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  * Copyright 2009 Henri Verbeet for CodeWeavers
  11  *
  12  * This library is free software; you can redistribute it and/or
  13  * modify it under the terms of the GNU Lesser General Public
  14  * License as published by the Free Software Foundation; either
  15  * version 2.1 of the License, or (at your option) any later version.
  16  *
  17  * This library is distributed in the hope that it will be useful,
  18  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  19  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  20  * Lesser General Public License for more details.
  21  *
  22  * You should have received a copy of the GNU Lesser General Public
  23  * License along with this library; if not, write to the Free Software
  24  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301, USA
  25  */
  26
  27 #include "config.h"
  28
  29 #include <math.h>
  30 #include <stdio.h>
  31
  32 #include "wined3d_private.h"
  33
  34 WINE_DEFAULT_DEBUG_CHANNEL(d3d_shader);
  35
  36 static HRESULT  WINAPI IWineD3DPixelShaderImpl_QueryInterface(IWineD3DPixelShader *iface, REFIID riid, LPVOID *ppobj) {
  37     TRACE("iface %p, riid %s, ppobj %p\n", iface, debugstr_guid(riid), ppobj);
  38
  39     if (IsEqualGUID(riid, &IID_IWineD3DPixelShader)
  40             || IsEqualGUID(riid, &IID_IWineD3DBaseShader)
  41             || IsEqualGUID(riid, &IID_IWineD3DBase)
  42             || IsEqualGUID(riid, &IID_IUnknown))
  43     {
  44         IUnknown_AddRef(iface);
  45         *ppobj = iface;
  46         return S_OK;
  47     }
  48
  49     WARN("%s not implemented, returning E_NOINTERFACE\n", debugstr_guid(riid));
  50
  51     *ppobj = NULL;
  52     return E_NOINTERFACE;
  53 }
  54
  55 static ULONG  WINAPI IWineD3DPixelShaderImpl_AddRef(IWineD3DPixelShader *iface) {
  56     IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)iface;
  57     ULONG refcount = InterlockedIncrement(&This->baseShader.ref);
  58
  59     TRACE("%p increasing refcount to %u\n", This, refcount);
  60
  61     return refcount;
  62 }
  63
  64 static ULONG  WINAPI IWineD3DPixelShaderImpl_Release(IWineD3DPixelShader *iface) {
  65     IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)iface;
  66     ULONG refcount = InterlockedDecrement(&This->baseShader.ref);
  67
  68     TRACE("%p decreasing refcount to %u\n", This, refcount);
  69
  70     if (!refcount)
  71     {
  72         shader_cleanup((IWineD3DBaseShader *)iface);
  73         This->baseShader.parent_ops->wined3d_object_destroyed(This->baseShader.parent);
  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->baseShader.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(4,0):
 182             FIXME("Using 3.0 limits for 4.0 shader\n");
 183             /* Fall through */
 184
 185         case WINED3D_SHADER_VERSION(3,0):
 186             This->baseShader.limits.temporary = 32;
 187             This->baseShader.limits.constant_float = 224;
 188             This->baseShader.limits.constant_int = 16;
 189             This->baseShader.limits.constant_bool = 16;
 190             This->baseShader.limits.texcoord = 0;
 191             This->baseShader.limits.sampler = 16;
 192             This->baseShader.limits.packed_input = 12;
 193             This->baseShader.limits.label = 16; /* FIXME: 2048 */
 194             break;
 195
 196         default:
 197             This->baseShader.limits.temporary = 32;
 198             This->baseShader.limits.constant_float = 32;
 199             This->baseShader.limits.constant_int = 16;
 200             This->baseShader.limits.constant_bool = 16;
 201             This->baseShader.limits.texcoord = 8;
 202             This->baseShader.limits.sampler = 16;
 203             This->baseShader.limits.packed_input = 0;
 204             This->baseShader.limits.label = 0;
 205             FIXME("Unrecognized pixel shader version %u.%u\n",
 206                     This->baseShader.reg_maps.shader_version.major,
 207                     This->baseShader.reg_maps.shader_version.minor);
 208     }
 209 }
 210
 211 static HRESULT pixelshader_set_function(IWineD3DPixelShaderImpl *shader,
 212         const DWORD *byte_code, const struct wined3d_shader_signature *output_signature)
 213 {
 214     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *)shader->baseShader.device;
 215     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
 216     unsigned int i, highest_reg_used = 0, num_regs_used = 0;
 217     shader_reg_maps *reg_maps = &shader->baseShader.reg_maps;
