dlls/wined3d/pixelshader.c

   1 /*
   2  * shaders implementation
   3  *
   4  * Copyright 2002-2003 Jason Edmeades
   5  * Copyright 2002-2003 Raphael Junqueira
   6  * Copyright 2004 Christian Costa
   7  * Copyright 2005 Oliver Stieber
   8  * Copyright 2006 Ivan Gyurdiev
   9  * Copyright 2007-2008 Stefan Dösinger for CodeWeavers
  10  *
  11  * This library is free software; you can redistribute it and/or
  12  * modify it under the terms of the GNU Lesser General Public
  13  * License as published by the Free Software Foundation; either
  14  * version 2.1 of the License, or (at your option) any later version.
  15  *
  16  * This library is distributed in the hope that it will be useful,
  17  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  18  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  19  * Lesser General Public License for more details.
  20  *
  21  * You should have received a copy of the GNU Lesser General Public
  22  * License along with this library; if not, write to the Free Software
  23  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301, USA
  24  */
  25
  26 #include "config.h"
  27
  28 #include <math.h>
  29 #include <stdio.h>
  30
  31 #include "wined3d_private.h"
  32
  33 WINE_DEFAULT_DEBUG_CHANNEL(d3d_shader);
  34
  35 #define GLINFO_LOCATION ((IWineD3DDeviceImpl *) This->baseShader.device)->adapter->gl_info
  36
  37 static HRESULT  WINAPI IWineD3DPixelShaderImpl_QueryInterface(IWineD3DPixelShader *iface, REFIID riid, LPVOID *ppobj) {
  38     TRACE("iface %p, riid %s, ppobj %p\n", iface, debugstr_guid(riid), ppobj);
  39
  40     if (IsEqualGUID(riid, &IID_IWineD3DPixelShader)
  41             || IsEqualGUID(riid, &IID_IWineD3DBaseShader)
  42             || IsEqualGUID(riid, &IID_IWineD3DBase)
  43             || IsEqualGUID(riid, &IID_IUnknown))
  44     {
  45         IUnknown_AddRef(iface);
  46         *ppobj = iface;
  47         return S_OK;
  48     }
  49
  50     WARN("%s not implemented, returning E_NOINTERFACE\n", debugstr_guid(riid));
  51
  52     *ppobj = NULL;
  53     return E_NOINTERFACE;
  54 }
  55
  56 static ULONG  WINAPI IWineD3DPixelShaderImpl_AddRef(IWineD3DPixelShader *iface) {
  57     IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)iface;
  58     ULONG refcount = InterlockedIncrement(&This->baseShader.ref);
  59
  60     TRACE("%p increasing refcount to %u\n", This, refcount);
  61
  62     return refcount;
  63 }
  64
  65 static ULONG  WINAPI IWineD3DPixelShaderImpl_Release(IWineD3DPixelShader *iface) {
  66     IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)iface;
  67     ULONG refcount = InterlockedDecrement(&This->baseShader.ref);
  68
  69     TRACE("%p decreasing refcount to %u\n", This, refcount);
  70
  71     if (!refcount)
  72     {
  73         shader_cleanup((IWineD3DBaseShader *)iface);
  74         HeapFree(GetProcessHeap(), 0, This);
  75     }
  76
  77     return refcount;
  78 }
  79
  80 /* *******************************************
  81    IWineD3DPixelShader IWineD3DPixelShader parts follow
  82    ******************************************* */
  83
  84 static HRESULT  WINAPI IWineD3DPixelShaderImpl_GetParent(IWineD3DPixelShader *iface, IUnknown** parent){
  85     IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)iface;
  86
  87     *parent = This->parent;
  88     IUnknown_AddRef(*parent);
  89     TRACE("(%p) : returning %p\n", This, *parent);
  90     return WINED3D_OK;
  91 }
  92
  93 static HRESULT  WINAPI IWineD3DPixelShaderImpl_GetDevice(IWineD3DPixelShader* iface, IWineD3DDevice **pDevice){
  94     IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)iface;
  95     IWineD3DDevice_AddRef(This->baseShader.device);
  96     *pDevice = This->baseShader.device;
  97     TRACE("(%p) returning %p\n", This, *pDevice);
  98     return WINED3D_OK;
  99 }
 100
 101
 102 static HRESULT  WINAPI IWineD3DPixelShaderImpl_GetFunction(IWineD3DPixelShader* impl, VOID* pData, UINT* pSizeOfData) {
