for(; i < max_constants; i++) {
 352             if(dirty_consts[i]) {
 353                 dirty_consts[i] = 0;
 354                 GL_EXTCALL(glProgramEnvParameter4fvARB(target_type, i, constants + (i * 4)));
 355             }
 356         }
 357     }
 358     checkGLcall("glProgramEnvParameter4fvARB()");
 359
 360     /* Load immediate constants */
 361     if(This->baseShader.load_local_constsF) {
 362         if (TRACE_ON(d3d_shader)) {
 363             LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry) {
 364                 GLfloat* values = (GLfloat*)lconst->value;
 365                 TRACE_(d3d_constants)("Loading local constants %i: %f, %f, %f, %f\n", lconst->idx,
 366                         values[0], values[1], values[2], values[3]);
 367             }
 368         }
 369         /* Immediate constants are clamped for 1.X shaders at loading times */
 370         ret = 0;
 371         LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry) {
 372             dirty_consts[lconst->idx] = 1; /* Dirtify so the non-immediate constant overwrites it next time */
 373             ret = max(ret, lconst->idx + 1);
 374             GL_EXTCALL(glProgramEnvParameter4fvARB(target_type, lconst->idx, (GLfloat*)lconst->value));
 375         }
 376         checkGLcall("glProgramEnvParameter4fvARB()");
 377         return ret; /* The loaded immediate constants need reloading for the next shader */
 378     } else {
 379         return 0; /* No constants are dirty now */
 380     }
 381 }
 382
 383 /**
 384  * Loads the texture dimensions for NP2 fixup into the currently set ARB_[vertex/fragment]_programs.
 385  */
 386 static void shader_arb_load_np2fixup_constants(
 387     IWineD3DDevice* device,
 388     char usePixelShader,
 389     char useVertexShader) {
 390
 391     IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl *) device;
 392     const struct shader_arb_priv* const priv = (const struct shader_arb_priv *) deviceImpl->shader_priv;
 393     IWineD3DStateBlockImpl* stateBlock = deviceImpl->stateBlock;
 394     const struct wined3d_gl_info *gl_info = &deviceImpl->adapter->gl_info;
 395
 396     if (!usePixelShader) {
 397         /* NP2 texcoord fixup is (currently) only done for pixelshaders. */
 398         return;
 399     }
 400
 401     if (priv->compiled_fprog && priv->compiled_fprog->np2fixup_info.super.active) {
 402         const struct arb_ps_np2fixup_info* const fixup = &priv->compiled_fprog->np2fixup_info;
 403         UINT i;
 404         WORD active = fixup->super.active;
 405         GLfloat np2fixup_constants[4 * MAX_FRAGMENT_SAMPLERS];
 406
 407         for (i = 0; active; active >>= 1, ++i) {
 408             const unsigned char idx = fixup->super.idx[i];
 409             const IWineD3DTextureImpl* const tex = (const IWineD3DTextureImpl*) stateBlock->textures[i];
 410             GLfloat* tex_dim = &np2fixup_constants[(idx >> 1) * 4];
 411
 412             if (!(active & 1)) continue;
 413
 414             if (!tex) {
 415                 FIXME("Nonexistent texture is flagged for NP2 texcoord fixup\n");
 416                 continue;
 417             }
 418
 419             if (idx % 2) {
 420                 tex_dim[2] = tex->baseTexture.pow2Matrix[0]; tex_dim[3] = tex->baseTexture.pow2Matrix[5];
 421             } else {
 422                 tex_dim[0] = tex->baseTexture.pow2Matrix[0]; tex_dim[1] = tex->baseTexture.pow2Matrix[5];
 423             }
 424         }
 425
 426         for (i = 0; i < fixup->super.num_consts; ++i) {
 427             GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB,
 428                                                    fixup->offset + i, &np2fixup_constants[i * 4]));
 429         }
 430     }
 431 }
 432
 433 /* GL locking is done by the caller. */
 434 static inline void shader_arb_ps_local_constants(IWineD3DDeviceImpl* deviceImpl)
 435 {
 436     const struct wined3d_context *context = context_get_current();
 437     IWineD3DStateBlockImpl* stateBlock = deviceImpl->stateBlock;
 438     const struct wined3d_gl_info *gl_info = context->gl_info;
 439     unsigned char i;
 440     struct shader_arb_priv *priv = deviceImpl->shader_priv;
 441     const struct arb_ps_compiled_shader *gl_shader = priv->compiled_fprog;
 442
 443     for(i = 0; i < gl_shader->numbumpenvmatconsts; i++)
 444     {
 445         int texunit = gl_shader->bumpenvmatconst[i].texunit;
 446
 447         /* The state manager takes care that this function is always called if the bump env matrix changes */
 448         const float *data = (const float *)&stateBlock->textureState[texunit][WINED3DTSS_BUMPENVMAT00];
 449         GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, gl_shader->bumpenvmatconst[i].const_num, data));
 450
 451         if (gl_shader->luminanceconst[i].const_num != WINED3D_CONST_NUM_UNUSED)
 452         {
 453             /* WINED3DTSS_BUMPENVLSCALE and WINED3DTSS_BUMPENVLOFFSET are next to each other.
 454              * point gl to the scale, and load 4 floats. x = scale, y = offset, z and w are junk, we
 455              * don't care about them. The pointers are valid for sure because the stateblock is bigger.
 456              * (they're WINED3DTSS_TEXTURETRANSFORMFLAGS and WINED3DTSS_ADDRESSW, so most likely 0 or NaN
 457             */
 458             const float *scale = (const float *)&stateBlock->textureState[texunit][WINED3DTSS_BUMPENVLSCALE];
 459             GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, gl_shader->luminanceconst[i].const_num, scale));
 460         }
 461     }
 462     checkGLcall("Load bumpmap consts");
 463
 464     if(gl_shader->ycorrection != WINED3D_CONST_NUM_UNUSED)
 465     {
 466         /* ycorrection.x: Backbuffer height(onscreen) or 0(offscreen).
 467         * ycorrection.y: -1.0(onscreen), 1.0(offscreen)
 468         * ycorrection.z: 1.0
 469         * ycorrection.w: 0.0
 470         */
 471         float val[4];
 472         val[0] = context->render_offscreen ? 0.0f
 473                 : ((IWineD3DSurfaceImpl *) deviceImpl->render_targets[0])->currentDesc.Height;
 474         val[1] = context->render_offscreen ? 1.0f : -1.0f;
 475         val[2] = 1.0f;
 476         val[3] = 0.0f;
 477         GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, gl_shader->ycorrection, val));
 478         checkGLcall("y correction loading");
 479     }
 480
 481     if(gl_shader->num_int_consts == 0) return;
 482
 483     for(i = 0; i < MAX_CONST_I; i++)
 484     {
 485         if(gl_shader->int_consts[i] != WINED3D_CONST_NUM_UNUSED)
 486         {
 487             float val[4];
 488             val[0] = stateBlock->pixelShaderConstantI[4 * i];
 489             val[1] = stateBlock->pixelShaderConstantI[4 * i + 1];
 490             val[2] = stateBlock->pixelShaderConstantI[4 * i + 2];
 491             val[3] = -1.0f;
 492
 493             GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, gl_shader->int_consts[i], val));
 494         }
 495     }
 496     checkGLcall("Load ps int consts");
 497 }
 498
 499 /* GL locking is done by the caller. */
 500 static inline void shader_arb_vs_local_constants(IWineD3DDeviceImpl* deviceImpl)
 501 {
 502     IWineD3DStateBlockImpl* stateBlock;
 503     const struct wined3d_gl_info *gl_info = &deviceImpl->adapter->gl_info;
 504     unsigned char i;
 505     struct shader_arb_priv *priv = deviceImpl->shader_priv;
 506     const struct arb_vs_compiled_shader *gl_shader = priv->compiled_vprog;
 507
 508     /* Upload the position fixup */
 509     GL_EXTCALL(glProgramLocalParameter4fvARB(GL_VERTEX_PROGRAM_ARB, gl_shader->pos_fixup, deviceImpl->posFixup));
 510
 511     if(gl_shader->num_int_consts == 0) return;
 512
 513     stateBlock = deviceImpl->stateBlock;
 514
 515     for(i = 0; i < MAX_CONST_I; i++)
 516     {
 517         if(gl_shader->int_consts[i] != WINED3D_CONST_NUM_UNUSED)
 518         {
 519             float val[4];
 520             val[0] = stateBlock->vertexShaderConstantI[4 * i];
 521             val[1] = stateBlock->vertexShaderConstantI[4 * i + 1];
 522             val[2] = stateBlock->vertexShaderConstantI[4 * i + 2];
 523             val[3] = -1.0f;
 524
 525             GL_EXTCALL(glProgramLocalParameter4fvARB(GL_VERTEX_PROGRAM_ARB, gl_shader->int_consts[i], val));
 526         }
 527     }
 528     checkGLcall("Load vs int consts");
 529 }
 530
 531 /**
 532  * Loads the app-supplied constants into the currently set ARB_[vertex/fragment]_programs.
 533  *
 534  * We only support float constants in ARB at the moment, so don't
 535  * worry about the Integers or Booleans
 536  */
 537 /* GL locking is done by the caller (state handler) */
 538 static void shader_arb_load_constants(const struct wined3d_context *context, char usePixelShader, char useVertexShader)
 539 {
 540     IWineD3DDeviceImpl *device = ((IWineD3DSurfaceImpl *)context->surface)->resource.device;
 541     IWineD3DStateBlockImpl* stateBlock = device->stateBlock;
 542     const struct wined3d_gl_info *gl_info = context->gl_info;
 543
 544     if (useVertexShader) {
 545         IWineD3DBaseShaderImpl* vshader = (IWineD3DBaseShaderImpl*) stateBlock->vertexShader;
 546
 547         /* Load DirectX 9 float constants for vertex shader */
 548         device->highest_dirty_vs_const = shader_arb_load_constantsF(vshader, gl_info, GL_VERTEX_PROGRAM_ARB,
 549                 device->highest_dirty_vs_const, stateBlock->vertexShaderConstantF, context->vshader_const_dirty);
 550         shader_arb_vs_local_constants(device);
 551     }
 552
 553     if (usePixelShader) {
 554         IWineD3DBaseShaderImpl* pshader = (IWineD3DBaseShaderImpl*) stateBlock->pixelShader;
 555
 556         /* Load DirectX 9 float constants for pixel shader */
 557         device->highest_dirty_ps_const = shader_arb_load_constantsF(pshader, gl_info, GL_FRAGMENT_PROGRAM_ARB,
 558                 device->highest_dirty_ps_const, stateBlock->pixelShaderConstantF, context->pshader_const_dirty);
 559         shader_arb_ps_local_constants(device);
 560     }
 561 }
 562
 563 static void shader_arb_update_float_vertex_constants(IWineD3DDevice *iface, UINT start, UINT count)
 564 {
 565     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
 566     struct wined3d_context *context = context_get_current();
 567
 568     /* We don't want shader constant dirtification to be an O(contexts), so just dirtify the active
 569      * context. On a context switch the old context will be fully dirtified */
 570     if (!context || ((IWineD3DSurfaceImpl *)context->surface)->resource.device != This) return;
 571
 572     memset(context->vshader_const_dirty + start, 1, sizeof(*context->vshader_const_dirty) * count);
 573     This->highest_dirty_vs_const = max(This->highest_dirty_vs_const, start + count);
 574 }
 575
 576 static void shader_arb_update_float_pixel_constants(IWineD3DDevice *iface, UINT start, UINT count)
 577 {
 578     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
 579     struct wined3d_context *context = context_get_current();
 580
 581     /* We don't want shader constant dirtification to be an O(contexts), so just dirtify the active
 582      * context. On a context switch the old context will be fully dirtified */
 583     if (!context || ((IWineD3DSurfaceImpl *)context->surface)->resource.device != This) return;
 584
 585     memset(context->pshader_const_dirty + start, 1, sizeof(*context->pshader_const_dirty) * count);
 586     This->highest_dirty_ps_const = max(This->highest_dirty_ps_const, start + count);
 587 }
 588
 589 static DWORD *local_const_mapping(IWineD3DBaseShaderImpl *This)
 590 {
 591     DWORD *ret;
 592     DWORD idx = 0;
 593     const local_constant *lconst;
 594
 595     if(This->baseShader.load_local_constsF || list_empty(&This->baseShader.constantsF)) return NULL;
 596
 597     ret = HeapAlloc(GetProcessHeap(), 0, sizeof(DWORD) * This->baseShader.limits.constant_float);
 598     if(!ret) {
 599         ERR("Out of memory\n");
 600         return NULL;
 601     }
 602
 603     LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry) {
 604         ret[lconst->idx] = idx++;
 605     }
 606     return ret;
 607 }
 608
 609 /* Generate the variable & register declarations for the ARB_vertex_program output target */
 610 static DWORD shader_generate_arb_declarations(IWineD3DBaseShader *iface, const shader_reg_maps *reg_maps,
 611         struct wined3d_shader_buffer *buffer, const struct wined3d_gl_info *gl_info, DWORD *lconst_map,
 612         DWORD *num_clipplanes, struct shader_arb_ctx_priv *ctx)
 613 {
 614     IWineD3DBaseShaderImpl* This = (IWineD3DBaseShaderImpl*) iface;
 615     DWORD i, next_local = 0;
 616     char pshader = shader_is_pshader_version(reg_maps->shader_version.type);
 617     unsigned max_constantsF;
 618     const local_constant *lconst;
 619     DWORD map;
 620
 621     /* In pixel shaders, all private constants are program local, we don't need anything
 622      * from program.env. Thus we can advertise the full set of constants in pixel shaders.
 623      * If we need a private constant the GL implementation will squeeze it in somewhere
 624      *
 625      * With vertex shaders we need the posFixup and on some GL implementations 4 helper
 626      * immediate values. The posFixup is loaded using program.env for now, so always
 627      * subtract one from the number of constants. If the shader uses indirect addressing,
 628      * account for the helper const too because we have to declare all availabke d3d constants
 629      * and don't know which are actually used.
 630      */
 631     if (pshader)
 632     {
 633         max_constantsF = gl_info->limits.arb_ps_native_constants;
 634     }
 635     else
 636     {
 637         if(This->baseShader.reg_maps.usesrelconstF) {
 638             DWORD highest_constf = 0, clip_limit;
 639             max_constantsF = gl_info->limits.arb_vs_native_constants - reserved_vs_const(iface, gl_info);
 640             max_constantsF -= count_bits(This->baseShader.reg_maps.integer_constants);
 641
 642             for(i = 0; i < This->baseShader.limits.constant_float; i++)
 643             {
 644                 DWORD idx = i >> 5;
 645                 DWORD shift = i & 0x1f;
 646                 if(reg_maps->constf[idx] & (1 << shift)) highest_constf = i;
 647             }
 648
 649             if(use_nv_clip(gl_info) && ctx->target_version >= NV2)
 650             {
 651                 if(ctx->cur_vs_args->super.clip_enabled)
 652                     clip_limit = gl_info->limits.clipplanes;
 653                 else
 654                     clip_limit = 0;
 655             }
 656             else
 657             {
 658                 unsigned int mask = ctx->cur_vs_args->clip.boolclip.clipplane_mask;
 659                 clip_limit = min(count_bits(mask), 4);
 660             }
 661             *num_clipplanes = min(clip_limit, max_constantsF - highest_constf - 1);
 662             max_constantsF -= *num_clipplanes;
 663             if(*num_clipplanes < clip_limit)
 664             {
 665                 WARN("Only %u clipplanes out of %u enabled\n", *num_clipplanes, gl_info->limits.clipplanes);
 666             }
 667         }
 668         else
 669         {
 670             if (ctx->target_version >= NV2) *num_clipplanes = gl_info->limits.clipplanes;
 671             else *num_clipplanes = min(gl_info->limits.clipplanes, 4);
 672             max_constantsF = gl_info->limits.arb_vs_native_constants;
 673         }
 674     }
 675
 676     for (i = 0, map = reg_maps->temporary; map; map >>= 1, ++i)
 677     {
 678         if (map & 1) shader_addline(buffer, "TEMP R%u;\n", i);
 679     }
 680
 681     for (i = 0, map = reg_maps->address; map; map >>= 1, ++i)
 682     {
 683         if (map & 1) shader_addline(buffer, "ADDRESS A%u;\n", i);
 684     }
 685
 686     if (pshader && reg_maps->shader_version.major == 1 && reg_maps->shader_version.minor <= 3)
 687     {
 688         for (i = 0, map = reg_maps->texcoord; map; map >>= 1, ++i)
 689         {
 690             if (map & 1) shader_addline(buffer, "TEMP T%u;\n", i);
 691         }
 692     }
 693
 694     /* Load local constants using the program-local space,
 695      * this avoids reloading them each time the shader is used
 696      */
 697     if(lconst_map) {
 698         LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry) {
 699             shader_addline(buffer, "PARAM C%u = program.local[%u];\n", lconst->idx,
 700                            lconst_map[lconst->idx]);
 701             next_local = max(next_local, lconst_map[lconst->idx] + 1);
 702         }
 703     }
 704
 705     /* After subtracting privately used constants from the hardware limit(they are loaded as
 706      * local constants), make sure the shader doesn't violate the env constant limit
 707      */
 708     if(pshader)
 709     {
 710         max_constantsF = min(max_constantsF, gl_info->limits.arb_ps_float_constants);
 711     }
 712     else
 713     {
 714         max_constantsF = min(max_constantsF, gl_info->limits.arb_vs_float_constants);
 715     }
 716
 717     /* Avoid declaring more constants than needed */
 718     max_constantsF = min(max_constantsF, This->baseShader.limits.constant_float);
 719
 720     /* we use the array-based constants array if the local constants are marked for loading,
 721      * because then we use indirect addressing, or when the local constant list is empty,
 722      * because then we don't know if we're using indirect addressing or not. If we're hardcoding
 723      * local constants do not declare the loaded constants as an array because ARB compilers usually
 724      * do not optimize unused constants away
 725      */
 726     if(This->baseShader.reg_maps.usesrelconstF) {
 727         /* Need to PARAM the environment parameters (constants) so we can use relative addressing */
 728         shader_addline(buffer, "PARAM C[%d] = { program.env[0..%d] };\n",
 729                     max_constantsF, max_constantsF - 1);
 730     } else {
 731         for(i = 0; i < max_constantsF; i++) {
 732             DWORD idx, mask;
 733             idx = i >> 5;
 734             mask = 1 << (i & 0x1f);
 735             if(!shader_constant_is_local(This, i) && (This->baseShader.reg_maps.constf[idx] & mask)) {
 736                 shader_addline(buffer, "PARAM C%d = program.env[%d];\n",i, i);
 737             }
 738         }
 739     }
 740
 741     return next_local;
 742 }
 743
 744 static const char * const shift_tab[] = {
 745     "dummy",     /*  0 (none) */
 746     "coefmul.x", /*  1 (x2)   */
 747     "coefmul.y", /*  2 (x4)   */
 748     "coefmul.z", /*  3 (x8)   */
 749     "coefmul.w", /*  4 (x16)  */
 750     "dummy",     /*  5 (x32)  */
 751     "dummy",     /*  6 (x64)  */
 752     "dummy",     /*  7 (x128) */
 753     "dummy",     /*  8 (d256) */
 754     "dummy",     /*  9 (d128) */
 755     "dummy",     /* 10 (d64)  */
 756     "dummy",     /* 11 (d32)  */
 757     "coefdiv.w", /* 12 (d16)  */
 758     "coefdiv.z", /* 13 (d8)   */
 759     "coefdiv.y", /* 14 (d4)   */
 760     "coefdiv.x"  /* 15 (d2)   */
 761 };
 762
 763 static void shader_arb_get_write_mask(const struct wined3d_shader_instruction *ins,
 764         const struct wined3d_shader_dst_param *dst, char *write_mask)
 765 {
 766     char *ptr = write_mask;
 767
 768     if (dst->write_mask != WINED3DSP_WRITEMASK_ALL)
 769     {
 770         *ptr++ = '.';
 771         if (dst->write_mask & WINED3DSP_WRITEMASK_0) *ptr++ = 'x';
 772         if (dst->write_mask & WINED3DSP_WRITEMASK_1) *ptr++ = 'y';
 773         if (dst->write_mask & WINED3DSP_WRITEMASK_2) *ptr++ = 'z';
 774         if (dst->write_mask & WINED3DSP_WRITEMASK_3) *ptr++ = 'w';
 775     }
 776
 777     *ptr = '\0';
 778 }
 779
 780 static void shader_arb_get_swizzle(const struct wined3d_shader_src_param *param, BOOL fixup, char *swizzle_str)
 781 {
 782     /* For registers of type WINED3DDECLTYPE_D3DCOLOR, data is stored as "bgra",
 783      * but addressed as "rgba". To fix this we need to swap the register's x
 784      * and z components. */
 785     const char *swizzle_chars = fixup ? "zyxw" : "xyzw";
 786     char *ptr = swizzle_str;
 787
 788     /* swizzle bits fields: wwzzyyxx */
 789     DWORD swizzle = param->swizzle;
 790     DWORD swizzle_x = swizzle & 0x03;
 791     DWORD swizzle_y = (swizzle >> 2) & 0x03;
 792     DWORD swizzle_z = (swizzle >> 4) & 0x03;
 793     DWORD swizzle_w = (swizzle >> 6) & 0x03;
 794
 795     /* If the swizzle is the default swizzle (ie, "xyzw"), we don't need to
 796      * generate a swizzle string. Unless we need to our own swizzling. */
 797     if (swizzle != WINED3DSP_NOSWIZZLE || fixup)
 798     {
 799         *ptr++ = '.';
 800         if (swizzle_x == swizzle_y && swizzle_x == swizzle_z && swizzle_x == swizzle_w) {
 801             *ptr++ = swizzle_chars[swizzle_x];
 802         } else {
 803             *ptr++ = swizzle_chars[swizzle_x];
 804             *ptr++ = swizzle_chars[swizzle_y];
 805             *ptr++ = swizzle_chars[swizzle_z];
 806             *ptr++ = swizzle_chars[swizzle_w];
 807         }
 808     }
 809
 810     *ptr = '\0';
 811 }
 812
 813 static void shader_arb_request_a0(const struct wined3d_shader_instruction *ins, const char *src)
 814 {
 815     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
 816     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
 817
 818     if(strcmp(priv->addr_reg, src) == 0) return;
 819
 820     strcpy(priv->addr_reg, src);
 821     shader_addline(buffer, "ARL A0.x, %s;\n", src);
 822 }
 823
 824 static void shader_arb_get_src_param(const struct wined3d_shader_instruction *ins,
 825         const struct wined3d_shader_src_param *src, unsigned int tmpreg, char *outregstr);
 826
 827 static void shader_arb_get_register_name(const struct wined3d_shader_instruction *ins,
 828         const struct wined3d_shader_register *reg, char *register_name, BOOL *is_color)
 829 {
 830     /* oPos, oFog and oPts in D3D */
 831     static const char * const rastout_reg_names[] = {"TMP_OUT", "result.fogcoord", "result.pointsize"};
 832     IWineD3DBaseShaderImpl *This = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
 833     BOOL pshader = shader_is_pshader_version(This->baseShader.reg_maps.shader_version.type);
 834     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
 835
 836     *is_color = FALSE;
 837
 838     switch (reg->type)
 839     {
 840         case WINED3DSPR_TEMP:
 841             sprintf(register_name, "R%u", reg->idx);
 842             break;
 843
 844         case WINED3DSPR_INPUT:
 845             if (pshader)
 846             {
 847                 if(This->baseShader.reg_maps.shader_version.major < 3)
 848                 {
 849                     if (reg->idx == 0) strcpy(register_name, "fragment.color.primary");
 850                     else strcpy(register_name, "fragment.color.secondary");
 851                 }
 852                 else
 853                 {
 854                     if(reg->rel_addr)
 855                     {
 856                         char rel_reg[50];
 857                         shader_arb_get_src_param(ins, reg->rel_addr, 0, rel_reg);
 858
 859                         if(strcmp(rel_reg, "**aL_emul**") == 0)
 860                         {
 861                             DWORD idx = ctx->aL + reg->idx;
 862                             if(idx < MAX_REG_INPUT)
 863                             {
 864                                 strcpy(register_name, ctx->ps_input[idx]);
 865                             }
 866                             else
 867                             {
 868                                 ERR("Pixel shader input register out of bounds: %u\n", idx);
 869                                 sprintf(register_name, "out_of_bounds_%u", idx);
 870                             }
 871                         }
 872                         else if(This->baseShader.reg_maps.input_registers & 0x0300)
 873                         {
 874                             /* There are two ways basically:
 875                              *
 876                              * 1) Use the unrolling code that is used for loop emulation and unroll the loop.
 877                              *    That means trouble if the loop also contains a breakc or if the control values
 878                              *    aren't local constants.
 879                              * 2) Generate an if block that checks if aL.y < 8, == 8 or == 9 and selects the
 880                              *    source dynamically. The trouble is that we cannot simply read aL.y because it
 881                              *    is an ADDRESS register. We could however push it, load .zw with a value and use
 882                              *    ADAC to load the condition code register and pop it again afterwards
 883                              */
 884                             FIXME("Relative input register addressing with more than 8 registers\n");
 885
 886                             /* This is better than nothing for now */
 887                             sprintf(register_name, "fragment.texcoord[%s + %u]", rel_reg, reg->idx);
 888                         }
 889                         else if(ctx->cur_ps_args->super.vp_mode != vertexshader)
 890                         {
 891                             /* This is problematic because we'd have to consult the ctx->ps_input strings
 892                              * for where to find the varying. Some may be "0.0", others can be texcoords or
 893                              * colors. This needs either a pipeline replacement to make the vertex shader feed
 894                              * proper varyings, or loop unrolling
 895                              *
 896                              * For now use the texcoords and hope for the best
 897                              */
 898                             FIXME("Non-vertex shader varying input with indirect addressing\n");
 899                             sprintf(register_name, "fragment.texcoord[%s + %u]", rel_reg, reg->idx);
 900                         }
 901                         else
 902                         {
 903                             /* D3D supports indirect addressing only with aL in loop registers. The loop instruction
 904                              * pulls GL_NV_fragment_program2 in
 905                              */
 906                             sprintf(register_name, "fragment.texcoord[%s + %u]", rel_reg, reg->idx);
 907                         }
 908                     }
 909                     else
 910                     {
 911                         if(reg->idx < MAX_REG_INPUT)
 912                         {
 913                             strcpy(register_name, ctx->ps_input[reg->idx]);
 914                         }
 915                         else
 916                         {
 917                             ERR("Pixel shader input register out of bounds: %u\n", reg->idx);
 918                             sprintf(register_name, "out_of_bounds_%u", reg->idx);
 919                         }
 920                     }
 921                 }
 922             }
 923             else
 924             {
 925                 if (ctx->cur_vs_args->super.swizzle_map & (1 << reg->idx)) *is_color = TRUE;
 926                 sprintf(register_name, "vertex.attrib[%u]", reg->idx);
 927             }
 928             break;
 929
 930         case WINED3DSPR_CONST:
 931             if (!pshader && reg->rel_addr)
 932             {
 933                 BOOL aL = FALSE;
 934                 char rel_reg[50];
 935                 UINT rel_offset = ((IWineD3DVertexShaderImpl *)This)->rel_offset;
 936                 if(This->baseShader.reg_maps.shader_version.major < 2) {
 937                     sprintf(rel_reg, "A0.x");
 938                 } else {
 939                     shader_arb_get_src_param(ins, reg->rel_addr, 0, rel_reg);
 940                     if(ctx->target_version == ARB) {
 941                         if(strcmp(rel_reg, "**aL_emul**") == 0) {
 942                             aL = TRUE;
 943                         } else {
 944                             shader_arb_request_a0(ins, rel_reg);
 945                             sprintf(rel_reg, "A0.x");
 946                         }
 947                     }
 948                 }
 949                 if(aL)
 950                     sprintf(register_name, "C[%u]", ctx->aL + reg->idx);
 951                 else if (reg->idx >= rel_offset)
 952                     sprintf(register_name, "C[%s + %u]", rel_reg, reg->idx - rel_offset);
 953                 else
 954                     sprintf(register_name, "C[%s - %u]", rel_reg, -reg->idx + rel_offset);
 955             }
 956             else
 957             {
 958                 if (This->baseShader.reg_maps.usesrelconstF)
 959                     sprintf(register_name, "C[%u]", reg->idx);
 960                 else
 961                     sprintf(register_name, "C%u", reg->idx);
 962             }
 963             break;
 964
 965         case WINED3DSPR_TEXTURE: /* case WINED3DSPR_ADDR: */
 966             if (pshader) {
 967                 if(This->baseShader.reg_maps.shader_version.major == 1 &&
 968                    This->baseShader.reg_maps.shader_version.minor <= 3) {
 969                     /* In ps <= 1.3, Tx is a temporary register as destination to all instructions,
 970                      * and as source to most instructions. For some instructions it is the texcoord
 971                      * input. Those instructions know about the special use
 972                      */
 973                     sprintf(register_name, "T%u", reg->idx);
 974                 } else {
 975                     /* in ps 1.4 and 2.x Tx is always a (read-only) varying */
 976                     sprintf(register_name, "fragment.texcoord[%u]", reg->idx);
 977                 }
 978             }
 979             else
 980             {
 981                 if(This->baseShader.reg_maps.shader_version.major == 1 || ctx->target_version >= NV2)
 982                 {
 983                     sprintf(register_name, "A%u", reg->idx);
 984                 }
 985                 else
 986                 {
 987                     sprintf(register_name, "A%u_SHADOW", reg->idx);
 988                 }
 989             }
 990             break;
 991
 992         case WINED3DSPR_COLOROUT:
 993             if(ctx->cur_ps_args->super.srgb_correction && reg->idx == 0)
 994             {
 995                 strcpy(register_name, "TMP_COLOR");
 996             }
 997             else
 998             {
 999                 if(ctx->cur_ps_args->super.srgb_correction) FIXME("sRGB correction on higher render targets\n");
1000                 if(This->baseShader.reg_maps.highest_render_target > 0)
1001                 {
1002                     sprintf(register_name, "result.color[%u]", reg->idx);
1003                 }
1004                 else
1005                 {
1006                     strcpy(register_name, "result.color");
1007                 }
1008             }
1009             break;
1010
1011         case WINED3DSPR_RASTOUT:
1012             if(reg->idx == 1) sprintf(register_name, "%s", ctx->fog_output);
1013             else sprintf(register_name, "%s", rastout_reg_names[reg->idx]);
1014             break;
1015
1016         case WINED3DSPR_DEPTHOUT:
1017             strcpy(register_name, "result.depth");
1018             break;
1019
1020         case WINED3DSPR_ATTROUT:
1021         /* case WINED3DSPR_OUTPUT: */
1022             if (pshader) sprintf(register_name, "oD[%u]", reg->idx);
1023             else strcpy(register_name, ctx->color_output[reg->idx]);
1024             break;
1025
1026         case WINED3DSPR_TEXCRDOUT:
1027             if (pshader)
1028             {
1029                 sprintf(register_name, "oT[%u]", reg->idx);
1030             }
1031             else
1032             {
1033                 if(This->baseShader.reg_maps.shader_version.major < 3)
1034                 {
1035                     strcpy(register_name, ctx->texcrd_output[reg->idx]);
1036                 }
1037                 else
1038                 {
1039                     strcpy(register_name, ctx->vs_output[reg->idx]);
1040                 }
1041             }
1042             break;
1043
1044         case WINED3DSPR_LOOP:
1045             if(ctx->target_version >= NV2)
1046             {
1047                 /* Pshader has an implicitly declared loop index counter A0.x that cannot be renamed */
1048                 if(pshader) sprintf(register_name, "A0.x");
1049                 else sprintf(register_name, "aL.y");
1050             }
1051             else
1052             {
1053                 /* Unfortunately this code cannot return the value of ctx->aL here. An immediate value
1054                  * would be valid, but if aL is used for indexing(its only use), there's likely an offset,
1055                  * thus the result would be something like C[15 + 30], which is not valid in the ARB program
1056                  * grammar. So return a marker for the emulated aL and intercept it in constant and varying
1057                  * indexing
1058                  */
1059                 sprintf(register_name, "**aL_emul**");
1060             }
1061
1062             break;
1063
1064         case WINED3DSPR_CONSTINT:
1065             sprintf(register_name, "I%u", reg->idx);
1066             break;
1067
1068         case WINED3DSPR_MISCTYPE:
1069             if(reg->idx == 0)
1070             {
1071                 sprintf(register_name, "vpos");
1072             }
1073             else if(reg->idx == 1)
1074             {
1075                 sprintf(register_name, "fragment.facing.x");
1076             }
1077             else
1078             {
1079                 FIXME("Unknown MISCTYPE register index %u\n", reg->idx);
1080             }
1081             break;
1082
1083         default:
1084             FIXME("Unhandled register type %#x[%u]\n", reg->type, reg->idx);
1085             sprintf(register_name, "unrecognized_register[%u]", reg->idx);
1086             break;
1087     }
1088 }
1089
1090 static void shader_arb_get_dst_param(const struct wined3d_shader_instruction *ins,
1091         const struct wined3d_shader_dst_param *wined3d_dst, char *str)
1092 {
1093     char register_name[255];
1094     char write_mask[6];
1095     BOOL is_color;
1096
1097     shader_arb_get_register_name(ins, &wined3d_dst->reg, register_name, &is_color);
1098     strcpy(str, register_name);
1099
1100     shader_arb_get_write_mask(ins, wined3d_dst, write_mask);
1101     strcat(str, write_mask);
1102 }
1103
1104 static const char *shader_arb_get_fixup_swizzle(enum fixup_channel_source channel_source)
1105 {
1106     switch(channel_source)
1107     {
1108         case CHANNEL_SOURCE_ZERO: return "0";
1109         case CHANNEL_SOURCE_ONE: return "1";
1110         case CHANNEL_SOURCE_X: return "x";
1111         case CHANNEL_SOURCE_Y: return "y";
1112         case CHANNEL_SOURCE_Z: return "z";
1113         case CHANNEL_SOURCE_W: return "w";
1114         default:
1115             FIXME("Unhandled channel source %#x\n", channel_source);
1116             return "undefined";
1117     }
1118 }
1119
1120 static void gen_color_correction(struct wined3d_shader_buffer *buffer, const char *reg,
1121         DWORD dst_mask, const char *one, const char *two, struct color_fixup_desc fixup)
1122 {
1123     DWORD mask;
1124
1125     if (is_yuv_fixup(fixup))
1126     {
1127         enum yuv_fixup yuv_fixup = get_yuv_fixup(fixup);
1128         FIXME("YUV fixup (%#x) not supported\n", yuv_fixup);
1129         return;
1130     }
1131
1132     mask = 0;
1133     if (fixup.x_source != CHANNEL_SOURCE_X) mask |= WINED3DSP_WRITEMASK_0;
1134     if (fixup.y_source != CHANNEL_SOURCE_Y) mask |= WINED3DSP_WRITEMASK_1;
1135     if (fixup.z_source != CHANNEL_SOURCE_Z) mask |= WINED3DSP_WRITEMASK_2;
1136     if (fixup.w_source != CHANNEL_SOURCE_W) mask |= WINED3DSP_WRITEMASK_3;
1137     mask &= dst_mask;
1138
1139     if (mask)
1140     {
1141         shader_addline(buffer, "SWZ %s, %s, %s, %s, %s, %s;\n", reg, reg,
1142                 shader_arb_get_fixup_swizzle(fixup.x_source), shader_arb_get_fixup_swizzle(fixup.y_source),
1143                 shader_arb_get_fixup_swizzle(fixup.z_source), shader_arb_get_fixup_swizzle(fixup.w_source));
1144     }
1145
1146     mask = 0;
1147     if (fixup.x_sign_fixup) mask |= WINED3DSP_WRITEMASK_0;
1148     if (fixup.y_sign_fixup) mask |= WINED3DSP_WRITEMASK_1;
1149     if (fixup.z_sign_fixup) mask |= WINED3DSP_WRITEMASK_2;
1150     if (fixup.w_sign_fixup) mask |= WINED3DSP_WRITEMASK_3;
1151     mask &= dst_mask;
1152
1153     if (mask)
1154     {
1155         char reg_mask[6];
1156         char *ptr = reg_mask;
1157
1158         if (mask != WINED3DSP_WRITEMASK_ALL)
1159         {
1160             *ptr++ = '.';
1161             if (mask & WINED3DSP_WRITEMASK_0) *ptr++ = 'x';
1162             if (mask & WINED3DSP_WRITEMASK_1) *ptr++ = 'y';
1163             if (mask & WINED3DSP_WRITEMASK_2) *ptr++ = 'z';
1164             if (mask & WINED3DSP_WRITEMASK_3) *ptr++ = 'w';
1165         }
1166         *ptr = '\0';
1167
1168         shader_addline(buffer, "MAD %s%s, %s, %s, -%s;\n", reg, reg_mask, reg, two, one);
1169     }
1170 }
1171
1172 static const char *shader_arb_get_modifier(const struct wined3d_shader_instruction *ins)
1173 {
1174     DWORD mod;
1175     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
1176     if (!ins->dst_count) return "";
1177
1178     mod = ins->dst[0].modifiers;
1179
1180     /* Silently ignore PARTIALPRECISION if its not supported */
1181     if(priv->target_version == ARB) mod &= ~WINED3DSPDM_PARTIALPRECISION;
1182
1183     if(mod & WINED3DSPDM_MSAMPCENTROID)
1184     {
1185         FIXME("Unhandled modifier WINED3DSPDM_MSAMPCENTROID\n");
1186         mod &= ~WINED3DSPDM_MSAMPCENTROID;
1187     }
1188
1189     switch(mod)
1190     {
1191         case WINED3DSPDM_SATURATE | WINED3DSPDM_PARTIALPRECISION:
1192             return "H_SAT";
1193
1194         case WINED3DSPDM_SATURATE:
1195             return "_SAT";
1196
1197         case WINED3DSPDM_PARTIALPRECISION:
1198             return "H";
1199
1200         case 0:
1201             return "";
1202
1203         default:
1204             FIXME("Unknown modifiers 0x%08x\n", mod);
1205             return "";
1206     }
1207 }
1208
1209 #define TEX_PROJ        0x1
1210 #define TEX_BIAS        0x2
1211 #define TEX_LOD         0x4
1212 #define TEX_DERIV       0x10
1213
1214 static void shader_hw_sample(const struct wined3d_shader_instruction *ins, DWORD sampler_idx,
1215         const char *dst_str, const char *coord_reg, WORD flags, const char *dsx, const char *dsy)
1216 {
1217     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1218     DWORD sampler_type = ins->ctx->reg_maps->sampler_type[sampler_idx];
1219     const char *tex_type;
1220     BOOL np2_fixup = FALSE;
1221     IWineD3DBaseShaderImpl *This = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
1222     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *) This->baseShader.device;
1223     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
1224     const char *mod;
1225     BOOL pshader = shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type);
1226
1227     /* D3D vertex shader sampler IDs are vertex samplers(0-3), not global d3d samplers */
1228     if(!pshader) sampler_idx += MAX_FRAGMENT_SAMPLERS;
1229
1230     switch(sampler_type) {
1231         case WINED3DSTT_1D:
1232             tex_type = "1D";
1233             break;
1234
1235         case WINED3DSTT_2D:
1236             if(device->stateBlock->textures[sampler_idx] &&
1237                IWineD3DBaseTexture_GetTextureDimensions(device->stateBlock->textures[sampler_idx]) == GL_TEXTURE_RECTANGLE_ARB) {
1238                 tex_type = "RECT";
1239             } else {
1240                 tex_type = "2D";
1241             }
1242             if (shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type))
1243             {
1244                 if (priv->cur_np2fixup_info->super.active & (1 << sampler_idx))
1245                 {
1246                     if (flags) FIXME("Only ordinary sampling from NP2 textures is supported.\n");
1247                     else np2_fixup = TRUE;
1248                 }
1249             }
1250             break;
1251
1252         case WINED3DSTT_VOLUME:
1253             tex_type = "3D";
1254             break;
1255
1256         case WINED3DSTT_CUBE:
1257             tex_type = "CUBE";
1258             break;
1259
1260         default:
1261             ERR("Unexpected texture type %d\n", sampler_type);
1262             tex_type = "";
1263     }
1264
1265     /* TEX, TXL, TXD and TXP do not support the "H" modifier,
1266      * so don't use shader_arb_get_modifier
1267      */
1268     if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE) mod = "_SAT";
1269     else mod = "";
1270
1271     /* Fragment samplers always have indentity mapping */
1272     if(sampler_idx >= MAX_FRAGMENT_SAMPLERS)
1273     {
1274         sampler_idx = priv->cur_vs_args->vertex.samplers[sampler_idx - MAX_FRAGMENT_SAMPLERS];
1275     }
1276
1277     if (flags & TEX_DERIV)
1278     {
1279         if(flags & TEX_PROJ) FIXME("Projected texture sampling with custom derivatives\n");
1280         if(flags & TEX_BIAS) FIXME("Biased texture sampling with custom derivatives\n");
1281         shader_addline(buffer, "TXD%s %s, %s, %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg,
1282                        dsx, dsy,sampler_idx, tex_type);
1283     }
1284     else if(flags & TEX_LOD)
1285     {
1286         if(flags & TEX_PROJ) FIXME("Projected texture sampling with explicit lod\n");
1287         if(flags & TEX_BIAS) FIXME("Biased texture sampling with explicit lod\n");
1288         shader_addline(buffer, "TXL%s %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg,
1289                        sampler_idx, tex_type);
1290     }
1291     else if (flags & TEX_BIAS)
1292     {
1293         /* Shouldn't be possible, but let's check for it */
1294         if(flags & TEX_PROJ) FIXME("Biased and Projected texture sampling\n");
1295         /* TXB takes the 4th component of the source vector automatically, as d3d. Nothing more to do */
1296         shader_addline(buffer, "TXB%s %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg, sampler_idx, tex_type);
1297     }
1298     else if (flags & TEX_PROJ)
1299     {
1300         shader_addline(buffer, "TXP%s %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg, sampler_idx, tex_type);
1301     }
1302     else
1303     {
1304         if (np2_fixup)
1305         {
1306             const unsigned char idx = priv->cur_np2fixup_info->super.idx[sampler_idx];
1307             shader_addline(buffer, "MUL TA, np2fixup[%u].%s, %s;\n", idx >> 1,
1308                            (idx % 2) ? "zwxy" : "xyzw", coord_reg);
1309
1310             shader_addline(buffer, "TEX%s %s, TA, texture[%u], %s;\n", mod, dst_str, sampler_idx, tex_type);
1311         }
1312         else
1313             shader_addline(buffer, "TEX%s %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg, sampler_idx, tex_type);
1314     }
1315
1316     if (pshader)
1317     {
1318         gen_color_correction(buffer, dst_str, ins->dst[0].write_mask,
1319                 "one", "coefmul.x", priv->cur_ps_args->super.color_fixup[sampler_idx]);
1320     }
1321 }
1322
1323 static void shader_arb_get_src_param(const struct wined3d_shader_instruction *ins,
1324         const struct wined3d_shader_src_param *src, unsigned int tmpreg, char *outregstr)
1325 {
1326     /* Generate a line that does the input modifier computation and return the input register to use */
1327     BOOL is_color = FALSE;
1328     char regstr[256];
1329     char swzstr[20];
1330     int insert_line;
1331     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1332     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
1333
1334     /* Assume a new line will be added */
1335     insert_line = 1;
1336
1337     /* Get register name */
1338     shader_arb_get_register_name(ins, &src->reg, regstr, &is_color);
1339     shader_arb_get_swizzle(src, is_color, swzstr);
1340
1341     switch (src->modifiers)
1342     {
1343     case WINED3DSPSM_NONE:
1344         sprintf(outregstr, "%s%s", regstr, swzstr);
1345         insert_line = 0;
1346         break;
1347     case WINED3DSPSM_NEG:
1348         sprintf(outregstr, "-%s%s", regstr, swzstr);
1349         insert_line = 0;
1350         break;
1351     case WINED3DSPSM_BIAS:
1352         shader_addline(buffer, "ADD T%c, %s, -coefdiv.x;\n", 'A' + tmpreg, regstr);
1353         break;
1354     case WINED3DSPSM_BIASNEG:
1355         shader_addline(buffer, "ADD T%c, -%s, coefdiv.x;\n", 'A' + tmpreg, regstr);
1356         break;
1357     case WINED3DSPSM_SIGN:
1358         shader_addline(buffer, "MAD T%c, %s, coefmul.x, -one.x;\n", 'A' + tmpreg, regstr);
1359         break;
1360     case WINED3DSPSM_SIGNNEG:
1361         shader_addline(buffer, "MAD T%c, %s, -coefmul.x, one.x;\n", 'A' + tmpreg, regstr);
1362         break;
1363     case WINED3DSPSM_COMP:
1364         shader_addline(buffer, "SUB T%c, one.x, %s;\n", 'A' + tmpreg, regstr);
1365         break;
1366     case WINED3DSPSM_X2:
1367         shader_addline(buffer, "ADD T%c, %s, %s;\n", 'A' + tmpreg, regstr, regstr);
1368         break;
1369     case WINED3DSPSM_X2NEG:
1370         shader_addline(buffer, "ADD T%c, -%s, -%s;\n", 'A' + tmpreg, regstr, regstr);
1371         break;
1372     case WINED3DSPSM_DZ:
1373         shader_addline(buffer, "RCP T%c, %s.z;\n", 'A' + tmpreg, regstr);
1374         shader_addline(buffer, "MUL T%c, %s, T%c;\n", 'A' + tmpreg, regstr, 'A' + tmpreg);
1375         break;
1376     case WINED3DSPSM_DW:
1377         shader_addline(buffer, "RCP T%c, %s.w;\n", 'A' + tmpreg, regstr);
1378         shader_addline(buffer, "MUL T%c, %s, T%c;\n", 'A' + tmpreg, regstr, 'A' + tmpreg);
1379         break;
1380     case WINED3DSPSM_ABS:
1381         if(ctx->target_version >= NV2) {
1382             sprintf(outregstr, "|%s%s|", regstr, swzstr);
1383             insert_line = 0;
1384         } else {
1385             shader_addline(buffer, "ABS T%c, %s;\n", 'A' + tmpreg, regstr);
1386         }
1387         break;
1388     case WINED3DSPSM_ABSNEG:
1389         if(ctx->target_version >= NV2) {
1390             sprintf(outregstr, "-|%s%s|", regstr, swzstr);
1391         } else {
1392             shader_addline(buffer, "ABS T%c, %s;\n", 'A' + tmpreg, regstr);
1393             sprintf(outregstr, "-T%c%s", 'A' + tmpreg, swzstr);
1394         }
1395         insert_line = 0;
1396         break;
1397     default:
1398         sprintf(outregstr, "%s%s", regstr, swzstr);
1399         insert_line = 0;
1400     }
1401
1402     /* Return modified or original register, with swizzle */
1403     if (insert_line)
1404         sprintf(outregstr, "T%c%s", 'A' + tmpreg, swzstr);
1405 }
1406
1407 static void pshader_hw_bem(const struct wined3d_shader_instruction *ins)
1408 {
1409     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1410     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1411     char dst_name[50];
1412     char src_name[2][50];
1413     DWORD sampler_code = dst->reg.idx;
1414
1415     shader_arb_get_dst_param(ins, dst, dst_name);
1416
1417     /* Sampling the perturbation map in Tsrc was done already, including the signedness correction if needed
1418      *
1419      * Keep in mind that src_name[1] can be "TB" and src_name[0] can be "TA" because modifiers like _x2 are valid
1420      * with bem. So delay loading the first parameter until after the perturbation calculation which needs two
1421      * temps is done.
