dlls/wined3d/vertexshader.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 void vshader_set_limits(IWineD3DVertexShaderImpl *This)
  37 {
  38     DWORD shader_version = WINED3D_SHADER_VERSION(This->baseShader.reg_maps.shader_version.major,
  39             This->baseShader.reg_maps.shader_version.minor);
  40     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *) This->baseShader.device;
  41
  42     This->baseShader.limits.texcoord = 0;
  43     This->baseShader.limits.attributes = 16;
  44     This->baseShader.limits.packed_input = 0;
  45
  46     switch (shader_version)
  47     {
  48         case WINED3D_SHADER_VERSION(1,0):
  49         case WINED3D_SHADER_VERSION(1,1):
  50             This->baseShader.limits.temporary = 12;
  51             This->baseShader.limits.constant_bool = 0;
  52             This->baseShader.limits.constant_int = 0;
  53             This->baseShader.limits.address = 1;
  54             This->baseShader.limits.packed_output = 0;
  55             This->baseShader.limits.sampler = 0;
  56             This->baseShader.limits.label = 0;
  57             /* TODO: vs_1_1 has a minimum of 96 constants. What happens if a vs_1_1 shader is used
  58              * on a vs_3_0 capable card that has 256 constants? */
  59             This->baseShader.limits.constant_float = min(256, device->d3d_vshader_constantF);
  60             break;
  61
  62         case WINED3D_SHADER_VERSION(2,0):
  63         case WINED3D_SHADER_VERSION(2,1):
  64             This->baseShader.limits.temporary = 12;
  65             This->baseShader.limits.constant_bool = 16;
  66             This->baseShader.limits.constant_int = 16;
  67             This->baseShader.limits.address = 1;
  68             This->baseShader.limits.packed_output = 0;
  69             This->baseShader.limits.sampler = 0;
  70             This->baseShader.limits.label = 16;
  71             This->baseShader.limits.constant_float = min(256, device->d3d_vshader_constantF);
  72             break;
  73
  74         case WINED3D_SHADER_VERSION(4,0):
  75             FIXME("Using 3.0 limits for 4.0 shader\n");
  76             /* Fall through */
  77
  78         case WINED3D_SHADER_VERSION(3,0):
  79             This->baseShader.limits.temporary = 32;
  80             This->baseShader.limits.constant_bool = 32;
  81             This->baseShader.limits.constant_int = 32;
  82             This->baseShader.limits.address = 1;
  83             This->baseShader.limits.packed_output = 12;
  84             This->baseShader.limits.sampler = 4;
  85             This->baseShader.limits.label = 16; /* FIXME: 2048 */
  86             /* DX10 cards on Windows advertise a d3d9 constant limit of 256 even though they are capable
  87              * of supporting much more(GL drivers advertise 1024). d3d9.dll and d3d8.dll clamp the
  88              * wined3d-advertised maximum. Clamp the constant limit for <= 3.0 shaders to 256.s
  89              * use constant buffers */
  90             This->baseShader.limits.constant_float = min(256, device->d3d_vshader_constantF);
  91             break;
  92
  93         default:
  94             This->baseShader.limits.temporary = 12;
  95             This->baseShader.limits.constant_bool = 16;
  96             This->baseShader.limits.constant_int = 16;
  97             This->baseShader.limits.address = 1;
  98             This->baseShader.limits.packed_output = 0;
  99             This->baseShader.limits.sampler = 0;
 100             This->baseShader.limits.label = 16;
 101             This->baseShader.limits.constant_float = min(256, device->d3d_vshader_constantF);
 102             FIXME("Unrecognized vertex shader version %u.%u\n",
 103                     This->baseShader.reg_maps.shader_version.major,
 104                     This->baseShader.reg_maps.shader_version.minor);