 218     const struct wined3d_shader_frontend *fe;
 219     HRESULT hr;
 220
 221     TRACE("shader %p, byte_code %p, output_signature %p.\n", shader, byte_code, output_signature);
 222
 223     fe = shader_select_frontend(*byte_code);
 224     if (!fe)
 225     {
 226         FIXME("Unable to find frontend for shader.\n");
 227         return WINED3DERR_INVALIDCALL;
 228     }
 229     shader->baseShader.frontend = fe;
 230     shader->baseShader.frontend_data = fe->shader_init(byte_code, output_signature);
 231     if (!shader->baseShader.frontend_data)
 232     {
 233         FIXME("Failed to initialize frontend.\n");
 234         return WINED3DERR_INVALIDCALL;
 235     }
 236
 237     /* First pass: trace shader */
 238     if (TRACE_ON(d3d_shader)) shader_trace_init(fe, shader->baseShader.frontend_data, byte_code);
 239
 240     /* Initialize immediate constant lists */
 241     list_init(&shader->baseShader.constantsF);
 242     list_init(&shader->baseShader.constantsB);
 243     list_init(&shader->baseShader.constantsI);
 244
 245     /* Second pass: figure out which registers are used, what the semantics are, etc.. */
 246     hr = shader_get_registers_used((IWineD3DBaseShader *)shader, fe,
 247             reg_maps, NULL, shader->input_signature, NULL,
 248             byte_code, device->d3d_pshader_constantF);
 249     if (FAILED(hr)) return hr;
 250
 251     pshader_set_limits(shader);
 252
 253     for (i = 0; i < MAX_REG_INPUT; ++i)
 254     {
 255         if (shader->input_reg_used[i])
 256         {
 257             ++num_regs_used;
 258             highest_reg_used = i;
 259         }
 260     }
 261
 262     /* Don't do any register mapping magic if it is not needed, or if we can't
 263      * achieve anything anyway */
 264     if (highest_reg_used < (gl_info->limits.glsl_varyings / 4)
 265             || num_regs_used > (gl_info->limits.glsl_varyings / 4))
 266     {
 267         if (num_regs_used > (gl_info->limits.glsl_varyings / 4))
 268         {
 269             /* This happens with relative addressing. The input mapper function
 270              * warns about this if the higher registers are declared too, so
 271              * don't write a FIXME here */
 272             WARN("More varying registers used than supported\n");
 273         }
 274
 275         for (i = 0; i < MAX_REG_INPUT; ++i)
 276         {
 277             shader->input_reg_map[i] = i;
 278         }
 279
 280         shader->declared_in_count = highest_reg_used + 1;
 281     }
 282     else
 283     {
 284         shader->declared_in_count = 0;
 285         for (i = 0; i < MAX_REG_INPUT; ++i)
 286         {
 287             if (shader->input_reg_used[i]) shader->input_reg_map[i] = shader->declared_in_count++;
 288             else shader->input_reg_map[i] = ~0U;
 289         }
 290     }
 291
 292     shader->baseShader.load_local_constsF = FALSE;
 293
 294     TRACE("(%p) : Copying byte code.\n", shader);
 295
 296     shader->baseShader.function = HeapAlloc(GetProcessHeap(), 0, shader->baseShader.functionLength);
 297     if (!shader->baseShader.function) return E_OUTOFMEMORY;
 298     memcpy(shader->baseShader.function, byte_code, shader->baseShader.functionLength);
 299
 300     return WINED3D_OK;
 301 }
 302
 303 void pixelshader_update_samplers(struct shader_reg_maps *reg_maps, IWineD3DBaseTexture * const *textures)
 304 {
 305     WINED3DSAMPLER_TEXTURE_TYPE *sampler_type = reg_maps->sampler_type;
 306     unsigned int i;
 307
 308     if (reg_maps->shader_version.major != 1) return;
 309
 310     for (i = 0; i < max(MAX_FRAGMENT_SAMPLERS, MAX_VERTEX_SAMPLERS); ++i)
 311     {
 312         /* We don't sample from this sampler */
 313         if (!sampler_type[i]) continue;
 314
 315         if (!textures[i])
 316         {
 317             ERR("No texture bound to sampler %u, using 2D\n", i);
 318             sampler_type[i] = WINED3DSTT_2D;
 319             continue;
 320         }
 321
 322         switch (IWineD3DBaseTexture_GetTextureDimensions(textures[i]))
 323         {
 324             case GL_TEXTURE_RECTANGLE_ARB:
 325             case GL_TEXTURE_2D:
 326                 /* We have to select between texture rectangles and 2D textures later because 2.0 and
 327                  * 3.0 shaders only have WINED3DSTT_2D as well */
 328                 sampler_type[i] = WINED3DSTT_2D;
 329                 break;
 330
 331             case GL_TEXTURE_3D:
 332                 sampler_type[i] = WINED3DSTT_VOLUME;
 333                 break;
 334