1422      */
1423     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1424     shader_addline(buffer, "SWZ TA, bumpenvmat%d, x, z, 0, 0;\n", sampler_code);
1425     shader_addline(buffer, "DP3 TC.r, TA, %s;\n", src_name[1]);
1426     shader_addline(buffer, "SWZ TA, bumpenvmat%d, y, w, 0, 0;\n", sampler_code);
1427     shader_addline(buffer, "DP3 TC.g, TA, %s;\n", src_name[1]);
1428
1429     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
1430     shader_addline(buffer, "ADD %s, %s, TC;\n", dst_name, src_name[0]);
1431 }
1432
1433 static DWORD negate_modifiers(DWORD mod, char *extra_char)
1434 {
1435     *extra_char = ' ';
1436     switch(mod)
1437     {
1438         case WINED3DSPSM_NONE:      return WINED3DSPSM_NEG;
1439         case WINED3DSPSM_NEG:       return WINED3DSPSM_NONE;
1440         case WINED3DSPSM_BIAS:      return WINED3DSPSM_BIASNEG;
1441         case WINED3DSPSM_BIASNEG:   return WINED3DSPSM_BIAS;
1442         case WINED3DSPSM_SIGN:      return WINED3DSPSM_SIGNNEG;
1443         case WINED3DSPSM_SIGNNEG:   return WINED3DSPSM_SIGN;
1444         case WINED3DSPSM_COMP:      *extra_char = '-'; return WINED3DSPSM_COMP;
1445         case WINED3DSPSM_X2:        return WINED3DSPSM_X2NEG;
1446         case WINED3DSPSM_X2NEG:     return WINED3DSPSM_X2;
1447         case WINED3DSPSM_DZ:        *extra_char = '-'; return WINED3DSPSM_DZ;
1448         case WINED3DSPSM_DW:        *extra_char = '-'; return WINED3DSPSM_DW;
1449         case WINED3DSPSM_ABS:       return WINED3DSPSM_ABSNEG;
1450         case WINED3DSPSM_ABSNEG:    return WINED3DSPSM_ABS;
1451     }
1452     FIXME("Unknown modifier %u\n", mod);
1453     return mod;
1454 }
1455
1456 static void pshader_hw_cnd(const struct wined3d_shader_instruction *ins)
1457 {
1458     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1459     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1460     char dst_name[50];
1461     char src_name[3][50];
1462     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
1463             ins->ctx->reg_maps->shader_version.minor);
1464     BOOL is_color;
1465
1466     shader_arb_get_dst_param(ins, dst, dst_name);
1467     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1468
1469     /* The coissue flag changes the semantic of the cnd instruction in <= 1.3 shaders */
1470     if (shader_version <= WINED3D_SHADER_VERSION(1, 3) && ins->coissue)
1471     {
1472         shader_addline(buffer, "MOV%s %s, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name[1]);
1473     } else {
1474         struct wined3d_shader_src_param src0_copy = ins->src[0];
1475         char extra_neg;
1476
1477         /* src0 may have a negate srcmod set, so we can't blindly add "-" to the name */
1478         src0_copy.modifiers = negate_modifiers(src0_copy.modifiers, &extra_neg);
1479
1480         shader_arb_get_src_param(ins, &src0_copy, 0, src_name[0]);
1481         shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
1482         shader_addline(buffer, "ADD TA, %c%s, coefdiv.x;\n", extra_neg, src_name[0]);
1483         /* No modifiers supported on CMP */
1484         shader_addline(buffer, "CMP %s, TA, %s, %s;\n", dst_name, src_name[1], src_name[2]);
1485
1486         /* _SAT on CMP doesn't make much sense, but it is not a pure NOP */
1487         if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE)
1488         {
1489             shader_arb_get_register_name(ins, &dst->reg, src_name[0], &is_color);
1490             shader_addline(buffer, "MOV_SAT %s, %s;\n", dst_name, dst_name);
1491         }
1492     }
1493 }
1494
1495 static void pshader_hw_cmp(const struct wined3d_shader_instruction *ins)
1496 {
1497     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1498     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1499     char dst_name[50];
1500     char src_name[3][50];
1501     BOOL is_color;
1502
1503     shader_arb_get_dst_param(ins, dst, dst_name);
1504
1505     /* Generate input register names (with modifiers) */
1506     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
1507     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1508     shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
1509
1510     /* No modifiers are supported on CMP */
1511     shader_addline(buffer, "CMP %s, %s, %s, %s;\n", dst_name,
1512                    src_name[0], src_name[2], src_name[1]);
1513
1514     if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE)
1515     {
1516         shader_arb_get_register_name(ins, &dst->reg, src_name[0], &is_color);
1517         shader_addline(buffer, "MOV_SAT %s, %s;\n", dst_name, src_name[0]);
1518     }
1519 }
1520
1521 /** Process the WINED3DSIO_DP2ADD instruction in ARB.
1522  * dst = dot2(src0, src1) + src2 */
1523 static void pshader_hw_dp2add(const struct wined3d_shader_instruction *ins)
1524 {
1525     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1526     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1527     char dst_name[50];
1528     char src_name[3][50];
1529     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
1530
1531     shader_arb_get_dst_param(ins, dst, dst_name);
1532     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
1533     shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
1534
1535     if(ctx->target_version >= NV3)
1536     {
1537         /* GL_NV_fragment_program2 has a 1:1 matching instruction */
1538         shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1539         shader_addline(buffer, "DP2A%s %s, %s, %s, %s;\n", shader_arb_get_modifier(ins),
1540                        dst_name, src_name[0], src_name[1], src_name[2]);
1541     }
1542     else if(ctx->target_version >= NV2)
1543     {
1544         /* dst.x = src2.?, src0.x, src1.x + src0.y * src1.y
1545          * dst.y = src2.?, src0.x, src1.z + src0.y * src1.w
1546          * dst.z = src2.?, src0.x, src1.x + src0.y * src1.y
1547          * dst.z = src2.?, src0.x, src1.z + src0.y * src1.w
1548          *
1549          * Make sure that src1.zw = src1.xy, then we get a classic dp2add
1550          *
1551          * .xyxy and other swizzles that we could get with this are not valid in
1552          * plain ARBfp, but luckily the NV extension grammar lifts this limitation.
1553          */
1554         struct wined3d_shader_src_param tmp_param = ins->src[1];
1555         DWORD swizzle = tmp_param.swizzle & 0xf; /* Selects .xy */
1556         tmp_param.swizzle = swizzle | (swizzle << 4); /* Creates .xyxy */
1557
1558         shader_arb_get_src_param(ins, &tmp_param, 1, src_name[1]);
1559
1560         shader_addline(buffer, "X2D%s %s, %s, %s, %s;\n", shader_arb_get_modifier(ins),
1561                        dst_name, src_name[2], src_name[0], src_name[1]);
1562     }
1563     else
1564     {
1565         shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1566         /* 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]
1567         * src_name[0] can be TA, but TA is a private temp for modifiers, so it is save to overwrite
1568         */
1569         shader_addline(buffer, "MOV TA, %s;\n", src_name[0]);
1570         shader_addline(buffer, "MOV TA.z, 0.0;\n");
1571         shader_addline(buffer, "DP3 TA, TA, %s;\n", src_name[1]);
1572         shader_addline(buffer, "ADD%s %s, TA, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name[2]);
1573     }
1574 }
1575
1576 /* Map the opcode 1-to-1 to the GL code */
1577 static void shader_hw_map2gl(const struct wined3d_shader_instruction *ins)
1578 {
1579     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1580     const char *instruction;
1581     char arguments[256], dst_str[50];
1582     unsigned int i;
1583     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1584
1585     switch (ins->handler_idx)
1586     {
1587         case WINED3DSIH_ABS: instruction = "ABS"; break;
1588         case WINED3DSIH_ADD: instruction = "ADD"; break;
1589         case WINED3DSIH_CRS: instruction = "XPD"; break;
1590         case WINED3DSIH_DP3: instruction = "DP3"; break;
1591         case WINED3DSIH_DP4: instruction = "DP4"; break;
1592         case WINED3DSIH_DST: instruction = "DST"; break;
1593         case WINED3DSIH_FRC: instruction = "FRC"; break;
1594         case WINED3DSIH_LIT: instruction = "LIT"; break;
1595         case WINED3DSIH_LRP: instruction = "LRP"; break;
1596         case WINED3DSIH_MAD: instruction = "MAD"; break;
1597         case WINED3DSIH_MAX: instruction = "MAX"; break;
1598         case WINED3DSIH_MIN: instruction = "MIN"; break;
1599         case WINED3DSIH_MOV: instruction = "MOV"; break;
1600         case WINED3DSIH_MUL: instruction = "MUL"; break;
1601         case WINED3DSIH_SGE: instruction = "SGE"; break;
1602         case WINED3DSIH_SLT: instruction = "SLT"; break;
1603         case WINED3DSIH_SUB: instruction = "SUB"; break;
1604         case WINED3DSIH_MOVA:instruction = "ARR"; break;
1605         case WINED3DSIH_DSX: instruction = "DDX"; break;
1606         default: instruction = "";
1607             FIXME("Unhandled opcode %#x\n", ins->handler_idx);
1608             break;
1609     }
1610
1611     /* Note that shader_arb_add_dst_param() adds spaces. */
1612     arguments[0] = '\0';
1613     shader_arb_get_dst_param(ins, dst, dst_str);
1614     for (i = 0; i < ins->src_count; ++i)
1615     {
1616         char operand[100];
1617         strcat(arguments, ", ");
1618         shader_arb_get_src_param(ins, &ins->src[i], i, operand);
1619         strcat(arguments, operand);
1620     }
1621     shader_addline(buffer, "%s%s %s%s;\n", instruction, shader_arb_get_modifier(ins), dst_str, arguments);
1622 }
1623
1624 static void shader_hw_nop(const struct wined3d_shader_instruction *ins)
1625 {
1626     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1627     shader_addline(buffer, "NOP;\n");
1628 }
1629
1630 static void shader_hw_mov(const struct wined3d_shader_instruction *ins)
1631 {
1632     IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
1633     BOOL pshader = shader_is_pshader_version(shader->baseShader.reg_maps.shader_version.type);
1634     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
1635
1636     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1637     char src0_param[256];
1638
1639     if(ins->handler_idx == WINED3DSIH_MOVA) {
1640         char write_mask[6];
1641
1642         if(ctx->target_version >= NV2) {
1643             shader_hw_map2gl(ins);
1644             return;
1645         }
1646         shader_arb_get_src_param(ins, &ins->src[0], 0, src0_param);
1647         shader_arb_get_write_mask(ins, &ins->dst[0], write_mask);
1648
1649         /* This implements the mova formula used in GLSL. The first two instructions
1650          * prepare the sign() part. Note that it is fine to have my_sign(0.0) = 1.0
1651          * in this case:
1652          * mova A0.x, 0.0
1653          *
1654          * A0.x = arl(floor(abs(0.0) + 0.5) * 1.0) = floor(0.5) = 0.0 since arl does a floor
1655          *
1656          * The ARL is performed when A0 is used - the requested component is read from A0_SHADOW into
1657          * A0.x. We can use the overwritten component of A0_shadow as temporary storage for the sign.
1658          */
1659         shader_addline(buffer, "SGE A0_SHADOW%s, %s, mova_const.y;\n", write_mask, src0_param);
1660         shader_addline(buffer, "MAD A0_SHADOW%s, A0_SHADOW, mova_const.z, -mova_const.w;\n", write_mask);
1661
1662         shader_addline(buffer, "ABS TA%s, %s;\n", write_mask, src0_param);
1663         shader_addline(buffer, "ADD TA%s, TA, mova_const.x;\n", write_mask);
1664         shader_addline(buffer, "FLR TA%s, TA;\n", write_mask);
1665         if (((IWineD3DVertexShaderImpl *)shader)->rel_offset)
1666         {
1667             shader_addline(buffer, "ADD TA%s, TA, helper_const.z;\n", write_mask);
1668         }
1669         shader_addline(buffer, "MUL A0_SHADOW%s, TA, A0_SHADOW;\n", write_mask);
1670
1671         ((struct shader_arb_ctx_priv *)ins->ctx->backend_data)->addr_reg[0] = '\0';
1672     } else if (ins->ctx->reg_maps->shader_version.major == 1
1673           && !shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type)
1674           && ins->dst[0].reg.type == WINED3DSPR_ADDR)
1675     {
1676         src0_param[0] = '\0';
1677         if (((IWineD3DVertexShaderImpl *)shader)->rel_offset)
1678         {
1679             shader_arb_get_src_param(ins, &ins->src[0], 0, src0_param);
1680             shader_addline(buffer, "ADD TA.x, %s, helper_const.z;\n", src0_param);
1681             shader_addline(buffer, "ARL A0.x, TA.x;\n");
1682         }
1683         else
1684         {
1685             /* Apple's ARB_vertex_program implementation does not accept an ARL source argument
1686              * with more than one component. Thus replicate the first source argument over all
1687              * 4 components. For example, .xyzw -> .x (or better: .xxxx), .zwxy -> .z, etc) */
1688             struct wined3d_shader_src_param tmp_src = ins->src[0];
1689             tmp_src.swizzle = (tmp_src.swizzle & 0x3) * 0x55;
1690             shader_arb_get_src_param(ins, &tmp_src, 0, src0_param);
1691             shader_addline(buffer, "ARL A0.x, %s;\n", src0_param);
1692         }
1693     }
1694     else if(ins->dst[0].reg.type == WINED3DSPR_COLOROUT && ins->dst[0].reg.idx == 0 && pshader)
1695     {
1696         IWineD3DPixelShaderImpl *ps = (IWineD3DPixelShaderImpl *) shader;
1697         if(ctx->cur_ps_args->super.srgb_correction && ps->color0_mov)
1698         {
1699             shader_addline(buffer, "#mov handled in srgb write code\n");
1700             return;
1701         }
1702         shader_hw_map2gl(ins);
1703     }
1704     else
1705     {
1706         shader_hw_map2gl(ins);
1707     }
1708 }
1709
1710 static void pshader_hw_texkill(const struct wined3d_shader_instruction *ins)
1711 {
1712     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1713     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1714     char reg_dest[40];
1715
1716     /* No swizzles are allowed in d3d's texkill. PS 1.x ignores the 4th component as documented,
1717      * but >= 2.0 honors it(undocumented, but tested by the d3d9 testsuit)
1718      */
1719     shader_arb_get_dst_param(ins, dst, reg_dest);
1720
1721     if (ins->ctx->reg_maps->shader_version.major >= 2)
1722     {
1723         const char *kilsrc = "TA";
1724         BOOL is_color;
1725
1726         shader_arb_get_register_name(ins, &dst->reg, reg_dest, &is_color);
1727         if(dst->write_mask == WINED3DSP_WRITEMASK_ALL)
1728         {
1729             kilsrc = reg_dest;
1730         }
1731         else
1732         {
1733             /* Sigh. KIL doesn't support swizzles/writemasks. KIL passes a writemask, but ".xy" for example
1734              * is not valid as a swizzle in ARB (needs ".xyyy"). Use SWZ to load the register properly, and set
1735              * masked out components to 0(won't kill)
1736              */
1737             char x = '0', y = '0', z = '0', w = '0';
1738             if(dst->write_mask & WINED3DSP_WRITEMASK_0) x = 'x';
1739             if(dst->write_mask & WINED3DSP_WRITEMASK_1) y = 'y';
1740             if(dst->write_mask & WINED3DSP_WRITEMASK_2) z = 'z';
1741             if(dst->write_mask & WINED3DSP_WRITEMASK_3) w = 'w';
1742             shader_addline(buffer, "SWZ TA, %s, %c, %c, %c, %c;\n", reg_dest, x, y, z, w);
1743         }
1744         shader_addline(buffer, "KIL %s;\n", kilsrc);
1745     } else {
1746         /* ARB fp doesn't like swizzles on the parameter of the KIL instruction. To mask the 4th component,
1747          * copy the register into our general purpose TMP variable, overwrite .w and pass TMP to KIL
1748          *
1749          * ps_1_3 shaders use the texcoord incarnation of the Tx register. ps_1_4 shaders can use the same,
1750          * or pass in any temporary register(in shader phase 2)
1751          */
1752         if(ins->ctx->reg_maps->shader_version.minor <= 3) {
1753             sprintf(reg_dest, "fragment.texcoord[%u]", dst->reg.idx);
1754         } else {
1755             shader_arb_get_dst_param(ins, dst, reg_dest);
1756         }
1757         shader_addline(buffer, "SWZ TA, %s, x, y, z, 1;\n", reg_dest);
1758         shader_addline(buffer, "KIL TA;\n");
1759     }
1760 }
1761
1762 static void pshader_hw_tex(const struct wined3d_shader_instruction *ins)
1763 {
1764     IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
1765     IWineD3DDeviceImpl *deviceImpl = (IWineD3DDeviceImpl *)shader->baseShader.device;
1766     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1767     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
1768             ins->ctx->reg_maps->shader_version.minor);
1769     struct wined3d_shader_src_param src;
1770
1771     char reg_dest[40];
1772     char reg_coord[40];
1773     DWORD reg_sampler_code;
1774     DWORD myflags = 0;
1775
1776     /* All versions have a destination register */
1777     shader_arb_get_dst_param(ins, dst, reg_dest);
1778
1779     /* 1.0-1.4: Use destination register number as texture code.
1780        2.0+: Use provided sampler number as texure code. */
1781     if (shader_version < WINED3D_SHADER_VERSION(2,0))
1782         reg_sampler_code = dst->reg.idx;
1783     else
1784         reg_sampler_code = ins->src[1].reg.idx;
1785
1786     /* 1.0-1.3: Use the texcoord varying.
1787        1.4+: Use provided coordinate source register. */
1788     if (shader_version < WINED3D_SHADER_VERSION(1,4))
1789         sprintf(reg_coord, "fragment.texcoord[%u]", reg_sampler_code);
1790     else {
1791         /* TEX is the only instruction that can handle DW and DZ natively */
1792         src = ins->src[0];
1793         if(src.modifiers == WINED3DSPSM_DW) src.modifiers = WINED3DSPSM_NONE;
1794         if(src.modifiers == WINED3DSPSM_DZ) src.modifiers = WINED3DSPSM_NONE;
1795         shader_arb_get_src_param(ins, &src, 0, reg_coord);
1796     }
1797
1798     /* projection flag:
1799      * 1.1, 1.2, 1.3: Use WINED3DTSS_TEXTURETRANSFORMFLAGS
1800      * 1.4: Use WINED3DSPSM_DZ or WINED3DSPSM_DW on src[0]
1801      * 2.0+: Use WINED3DSI_TEXLD_PROJECT on the opcode
1802      */
1803     if (shader_version < WINED3D_SHADER_VERSION(1,4))
1804     {
1805         DWORD flags = 0;
1806         if(reg_sampler_code < MAX_TEXTURES) {
1807             flags = deviceImpl->stateBlock->textureState[reg_sampler_code][WINED3DTSS_TEXTURETRANSFORMFLAGS];
1808         }
1809         if (flags & WINED3DTTFF_PROJECTED) {
1810             myflags |= TEX_PROJ;
1811         }
1812     }
1813     else if (shader_version < WINED3D_SHADER_VERSION(2,0))
1814     {
1815         DWORD src_mod = ins->src[0].modifiers;
1816         if (src_mod == WINED3DSPSM_DZ) {
1817             /* TXP cannot handle DZ natively, so move the z coordinate to .w. reg_coord is a read-only
1818              * varying register, so we need a temp reg
1819              */
1820             shader_addline(ins->ctx->buffer, "SWZ TA, %s, x, y, z, z;\n", reg_coord);
1821             strcpy(reg_coord, "TA");
1822             myflags |= TEX_PROJ;
1823         } else if(src_mod == WINED3DSPSM_DW) {
1824             myflags |= TEX_PROJ;
1825         }
1826     } else {
1827         if (ins->flags & WINED3DSI_TEXLD_PROJECT) myflags |= TEX_PROJ;
1828         if (ins->flags & WINED3DSI_TEXLD_BIAS) myflags |= TEX_BIAS;
1829     }
1830     shader_hw_sample(ins, reg_sampler_code, reg_dest, reg_coord, myflags, NULL, NULL);
1831 }
1832
1833 static void pshader_hw_texcoord(const struct wined3d_shader_instruction *ins)
1834 {
1835     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1836     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1837     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
1838             ins->ctx->reg_maps->shader_version.minor);
1839     char dst_str[50];
1840
1841     if (shader_version < WINED3D_SHADER_VERSION(1,4))
1842     {
1843         DWORD reg = dst->reg.idx;
1844
1845         shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
1846         shader_addline(buffer, "MOV_SAT %s, fragment.texcoord[%u];\n", dst_str, reg);
1847     } else {
1848         char reg_src[40];
1849
1850         shader_arb_get_src_param(ins, &ins->src[0], 0, reg_src);
1851         shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
1852         shader_addline(buffer, "MOV %s, %s;\n", dst_str, reg_src);
1853    }
1854 }
1855
1856 static void pshader_hw_texreg2ar(const struct wined3d_shader_instruction *ins)
1857 {
1858      struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1859      IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
1860      IWineD3DDeviceImpl *deviceImpl = (IWineD3DDeviceImpl *)shader->baseShader.device;
1861      DWORD flags;
1862
1863      DWORD reg1 = ins->dst[0].reg.idx;
1864      char dst_str[50];
1865      char src_str[50];
1866
1867      /* Note that texreg2ar treats Tx as a temporary register, not as a varying */
1868      shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
1869      shader_arb_get_src_param(ins, &ins->src[0], 0, src_str);
1870      /* Move .x first in case src_str is "TA" */
1871      shader_addline(buffer, "MOV TA.y, %s.x;\n", src_str);
1872      shader_addline(buffer, "MOV TA.x, %s.w;\n", src_str);
1873      flags = reg1 < MAX_TEXTURES ? deviceImpl->stateBlock->textureState[reg1][WINED3DTSS_TEXTURETRANSFORMFLAGS] : 0;
1874      shader_hw_sample(ins, reg1, dst_str, "TA", flags & WINED3DTTFF_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
1875 }
1876
1877 static void pshader_hw_texreg2gb(const struct wined3d_shader_instruction *ins)
1878 {
1879      struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1880
1881      DWORD reg1 = ins->dst[0].reg.idx;
1882      char dst_str[50];
1883      char src_str[50];
1884
1885      /* Note that texreg2gb treats Tx as a temporary register, not as a varying */
1886      shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
1887      shader_arb_get_src_param(ins, &ins->src[0], 0, src_str);
1888      shader_addline(buffer, "MOV TA.x, %s.y;\n", src_str);
1889      shader_addline(buffer, "MOV TA.y, %s.z;\n", src_str);
1890      shader_hw_sample(ins, reg1, dst_str, "TA", 0, NULL, NULL);
1891 }
1892
1893 static void pshader_hw_texreg2rgb(const struct wined3d_shader_instruction *ins)
1894 {
1895     DWORD reg1 = ins->dst[0].reg.idx;
1896     char dst_str[50];
1897     char src_str[50];
1898
1899     /* Note that texreg2rg treats Tx as a temporary register, not as a varying */
1900     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
1901     shader_arb_get_src_param(ins, &ins->src[0], 0, src_str);
1902     shader_hw_sample(ins, reg1, dst_str, src_str, 0, NULL, NULL);
1903 }
1904
1905 static void pshader_hw_texbem(const struct wined3d_shader_instruction *ins)
1906 {
1907     IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
1908     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *)shader->baseShader.device;
1909     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1910     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1911     char reg_coord[40], dst_reg[50], src_reg[50];
1912     DWORD reg_dest_code;
1913
1914     /* All versions have a destination register. The Tx where the texture coordinates come
1915      * from is the varying incarnation of the texture register
1916      */
1917     reg_dest_code = dst->reg.idx;
1918     shader_arb_get_dst_param(ins, &ins->dst[0], dst_reg);
1919     shader_arb_get_src_param(ins, &ins->src[0], 0, src_reg);
1920     sprintf(reg_coord, "fragment.texcoord[%u]", reg_dest_code);
1921
1922     /* Sampling the perturbation map in Tsrc was done already, including the signedness correction if needed
1923      * The Tx in which the perturbation map is stored is the tempreg incarnation of the texture register
1924      *
1925      * GL_NV_fragment_program_option could handle this in one instruction via X2D:
1926      * X2D TA.xy, fragment.texcoord, T%u, bumpenvmat%u.xzyw
1927      *
1928      * However, the NV extensions are never enabled for <= 2.0 shaders because of the performance penalty that
1929      * comes with it, and texbem is an 1.x only instruction. No 1.x instruction forces us to enable the NV
1930      * extension.
1931      */
1932     shader_addline(buffer, "SWZ TB, bumpenvmat%d, x, z, 0, 0;\n", reg_dest_code);
1933     shader_addline(buffer, "DP3 TA.x, TB, %s;\n", src_reg);
1934     shader_addline(buffer, "SWZ TB, bumpenvmat%d, y, w, 0, 0;\n", reg_dest_code);
1935     shader_addline(buffer, "DP3 TA.y, TB, %s;\n", src_reg);
1936
1937     /* with projective textures, texbem only divides the static texture coord, not the displacement,
1938      * so we can't let the GL handle this.