 105     }
 106 }
 107
 108 static BOOL match_usage(BYTE usage1, BYTE usage_idx1, BYTE usage2, BYTE usage_idx2) {
 109     if (usage_idx1 != usage_idx2) return FALSE;
 110     if (usage1 == usage2) return TRUE;
 111     if (usage1 == WINED3DDECLUSAGE_POSITION && usage2 == WINED3DDECLUSAGE_POSITIONT) return TRUE;
 112     if (usage2 == WINED3DDECLUSAGE_POSITION && usage1 == WINED3DDECLUSAGE_POSITIONT) return TRUE;
 113
 114     return FALSE;
 115 }
 116
 117 BOOL vshader_get_input(IWineD3DVertexShader* iface, BYTE usage_req, BYTE usage_idx_req, unsigned int *regnum)
 118 {
 119     IWineD3DVertexShaderImpl *This = (IWineD3DVertexShaderImpl *)iface;
 120     WORD map = This->baseShader.reg_maps.input_registers;
 121     unsigned int i;
 122
 123     for (i = 0; map; map >>= 1, ++i)
 124     {
 125         if (!(map & 1)) continue;
 126
 127         if (match_usage(This->attributes[i].usage,
 128                 This->attributes[i].usage_idx, usage_req, usage_idx_req))
 129         {
 130             *regnum = i;
 131             return TRUE;
 132         }
 133     }
 134     return FALSE;
 135 }
 136
 137 /* *******************************************
 138    IWineD3DVertexShader IUnknown parts follow
 139    ******************************************* */
 140 static HRESULT  WINAPI IWineD3DVertexShaderImpl_QueryInterface(IWineD3DVertexShader *iface, REFIID riid, LPVOID *ppobj) {
 141     TRACE("iface %p, riid %s, ppobj %p\n", iface, debugstr_guid(riid), ppobj);
 142
 143     if (IsEqualGUID(riid, &IID_IWineD3DVertexShader)
 144             || IsEqualGUID(riid, &IID_IWineD3DBaseShader)
 145             || IsEqualGUID(riid, &IID_IWineD3DBase)
 146             || IsEqualGUID(riid, &IID_IUnknown))
 147     {
 148         IUnknown_AddRef(iface);
 149         *ppobj = iface;
 150         return S_OK;
 151     }
 152
 153     WARN("%s not implemented, returning E_NOINTERFACE\n", debugstr_guid(riid));
 154
 155     *ppobj = NULL;
 156     return E_NOINTERFACE;
 157 }
 158
 159 static ULONG  WINAPI IWineD3DVertexShaderImpl_AddRef(IWineD3DVertexShader *iface) {
 160     IWineD3DVertexShaderImpl *This = (IWineD3DVertexShaderImpl *)iface;
 161     ULONG refcount = InterlockedIncrement(&This->baseShader.ref);
 162
 163     TRACE("%p increasing refcount to %u\n", This, refcount);
 164
 165     return refcount;
 166 }
 167
 168 static ULONG WINAPI IWineD3DVertexShaderImpl_Release(IWineD3DVertexShader *iface) {
 169     IWineD3DVertexShaderImpl *This = (IWineD3DVertexShaderImpl *)iface;
 170     ULONG refcount = InterlockedDecrement(&This->baseShader.ref);
 171
 172     TRACE("%p decreasing refcount to %u\n", This, refcount);
 173
 174     if (!refcount)
 175     {
 176         shader_cleanup((IWineD3DBaseShader *)iface);
 177         This->baseShader.parent_ops->wined3d_object_destroyed(This->baseShader.parent);
 178         HeapFree(GetProcessHeap(), 0, This);
 179     }
 180
 181     return refcount;
 182 }
 183
 184 /* *******************************************
 185    IWineD3DVertexShader IWineD3DVertexShader parts follow
 186    ******************************************* */
 187
 188 static HRESULT WINAPI IWineD3DVertexShaderImpl_GetParent(IWineD3DVertexShader *iface, IUnknown** parent){
 189     IWineD3DVertexShaderImpl *This = (IWineD3DVertexShaderImpl *)iface;
 190
 191     *parent = This->baseShader.parent;
 192     IUnknown_AddRef(*parent);
 193     TRACE("(%p) : returning %p\n", This, *parent);
 194     return WINED3D_OK;
 195 }
 196
 197 static HRESULT WINAPI IWineD3DVertexShaderImpl_GetFunction(IWineD3DVertexShader* impl, VOID* pData, UINT* pSizeOfData) {
 198     IWineD3DVertexShaderImpl *This = (IWineD3DVertexShaderImpl *)impl;
 199     TRACE("(%p) : pData(%p), pSizeOfData(%p)\n", This, pData, pSizeOfData);
 200
 201     if (NULL == pData) {
