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(
 124       IWineD3DPixelShaderImpl *This) {
 125
 126       This->baseShader.limits.attributes = 0;
 127       This->baseShader.limits.address = 0;
 128       This->baseShader.limits.packed_output = 0;
 129
 130       switch (This->baseShader.reg_maps.shader_version)
 131       {
 132           case WINED3DPS_VERSION(1,0):
 133           case WINED3DPS_VERSION(1,1):
 134           case WINED3DPS_VERSION(1,2):
 135           case WINED3DPS_VERSION(1,3):
 136                    This->baseShader.limits.temporary = 2;
 137                    This->baseShader.limits.constant_float = 8;
 138                    This->baseShader.limits.constant_int = 0;
 139                    This->baseShader.limits.constant_bool = 0;
 140                    This->baseShader.limits.texcoord = 4;
 141                    This->baseShader.limits.sampler = 4;
 142                    This->baseShader.limits.packed_input = 0;
 143                    This->baseShader.limits.label = 0;
 144                    break;
 145
 146           case WINED3DPS_VERSION(1,4):
 147                    This->baseShader.limits.temporary = 6;
 148                    This->baseShader.limits.constant_float = 8;
 149                    This->baseShader.limits.constant_int = 0;
 150                    This->baseShader.limits.constant_bool = 0;
 151                    This->baseShader.limits.texcoord = 6;
 152                    This->baseShader.limits.sampler = 6;
 153                    This->baseShader.limits.packed_input = 0;
 154                    This->baseShader.limits.label = 0;
 155                    break;
 156
 157           /* FIXME: temporaries must match D3DPSHADERCAPS2_0.NumTemps */
 158           case WINED3DPS_VERSION(2,0):
 159                    This->baseShader.limits.temporary = 32;
 160                    This->baseShader.limits.constant_float = 32;
 161                    This->baseShader.limits.constant_int = 16;
 162                    This->baseShader.limits.constant_bool = 16;
 163                    This->baseShader.limits.texcoord = 8;
 164                    This->baseShader.limits.sampler = 16;
 165                    This->baseShader.limits.packed_input = 0;
 166                    break;
 167
 168           case WINED3DPS_VERSION(2,1):
 169                    This->baseShader.limits.temporary = 32;
 170                    This->baseShader.limits.constant_float = 32;
 171                    This->baseShader.limits.constant_int = 16;
 172                    This->baseShader.limits.constant_bool = 16;
 173                    This->baseShader.limits.texcoord = 8;
 174                    This->baseShader.limits.sampler = 16;
 175                    This->baseShader.limits.packed_input = 0;
 176                    This->baseShader.limits.label = 16;
 177                    break;
 178
 179           case WINED3DPS_VERSION(3,0):
 180                    This->baseShader.limits.temporary = 32;
 181                    This->baseShader.limits.constant_float = 224;
 182                    This->baseShader.limits.constant_int = 16;
 183                    This->baseShader.limits.constant_bool = 16;
 184                    This->baseShader.limits.texcoord = 0;
 185                    This->baseShader.limits.sampler = 16;
 186                    This->baseShader.limits.packed_input = 12;
 187                    This->baseShader.limits.label = 16; /* FIXME: 2048 */
 188                    break;
 189
 190           default: This->baseShader.limits.temporary = 32;
 191                    This->baseShader.limits.constant_float = 32;
 192                    This->baseShader.limits.constant_int = 16;
 193                    This->baseShader.limits.constant_bool = 16;
 194                    This->baseShader.limits.texcoord = 8;
 195                    This->baseShader.limits.sampler = 16;