1939      */
1940     if (device->stateBlock->textureState[reg_dest_code][WINED3DTSS_TEXTURETRANSFORMFLAGS] & WINED3DTTFF_PROJECTED)
1941     {
1942         shader_addline(buffer, "RCP TB.w, %s.w;\n", reg_coord);
1943         shader_addline(buffer, "MUL TB.xy, %s, TB.w;\n", reg_coord);
1944         shader_addline(buffer, "ADD TA.xy, TA, TB;\n");
1945     } else {
1946         shader_addline(buffer, "ADD TA.xy, TA, %s;\n", reg_coord);
1947     }
1948
1949     shader_hw_sample(ins, reg_dest_code, dst_reg, "TA", 0, NULL, NULL);
1950
1951     if (ins->handler_idx == WINED3DSIH_TEXBEML)
1952     {
1953         /* No src swizzles are allowed, so this is ok */
1954         shader_addline(buffer, "MAD TA, %s.z, luminance%d.x, luminance%d.y;\n",
1955                        src_reg, reg_dest_code, reg_dest_code);
1956         shader_addline(buffer, "MUL %s, %s, TA;\n", dst_reg, dst_reg);
1957     }
1958 }
1959
1960 static void pshader_hw_texm3x2pad(const struct wined3d_shader_instruction *ins)
1961 {
1962     DWORD reg = ins->dst[0].reg.idx;
1963     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1964     char src0_name[50], dst_name[50];
1965     BOOL is_color;
1966     struct wined3d_shader_register tmp_reg = ins->dst[0].reg;
1967
1968     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
1969     /* The next instruction will be a texm3x2tex or texm3x2depth that writes to the uninitialized
1970      * T<reg+1> register. Use this register to store the calculated vector
1971      */
1972     tmp_reg.idx = reg + 1;
1973     shader_arb_get_register_name(ins, &tmp_reg, dst_name, &is_color);
1974     shader_addline(buffer, "DP3 %s.x, fragment.texcoord[%u], %s;\n", dst_name, reg, src0_name);
1975 }
1976
1977 static void pshader_hw_texm3x2tex(const struct wined3d_shader_instruction *ins)
1978 {
1979     IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
1980     IWineD3DDeviceImpl *deviceImpl = (IWineD3DDeviceImpl *)shader->baseShader.device;
1981     DWORD flags;
1982     DWORD reg = ins->dst[0].reg.idx;
1983     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1984     char dst_str[50];
1985     char src0_name[50];
1986     char dst_reg[50];
1987     BOOL is_color;
1988
1989     /* We know that we're writing to the uninitialized T<reg> register, so use it for temporary storage */
1990     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_reg, &is_color);
1991
1992     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
1993     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
1994     shader_addline(buffer, "DP3 %s.y, fragment.texcoord[%u], %s;\n", dst_reg, reg, src0_name);
1995     flags = reg < MAX_TEXTURES ? deviceImpl->stateBlock->textureState[reg][WINED3DTSS_TEXTURETRANSFORMFLAGS] : 0;
1996     shader_hw_sample(ins, reg, dst_str, dst_reg, flags & WINED3DTTFF_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
1997 }
1998
1999 static void pshader_hw_texm3x3pad(const struct wined3d_shader_instruction *ins)
2000 {
2001     IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
2002     SHADER_PARSE_STATE *current_state = &shader->baseShader.parse_state;
2003     DWORD reg = ins->dst[0].reg.idx;
2004     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2005     char src0_name[50], dst_name[50];
2006     struct wined3d_shader_register tmp_reg = ins->dst[0].reg;
2007     BOOL is_color;
2008
2009     /* There are always 2 texm3x3pad instructions followed by one texm3x3[tex,vspec, ...] instruction, with
2010      * incrementing ins->dst[0].register_idx numbers. So the pad instruction already knows the final destination
2011      * register, and this register is uninitialized(otherwise the assembler complains that it is 'redeclared')
2012      */
2013     tmp_reg.idx = reg + 2 - current_state->current_row;
2014     shader_arb_get_register_name(ins, &tmp_reg, dst_name, &is_color);
2015
2016     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2017     shader_addline(buffer, "DP3 %s.%c, fragment.texcoord[%u], %s;\n",
2018                    dst_name, 'x' + current_state->current_row, reg, src0_name);
2019     current_state->texcoord_w[current_state->current_row++] = reg;
2020 }
2021
2022 static void pshader_hw_texm3x3tex(const struct wined3d_shader_instruction *ins)
2023 {
2024     IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
2025     IWineD3DDeviceImpl *deviceImpl = (IWineD3DDeviceImpl *)shader->baseShader.device;
2026     SHADER_PARSE_STATE *current_state = &shader->baseShader.parse_state;
2027     DWORD flags;
2028     DWORD reg = ins->dst[0].reg.idx;
2029     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2030     char dst_str[50];
2031     char src0_name[50], dst_name[50];
2032     BOOL is_color;
2033
2034     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2035     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2036     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_name, reg, src0_name);
2037
2038     /* Sample the texture using the calculated coordinates */
2039     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2040     flags = reg < MAX_TEXTURES ? deviceImpl->stateBlock->textureState[reg][WINED3DTSS_TEXTURETRANSFORMFLAGS] : 0;
2041     shader_hw_sample(ins, reg, dst_str, dst_name, flags & WINED3DTTFF_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2042     current_state->current_row = 0;
2043 }
2044
2045 static void pshader_hw_texm3x3vspec(const struct wined3d_shader_instruction *ins)
2046 {
2047     IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
2048     IWineD3DDeviceImpl *deviceImpl = (IWineD3DDeviceImpl *)shader->baseShader.device;
2049     SHADER_PARSE_STATE *current_state = &shader->baseShader.parse_state;
2050     DWORD flags;
2051     DWORD reg = ins->dst[0].reg.idx;
2052     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2053     char dst_str[50];
2054     char src0_name[50];
2055     char dst_reg[50];
2056     BOOL is_color;
2057
2058     /* Get the dst reg without writemask strings. We know this register is uninitialized, so we can use all
2059      * components for temporary data storage
2060      */
2061     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_reg, &is_color);
2062     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2063     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_reg, reg, src0_name);
2064
2065     /* Construct the eye-ray vector from w coordinates */
2066     shader_addline(buffer, "MOV TB.x, fragment.texcoord[%u].w;\n", current_state->texcoord_w[0]);
2067     shader_addline(buffer, "MOV TB.y, fragment.texcoord[%u].w;\n", current_state->texcoord_w[1]);
2068     shader_addline(buffer, "MOV TB.z, fragment.texcoord[%u].w;\n", reg);
2069
2070     /* Calculate reflection vector
2071      */
2072     shader_addline(buffer, "DP3 %s.w, %s, TB;\n", dst_reg, dst_reg);
2073     /* The .w is ignored when sampling, so I can use TB.w to calculate dot(N, N) */
2074     shader_addline(buffer, "DP3 TB.w, %s, %s;\n", dst_reg, dst_reg);
2075     shader_addline(buffer, "RCP TB.w, TB.w;\n");
2076     shader_addline(buffer, "MUL %s.w, %s.w, TB.w;\n", dst_reg, dst_reg);
2077     shader_addline(buffer, "MUL %s, %s.w, %s;\n", dst_reg, dst_reg, dst_reg);
2078     shader_addline(buffer, "MAD %s, coefmul.x, %s, -TB;\n", dst_reg, dst_reg);
2079
2080     /* Sample the texture using the calculated coordinates */
2081     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2082     flags = reg < MAX_TEXTURES ? deviceImpl->stateBlock->textureState[reg][WINED3DTSS_TEXTURETRANSFORMFLAGS] : 0;
2083     shader_hw_sample(ins, reg, dst_str, dst_reg, flags & WINED3DTTFF_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2084     current_state->current_row = 0;
2085 }
2086
2087 static void pshader_hw_texm3x3spec(const struct wined3d_shader_instruction *ins)
2088 {
2089     IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
2090     IWineD3DDeviceImpl *deviceImpl = (IWineD3DDeviceImpl *)shader->baseShader.device;
2091     SHADER_PARSE_STATE *current_state = &shader->baseShader.parse_state;
2092     DWORD flags;
2093     DWORD reg = ins->dst[0].reg.idx;
2094     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2095     char dst_str[50];
2096     char src0_name[50];
2097     char src1_name[50];
2098     char dst_reg[50];
2099     BOOL is_color;
2100
2101     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2102     shader_arb_get_src_param(ins, &ins->src[0], 1, src1_name);
2103     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_reg, &is_color);
2104     /* Note: dst_reg.xy is input here, generated by two texm3x3pad instructions */
2105     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_reg, reg, src0_name);
2106
2107     /* Calculate reflection vector.
2108      *
2109      *                   dot(N, E)
2110      * dst_reg.xyz = 2 * --------- * N - E
2111      *                   dot(N, N)
2112      *
2113      * Which normalizes the normal vector
2114      */
2115     shader_addline(buffer, "DP3 %s.w, %s, %s;\n", dst_reg, dst_reg, src1_name);
2116     shader_addline(buffer, "DP3 TC.w, %s, %s;\n", dst_reg, dst_reg);
2117     shader_addline(buffer, "RCP TC.w, TC.w;\n");
2118     shader_addline(buffer, "MUL %s.w, %s.w, TC.w;\n", dst_reg, dst_reg);
2119     shader_addline(buffer, "MUL %s, %s.w, %s;\n", dst_reg, dst_reg, dst_reg);
2120     shader_addline(buffer, "MAD %s, coefmul.x, %s, -%s;\n", dst_reg, dst_reg, src1_name);
2121
2122     /* Sample the texture using the calculated coordinates */
2123     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2124     flags = reg < MAX_TEXTURES ? deviceImpl->stateBlock->textureState[reg][WINED3DTSS_TEXTURETRANSFORMFLAGS] : 0;
2125     shader_hw_sample(ins, reg, dst_str, dst_reg, flags & WINED3DTTFF_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2126     current_state->current_row = 0;
2127 }
2128
2129 static void pshader_hw_texdepth(const struct wined3d_shader_instruction *ins)
2130 {
2131     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2132     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2133     char dst_name[50];
2134
2135     /* texdepth has an implicit destination, the fragment depth value. It's only parameter,
2136      * which is essentially an input, is the destination register because it is the first
2137      * parameter. According to the msdn, this must be register r5, but let's keep it more flexible
2138      * here(writemasks/swizzles are not valid on texdepth)
2139      */
2140     shader_arb_get_dst_param(ins, dst, dst_name);
2141
2142     /* According to the msdn, the source register(must be r5) is unusable after
2143      * the texdepth instruction, so we're free to modify it
2144      */
2145     shader_addline(buffer, "MIN %s.y, %s.y, one.y;\n", dst_name, dst_name);
2146
2147     /* How to deal with the special case dst_name.g == 0? if r != 0, then
2148      * the r * (1 / 0) will give infinity, which is clamped to 1.0, the correct
2149      * result. But if r = 0.0, then 0 * inf = 0, which is incorrect.
2150      */
2151     shader_addline(buffer, "RCP %s.y, %s.y;\n", dst_name, dst_name);
2152     shader_addline(buffer, "MUL TA.x, %s.x, %s.y;\n", dst_name, dst_name);
2153     shader_addline(buffer, "MIN TA.x, TA.x, one.x;\n");
2154     shader_addline(buffer, "MAX result.depth, TA.x, 0.0;\n");
2155 }
2156
2157 /** Process the WINED3DSIO_TEXDP3TEX instruction in ARB:
2158  * Take a 3-component dot product of the TexCoord[dstreg] and src,
2159  * then perform a 1D texture lookup from stage dstregnum, place into dst. */
2160 static void pshader_hw_texdp3tex(const struct wined3d_shader_instruction *ins)
2161 {
2162     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2163     DWORD sampler_idx = ins->dst[0].reg.idx;
2164     char src0[50];
2165     char dst_str[50];
2166
2167     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2168     shader_addline(buffer, "MOV TB, 0.0;\n");
2169     shader_addline(buffer, "DP3 TB.x, fragment.texcoord[%u], %s;\n", sampler_idx, src0);
2170
2171     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2172     shader_hw_sample(ins, sampler_idx, dst_str, "TB", 0 /* Only one coord, can't be projected */, NULL, NULL);
2173 }
2174
2175 /** Process the WINED3DSIO_TEXDP3 instruction in ARB:
2176  * Take a 3-component dot product of the TexCoord[dstreg] and src. */
2177 static void pshader_hw_texdp3(const struct wined3d_shader_instruction *ins)
2178 {
2179     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2180     char src0[50];
2181     char dst_str[50];
2182     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2183
2184     /* Handle output register */
2185     shader_arb_get_dst_param(ins, dst, dst_str);
2186     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2187     shader_addline(buffer, "DP3 %s, fragment.texcoord[%u], %s;\n", dst_str, dst->reg.idx, src0);
2188 }
2189
2190 /** Process the WINED3DSIO_TEXM3X3 instruction in ARB
2191  * Perform the 3rd row of a 3x3 matrix multiply */
2192 static void pshader_hw_texm3x3(const struct wined3d_shader_instruction *ins)
2193 {
2194     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2195     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2196     char dst_str[50], dst_name[50];
2197     char src0[50];
2198     BOOL is_color;
2199
2200     shader_arb_get_dst_param(ins, dst, dst_str);
2201     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2202     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2203     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_name, dst->reg.idx, src0);
2204     shader_addline(buffer, "MOV %s, %s;\n", dst_str, dst_name);
2205 }
2206
2207 /** Process the WINED3DSIO_TEXM3X2DEPTH instruction in ARB:
2208  * Last row of a 3x2 matrix multiply, use the result to calculate the depth:
2209  * Calculate tmp0.y = TexCoord[dstreg] . src.xyz;  (tmp0.x has already been calculated)
2210  * depth = (tmp0.y == 0.0) ? 1.0 : tmp0.x / tmp0.y
2211  */
2212 static void pshader_hw_texm3x2depth(const struct wined3d_shader_instruction *ins)
2213 {
2214     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2215     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2216     char src0[50], dst_name[50];
2217     BOOL is_color;
2218
2219     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2220     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2221     shader_addline(buffer, "DP3 %s.y, fragment.texcoord[%u], %s;\n", dst_name, dst->reg.idx, src0);
2222
2223     /* How to deal with the special case dst_name.g == 0? if r != 0, then
2224      * the r * (1 / 0) will give infinity, which is clamped to 1.0, the correct
2225      * result. But if r = 0.0, then 0 * inf = 0, which is incorrect.
2226      */
2227     shader_addline(buffer, "RCP %s.y, %s.y;\n", dst_name, dst_name);
2228     shader_addline(buffer, "MUL %s.x, %s.x, %s.y;\n", dst_name, dst_name, dst_name);
2229     shader_addline(buffer, "MIN %s.x, %s.x, one.x;\n", dst_name, dst_name);
2230     shader_addline(buffer, "MAX result.depth, %s.x, 0.0;\n", dst_name);
2231 }
2232
2233 /** Handles transforming all WINED3DSIO_M?x? opcodes for
2234     Vertex/Pixel shaders to ARB_vertex_program codes */
2235 static void shader_hw_mnxn(const struct wined3d_shader_instruction *ins)
2236 {
2237     int i;
2238     int nComponents = 0;
2239     struct wined3d_shader_dst_param tmp_dst = {{0}};
2240     struct wined3d_shader_src_param tmp_src[2] = {{{0}}};
2241     struct wined3d_shader_instruction tmp_ins;
2242
2243     memset(&tmp_ins, 0, sizeof(tmp_ins));
2244
2245     /* Set constants for the temporary argument */
2246     tmp_ins.ctx = ins->ctx;
2247     tmp_ins.dst_count = 1;
2248     tmp_ins.dst = &tmp_dst;
2249     tmp_ins.src_count = 2;
2250     tmp_ins.src = tmp_src;
2251
2252     switch(ins->handler_idx)
2253     {
2254         case WINED3DSIH_M4x4:
2255             nComponents = 4;
2256             tmp_ins.handler_idx = WINED3DSIH_DP4;
2257             break;
2258         case WINED3DSIH_M4x3:
2259             nComponents = 3;
2260             tmp_ins.handler_idx = WINED3DSIH_DP4;
2261             break;
2262         case WINED3DSIH_M3x4:
2263             nComponents = 4;
2264             tmp_ins.handler_idx = WINED3DSIH_DP3;
2265             break;
2266         case WINED3DSIH_M3x3:
2267             nComponents = 3;
2268             tmp_ins.handler_idx = WINED3DSIH_DP3;
2269             break;
2270         case WINED3DSIH_M3x2:
2271             nComponents = 2;
2272             tmp_ins.handler_idx = WINED3DSIH_DP3;
2273             break;
2274         default:
2275             FIXME("Unhandled opcode %#x\n", ins->handler_idx);
2276             break;
2277     }
2278
2279     tmp_dst = ins->dst[0];
2280     tmp_src[0] = ins->src[0];
2281     tmp_src[1] = ins->src[1];
2282     for (i = 0; i < nComponents; i++) {
2283         tmp_dst.write_mask = WINED3DSP_WRITEMASK_0 << i;
2284         shader_hw_map2gl(&tmp_ins);
2285         ++tmp_src[1].reg.idx;
2286     }
2287 }
2288
2289 static void shader_hw_scalar_op(const struct wined3d_shader_instruction *ins)
2290 {
2291     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2292     const char *instruction;
2293
2294     char dst[50];
2295     char src[50];
2296
2297     switch(ins->handler_idx)
2298     {
2299         case WINED3DSIH_RSQ:  instruction = "RSQ"; break;
2300         case WINED3DSIH_RCP:  instruction = "RCP"; break;
2301         case WINED3DSIH_EXP:  instruction = "EX2"; break;
2302         case WINED3DSIH_EXPP: instruction = "EXP"; break;
2303         default: instruction = "";
2304             FIXME("Unhandled opcode %#x\n", ins->handler_idx);
2305             break;
2306     }
2307
2308     shader_arb_get_dst_param(ins, &ins->dst[0], dst); /* Destination */
2309     shader_arb_get_src_param(ins, &ins->src[0], 0, src);
2310     if (ins->src[0].swizzle == WINED3DSP_NOSWIZZLE)
2311     {
2312         /* Dx sdk says .x is used if no swizzle is given, but our test shows that
2313          * .w is used
2314          */
2315         strcat(src, ".w");
2316     }
2317
2318     shader_addline(buffer, "%s%s %s, %s;\n", instruction, shader_arb_get_modifier(ins), dst, src);
2319 }
2320
2321 static void shader_hw_nrm(const struct wined3d_shader_instruction *ins)
2322 {
2323     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2324     char dst_name[50];
2325     char src_name[50];
2326     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2327     BOOL pshader = shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type);
2328
2329     shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2330     shader_arb_get_src_param(ins, &ins->src[0], 1 /* Use TB */, src_name);
2331
2332     if(pshader && priv->target_version >= NV3)
2333     {
2334         shader_addline(buffer, "NRM%s %s, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name);
2335     }
2336     else
2337     {
2338         shader_addline(buffer, "DP3 TA, %s, %s;\n", src_name, src_name);
2339         shader_addline(buffer, "RSQ TA, TA.x;\n");
2340         /* dst.w = src[0].w * 1 / (src.x^2 + src.y^2 + src.z^2)^(1/2) according to msdn*/
2341         shader_addline(buffer, "MUL%s %s, %s, TA;\n", shader_arb_get_modifier(ins), dst_name,
2342                     src_name);
2343     }
2344 }
2345
2346 static void shader_hw_lrp(const struct wined3d_shader_instruction *ins)
2347 {
2348     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2349     char dst_name[50];
2350     char src_name[3][50];
2351
2352     /* ARB_fragment_program has a convenient LRP instruction */
2353     if(shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type)) {
2354         shader_hw_map2gl(ins);
2355         return;
2356     }
2357
2358     shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2359     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
2360     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
2361     shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
2362
2363     shader_addline(buffer, "SUB TA, %s, %s;\n", src_name[1], src_name[2]);
2364     shader_addline(buffer, "MAD%s %s, %s, TA, %s;\n", shader_arb_get_modifier(ins),
2365                    dst_name, src_name[0], src_name[2]);
2366 }
2367
2368 static void shader_hw_sincos(const struct wined3d_shader_instruction *ins)
2369 {
2370     /* This instruction exists in ARB, but the d3d instruction takes two extra parameters which
2371      * must contain fixed constants. So we need a separate function to filter those constants and
2372      * can't use map2gl
2373      */
2374     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2375     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2376     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2377     char dst_name[50];
2378     char src_name0[50], src_name1[50], src_name2[50];
2379     BOOL is_color;
2380
2381     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name0);
2382     if(shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type)) {
2383         shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2384         /* No modifiers are supported on SCS */
2385         shader_addline(buffer, "SCS %s, %s;\n", dst_name, src_name0);
2386
2387         if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE)
2388         {
2389             shader_arb_get_register_name(ins, &dst->reg, src_name0, &is_color);
2390             shader_addline(buffer, "MOV_SAT %s, %s;\n", dst_name, src_name0);
2391         }
2392     } else if(priv->target_version >= NV2) {
2393         shader_arb_get_register_name(ins, &dst->reg, dst_name, &is_color);
2394
2395         /* Sincos writemask must be .x, .y or .xy */
2396         if(dst->write_mask & WINED3DSP_WRITEMASK_0)
2397             shader_addline(buffer, "COS%s %s.x, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name0);
2398         if(dst->write_mask & WINED3DSP_WRITEMASK_1)
2399             shader_addline(buffer, "SIN%s %s.y, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name0);
2400     } else {
2401         /* Approximate sine and cosine with a taylor series, as per math textbook. The application passes 8
2402          * helper constants(D3DSINCOSCONST1 and D3DSINCOSCONST2) in src1 and src2.
2403          *
2404          * sin(x) = x - x^3/3! + x^5/5! - x^7/7! + ...
2405          * cos(x) = 1 - x^2/2! + x^4/4! - x^6/6! + ...
2406          *
2407          * The constants we get are:
2408          *
2409          *  +1   +1,     -1     -1     +1      +1      -1       -1
2410          *      ---- ,  ---- , ---- , ----- , ----- , ----- , ------
2411          *      1!*2    2!*4   3!*8   4!*16   5!*32   6!*64   7!*128
2412          *
2413          * If used with x^2, x^3, x^4 etc they calculate sin(x/2) and cos(x/2):
2414          *
2415          * (x/2)^2 = x^2 / 4
2416          * (x/2)^3 = x^3 / 8
2417          * (x/2)^4 = x^4 / 16
2418          * (x/2)^5 = x^5 / 32
2419          * etc
2420          *
2421          * To get the final result:
2422          * sin(x) = 2 * sin(x/2) * cos(x/2)
2423          * cos(x) = cos(x/2)^2 - sin(x/2)^2
2424          * (from sin(x+y) and cos(x+y) rules)
2425          *
2426          * As per MSDN, dst.z is undefined after the operation, and so is
2427          * dst.x and dst.y if they're masked out by the writemask. Ie
2428          * sincos dst.y, src1, c0, c1
2429          * returns the sine in dst.y. dst.x and dst.z are undefined, dst.w is not touched. The assembler
2430          * vsa.exe also stops with an error if the dest register is the same register as the source
2431          * register. This means we can use dest.xyz as temporary storage. The assembler vsa.exe output also
2432          * indicates that sincos consumes 8 instruction slots in vs_2_0(and, strangely, in vs_3_0).
2433          */
2434         shader_arb_get_src_param(ins, &ins->src[1], 1, src_name1);
2435         shader_arb_get_src_param(ins, &ins->src[2], 2, src_name2);
2436         shader_arb_get_register_name(ins, &dst->reg, dst_name, &is_color);
2437
2438         shader_addline(buffer, "MUL %s.x, %s, %s;\n", dst_name, src_name0, src_name0);  /* x ^ 2 */
2439         shader_addline(buffer, "MUL TA.y, %s.x, %s;\n", dst_name, src_name0);           /* x ^ 3 */
2440         shader_addline(buffer, "MUL %s.y, TA.y, %s;\n", dst_name, src_name0);           /* x ^ 4 */
2441         shader_addline(buffer, "MUL TA.z, %s.y, %s;\n", dst_name, src_name0);           /* x ^ 5 */
2442         shader_addline(buffer, "MUL %s.z, TA.z, %s;\n", dst_name, src_name0);           /* x ^ 6 */
2443         shader_addline(buffer, "MUL TA.w, %s.z, %s;\n", dst_name, src_name0);           /* x ^ 7 */
2444
2445         /* sin(x/2)
2446          *
2447          * Unfortunately we don't get the constants in a DP4-capable form. Is there a way to
2448          * properly merge that with MULs in the code above?
2449          * The swizzles .yz and xw however fit into the .yzxw swizzle added to ps_2_0. Maybe
2450          * we can merge the sine and cosine MAD rows to calculate them together.
2451          */
2452         shader_addline(buffer, "MUL TA.x, %s, %s.w;\n", src_name0, src_name2); /* x^1, +1/(1!*2) */
2453         shader_addline(buffer, "MAD TA.x, TA.y, %s.x, TA.x;\n", src_name2); /* -1/(3!*8) */
2454         shader_addline(buffer, "MAD TA.x, TA.z, %s.w, TA.x;\n", src_name1); /* +1/(5!*32) */
2455         shader_addline(buffer, "MAD TA.x, TA.w, %s.x, TA.x;\n", src_name1); /* -1/(7!*128) */
2456
2457         /* cos(x/2) */
2458         shader_addline(buffer, "MAD TA.y, %s.x, %s.y, %s.z;\n", dst_name, src_name2, src_name2); /* -1/(2!*4), +1.0 */
2459         shader_addline(buffer, "MAD TA.y, %s.y, %s.z, TA.y;\n", dst_name, src_name1); /* +1/(4!*16) */
2460         shader_addline(buffer, "MAD TA.y, %s.z, %s.y, TA.y;\n", dst_name, src_name1); /* -1/(6!*64) */
2461
2462         if(dst->write_mask & WINED3DSP_WRITEMASK_0) {
2463             /* cos x */
2464             shader_addline(buffer, "MUL TA.z, TA.y, TA.y;\n");
2465             shader_addline(buffer, "MAD %s.x, -TA.x, TA.x, TA.z;\n", dst_name);
2466         }
2467         if(dst->write_mask & WINED3DSP_WRITEMASK_1) {
2468             /* sin x */
2469             shader_addline(buffer, "MUL %s.y, TA.x, TA.y;\n", dst_name);
2470             shader_addline(buffer, "ADD %s.y, %s.y, %s.y;\n", dst_name, dst_name, dst_name);
2471         }
2472     }
2473 }
2474
2475 static void shader_hw_sgn(const struct wined3d_shader_instruction *ins)
2476 {
2477     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2478     char dst_name[50];
2479     char src_name[50];
2480     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
2481
2482     shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2483     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name);
2484
2485     /* SGN is only valid in vertex shaders */
2486     if(ctx->target_version >= NV2) {
2487         shader_addline(buffer, "SSG%s %s, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name);
2488         return;
2489     }
2490
2491     /* If SRC > 0.0, -SRC < SRC = TRUE, otherwise false.
2492      * if SRC < 0.0,  SRC < -SRC = TRUE. If neither is true, src = 0.0
2493      */
2494     if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE) {
2495         shader_addline(buffer, "SLT %s, -%s, %s;\n", dst_name, src_name, src_name);
2496     } else {
2497         /* src contains TA? Write to the dest first. This won't overwrite our destination.
2498          * Then use TA, and calculate the final result
2499          *
2500          * Not reading from TA? Store the first result in TA to avoid overwriting the
2501          * destination if src reg = dst reg
2502          */
2503         if(strstr(src_name, "TA"))
2504         {
2505             shader_addline(buffer, "SLT %s,  %s, -%s;\n", dst_name, src_name, src_name);
2506             shader_addline(buffer, "SLT TA, -%s, %s;\n", src_name, src_name);
2507             shader_addline(buffer, "ADD %s, %s, -TA;\n", dst_name, dst_name);
2508         }
2509         else
2510         {
2511             shader_addline(buffer, "SLT TA, -%s, %s;\n", src_name, src_name);
2512             shader_addline(buffer, "SLT %s,  %s, -%s;\n", dst_name, src_name, src_name);
2513             shader_addline(buffer, "ADD %s, TA, -%s;\n", dst_name, dst_name);
2514         }
2515     }
2516 }
2517
2518 static void shader_hw_dsy(const struct wined3d_shader_instruction *ins)
2519 {
2520     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2521     char src[50];
2522     char dst[50];
2523     char dst_name[50];
2524     BOOL is_color;
2525
2526     shader_arb_get_dst_param(ins, &ins->dst[0], dst);
2527     shader_arb_get_src_param(ins, &ins->src[0], 0, src);
2528     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2529
2530     shader_addline(buffer, "DDY %s, %s;\n", dst, src);
2531     shader_addline(buffer, "MUL%s %s, %s, ycorrection.y;\n", shader_arb_get_modifier(ins), dst, dst_name);
2532 }
2533
2534 static DWORD abs_modifier(DWORD mod, BOOL *need_abs)
2535 {
2536     *need_abs = FALSE;
2537
2538     switch(mod)
2539     {
2540         case WINED3DSPSM_NONE:      return WINED3DSPSM_ABS;
2541         case WINED3DSPSM_NEG:       return WINED3DSPSM_ABS;
2542         case WINED3DSPSM_BIAS:      *need_abs = TRUE; return WINED3DSPSM_BIAS;
2543         case WINED3DSPSM_BIASNEG:   *need_abs = TRUE; return WINED3DSPSM_BIASNEG;
2544         case WINED3DSPSM_SIGN:      *need_abs = TRUE; return WINED3DSPSM_SIGN;
2545         case WINED3DSPSM_SIGNNEG:   *need_abs = TRUE; return WINED3DSPSM_SIGNNEG;
2546         case WINED3DSPSM_COMP:      *need_abs = TRUE; return WINED3DSPSM_COMP;
2547         case WINED3DSPSM_X2:        *need_abs = TRUE; return WINED3DSPSM_X2;
2548         case WINED3DSPSM_X2NEG:     *need_abs = TRUE; return WINED3DSPSM_X2NEG;
2549         case WINED3DSPSM_DZ:        *need_abs = TRUE; return WINED3DSPSM_DZ;
2550         case WINED3DSPSM_DW:        *need_abs = TRUE; return WINED3DSPSM_DW;
2551         case WINED3DSPSM_ABS:       return WINED3DSPSM_ABS;
2552         case WINED3DSPSM_ABSNEG:    return WINED3DSPSM_ABS;
2553     }
2554     FIXME("Unknown modifier %u\n", mod);
2555     return mod;
2556 }
2557
2558 static void shader_hw_log_pow(const struct wined3d_shader_instruction *ins)
2559 {
2560     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2561     char src0[50], src1[50], dst[50];
2562     struct wined3d_shader_src_param src0_copy = ins->src[0];
2563     BOOL need_abs = FALSE;
2564     const char *instr;
2565     BOOL arg2 = FALSE;
2566
2567     switch(ins->handler_idx)
2568     {
2569         case WINED3DSIH_LOG:  instr = "LG2"; break;
2570         case WINED3DSIH_LOGP: instr = "LOG"; break;
2571         case WINED3DSIH_POW:  instr = "POW"; arg2 = TRUE; break;
2572         default:
2573             ERR("Unexpected instruction %d\n", ins->handler_idx);
2574             return;
2575     }
2576
2577     /* LOG, LOGP and POW operate on the absolute value of the input */
2578     src0_copy.modifiers = abs_modifier(src0_copy.modifiers, &need_abs);
2579
2580     shader_arb_get_dst_param(ins, &ins->dst[0], dst);
2581     shader_arb_get_src_param(ins, &src0_copy, 0, src0);
2582     if(arg2) shader_arb_get_src_param(ins, &ins->src[1], 1, src1);
2583
2584     if(need_abs)
2585     {
2586         shader_addline(buffer, "ABS TA, %s;\n", src0);
2587         if(arg2)
2588         {
2589             shader_addline(buffer, "%s%s %s, TA, %s;\n", instr, shader_arb_get_modifier(ins), dst, src1);
2590         }
2591         else
2592         {
2593             shader_addline(buffer, "%s%s %s, TA;\n", instr, shader_arb_get_modifier(ins), dst);
2594         }
2595     }
2596     else if(arg2)
2597     {
2598         shader_addline(buffer, "%s%s %s, %s, %s;\n", instr, shader_arb_get_modifier(ins), dst, src0, src1);
2599     }
2600     else
2601     {
2602         shader_addline(buffer, "%s%s %s, %s;\n", instr, shader_arb_get_modifier(ins), dst, src0);
2603     }
2604 }
2605
2606 static void shader_hw_loop(const struct wined3d_shader_instruction *ins)
2607 {
2608     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2609     char src_name[50];
2610     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2611
2612     /* src0 is aL */
2613     shader_arb_get_src_param(ins, &ins->src[1], 0, src_name);
2614
2615     if(vshader)
2616     {
2617         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2618         struct list *e = list_head(&priv->control_frames);
2619         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2620
2621         if(priv->loop_depth > 1) shader_addline(buffer, "PUSHA aL;\n");
2622         /* The constant loader makes sure to load -1 into iX.w */
2623         shader_addline(buffer, "ARLC aL, %s.xywz;\n", src_name);
2624         shader_addline(buffer, "BRA loop_%u_end (LE.x);\n", control_frame->no.loop);
2625         shader_addline(buffer, "loop_%u_start:\n", control_frame->no.loop);
2626     }
2627     else
2628     {
2629         shader_addline(buffer, "LOOP %s;\n", src_name);
2630     }
2631 }
2632
2633 static void shader_hw_rep(const struct wined3d_shader_instruction *ins)
2634 {
2635     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2636     char src_name[50];
2637     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2638
2639     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name);
2640
2641     /* The constant loader makes sure to load -1 into iX.w */
2642     if(vshader)
2643     {
2644         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2645         struct list *e = list_head(&priv->control_frames);
2646         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2647
2648         if(priv->loop_depth > 1) shader_addline(buffer, "PUSHA aL;\n");
2649
2650         shader_addline(buffer, "ARLC aL, %s.xywz;\n", src_name);
2651         shader_addline(buffer, "BRA loop_%u_end (LE.x);\n", control_frame->no.loop);
2652         shader_addline(buffer, "loop_%u_start:\n", control_frame->no.loop);
2653     }
2654     else
2655     {
2656         shader_addline(buffer, "REP %s;\n", src_name);
2657     }
2658 }
2659
2660 static void shader_hw_endloop(const struct wined3d_shader_instruction *ins)
2661 {
2662     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2663     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2664
2665     if(vshader)
2666     {
2667         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2668         struct list *e = list_head(&priv->control_frames);
2669         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2670
2671         shader_addline(buffer, "ARAC aL.xy, aL;\n");
2672         shader_addline(buffer, "BRA loop_%u_start (GT.x);\n", control_frame->no.loop);
2673         shader_addline(buffer, "loop_%u_end:\n", control_frame->no.loop);
2674
2675         if(priv->loop_depth > 1) shader_addline(buffer, "POPA aL;\n");
2676     }
2677     else
2678     {
2679         shader_addline(buffer, "ENDLOOP;\n");
2680     }
2681 }
2682
2683 static void shader_hw_endrep(const struct wined3d_shader_instruction *ins)
2684 {
2685     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2686     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2687
2688     if(vshader)
2689     {
2690         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2691         struct list *e = list_head(&priv->control_frames);
2692         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2693
2694         shader_addline(buffer, "ARAC aL.xy, aL;\n");
2695         shader_addline(buffer, "BRA loop_%u_start (GT.x);\n", control_frame->no.loop);
2696         shader_addline(buffer, "loop_%u_end:\n", control_frame->no.loop);
2697
2698         if(priv->loop_depth > 1) shader_addline(buffer, "POPA aL;\n");
2699     }
2700     else
2701     {
2702         shader_addline(buffer, "ENDREP;\n");
2703     }
2704 }
2705
2706 static const struct control_frame *find_last_loop(const struct shader_arb_ctx_priv *priv)
2707 {
2708     struct control_frame *control_frame;
2709
2710     LIST_FOR_EACH_ENTRY(control_frame, &priv->control_frames, struct control_frame, entry)
2711     {
2712         if(control_frame->type == LOOP || control_frame->type == REP) return control_frame;
2713     }
2714     ERR("Could not find loop for break\n");
2715     return NULL;
2716 }
2717
2718 static void shader_hw_break(const struct wined3d_shader_instruction *ins)
2719 {
2720     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2721     const struct control_frame *control_frame = find_last_loop(ins->ctx->backend_data);
2722     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2723
2724     if(vshader)
2725     {
2726         shader_addline(buffer, "BRA loop_%u_end;\n", control_frame->no.loop);
2727     }
2728     else
2729     {
2730         shader_addline(buffer, "BRK;\n");
2731     }
2732 }
2733
2734 static const char *get_compare(COMPARISON_TYPE flags)
2735 {
2736     switch (flags)
2737     {
2738         case COMPARISON_GT: return "GT";
2739         case COMPARISON_EQ: return "EQ";
2740         case COMPARISON_GE: return "GE";
2741         case COMPARISON_LT: return "LT";
2742         case COMPARISON_NE: return "NE";
2743         case COMPARISON_LE: return "LE";
2744         default:
2745             FIXME("Unrecognized comparison value: %u\n", flags);
2746             return "(\?\?)";
2747     }
2748 }
2749
2750 static COMPARISON_TYPE invert_compare(COMPARISON_TYPE flags)
2751 {
2752     switch (flags)
2753     {
2754         case COMPARISON_GT: return COMPARISON_LE;
2755         case COMPARISON_EQ: return COMPARISON_NE;
2756         case COMPARISON_GE: return COMPARISON_LT;
2757         case COMPARISON_LT: return COMPARISON_GE;
2758         case COMPARISON_NE: return COMPARISON_EQ;
2759         case COMPARISON_LE: return COMPARISON_GT;
2760         default:
2761             FIXME("Unrecognized comparison value: %u\n", flags);
2762             return -1;
2763     }
2764 }
2765
2766 static void shader_hw_breakc(const struct wined3d_shader_instruction *ins)
2767 {
2768     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2769     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2770     const struct control_frame *control_frame = find_last_loop(ins->ctx->backend_data);
2771     char src_name0[50];
2772     char src_name1[50];
2773     const char *comp = get_compare(ins->flags);
2774
2775     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name0);
2776     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name1);
2777
2778     if(vshader)
2779     {
2780         /* SUBC CC, src0, src1" works only in pixel shaders, so use TA to throw
2781          * away the subtraction result
2782          */
2783         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
2784         shader_addline(buffer, "BRA loop_%u_end (%s.x);\n", control_frame->no.loop, comp);
2785     }
2786     else
2787     {
2788         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
2789         shader_addline(buffer, "BRK (%s.x);\n", comp);
2790     }
2791 }
2792
2793 static void shader_hw_ifc(const struct wined3d_shader_instruction *ins)
2794 {
2795     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2796     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2797     struct list *e = list_head(&priv->control_frames);
2798     struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2799     const char *comp;
2800     char src_name0[50];
2801     char src_name1[50];
2802     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2803
2804     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name0);
2805     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name1);
2806
2807     if(vshader)
2808     {
2809         /* Invert the flag. We jump to the else label if the condition is NOT true */
2810         comp = get_compare(invert_compare(ins->flags));
2811         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
2812         shader_addline(buffer, "BRA ifc_%u_else (%s.x);\n", control_frame->no.ifc, comp);
2813     }
2814     else
2815     {
2816         comp = get_compare(ins->flags);
2817         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
2818         shader_addline(buffer, "IF %s.x;\n", comp);
2819     }
2820 }
2821
2822 static void shader_hw_else(const struct wined3d_shader_instruction *ins)
2823 {
2824     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2825     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2826     struct list *e = list_head(&priv->control_frames);
2827     struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2828     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2829
2830     if(vshader)
2831     {
2832         shader_addline(buffer, "BRA ifc_%u_endif;\n", control_frame->no.ifc);
2833         shader_addline(buffer, "ifc_%u_else:\n", control_frame->no.ifc);
2834         control_frame->had_else = TRUE;
2835     }
2836     else
2837     {
2838         shader_addline(buffer, "ELSE;\n");
2839     }
2840 }
2841
2842 static void shader_hw_endif(const struct wined3d_shader_instruction *ins)
2843 {
2844     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2845     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2846     struct list *e = list_head(&priv->control_frames);
2847     struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2848     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2849
2850     if(vshader)
2851     {
2852         if(control_frame->had_else)
2853         {
2854             shader_addline(buffer, "ifc_%u_endif:\n", control_frame->no.ifc);
2855         }
2856         else
2857         {
2858             shader_addline(buffer, "#No else branch. else is endif\n");
2859             shader_addline(buffer, "ifc_%u_else:\n", control_frame->no.ifc);
2860         }
2861     }
2862     else
2863     {
2864         shader_addline(buffer, "ENDIF;\n");
2865     }
2866 }
2867
2868 static void shader_hw_texldd(const struct wined3d_shader_instruction *ins)
2869 {
2870     DWORD sampler_idx = ins->src[1].reg.idx;
2871     char reg_dest[40];
2872     char reg_src[3][40];
2873     DWORD flags = TEX_DERIV;
2874
2875     shader_arb_get_dst_param(ins, &ins->dst[0], reg_dest);
2876     shader_arb_get_src_param(ins, &ins->src[0], 0, reg_src[0]);
2877     shader_arb_get_src_param(ins, &ins->src[2], 1, reg_src[1]);
2878     shader_arb_get_src_param(ins, &ins->src[3], 2, reg_src[2]);
2879
2880     if (ins->flags & WINED3DSI_TEXLD_PROJECT) flags |= TEX_PROJ;
2881     if (ins->flags & WINED3DSI_TEXLD_BIAS) flags |= TEX_BIAS;
2882
2883     shader_hw_sample(ins, sampler_idx, reg_dest, reg_src[0], flags, reg_src[1], reg_src[2]);
2884 }
2885
2886 static void shader_hw_texldl(const struct wined3d_shader_instruction *ins)
2887 {
2888     DWORD sampler_idx = ins->src[1].reg.idx;
2889     char reg_dest[40];
2890     char reg_coord[40];
2891     DWORD flags = TEX_LOD;
2892
2893     shader_arb_get_dst_param(ins, &ins->dst[0], reg_dest);
2894     shader_arb_get_src_param(ins, &ins->src[0], 0, reg_coord);
2895
2896     if (ins->flags & WINED3DSI_TEXLD_PROJECT) flags |= TEX_PROJ;
2897     if (ins->flags & WINED3DSI_TEXLD_BIAS) flags |= TEX_BIAS;
2898
2899     shader_hw_sample(ins, sampler_idx, reg_dest, reg_coord, flags, NULL, NULL);
2900 }
2901
2902 static void shader_hw_label(const struct wined3d_shader_instruction *ins)
2903 {
2904     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2905     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2906
2907     priv->in_main_func = FALSE;
2908     /* Call instructions activate the NV extensions, not labels and rets. If there is an uncalled
2909      * subroutine, don't generate a label that will make GL complain
2910      */
2911     if(priv->target_version == ARB) return;
2912
2913     shader_addline(buffer, "l%u:\n", ins->src[0].reg.idx);
2914 }
2915
2916 static void vshader_add_footer(IWineD3DVertexShaderImpl *This, struct wined3d_shader_buffer *buffer,
2917         const struct arb_vs_compile_args *args, struct shader_arb_ctx_priv *priv_ctx)
2918 {
2919     const shader_reg_maps *reg_maps = &This->baseShader.reg_maps;
2920     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *)This->baseShader.device;
2921     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
2922     unsigned int i;
2923
2924     /* The D3DRS_FOGTABLEMODE render state defines if the shader-generated fog coord is used
2925      * or if the fragment depth is used. If the fragment depth is used(FOGTABLEMODE != NONE),
2926      * the fog frag coord is thrown away. If the fog frag coord is used, but not written by
2927      * the shader, it is set to 0.0(fully fogged, since start = 1.0, end = 0.0)
2928      */
2929     if(args->super.fog_src == VS_FOG_Z) {
2930         shader_addline(buffer, "MOV result.fogcoord, TMP_OUT.z;\n");
2931     } else if (!reg_maps->fog) {
2932         /* posFixup.x is always 1.0, so we can savely use it */
2933         shader_addline(buffer, "ADD result.fogcoord, posFixup.x, -posFixup.x;\n");
2934     }
2935
2936     /* Write the final position.
