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  *
  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 void vshader_set_limits(IWineD3DVertexShaderImpl *This)
  38 {
  39     DWORD shader_version = WINED3D_SHADER_VERSION(This->baseShader.reg_maps.shader_version.major,
  40             This->baseShader.reg_maps.shader_version.minor);
  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, GL_LIMITS(vshader_constantsF));
  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, GL_LIMITS(vshader_constantsF));
  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, GL_LIMITS(vshader_constantsF));
  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, GL_LIMITS(vshader_constantsF));
 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         HeapFree(GetProcessHeap(), 0, This);
 178     }
 179
 180     return refcount;
 181 }
 182
 183 /* *******************************************
 184    IWineD3DVertexShader IWineD3DVertexShader parts follow
 185    ******************************************* */
 186
 187 static HRESULT WINAPI IWineD3DVertexShaderImpl_GetParent(IWineD3DVertexShader *iface, IUnknown** parent){
 188     IWineD3DVertexShaderImpl *This = (IWineD3DVertexShaderImpl *)iface;
 189
 190     *parent = This->parent;
 191     IUnknown_AddRef(*parent);
 192     TRACE("(%p) : returning %p\n", This, *parent);
 193     return WINED3D_OK;
 194 }
 195
 196 static HRESULT WINAPI IWineD3DVertexShaderImpl_GetDevice(IWineD3DVertexShader* iface, IWineD3DDevice **pDevice){
 197     IWineD3DVertexShaderImpl *This = (IWineD3DVertexShaderImpl *)iface;
 198     IWineD3DDevice_AddRef(This->baseShader.device);
 199     *pDevice = This->baseShader.device;
 200     TRACE("(%p) returning %p\n", This, *pDevice);
 201     return WINED3D_OK;
 202 }
 203
 204 static HRESULT WINAPI IWineD3DVertexShaderImpl_GetFunction(IWineD3DVertexShader* impl, VOID* pData, UINT* pSizeOfData) {
 205     IWineD3DVertexShaderImpl *This = (IWineD3DVertexShaderImpl *)impl;
 206     TRACE("(%p) : pData(%p), pSizeOfData(%p)\n", This, pData, pSizeOfData);
 207
 208     if (NULL == pData) {
 209         *pSizeOfData = This->baseShader.functionLength;
 210         return WINED3D_OK;
 211     }
 212     if (*pSizeOfData < This->baseShader.functionLength) {
 213         /* MSDN claims (for d3d8 at least) that if *pSizeOfData is smaller
 214          * than the required size we should write the required size and
 215          * return D3DERR_MOREDATA. That's not actually true. */
 216         return WINED3DERR_INVALIDCALL;
 217     }
 218
 219     TRACE("(%p) : GetFunction copying to %p\n", This, pData);
 220     memcpy(pData, This->baseShader.function, This->baseShader.functionLength);
 221
 222     return WINED3D_OK;
 223 }
 224
 225 /* Note that for vertex shaders CompileShader isn't called until the
 226  * shader is first used. The reason for this is that we need the vertex
 227  * declaration the shader will be used with in order to determine if
 228  * the data in a register is of type D3DCOLOR, and needs swizzling. */
 229 static HRESULT WINAPI IWineD3DVertexShaderImpl_SetFunction(IWineD3DVertexShader *iface,
 230         const DWORD *pFunction, const struct wined3d_shader_signature *output_signature)
 231 {
 232     IWineD3DVertexShaderImpl *This =(IWineD3DVertexShaderImpl *)iface;
 233     IWineD3DDeviceImpl *deviceImpl = (IWineD3DDeviceImpl *) This->baseShader.device;
 234     const struct wined3d_shader_frontend *fe;
 235     unsigned int i;
 236     HRESULT hr;
 237     shader_reg_maps *reg_maps = &This->baseShader.reg_maps;
 238
 239     TRACE("(%p) : pFunction %p\n", iface, pFunction);
 240
 241     fe = shader_select_frontend(*pFunction);
 242     if (!fe)
 243     {
 244         FIXME("Unable to find frontend for shader.\n");
 245         return WINED3DERR_INVALIDCALL;
 246     }
 247     This->baseShader.frontend = fe;
 248     This->baseShader.frontend_data = fe->shader_init(pFunction, output_signature);
 249     if (!This->baseShader.frontend_data)
 250     {
 251         FIXME("Failed to initialize frontend.\n");
 252         return WINED3DERR_INVALIDCALL;
 253     }
 254
 255     /* First pass: trace shader */
 256     if (TRACE_ON(d3d_shader)) shader_trace_init(fe, This->baseShader.frontend_data, pFunction);
 257
 258     /* Initialize immediate constant lists */
 259     list_init(&This->baseShader.constantsF);
 260     list_init(&This->baseShader.constantsB);