 103   IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)impl;
 104   TRACE("(%p) : pData(%p), pSizeOfData(%p)\n", This, pData, pSizeOfData);
 105
 106   if (NULL == pData) {
 107     *pSizeOfData = This->baseShader.functionLength;
 108     return WINED3D_OK;
 109   }
 110   if (*pSizeOfData < This->baseShader.functionLength) {
 111     /* MSDN claims (for d3d8 at least) that if *pSizeOfData is smaller
 112      * than the required size we should write the required size and
 113      * return D3DERR_MOREDATA. That's not actually true. */
 114     return WINED3DERR_INVALIDCALL;
 115   }
 116
 117   TRACE("(%p) : GetFunction copying to %p\n", This, pData);
 118   memcpy(pData, This->baseShader.function, This->baseShader.functionLength);
 119
 120   return WINED3D_OK;
 121 }
 122
 123 static void pshader_set_limits(IWineD3DPixelShaderImpl *This)
 124 {
 125     DWORD shader_version = WINED3D_SHADER_VERSION(This->baseShader.reg_maps.shader_version.major,
 126             This->baseShader.reg_maps.shader_version.minor);
 127
 128     This->baseShader.limits.attributes = 0;
 129     This->baseShader.limits.address = 0;
 130     This->baseShader.limits.packed_output = 0;
 131
 132     switch (shader_version)
 133     {
 134         case WINED3D_SHADER_VERSION(1,0):
 135         case WINED3D_SHADER_VERSION(1,1):
 136         case WINED3D_SHADER_VERSION(1,2):
 137         case WINED3D_SHADER_VERSION(1,3):
 138             This->baseShader.limits.temporary = 2;
 139             This->baseShader.limits.constant_float = 8;
 140             This->baseShader.limits.constant_int = 0;
 141             This->baseShader.limits.constant_bool = 0;
 142             This->baseShader.limits.texcoord = 4;
 143             This->baseShader.limits.sampler = 4;
 144             This->baseShader.limits.packed_input = 0;
 145             This->baseShader.limits.label = 0;
 146             break;
 147
 148         case WINED3D_SHADER_VERSION(1,4):
 149             This->baseShader.limits.temporary = 6;
 150             This->baseShader.limits.constant_float = 8;
 151             This->baseShader.limits.constant_int = 0;
 152             This->baseShader.limits.constant_bool = 0;
 153             This->baseShader.limits.texcoord = 6;
 154             This->baseShader.limits.sampler = 6;
 155             This->baseShader.limits.packed_input = 0;
 156             This->baseShader.limits.label = 0;
 157             break;
 158
 159         /* FIXME: temporaries must match D3DPSHADERCAPS2_0.NumTemps */
 160         case WINED3D_SHADER_VERSION(2,0):
 161             This->baseShader.limits.temporary = 32;
 162             This->baseShader.limits.constant_float = 32;
 163             This->baseShader.limits.constant_int = 16;
 164             This->baseShader.limits.constant_bool = 16;
 165             This->baseShader.limits.texcoord = 8;
 166             This->baseShader.limits.sampler = 16;
 167             This->baseShader.limits.packed_input = 0;
 168             break;
 169
 170         case WINED3D_SHADER_VERSION(2,1):
 171             This->baseShader.limits.temporary = 32;
 172             This->baseShader.limits.constant_float = 32;
 173             This->baseShader.limits.constant_int = 16;
 174             This->baseShader.limits.constant_bool = 16;
 175             This->baseShader.limits.texcoord = 8;
 176             This->baseShader.limits.sampler = 16;
 177             This->baseShader.limits.packed_input = 0;
 178             This->baseShader.limits.label = 16;
 179             break;
 180
 181         case WINED3D_SHADER_VERSION(3,0):
 182             This->baseShader.limits.temporary = 32;
 183             This->baseShader.limits.constant_float = 224;
 184             This->baseShader.limits.constant_int = 16;
 185             This->baseShader.limits.constant_bool = 16;
 186             This->baseShader.limits.texcoord = 0;
 187             This->baseShader.limits.sampler = 16;
 188             This->baseShader.limits.packed_input = 12;