2937      *
2938      * OpenGL coordinates specify the center of the pixel while d3d coords specify
2939      * the corner. The offsets are stored in z and w in posFixup. posFixup.y contains
2940      * 1.0 or -1.0 to turn the rendering upside down for offscreen rendering. PosFixup.x
2941      * contains 1.0 to allow a mad, but arb vs swizzles are too restricted for that.
2942      */
2943     shader_addline(buffer, "MUL TA, posFixup, TMP_OUT.w;\n");
2944     shader_addline(buffer, "ADD TMP_OUT.x, TMP_OUT.x, TA.z;\n");
2945     shader_addline(buffer, "MAD TMP_OUT.y, TMP_OUT.y, posFixup.y, TA.w;\n");
2946
2947     if(use_nv_clip(gl_info) && priv_ctx->target_version >= NV2)
2948     {
2949         if(args->super.clip_enabled)
2950         {
2951             for(i = 0; i < priv_ctx->vs_clipplanes; i++)
2952             {
2953                 shader_addline(buffer, "DP4 result.clip[%u].x, TMP_OUT, state.clip[%u].plane;\n", i, i);
2954             }
2955         }
2956     }
2957     else if(args->clip.boolclip.clip_texcoord)
2958     {
2959         unsigned int cur_clip = 0;
2960         char component[4] = {'x', 'y', 'z', 'w'};
2961
2962         for (i = 0; i < gl_info->limits.clipplanes; ++i)
2963         {
2964             if(args->clip.boolclip.clipplane_mask & (1 << i))
2965             {
2966                 shader_addline(buffer, "DP4 TA.%c, TMP_OUT, state.clip[%u].plane;\n",
2967                                component[cur_clip++], i);
2968             }
2969         }
2970         switch(cur_clip)
2971         {
2972             case 0:
2973                 shader_addline(buffer, "MOV TA, -helper_const.w;\n");
2974                 break;
2975             case 1:
2976                 shader_addline(buffer, "MOV TA.yzw, -helper_const.w;\n");
2977                 break;
2978             case 2:
2979                 shader_addline(buffer, "MOV TA.zw, -helper_const.w;\n");
2980                 break;
2981             case 3:
2982                 shader_addline(buffer, "MOV TA.w, -helper_const.w;\n");
2983                 break;
2984         }
2985         shader_addline(buffer, "MOV result.texcoord[%u], TA;\n",
2986                        args->clip.boolclip.clip_texcoord - 1);
2987     }
2988
2989     /* Z coord [0;1]->[-1;1] mapping, see comment in transform_projection in state.c
2990      * and the glsl equivalent
2991      */
2992     if(need_helper_const(gl_info)) {
2993         shader_addline(buffer, "MAD TMP_OUT.z, TMP_OUT.z, helper_const.x, -TMP_OUT.w;\n");
2994     } else {
2995         shader_addline(buffer, "ADD TMP_OUT.z, TMP_OUT.z, TMP_OUT.z;\n");
2996         shader_addline(buffer, "ADD TMP_OUT.z, TMP_OUT.z, -TMP_OUT.w;\n");
2997     }
2998
2999     shader_addline(buffer, "MOV result.position, TMP_OUT;\n");
3000
3001     priv_ctx->footer_written = TRUE;
3002 }
3003
3004 static void shader_hw_ret(const struct wined3d_shader_instruction *ins)
3005 {
3006     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3007     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3008     IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *) ins->ctx->shader;
3009     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
3010
3011     if(priv->target_version == ARB) return;
3012
3013     if(vshader)
3014     {
3015         if(priv->in_main_func) vshader_add_footer((IWineD3DVertexShaderImpl *) shader, buffer, priv->cur_vs_args, priv);
3016     }
3017
3018     shader_addline(buffer, "RET;\n");
3019 }
3020
3021 static void shader_hw_call(const struct wined3d_shader_instruction *ins)
3022 {
3023     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3024     shader_addline(buffer, "CAL l%u;\n", ins->src[0].reg.idx);
3025 }
3026
3027 /* GL locking is done by the caller */
3028 static GLuint create_arb_blt_vertex_program(const struct wined3d_gl_info *gl_info)
3029 {
3030     GLuint program_id = 0;
3031     GLint pos;
3032
3033     const char *blt_vprogram =
3034         "!!ARBvp1.0\n"
3035         "PARAM c[1] = { { 1, 0.5 } };\n"
3036         "MOV result.position, vertex.position;\n"
3037         "MOV result.color, c[0].x;\n"
3038         "MOV result.texcoord[0], vertex.texcoord[0];\n"
3039         "END\n";
3040
3041     GL_EXTCALL(glGenProgramsARB(1, &program_id));
3042     GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, program_id));
3043     GL_EXTCALL(glProgramStringARB(GL_VERTEX_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
3044             strlen(blt_vprogram), blt_vprogram));
3045     checkGLcall("glProgramStringARB()");
3046
3047     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
3048     if (pos != -1)
3049     {
3050         FIXME("Vertex program error at position %d: %s\n", pos,
3051             debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
3052     }
3053     else
3054     {
3055         GLint native;
3056
3057         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
3058         checkGLcall("glGetProgramivARB()");
3059         if (!native) WARN("Program exceeds native resource limits.\n");
3060     }
3061
3062     return program_id;
3063 }
3064
3065 /* GL locking is done by the caller */
3066 static GLuint create_arb_blt_fragment_program(const struct wined3d_gl_info *gl_info, enum tex_types tex_type)
3067 {
3068     GLuint program_id = 0;
3069     GLint pos;
3070
3071     static const char * const blt_fprograms[tex_type_count] =
3072     {
3073         /* tex_1d */
3074         NULL,
3075         /* tex_2d */
3076         "!!ARBfp1.0\n"
3077         "TEMP R0;\n"
3078         "TEX R0.x, fragment.texcoord[0], texture[0], 2D;\n"
3079         "MOV result.depth.z, R0.x;\n"
3080         "END\n",
3081         /* tex_3d */
3082         NULL,
3083         /* tex_cube */
3084         "!!ARBfp1.0\n"
3085         "TEMP R0;\n"
3086         "TEX R0.x, fragment.texcoord[0], texture[0], CUBE;\n"
3087         "MOV result.depth.z, R0.x;\n"
3088         "END\n",
3089         /* tex_rect */
3090         "!!ARBfp1.0\n"
3091         "TEMP R0;\n"
3092         "TEX R0.x, fragment.texcoord[0], texture[0], RECT;\n"
3093         "MOV result.depth.z, R0.x;\n"
3094         "END\n",
3095     };
3096
3097     if (!blt_fprograms[tex_type])
3098     {
3099         FIXME("tex_type %#x not supported\n", tex_type);
3100         tex_type = tex_2d;
3101     }
3102
3103     GL_EXTCALL(glGenProgramsARB(1, &program_id));
3104     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, program_id));
3105     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
3106             strlen(blt_fprograms[tex_type]), blt_fprograms[tex_type]));
3107     checkGLcall("glProgramStringARB()");
3108
3109     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
3110     if (pos != -1)
3111     {
3112         FIXME("Fragment program error at position %d: %s\n", pos,
3113             debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
3114     }
3115     else
3116     {
3117         GLint native;
3118
3119         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
3120         checkGLcall("glGetProgramivARB()");
3121         if (!native) WARN("Program exceeds native resource limits.\n");
3122     }
3123
3124     return program_id;
3125 }
3126
3127 static void arbfp_add_sRGB_correction(struct wined3d_shader_buffer *buffer, const char *fragcolor,
3128         const char *tmp1, const char *tmp2, const char *tmp3, const char *tmp4, BOOL condcode)
3129 {
3130     /* Perform sRGB write correction. See GLX_EXT_framebuffer_sRGB */
3131
3132     if(condcode)
3133     {
3134         /* Sigh. MOVC CC doesn't work, so use one of the temps as dummy dest */
3135         shader_addline(buffer, "SUBC %s, %s.x, srgb_consts1.y;\n", tmp1, fragcolor);
3136         /* Calculate the > 0.0031308 case */
3137         shader_addline(buffer, "POW %s.x (GE), %s.x, srgb_consts1.z;\n", fragcolor, fragcolor);
3138         shader_addline(buffer, "POW %s.y (GE), %s.y, srgb_consts1.z;\n", fragcolor, fragcolor);
3139         shader_addline(buffer, "POW %s.z (GE), %s.z, srgb_consts1.z;\n", fragcolor, fragcolor);
3140         shader_addline(buffer, "MUL %s.xyz (GE), %s, srgb_consts1.w;\n", fragcolor, fragcolor);
3141         shader_addline(buffer, "SUB %s.xyz (GE), %s, srgb_consts2.x;\n", fragcolor, fragcolor);
3142         /* Calculate the < case */
3143         shader_addline(buffer, "MUL %s.xyz (LT), srgb_consts1.x, %s;\n", fragcolor, fragcolor);
3144     }
3145     else
3146     {
3147         /* Calculate the > 0.0031308 case */
3148         shader_addline(buffer, "POW %s.x, %s.x, srgb_consts1.z;\n", tmp1, fragcolor);
3149         shader_addline(buffer, "POW %s.y, %s.y, srgb_consts1.z;\n", tmp1, fragcolor);
3150         shader_addline(buffer, "POW %s.z, %s.z, srgb_consts1.z;\n", tmp1, fragcolor);
3151         shader_addline(buffer, "MUL %s, %s, srgb_consts1.w;\n", tmp1, tmp1);
3152         shader_addline(buffer, "SUB %s, %s, srgb_consts2.x;\n", tmp1, tmp1);
3153         /* Calculate the < case */
3154         shader_addline(buffer, "MUL %s, srgb_consts1.x, %s;\n", tmp2, fragcolor);
3155         /* Get 1.0 / 0.0 masks for > 0.0031308 and < 0.0031308 */
3156         shader_addline(buffer, "SLT %s, srgb_consts1.y, %s;\n", tmp3, fragcolor);
3157         shader_addline(buffer, "SGE %s, srgb_consts1.y, %s;\n", tmp4, fragcolor);
3158         /* Store the components > 0.0031308 in the destination */
3159         shader_addline(buffer, "MUL %s.xyz, %s, %s;\n", fragcolor, tmp1, tmp3);
3160         /* Add the components that are < 0.0031308 */
3161         shader_addline(buffer, "MAD %s.xyz, %s, %s, %s;\n", fragcolor, tmp2, tmp4, fragcolor);
3162         /* Move everything into result.color at once. Nvidia hardware cannot handle partial
3163         * result.color writes(.rgb first, then .a), or handle overwriting already written
3164         * components. The assembler uses a temporary register in this case, which is usually
3165         * not allocated from one of our registers that were used earlier.
3166         */
3167     }
3168     /* [0.0;1.0] clamping. Not needed, this is done implicitly */
3169 }
3170
3171 static const DWORD *find_loop_control_values(IWineD3DBaseShaderImpl *This, DWORD idx)
3172 {
3173     const local_constant *constant;
3174
3175     LIST_FOR_EACH_ENTRY(constant, &This->baseShader.constantsI, local_constant, entry)
3176     {
3177         if (constant->idx == idx)
3178         {
3179             return constant->value;
3180         }
3181     }
3182     return NULL;
3183 }
3184
3185 static void init_ps_input(const IWineD3DPixelShaderImpl *This, const struct arb_ps_compile_args *args,
3186                           struct shader_arb_ctx_priv *priv)
3187 {
3188     const char *texcoords[8] =
3189     {
3190         "fragment.texcoord[0]", "fragment.texcoord[1]", "fragment.texcoord[2]", "fragment.texcoord[3]",
3191         "fragment.texcoord[4]", "fragment.texcoord[5]", "fragment.texcoord[6]", "fragment.texcoord[7]"
3192     };
3193     unsigned int i;
3194     const struct wined3d_shader_signature_element *sig = This->baseShader.input_signature;
3195     const char *semantic_name;
3196     DWORD semantic_idx;
3197
3198     switch(args->super.vp_mode)
3199     {
3200         case pretransformed:
3201         case fixedfunction:
3202             /* The pixelshader has to collect the varyings on its own. In any case properly load
3203              * color0 and color1. In the case of pretransformed vertices also load texcoords. Set
3204              * other attribs to 0.0.
3205              *
3206              * For fixedfunction this behavior is correct, according to the tests. For pretransformed
3207              * we'd either need a replacement shader that can load other attribs like BINORMAL, or
3208              * load the texcoord attrib pointers to match the pixel shader signature
3209              */
3210             for(i = 0; i < MAX_REG_INPUT; i++)
3211             {
3212                 semantic_name = sig[i].semantic_name;
3213                 semantic_idx = sig[i].semantic_idx;
3214                 if(semantic_name == NULL) continue;
3215
3216                 if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_COLOR))
3217                 {
3218                     if(semantic_idx == 0) priv->ps_input[i] = "fragment.color.primary";
3219                     else if(semantic_idx == 1) priv->ps_input[i] = "fragment.color.secondary";
3220                     else priv->ps_input[i] = "0.0";
3221                 }
3222                 else if(args->super.vp_mode == fixedfunction)
3223                 {
3224                     priv->ps_input[i] = "0.0";
3225                 }
3226                 else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_TEXCOORD))
3227                 {
3228                     if(semantic_idx < 8) priv->ps_input[i] = texcoords[semantic_idx];
3229                     else priv->ps_input[i] = "0.0";
3230                 }
3231                 else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_FOG))
3232                 {
3233                     if(semantic_idx == 0) priv->ps_input[i] = "fragment.fogcoord";
3234                     else priv->ps_input[i] = "0.0";
3235                 }
3236                 else
3237                 {
3238                     priv->ps_input[i] = "0.0";
3239                 }
3240
3241                 TRACE("v%u, semantic %s%u is %s\n", i, semantic_name, semantic_idx, priv->ps_input[i]);
3242             }
3243             break;
3244
3245         case vertexshader:
3246             /* That one is easy. The vertex shaders provide v0-v7 in fragment.texcoord and v8 and v9 in
3247              * fragment.color
3248              */
3249             for(i = 0; i < 8; i++)
3250             {
3251                 priv->ps_input[i] = texcoords[i];
3252             }
3253             priv->ps_input[8] = "fragment.color.primary";
3254             priv->ps_input[9] = "fragment.color.secondary";
3255             break;
3256     }
3257 }
3258
3259 /* GL locking is done by the caller */
3260 static GLuint shader_arb_generate_pshader(IWineD3DPixelShaderImpl *This, struct wined3d_shader_buffer *buffer,
3261         const struct arb_ps_compile_args *args, struct arb_ps_compiled_shader *compiled)
3262 {
3263     const shader_reg_maps* reg_maps = &This->baseShader.reg_maps;
3264     CONST DWORD *function = This->baseShader.function;
3265     const struct wined3d_gl_info *gl_info = &((IWineD3DDeviceImpl *)This->baseShader.device)->adapter->gl_info;
3266     const local_constant *lconst;
3267     GLuint retval;
3268     char fragcolor[16];
3269     DWORD *lconst_map = local_const_mapping((IWineD3DBaseShaderImpl *) This), next_local, cur;
3270     struct shader_arb_ctx_priv priv_ctx;
3271     BOOL dcl_tmp = args->super.srgb_correction, dcl_td = FALSE;
3272     BOOL want_nv_prog = FALSE;
3273     struct arb_pshader_private *shader_priv = This->baseShader.backend_data;
3274     GLint errPos;
3275     DWORD map;
3276
3277     char srgbtmp[4][4];
3278     unsigned int i, found = 0;
3279
3280     for (i = 0, map = reg_maps->temporary; map; map >>= 1, ++i)
3281     {
3282         if (!(map & 1)
3283                 || (This->color0_mov && i == This->color0_reg)
3284                 || (reg_maps->shader_version.major < 2 && i == 0))
3285             continue;
3286
3287         sprintf(srgbtmp[found], "R%u", i);
3288         ++found;
3289         if (found == 4) break;
3290     }
3291
3292     switch(found) {
3293         case 4: dcl_tmp = FALSE; break;
3294         case 0:
3295             sprintf(srgbtmp[0], "TA");
3296             sprintf(srgbtmp[1], "TB");
3297             sprintf(srgbtmp[2], "TC");
3298             sprintf(srgbtmp[3], "TD");
3299             dcl_td = TRUE;
3300             break;
3301         case 1:
3302             sprintf(srgbtmp[1], "TA");
3303             sprintf(srgbtmp[2], "TB");
3304             sprintf(srgbtmp[3], "TC");
3305             break;
3306         case 2:
3307             sprintf(srgbtmp[2], "TA");
3308             sprintf(srgbtmp[3], "TB");
3309             break;
3310         case 3:
3311             sprintf(srgbtmp[3], "TA");
3312             break;
3313     }
3314
3315     /*  Create the hw ARB shader */
3316     memset(&priv_ctx, 0, sizeof(priv_ctx));
3317     priv_ctx.cur_ps_args = args;
3318     priv_ctx.compiled_fprog = compiled;
3319     priv_ctx.cur_np2fixup_info = &compiled->np2fixup_info;
3320     init_ps_input(This, args, &priv_ctx);
3321     list_init(&priv_ctx.control_frames);
3322
3323     /* Avoid enabling NV_fragment_program* if we do not need it.
3324      *
3325      * Enabling GL_NV_fragment_program_option causes the driver to occupy a temporary register,
3326      * and it slows down the shader execution noticeably(about 5%). Usually our instruction emulation
3327      * is faster than what we gain from using higher native instructions. There are some things though
3328      * that cannot be emulated. In that case enable the extensions.
3329      * If the extension is enabled, instruction handlers that support both ways will use it.
3330      *
3331      * Testing shows no performance difference between OPTION NV_fragment_program2 and NV_fragment_program.
3332      * So enable the best we can get.
3333      */
3334     if(reg_maps->usesdsx || reg_maps->usesdsy || reg_maps->loop_depth > 0 || reg_maps->usestexldd ||
3335        reg_maps->usestexldl || reg_maps->usesfacing || reg_maps->usesifc || reg_maps->usescall)
3336     {
3337         want_nv_prog = TRUE;
3338     }
3339
3340     shader_addline(buffer, "!!ARBfp1.0\n");
3341     if (want_nv_prog && gl_info->supported[NV_FRAGMENT_PROGRAM2])
3342     {
3343         shader_addline(buffer, "OPTION NV_fragment_program2;\n");
3344         priv_ctx.target_version = NV3;
3345     }
3346     else if (want_nv_prog && gl_info->supported[NV_FRAGMENT_PROGRAM_OPTION])
3347     {
3348         shader_addline(buffer, "OPTION NV_fragment_program;\n");
3349         priv_ctx.target_version = NV2;
3350     } else {
3351         if(want_nv_prog)
3352         {
3353             /* This is an error - either we're advertising the wrong shader version, or aren't enforcing some
3354              * limits properly
3355              */
3356             ERR("The shader requires instructions that are not available in plain GL_ARB_fragment_program\n");
3357             ERR("Try GLSL\n");
3358         }
3359         priv_ctx.target_version = ARB;
3360     }
3361
3362     if(This->baseShader.reg_maps.highest_render_target > 0)
3363     {
3364         shader_addline(buffer, "OPTION ARB_draw_buffers;\n");
3365     }
3366
3367     if (reg_maps->shader_version.major < 3)
3368     {
3369         switch(args->super.fog) {
3370             case FOG_OFF:
3371                 break;
3372             case FOG_LINEAR:
3373                 shader_addline(buffer, "OPTION ARB_fog_linear;\n");
3374                 break;
3375             case FOG_EXP:
3376                 shader_addline(buffer, "OPTION ARB_fog_exp;\n");
3377                 break;
3378             case FOG_EXP2:
3379                 shader_addline(buffer, "OPTION ARB_fog_exp2;\n");
3380                 break;
3381         }
3382     }
3383
3384     /* For now always declare the temps. At least the Nvidia assembler optimizes completely
3385      * unused temps away(but occupies them for the whole shader if they're used once). Always
3386      * declaring them avoids tricky bookkeeping work
3387      */
3388     shader_addline(buffer, "TEMP TA;\n");      /* Used for modifiers */
3389     shader_addline(buffer, "TEMP TB;\n");      /* Used for modifiers */
3390     shader_addline(buffer, "TEMP TC;\n");      /* Used for modifiers */
3391     if(dcl_td) shader_addline(buffer, "TEMP TD;\n"); /* Used for sRGB writing */
3392     shader_addline(buffer, "PARAM coefdiv = { 0.5, 0.25, 0.125, 0.0625 };\n");
3393     shader_addline(buffer, "PARAM coefmul = { 2, 4, 8, 16 };\n");
3394     shader_addline(buffer, "PARAM one = { 1.0, 1.0, 1.0, 1.0 };\n");
3395
3396     if (reg_maps->shader_version.major < 2)
3397     {
3398         strcpy(fragcolor, "R0");
3399     } else {
3400         if(args->super.srgb_correction) {
3401             if(This->color0_mov) {
3402                 sprintf(fragcolor, "R%u", This->color0_reg);
3403             } else {
3404                 shader_addline(buffer, "TEMP TMP_COLOR;\n");
3405                 strcpy(fragcolor, "TMP_COLOR");
3406             }
3407         } else {
3408             strcpy(fragcolor, "result.color");
3409         }
3410     }
3411
3412     if(args->super.srgb_correction) {
3413         shader_addline(buffer, "PARAM srgb_consts1 = {%f, %f, %f, %f};\n",
3414                        srgb_mul_low, srgb_cmp, srgb_pow, srgb_mul_high);
3415         shader_addline(buffer, "PARAM srgb_consts2 = {%f, %f, %f, %f};\n",
3416                        srgb_sub_high, 0.0, 0.0, 0.0);
3417     }
3418
3419     /* Base Declarations */
3420     next_local = shader_generate_arb_declarations((IWineD3DBaseShader *)This,
3421             reg_maps, buffer, gl_info, lconst_map, NULL, &priv_ctx);
3422
3423     for (i = 0, map = reg_maps->bumpmat; map; map >>= 1, ++i)
3424     {
3425         if (!(map & 1)) continue;
3426
3427         cur = compiled->numbumpenvmatconsts;
3428         compiled->bumpenvmatconst[cur].const_num = WINED3D_CONST_NUM_UNUSED;
3429         compiled->bumpenvmatconst[cur].texunit = i;
3430         compiled->luminanceconst[cur].const_num = WINED3D_CONST_NUM_UNUSED;
3431         compiled->luminanceconst[cur].texunit = i;
3432
3433         /* We can fit the constants into the constant limit for sure because texbem, texbeml, bem and beml are only supported
3434          * in 1.x shaders, and GL_ARB_fragment_program has a constant limit of 24 constants. So in the worst case we're loading
3435          * 8 shader constants, 8 bump matrices and 8 luminance parameters and are perfectly fine. (No NP2 fixup on bumpmapped
3436          * textures due to conditional NP2 restrictions)
3437          *
3438          * Use local constants to load the bump env parameters, not program.env. This avoids collisions with d3d constants of
3439          * shaders in newer shader models. Since the bump env parameters have to share their space with NP2 fixup constants,
3440          * their location is shader dependent anyway and they cannot be loaded globally.
3441          */
3442         compiled->bumpenvmatconst[cur].const_num = next_local++;
3443         shader_addline(buffer, "PARAM bumpenvmat%d = program.local[%d];\n",
3444                        i, compiled->bumpenvmatconst[cur].const_num);
3445         compiled->numbumpenvmatconsts = cur + 1;
3446
3447         if (!(reg_maps->luminanceparams & (1 << i))) continue;
3448
3449         compiled->luminanceconst[cur].const_num = next_local++;
3450         shader_addline(buffer, "PARAM luminance%d = program.local[%d];\n",
3451                        i, compiled->luminanceconst[cur].const_num);
3452     }
3453
3454     for(i = 0; i < MAX_CONST_I; i++)
3455     {
3456         compiled->int_consts[i] = WINED3D_CONST_NUM_UNUSED;
3457         if (reg_maps->integer_constants & (1 << i) && priv_ctx.target_version >= NV2)
3458         {
3459             const DWORD *control_values = find_loop_control_values((IWineD3DBaseShaderImpl *) This, i);
3460
3461             if(control_values)
3462             {
3463                 shader_addline(buffer, "PARAM I%u = {%u, %u, %u, -1};\n", i,
3464                                 control_values[0], control_values[1], control_values[2]);
3465             }
3466             else
3467             {
3468                 compiled->int_consts[i] = next_local;
3469                 compiled->num_int_consts++;
3470                 shader_addline(buffer, "PARAM I%u = program.local[%u];\n", i, next_local++);
3471             }
3472         }
3473     }
3474
3475     if(reg_maps->vpos || reg_maps->usesdsy)
3476     {
3477         compiled->ycorrection = next_local;
3478         shader_addline(buffer, "PARAM ycorrection = program.local[%u];\n", next_local++);
3479
3480         if(reg_maps->vpos)
3481         {
3482             shader_addline(buffer, "TEMP vpos;\n");
3483             /* ycorrection.x: Backbuffer height(onscreen) or 0(offscreen).
3484              * ycorrection.y: -1.0(onscreen), 1.0(offscreen)
3485              * ycorrection.z: 1.0
3486              * ycorrection.w: 0.0
3487              */
3488             shader_addline(buffer, "MAD vpos, fragment.position, ycorrection.zyww, ycorrection.wxww;\n");
3489             shader_addline(buffer, "FLR vpos.xy, vpos;\n");
3490         }
3491     }
3492     else
3493     {
3494         compiled->ycorrection = WINED3D_CONST_NUM_UNUSED;
3495     }
3496
3497     /* Load constants to fixup NP2 texcoords if there are still free constants left:
3498      * Constants (texture dimensions) for the NP2 fixup are loaded as local program parameters. This will consume
3499      * at most 8 (MAX_FRAGMENT_SAMPLERS / 2) parameters, which is highly unlikely, since the application had to
3500      * use 16 NP2 textures at the same time. In case that we run out of constants the fixup is simply not
3501      * applied / activated. This will probably result in wrong rendering of the texture, but will save us from
3502      * shader compilation errors and the subsequent errors when drawing with this shader. */
3503     if (priv_ctx.cur_ps_args->super.np2_fixup) {
3504
3505         struct arb_ps_np2fixup_info* const fixup = priv_ctx.cur_np2fixup_info;
3506         const WORD map = priv_ctx.cur_ps_args->super.np2_fixup;
3507         const UINT max_lconsts = gl_info->limits.arb_ps_local_constants;
3508
3509         fixup->offset = next_local;
3510         fixup->super.active = 0;
3511
3512         cur = 0;
3513         for (i = 0; i < MAX_FRAGMENT_SAMPLERS; ++i) {
3514             if (!(map & (1 << i))) continue;
3515
3516             if (fixup->offset + (cur >> 1) < max_lconsts) {
3517                 fixup->super.active |= (1 << i);
3518                 fixup->super.idx[i] = cur++;
3519             } else {
3520                 FIXME("No free constant found to load NP2 fixup data into shader. "
3521                       "Sampling from this texture will probably look wrong.\n");
3522                 break;
3523             }
3524         }
3525
3526         fixup->super.num_consts = (cur + 1) >> 1;
3527         if (fixup->super.num_consts) {
3528             shader_addline(buffer, "PARAM np2fixup[%u] = { program.env[%u..%u] };\n",
3529                            fixup->super.num_consts, fixup->offset, fixup->super.num_consts + fixup->offset - 1);
3530         }
3531
3532         next_local += fixup->super.num_consts;
3533     }
3534
3535     if (shader_priv->clipplane_emulation != ~0U && args->clip)
3536     {
3537         shader_addline(buffer, "KIL fragment.texcoord[%u];\n", shader_priv->clipplane_emulation);
3538     }
3539
3540     /* Base Shader Body */
3541     shader_generate_main((IWineD3DBaseShader *)This, buffer, reg_maps, function, &priv_ctx);
3542
3543     if(args->super.srgb_correction) {
3544         arbfp_add_sRGB_correction(buffer, fragcolor, srgbtmp[0], srgbtmp[1], srgbtmp[2], srgbtmp[3],
3545                                   priv_ctx.target_version >= NV2);
3546     }
3547
3548     if(strcmp(fragcolor, "result.color")) {
3549         shader_addline(buffer, "MOV result.color, %s;\n", fragcolor);
3550     }
3551     shader_addline(buffer, "END\n");
3552
3553     /* TODO: change to resource.glObjectHandle or something like that */
3554     GL_EXTCALL(glGenProgramsARB(1, &retval));
3555
3556     TRACE("Creating a hw pixel shader, prg=%d\n", retval);
3557     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, retval));
3558
3559     TRACE("Created hw pixel shader, prg=%d\n", retval);
3560     /* Create the program and check for errors */
3561     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
3562                buffer->bsize, buffer->buffer));
3563     checkGLcall("glProgramStringARB()");
3564
3565     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &errPos);
3566     if (errPos != -1)
3567     {
3568         FIXME("HW PixelShader Error at position %d: %s\n",
3569               errPos, debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
3570         retval = 0;
3571     }
3572     else
3573     {
3574         GLint native;
3575
3576         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
3577         checkGLcall("glGetProgramivARB()");
3578         if (!native) WARN("Program exceeds native resource limits.\n");
3579     }
3580
3581     /* Load immediate constants */
3582     if(lconst_map) {
3583         LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry) {
3584             const float *value = (const float *)lconst->value;
3585             GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, lconst_map[lconst->idx], value));
3586             checkGLcall("glProgramLocalParameter4fvARB");
3587         }
3588         HeapFree(GetProcessHeap(), 0, lconst_map);
3589     }
3590
3591     return retval;
3592 }
3593
3594 static int compare_sig(const struct wined3d_shader_signature_element *sig1, const struct wined3d_shader_signature_element *sig2)
3595 {
3596     unsigned int i;
3597     int ret;
3598
3599     for(i = 0; i < MAX_REG_INPUT; i++)
3600     {
3601         if(sig1[i].semantic_name == NULL || sig2[i].semantic_name == NULL)
3602         {
3603             /* Compare pointers, not contents. One string is NULL(element does not exist), the other one is not NULL */
3604             if(sig1[i].semantic_name != sig2[i].semantic_name) return sig1[i].semantic_name < sig2[i].semantic_name ? -1 : 1;
3605             continue;
3606         }
3607
3608         ret = strcmp(sig1[i].semantic_name, sig2[i].semantic_name);
3609         if(ret != 0) return ret;
3610         if(sig1[i].semantic_idx    != sig2[i].semantic_idx)    return sig1[i].semantic_idx    < sig2[i].semantic_idx    ? -1 : 1;
3611         if(sig1[i].sysval_semantic != sig2[i].sysval_semantic) return sig1[i].sysval_semantic < sig2[i].sysval_semantic ? -1 : 1;
3612         if(sig1[i].component_type  != sig2[i].component_type)  return sig1[i].sysval_semantic < sig2[i].component_type  ? -1 : 1;
3613         if(sig1[i].register_idx    != sig2[i].register_idx)    return sig1[i].register_idx    < sig2[i].register_idx    ? -1 : 1;
3614         if(sig1[i].mask            != sig2->mask)              return sig1[i].mask            < sig2[i].mask            ? -1 : 1;
3615     }
3616     return 0;
3617 }
3618
3619 static struct wined3d_shader_signature_element *clone_sig(const struct wined3d_shader_signature_element *sig)
3620 {
3621     struct wined3d_shader_signature_element *new;
3622     int i;
3623     char *name;
3624
3625     new = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*new) * MAX_REG_INPUT);
3626     for(i = 0; i < MAX_REG_INPUT; i++)
3627     {
3628         if(sig[i].semantic_name == NULL)
3629         {
3630             continue;
3631         }
3632
3633         new[i] = sig[i];
3634         /* Clone the semantic string */
3635         name = HeapAlloc(GetProcessHeap(), 0, strlen(sig[i].semantic_name) + 1);
3636         strcpy(name, sig[i].semantic_name);
3637         new[i].semantic_name = name;
3638     }
3639     return new;
3640 }
3641
3642 static DWORD find_input_signature(struct shader_arb_priv *priv, const struct wined3d_shader_signature_element *sig)
3643 {
3644     struct wine_rb_entry *entry = wine_rb_get(&priv->signature_tree, sig);
3645     struct ps_signature *found_sig;
3646
3647     if(entry != NULL)
3648     {
3649         found_sig = WINE_RB_ENTRY_VALUE(entry, struct ps_signature, entry);
3650         TRACE("Found existing signature %u\n", found_sig->idx);
3651         return found_sig->idx;
3652     }
3653     found_sig = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*sig));
3654     found_sig->sig = clone_sig(sig);
3655     found_sig->idx = priv->ps_sig_number++;
3656     TRACE("New signature stored and assigned number %u\n", found_sig->idx);
3657     if(wine_rb_put(&priv->signature_tree, sig, &found_sig->entry) == -1)
3658     {
3659         ERR("Failed to insert program entry.\n");
3660     }
3661     return found_sig->idx;
3662 }
3663
3664 static void init_output_registers(IWineD3DVertexShaderImpl *shader, DWORD sig_num, struct shader_arb_ctx_priv *priv_ctx,
3665                                   struct arb_vs_compiled_shader *compiled)
3666 {
3667     unsigned int i, j;
3668     static const char *texcoords[8] =
3669     {
3670         "result.texcoord[0]", "result.texcoord[1]", "result.texcoord[2]", "result.texcoord[3]",
3671         "result.texcoord[4]", "result.texcoord[5]", "result.texcoord[6]", "result.texcoord[7]"
3672     };
3673     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *) shader->baseShader.device;
3674     IWineD3DBaseShaderClass *baseshader = &shader->baseShader;
3675     const struct wined3d_shader_signature_element *sig;
3676     const char *semantic_name;
3677     DWORD semantic_idx, reg_idx;
3678
3679     /* Write generic input varyings 0 to 7 to result.texcoord[], varying 8 to result.color.primary
3680      * and varying 9 to result.color.secondary
3681      */
3682     const char *decl_idx_to_string[MAX_REG_INPUT] =
3683     {
3684         texcoords[0], texcoords[1], texcoords[2], texcoords[3],
3685         texcoords[4], texcoords[5], texcoords[6], texcoords[7],
3686         "result.color.primary", "result.color.secondary"
3687     };
3688
3689     if(sig_num == ~0)
3690     {
3691         TRACE("Pixel shader uses builtin varyings\n");
3692         /* Map builtins to builtins */
3693         for(i = 0; i < 8; i++)
3694         {
3695             priv_ctx->texcrd_output[i] = texcoords[i];
3696         }
3697         priv_ctx->color_output[0] = "result.color.primary";
3698         priv_ctx->color_output[1] = "result.color.secondary";
3699         priv_ctx->fog_output = "result.fogcoord";
3700
3701         /* Map declared regs to builtins. Use "TA" to /dev/null unread output */
3702         for (i = 0; i < (sizeof(baseshader->output_signature) / sizeof(*baseshader->output_signature)); ++i)
3703         {
3704             semantic_name = baseshader->output_signature[i].semantic_name;
3705             if(semantic_name == NULL) continue;
3706
3707             if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_POSITION))
3708             {
3709                 TRACE("o%u is TMP_OUT\n", i);
3710                 if (baseshader->output_signature[i].semantic_idx == 0) priv_ctx->vs_output[i] = "TMP_OUT";
3711                 else priv_ctx->vs_output[i] = "TA";
3712             }
3713             else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_PSIZE))
3714             {
3715                 TRACE("o%u is result.pointsize\n", i);
3716                 if (baseshader->output_signature[i].semantic_idx == 0) priv_ctx->vs_output[i] = "result.pointsize";
3717                 else priv_ctx->vs_output[i] = "TA";
3718             }
3719             else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_COLOR))
3720             {
3721                 TRACE("o%u is result.color.?, idx %u\n", i, baseshader->output_signature[i].semantic_idx);
3722                 if (baseshader->output_signature[i].semantic_idx == 0)
3723                     priv_ctx->vs_output[i] = "result.color.primary";
3724                 else if (baseshader->output_signature[i].semantic_idx == 1)
3725                     priv_ctx->vs_output[i] = "result.color.secondary";
3726                 else priv_ctx->vs_output[i] = "TA";
3727             }
3728             else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_TEXCOORD))
3729             {
3730                 TRACE("o%u is %s\n", i, texcoords[baseshader->output_signature[i].semantic_idx]);
3731                 if (baseshader->output_signature[i].semantic_idx >= 8) priv_ctx->vs_output[i] = "TA";
3732                 else priv_ctx->vs_output[i] = texcoords[baseshader->output_signature[i].semantic_idx];
3733             }
3734             else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_FOG))
3735             {
3736                 TRACE("o%u is result.fogcoord\n", i);
3737                 if (baseshader->output_signature[i].semantic_idx > 0) priv_ctx->vs_output[i] = "TA";
3738                 else priv_ctx->vs_output[i] = "result.fogcoord";
3739             }
3740             else
3741             {
3742                 priv_ctx->vs_output[i] = "TA";
3743             }
3744         }
3745         return;
3746     }
3747
3748     /* Instead of searching for the signature in the signature list, read the one from the current pixel shader.