 261     list_init(&This->baseShader.constantsI);
 262
 263     /* Second pass: figure out registers used, semantics, etc.. */
 264     This->min_rel_offset = GL_LIMITS(vshader_constantsF);
 265     This->max_rel_offset = 0;
 266     hr = shader_get_registers_used((IWineD3DBaseShader*) This, fe,
 267             reg_maps, This->attributes, NULL, This->output_signature,
 268             pFunction, GL_LIMITS(vshader_constantsF));
 269     if (hr != WINED3D_OK) return hr;
 270
 271     if (output_signature)
 272     {
 273         for (i = 0; i < output_signature->element_count; ++i)
 274         {
 275             struct wined3d_shader_signature_element *e = &output_signature->elements[i];
 276             reg_maps->output_registers |= 1 << e->register_idx;
 277             This->output_signature[e->register_idx] = *e;
 278         }
 279     }
 280
 281     vshader_set_limits(This);
 282
 283     if(deviceImpl->vs_selected_mode == SHADER_ARB &&
 284        (GLINFO_LOCATION).arb_vs_offset_limit      &&
 285        This->min_rel_offset <= This->max_rel_offset) {
 286
 287         if(This->max_rel_offset - This->min_rel_offset > 127) {
 288             FIXME("The difference between the minimum and maximum relative offset is > 127\n");
 289             FIXME("Which this OpenGL implementation does not support. Try using GLSL\n");
 290             FIXME("Min: %d, Max: %d\n", This->min_rel_offset, This->max_rel_offset);
 291         } else if(This->max_rel_offset - This->min_rel_offset > 63) {
 292             This->rel_offset = This->min_rel_offset + 63;
 293         } else if(This->max_rel_offset > 63) {
 294             This->rel_offset = This->min_rel_offset;
 295         } else {
 296             This->rel_offset = 0;
 297         }
 298     }
 299     This->baseShader.load_local_constsF = This->baseShader.reg_maps.usesrelconstF && !list_empty(&This->baseShader.constantsF);
 300
 301     /* copy the function ... because it will certainly be released by application */
 302     This->baseShader.function = HeapAlloc(GetProcessHeap(), 0, This->baseShader.functionLength);
 303     if (!This->baseShader.function) return E_OUTOFMEMORY;
 304     memcpy(This->baseShader.function, pFunction, This->baseShader.functionLength);
 305
 306     return WINED3D_OK;
 307 }
 308
 309 /* Set local constants for d3d8 shaders */
 310 static HRESULT WINAPI IWIneD3DVertexShaderImpl_SetLocalConstantsF(IWineD3DVertexShader *iface,
 311         UINT start_idx, const float *src_data, UINT count) {
 312     IWineD3DVertexShaderImpl *This =(IWineD3DVertexShaderImpl *)iface;
 313     UINT i, end_idx;
 314
 315     TRACE("(%p) : start_idx %u, src_data %p, count %u\n", This, start_idx, src_data, count);
 316
 317     end_idx = start_idx + count;
 318     if (end_idx > GL_LIMITS(vshader_constantsF)) {
 319         WARN("end_idx %u > float constants limit %u\n", end_idx, GL_LIMITS(vshader_constantsF));
 320         end_idx = GL_LIMITS(vshader_constantsF);
 321     }
 322
 323     for (i = start_idx; i < end_idx; ++i) {
 324         local_constant* lconst = HeapAlloc(GetProcessHeap(), 0, sizeof(local_constant));
 325         if (!lconst) return E_OUTOFMEMORY;
 326
 327         lconst->idx = i;
 328         memcpy(lconst->value, src_data + (i - start_idx) * 4 /* 4 components */, 4 * sizeof(float));
 329         list_add_head(&This->baseShader.constantsF, &lconst->entry);
 330     }
 331
 332     return WINED3D_OK;
 333 }
 334
 335 const IWineD3DVertexShaderVtbl IWineD3DVertexShader_Vtbl =
 336 {
 337     /*** IUnknown methods ***/
 338     IWineD3DVertexShaderImpl_QueryInterface,
 339     IWineD3DVertexShaderImpl_AddRef,
 340     IWineD3DVertexShaderImpl_Release,
 341     /*** IWineD3DBase methods ***/
 342     IWineD3DVertexShaderImpl_GetParent,
 343     /*** IWineD3DBaseShader methods ***/
 344     IWineD3DVertexShaderImpl_SetFunction,
 345     /*** IWineD3DVertexShader methods ***/
 346     IWineD3DVertexShaderImpl_GetDevice,
 347     IWineD3DVertexShaderImpl_GetFunction,
 348     IWIneD3DVertexShaderImpl_SetLocalConstantsF
 349 };
 350
 351 void find_vs_compile_args(IWineD3DVertexShaderImpl *shader, IWineD3DStateBlockImpl *stateblock, struct vs_compile_args *args) {
 352     args->fog_src = stateblock->renderState[WINED3DRS_FOGTABLEMODE] == WINED3DFOG_NONE ? VS_FOG_COORD : VS_FOG_Z;
 353     args->swizzle_map = ((IWineD3DDeviceImpl *)shader->baseShader.device)->strided_streams.swizzle_map;
 354 }