 202         *pSizeOfData = This->baseShader.functionLength;
 203         return WINED3D_OK;
 204     }
 205     if (*pSizeOfData < This->baseShader.functionLength) {
 206         /* MSDN claims (for d3d8 at least) that if *pSizeOfData is smaller
 207          * than the required size we should write the required size and
 208          * return D3DERR_MOREDATA. That's not actually true. */
 209         return WINED3DERR_INVALIDCALL;
 210     }
 211
 212     TRACE("(%p) : GetFunction copying to %p\n", This, pData);
 213     memcpy(pData, This->baseShader.function, This->baseShader.functionLength);
 214
 215     return WINED3D_OK;
 216 }
 217
 218 static HRESULT vertexshader_set_function(IWineD3DVertexShaderImpl *shader,
 219         const DWORD *byte_code, const struct wined3d_shader_signature *output_signature)
 220 {
 221     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *)shader->baseShader.device;
 222     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
 223     const struct wined3d_shader_frontend *fe;
 224     unsigned int i;
 225     HRESULT hr;
 226     shader_reg_maps *reg_maps = &shader->baseShader.reg_maps;
 227
 228     TRACE("shader %p, byte_code %p, output_signature %p.\n", shader, byte_code, output_signature);
 229
 230     fe = shader_select_frontend(*byte_code);
 231     if (!fe)
 232     {
 233         FIXME("Unable to find frontend for shader.\n");
 234         return WINED3DERR_INVALIDCALL;
 235     }
 236     shader->baseShader.frontend = fe;
 237     shader->baseShader.frontend_data = fe->shader_init(byte_code, output_signature);
 238     if (!shader->baseShader.frontend_data)
 239     {
 240         FIXME("Failed to initialize frontend.\n");
 241         return WINED3DERR_INVALIDCALL;
 242     }
 243
 244     /* First pass: trace shader */
 245     if (TRACE_ON(d3d_shader)) shader_trace_init(fe, shader->baseShader.frontend_data, byte_code);
 246
 247     /* Initialize immediate constant lists */
 248     list_init(&shader->baseShader.constantsF);
 249     list_init(&shader->baseShader.constantsB);
 250     list_init(&shader->baseShader.constantsI);
 251
 252     /* Second pass: figure out registers used, semantics, etc.. */
 253     shader->min_rel_offset = device->d3d_vshader_constantF;
 254     shader->max_rel_offset = 0;
 255     hr = shader_get_registers_used((IWineD3DBaseShader *)shader, fe,
 256             reg_maps, shader->attributes, NULL, shader->output_signature,
 257             byte_code, device->d3d_vshader_constantF);
 258     if (hr != WINED3D_OK) return hr;
 259
 260     if (output_signature)
 261     {
 262         for (i = 0; i < output_signature->element_count; ++i)
 263         {
 264             struct wined3d_shader_signature_element *e = &output_signature->elements[i];
 265             reg_maps->output_registers |= 1 << e->register_idx;
 266             shader->output_signature[e->register_idx] = *e;
 267         }
 268     }
 269
 270     vshader_set_limits(shader);
 271
 272     if (device->vs_selected_mode == SHADER_ARB
 273             && (gl_info->quirks & WINED3D_QUIRK_ARB_VS_OFFSET_LIMIT)
 274             && shader->min_rel_offset <= shader->max_rel_offset)
 275     {
 276         if (shader->max_rel_offset - shader->min_rel_offset > 127)
 277         {
 278             FIXME("The difference between the minimum and maximum relative offset is > 127\n");
 279             FIXME("Which this OpenGL implementation does not support. Try using GLSL\n");
 280             FIXME("Min: %d, Max: %d\n", shader->min_rel_offset, shader->max_rel_offset);
 281         }
 282         else if (shader->max_rel_offset - shader->min_rel_offset > 63)
 283         {
 284             shader->rel_offset = shader->min_rel_offset + 63;
 285         }
 286         else if (shader->max_rel_offset > 63)
 287         {