3749      * Its maybe not the shader where the signature came from, but it is the same signature and faster to find
3750      */
3751     sig = ((IWineD3DPixelShaderImpl *)device->stateBlock->pixelShader)->baseShader.input_signature;
3752     TRACE("Pixel shader uses declared varyings\n");
3753
3754     /* Map builtin to declared. /dev/null the results by default to the TA temp reg */
3755     for(i = 0; i < 8; i++)
3756     {
3757         priv_ctx->texcrd_output[i] = "TA";
3758     }
3759     priv_ctx->color_output[0] = "TA";
3760     priv_ctx->color_output[1] = "TA";
3761     priv_ctx->fog_output = "TA";
3762
3763     for(i = 0; i < MAX_REG_INPUT; i++)
3764     {
3765         semantic_name = sig[i].semantic_name;
3766         semantic_idx = sig[i].semantic_idx;
3767         reg_idx = sig[i].register_idx;
3768         if(semantic_name == NULL) continue;
3769
3770         /* If a declared input register is not written by builtin arguments, don't write to it.
3771          * GL_NV_vertex_program makes sure the input defaults to 0.0, which is correct with D3D
3772          *
3773          * Don't care about POSITION and PSIZE here - this is a builtin vertex shader, position goes
3774          * to TMP_OUT in any case
3775          */
3776         if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_TEXCOORD))
3777         {
3778             if(semantic_idx < 8) priv_ctx->texcrd_output[semantic_idx] = decl_idx_to_string[reg_idx];
3779         }
3780         else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_COLOR))
3781         {
3782             if(semantic_idx < 2) priv_ctx->color_output[semantic_idx] = decl_idx_to_string[reg_idx];
3783         }
3784         else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_FOG))
3785         {
3786             if(semantic_idx == 0) priv_ctx->fog_output = decl_idx_to_string[reg_idx];
3787         }
3788         else
3789         {
3790             continue;
3791         }
3792
3793         if(strcmp(decl_idx_to_string[reg_idx], "result.color.primary") == 0 ||
3794            strcmp(decl_idx_to_string[reg_idx], "result.color.secondary") == 0)
3795         {
3796             compiled->need_color_unclamp = TRUE;
3797         }
3798     }
3799
3800     /* Map declared to declared */
3801     for (i = 0; i < (sizeof(baseshader->output_signature) / sizeof(*baseshader->output_signature)); ++i)
3802     {
3803         /* Write unread output to TA to throw them away */
3804         priv_ctx->vs_output[i] = "TA";
3805         semantic_name = baseshader->output_signature[i].semantic_name;
3806         if(semantic_name == NULL)
3807         {
3808             continue;
3809         }
3810
3811         if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_POSITION)
3812                 && baseshader->output_signature[i].semantic_idx == 0)
3813         {
3814             priv_ctx->vs_output[i] = "TMP_OUT";
3815             continue;
3816         }
3817         else if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_PSIZE)
3818                 && baseshader->output_signature[i].semantic_idx == 0)
3819         {
3820             priv_ctx->vs_output[i] = "result.pointsize";
3821             continue;
3822         }
3823
3824         for(j = 0; j < MAX_REG_INPUT; j++)
3825         {
3826             if(sig[j].semantic_name == NULL)
3827             {
3828                 continue;
3829             }
3830
3831             if (strcmp(sig[j].semantic_name, semantic_name) == 0
3832                     && sig[j].semantic_idx == baseshader->output_signature[i].semantic_idx)
3833             {
3834                 priv_ctx->vs_output[i] = decl_idx_to_string[sig[j].register_idx];
3835
3836                 if(strcmp(priv_ctx->vs_output[i], "result.color.primary") == 0 ||
3837                    strcmp(priv_ctx->vs_output[i], "result.color.secondary") == 0)
3838                 {
3839                     compiled->need_color_unclamp = TRUE;
3840                 }
3841             }
3842         }
3843     }
3844 }
3845
3846 /* GL locking is done by the caller */
3847 static GLuint shader_arb_generate_vshader(IWineD3DVertexShaderImpl *This, struct wined3d_shader_buffer *buffer,
3848         const struct arb_vs_compile_args *args, struct arb_vs_compiled_shader *compiled)
3849 {
3850     const shader_reg_maps *reg_maps = &This->baseShader.reg_maps;
3851     CONST DWORD *function = This->baseShader.function;
3852     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *)This->baseShader.device;
3853     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
3854     const local_constant *lconst;
3855     GLuint ret;
3856     DWORD next_local, *lconst_map = local_const_mapping((IWineD3DBaseShaderImpl *) This);
3857     struct shader_arb_ctx_priv priv_ctx;
3858     unsigned int i;
3859     GLint errPos;
3860
3861     memset(&priv_ctx, 0, sizeof(priv_ctx));
3862     priv_ctx.cur_vs_args = args;
3863     list_init(&priv_ctx.control_frames);
3864     init_output_registers(This, args->ps_signature, &priv_ctx, compiled);
3865
3866     /*  Create the hw ARB shader */
3867     shader_addline(buffer, "!!ARBvp1.0\n");
3868
3869     /* Always enable the NV extension if available. Unlike fragment shaders, there is no
3870      * mesurable performance penalty, and we can always make use of it for clipplanes.
3871      */
3872     if (gl_info->supported[NV_VERTEX_PROGRAM3])
3873     {
3874         shader_addline(buffer, "OPTION NV_vertex_program3;\n");
3875         priv_ctx.target_version = NV3;
3876         shader_addline(buffer, "ADDRESS aL;\n");
3877     }
3878     else if (gl_info->supported[NV_VERTEX_PROGRAM2_OPTION])
3879     {
3880         shader_addline(buffer, "OPTION NV_vertex_program2;\n");
3881         priv_ctx.target_version = NV2;
3882         shader_addline(buffer, "ADDRESS aL;\n");
3883     } else {
3884         priv_ctx.target_version = ARB;
3885     }
3886
3887     shader_addline(buffer, "TEMP TMP_OUT;\n");
3888     if(need_helper_const(gl_info)) {
3889         shader_addline(buffer, "PARAM helper_const = { 2.0, -1.0, %d.0, 0.0 };\n", This->rel_offset);
3890     }
3891     if(need_mova_const((IWineD3DBaseShader *) This, gl_info)) {
3892         shader_addline(buffer, "PARAM mova_const = { 0.5, 0.0, 2.0, 1.0 };\n");
3893         shader_addline(buffer, "TEMP A0_SHADOW;\n");
3894     }
3895
3896     shader_addline(buffer, "TEMP TA;\n");
3897
3898     /* Base Declarations */
3899     next_local = shader_generate_arb_declarations((IWineD3DBaseShader *)This,
3900             reg_maps, buffer, gl_info, lconst_map, &priv_ctx.vs_clipplanes, &priv_ctx);
3901
3902     for(i = 0; i < MAX_CONST_I; i++)
3903     {
3904         compiled->int_consts[i] = WINED3D_CONST_NUM_UNUSED;
3905         if(reg_maps->integer_constants & (1 << i) && priv_ctx.target_version >= NV2)
3906         {
3907             const DWORD *control_values = find_loop_control_values((IWineD3DBaseShaderImpl *) This, i);
3908
3909             if(control_values)
3910             {
3911                 shader_addline(buffer, "PARAM I%u = {%u, %u, %u, -1};\n", i,
3912                                 control_values[0], control_values[1], control_values[2]);
3913             }
3914             else
3915             {
3916                 compiled->int_consts[i] = next_local;
3917                 compiled->num_int_consts++;
3918                 shader_addline(buffer, "PARAM I%u = program.local[%u];\n", i, next_local++);
3919             }
3920         }
3921     }
3922
3923     /* We need a constant to fixup the final position */
3924     shader_addline(buffer, "PARAM posFixup = program.local[%u];\n", next_local);
3925     compiled->pos_fixup = next_local++;
3926
3927     /* Initialize output parameters. GL_ARB_vertex_program does not require special initialization values
3928      * for output parameters. D3D in theory does not do that either, but some applications depend on a
3929      * proper initialization of the secondary color, and programs using the fixed function pipeline without
3930      * a replacement shader depend on the texcoord.w being set properly.
3931      *
3932      * GL_NV_vertex_program defines that all output values are initialized to {0.0, 0.0, 0.0, 1.0}. This
3933      * assertion is in effect even when using GL_ARB_vertex_program without any NV specific additions. So
3934      * skip this if NV_vertex_program is supported. Otherwise, initialize the secondary color. For the tex-
3935      * coords, we have a flag in the opengl caps. Many cards do not require the texcoord being set, and
3936      * this can eat a number of instructions, so skip it unless this cap is set as well
3937      */
3938     if (!gl_info->supported[NV_VERTEX_PROGRAM])
3939     {
3940         shader_addline(buffer, "MOV result.color.secondary, -helper_const.wwwy;\n");
3941
3942         if (gl_info->quirks & WINED3D_QUIRK_SET_TEXCOORD_W && !device->frag_pipe->ffp_proj_control)
3943         {
3944             int i;
3945             for(i = 0; i < min(8, MAX_REG_TEXCRD); i++) {
3946                 if(This->baseShader.reg_maps.texcoord_mask[i] != 0 &&
3947                 This->baseShader.reg_maps.texcoord_mask[i] != WINED3DSP_WRITEMASK_ALL) {
3948                     shader_addline(buffer, "MOV result.texcoord[%u].w, -helper_const.y;\n", i);
3949                 }
3950             }
3951         }
3952     }
3953
3954     /* The shader starts with the main function */
3955     priv_ctx.in_main_func = TRUE;
3956     /* Base Shader Body */
3957     shader_generate_main((IWineD3DBaseShader *)This, buffer, reg_maps, function, &priv_ctx);
3958
3959     if(!priv_ctx.footer_written) vshader_add_footer(This, buffer, args, &priv_ctx);
3960
3961     shader_addline(buffer, "END\n");
3962
3963     /* TODO: change to resource.glObjectHandle or something like that */
3964     GL_EXTCALL(glGenProgramsARB(1, &ret));
3965
3966     TRACE("Creating a hw vertex shader, prg=%d\n", ret);
3967     GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, ret));
3968
3969     TRACE("Created hw vertex shader, prg=%d\n", ret);
3970     /* Create the program and check for errors */
3971     GL_EXTCALL(glProgramStringARB(GL_VERTEX_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
3972                buffer->bsize, buffer->buffer));
3973     checkGLcall("glProgramStringARB()");
3974
3975     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &errPos);
3976     if (errPos != -1)
3977     {
3978         FIXME("HW VertexShader Error at position %d: %s\n",
3979               errPos, debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
3980         ret = -1;
3981     }
3982     else
3983     {
3984         GLint native;
3985
3986         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
3987         checkGLcall("glGetProgramivARB()");
3988         if (!native) WARN("Program exceeds native resource limits.\n");
3989
3990         /* Load immediate constants */
3991         if(lconst_map) {
3992             LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry) {
3993                 const float *value = (const float *)lconst->value;
3994                 GL_EXTCALL(glProgramLocalParameter4fvARB(GL_VERTEX_PROGRAM_ARB, lconst_map[lconst->idx], value));
3995             }
3996         }
3997     }
3998     HeapFree(GetProcessHeap(), 0, lconst_map);
3999
4000     return ret;
4001 }
4002
4003 /* GL locking is done by the caller */
4004 static struct arb_ps_compiled_shader *find_arb_pshader(IWineD3DPixelShaderImpl *shader, const struct arb_ps_compile_args *args)
4005 {
4006     UINT i;
4007     DWORD new_size;
4008     struct arb_ps_compiled_shader *new_array;
4009     struct wined3d_shader_buffer buffer;
4010     struct arb_pshader_private *shader_data;
4011     GLuint ret;
4012
4013     if (!shader->baseShader.backend_data)
4014     {
4015         IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *) shader->baseShader.device;
4016         const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4017         struct shader_arb_priv *priv = device->shader_priv;
4018
4019         shader->baseShader.backend_data = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data));
4020         shader_data = shader->baseShader.backend_data;
4021         shader_data->clamp_consts = shader->baseShader.reg_maps.shader_version.major == 1;
4022
4023         if(shader->baseShader.reg_maps.shader_version.major < 3) shader_data->input_signature_idx = ~0;
4024         else shader_data->input_signature_idx = find_input_signature(priv, shader->baseShader.input_signature);
4025
4026         shader_data->has_signature_idx = TRUE;
4027         TRACE("Shader got assigned input signature index %u\n", shader_data->input_signature_idx);
4028
4029         if (!device->vs_clipping)
4030             shader_data->clipplane_emulation = shader_find_free_input_register(&shader->baseShader.reg_maps,
4031                     gl_info->limits.texture_stages - 1);
4032         else
4033             shader_data->clipplane_emulation = ~0U;
4034     }
4035     shader_data = shader->baseShader.backend_data;
4036
4037     /* Usually we have very few GL shaders for each d3d shader(just 1 or maybe 2),
4038      * so a linear search is more performant than a hashmap or a binary search
4039      * (cache coherency etc)
4040      */
4041     for(i = 0; i < shader_data->num_gl_shaders; i++) {
4042         if(memcmp(&shader_data->gl_shaders[i].args, args, sizeof(*args)) == 0) {
4043             return &shader_data->gl_shaders[i];
4044         }
4045     }
4046
4047     TRACE("No matching GL shader found, compiling a new shader\n");
4048     if(shader_data->shader_array_size == shader_data->num_gl_shaders) {
4049         if (shader_data->num_gl_shaders)
4050         {
4051             new_size = shader_data->shader_array_size + max(1, shader_data->shader_array_size / 2);
4052             new_array = HeapReAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, shader_data->gl_shaders,
4053                                     new_size * sizeof(*shader_data->gl_shaders));
4054         } else {
4055             new_array = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data->gl_shaders));
4056             new_size = 1;
4057         }
4058
4059         if(!new_array) {
4060             ERR("Out of memory\n");
4061             return 0;
4062         }
4063         shader_data->gl_shaders = new_array;
4064         shader_data->shader_array_size = new_size;
4065     }
4066
4067     shader_data->gl_shaders[shader_data->num_gl_shaders].args = *args;
4068
4069     pixelshader_update_samplers(&shader->baseShader.reg_maps,
4070             ((IWineD3DDeviceImpl *)shader->baseShader.device)->stateBlock->textures);
4071
4072     if (!shader_buffer_init(&buffer))
4073     {
4074         ERR("Failed to initialize shader buffer.\n");
4075         return 0;
4076     }
4077
4078     ret = shader_arb_generate_pshader(shader, &buffer, args,
4079                                       &shader_data->gl_shaders[shader_data->num_gl_shaders]);
4080     shader_buffer_free(&buffer);
4081     shader_data->gl_shaders[shader_data->num_gl_shaders].prgId = ret;
4082
4083     return &shader_data->gl_shaders[shader_data->num_gl_shaders++];
4084 }
4085
4086 static inline BOOL vs_args_equal(const struct arb_vs_compile_args *stored, const struct arb_vs_compile_args *new,
4087                                  const DWORD use_map, BOOL skip_int) {
4088     if((stored->super.swizzle_map & use_map) != new->super.swizzle_map) return FALSE;
4089     if(stored->super.clip_enabled != new->super.clip_enabled) return FALSE;
4090     if(stored->super.fog_src != new->super.fog_src) return FALSE;
4091     if(stored->clip.boolclip_compare != new->clip.boolclip_compare) return FALSE;
4092     if(stored->ps_signature != new->ps_signature) return FALSE;
4093     if(stored->vertex.samplers_compare != new->vertex.samplers_compare) return FALSE;
4094     if(skip_int) return TRUE;
4095
4096     return memcmp(stored->loop_ctrl, new->loop_ctrl, sizeof(stored->loop_ctrl)) == 0;
4097 }
4098
4099 static struct arb_vs_compiled_shader *find_arb_vshader(IWineD3DVertexShaderImpl *shader, const struct arb_vs_compile_args *args)
4100 {
4101     UINT i;
4102     DWORD new_size;
4103     struct arb_vs_compiled_shader *new_array;
4104     DWORD use_map = ((IWineD3DDeviceImpl *)shader->baseShader.device)->strided_streams.use_map;
4105     struct wined3d_shader_buffer buffer;
4106     struct arb_vshader_private *shader_data;
4107     GLuint ret;
4108     const struct wined3d_gl_info *gl_info = &((IWineD3DDeviceImpl *)shader->baseShader.device)->adapter->gl_info;
4109
4110     if (!shader->baseShader.backend_data)
4111     {
4112         shader->baseShader.backend_data = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data));
4113     }
4114     shader_data = shader->baseShader.backend_data;
4115
4116     /* Usually we have very few GL shaders for each d3d shader(just 1 or maybe 2),
4117      * so a linear search is more performant than a hashmap or a binary search
4118      * (cache coherency etc)
4119      */
4120     for(i = 0; i < shader_data->num_gl_shaders; i++) {
4121         if (vs_args_equal(&shader_data->gl_shaders[i].args, args,
4122                 use_map, gl_info->supported[NV_VERTEX_PROGRAM2_OPTION]))
4123         {
4124             return &shader_data->gl_shaders[i];
4125         }
4126     }
4127
4128     TRACE("No matching GL shader found, compiling a new shader\n");
4129
4130     if(shader_data->shader_array_size == shader_data->num_gl_shaders) {
4131         if (shader_data->num_gl_shaders)
4132         {
4133             new_size = shader_data->shader_array_size + max(1, shader_data->shader_array_size / 2);
4134             new_array = HeapReAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, shader_data->gl_shaders,
4135                                     new_size * sizeof(*shader_data->gl_shaders));
4136         } else {
4137             new_array = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data->gl_shaders));
4138             new_size = 1;
4139         }
4140
4141         if(!new_array) {
4142             ERR("Out of memory\n");
4143             return 0;
4144         }
4145         shader_data->gl_shaders = new_array;
4146         shader_data->shader_array_size = new_size;
4147     }
4148
4149     shader_data->gl_shaders[shader_data->num_gl_shaders].args = *args;
4150
4151     if (!shader_buffer_init(&buffer))
4152     {
4153         ERR("Failed to initialize shader buffer.\n");
4154         return 0;
4155     }
4156
4157     ret = shader_arb_generate_vshader(shader, &buffer, args,
4158             &shader_data->gl_shaders[shader_data->num_gl_shaders]);
4159     shader_buffer_free(&buffer);
4160     shader_data->gl_shaders[shader_data->num_gl_shaders].prgId = ret;
4161
4162     return &shader_data->gl_shaders[shader_data->num_gl_shaders++];
4163 }
4164
4165 static inline void find_arb_ps_compile_args(IWineD3DPixelShaderImpl *shader, IWineD3DStateBlockImpl *stateblock,
4166         struct arb_ps_compile_args *args)
4167 {
4168     int i;
4169     WORD int_skip;
4170     const struct wined3d_gl_info *gl_info = &((IWineD3DDeviceImpl *)shader->baseShader.device)->adapter->gl_info;
4171     find_ps_compile_args(shader, stateblock, &args->super);
4172
4173     /* This forces all local boolean constants to 1 to make them stateblock independent */
4174     args->bools = shader->baseShader.reg_maps.local_bool_consts;
4175
4176     for(i = 0; i < MAX_CONST_B; i++)
4177     {
4178         if(stateblock->pixelShaderConstantB[i]) args->bools |= ( 1 << i);
4179     }
4180
4181     /* Only enable the clip plane emulation KIL if at least one clipplane is enabled. The KIL instruction
4182      * is quite expensive because it forces the driver to disable early Z discards. It is cheaper to
4183      * duplicate the shader than have a no-op KIL instruction in every shader
4184      */
4185     if((!((IWineD3DDeviceImpl *) shader->baseShader.device)->vs_clipping) && use_vs(stateblock) &&
4186        stateblock->renderState[WINED3DRS_CLIPPING] && stateblock->renderState[WINED3DRS_CLIPPLANEENABLE])
4187     {
4188         args->clip = 1;
4189     }
4190     else
4191     {
4192         args->clip = 0;
4193     }
4194
4195     /* Skip if unused or local, or supported natively */
4196     int_skip = ~shader->baseShader.reg_maps.integer_constants | shader->baseShader.reg_maps.local_int_consts;
4197     if (int_skip == 0xffff || gl_info->supported[NV_FRAGMENT_PROGRAM_OPTION])
4198     {
4199         memset(&args->loop_ctrl, 0, sizeof(args->loop_ctrl));
4200         return;
4201     }
4202
4203     for(i = 0; i < MAX_CONST_I; i++)
4204     {
4205         if(int_skip & (1 << i))
4206         {
4207             args->loop_ctrl[i][0] = 0;
4208             args->loop_ctrl[i][1] = 0;
4209             args->loop_ctrl[i][2] = 0;
4210         }
4211         else
4212         {
4213             args->loop_ctrl[i][0] = stateblock->pixelShaderConstantI[i * 4];
4214             args->loop_ctrl[i][1] = stateblock->pixelShaderConstantI[i * 4 + 1];
4215             args->loop_ctrl[i][2] = stateblock->pixelShaderConstantI[i * 4 + 2];
4216         }
4217     }
4218 }
4219
4220 static inline void find_arb_vs_compile_args(IWineD3DVertexShaderImpl *shader, IWineD3DStateBlockImpl *stateblock,
4221         struct arb_vs_compile_args *args)
4222 {
4223     int i;
4224     WORD int_skip;
4225     IWineD3DDeviceImpl *dev = (IWineD3DDeviceImpl *)shader->baseShader.device;
4226     const struct wined3d_gl_info *gl_info = &dev->adapter->gl_info;
4227     find_vs_compile_args(shader, stateblock, &args->super);
4228
4229     args->clip.boolclip_compare = 0;
4230     if(use_ps(stateblock))
4231     {
4232         IWineD3DPixelShaderImpl *ps = (IWineD3DPixelShaderImpl *) stateblock->pixelShader;
4233         struct arb_pshader_private *shader_priv = ps->baseShader.backend_data;
4234         args->ps_signature = shader_priv->input_signature_idx;
4235
4236         args->clip.boolclip.clip_texcoord = shader_priv->clipplane_emulation + 1;
4237     }
4238     else
4239     {
4240         args->ps_signature = ~0;
4241         if(!dev->vs_clipping)
4242         {
4243             args->clip.boolclip.clip_texcoord = ffp_clip_emul(stateblock) ? gl_info->limits.texture_stages : 0;
4244         }
4245         /* Otherwise: Setting boolclip_compare set clip_texcoord to 0 */
4246     }
4247
4248     if(args->clip.boolclip.clip_texcoord)
4249     {
4250         if(stateblock->renderState[WINED3DRS_CLIPPING])
4251         {
4252             args->clip.boolclip.clipplane_mask = stateblock->renderState[WINED3DRS_CLIPPLANEENABLE];
4253         }
4254         /* clipplane_mask was set to 0 by setting boolclip_compare to 0 */
4255     }
4256
4257     /* This forces all local boolean constants to 1 to make them stateblock independent */
4258     args->clip.boolclip.bools = shader->baseShader.reg_maps.local_bool_consts;
4259     /* TODO: Figure out if it would be better to store bool constants as bitmasks in the stateblock */
4260     for(i = 0; i < MAX_CONST_B; i++)
4261     {
4262         if(stateblock->vertexShaderConstantB[i]) args->clip.boolclip.bools |= ( 1 << i);
4263     }
4264
4265     args->vertex.samplers[0] = dev->texUnitMap[MAX_FRAGMENT_SAMPLERS + 0];
4266     args->vertex.samplers[1] = dev->texUnitMap[MAX_FRAGMENT_SAMPLERS + 1];
4267     args->vertex.samplers[2] = dev->texUnitMap[MAX_FRAGMENT_SAMPLERS + 2];
4268     args->vertex.samplers[3] = 0;
4269
4270     /* Skip if unused or local */
4271     int_skip = ~shader->baseShader.reg_maps.integer_constants | shader->baseShader.reg_maps.local_int_consts;
4272     /* This is about flow control, not clipping. */
4273     if (int_skip == 0xffff || gl_info->supported[NV_VERTEX_PROGRAM2_OPTION])
4274     {
4275         memset(&args->loop_ctrl, 0, sizeof(args->loop_ctrl));
4276         return;
4277     }
4278
4279     for(i = 0; i < MAX_CONST_I; i++)
4280     {
4281         if(int_skip & (1 << i))
4282         {
4283             args->loop_ctrl[i][0] = 0;
4284             args->loop_ctrl[i][1] = 0;
4285             args->loop_ctrl[i][2] = 0;
4286         }
4287         else
4288         {
4289             args->loop_ctrl[i][0] = stateblock->vertexShaderConstantI[i * 4];
4290             args->loop_ctrl[i][1] = stateblock->vertexShaderConstantI[i * 4 + 1];
4291             args->loop_ctrl[i][2] = stateblock->vertexShaderConstantI[i * 4 + 2];
4292         }
4293     }
4294 }
4295
4296 /* GL locking is done by the caller */
4297 static void shader_arb_select(const struct wined3d_context *context, BOOL usePS, BOOL useVS)
4298 {
4299     IWineD3DDeviceImpl *This = ((IWineD3DSurfaceImpl *)context->surface)->resource.device;
4300     struct shader_arb_priv *priv = This->shader_priv;
4301     const struct wined3d_gl_info *gl_info = context->gl_info;
4302     int i;
4303
4304     /* Deal with pixel shaders first so the vertex shader arg function has the input signature ready */
4305     if (usePS) {
4306         struct arb_ps_compile_args compile_args;
4307         struct arb_ps_compiled_shader *compiled;
4308         IWineD3DPixelShaderImpl *ps = (IWineD3DPixelShaderImpl *) This->stateBlock->pixelShader;
4309
4310         TRACE("Using pixel shader %p\n", This->stateBlock->pixelShader);
4311         find_arb_ps_compile_args(ps, This->stateBlock, &compile_args);
4312         compiled = find_arb_pshader(ps, &compile_args);
4313         priv->current_fprogram_id = compiled->prgId;
4314         priv->compiled_fprog = compiled;
4315
4316         /* Bind the fragment program */
4317         GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, priv->current_fprogram_id));
4318         checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, priv->current_fprogram_id);");
4319
4320         if(!priv->use_arbfp_fixed_func) {
4321             /* Enable OpenGL fragment programs */
4322             glEnable(GL_FRAGMENT_PROGRAM_ARB);
4323             checkGLcall("glEnable(GL_FRAGMENT_PROGRAM_ARB);");
4324         }
4325         TRACE("(%p) : Bound fragment program %u and enabled GL_FRAGMENT_PROGRAM_ARB\n", This, priv->current_fprogram_id);
4326
4327         /* Pixel Shader 1.x constants are clamped to [-1;1], Pixel Shader 2.0 constants are not. If switching between
4328          * a 1.x and newer shader, reload the first 8 constants
4329          */
4330         if(priv->last_ps_const_clamped != ((struct arb_pshader_private *)ps->baseShader.backend_data)->clamp_consts)
4331         {
4332             priv->last_ps_const_clamped = ((struct arb_pshader_private *)ps->baseShader.backend_data)->clamp_consts;
4333             This->highest_dirty_ps_const = max(This->highest_dirty_ps_const, 8);
4334             for(i = 0; i < 8; i++)
4335             {
4336                 context->pshader_const_dirty[i] = 1;
4337             }
4338             /* Also takes care of loading local constants */
4339             shader_arb_load_constants(context, TRUE, FALSE);
4340         }
4341         else
4342         {
4343             shader_arb_ps_local_constants(This);
4344         }
4345
4346         /* Force constant reloading for the NP2 fixup (see comment in shader_glsl_select for more info) */
4347         if (compiled->np2fixup_info.super.active)
4348             shader_arb_load_np2fixup_constants((IWineD3DDevice *)This, usePS, useVS);
4349     }
4350     else if (gl_info->supported[ARB_FRAGMENT_PROGRAM] && !priv->use_arbfp_fixed_func)
4351     {
4352         /* Disable only if we're not using arbfp fixed function fragment processing. If this is used,
4353         * keep GL_FRAGMENT_PROGRAM_ARB enabled, and the fixed function pipeline will bind the fixed function
4354         * replacement shader
4355         */
4356         glDisable(GL_FRAGMENT_PROGRAM_ARB);
4357         checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
4358         priv->current_fprogram_id = 0;
4359     }
4360
4361     if (useVS) {
4362         struct arb_vs_compile_args compile_args;
4363         struct arb_vs_compiled_shader *compiled;
4364         IWineD3DVertexShaderImpl *vs = (IWineD3DVertexShaderImpl *) This->stateBlock->vertexShader;
4365
4366         TRACE("Using vertex shader %p\n", This->stateBlock->vertexShader);
4367         find_arb_vs_compile_args(vs, This->stateBlock, &compile_args);
4368         compiled = find_arb_vshader(vs, &compile_args);
4369         priv->current_vprogram_id = compiled->prgId;
4370         priv->compiled_vprog = compiled;
4371
4372         /* Bind the vertex program */
4373         GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->current_vprogram_id));
4374         checkGLcall("glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->current_vprogram_id);");
4375
4376         /* Enable OpenGL vertex programs */
4377         glEnable(GL_VERTEX_PROGRAM_ARB);
4378         checkGLcall("glEnable(GL_VERTEX_PROGRAM_ARB);");
4379         TRACE("(%p) : Bound vertex program %u and enabled GL_VERTEX_PROGRAM_ARB\n", This, priv->current_vprogram_id);
4380         shader_arb_vs_local_constants(This);
4381
4382         if(priv->last_vs_color_unclamp != compiled->need_color_unclamp) {
4383             priv->last_vs_color_unclamp = compiled->need_color_unclamp;
4384
4385             if (gl_info->supported[ARB_COLOR_BUFFER_FLOAT])
4386             {
4387                 GL_EXTCALL(glClampColorARB(GL_CLAMP_VERTEX_COLOR_ARB, !compiled->need_color_unclamp));
4388                 checkGLcall("glClampColorARB");
4389             } else {
4390                 FIXME("vertex color clamp needs to be changed, but extension not supported.\n");
4391             }
4392         }
4393     }
4394     else if (gl_info->supported[ARB_VERTEX_PROGRAM])
4395     {
4396         priv->current_vprogram_id = 0;
4397         glDisable(GL_VERTEX_PROGRAM_ARB);
4398         checkGLcall("glDisable(GL_VERTEX_PROGRAM_ARB)");
4399     }
4400 }
4401
4402 /* GL locking is done by the caller */
4403 static void shader_arb_select_depth_blt(IWineD3DDevice *iface, enum tex_types tex_type) {
4404     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
4405     struct shader_arb_priv *priv = This->shader_priv;
4406     GLuint *blt_fprogram = &priv->depth_blt_fprogram_id[tex_type];
4407     const struct wined3d_gl_info *gl_info = &This->adapter->gl_info;
4408
4409     if (!priv->depth_blt_vprogram_id) priv->depth_blt_vprogram_id = create_arb_blt_vertex_program(gl_info);
4410     GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->depth_blt_vprogram_id));
4411     glEnable(GL_VERTEX_PROGRAM_ARB);
4412
4413     if (!*blt_fprogram) *blt_fprogram = create_arb_blt_fragment_program(gl_info, tex_type);
4414     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, *blt_fprogram));
4415     glEnable(GL_FRAGMENT_PROGRAM_ARB);
4416 }
4417
4418 /* GL locking is done by the caller */
4419 static void shader_arb_deselect_depth_blt(IWineD3DDevice *iface) {
4420     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
4421     struct shader_arb_priv *priv = This->shader_priv;
4422     const struct wined3d_gl_info *gl_info = &This->adapter->gl_info;
4423
4424     if (priv->current_vprogram_id) {
4425         GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->current_vprogram_id));
4426         checkGLcall("glBindProgramARB(GL_VERTEX_PROGRAM_ARB, vertexShader->prgId);");
4427
4428         TRACE("(%p) : Bound vertex program %u and enabled GL_VERTEX_PROGRAM_ARB\n", This, priv->current_vprogram_id);
4429     } else {
4430         glDisable(GL_VERTEX_PROGRAM_ARB);
4431         checkGLcall("glDisable(GL_VERTEX_PROGRAM_ARB)");
4432     }
4433
4434     if (priv->current_fprogram_id) {
4435         GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, priv->current_fprogram_id));
4436         checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, pixelShader->prgId);");
4437
4438         TRACE("(%p) : Bound fragment program %u and enabled GL_FRAGMENT_PROGRAM_ARB\n", This, priv->current_fprogram_id);
4439     } else if(!priv->use_arbfp_fixed_func) {
4440         glDisable(GL_FRAGMENT_PROGRAM_ARB);
4441         checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
4442     }
4443 }
4444
4445 static void shader_arb_destroy(IWineD3DBaseShader *iface) {
4446     IWineD3DBaseShaderImpl *baseShader = (IWineD3DBaseShaderImpl *) iface;
4447     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *)baseShader->baseShader.device;
4448     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4449
4450     if (shader_is_pshader_version(baseShader->baseShader.reg_maps.shader_version.type))
4451     {
4452         struct arb_pshader_private *shader_data = baseShader->baseShader.backend_data;
4453         UINT i;
4454
4455         if(!shader_data) return; /* This can happen if a shader was never compiled */
4456
4457         if (shader_data->num_gl_shaders)
4458         {
4459             struct wined3d_context *context = context_acquire(device, NULL, CTXUSAGE_RESOURCELOAD);
4460
4461             ENTER_GL();
4462             for (i = 0; i < shader_data->num_gl_shaders; ++i)
4463             {
4464                 GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId));
4465                 checkGLcall("GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId))");
4466             }
4467             LEAVE_GL();
4468
4469             context_release(context);
4470         }
4471
4472         HeapFree(GetProcessHeap(), 0, shader_data->gl_shaders);
4473         HeapFree(GetProcessHeap(), 0, shader_data);
4474         baseShader->baseShader.backend_data = NULL;
4475     }
4476     else
4477     {
4478         struct arb_vshader_private *shader_data = baseShader->baseShader.backend_data;
4479         UINT i;
4480
4481         if(!shader_data) return; /* This can happen if a shader was never compiled */
4482
4483         if (shader_data->num_gl_shaders)
4484         {
4485             struct wined3d_context *context = context_acquire(device, NULL, CTXUSAGE_RESOURCELOAD);