 196                    This->baseShader.limits.packed_input = 0;
 197                    This->baseShader.limits.label = 0;
 198                    FIXME("Unrecognized pixel shader version %#x\n",
 199                            This->baseShader.reg_maps.shader_version);
 200       }
 201 }
 202
 203 static HRESULT WINAPI IWineD3DPixelShaderImpl_SetFunction(IWineD3DPixelShader *iface, CONST DWORD *pFunction) {
 204
 205     IWineD3DPixelShaderImpl *This =(IWineD3DPixelShaderImpl *)iface;
 206     IWineD3DDeviceImpl *deviceImpl = (IWineD3DDeviceImpl *) This->baseShader.device;
 207     unsigned int i, highest_reg_used = 0, num_regs_used = 0;
 208     shader_reg_maps *reg_maps = &This->baseShader.reg_maps;
 209     const struct wined3d_shader_frontend *fe;
 210     HRESULT hr;
 211
 212     TRACE("(%p) : pFunction %p\n", iface, pFunction);
 213
 214     fe = shader_select_frontend(*pFunction);
 215     if (!fe)
 216     {
 217         FIXME("Unable to find frontend for shader.\n");
 218         return WINED3DERR_INVALIDCALL;
 219     }
 220     This->baseShader.frontend = fe;
 221     This->baseShader.frontend_data = fe->shader_init(pFunction);
 222     if (!This->baseShader.frontend_data)
 223     {
 224         FIXME("Failed to initialize frontend.\n");
 225         return WINED3DERR_INVALIDCALL;
 226     }
 227
 228     /* First pass: trace shader */
 229     if (TRACE_ON(d3d_shader)) shader_trace_init(fe, This->baseShader.frontend_data, pFunction);
 230
 231     /* Initialize immediate constant lists */
 232     list_init(&This->baseShader.constantsF);
 233     list_init(&This->baseShader.constantsB);
 234     list_init(&This->baseShader.constantsI);
 235
 236     /* Second pass: figure out which registers are used, what the semantics are, etc.. */
 237     hr = shader_get_registers_used((IWineD3DBaseShader *)This, fe, reg_maps, This->semantics_in, NULL, pFunction);
 238     if (FAILED(hr)) return hr;
 239
 240     pshader_set_limits(This);
 241
 242     for (i = 0; i < MAX_REG_INPUT; ++i)
 243     {
 244         if (This->input_reg_used[i])
 245         {
 246             ++num_regs_used;
 247             highest_reg_used = i;
 248         }
 249     }
 250
 251     /* Don't do any register mapping magic if it is not needed, or if we can't
 252      * achieve anything anyway */
 253     if (highest_reg_used < (GL_LIMITS(glsl_varyings) / 4)
 254             || num_regs_used > (GL_LIMITS(glsl_varyings) / 4))
 255     {
 256         if (num_regs_used > (GL_LIMITS(glsl_varyings) / 4))
 257         {
 258             /* This happens with relative addressing. The input mapper function
 259              * warns about this if the higher registers are declared too, so
 260              * don't write a FIXME here */
 261             WARN("More varying registers used than supported\n");
 262         }
 263
 264         for (i = 0; i < MAX_REG_INPUT; ++i)
 265         {
 266             This->input_reg_map[i] = i;
 267         }
 268
 269         This->declared_in_count = highest_reg_used + 1;
 270     }
 271     else
 272     {
 273         This->declared_in_count = 0;
 274         for (i = 0; i < MAX_REG_INPUT; ++i)
 275         {
 276             if (This->input_reg_used[i]) This->input_reg_map[i] = This->declared_in_count++;
 277             else This->input_reg_map[i] = ~0U;
 278         }
 279     }
 280
 281     This->baseShader.load_local_constsF = FALSE;
 282
 283     This->baseShader.shader_mode = deviceImpl->ps_selected_mode;
 284
 285     TRACE("(%p) : Copying the function\n", This);
 286
 287     This->baseShader.function = HeapAlloc(GetProcessHeap(), 0, This->baseShader.functionLength);
 288     if (!This->baseShader.function) return E_OUTOFMEMORY;
 289     memcpy(This->baseShader.function, pFunction, This->baseShader.functionLength);
 290
 291     return WINED3D_OK;
 292 }
 293