 189             This->baseShader.limits.label = 16; /* FIXME: 2048 */
 190             break;
 191
 192         default:
 193             This->baseShader.limits.temporary = 32;
 194             This->baseShader.limits.constant_float = 32;
 195             This->baseShader.limits.constant_int = 16;
 196             This->baseShader.limits.constant_bool = 16;
 197             This->baseShader.limits.texcoord = 8;
 198             This->baseShader.limits.sampler = 16;
 199             This->baseShader.limits.packed_input = 0;
 200             This->baseShader.limits.label = 0;
 201             FIXME("Unrecognized pixel shader version %u.%u\n",
 202                     This->baseShader.reg_maps.shader_version.major,
 203                     This->baseShader.reg_maps.shader_version.minor);
 204     }
 205 }
 206
 207 static HRESULT WINAPI IWineD3DPixelShaderImpl_SetFunction(IWineD3DPixelShader *iface, CONST DWORD *pFunction) {
 208
 209     IWineD3DPixelShaderImpl *This =(IWineD3DPixelShaderImpl *)iface;
 210     unsigned int i, highest_reg_used = 0, num_regs_used = 0;
 211     shader_reg_maps *reg_maps = &This->baseShader.reg_maps;
 212     const struct wined3d_shader_frontend *fe;
 213     HRESULT hr;
 214
 215     TRACE("(%p) : pFunction %p\n", iface, pFunction);
 216
 217     fe = shader_select_frontend(*pFunction);
 218     if (!fe)
 219     {
 220         FIXME("Unable to find frontend for shader.\n");
 221         return WINED3DERR_INVALIDCALL;
 222     }
 223     This->baseShader.frontend = fe;
 224     This->baseShader.frontend_data = fe->shader_init(pFunction);
 225     if (!This->baseShader.frontend_data)
 226     {
 227         FIXME("Failed to initialize frontend.\n");
 228         return WINED3DERR_INVALIDCALL;
 229     }
 230
 231     /* First pass: trace shader */
 232     if (TRACE_ON(d3d_shader)) shader_trace_init(fe, This->baseShader.frontend_data, pFunction);
 233
 234     /* Initialize immediate constant lists */
 235     list_init(&This->baseShader.constantsF);
 236     list_init(&This->baseShader.constantsB);
 237     list_init(&This->baseShader.constantsI);
 238
 239     /* Second pass: figure out which registers are used, what the semantics are, etc.. */
 240     hr = shader_get_registers_used((IWineD3DBaseShader *)This, fe, reg_maps, This->semantics_in, NULL, pFunction,
 241                                     GL_LIMITS(pshader_constantsF));
 242     if (FAILED(hr)) return hr;
 243
 244     pshader_set_limits(This);
 245
 246     for (i = 0; i < MAX_REG_INPUT; ++i)
 247     {
 248         if (This->input_reg_used[i])
 249         {
 250             ++num_regs_used;
 251             highest_reg_used = i;
 252         }
 253     }
 254
 255     /* Don't do any register mapping magic if it is not needed, or if we can't
 256      * achieve anything anyway */
 257     if (highest_reg_used < (GL_LIMITS(glsl_varyings) / 4)
 258             || num_regs_used > (GL_LIMITS(glsl_varyings) / 4))
 259     {
 260         if (num_regs_used > (GL_LIMITS(glsl_varyings) / 4))
 261         {
 262             /* This happens with relative addressing. The input mapper function
 263              * warns about this if the higher registers are declared too, so
 264              * don't write a FIXME here */
 265             WARN("More varying registers used than supported\n");
 266         }
 267
 268         for (i = 0; i < MAX_REG_INPUT; ++i)
 269         {
 270             This->input_reg_map[i] = i;
 271         }
 272
 273         This->declared_in_count = highest_reg_used + 1;
 274     }
 275     else
 276     {
 277         This->declared_in_count = 0;
 278         for (i = 0; i < MAX_REG_INPUT; ++i)
 279         {
 280             if (This->input_reg_used[i]) This->input_reg_map[i] = This->declared_in_count++;
 281             else This->input_reg_map[i] = ~0U;
 282         }
 283     }
 284
 285     This->baseShader.load_local_constsF = FALSE;
 286
 287     TRACE("(%p) : Copying the function\n", This);
 288