4486
4487             ENTER_GL();
4488             for (i = 0; i < shader_data->num_gl_shaders; ++i)
4489             {
4490                 GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId));
4491                 checkGLcall("GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId))");
4492             }
4493             LEAVE_GL();
4494
4495             context_release(context);
4496         }
4497
4498         HeapFree(GetProcessHeap(), 0, shader_data->gl_shaders);
4499         HeapFree(GetProcessHeap(), 0, shader_data);
4500         baseShader->baseShader.backend_data = NULL;
4501     }
4502 }
4503
4504 static int sig_tree_compare(const void *key, const struct wine_rb_entry *entry)
4505 {
4506     struct ps_signature *e = WINE_RB_ENTRY_VALUE(entry, struct ps_signature, entry);
4507     return compare_sig(key, e->sig);
4508 }
4509
4510 static const struct wine_rb_functions sig_tree_functions =
4511 {
4512     wined3d_rb_alloc,
4513     wined3d_rb_realloc,
4514     wined3d_rb_free,
4515     sig_tree_compare
4516 };
4517
4518 static HRESULT shader_arb_alloc(IWineD3DDevice *iface) {
4519     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
4520     struct shader_arb_priv *priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*priv));
4521     if(wine_rb_init(&priv->signature_tree, &sig_tree_functions) == -1)
4522     {
4523         ERR("RB tree init failed\n");
4524         HeapFree(GetProcessHeap(), 0, priv);
4525         return E_OUTOFMEMORY;
4526     }
4527     This->shader_priv = priv;
4528     return WINED3D_OK;
4529 }
4530
4531 static void release_signature(struct wine_rb_entry *entry, void *context)
4532 {
4533     struct ps_signature *sig = WINE_RB_ENTRY_VALUE(entry, struct ps_signature, entry);
4534     int i;
4535     for(i = 0; i < MAX_REG_INPUT; i++)
4536     {
4537         HeapFree(GetProcessHeap(), 0, (char *) sig->sig[i].semantic_name);
4538     }
4539     HeapFree(GetProcessHeap(), 0, sig->sig);
4540     HeapFree(GetProcessHeap(), 0, sig);
4541 }
4542
4543 /* Context activation is done by the caller. */
4544 static void shader_arb_free(IWineD3DDevice *iface) {
4545     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
4546     const struct wined3d_gl_info *gl_info = &This->adapter->gl_info;
4547     struct shader_arb_priv *priv = This->shader_priv;
4548     int i;
4549
4550     ENTER_GL();
4551     if(priv->depth_blt_vprogram_id) {
4552         GL_EXTCALL(glDeleteProgramsARB(1, &priv->depth_blt_vprogram_id));
4553     }
4554     for (i = 0; i < tex_type_count; ++i) {
4555         if (priv->depth_blt_fprogram_id[i]) {
4556             GL_EXTCALL(glDeleteProgramsARB(1, &priv->depth_blt_fprogram_id[i]));
4557         }
4558     }
4559     LEAVE_GL();
4560
4561     wine_rb_destroy(&priv->signature_tree, release_signature, NULL);
4562     HeapFree(GetProcessHeap(), 0, This->shader_priv);
4563 }
4564
4565 static BOOL shader_arb_dirty_const(IWineD3DDevice *iface) {
4566     return TRUE;
4567 }
4568
4569 static void shader_arb_get_caps(WINED3DDEVTYPE devtype, const struct wined3d_gl_info *gl_info,
4570         struct shader_caps *pCaps)
4571 {
4572     DWORD vs_consts = min(gl_info->limits.arb_vs_float_constants, gl_info->limits.arb_vs_native_constants);
4573     DWORD ps_consts = min(gl_info->limits.arb_ps_float_constants, gl_info->limits.arb_ps_native_constants);
4574
4575     /* We don't have an ARB fixed function pipeline yet, so let the none backend set its caps,
4576      * then overwrite the shader specific ones
4577      */
4578     none_shader_backend.shader_get_caps(devtype, gl_info, pCaps);
4579
4580     if (gl_info->supported[ARB_VERTEX_PROGRAM])
4581     {
4582         if (gl_info->supported[NV_VERTEX_PROGRAM3])
4583         {
4584             pCaps->VertexShaderVersion = WINED3DVS_VERSION(3,0);
4585             TRACE_(d3d_caps)("Hardware vertex shader version 3.0 enabled (NV_VERTEX_PROGRAM3)\n");
4586         }
4587         else if (vs_consts >= 256)
4588         {
4589             /* Shader Model 2.0 requires at least 256 vertex shader constants */
4590             pCaps->VertexShaderVersion = WINED3DVS_VERSION(2,0);
4591             TRACE_(d3d_caps)("Hardware vertex shader version 2.0 enabled (ARB_PROGRAM)\n");
4592         }
4593         else
4594         {
4595             pCaps->VertexShaderVersion = WINED3DVS_VERSION(1,1);
4596             TRACE_(d3d_caps)("Hardware vertex shader version 1.1 enabled (ARB_PROGRAM)\n");
4597         }
4598         pCaps->MaxVertexShaderConst = vs_consts;
4599     }
4600
4601     if (gl_info->supported[ARB_FRAGMENT_PROGRAM])
4602     {
4603         if (gl_info->supported[NV_FRAGMENT_PROGRAM2])
4604         {
4605             pCaps->PixelShaderVersion    = WINED3DPS_VERSION(3,0);
4606             TRACE_(d3d_caps)("Hardware pixel shader version 3.0 enabled (NV_FRAGMENT_PROGRAM2)\n");
4607         }
4608         else if (ps_consts >= 32)
4609         {
4610             /* Shader Model 2.0 requires at least 32 pixel shader constants */
4611             pCaps->PixelShaderVersion    = WINED3DPS_VERSION(2,0);
4612             TRACE_(d3d_caps)("Hardware pixel shader version 2.0 enabled (ARB_PROGRAM)\n");
4613         }
4614         else
4615         {
4616             pCaps->PixelShaderVersion    = WINED3DPS_VERSION(1,4);
4617             TRACE_(d3d_caps)("Hardware pixel shader version 1.4 enabled (ARB_PROGRAM)\n");
4618         }
4619         pCaps->PixelShader1xMaxValue = 8.0f;
4620         pCaps->MaxPixelShaderConst = ps_consts;
4621     }
4622
4623     pCaps->VSClipping = use_nv_clip(gl_info);
4624 }
4625
4626 static BOOL shader_arb_color_fixup_supported(struct color_fixup_desc fixup)
4627 {
4628     if (TRACE_ON(d3d_shader) && TRACE_ON(d3d))
4629     {
4630         TRACE("Checking support for color_fixup:\n");
4631         dump_color_fixup_desc(fixup);
4632     }
4633
4634     /* We support everything except YUV conversions. */
4635     if (!is_yuv_fixup(fixup))
4636     {
4637         TRACE("[OK]\n");
4638         return TRUE;
4639     }
4640
4641     TRACE("[FAILED]\n");
4642     return FALSE;
4643 }
4644
4645 static void shader_arb_add_instruction_modifiers(const struct wined3d_shader_instruction *ins) {
4646     DWORD shift;
4647     char write_mask[20], regstr[50];
4648     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
4649     BOOL is_color = FALSE;
4650     const struct wined3d_shader_dst_param *dst;
4651
4652     if (!ins->dst_count) return;
4653
4654     dst = &ins->dst[0];
4655     shift = dst->shift;
4656     if(shift == 0) return; /* Saturate alone is handled by the instructions */
4657
4658     shader_arb_get_write_mask(ins, dst, write_mask);
4659     shader_arb_get_register_name(ins, &dst->reg, regstr, &is_color);
4660
4661     /* Generate a line that does the output modifier computation
4662      * FIXME: _SAT vs shift? _SAT alone is already handled in the instructions, if this
4663      * maps problems in e.g. _d4_sat modify shader_arb_get_modifier
4664      */
4665     shader_addline(buffer, "MUL%s %s%s, %s, %s;\n", shader_arb_get_modifier(ins),
4666                    regstr, write_mask, regstr, shift_tab[shift]);
4667 }
4668
4669 static const SHADER_HANDLER shader_arb_instruction_handler_table[WINED3DSIH_TABLE_SIZE] =
4670 {
4671     /* WINED3DSIH_ABS           */ shader_hw_map2gl,
4672     /* WINED3DSIH_ADD           */ shader_hw_map2gl,
4673     /* WINED3DSIH_BEM           */ pshader_hw_bem,
4674     /* WINED3DSIH_BREAK         */ shader_hw_break,
4675     /* WINED3DSIH_BREAKC        */ shader_hw_breakc,
4676     /* WINED3DSIH_BREAKP        */ NULL,
4677     /* WINED3DSIH_CALL          */ shader_hw_call,
4678     /* WINED3DSIH_CALLNZ        */ NULL,
4679     /* WINED3DSIH_CMP           */ pshader_hw_cmp,
4680     /* WINED3DSIH_CND           */ pshader_hw_cnd,
4681     /* WINED3DSIH_CRS           */ shader_hw_map2gl,
4682     /* WINED3DSIH_CUT           */ NULL,
4683     /* WINED3DSIH_DCL           */ NULL,
4684     /* WINED3DSIH_DEF           */ NULL,
4685     /* WINED3DSIH_DEFB          */ NULL,
4686     /* WINED3DSIH_DEFI          */ NULL,
4687     /* WINED3DSIH_DP2ADD        */ pshader_hw_dp2add,
4688     /* WINED3DSIH_DP3           */ shader_hw_map2gl,
4689     /* WINED3DSIH_DP4           */ shader_hw_map2gl,
4690     /* WINED3DSIH_DST           */ shader_hw_map2gl,
4691     /* WINED3DSIH_DSX           */ shader_hw_map2gl,
4692     /* WINED3DSIH_DSY           */ shader_hw_dsy,
4693     /* WINED3DSIH_ELSE          */ shader_hw_else,
4694     /* WINED3DSIH_EMIT          */ NULL,
4695     /* WINED3DSIH_ENDIF         */ shader_hw_endif,
4696     /* WINED3DSIH_ENDLOOP       */ shader_hw_endloop,
4697     /* WINED3DSIH_ENDREP        */ shader_hw_endrep,
4698     /* WINED3DSIH_EXP           */ shader_hw_scalar_op,
4699     /* WINED3DSIH_EXPP          */ shader_hw_scalar_op,
4700     /* WINED3DSIH_FRC           */ shader_hw_map2gl,
4701     /* WINED3DSIH_IADD          */ NULL,
4702     /* WINED3DSIH_IF            */ NULL /* Hardcoded into the shader */,
4703     /* WINED3DSIH_IFC           */ shader_hw_ifc,
4704     /* WINED3DSIH_IGE           */ NULL,
4705     /* WINED3DSIH_LABEL         */ shader_hw_label,
4706     /* WINED3DSIH_LIT           */ shader_hw_map2gl,
4707     /* WINED3DSIH_LOG           */ shader_hw_log_pow,
4708     /* WINED3DSIH_LOGP          */ shader_hw_log_pow,
4709     /* WINED3DSIH_LOOP          */ shader_hw_loop,
4710     /* WINED3DSIH_LRP           */ shader_hw_lrp,
4711     /* WINED3DSIH_LT            */ NULL,
4712     /* WINED3DSIH_M3x2          */ shader_hw_mnxn,
4713     /* WINED3DSIH_M3x3          */ shader_hw_mnxn,
4714     /* WINED3DSIH_M3x4          */ shader_hw_mnxn,
4715     /* WINED3DSIH_M4x3          */ shader_hw_mnxn,
4716     /* WINED3DSIH_M4x4          */ shader_hw_mnxn,
4717     /* WINED3DSIH_MAD           */ shader_hw_map2gl,
4718     /* WINED3DSIH_MAX           */ shader_hw_map2gl,
4719     /* WINED3DSIH_MIN           */ shader_hw_map2gl,
4720     /* WINED3DSIH_MOV           */ shader_hw_mov,
4721     /* WINED3DSIH_MOVA          */ shader_hw_mov,
4722     /* WINED3DSIH_MUL           */ shader_hw_map2gl,
4723     /* WINED3DSIH_NOP           */ shader_hw_nop,
4724     /* WINED3DSIH_NRM           */ shader_hw_nrm,
4725     /* WINED3DSIH_PHASE         */ NULL,
4726     /* WINED3DSIH_POW           */ shader_hw_log_pow,
4727     /* WINED3DSIH_RCP           */ shader_hw_scalar_op,
4728     /* WINED3DSIH_REP           */ shader_hw_rep,
4729     /* WINED3DSIH_RET           */ shader_hw_ret,
4730     /* WINED3DSIH_RSQ           */ shader_hw_scalar_op,
4731     /* WINED3DSIH_SETP          */ NULL,
4732     /* WINED3DSIH_SGE           */ shader_hw_map2gl,
4733     /* WINED3DSIH_SGN           */ shader_hw_sgn,
4734     /* WINED3DSIH_SINCOS        */ shader_hw_sincos,
4735     /* WINED3DSIH_SLT           */ shader_hw_map2gl,
4736     /* WINED3DSIH_SUB           */ shader_hw_map2gl,
4737     /* WINED3DSIH_TEX           */ pshader_hw_tex,
4738     /* WINED3DSIH_TEXBEM        */ pshader_hw_texbem,
4739     /* WINED3DSIH_TEXBEML       */ pshader_hw_texbem,
4740     /* WINED3DSIH_TEXCOORD      */ pshader_hw_texcoord,
4741     /* WINED3DSIH_TEXDEPTH      */ pshader_hw_texdepth,
4742     /* WINED3DSIH_TEXDP3        */ pshader_hw_texdp3,
4743     /* WINED3DSIH_TEXDP3TEX     */ pshader_hw_texdp3tex,
4744     /* WINED3DSIH_TEXKILL       */ pshader_hw_texkill,
4745     /* WINED3DSIH_TEXLDD        */ shader_hw_texldd,
4746     /* WINED3DSIH_TEXLDL        */ shader_hw_texldl,
4747     /* WINED3DSIH_TEXM3x2DEPTH  */ pshader_hw_texm3x2depth,
4748     /* WINED3DSIH_TEXM3x2PAD    */ pshader_hw_texm3x2pad,
4749     /* WINED3DSIH_TEXM3x2TEX    */ pshader_hw_texm3x2tex,
4750     /* WINED3DSIH_TEXM3x3       */ pshader_hw_texm3x3,
4751     /* WINED3DSIH_TEXM3x3DIFF   */ NULL,
4752     /* WINED3DSIH_TEXM3x3PAD    */ pshader_hw_texm3x3pad,
4753     /* WINED3DSIH_TEXM3x3SPEC   */ pshader_hw_texm3x3spec,
4754     /* WINED3DSIH_TEXM3x3TEX    */ pshader_hw_texm3x3tex,
4755     /* WINED3DSIH_TEXM3x3VSPEC  */ pshader_hw_texm3x3vspec,
4756     /* WINED3DSIH_TEXREG2AR     */ pshader_hw_texreg2ar,
4757     /* WINED3DSIH_TEXREG2GB     */ pshader_hw_texreg2gb,
4758     /* WINED3DSIH_TEXREG2RGB    */ pshader_hw_texreg2rgb,
4759 };
4760
4761 static inline BOOL get_bool_const(const struct wined3d_shader_instruction *ins, IWineD3DBaseShaderImpl *This, DWORD idx)
4762 {
4763     BOOL vshader = shader_is_vshader_version(This->baseShader.reg_maps.shader_version.type);
4764     WORD bools = 0;
4765     WORD flag = (1 << idx);
4766     const local_constant *constant;
4767     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
4768
4769     if(This->baseShader.reg_maps.local_bool_consts & flag)
4770     {
4771         /* What good is a if(bool) with a hardcoded local constant? I don't know, but handle it */
4772         LIST_FOR_EACH_ENTRY(constant, &This->baseShader.constantsB, local_constant, entry)
4773         {
4774             if (constant->idx == idx)
4775             {
4776                 return constant->value[0];
4777             }
4778         }
4779         ERR("Local constant not found\n");
4780         return FALSE;
4781     }
4782     else
4783     {
4784         if(vshader) bools = priv->cur_vs_args->clip.boolclip.bools;
4785         else bools = priv->cur_ps_args->bools;
4786         return bools & flag;
4787     }
4788 }
4789
4790 static void get_loop_control_const(const struct wined3d_shader_instruction *ins,
4791         IWineD3DBaseShaderImpl *This, UINT idx, struct wined3d_shader_loop_control *loop_control)
4792 {
4793     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
4794
4795     /* Integer constants can either be a local constant, or they can be stored in the shader
4796      * type specific compile args. */
4797     if (This->baseShader.reg_maps.local_int_consts & (1 << idx))
4798     {
4799         const local_constant *constant;
4800
4801         LIST_FOR_EACH_ENTRY(constant, &This->baseShader.constantsI, local_constant, entry)
4802         {
4803             if (constant->idx == idx)
4804             {
4805                 loop_control->count = constant->value[0];
4806                 loop_control->start = constant->value[1];
4807                 /* Step is signed. */
4808                 loop_control->step = (int)constant->value[2];
4809                 return;
4810             }
4811         }
4812         /* If this happens the flag was set incorrectly */
4813         ERR("Local constant not found\n");
4814         loop_control->count = 0;
4815         loop_control->start = 0;
4816         loop_control->step = 0;
4817         return;
4818     }
4819
4820     switch (This->baseShader.reg_maps.shader_version.type)
4821     {
4822         case WINED3D_SHADER_TYPE_VERTEX:
4823             /* Count and aL start value are unsigned */
4824             loop_control->count = priv->cur_vs_args->loop_ctrl[idx][0];
4825             loop_control->start = priv->cur_vs_args->loop_ctrl[idx][1];
4826             /* Step is signed. */
4827             loop_control->step = ((char)priv->cur_vs_args->loop_ctrl[idx][2]);
4828             break;
4829
4830         case WINED3D_SHADER_TYPE_PIXEL:
4831             loop_control->count = priv->cur_ps_args->loop_ctrl[idx][0];
4832             loop_control->start = priv->cur_ps_args->loop_ctrl[idx][1];
4833             loop_control->step = ((char)priv->cur_ps_args->loop_ctrl[idx][2]);
4834             break;
4835
4836         default:
4837             FIXME("Unhandled shader type %#x.\n", This->baseShader.reg_maps.shader_version.type);
4838             break;
4839     }
4840 }
4841
4842 static void record_instruction(struct list *list, const struct wined3d_shader_instruction *ins)
4843 {
4844     unsigned int i;
4845     struct wined3d_shader_dst_param *dst_param = NULL;
4846     struct wined3d_shader_src_param *src_param = NULL, *rel_addr = NULL;
4847     struct recorded_instruction *rec = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*rec));
4848     if(!rec)
4849     {
4850         ERR("Out of memory\n");
4851         return;
4852     }
4853
4854     rec->ins = *ins;
4855     dst_param = HeapAlloc(GetProcessHeap(), 0, sizeof(*dst_param));
4856     if(!dst_param) goto free;
4857     *dst_param = *ins->dst;
4858     if(ins->dst->reg.rel_addr)
4859     {
4860         rel_addr = HeapAlloc(GetProcessHeap(), 0, sizeof(*dst_param->reg.rel_addr));
4861         if(!rel_addr) goto free;
4862         *rel_addr = *ins->dst->reg.rel_addr;
4863         dst_param->reg.rel_addr = rel_addr;
4864     }
4865     rec->ins.dst = dst_param;
4866
4867     src_param = HeapAlloc(GetProcessHeap(), 0, sizeof(*src_param) * ins->src_count);
4868     if(!src_param) goto free;
4869     for(i = 0; i < ins->src_count; i++)
4870     {
4871         src_param[i] = ins->src[i];
4872         if(ins->src[i].reg.rel_addr)
4873         {
4874             rel_addr = HeapAlloc(GetProcessHeap(), 0, sizeof(*rel_addr));
4875             if(!rel_addr) goto free;
4876             *rel_addr = *ins->src[i].reg.rel_addr;
4877             src_param[i].reg.rel_addr = rel_addr;
4878         }
4879     }
4880     rec->ins.src = src_param;
4881     list_add_tail(list, &rec->entry);
4882     return;
4883
4884 free:
4885     ERR("Out of memory\n");
4886     if(dst_param)
4887     {
4888         HeapFree(GetProcessHeap(), 0, (void *) dst_param->reg.rel_addr);
4889         HeapFree(GetProcessHeap(), 0, dst_param);
4890     }
4891     if(src_param)
4892     {
4893         for(i = 0; i < ins->src_count; i++)
4894         {
4895             HeapFree(GetProcessHeap(), 0, (void *) src_param[i].reg.rel_addr);
4896         }
4897         HeapFree(GetProcessHeap(), 0, src_param);
4898     }
4899     HeapFree(GetProcessHeap(), 0, rec);
4900 }
4901
4902 static void free_recorded_instruction(struct list *list)
4903 {
4904     struct recorded_instruction *rec_ins, *entry2;
4905     unsigned int i;
4906
4907     LIST_FOR_EACH_ENTRY_SAFE(rec_ins, entry2, list, struct recorded_instruction, entry)
4908     {
4909         list_remove(&rec_ins->entry);
4910         if(rec_ins->ins.dst)
4911         {
4912             HeapFree(GetProcessHeap(), 0, (void *) rec_ins->ins.dst->reg.rel_addr);
4913             HeapFree(GetProcessHeap(), 0, (void *) rec_ins->ins.dst);
4914         }
4915         if(rec_ins->ins.src)
4916         {
4917             for(i = 0; i < rec_ins->ins.src_count; i++)
4918             {
4919                 HeapFree(GetProcessHeap(), 0, (void *) rec_ins->ins.src[i].reg.rel_addr);
4920             }
4921             HeapFree(GetProcessHeap(), 0, (void *) rec_ins->ins.src);
4922         }
4923         HeapFree(GetProcessHeap(), 0, rec_ins);
4924     }
4925 }
4926
4927 static void shader_arb_handle_instruction(const struct wined3d_shader_instruction *ins) {
4928     SHADER_HANDLER hw_fct;
4929     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
4930     IWineD3DBaseShaderImpl *This = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
4931     struct control_frame *control_frame;
4932     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
4933     BOOL bool_const;
4934
4935     if(ins->handler_idx == WINED3DSIH_LOOP || ins->handler_idx == WINED3DSIH_REP)
4936     {
4937         control_frame = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*control_frame));
4938         list_add_head(&priv->control_frames, &control_frame->entry);
4939
4940         if(ins->handler_idx == WINED3DSIH_LOOP) control_frame->type = LOOP;
4941         if(ins->handler_idx == WINED3DSIH_REP) control_frame->type = REP;
4942
4943         if(priv->target_version >= NV2)
4944         {
4945             control_frame->no.loop = priv->num_loops++;
4946             priv->loop_depth++;
4947         }
4948         else
4949         {
4950             /* Don't bother recording when we're in a not used if branch */
4951             if(priv->muted)
4952             {
4953                 return;
4954             }
4955
4956             if(!priv->recording)
4957             {
4958                 list_init(&priv->record);
4959                 priv->recording = TRUE;
4960                 control_frame->outer_loop = TRUE;
4961                 get_loop_control_const(ins, This, ins->src[0].reg.idx, &control_frame->loop_control);
4962                 return; /* Instruction is handled */
4963             }
4964             /* Record this loop in the outer loop's recording */
4965         }
4966     }
4967     else if(ins->handler_idx == WINED3DSIH_ENDLOOP || ins->handler_idx == WINED3DSIH_ENDREP)
4968     {
4969         if(priv->target_version >= NV2)
4970         {
4971             /* Nothing to do. The control frame is popped after the HW instr handler */
4972         }
4973         else
4974         {
4975             struct list *e = list_head(&priv->control_frames);
4976             control_frame = LIST_ENTRY(e, struct control_frame, entry);
4977             list_remove(&control_frame->entry);
4978
4979             if(control_frame->outer_loop)
4980             {
4981                 int iteration, aL = 0;
4982                 struct list copy;
4983
4984                 /* Turn off recording before playback */
4985                 priv->recording = FALSE;
4986
4987                 /* Move the recorded instructions to a separate list and get them out of the private data
4988                  * structure. If there are nested loops, the shader_arb_handle_instruction below will
4989                  * be recorded again, thus priv->record might be overwritten
4990                  */
4991                 list_init(&copy);
4992                 list_move_tail(&copy, &priv->record);
4993                 list_init(&priv->record);
4994
4995                 if(ins->handler_idx == WINED3DSIH_ENDLOOP)
4996                 {
4997                     shader_addline(buffer, "#unrolling loop: %u iterations, aL=%u, inc %d\n",
4998                                    control_frame->loop_control.count, control_frame->loop_control.start,
4999                                    control_frame->loop_control.step);
5000                     aL = control_frame->loop_control.start;
5001                 }
5002                 else
5003                 {
5004                     shader_addline(buffer, "#unrolling rep: %u iterations\n", control_frame->loop_control.count);
5005                 }
5006
5007                 for (iteration = 0; iteration < control_frame->loop_control.count; ++iteration)
5008                 {
5009                     struct recorded_instruction *rec_ins;
5010                     if(ins->handler_idx == WINED3DSIH_ENDLOOP)
5011                     {
5012                         priv->aL = aL;
5013                         shader_addline(buffer, "#Iteration %d, aL=%d\n", iteration, aL);
5014                     }
5015                     else
5016                     {
5017                         shader_addline(buffer, "#Iteration %d\n", iteration);
5018                     }
5019
5020                     LIST_FOR_EACH_ENTRY(rec_ins, &copy, struct recorded_instruction, entry)
5021                     {
5022                         shader_arb_handle_instruction(&rec_ins->ins);
5023                     }
5024
5025                     if(ins->handler_idx == WINED3DSIH_ENDLOOP)
5026                     {
5027                         aL += control_frame->loop_control.step;
5028                     }
5029                 }
5030                 shader_addline(buffer, "#end loop/rep\n");
5031
5032                 free_recorded_instruction(&copy);
5033                 HeapFree(GetProcessHeap(), 0, control_frame);
5034                 return; /* Instruction is handled */
5035             }
5036             else
5037             {
5038                 /* This is a nested loop. Proceed to the normal recording function */
5039                 HeapFree(GetProcessHeap(), 0, control_frame);
5040             }
5041         }
5042     }
5043
5044     if(priv->recording)
5045     {
5046         record_instruction(&priv->record, ins);
5047         return;
5048     }
5049
5050     /* boolean if */
5051     if(ins->handler_idx == WINED3DSIH_IF)
5052     {
5053         control_frame = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*control_frame));
5054         list_add_head(&priv->control_frames, &control_frame->entry);
5055         control_frame->type = IF;
5056
5057         bool_const = get_bool_const(ins, This, ins->src[0].reg.idx);
5058         if(ins->src[0].modifiers == WINED3DSPSM_NOT) bool_const = !bool_const;
5059         if(!priv->muted && bool_const == FALSE)
5060         {
5061             shader_addline(buffer, "#if(FALSE){\n");
5062             priv->muted = TRUE;
5063             control_frame->muting = TRUE;
5064         }
5065         else shader_addline(buffer, "#if(TRUE) {\n");
5066
5067         return; /* Instruction is handled */
5068     }
5069     else if(ins->handler_idx == WINED3DSIH_IFC)
5070     {
5071         /* IF(bool) and if_cond(a, b) use the same ELSE and ENDIF tokens */
5072         control_frame = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*control_frame));
5073         control_frame->type = IFC;
5074         control_frame->no.ifc = priv->num_ifcs++;
5075         list_add_head(&priv->control_frames, &control_frame->entry);
5076     }
5077     else if(ins->handler_idx == WINED3DSIH_ELSE)
5078     {
5079         struct list *e = list_head(&priv->control_frames);
5080         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5081
5082         if(control_frame->type == IF)
5083         {
5084             shader_addline(buffer, "#} else {\n");
5085             if(!priv->muted && !control_frame->muting)
5086             {
5087                 priv->muted = TRUE;
5088                 control_frame->muting = TRUE;
5089             }
5090             else if(control_frame->muting) priv->muted = FALSE;
5091             return; /* Instruction is handled. */
5092         }
5093         /* In case of an ifc, generate a HW shader instruction */
5094     }
5095     else if(ins->handler_idx == WINED3DSIH_ENDIF)
5096     {
5097         struct list *e = list_head(&priv->control_frames);
5098         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5099
5100         if(control_frame->type == IF)
5101         {
5102             shader_addline(buffer, "#} endif\n");
5103             if(control_frame->muting) priv->muted = FALSE;
5104             list_remove(&control_frame->entry);
5105             HeapFree(GetProcessHeap(), 0, control_frame);
5106             return; /* Instruction is handled */
5107         }
5108     }
5109
5110     if(priv->muted) return;
5111
5112     /* Select handler */
5113     hw_fct = shader_arb_instruction_handler_table[ins->handler_idx];
5114
5115     /* Unhandled opcode */
5116     if (!hw_fct)
5117     {
5118         FIXME("Backend can't handle opcode %#x\n", ins->handler_idx);
5119         return;
5120     }
5121     hw_fct(ins);
5122
5123     if(ins->handler_idx == WINED3DSIH_ENDLOOP || ins->handler_idx == WINED3DSIH_ENDREP)
5124     {
5125         struct list *e = list_head(&priv->control_frames);
5126         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5127         list_remove(&control_frame->entry);
5128         HeapFree(GetProcessHeap(), 0, control_frame);
5129         priv->loop_depth--;
5130     }
5131     else if(ins->handler_idx == WINED3DSIH_ENDIF)
5132     {
5133         /* Non-ifc ENDIFs don't reach that place because of the return in the if block above */
5134         struct list *e = list_head(&priv->control_frames);
5135         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5136         list_remove(&control_frame->entry);
5137         HeapFree(GetProcessHeap(), 0, control_frame);
5138     }
5139
5140
5141     shader_arb_add_instruction_modifiers(ins);
5142 }
5143
5144 const shader_backend_t arb_program_shader_backend = {
5145     shader_arb_handle_instruction,
5146     shader_arb_select,
5147     shader_arb_select_depth_blt,
5148     shader_arb_deselect_depth_blt,
5149     shader_arb_update_float_vertex_constants,
5150     shader_arb_update_float_pixel_constants,
5151     shader_arb_load_constants,
5152     shader_arb_load_np2fixup_constants,
5153     shader_arb_destroy,
5154     shader_arb_alloc,
5155     shader_arb_free,
5156     shader_arb_dirty_const,
5157     shader_arb_get_caps,
5158     shader_arb_color_fixup_supported,
5159 };
5160
5161 /* ARB_fragment_program fixed function pipeline replacement definitions */
5162 #define ARB_FFP_CONST_TFACTOR           0
5163 #define ARB_FFP_CONST_SPECULAR_ENABLE   ((ARB_FFP_CONST_TFACTOR) + 1)
5164 #define ARB_FFP_CONST_CONSTANT(i)       ((ARB_FFP_CONST_SPECULAR_ENABLE) + 1 + i)
5165 #define ARB_FFP_CONST_BUMPMAT(i)        ((ARB_FFP_CONST_CONSTANT(7)) + 1 + i)
5166 #define ARB_FFP_CONST_LUMINANCE(i)      ((ARB_FFP_CONST_BUMPMAT(7)) + 1 + i)
5167
5168 struct arbfp_ffp_desc
5169 {
5170     struct ffp_frag_desc parent;
5171     GLuint shader;
5172     unsigned int num_textures_used;
5173 };
5174
5175 /* Context activation is done by the caller. */
5176 static void arbfp_enable(IWineD3DDevice *iface, BOOL enable) {
5177     ENTER_GL();
5178     if(enable) {
5179         glEnable(GL_FRAGMENT_PROGRAM_ARB);
5180         checkGLcall("glEnable(GL_FRAGMENT_PROGRAM_ARB)");
5181     } else {
5182         glDisable(GL_FRAGMENT_PROGRAM_ARB);
5183         checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
5184     }
5185     LEAVE_GL();
5186 }
5187
5188 static HRESULT arbfp_alloc(IWineD3DDevice *iface) {
5189     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *) iface;
5190     struct shader_arb_priv *priv;
5191     /* Share private data between the shader backend and the pipeline replacement, if both
5192      * are the arb implementation. This is needed to figure out whether ARBfp should be disabled
5193      * if no pixel shader is bound or not
5194      */
5195     if(This->shader_backend == &arb_program_shader_backend) {
5196         This->fragment_priv = This->shader_priv;
5197     } else {
5198         This->fragment_priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(struct shader_arb_priv));
5199         if(!This->fragment_priv) return E_OUTOFMEMORY;
5200     }
5201     priv = This->fragment_priv;
5202     if (wine_rb_init(&priv->fragment_shaders, &wined3d_ffp_frag_program_rb_functions) == -1)
5203     {
5204         ERR("Failed to initialize rbtree.\n");
5205         HeapFree(GetProcessHeap(), 0, This->fragment_priv);
5206         return E_OUTOFMEMORY;
5207     }
5208     priv->use_arbfp_fixed_func = TRUE;
5209     return WINED3D_OK;
5210 }
5211
5212 /* Context activation is done by the caller. */
5213 static void arbfp_free_ffpshader(struct wine_rb_entry *entry, void *context)
5214 {
5215     const struct wined3d_gl_info *gl_info = context;
5216     struct arbfp_ffp_desc *entry_arb = WINE_RB_ENTRY_VALUE(entry, struct arbfp_ffp_desc, parent.entry);
5217
5218     ENTER_GL();
5219     GL_EXTCALL(glDeleteProgramsARB(1, &entry_arb->shader));
5220     checkGLcall("glDeleteProgramsARB(1, &entry_arb->shader)");
5221     HeapFree(GetProcessHeap(), 0, entry_arb);
5222     LEAVE_GL();
5223 }
5224
5225 /* Context activation is done by the caller. */
5226 static void arbfp_free(IWineD3DDevice *iface) {
5227     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *) iface;
5228     struct shader_arb_priv *priv = This->fragment_priv;
5229
5230     wine_rb_destroy(&priv->fragment_shaders, arbfp_free_ffpshader, &This->adapter->gl_info);
5231     priv->use_arbfp_fixed_func = FALSE;
5232
5233     if(This->shader_backend != &arb_program_shader_backend) {
5234         HeapFree(GetProcessHeap(), 0, This->fragment_priv);
5235     }
5236 }
5237
5238 static void arbfp_get_caps(WINED3DDEVTYPE devtype, const struct wined3d_gl_info *gl_info, struct fragment_caps *caps)
5239 {
5240     caps->TextureOpCaps =  WINED3DTEXOPCAPS_DISABLE                     |
5241                            WINED3DTEXOPCAPS_SELECTARG1                  |
5242                            WINED3DTEXOPCAPS_SELECTARG2                  |
5243                            WINED3DTEXOPCAPS_MODULATE4X                  |
5244                            WINED3DTEXOPCAPS_MODULATE2X                  |
5245                            WINED3DTEXOPCAPS_MODULATE                    |
5246                            WINED3DTEXOPCAPS_ADDSIGNED2X                 |
5247                            WINED3DTEXOPCAPS_ADDSIGNED                   |
5248                            WINED3DTEXOPCAPS_ADD                         |
5249                            WINED3DTEXOPCAPS_SUBTRACT                    |
5250                            WINED3DTEXOPCAPS_ADDSMOOTH                   |
5251                            WINED3DTEXOPCAPS_BLENDCURRENTALPHA           |
5252                            WINED3DTEXOPCAPS_BLENDFACTORALPHA            |
5253                            WINED3DTEXOPCAPS_BLENDTEXTUREALPHA           |
5254                            WINED3DTEXOPCAPS_BLENDDIFFUSEALPHA           |
5255                            WINED3DTEXOPCAPS_BLENDTEXTUREALPHAPM         |
5256                            WINED3DTEXOPCAPS_MODULATEALPHA_ADDCOLOR      |
5257                            WINED3DTEXOPCAPS_MODULATECOLOR_ADDALPHA      |
5258                            WINED3DTEXOPCAPS_MODULATEINVCOLOR_ADDALPHA   |
5259                            WINED3DTEXOPCAPS_MODULATEINVALPHA_ADDCOLOR   |
5260                            WINED3DTEXOPCAPS_DOTPRODUCT3                 |
5261                            WINED3DTEXOPCAPS_MULTIPLYADD                 |
5262                            WINED3DTEXOPCAPS_LERP                        |
5263                            WINED3DTEXOPCAPS_BUMPENVMAP                  |
5264                            WINED3DTEXOPCAPS_BUMPENVMAPLUMINANCE;
5265
5266     /* TODO: Implement WINED3DTEXOPCAPS_PREMODULATE */