 335             case GL_TEXTURE_CUBE_MAP_ARB:
 336                 sampler_type[i] = WINED3DSTT_CUBE;
 337                 break;
 338
 339             default:
 340                 FIXME("Unrecognized texture type %#x, using 2D\n",
 341                         IWineD3DBaseTexture_GetTextureDimensions(textures[i]));
 342                 sampler_type[i] = WINED3DSTT_2D;
 343         }
 344     }
 345 }
 346
 347 static const IWineD3DPixelShaderVtbl IWineD3DPixelShader_Vtbl =
 348 {
 349     /*** IUnknown methods ***/
 350     IWineD3DPixelShaderImpl_QueryInterface,
 351     IWineD3DPixelShaderImpl_AddRef,
 352     IWineD3DPixelShaderImpl_Release,
 353     /*** IWineD3DBase methods ***/
 354     IWineD3DPixelShaderImpl_GetParent,
 355     /*** IWineD3DBaseShader methods ***/
 356     IWineD3DPixelShaderImpl_GetDevice,
 357     IWineD3DPixelShaderImpl_GetFunction
 358     /*** IWineD3DPixelShader methods ***/
 359 };
 360
 361 void find_ps_compile_args(IWineD3DPixelShaderImpl *shader, IWineD3DStateBlockImpl *stateblock, struct ps_compile_args *args) {
 362     UINT i;
 363     IWineD3DBaseTextureImpl *tex;
 364
 365     memset(args, 0, sizeof(*args)); /* FIXME: Make sure all bits are set */
 366     args->srgb_correction = stateblock->renderState[WINED3DRS_SRGBWRITEENABLE] ? 1 : 0;
 367     args->np2_fixup = 0;
 368
 369     for(i = 0; i < MAX_FRAGMENT_SAMPLERS; i++) {
 370         if (!shader->baseShader.reg_maps.sampler_type[i]) continue;
 371         tex = (IWineD3DBaseTextureImpl *) stateblock->textures[i];
 372         if(!tex) {
 373             args->color_fixup[i] = COLOR_FIXUP_IDENTITY;
 374             continue;
 375         }
 376         args->color_fixup[i] = tex->resource.format_desc->color_fixup;
 377
 378         /* Flag samplers that need NP2 texcoord fixup. */
 379         if(!tex->baseTexture.pow2Matrix_identity) {
 380             args->np2_fixup |= (1 << i);
 381         }
 382     }
 383     if (shader->baseShader.reg_maps.shader_version.major >= 3)
 384     {
 385         if (((IWineD3DDeviceImpl *)shader->baseShader.device)->strided_streams.position_transformed)
 386         {
 387             args->vp_mode = pretransformed;
 388         }
 389         else if (use_vs(stateblock))
 390         {
 391             args->vp_mode = vertexshader;
 392         } else {
 393             args->vp_mode = fixedfunction;
 394         }
 395         args->fog = FOG_OFF;
 396     } else {
 397         args->vp_mode = vertexshader;
 398         if(stateblock->renderState[WINED3DRS_FOGENABLE]) {
 399             switch(stateblock->renderState[WINED3DRS_FOGTABLEMODE]) {
 400                 case WINED3DFOG_NONE:
 401                     if (((IWineD3DDeviceImpl *)shader->baseShader.device)->strided_streams.position_transformed
 402                             || use_vs(stateblock))
 403                     {
 404                         args->fog = FOG_LINEAR;
 405                         break;
 406                     }
 407                     switch(stateblock->renderState[WINED3DRS_FOGVERTEXMODE]) {
 408                         case WINED3DFOG_NONE: /* Drop through */
 409                         case WINED3DFOG_LINEAR: args->fog = FOG_LINEAR; break;
 410                         case WINED3DFOG_EXP:    args->fog = FOG_EXP;    break;
 411                         case WINED3DFOG_EXP2:   args->fog = FOG_EXP2;   break;
 412                     }
 413                     break;
 414
 415                 case WINED3DFOG_LINEAR: args->fog = FOG_LINEAR; break;
 416                 case WINED3DFOG_EXP:    args->fog = FOG_EXP;    break;
 417                 case WINED3DFOG_EXP2:   args->fog = FOG_EXP2;   break;
 418             }
 419         } else {
 420             args->fog = FOG_OFF;
 421         }
 422     }
 423 }
 424
 425 HRESULT pixelshader_init(IWineD3DPixelShaderImpl *shader, IWineD3DDeviceImpl *device,
 426         const DWORD *byte_code, const struct wined3d_shader_signature *output_signature,
 427         IUnknown *parent, const struct wined3d_parent_ops *parent_ops)
 428 {
 429     HRESULT hr;
 430
 431     if (!byte_code) return WINED3DERR_INVALIDCALL;
 432
 433     shader->lpVtbl = &IWineD3DPixelShader_Vtbl;
 434     shader_init(&shader->baseShader, device, parent, parent_ops);
 435
 436     hr = pixelshader_set_function(shader, byte_code, output_signature);
 437     if (FAILED(hr))
 438     {
 439         WARN("Failed to set function, hr %#x.\n", hr);
 440         shader_cleanup((IWineD3DBaseShader *)shader);
 441         return hr;
 442     }
 443
 444     return WINED3D_OK;
 445 }