 289     This->baseShader.function = HeapAlloc(GetProcessHeap(), 0, This->baseShader.functionLength);
 290     if (!This->baseShader.function) return E_OUTOFMEMORY;
 291     memcpy(This->baseShader.function, pFunction, This->baseShader.functionLength);
 292
 293     return WINED3D_OK;
 294 }
 295
 296 static void pixelshader_update_samplers(struct shader_reg_maps *reg_maps, IWineD3DBaseTexture * const *textures)
 297 {
 298     WINED3DSAMPLER_TEXTURE_TYPE *sampler_type = reg_maps->sampler_type;
 299     unsigned int i;
 300
 301     if (reg_maps->shader_version.major != 1) return;
 302
 303     for (i = 0; i < max(MAX_FRAGMENT_SAMPLERS, MAX_VERTEX_SAMPLERS); ++i)
 304     {
 305         /* We don't sample from this sampler */
 306         if (!sampler_type[i]) continue;
 307
 308         if (!textures[i])
 309         {
 310             ERR("No texture bound to sampler %u, using 2D\n", i);
 311             sampler_type[i] = WINED3DSTT_2D;
 312             continue;
 313         }
 314
 315         switch (IWineD3DBaseTexture_GetTextureDimensions(textures[i]))
 316         {
 317             case GL_TEXTURE_RECTANGLE_ARB:
 318             case GL_TEXTURE_2D:
 319                 /* We have to select between texture rectangles and 2D textures later because 2.0 and
 320                  * 3.0 shaders only have WINED3DSTT_2D as well */
 321                 sampler_type[i] = WINED3DSTT_2D;
 322                 break;
 323
 324             case GL_TEXTURE_3D:
 325                 sampler_type[i] = WINED3DSTT_VOLUME;
 326                 break;
 327
 328             case GL_TEXTURE_CUBE_MAP_ARB:
 329                 sampler_type[i] = WINED3DSTT_CUBE;
 330                 break;
 331
 332             default:
 333                 FIXME("Unrecognized texture type %#x, using 2D\n",
 334                         IWineD3DBaseTexture_GetTextureDimensions(textures[i]));
 335                 sampler_type[i] = WINED3DSTT_2D;
 336         }
 337     }
 338 }
 339
 340 static GLuint pixelshader_compile(IWineD3DPixelShaderImpl *This, const struct ps_compile_args *args)
 341 {
 342     CONST DWORD *function = This->baseShader.function;
 343     GLuint retval;
 344     SHADER_BUFFER buffer;
 345     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *) This->baseShader.device;
 346
 347     TRACE("(%p) : function %p\n", This, function);
 348
 349     pixelshader_update_samplers(&This->baseShader.reg_maps,
 350             ((IWineD3DDeviceImpl *)This->baseShader.device)->stateBlock->textures);
 351
 352     /* Generate the HW shader */
 353     TRACE("(%p) : Generating hardware program\n", This);
 354     This->cur_args = args;
 355     shader_buffer_init(&buffer);
 356     retval = device->shader_backend->shader_generate_pshader((IWineD3DPixelShader *)This, &buffer, args);
 357     shader_buffer_free(&buffer);
 358     This->cur_args = NULL;
 359
 360     return retval;
 361 }
 362
 363 const IWineD3DPixelShaderVtbl IWineD3DPixelShader_Vtbl =
 364 {
 365     /*** IUnknown methods ***/
 366     IWineD3DPixelShaderImpl_QueryInterface,
 367     IWineD3DPixelShaderImpl_AddRef,
 368     IWineD3DPixelShaderImpl_Release,
 369     /*** IWineD3DBase methods ***/
 370     IWineD3DPixelShaderImpl_GetParent,
 371     /*** IWineD3DBaseShader methods ***/
 372     IWineD3DPixelShaderImpl_SetFunction,
 373     /*** IWineD3DPixelShader methods ***/
 374     IWineD3DPixelShaderImpl_GetDevice,
 375     IWineD3DPixelShaderImpl_GetFunction
 376 };
 377
 378 void find_ps_compile_args(IWineD3DPixelShaderImpl *shader, IWineD3DStateBlockImpl *stateblock, struct ps_compile_args *args) {
 379     UINT i;
 380     IWineD3DBaseTextureImpl *tex;
 381
 382     memset(args, 0, sizeof(*args)); /* FIXME: Make sure all bits are set */
 383     args->srgb_correction = stateblock->renderState[WINED3DRS_SRGBWRITEENABLE] ? 1 : 0;
 384     args->np2_fixup = 0;
 385
 386     for(i = 0; i < MAX_FRAGMENT_SAMPLERS; i++) {
 387         if (!shader->baseShader.reg_maps.sampler_type[i]) continue;
 388         tex = (IWineD3DBaseTextureImpl *) stateblock->textures[i];