5267
5268     caps->MaxTextureBlendStages   = 8;
5269     caps->MaxSimultaneousTextures = min(gl_info->limits.fragment_samplers, 8);
5270
5271     caps->PrimitiveMiscCaps |= WINED3DPMISCCAPS_TSSARGTEMP;
5272 }
5273 #undef GLINFO_LOCATION
5274
5275 #define GLINFO_LOCATION stateblock->device->adapter->gl_info
5276 static void state_texfactor_arbfp(DWORD state, IWineD3DStateBlockImpl *stateblock, struct wined3d_context *context)
5277 {
5278     IWineD3DDeviceImpl *device = stateblock->device;
5279     float col[4];
5280
5281     /* Don't load the parameter if we're using an arbfp pixel shader, otherwise we'll overwrite
5282      * application provided constants
5283      */
5284     if(device->shader_backend == &arb_program_shader_backend) {
5285         if (use_ps(stateblock)) return;
5286
5287         context->pshader_const_dirty[ARB_FFP_CONST_TFACTOR] = 1;
5288         device->highest_dirty_ps_const = max(device->highest_dirty_ps_const, ARB_FFP_CONST_TFACTOR + 1);
5289     }
5290
5291     D3DCOLORTOGLFLOAT4(stateblock->renderState[WINED3DRS_TEXTUREFACTOR], col);
5292     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_TFACTOR, col));
5293     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_TFACTOR, col)");
5294
5295 }
5296
5297 static void state_arb_specularenable(DWORD state, IWineD3DStateBlockImpl *stateblock, struct wined3d_context *context)
5298 {
5299     IWineD3DDeviceImpl *device = stateblock->device;
5300     float col[4];
5301
5302     /* Don't load the parameter if we're using an arbfp pixel shader, otherwise we'll overwrite
5303      * application provided constants
5304      */
5305     if(device->shader_backend == &arb_program_shader_backend) {
5306         if (use_ps(stateblock)) return;
5307
5308         context->pshader_const_dirty[ARB_FFP_CONST_SPECULAR_ENABLE] = 1;
5309         device->highest_dirty_ps_const = max(device->highest_dirty_ps_const, ARB_FFP_CONST_SPECULAR_ENABLE + 1);
5310     }
5311
5312     if(stateblock->renderState[WINED3DRS_SPECULARENABLE]) {
5313         /* The specular color has no alpha */
5314         col[0] = 1.0f; col[1] = 1.0f;
5315         col[2] = 1.0f; col[3] = 0.0f;
5316     } else {
5317         col[0] = 0.0f; col[1] = 0.0f;
5318         col[2] = 0.0f; col[3] = 0.0f;
5319     }
5320     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_SPECULAR_ENABLE, col));
5321     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_SPECULAR_ENABLE, col)");
5322 }
5323
5324 static void set_bumpmat_arbfp(DWORD state, IWineD3DStateBlockImpl *stateblock, struct wined3d_context *context)
5325 {
5326     DWORD stage = (state - STATE_TEXTURESTAGE(0, 0)) / (WINED3D_HIGHEST_TEXTURE_STATE + 1);
5327     IWineD3DDeviceImpl *device = stateblock->device;
5328     float mat[2][2];
5329
5330     if (use_ps(stateblock))
5331     {
5332         if (stage != 0
5333                 && (((IWineD3DPixelShaderImpl *)stateblock->pixelShader)->baseShader.reg_maps.bumpmat & (1 << stage)))
5334         {
5335             /* The pixel shader has to know the bump env matrix. Do a constants update if it isn't scheduled
5336              * anyway
5337              */
5338             if(!isStateDirty(context, STATE_PIXELSHADERCONSTANT)) {
5339                 device->StateTable[STATE_PIXELSHADERCONSTANT].apply(STATE_PIXELSHADERCONSTANT, stateblock, context);
5340             }
5341         }
5342
5343         if(device->shader_backend == &arb_program_shader_backend) {
5344             /* Exit now, don't set the bumpmat below, otherwise we may overwrite pixel shader constants */
5345             return;
5346         }
5347     } else if(device->shader_backend == &arb_program_shader_backend) {
5348         context->pshader_const_dirty[ARB_FFP_CONST_BUMPMAT(stage)] = 1;
5349         device->highest_dirty_ps_const = max(device->highest_dirty_ps_const, ARB_FFP_CONST_BUMPMAT(stage) + 1);
5350     }
5351
5352     mat[0][0] = *((float *) &stateblock->textureState[stage][WINED3DTSS_BUMPENVMAT00]);
5353     mat[0][1] = *((float *) &stateblock->textureState[stage][WINED3DTSS_BUMPENVMAT01]);
5354     mat[1][0] = *((float *) &stateblock->textureState[stage][WINED3DTSS_BUMPENVMAT10]);
5355     mat[1][1] = *((float *) &stateblock->textureState[stage][WINED3DTSS_BUMPENVMAT11]);
5356
5357     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_BUMPMAT(stage), &mat[0][0]));
5358     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_BUMPMAT(stage), &mat[0][0])");
5359 }
5360
5361 static void tex_bumpenvlum_arbfp(DWORD state, IWineD3DStateBlockImpl *stateblock, struct wined3d_context *context)
5362 {
5363     DWORD stage = (state - STATE_TEXTURESTAGE(0, 0)) / (WINED3D_HIGHEST_TEXTURE_STATE + 1);
5364     IWineD3DDeviceImpl *device = stateblock->device;
5365     float param[4];
5366
5367     if (use_ps(stateblock))
5368     {
5369         if (stage != 0
5370                 && (((IWineD3DPixelShaderImpl *)stateblock->pixelShader)->baseShader.reg_maps.luminanceparams & (1 << stage)))
5371         {
5372             /* The pixel shader has to know the luminance offset. Do a constants update if it
5373              * isn't scheduled anyway
5374              */
5375             if(!isStateDirty(context, STATE_PIXELSHADERCONSTANT)) {
5376                 device->StateTable[STATE_PIXELSHADERCONSTANT].apply(STATE_PIXELSHADERCONSTANT, stateblock, context);
5377             }
5378         }
5379
5380         if(device->shader_backend == &arb_program_shader_backend) {
5381             /* Exit now, don't set the bumpmat below, otherwise we may overwrite pixel shader constants */
5382             return;
5383         }
5384     } else if(device->shader_backend == &arb_program_shader_backend) {
5385         context->pshader_const_dirty[ARB_FFP_CONST_LUMINANCE(stage)] = 1;
5386         device->highest_dirty_ps_const = max(device->highest_dirty_ps_const, ARB_FFP_CONST_LUMINANCE(stage) + 1);
5387     }
5388
5389     param[0] = *((float *) &stateblock->textureState[stage][WINED3DTSS_BUMPENVLSCALE]);
5390     param[1] = *((float *) &stateblock->textureState[stage][WINED3DTSS_BUMPENVLOFFSET]);
5391     param[2] = 0.0f;
5392     param[3] = 0.0f;
5393
5394     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_LUMINANCE(stage), param));
5395     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_LUMINANCE(stage), param)");
5396 }
5397
5398 static const char *get_argreg(struct wined3d_shader_buffer *buffer, DWORD argnum, unsigned int stage, DWORD arg)
5399 {
5400     const char *ret;
5401
5402     if(arg == ARG_UNUSED) return "unused"; /* This is the marker for unused registers */
5403
5404     switch(arg & WINED3DTA_SELECTMASK) {
5405         case WINED3DTA_DIFFUSE:
5406             ret = "fragment.color.primary"; break;
5407
5408         case WINED3DTA_CURRENT:
5409             if(stage == 0) ret = "fragment.color.primary";
5410             else ret = "ret";
5411             break;
5412
5413         case WINED3DTA_TEXTURE:
5414             switch(stage) {
5415                 case 0: ret = "tex0"; break;
5416                 case 1: ret = "tex1"; break;
5417                 case 2: ret = "tex2"; break;
5418                 case 3: ret = "tex3"; break;
5419                 case 4: ret = "tex4"; break;
5420                 case 5: ret = "tex5"; break;
5421                 case 6: ret = "tex6"; break;
5422                 case 7: ret = "tex7"; break;
5423                 default: ret = "unknown texture";
5424             }
5425             break;
5426
5427         case WINED3DTA_TFACTOR:
5428             ret = "tfactor"; break;
5429
5430         case WINED3DTA_SPECULAR:
5431             ret = "fragment.color.secondary"; break;
5432
5433         case WINED3DTA_TEMP:
5434             ret = "tempreg"; break;
5435
5436         case WINED3DTA_CONSTANT:
5437             FIXME("Implement perstage constants\n");
5438             switch(stage) {
5439                 case 0: ret = "const0"; break;
5440                 case 1: ret = "const1"; break;
5441                 case 2: ret = "const2"; break;
5442                 case 3: ret = "const3"; break;
5443                 case 4: ret = "const4"; break;
5444                 case 5: ret = "const5"; break;
5445                 case 6: ret = "const6"; break;
5446                 case 7: ret = "const7"; break;
5447                 default: ret = "unknown constant";
5448             }
5449             break;
5450
5451         default:
5452             return "unknown";
5453     }
5454
5455     if(arg & WINED3DTA_COMPLEMENT) {
5456         shader_addline(buffer, "SUB arg%u, const.x, %s;\n", argnum, ret);
5457         if(argnum == 0) ret = "arg0";
5458         if(argnum == 1) ret = "arg1";
5459         if(argnum == 2) ret = "arg2";
5460     }
5461     if(arg & WINED3DTA_ALPHAREPLICATE) {
5462         shader_addline(buffer, "MOV arg%u, %s.w;\n", argnum, ret);
5463         if(argnum == 0) ret = "arg0";
5464         if(argnum == 1) ret = "arg1";
5465         if(argnum == 2) ret = "arg2";
5466     }
5467     return ret;
5468 }
5469
5470 static void gen_ffp_instr(struct wined3d_shader_buffer *buffer, unsigned int stage, BOOL color,
5471         BOOL alpha, DWORD dst, DWORD op, DWORD dw_arg0, DWORD dw_arg1, DWORD dw_arg2)
5472 {
5473     const char *dstmask, *dstreg, *arg0, *arg1, *arg2;
5474     unsigned int mul = 1;
5475     BOOL mul_final_dest = FALSE;
5476
5477     if(color && alpha) dstmask = "";
5478     else if(color) dstmask = ".xyz";
5479     else dstmask = ".w";
5480
5481     if(dst == tempreg) dstreg = "tempreg";
5482     else dstreg = "ret";
5483
5484     arg0 = get_argreg(buffer, 0, stage, dw_arg0);
5485     arg1 = get_argreg(buffer, 1, stage, dw_arg1);
5486     arg2 = get_argreg(buffer, 2, stage, dw_arg2);
5487
5488     switch(op) {
5489         case WINED3DTOP_DISABLE:
5490             if(stage == 0) shader_addline(buffer, "MOV %s%s, fragment.color.primary;\n", dstreg, dstmask);
5491             break;
5492
5493         case WINED3DTOP_SELECTARG2:
5494             arg1 = arg2;
5495         case WINED3DTOP_SELECTARG1:
5496             shader_addline(buffer, "MOV %s%s, %s;\n", dstreg, dstmask, arg1);
5497             break;
5498
5499         case WINED3DTOP_MODULATE4X:
5500             mul = 2;
5501         case WINED3DTOP_MODULATE2X:
5502             mul *= 2;
5503             if(strcmp(dstreg, "result.color") == 0) {
5504                 dstreg = "ret";
5505                 mul_final_dest = TRUE;
5506             }
5507         case WINED3DTOP_MODULATE:
5508             shader_addline(buffer, "MUL %s%s, %s, %s;\n", dstreg, dstmask, arg1, arg2);
5509             break;
5510
5511         case WINED3DTOP_ADDSIGNED2X:
5512             mul = 2;
5513             if(strcmp(dstreg, "result.color") == 0) {
5514                 dstreg = "ret";
5515                 mul_final_dest = TRUE;
5516             }
5517         case WINED3DTOP_ADDSIGNED:
5518             shader_addline(buffer, "SUB arg2, %s, const.w;\n", arg2);
5519             arg2 = "arg2";
5520         case WINED3DTOP_ADD:
5521             shader_addline(buffer, "ADD_SAT %s%s, %s, %s;\n", dstreg, dstmask, arg1, arg2);
5522             break;
5523
5524         case WINED3DTOP_SUBTRACT:
5525             shader_addline(buffer, "SUB_SAT %s%s, %s, %s;\n", dstreg, dstmask, arg1, arg2);
5526             break;
5527
5528         case WINED3DTOP_ADDSMOOTH:
5529             shader_addline(buffer, "SUB arg1, const.x, %s;\n", arg1);
5530             shader_addline(buffer, "MAD_SAT %s%s, arg1, %s, %s;\n", dstreg, dstmask, arg2, arg1);
5531             break;
5532
5533         case WINED3DTOP_BLENDCURRENTALPHA:
5534             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_CURRENT);
5535             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
5536             break;
5537         case WINED3DTOP_BLENDFACTORALPHA:
5538             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_TFACTOR);
5539             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
5540             break;
5541         case WINED3DTOP_BLENDTEXTUREALPHA:
5542             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_TEXTURE);
5543             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
5544             break;
5545         case WINED3DTOP_BLENDDIFFUSEALPHA:
5546             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_DIFFUSE);
5547             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
5548             break;
5549
5550         case WINED3DTOP_BLENDTEXTUREALPHAPM:
5551             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_TEXTURE);
5552             shader_addline(buffer, "SUB arg0.w, const.x, %s.w;\n", arg0);
5553             shader_addline(buffer, "MAD_SAT %s%s, %s, arg0.w, %s;\n", dstreg, dstmask, arg2, arg1);
5554             break;
5555
5556         /* D3DTOP_PREMODULATE ???? */
5557
5558         case WINED3DTOP_MODULATEINVALPHA_ADDCOLOR:
5559             shader_addline(buffer, "SUB arg0.w, const.x, %s;\n", arg1);
5560             shader_addline(buffer, "MAD_SAT %s%s, arg0.w, %s, %s;\n", dstreg, dstmask, arg2, arg1);
5561             break;
5562         case WINED3DTOP_MODULATEALPHA_ADDCOLOR:
5563             shader_addline(buffer, "MAD_SAT %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg1, arg2, arg1);
5564             break;
5565         case WINED3DTOP_MODULATEINVCOLOR_ADDALPHA:
5566             shader_addline(buffer, "SUB arg0, const.x, %s;\n", arg1);
5567             shader_addline(buffer, "MAD_SAT %s%s, arg0, %s, %s.w;\n", dstreg, dstmask, arg2, arg1);
5568             break;
5569         case WINED3DTOP_MODULATECOLOR_ADDALPHA:
5570             shader_addline(buffer, "MAD_SAT %s%s, %s, %s, %s.w;\n", dstreg, dstmask, arg1, arg2, arg1);
5571             break;
5572
5573         case WINED3DTOP_DOTPRODUCT3:
5574             mul = 4;
5575             if(strcmp(dstreg, "result.color") == 0) {
5576                 dstreg = "ret";
5577                 mul_final_dest = TRUE;
5578             }
5579             shader_addline(buffer, "SUB arg1, %s, const.w;\n", arg1);
5580             shader_addline(buffer, "SUB arg2, %s, const.w;\n", arg2);
5581             shader_addline(buffer, "DP3_SAT %s%s, arg1, arg2;\n", dstreg, dstmask);
5582             break;
5583
5584         case WINED3DTOP_MULTIPLYADD:
5585             shader_addline(buffer, "MAD_SAT %s%s, %s, %s, %s;\n", dstreg, dstmask, arg1, arg2, arg0);
5586             break;
5587
5588         case WINED3DTOP_LERP:
5589             /* The msdn is not quite right here */
5590             shader_addline(buffer, "LRP %s%s, %s, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
5591             break;
5592
5593         case WINED3DTOP_BUMPENVMAP:
5594         case WINED3DTOP_BUMPENVMAPLUMINANCE:
5595             /* Those are handled in the first pass of the shader(generation pass 1 and 2) already */
5596             break;
5597
5598         default:
5599             FIXME("Unhandled texture op %08x\n", op);
5600     }
5601
5602     if(mul == 2) {
5603         shader_addline(buffer, "MUL_SAT %s%s, %s, const.y;\n", mul_final_dest ? "result.color" : dstreg, dstmask, dstreg);
5604     } else if(mul == 4) {
5605         shader_addline(buffer, "MUL_SAT %s%s, %s, const.z;\n", mul_final_dest ? "result.color" : dstreg, dstmask, dstreg);
5606     }
5607 }
5608
5609 /* The stateblock is passed for GLINFO_LOCATION */
5610 static GLuint gen_arbfp_ffp_shader(const struct ffp_frag_settings *settings, IWineD3DStateBlockImpl *stateblock)
5611 {
5612     unsigned int stage;
5613     struct wined3d_shader_buffer buffer;
5614     BOOL tex_read[MAX_TEXTURES] = {FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE};
5615     BOOL bump_used[MAX_TEXTURES] = {FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE};
5616     BOOL luminance_used[MAX_TEXTURES] = {FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE};
5617     const char *textype;
5618     const char *instr, *sat;
5619     char colorcor_dst[8];
5620     GLuint ret;
5621     DWORD arg0, arg1, arg2;
5622     BOOL tempreg_used = FALSE, tfactor_used = FALSE;
5623     BOOL op_equal;
5624     const char *final_combiner_src = "ret";
5625     GLint pos;
5626
5627     /* Find out which textures are read */
5628     for(stage = 0; stage < MAX_TEXTURES; stage++) {
5629         if(settings->op[stage].cop == WINED3DTOP_DISABLE) break;
5630         arg0 = settings->op[stage].carg0 & WINED3DTA_SELECTMASK;
5631         arg1 = settings->op[stage].carg1 & WINED3DTA_SELECTMASK;
5632         arg2 = settings->op[stage].carg2 & WINED3DTA_SELECTMASK;
5633         if(arg0 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
5634         if(arg1 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
5635         if(arg2 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
5636
5637         if(settings->op[stage].cop == WINED3DTOP_BLENDTEXTUREALPHA) tex_read[stage] = TRUE;
5638         if(settings->op[stage].cop == WINED3DTOP_BLENDTEXTUREALPHAPM) tex_read[stage] = TRUE;
5639         if(settings->op[stage].cop == WINED3DTOP_BUMPENVMAP) {
5640             bump_used[stage] = TRUE;
5641             tex_read[stage] = TRUE;
5642         }
5643         if(settings->op[stage].cop == WINED3DTOP_BUMPENVMAPLUMINANCE) {
5644             bump_used[stage] = TRUE;
5645             tex_read[stage] = TRUE;
5646             luminance_used[stage] = TRUE;
5647         } else if(settings->op[stage].cop == WINED3DTOP_BLENDFACTORALPHA) {
5648             tfactor_used = TRUE;
5649         }
5650
5651         if(arg0 == WINED3DTA_TFACTOR || arg1 == WINED3DTA_TFACTOR || arg2 == WINED3DTA_TFACTOR) {
5652             tfactor_used = TRUE;
5653         }
5654
5655         if(settings->op[stage].dst == tempreg) tempreg_used = TRUE;
5656         if(arg0 == WINED3DTA_TEMP || arg1 == WINED3DTA_TEMP || arg2 == WINED3DTA_TEMP) {
5657             tempreg_used = TRUE;
5658         }
5659
5660         if(settings->op[stage].aop == WINED3DTOP_DISABLE) continue;
5661         arg0 = settings->op[stage].aarg0 & WINED3DTA_SELECTMASK;
5662         arg1 = settings->op[stage].aarg1 & WINED3DTA_SELECTMASK;
5663         arg2 = settings->op[stage].aarg2 & WINED3DTA_SELECTMASK;
5664         if(arg0 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
5665         if(arg1 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
5666         if(arg2 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
5667
5668         if(arg0 == WINED3DTA_TEMP || arg1 == WINED3DTA_TEMP || arg2 == WINED3DTA_TEMP) {
5669             tempreg_used = TRUE;
5670         }
5671         if(arg0 == WINED3DTA_TFACTOR || arg1 == WINED3DTA_TFACTOR || arg2 == WINED3DTA_TFACTOR) {
5672             tfactor_used = TRUE;
5673         }
5674     }
5675
5676     /* Shader header */
5677     if (!shader_buffer_init(&buffer))
5678     {
5679         ERR("Failed to initialize shader buffer.\n");
5680         return 0;
5681     }
5682
5683     shader_addline(&buffer, "!!ARBfp1.0\n");
5684
5685     switch(settings->fog) {
5686         case FOG_OFF:                                                         break;
5687         case FOG_LINEAR: shader_addline(&buffer, "OPTION ARB_fog_linear;\n"); break;
5688         case FOG_EXP:    shader_addline(&buffer, "OPTION ARB_fog_exp;\n");    break;
5689         case FOG_EXP2:   shader_addline(&buffer, "OPTION ARB_fog_exp2;\n");   break;
5690         default: FIXME("Unexpected fog setting %d\n", settings->fog);
5691     }
5692
5693     shader_addline(&buffer, "PARAM const = {1, 2, 4, 0.5};\n");
5694     shader_addline(&buffer, "TEMP TMP;\n");
5695     shader_addline(&buffer, "TEMP ret;\n");
5696     if(tempreg_used || settings->sRGB_write) shader_addline(&buffer, "TEMP tempreg;\n");
5697     shader_addline(&buffer, "TEMP arg0;\n");
5698     shader_addline(&buffer, "TEMP arg1;\n");
5699     shader_addline(&buffer, "TEMP arg2;\n");
5700     for(stage = 0; stage < MAX_TEXTURES; stage++) {
5701         if(!tex_read[stage]) continue;
5702         shader_addline(&buffer, "TEMP tex%u;\n", stage);
5703         if(!bump_used[stage]) continue;
5704         shader_addline(&buffer, "PARAM bumpmat%u = program.env[%u];\n", stage, ARB_FFP_CONST_BUMPMAT(stage));
5705         if(!luminance_used[stage]) continue;
5706         shader_addline(&buffer, "PARAM luminance%u = program.env[%u];\n", stage, ARB_FFP_CONST_LUMINANCE(stage));
5707     }
5708     if(tfactor_used) {
5709         shader_addline(&buffer, "PARAM tfactor = program.env[%u];\n", ARB_FFP_CONST_TFACTOR);
5710     }
5711         shader_addline(&buffer, "PARAM specular_enable = program.env[%u];\n", ARB_FFP_CONST_SPECULAR_ENABLE);
5712
5713     if(settings->sRGB_write) {
5714         shader_addline(&buffer, "PARAM srgb_consts1 = {%f, %f, %f, %f};\n",
5715                        srgb_mul_low, srgb_cmp, srgb_pow, srgb_mul_high);
5716         shader_addline(&buffer, "PARAM srgb_consts2 = {%f, %f, %f, %f};\n",
5717                        srgb_sub_high, 0.0, 0.0, 0.0);
5718     }
5719
5720     if(ffp_clip_emul(stateblock) && settings->emul_clipplanes) shader_addline(&buffer, "KIL fragment.texcoord[7];\n");
5721
5722     /* Generate texture sampling instructions) */
5723     for(stage = 0; stage < MAX_TEXTURES && settings->op[stage].cop != WINED3DTOP_DISABLE; stage++) {
5724         if(!tex_read[stage]) continue;
5725
5726         switch(settings->op[stage].tex_type) {
5727             case tex_1d:                    textype = "1D";     break;
5728             case tex_2d:                    textype = "2D";     break;
5729             case tex_3d:                    textype = "3D";     break;
5730             case tex_cube:                  textype = "CUBE";   break;
5731             case tex_rect:                  textype = "RECT";   break;
5732             default: textype = "unexpected_textype";   break;
5733         }
5734
5735         if(settings->op[stage].cop == WINED3DTOP_BUMPENVMAP ||
5736            settings->op[stage].cop == WINED3DTOP_BUMPENVMAPLUMINANCE) {
5737             sat = "";
5738         } else {
5739             sat = "_SAT";
5740         }
5741
5742         if(settings->op[stage].projected == proj_none) {
5743             instr = "TEX";
5744         } else if(settings->op[stage].projected == proj_count4 ||
5745                   settings->op[stage].projected == proj_count3) {
5746             instr = "TXP";
5747         } else {
5748             FIXME("Unexpected projection mode %d\n", settings->op[stage].projected);
5749             instr = "TXP";
5750         }
5751
5752         if(stage > 0 &&
5753            (settings->op[stage - 1].cop == WINED3DTOP_BUMPENVMAP ||
5754             settings->op[stage - 1].cop == WINED3DTOP_BUMPENVMAPLUMINANCE)) {
5755             shader_addline(&buffer, "SWZ arg1, bumpmat%u, x, z, 0, 0;\n", stage - 1);
5756             shader_addline(&buffer, "DP3 ret.x, arg1, tex%u;\n", stage - 1);
5757             shader_addline(&buffer, "SWZ arg1, bumpmat%u, y, w, 0, 0;\n", stage - 1);
5758             shader_addline(&buffer, "DP3 ret.y, arg1, tex%u;\n", stage - 1);
5759
5760             /* with projective textures, texbem only divides the static texture coord, not the displacement,
5761              * so multiply the displacement with the dividing parameter before passing it to TXP
5762              */
5763             if (settings->op[stage].projected != proj_none) {
5764                 if(settings->op[stage].projected == proj_count4) {
5765                     shader_addline(&buffer, "MOV ret.w, fragment.texcoord[%u].w;\n", stage);
5766                     shader_addline(&buffer, "MUL ret.xyz, ret, fragment.texcoord[%u].w, fragment.texcoord[%u];\n", stage, stage);
5767                 } else {
5768                     shader_addline(&buffer, "MOV ret.w, fragment.texcoord[%u].z;\n", stage);
5769                     shader_addline(&buffer, "MAD ret.xyz, ret, fragment.texcoord[%u].z, fragment.texcoord[%u];\n", stage, stage);
5770                 }
5771             } else {
5772                 shader_addline(&buffer, "ADD ret, ret, fragment.texcoord[%u];\n", stage);
5773             }
5774
5775             shader_addline(&buffer, "%s%s tex%u, ret, texture[%u], %s;\n",
5776                            instr, sat, stage, stage, textype);
5777             if(settings->op[stage - 1].cop == WINED3DTOP_BUMPENVMAPLUMINANCE) {
5778                 shader_addline(&buffer, "MAD_SAT ret.x, tex%u.z, luminance%u.x, luminance%u.y;\n",
5779                                stage - 1, stage - 1, stage - 1);
5780                 shader_addline(&buffer, "MUL tex%u, tex%u, ret.x;\n", stage, stage);
5781             }
5782         } else if(settings->op[stage].projected == proj_count3) {
5783             shader_addline(&buffer, "MOV ret, fragment.texcoord[%u];\n", stage);
5784             shader_addline(&buffer, "MOV ret.w, ret.z;\n");
5785             shader_addline(&buffer, "%s%s tex%u, ret, texture[%u], %s;\n",
5786                             instr, sat, stage, stage, textype);
5787         } else {
5788             shader_addline(&buffer, "%s%s tex%u, fragment.texcoord[%u], texture[%u], %s;\n",
5789                             instr, sat, stage, stage, stage, textype);
5790         }
5791
5792         sprintf(colorcor_dst, "tex%u", stage);
5793         gen_color_correction(&buffer, colorcor_dst, WINED3DSP_WRITEMASK_ALL, "const.x", "const.y",
5794                 settings->op[stage].color_fixup);
5795     }
5796
5797     /* Generate the main shader */
5798     for(stage = 0; stage < MAX_TEXTURES; stage++) {
5799         if(settings->op[stage].cop == WINED3DTOP_DISABLE) {
5800             if(stage == 0) {
5801                 final_combiner_src = "fragment.color.primary";
5802             }
5803             break;
5804         }
5805
5806         if(settings->op[stage].cop == WINED3DTOP_SELECTARG1 &&
5807            settings->op[stage].aop == WINED3DTOP_SELECTARG1) {
5808             op_equal = settings->op[stage].carg1 == settings->op[stage].aarg1;
5809         } else if(settings->op[stage].cop == WINED3DTOP_SELECTARG1 &&
5810                   settings->op[stage].aop == WINED3DTOP_SELECTARG2) {
5811             op_equal = settings->op[stage].carg1 == settings->op[stage].aarg2;
5812         } else if(settings->op[stage].cop == WINED3DTOP_SELECTARG2 &&
5813                   settings->op[stage].aop == WINED3DTOP_SELECTARG1) {
5814             op_equal = settings->op[stage].carg2 == settings->op[stage].aarg1;
5815         } else if(settings->op[stage].cop == WINED3DTOP_SELECTARG2 &&
5816                   settings->op[stage].aop == WINED3DTOP_SELECTARG2) {
5817             op_equal = settings->op[stage].carg2 == settings->op[stage].aarg2;
5818         } else {
5819             op_equal = settings->op[stage].aop   == settings->op[stage].cop &&
5820                        settings->op[stage].carg0 == settings->op[stage].aarg0 &&
5821                        settings->op[stage].carg1 == settings->op[stage].aarg1 &&
5822                        settings->op[stage].carg2 == settings->op[stage].aarg2;
5823         }
5824
5825         if(settings->op[stage].aop == WINED3DTOP_DISABLE) {
5826             gen_ffp_instr(&buffer, stage, TRUE, FALSE, settings->op[stage].dst,
5827                           settings->op[stage].cop, settings->op[stage].carg0,
5828                           settings->op[stage].carg1, settings->op[stage].carg2);
5829             if(stage == 0) {
5830                 shader_addline(&buffer, "MOV ret.w, fragment.color.primary.w;\n");
5831             }
5832         } else if(op_equal) {
5833             gen_ffp_instr(&buffer, stage, TRUE, TRUE, settings->op[stage].dst,
5834                           settings->op[stage].cop, settings->op[stage].carg0,
5835                           settings->op[stage].carg1, settings->op[stage].carg2);
5836         } else {
5837             gen_ffp_instr(&buffer, stage, TRUE, FALSE, settings->op[stage].dst,
5838                           settings->op[stage].cop, settings->op[stage].carg0,
5839                           settings->op[stage].carg1, settings->op[stage].carg2);
5840             gen_ffp_instr(&buffer, stage, FALSE, TRUE, settings->op[stage].dst,
5841                           settings->op[stage].aop, settings->op[stage].aarg0,
5842                           settings->op[stage].aarg1, settings->op[stage].aarg2);
5843         }
5844     }
5845
5846     if(settings->sRGB_write) {
5847         shader_addline(&buffer, "MAD ret, fragment.color.secondary, specular_enable, %s;\n", final_combiner_src);
5848         arbfp_add_sRGB_correction(&buffer, "ret", "arg0", "arg1", "arg2", "tempreg", FALSE);
5849         shader_addline(&buffer, "MOV result.color, ret;\n");
5850     } else {
5851         shader_addline(&buffer, "MAD result.color, fragment.color.secondary, specular_enable, %s;\n", final_combiner_src);
5852     }
5853
5854     /* Footer */
5855     shader_addline(&buffer, "END\n");
5856
5857     /* Generate the shader */
5858     GL_EXTCALL(glGenProgramsARB(1, &ret));
5859     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, ret));
5860     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
5861             strlen(buffer.buffer), buffer.buffer));
5862     checkGLcall("glProgramStringARB()");
5863
5864     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
5865     if (pos != -1)
5866     {
5867         FIXME("Fragment program error at position %d: %s\n", pos,
5868               debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
5869     }
5870     else
5871     {
5872         GLint native;
5873
5874         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
5875         checkGLcall("glGetProgramivARB()");
5876         if (!native) WARN("Program exceeds native resource limits.\n");
5877     }
5878
5879     shader_buffer_free(&buffer);
5880     return ret;
5881 }
5882
5883 static void fragment_prog_arbfp(DWORD state, IWineD3DStateBlockImpl *stateblock, struct wined3d_context *context)
5884 {
5885     IWineD3DDeviceImpl *device = stateblock->device;
5886     struct shader_arb_priv *priv = device->fragment_priv;
5887     BOOL use_pshader = use_ps(stateblock);
5888     BOOL use_vshader = use_vs(stateblock);
5889     struct ffp_frag_settings settings;
5890     const struct arbfp_ffp_desc *desc;
5891     unsigned int i;
5892
5893     TRACE("state %#x, stateblock %p, context %p\n", state, stateblock, context);
5894
5895     if(isStateDirty(context, STATE_RENDER(WINED3DRS_FOGENABLE))) {
5896         if(!use_pshader && device->shader_backend == &arb_program_shader_backend && context->last_was_pshader) {
5897             /* Reload fixed function constants since they collide with the pixel shader constants */
5898             for(i = 0; i < MAX_TEXTURES; i++) {
5899                 set_bumpmat_arbfp(STATE_TEXTURESTAGE(i, WINED3DTSS_BUMPENVMAT00), stateblock, context);
5900             }
5901             state_texfactor_arbfp(STATE_RENDER(WINED3DRS_TEXTUREFACTOR), stateblock, context);
5902             state_arb_specularenable(STATE_RENDER(WINED3DRS_SPECULARENABLE), stateblock, context);
5903         } else if(use_pshader && !isStateDirty(context, device->StateTable[STATE_VSHADER].representative)) {
5904             device->shader_backend->shader_select(context, use_pshader, use_vshader);
5905         }
5906         return;
5907     }
5908
5909     if(!use_pshader) {
5910         /* Find or create a shader implementing the fixed function pipeline settings, then activate it */
5911         gen_ffp_frag_op(stateblock, &settings, FALSE);
5912         desc = (const struct arbfp_ffp_desc *)find_ffp_frag_shader(&priv->fragment_shaders, &settings);
5913         if(!desc) {
5914             struct arbfp_ffp_desc *new_desc = HeapAlloc(GetProcessHeap(), 0, sizeof(*new_desc));
5915             if (!new_desc)
5916             {
5917                 ERR("Out of memory\n");
5918                 return;
5919             }
5920             new_desc->num_textures_used = 0;
5921             for (i = 0; i < context->gl_info->limits.texture_stages; ++i)
5922             {
5923                 if(settings.op[i].cop == WINED3DTOP_DISABLE) break;
5924                 new_desc->num_textures_used = i;
5925             }
5926
5927             memcpy(&new_desc->parent.settings, &settings, sizeof(settings));
5928             new_desc->shader = gen_arbfp_ffp_shader(&settings, stateblock);
5929             add_ffp_frag_shader(&priv->fragment_shaders, &new_desc->parent);
5930             TRACE("Allocated fixed function replacement shader descriptor %p\n", new_desc);
5931             desc = new_desc;
5932         }
5933
5934         /* Now activate the replacement program. GL_FRAGMENT_PROGRAM_ARB is already active(however, note the
5935          * comment above the shader_select call below). If e.g. GLSL is active, the shader_select call will
5936          * deactivate it.
5937          */
5938         GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, desc->shader));
5939         checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, desc->shader)");
5940         priv->current_fprogram_id = desc->shader;
5941
5942         if(device->shader_backend == &arb_program_shader_backend && context->last_was_pshader) {
5943             /* Reload fixed function constants since they collide with the pixel shader constants */
5944             for(i = 0; i < MAX_TEXTURES; i++) {
5945                 set_bumpmat_arbfp(STATE_TEXTURESTAGE(i, WINED3DTSS_BUMPENVMAT00), stateblock, context);
5946             }
5947             state_texfactor_arbfp(STATE_RENDER(WINED3DRS_TEXTUREFACTOR), stateblock, context);
5948             state_arb_specularenable(STATE_RENDER(WINED3DRS_SPECULARENABLE), stateblock, context);
5949         }
5950         context->last_was_pshader = FALSE;
5951     } else {
5952         context->last_was_pshader = TRUE;
5953     }
5954
5955     /* Finally, select the shader. If a pixel shader is used, it will be set and enabled by the shader backend.