 294 static void pixelshader_update_samplers(struct shader_reg_maps *reg_maps, IWineD3DBaseTexture * const *textures)
 295 {
 296     DWORD shader_version = reg_maps->shader_version;
 297     WINED3DSAMPLER_TEXTURE_TYPE *sampler_type = reg_maps->sampler_type;
 298     unsigned int i;
 299
 300     if (WINED3DSHADER_VERSION_MAJOR(shader_version) != 1) return;
 301
 302     for (i = 0; i < max(MAX_FRAGMENT_SAMPLERS, MAX_VERTEX_SAMPLERS); ++i)
 303     {
 304         /* We don't sample from this sampler */
 305         if (!sampler_type[i]) continue;
 306
 307         if (!textures[i])
 308         {
 309             ERR("No texture bound to sampler %u, using 2D\n", i);
 310             sampler_type[i] = WINED3DSTT_2D;
 311             continue;
 312         }
 313
 314         switch (IWineD3DBaseTexture_GetTextureDimensions(textures[i]))
 315         {
 316             case GL_TEXTURE_RECTANGLE_ARB:
 317             case GL_TEXTURE_2D:
 318                 /* We have to select between texture rectangles and 2D textures later because 2.0 and
 319                  * 3.0 shaders only have WINED3DSTT_2D as well */
 320                 sampler_type[i] = WINED3DSTT_2D;
 321                 break;
 322
 323             case GL_TEXTURE_3D:
 324                 sampler_type[i] = WINED3DSTT_VOLUME;
 325                 break;
 326
 327             case GL_TEXTURE_CUBE_MAP_ARB:
 328                 sampler_type[i] = WINED3DSTT_CUBE;
 329                 break;
 330
 331             default:
 332                 FIXME("Unrecognized texture type %#x, using 2D\n",
 333                         IWineD3DBaseTexture_GetTextureDimensions(textures[i]));
 334                 sampler_type[i] = WINED3DSTT_2D;
 335         }
 336     }
 337 }
 338
 339 static GLuint pixelshader_compile(IWineD3DPixelShaderImpl *This, const struct ps_compile_args *args)
 340 {
 341     CONST DWORD *function = This->baseShader.function;
 342     GLuint retval;
 343     SHADER_BUFFER buffer;
 344     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *) This->baseShader.device;
 345
 346     TRACE("(%p) : function %p\n", This, function);
 347
 348     pixelshader_update_samplers(&This->baseShader.reg_maps,
 349             ((IWineD3DDeviceImpl *)This->baseShader.device)->stateBlock->textures);
 350
 351     /* Generate the HW shader */
 352     TRACE("(%p) : Generating hardware program\n", This);
 353     This->cur_args = args;
 354     shader_buffer_init(&buffer);
 355     retval = device->shader_backend->shader_generate_pshader((IWineD3DPixelShader *)This, &buffer, args);
 356     shader_buffer_free(&buffer);
 357     This->cur_args = NULL;
 358
 359     return retval;
 360 }
 361
 362 const IWineD3DPixelShaderVtbl IWineD3DPixelShader_Vtbl =
 363 {
 364     /*** IUnknown methods ***/
 365     IWineD3DPixelShaderImpl_QueryInterface,
 366     IWineD3DPixelShaderImpl_AddRef,
 367     IWineD3DPixelShaderImpl_Release,
 368     /*** IWineD3DBase methods ***/
 369     IWineD3DPixelShaderImpl_GetParent,
 370     /*** IWineD3DBaseShader methods ***/
 371     IWineD3DPixelShaderImpl_SetFunction,
 372     /*** IWineD3DPixelShader methods ***/
 373     IWineD3DPixelShaderImpl_GetDevice,
 374     IWineD3DPixelShaderImpl_GetFunction
 375 };
 376
 377 void find_ps_compile_args(IWineD3DPixelShaderImpl *shader, IWineD3DStateBlockImpl *stateblock, struct ps_compile_args *args) {
 378     UINT i;
 379     IWineD3DBaseTextureImpl *tex;
 380
 381     memset(args, 0, sizeof(*args)); /* FIXME: Make sure all bits are set */
 382     args->srgb_correction = stateblock->renderState[WINED3DRS_SRGBWRITEENABLE] ? 1 : 0;
 383     args->np2_fixup = 0;
 384
 385     for(i = 0; i < MAX_FRAGMENT_SAMPLERS; i++) {
 386         if (!shader->baseShader.reg_maps.sampler_type[i]) continue;
 387         tex = (IWineD3DBaseTextureImpl *) stateblock->textures[i];
 388         if(!tex) {
 389             args->color_fixup[i] = COLOR_FIXUP_IDENTITY;