 288             shader->rel_offset = shader->min_rel_offset;
 289         }
 290         else
 291         {
 292             shader->rel_offset = 0;
 293         }
 294     }
 295     shader->baseShader.load_local_constsF = shader->baseShader.reg_maps.usesrelconstF
 296             && !list_empty(&shader->baseShader.constantsF);
 297
 298     /* copy the function ... because it will certainly be released by application */
 299     shader->baseShader.function = HeapAlloc(GetProcessHeap(), 0, shader->baseShader.functionLength);
 300     if (!shader->baseShader.function) return E_OUTOFMEMORY;
 301     memcpy(shader->baseShader.function, byte_code, shader->baseShader.functionLength);
 302
 303     return WINED3D_OK;
 304 }
 305
 306 /* Set local constants for d3d8 shaders */
 307 static HRESULT WINAPI IWIneD3DVertexShaderImpl_SetLocalConstantsF(IWineD3DVertexShader *iface,
 308         UINT start_idx, const float *src_data, UINT count) {
 309     IWineD3DVertexShaderImpl *This =(IWineD3DVertexShaderImpl *)iface;
 310     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *) This->baseShader.device;
 311     UINT i, end_idx;
 312
 313     TRACE("(%p) : start_idx %u, src_data %p, count %u\n", This, start_idx, src_data, count);
 314
 315     end_idx = start_idx + count;
 316     if (end_idx > device->d3d_vshader_constantF)
 317     {
 318         WARN("end_idx %u > float constants limit %u\n", end_idx, device->d3d_vshader_constantF);
 319         end_idx = device->d3d_vshader_constantF;
 320     }
 321
 322     for (i = start_idx; i < end_idx; ++i) {
 323         local_constant* lconst = HeapAlloc(GetProcessHeap(), 0, sizeof(local_constant));
 324         if (!lconst) return E_OUTOFMEMORY;
 325
 326         lconst->idx = i;
 327         memcpy(lconst->value, src_data + (i - start_idx) * 4 /* 4 components */, 4 * sizeof(float));
 328         list_add_head(&This->baseShader.constantsF, &lconst->entry);
 329     }
 330
 331     return WINED3D_OK;
 332 }
 333
 334 static const IWineD3DVertexShaderVtbl IWineD3DVertexShader_Vtbl =
 335 {
 336     /*** IUnknown methods ***/
 337     IWineD3DVertexShaderImpl_QueryInterface,
 338     IWineD3DVertexShaderImpl_AddRef,
 339     IWineD3DVertexShaderImpl_Release,
 340     /*** IWineD3DBase methods ***/
 341     IWineD3DVertexShaderImpl_GetParent,
 342     /*** IWineD3DBaseShader methods ***/
 343     IWineD3DVertexShaderImpl_GetFunction,
 344     /*** IWineD3DVertexShader methods ***/
 345     IWIneD3DVertexShaderImpl_SetLocalConstantsF
 346 };
 347
 348 void find_vs_compile_args(IWineD3DVertexShaderImpl *shader, IWineD3DStateBlockImpl *stateblock, struct vs_compile_args *args) {
 349     args->fog_src = stateblock->renderState[WINED3DRS_FOGTABLEMODE] == WINED3DFOG_NONE ? VS_FOG_COORD : VS_FOG_Z;
 350     args->clip_enabled = stateblock->renderState[WINED3DRS_CLIPPING] && stateblock->renderState[WINED3DRS_CLIPPLANEENABLE];
 351     args->swizzle_map = ((IWineD3DDeviceImpl *)shader->baseShader.device)->strided_streams.swizzle_map;
 352 }
 353
 354 HRESULT vertexshader_init(IWineD3DVertexShaderImpl *shader, IWineD3DDeviceImpl *device,
 355         const DWORD *byte_code, const struct wined3d_shader_signature *output_signature,
 356         IUnknown *parent, const struct wined3d_parent_ops *parent_ops)
 357 {
 358     HRESULT hr;
 359
 360     if (!byte_code) return WINED3DERR_INVALIDCALL;
 361
 362     shader->lpVtbl = &IWineD3DVertexShader_Vtbl;
 363     shader_init(&shader->baseShader, device, parent, parent_ops);
 364
 365     hr = vertexshader_set_function(shader, byte_code, output_signature);
 366     if (FAILED(hr))
 367     {
 368         WARN("Failed to set function, hr %#x.\n", hr);
 369         shader_cleanup((IWineD3DBaseShader *)shader);
 370         return hr;
 371     }
 372
 373     return WINED3D_OK;
 374 }