5956      * If this shader backend is arbfp(most likely), then it will simply overwrite the last fixed function replace-
5957      * ment shader. If the shader backend is not ARB, it currently is important that the opengl implementation
5958      * type overwrites GL_ARB_fragment_program. This is currently the case with GLSL. If we really want to use
5959      * atifs or nvrc pixel shaders with arb fragment programs we'd have to disable GL_FRAGMENT_PROGRAM_ARB here
5960      *
5961      * Don't call shader_select if the vertex shader is dirty, because it will be called later on by the vertex
5962      * shader handler
5963      */
5964     if(!isStateDirty(context, device->StateTable[STATE_VSHADER].representative)) {
5965         device->shader_backend->shader_select(context, use_pshader, use_vshader);
5966
5967         if (!isStateDirty(context, STATE_VERTEXSHADERCONSTANT) && (use_vshader || use_pshader)) {
5968             device->StateTable[STATE_VERTEXSHADERCONSTANT].apply(STATE_VERTEXSHADERCONSTANT, stateblock, context);
5969         }
5970     }
5971     if(use_pshader) {
5972         device->StateTable[STATE_PIXELSHADERCONSTANT].apply(STATE_PIXELSHADERCONSTANT, stateblock, context);
5973     }
5974 }
5975
5976 /* We can't link the fog states to the fragment state directly since the vertex pipeline links them
5977  * to FOGENABLE. A different linking in different pipeline parts can't be expressed in the combined
5978  * state table, so we need to handle that with a forwarding function. The other invisible side effect
5979  * is that changing the fog start and fog end(which links to FOGENABLE in vertex) results in the
5980  * fragment_prog_arbfp function being called because FOGENABLE is dirty, which calls this function here
5981  */
5982 static void state_arbfp_fog(DWORD state, IWineD3DStateBlockImpl *stateblock, struct wined3d_context *context)
5983 {
5984     enum fogsource new_source;
5985
5986     TRACE("state %#x, stateblock %p, context %p\n", state, stateblock, context);
5987
5988     if(!isStateDirty(context, STATE_PIXELSHADER)) {
5989         fragment_prog_arbfp(state, stateblock, context);
5990     }
5991
5992     if(!stateblock->renderState[WINED3DRS_FOGENABLE]) return;
5993
5994     if(stateblock->renderState[WINED3DRS_FOGTABLEMODE] == WINED3DFOG_NONE) {
5995         if(use_vs(stateblock)) {
5996             new_source = FOGSOURCE_VS;
5997         } else {
5998             if(stateblock->renderState[WINED3DRS_FOGVERTEXMODE] == WINED3DFOG_NONE || context->last_was_rhw) {
5999                 new_source = FOGSOURCE_COORD;
6000             } else {
6001                 new_source = FOGSOURCE_FFP;
6002             }
6003         }
6004     } else {
6005         new_source = FOGSOURCE_FFP;
6006     }
6007     if(new_source != context->fog_source) {
6008         context->fog_source = new_source;
6009         state_fogstartend(STATE_RENDER(WINED3DRS_FOGSTART), stateblock, context);
6010     }
6011 }
6012
6013 static void textransform(DWORD state, IWineD3DStateBlockImpl *stateblock, struct wined3d_context *context)
6014 {
6015     if(!isStateDirty(context, STATE_PIXELSHADER)) {
6016         fragment_prog_arbfp(state, stateblock, context);
6017     }
6018 }
6019
6020 #undef GLINFO_LOCATION
6021
6022 static const struct StateEntryTemplate arbfp_fragmentstate_template[] = {
6023     {STATE_RENDER(WINED3DRS_TEXTUREFACTOR),               { STATE_RENDER(WINED3DRS_TEXTUREFACTOR),              state_texfactor_arbfp   }, WINED3D_GL_EXT_NONE             },
6024     {STATE_TEXTURESTAGE(0, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6025     {STATE_TEXTURESTAGE(0, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6026     {STATE_TEXTURESTAGE(0, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6027     {STATE_TEXTURESTAGE(0, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6028     {STATE_TEXTURESTAGE(0, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6029     {STATE_TEXTURESTAGE(0, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6030     {STATE_TEXTURESTAGE(0, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6031     {STATE_TEXTURESTAGE(0, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6032     {STATE_TEXTURESTAGE(0, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6033     {STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6034     {STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6035     {STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6036     {STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6037     {STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6038     {STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6039     {STATE_TEXTURESTAGE(1, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6040     {STATE_TEXTURESTAGE(1, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6041     {STATE_TEXTURESTAGE(1, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6042     {STATE_TEXTURESTAGE(1, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6043     {STATE_TEXTURESTAGE(1, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6044     {STATE_TEXTURESTAGE(1, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6045     {STATE_TEXTURESTAGE(1, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6046     {STATE_TEXTURESTAGE(1, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6047     {STATE_TEXTURESTAGE(1, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6048     {STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6049     {STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6050     {STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6051     {STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6052     {STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6053     {STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6054     {STATE_TEXTURESTAGE(2, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6055     {STATE_TEXTURESTAGE(2, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6056     {STATE_TEXTURESTAGE(2, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6057     {STATE_TEXTURESTAGE(2, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6058     {STATE_TEXTURESTAGE(2, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6059     {STATE_TEXTURESTAGE(2, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6060     {STATE_TEXTURESTAGE(2, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6061     {STATE_TEXTURESTAGE(2, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6062     {STATE_TEXTURESTAGE(2, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6063     {STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6064     {STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6065     {STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6066     {STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6067     {STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6068     {STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6069     {STATE_TEXTURESTAGE(3, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6070     {STATE_TEXTURESTAGE(3, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6071     {STATE_TEXTURESTAGE(3, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6072     {STATE_TEXTURESTAGE(3, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6073     {STATE_TEXTURESTAGE(3, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6074     {STATE_TEXTURESTAGE(3, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6075     {STATE_TEXTURESTAGE(3, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6076     {STATE_TEXTURESTAGE(3, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6077     {STATE_TEXTURESTAGE(3, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6078     {STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6079     {STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6080     {STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6081     {STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6082     {STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6083     {STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6084     {STATE_TEXTURESTAGE(4, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6085     {STATE_TEXTURESTAGE(4, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6086     {STATE_TEXTURESTAGE(4, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6087     {STATE_TEXTURESTAGE(4, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6088     {STATE_TEXTURESTAGE(4, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6089     {STATE_TEXTURESTAGE(4, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6090     {STATE_TEXTURESTAGE(4, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6091     {STATE_TEXTURESTAGE(4, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6092     {STATE_TEXTURESTAGE(4, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6093     {STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6094     {STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6095     {STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6096     {STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6097     {STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6098     {STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6099     {STATE_TEXTURESTAGE(5, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6100     {STATE_TEXTURESTAGE(5, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6101     {STATE_TEXTURESTAGE(5, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6102     {STATE_TEXTURESTAGE(5, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6103     {STATE_TEXTURESTAGE(5, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6104     {STATE_TEXTURESTAGE(5, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6105     {STATE_TEXTURESTAGE(5, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6106     {STATE_TEXTURESTAGE(5, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6107     {STATE_TEXTURESTAGE(5, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6108     {STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6109     {STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6110     {STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6111     {STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6112     {STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6113     {STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6114     {STATE_TEXTURESTAGE(6, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6115     {STATE_TEXTURESTAGE(6, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6116     {STATE_TEXTURESTAGE(6, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6117     {STATE_TEXTURESTAGE(6, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6118     {STATE_TEXTURESTAGE(6, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6119     {STATE_TEXTURESTAGE(6, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6120     {STATE_TEXTURESTAGE(6, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6121     {STATE_TEXTURESTAGE(6, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6122     {STATE_TEXTURESTAGE(6, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6123     {STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6124     {STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6125     {STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6126     {STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6127     {STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6128     {STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6129     {STATE_TEXTURESTAGE(7, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6130     {STATE_TEXTURESTAGE(7, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6131     {STATE_TEXTURESTAGE(7, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6132     {STATE_TEXTURESTAGE(7, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6133     {STATE_TEXTURESTAGE(7, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6134     {STATE_TEXTURESTAGE(7, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6135     {STATE_TEXTURESTAGE(7, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6136     {STATE_TEXTURESTAGE(7, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6137     {STATE_TEXTURESTAGE(7, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6138     {STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6139     {STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6140     {STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6141     {STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6142     {STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6143     {STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6144     {STATE_SAMPLER(0),                                    { STATE_SAMPLER(0),                                   sampler_texdim          }, WINED3D_GL_EXT_NONE             },
6145     {STATE_SAMPLER(1),                                    { STATE_SAMPLER(1),                                   sampler_texdim          }, WINED3D_GL_EXT_NONE             },
6146     {STATE_SAMPLER(2),                                    { STATE_SAMPLER(2),                                   sampler_texdim          }, WINED3D_GL_EXT_NONE             },
6147     {STATE_SAMPLER(3),                                    { STATE_SAMPLER(3),                                   sampler_texdim          }, WINED3D_GL_EXT_NONE             },
6148     {STATE_SAMPLER(4),                                    { STATE_SAMPLER(4),                                   sampler_texdim          }, WINED3D_GL_EXT_NONE             },
6149     {STATE_SAMPLER(5),                                    { STATE_SAMPLER(5),                                   sampler_texdim          }, WINED3D_GL_EXT_NONE             },
6150     {STATE_SAMPLER(6),                                    { STATE_SAMPLER(6),                                   sampler_texdim          }, WINED3D_GL_EXT_NONE             },
6151     {STATE_SAMPLER(7),                                    { STATE_SAMPLER(7),                                   sampler_texdim          }, WINED3D_GL_EXT_NONE             },
6152     {STATE_PIXELSHADER,                                   { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6153     {STATE_RENDER(WINED3DRS_FOGENABLE),                   { STATE_RENDER(WINED3DRS_FOGENABLE),                  state_arbfp_fog         }, WINED3D_GL_EXT_NONE             },
6154     {STATE_RENDER(WINED3DRS_FOGTABLEMODE),                { STATE_RENDER(WINED3DRS_FOGENABLE),                  state_arbfp_fog         }, WINED3D_GL_EXT_NONE             },
6155     {STATE_RENDER(WINED3DRS_FOGVERTEXMODE),               { STATE_RENDER(WINED3DRS_FOGENABLE),                  state_arbfp_fog         }, WINED3D_GL_EXT_NONE             },
6156     {STATE_RENDER(WINED3DRS_FOGSTART),                    { STATE_RENDER(WINED3DRS_FOGSTART),                   state_fogstartend       }, WINED3D_GL_EXT_NONE             },
6157     {STATE_RENDER(WINED3DRS_FOGEND),                      { STATE_RENDER(WINED3DRS_FOGSTART),                   state_fogstartend       }, WINED3D_GL_EXT_NONE             },
6158     {STATE_RENDER(WINED3DRS_SRGBWRITEENABLE),             { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6159     {STATE_RENDER(WINED3DRS_FOGCOLOR),                    { STATE_RENDER(WINED3DRS_FOGCOLOR),                   state_fogcolor          }, WINED3D_GL_EXT_NONE             },
6160     {STATE_RENDER(WINED3DRS_FOGDENSITY),                  { STATE_RENDER(WINED3DRS_FOGDENSITY),                 state_fogdensity        }, WINED3D_GL_EXT_NONE             },
6161     {STATE_TEXTURESTAGE(0,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(0, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6162     {STATE_TEXTURESTAGE(1,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(1, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6163     {STATE_TEXTURESTAGE(2,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(2, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6164     {STATE_TEXTURESTAGE(3,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(3, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6165     {STATE_TEXTURESTAGE(4,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(4, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6166     {STATE_TEXTURESTAGE(5,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(5, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6167     {STATE_TEXTURESTAGE(6,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(6, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6168     {STATE_TEXTURESTAGE(7,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(7, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6169     {STATE_RENDER(WINED3DRS_SPECULARENABLE),              { STATE_RENDER(WINED3DRS_SPECULARENABLE),             state_arb_specularenable}, WINED3D_GL_EXT_NONE             },
6170     {0 /* Terminate */,                                   { 0,                                                  0                       }, WINED3D_GL_EXT_NONE             },
6171 };
6172
6173 const struct fragment_pipeline arbfp_fragment_pipeline = {
6174     arbfp_enable,
6175     arbfp_get_caps,
6176     arbfp_alloc,
6177     arbfp_free,
6178     shader_arb_color_fixup_supported,
6179     arbfp_fragmentstate_template,
6180     TRUE /* We can disable projected textures */
6181 };
6182
6183 #define GLINFO_LOCATION device->adapter->gl_info
6184
6185 struct arbfp_blit_priv {
6186     GLenum yuy2_rect_shader, yuy2_2d_shader;
6187     GLenum uyvy_rect_shader, uyvy_2d_shader;
6188     GLenum yv12_rect_shader, yv12_2d_shader;
6189 };
6190
6191 static HRESULT arbfp_blit_alloc(IWineD3DDevice *iface) {
6192     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *) iface;
6193     device->blit_priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(struct arbfp_blit_priv));
6194     if(!device->blit_priv) {
6195         ERR("Out of memory\n");
6196         return E_OUTOFMEMORY;
6197     }
6198     return WINED3D_OK;
6199 }
6200
6201 /* Context activation is done by the caller. */
6202 static void arbfp_blit_free(IWineD3DDevice *iface) {
6203     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *) iface;
6204     struct arbfp_blit_priv *priv = device->blit_priv;
6205
6206     ENTER_GL();
6207     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yuy2_rect_shader));
6208     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yuy2_2d_shader));
6209     GL_EXTCALL(glDeleteProgramsARB(1, &priv->uyvy_rect_shader));
6210     GL_EXTCALL(glDeleteProgramsARB(1, &priv->uyvy_2d_shader));
6211     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yv12_rect_shader));
6212     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yv12_2d_shader));
6213     checkGLcall("Delete yuv programs");
6214     LEAVE_GL();
6215
6216     HeapFree(GetProcessHeap(), 0, device->blit_priv);
6217     device->blit_priv = NULL;
6218 }
6219
6220 static BOOL gen_planar_yuv_read(struct wined3d_shader_buffer *buffer, enum yuv_fixup yuv_fixup,
6221         GLenum textype, char *luminance)
6222 {
6223     char chroma;
6224     const char *tex, *texinstr;
6225
6226     if (yuv_fixup == YUV_FIXUP_UYVY) {
6227         chroma = 'x';
6228         *luminance = 'w';
6229     } else {
6230         chroma = 'w';
6231         *luminance = 'x';
6232     }
6233     switch(textype) {
6234         case GL_TEXTURE_2D:             tex = "2D";     texinstr = "TXP"; break;
6235         case GL_TEXTURE_RECTANGLE_ARB:  tex = "RECT";   texinstr = "TEX"; break;
6236         default:
6237             /* This is more tricky than just replacing the texture type - we have to navigate
6238              * properly in the texture to find the correct chroma values
6239              */
6240             FIXME("Implement yuv correction for non-2d, non-rect textures\n");
6241             return FALSE;
6242     }
6243
6244     /* First we have to read the chroma values. This means we need at least two pixels(no filtering),
6245      * or 4 pixels(with filtering). To get the unmodified chromas, we have to rid ourselves of the
6246      * filtering when we sample the texture.
6247      *
6248      * These are the rules for reading the chroma:
6249      *
6250      * Even pixel: Cr
6251      * Even pixel: U
6252      * Odd pixel: V
6253      *
6254      * So we have to get the sampling x position in non-normalized coordinates in integers
6255      */
6256     if(textype != GL_TEXTURE_RECTANGLE_ARB) {
6257         shader_addline(buffer, "MUL texcrd.xy, fragment.texcoord[0], size.x;\n");
6258         shader_addline(buffer, "MOV texcrd.w, size.x;\n");
6259     } else {
6260         shader_addline(buffer, "MOV texcrd, fragment.texcoord[0];\n");
6261     }
6262     /* We must not allow filtering between pixel x and x+1, this would mix U and V
6263      * Vertical filtering is ok. However, bear in mind that the pixel center is at
6264      * 0.5, so add 0.5.
6265      */
6266     shader_addline(buffer, "FLR texcrd.x, texcrd.x;\n");
6267     shader_addline(buffer, "ADD texcrd.x, texcrd.x, coef.y;\n");
6268
6269     /* Divide the x coordinate by 0.5 and get the fraction. This gives 0.25 and 0.75 for the
6270      * even and odd pixels respectively
6271      */
6272     shader_addline(buffer, "MUL texcrd2, texcrd, coef.y;\n");
6273     shader_addline(buffer, "FRC texcrd2, texcrd2;\n");
6274
6275     /* Sample Pixel 1 */
6276     shader_addline(buffer, "%s luminance, texcrd, texture[0], %s;\n", texinstr, tex);
6277
6278     /* Put the value into either of the chroma values */
6279     shader_addline(buffer, "SGE temp.x, texcrd2.x, coef.y;\n");
6280     shader_addline(buffer, "MUL chroma.x, luminance.%c, temp.x;\n", chroma);
6281     shader_addline(buffer, "SLT temp.x, texcrd2.x, coef.y;\n");
6282     shader_addline(buffer, "MUL chroma.y, luminance.%c, temp.x;\n", chroma);
6283
6284     /* Sample pixel 2. If we read an even pixel(SLT above returned 1), sample
6285      * the pixel right to the current one. Otherwise, sample the left pixel.
6286      * Bias and scale the SLT result to -1;1 and add it to the texcrd.x.
6287      */
6288     shader_addline(buffer, "MAD temp.x, temp.x, coef.z, -coef.x;\n");
6289     shader_addline(buffer, "ADD texcrd.x, texcrd, temp.x;\n");
6290     shader_addline(buffer, "%s luminance, texcrd, texture[0], %s;\n", texinstr, tex);
6291
6292     /* Put the value into the other chroma */
6293     shader_addline(buffer, "SGE temp.x, texcrd2.x, coef.y;\n");
6294     shader_addline(buffer, "MAD chroma.y, luminance.%c, temp.x, chroma.y;\n", chroma);
6295     shader_addline(buffer, "SLT temp.x, texcrd2.x, coef.y;\n");
6296     shader_addline(buffer, "MAD chroma.x, luminance.%c, temp.x, chroma.x;\n", chroma);
6297
6298     /* TODO: If filtering is enabled, sample a 2nd pair of pixels left or right of
6299      * the current one and lerp the two U and V values
6300      */
6301
6302     /* This gives the correctly filtered luminance value */
6303     shader_addline(buffer, "TEX luminance, fragment.texcoord[0], texture[0], %s;\n", tex);
6304
6305     return TRUE;
6306 }
6307
6308 static BOOL gen_yv12_read(struct wined3d_shader_buffer *buffer, GLenum textype, char *luminance)
6309 {
6310     const char *tex;
6311
6312     switch(textype) {
6313         case GL_TEXTURE_2D:             tex = "2D";     break;
6314         case GL_TEXTURE_RECTANGLE_ARB:  tex = "RECT";   break;
6315         default:
6316             FIXME("Implement yv12 correction for non-2d, non-rect textures\n");
6317             return FALSE;
6318     }
6319
6320     /* YV12 surfaces contain a WxH sized luminance plane, followed by a (W/2)x(H/2)
6321      * V and a (W/2)x(H/2) U plane, each with 8 bit per pixel. So the effective
6322      * bitdepth is 12 bits per pixel. Since the U and V planes have only half the
6323      * pitch of the luminance plane, the packing into the gl texture is a bit
6324      * unfortunate. If the whole texture is interpreted as luminance data it looks
6325      * approximately like this:
6326      *
6327      *        +----------------------------------+----
6328      *        |                                  |
6329      *        |                                  |
6330      *        |                                  |
6331      *        |                                  |
6332      *        |                                  |   2
6333      *        |            LUMINANCE             |   -
6334      *        |                                  |   3
6335      *        |                                  |
6336      *        |                                  |
6337      *        |                                  |
6338      *        |                                  |
6339      *        +----------------+-----------------+----
6340      *        |                |                 |
6341      *        |  U even rows   |  U odd rows     |
6342      *        |                |                 |   1
6343      *        +----------------+------------------   -
6344      *        |                |                 |   3
6345      *        |  V even rows   |  V odd rows     |
6346      *        |                |                 |
6347      *        +----------------+-----------------+----
6348      *        |                |                 |
6349      *        |     0.5        |       0.5       |
6350      *
6351      * So it appears as if there are 4 chroma images, but in fact the odd rows
6352      * in the chroma images are in the same row as the even ones. So its is
6353      * kinda tricky to read
6354      *
6355      * When reading from rectangle textures, keep in mind that the input y coordinates
6356      * go from 0 to d3d_height, whereas the opengl texture height is 1.5 * d3d_height
6357      */
6358     shader_addline(buffer, "PARAM yv12_coef = {%f, %f, %f, %f};\n",
6359             2.0f / 3.0f, 1.0f / 6.0f, (2.0f / 3.0f) + (1.0f / 6.0f), 1.0f / 3.0f);
6360
6361     shader_addline(buffer, "MOV texcrd, fragment.texcoord[0];\n");
6362     /* the chroma planes have only half the width */
6363     shader_addline(buffer, "MUL texcrd.x, texcrd.x, coef.y;\n");
6364
6365     /* The first value is between 2/3 and 5/6th of the texture's height, so scale+bias
6366      * the coordinate. Also read the right side of the image when reading odd lines
6367      *
6368      * Don't forget to clamp the y values in into the range, otherwise we'll get filtering
6369      * bleeding
6370      */
6371     if(textype == GL_TEXTURE_2D) {
6372
6373         shader_addline(buffer, "RCP chroma.w, size.y;\n");
6374
6375         shader_addline(buffer, "MUL texcrd2.y, texcrd.y, size.y;\n");
6376
6377         shader_addline(buffer, "FLR texcrd2.y, texcrd2.y;\n");
6378         shader_addline(buffer, "MAD texcrd.y, texcrd.y, yv12_coef.y, yv12_coef.x;\n");
6379
6380         /* Read odd lines from the right side(add size * 0.5 to the x coordinate */
6381         shader_addline(buffer, "ADD texcrd2.x, texcrd2.y, yv12_coef.y;\n"); /* To avoid 0.5 == 0.5 comparisons */
6382         shader_addline(buffer, "FRC texcrd2.x, texcrd2.x;\n");
6383         shader_addline(buffer, "SGE texcrd2.x, texcrd2.x, coef.y;\n");
6384         shader_addline(buffer, "MAD texcrd.x, texcrd2.x, coef.y, texcrd.x;\n");
6385
6386         /* clamp, keep the half pixel origin in mind */
6387         shader_addline(buffer, "MAD temp.y, coef.y, chroma.w, yv12_coef.x;\n");
6388         shader_addline(buffer, "MAX texcrd.y, temp.y, texcrd.y;\n");
6389         shader_addline(buffer, "MAD temp.y, -coef.y, chroma.w, yv12_coef.z;\n");
6390         shader_addline(buffer, "MIN texcrd.y, temp.y, texcrd.y;\n");
6391     } else {
6392         /* Read from [size - size+size/4] */
6393         shader_addline(buffer, "FLR texcrd.y, texcrd.y;\n");
6394         shader_addline(buffer, "MAD texcrd.y, texcrd.y, coef.w, size.y;\n");
6395
6396         /* Read odd lines from the right side(add size * 0.5 to the x coordinate */
6397         shader_addline(buffer, "ADD texcrd2.x, texcrd.y, yv12_coef.y;\n"); /* To avoid 0.5 == 0.5 comparisons */
6398         shader_addline(buffer, "FRC texcrd2.x, texcrd2.x;\n");
6399         shader_addline(buffer, "SGE texcrd2.x, texcrd2.x, coef.y;\n");
6400         shader_addline(buffer, "MUL texcrd2.x, texcrd2.x, size.x;\n");
6401         shader_addline(buffer, "MAD texcrd.x, texcrd2.x, coef.y, texcrd.x;\n");
6402
6403         /* Make sure to read exactly from the pixel center */
6404         shader_addline(buffer, "FLR texcrd.y, texcrd.y;\n");
6405         shader_addline(buffer, "ADD texcrd.y, texcrd.y, coef.y;\n");
6406
6407         /* Clamp */
6408         shader_addline(buffer, "MAD temp.y, size.y, coef.w, size.y;\n");
6409         shader_addline(buffer, "ADD temp.y, temp.y, -coef.y;\n");
6410         shader_addline(buffer, "MIN texcrd.y, temp.y, texcrd.y;\n");
6411         shader_addline(buffer, "ADD temp.y, size.y, -coef.y;\n");
6412         shader_addline(buffer, "MAX texcrd.y, temp.y, texcrd.y;\n");
6413     }
6414     /* Read the texture, put the result into the output register */
6415     shader_addline(buffer, "TEX temp, texcrd, texture[0], %s;\n", tex);
6416     shader_addline(buffer, "MOV chroma.x, temp.w;\n");
6417
6418     /* The other chroma value is 1/6th of the texture lower, from 5/6th to 6/6th
6419      * No need to clamp because we're just reusing the already clamped value from above
6420      */
6421     if(textype == GL_TEXTURE_2D) {
6422         shader_addline(buffer, "ADD texcrd.y, texcrd.y, yv12_coef.y;\n");
6423     } else {
6424         shader_addline(buffer, "MAD texcrd.y, size.y, coef.w, texcrd.y;\n");
6425     }
6426     shader_addline(buffer, "TEX temp, texcrd, texture[0], %s;\n", tex);
6427     shader_addline(buffer, "MOV chroma.y, temp.w;\n");
6428
6429     /* Sample the luminance value. It is in the top 2/3rd of the texture, so scale the y coordinate.
6430      * Clamp the y coordinate to prevent the chroma values from bleeding into the sampled luminance
6431      * values due to filtering
6432      */
6433     shader_addline(buffer, "MOV texcrd, fragment.texcoord[0];\n");
6434     if(textype == GL_TEXTURE_2D) {
6435         /* Multiply the y coordinate by 2/3 and clamp it */
6436         shader_addline(buffer, "MUL texcrd.y, texcrd.y, yv12_coef.x;\n");
6437         shader_addline(buffer, "MAD temp.y, -coef.y, chroma.w, yv12_coef.x;\n");
6438         shader_addline(buffer, "MIN texcrd.y, temp.y, texcrd.y;\n");
6439         shader_addline(buffer, "TEX luminance, texcrd, texture[0], %s;\n", tex);
6440     } else {
6441         /* Reading from texture_rectangles is pretty straightforward, just use the unmodified
6442          * texture coordinate. It is still a good idea to clamp it though, since the opengl texture
6443          * is bigger
6444          */
6445         shader_addline(buffer, "ADD temp.x, size.y, -coef.y;\n");
6446         shader_addline(buffer, "MIN texcrd.y, texcrd.y, size.x;\n");
6447         shader_addline(buffer, "TEX luminance, texcrd, texture[0], %s;\n", tex);
6448     }
6449     *luminance = 'a';
6450
6451     return TRUE;
6452 }
6453
6454 /* Context activation is done by the caller. */
6455 static GLuint gen_yuv_shader(IWineD3DDeviceImpl *device, enum yuv_fixup yuv_fixup, GLenum textype)
6456 {
6457     GLenum shader;
6458     struct wined3d_shader_buffer buffer;
6459     char luminance_component;
6460     struct arbfp_blit_priv *priv = device->blit_priv;
6461     GLint pos;
6462
6463     /* Shader header */
6464     if (!shader_buffer_init(&buffer))
6465     {
6466         ERR("Failed to initialize shader buffer.\n");
6467         return 0;
6468     }
6469
6470     ENTER_GL();
6471     GL_EXTCALL(glGenProgramsARB(1, &shader));
6472     checkGLcall("GL_EXTCALL(glGenProgramsARB(1, &shader))");
6473     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader));
6474     checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader)");
6475     LEAVE_GL();
6476     if(!shader) {
6477         shader_buffer_free(&buffer);
6478         return 0;
6479     }
6480
6481     /* The YUY2 and UYVY formats contain two pixels packed into a 32 bit macropixel,
6482      * giving effectively 16 bit per pixel. The color consists of a luminance(Y) and
6483      * two chroma(U and V) values. Each macropixel has two luminance values, one for
6484      * each single pixel it contains, and one U and one V value shared between both
6485      * pixels.
6486      *
6487      * The data is loaded into an A8L8 texture. With YUY2, the luminance component
6488      * contains the luminance and alpha the chroma. With UYVY it is vice versa. Thus
6489      * take the format into account when generating the read swizzles
6490      *
6491      * Reading the Y value is straightforward - just sample the texture. The hardware
6492      * takes care of filtering in the horizontal and vertical direction.
6493      *
6494      * Reading the U and V values is harder. We have to avoid filtering horizontally,
6495      * because that would mix the U and V values of one pixel or two adjacent pixels.
6496      * Thus floor the texture coordinate and add 0.5 to get an unfiltered read,
6497      * regardless of the filtering setting. Vertical filtering works automatically
6498      * though - the U and V values of two rows are mixed nicely.
6499      *
6500      * Appart of avoiding filtering issues, the code has to know which value it just
6501      * read, and where it can find the other one. To determine this, it checks if
6502      * it sampled an even or odd pixel, and shifts the 2nd read accordingly.
6503      *
6504      * Handling horizontal filtering of U and V values requires reading a 2nd pair
6505      * of pixels, extracting U and V and mixing them. This is not implemented yet.
6506      *
6507      * An alternative implementation idea is to load the texture as A8R8G8B8 texture,
6508      * with width / 2. This way one read gives all 3 values, finding U and V is easy
6509      * in an unfiltered situation. Finding the luminance on the other hand requires
6510      * finding out if it is an odd or even pixel. The real drawback of this approach
6511      * is filtering. This would have to be emulated completely in the shader, reading
6512      * up two 2 packed pixels in up to 2 rows and interpolating both horizontally and
6513      * vertically. Beyond that it would require adjustments to the texture handling
6514      * code to deal with the width scaling
6515      */
6516     shader_addline(&buffer, "!!ARBfp1.0\n");
6517     shader_addline(&buffer, "TEMP luminance;\n");
6518     shader_addline(&buffer, "TEMP temp;\n");
6519     shader_addline(&buffer, "TEMP chroma;\n");
6520     shader_addline(&buffer, "TEMP texcrd;\n");
6521     shader_addline(&buffer, "TEMP texcrd2;\n");
6522     shader_addline(&buffer, "PARAM coef = {1.0, 0.5, 2.0, 0.25};\n");
6523     shader_addline(&buffer, "PARAM yuv_coef = {1.403, 0.344, 0.714, 1.770};\n");
6524     shader_addline(&buffer, "PARAM size = program.local[0];\n");
6525
6526     switch (yuv_fixup)
6527     {
6528         case YUV_FIXUP_UYVY:
6529         case YUV_FIXUP_YUY2:
6530             if (!gen_planar_yuv_read(&buffer, yuv_fixup, textype, &luminance_component))
6531             {
6532                 shader_buffer_free(&buffer);
6533                 return 0;
6534             }
6535             break;
6536
6537         case YUV_FIXUP_YV12:
6538             if (!gen_yv12_read(&buffer, textype, &luminance_component))
6539             {
6540                 shader_buffer_free(&buffer);
6541                 return 0;
6542             }
6543             break;
6544
6545         default:
6546             FIXME("Unsupported YUV fixup %#x\n", yuv_fixup);
6547             shader_buffer_free(&buffer);
6548             return 0;
6549     }
6550
6551     /* Calculate the final result. Formula is taken from
6552      * http://www.fourcc.org/fccyvrgb.php. Note that the chroma
6553      * ranges from -0.5 to 0.5
6554      */
6555     shader_addline(&buffer, "SUB chroma.xy, chroma, coef.y;\n");
6556
6557     shader_addline(&buffer, "MAD result.color.x, chroma.x, yuv_coef.x, luminance.%c;\n", luminance_component);
6558     shader_addline(&buffer, "MAD temp.x, -chroma.y, yuv_coef.y, luminance.%c;\n", luminance_component);
6559     shader_addline(&buffer, "MAD result.color.y, -chroma.x, yuv_coef.z, temp.x;\n");
6560     shader_addline(&buffer, "MAD result.color.z, chroma.y, yuv_coef.w, luminance.%c;\n", luminance_component);
6561     shader_addline(&buffer, "END\n");
6562
6563     ENTER_GL();
6564     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
6565             strlen(buffer.buffer), buffer.buffer));
6566     checkGLcall("glProgramStringARB()");
6567
6568     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
6569     if (pos != -1)
6570     {
6571         FIXME("Fragment program error at position %d: %s\n", pos,
6572               debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
6573     }
6574     else
6575     {
6576         GLint native;
6577
6578         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
6579         checkGLcall("glGetProgramivARB()");
6580         if (!native) WARN("Program exceeds native resource limits.\n");
6581     }
6582
6583     shader_buffer_free(&buffer);
6584     LEAVE_GL();
6585
6586     switch (yuv_fixup)
6587     {
6588         case YUV_FIXUP_YUY2:
6589             if (textype == GL_TEXTURE_RECTANGLE_ARB) priv->yuy2_rect_shader = shader;
6590             else priv->yuy2_2d_shader = shader;
6591             break;
6592
6593         case YUV_FIXUP_UYVY:
6594             if (textype == GL_TEXTURE_RECTANGLE_ARB) priv->uyvy_rect_shader = shader;
6595             else priv->uyvy_2d_shader = shader;
6596             break;
6597
6598         case YUV_FIXUP_YV12:
6599             if (textype == GL_TEXTURE_RECTANGLE_ARB) priv->yv12_rect_shader = shader;
6600             else priv->yv12_2d_shader = shader;
6601             break;
6602     }
6603
6604     return shader;
6605 }
6606
6607 /* Context activation is done by the caller. */
6608 static HRESULT arbfp_blit_set(IWineD3DDevice *iface, const struct GlPixelFormatDesc *format_desc,
6609         GLenum textype, UINT width, UINT height)
6610 {
6611     GLenum shader;
6612     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *) iface;
6613     float size[4] = {width, height, 1, 1};
6614     struct arbfp_blit_priv *priv = device->blit_priv;
6615     enum yuv_fixup yuv_fixup;
6616
6617     if (!is_yuv_fixup(format_desc->color_fixup))
6618     {
6619         TRACE("Fixup:\n");
6620         dump_color_fixup_desc(format_desc->color_fixup);
6621         /* Don't bother setting up a shader for unconverted formats */
6622         ENTER_GL();
6623         glEnable(textype);
6624         checkGLcall("glEnable(textype)");
6625         LEAVE_GL();
6626         return WINED3D_OK;
6627     }
6628
6629     yuv_fixup = get_yuv_fixup(format_desc->color_fixup);
6630
6631     switch(yuv_fixup)
6632     {
6633         case YUV_FIXUP_YUY2:
6634             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->yuy2_rect_shader : priv->yuy2_2d_shader;
6635             break;
6636
6637         case YUV_FIXUP_UYVY:
6638             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->uyvy_rect_shader : priv->uyvy_2d_shader;
6639             break;
6640
6641         case YUV_FIXUP_YV12:
6642             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->yv12_rect_shader : priv->yv12_2d_shader;
6643             break;
6644
6645         default:
6646             FIXME("Unsupported YUV fixup %#x, not setting a shader\n", yuv_fixup);
6647             ENTER_GL();
6648             glEnable(textype);
6649             checkGLcall("glEnable(textype)");
6650             LEAVE_GL();
6651             return E_NOTIMPL;
6652     }
6653
6654     if (!shader) shader = gen_yuv_shader(device, yuv_fixup, textype);
6655
6656     ENTER_GL();
6657     glEnable(GL_FRAGMENT_PROGRAM_ARB);
6658     checkGLcall("glEnable(GL_FRAGMENT_PROGRAM_ARB)");
6659     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader));
6660     checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader)");
6661     GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, 0, size));
6662     checkGLcall("glProgramLocalParameter4fvARB");
6663     LEAVE_GL();
6664
6665     return WINED3D_OK;
6666 }
6667
6668 /* Context activation is done by the caller. */
6669 static void arbfp_blit_unset(IWineD3DDevice *iface) {
6670     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *) iface;
6671     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
6672
6673     ENTER_GL();
6674     glDisable(GL_FRAGMENT_PROGRAM_ARB);
6675     checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
6676     glDisable(GL_TEXTURE_2D);
6677     checkGLcall("glDisable(GL_TEXTURE_2D)");
6678     if (gl_info->supported[ARB_TEXTURE_CUBE_MAP])
6679     {
6680         glDisable(GL_TEXTURE_CUBE_MAP_ARB);
6681         checkGLcall("glDisable(GL_TEXTURE_CUBE_MAP_ARB)");
6682     }
6683     if (gl_info->supported[ARB_TEXTURE_RECTANGLE])
6684     {
6685         glDisable(GL_TEXTURE_RECTANGLE_ARB);
6686         checkGLcall("glDisable(GL_TEXTURE_RECTANGLE_ARB)");
6687     }
6688     LEAVE_GL();
6689 }
6690
6691 static BOOL arbfp_blit_color_fixup_supported(struct color_fixup_desc fixup)
6692 {
6693     enum yuv_fixup yuv_fixup;
6694
6695     if (TRACE_ON(d3d_shader) && TRACE_ON(d3d))
6696     {
6697         TRACE("Checking support for fixup:\n");
6698         dump_color_fixup_desc(fixup);
6699     }
6700
6701     if (is_identity_fixup(fixup))
6702     {
6703         TRACE("[OK]\n");
6704         return TRUE;
6705     }
6706
6707     /* We only support YUV conversions. */
6708     if (!is_yuv_fixup(fixup))
6709     {
6710         TRACE("[FAILED]\n");
6711         return FALSE;
6712     }
6713
6714     yuv_fixup = get_yuv_fixup(fixup);
6715     switch(yuv_fixup)
6716     {
6717         case YUV_FIXUP_YUY2:
6718         case YUV_FIXUP_UYVY:
6719         case YUV_FIXUP_YV12:
6720             TRACE("[OK]\n");
6721             return TRUE;
6722
6723         default:
6724             FIXME("Unsupported YUV fixup %#x\n", yuv_fixup);
6725             TRACE("[FAILED]\n");
6726             return FALSE;
6727     }
6728 }
6729
6730 const struct blit_shader arbfp_blit = {
6731     arbfp_blit_alloc,
6732     arbfp_blit_free,
6733     arbfp_blit_set,
6734     arbfp_blit_unset,
6735     arbfp_blit_color_fixup_supported,
6736 };
6737
6738 #undef GLINFO_LOCATION