 389         if(!tex) {
 390             args->color_fixup[i] = COLOR_FIXUP_IDENTITY;
 391             continue;
 392         }
 393         args->color_fixup[i] = tex->resource.format_desc->color_fixup;
 394
 395         /* Flag samplers that need NP2 texcoord fixup. */
 396         if(!tex->baseTexture.pow2Matrix_identity) {
 397             args->np2_fixup |= (1 << i);
 398         }
 399     }
 400     if (shader->baseShader.reg_maps.shader_version.major >= 3)
 401     {
 402         if (((IWineD3DDeviceImpl *)shader->baseShader.device)->strided_streams.position_transformed)
 403         {
 404             args->vp_mode = pretransformed;
 405         }
 406         else if (use_vs(stateblock))
 407         {
 408             args->vp_mode = vertexshader;
 409         } else {
 410             args->vp_mode = fixedfunction;
 411         }
 412         args->fog = FOG_OFF;
 413     } else {
 414         args->vp_mode = vertexshader;
 415         if(stateblock->renderState[WINED3DRS_FOGENABLE]) {
 416             switch(stateblock->renderState[WINED3DRS_FOGTABLEMODE]) {
 417                 case WINED3DFOG_NONE:
 418                     if (((IWineD3DDeviceImpl *)shader->baseShader.device)->strided_streams.position_transformed
 419                             || use_vs(stateblock))
 420                     {
 421                         args->fog = FOG_LINEAR;
 422                         break;
 423                     }
 424                     switch(stateblock->renderState[WINED3DRS_FOGVERTEXMODE]) {
 425                         case WINED3DFOG_NONE: /* Drop through */
 426                         case WINED3DFOG_LINEAR: args->fog = FOG_LINEAR; break;
 427                         case WINED3DFOG_EXP:    args->fog = FOG_EXP;    break;
 428                         case WINED3DFOG_EXP2:   args->fog = FOG_EXP2;   break;
 429                     }
 430                     break;
 431
 432                 case WINED3DFOG_LINEAR: args->fog = FOG_LINEAR; break;
 433                 case WINED3DFOG_EXP:    args->fog = FOG_EXP;    break;
 434                 case WINED3DFOG_EXP2:   args->fog = FOG_EXP2;   break;
 435             }
 436         } else {
 437             args->fog = FOG_OFF;
 438         }
 439     }
 440 }
 441
 442 GLuint find_gl_pshader(IWineD3DPixelShaderImpl *shader, const struct ps_compile_args *args)
 443 {
 444     UINT i;
 445     DWORD new_size;
 446     struct ps_compiled_shader *new_array;
 447
 448     /* Usually we have very few GL shaders for each d3d shader(just 1 or maybe 2),
 449      * so a linear search is more performant than a hashmap or a binary search
 450      * (cache coherency etc)
 451      */
 452     for(i = 0; i < shader->num_gl_shaders; i++) {
 453         if(memcmp(&shader->gl_shaders[i].args, args, sizeof(*args)) == 0) {
 454             return shader->gl_shaders[i].prgId;
 455         }
 456     }
 457
 458     TRACE("No matching GL shader found, compiling a new shader\n");
 459     if(shader->shader_array_size == shader->num_gl_shaders) {
 460         if (shader->num_gl_shaders)
 461         {
 462             new_size = shader->shader_array_size + max(1, shader->shader_array_size / 2);
 463             new_array = HeapReAlloc(GetProcessHeap(), 0, shader->gl_shaders,
 464                                     new_size * sizeof(*shader->gl_shaders));
 465         } else {
 466             new_array = HeapAlloc(GetProcessHeap(), 0, sizeof(*shader->gl_shaders));
 467             new_size = 1;
 468         }
 469
 470         if(!new_array) {
 471             ERR("Out of memory\n");
 472             return 0;
 473         }
 474         shader->gl_shaders = new_array;
 475         shader->shader_array_size = new_size;
 476     }
 477
 478     shader->gl_shaders[shader->num_gl_shaders].args = *args;
 479     shader->gl_shaders[shader->num_gl_shaders].prgId = pixelshader_compile(shader, args);
 480     return shader->gl_shaders[shader->num_gl_shaders++].prgId;
 481 }