 390             continue;
 391         }
 392         args->color_fixup[i] = tex->resource.format_desc->color_fixup;
 393
 394         /* Flag samplers that need NP2 texcoord fixup. */
 395         if(!tex->baseTexture.pow2Matrix_identity) {
 396             args->np2_fixup |= (1 << i);
 397         }
 398     }
 399     if (shader->baseShader.reg_maps.shader_version >= WINED3DPS_VERSION(3,0))
 400     {
 401         if (((IWineD3DDeviceImpl *)shader->baseShader.device)->strided_streams.position_transformed)
 402         {
 403             args->vp_mode = pretransformed;
 404         }
 405         else if (use_vs(stateblock))
 406         {
 407             args->vp_mode = vertexshader;
 408         } else {
 409             args->vp_mode = fixedfunction;
 410         }
 411         args->fog = FOG_OFF;
 412     } else {
 413         args->vp_mode = vertexshader;
 414         if(stateblock->renderState[WINED3DRS_FOGENABLE]) {
 415             switch(stateblock->renderState[WINED3DRS_FOGTABLEMODE]) {
 416                 case WINED3DFOG_NONE:
 417                     if (((IWineD3DDeviceImpl *)shader->baseShader.device)->strided_streams.position_transformed
 418                             || use_vs(stateblock))
 419                     {
 420                         args->fog = FOG_LINEAR;
 421                         break;
 422                     }
 423                     switch(stateblock->renderState[WINED3DRS_FOGVERTEXMODE]) {
 424                         case WINED3DFOG_NONE: /* Drop through */
 425                         case WINED3DFOG_LINEAR: args->fog = FOG_LINEAR; break;
 426                         case WINED3DFOG_EXP:    args->fog = FOG_EXP;    break;
 427                         case WINED3DFOG_EXP2:   args->fog = FOG_EXP2;   break;
 428                     }
 429                     break;
 430
 431                 case WINED3DFOG_LINEAR: args->fog = FOG_LINEAR; break;
 432                 case WINED3DFOG_EXP:    args->fog = FOG_EXP;    break;
 433                 case WINED3DFOG_EXP2:   args->fog = FOG_EXP2;   break;
 434             }
 435         } else {
 436             args->fog = FOG_OFF;
 437         }
 438     }
 439 }
 440
 441 GLuint find_gl_pshader(IWineD3DPixelShaderImpl *shader, const struct ps_compile_args *args)
 442 {
 443     UINT i;
 444     DWORD new_size;
 445     struct ps_compiled_shader *new_array;
 446
 447     /* Usually we have very few GL shaders for each d3d shader(just 1 or maybe 2),
 448      * so a linear search is more performant than a hashmap or a binary search
 449      * (cache coherency etc)
 450      */
 451     for(i = 0; i < shader->num_gl_shaders; i++) {
 452         if(memcmp(&shader->gl_shaders[i].args, args, sizeof(*args)) == 0) {
 453             return shader->gl_shaders[i].prgId;
 454         }
 455     }
 456
 457     TRACE("No matching GL shader found, compiling a new shader\n");
 458     if(shader->shader_array_size == shader->num_gl_shaders) {
 459         if (shader->num_gl_shaders)
 460         {
 461             new_size = shader->shader_array_size + max(1, shader->shader_array_size / 2);
 462             new_array = HeapReAlloc(GetProcessHeap(), 0, shader->gl_shaders,
 463                                     new_size * sizeof(*shader->gl_shaders));
 464         } else {
 465             new_array = HeapAlloc(GetProcessHeap(), 0, sizeof(*shader->gl_shaders));
 466             new_size = 1;
 467         }
 468
 469         if(!new_array) {
 470             ERR("Out of memory\n");
 471             return 0;
 472         }
 473         shader->gl_shaders = new_array;
 474         shader->shader_array_size = new_size;
 475     }
 476
 477     shader->gl_shaders[shader->num_gl_shaders].args = *args;
 478     shader->gl_shaders[shader->num_gl_shaders].prgId = pixelshader_compile(shader, args);
 479     return shader->gl_shaders[shader->num_gl_shaders++].prgId;
 480 }