push a18c97e3310ae68c9cf1d09430f25ca292f320eb
[wine/hacks.git] / dlls / wined3d / vertexbuffer.c
blob72cef079cf57c29cce0db55473a3452048866557
1 /*
2 * IWineD3DVertexBuffer Implementation
4 * Copyright 2002-2005 Jason Edmeades
5 * Raphael Junqueira
6 * Copyright 2004 Christian Costa
8 * This library is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU Lesser General Public
10 * License as published by the Free Software Foundation; either
11 * version 2.1 of the License, or (at your option) any later version.
13 * This library is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 * Lesser General Public License for more details.
18 * You should have received a copy of the GNU Lesser General Public
19 * License along with this library; if not, write to the Free Software
20 * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301, USA
23 #include "config.h"
24 #include "wined3d_private.h"
26 WINE_DEFAULT_DEBUG_CHANNEL(d3d);
27 #define GLINFO_LOCATION This->resource.wineD3DDevice->adapter->gl_info
29 #define VB_MAXDECLCHANGES 100 /* After that number we stop converting */
30 #define VB_RESETDECLCHANGE 1000 /* Reset the changecount after that number of draws */
32 /* *******************************************
33 IWineD3DVertexBuffer IUnknown parts follow
34 ******************************************* */
35 static HRESULT WINAPI IWineD3DVertexBufferImpl_QueryInterface(IWineD3DVertexBuffer *iface, REFIID riid, LPVOID *ppobj)
37 IWineD3DVertexBufferImpl *This = (IWineD3DVertexBufferImpl *)iface;
38 TRACE("(%p)->(%s,%p)\n",This,debugstr_guid(riid),ppobj);
39 if (IsEqualGUID(riid, &IID_IUnknown)
40 || IsEqualGUID(riid, &IID_IWineD3DBase)
41 || IsEqualGUID(riid, &IID_IWineD3DResource)
42 || IsEqualGUID(riid, &IID_IWineD3DVertexBuffer)){
43 IUnknown_AddRef(iface);
44 *ppobj = This;
45 return S_OK;
47 *ppobj = NULL;
48 return E_NOINTERFACE;
51 static ULONG WINAPI IWineD3DVertexBufferImpl_AddRef(IWineD3DVertexBuffer *iface) {
52 IWineD3DVertexBufferImpl *This = (IWineD3DVertexBufferImpl *)iface;
53 ULONG ref = InterlockedIncrement(&This->resource.ref);
54 TRACE("(%p) : AddRef increasing from %d\n", This, ref - 1);
55 return ref;
58 static ULONG WINAPI IWineD3DVertexBufferImpl_Release(IWineD3DVertexBuffer *iface) {
59 IWineD3DVertexBufferImpl *This = (IWineD3DVertexBufferImpl *)iface;
60 ULONG ref = InterlockedDecrement(&This->resource.ref);
61 TRACE("(%p) : Releasing from %d\n", This, ref + 1);
62 if (ref == 0) {
64 if(This->vbo) {
65 IWineD3DDeviceImpl *device = This->resource.wineD3DDevice;
67 ActivateContext(device, device->lastActiveRenderTarget, CTXUSAGE_RESOURCELOAD);
68 ENTER_GL();
69 GL_EXTCALL(glDeleteBuffersARB(1, &This->vbo));
70 checkGLcall("glDeleteBuffersARB");
71 LEAVE_GL();
74 IWineD3DResourceImpl_CleanUp((IWineD3DResource *)iface);
75 HeapFree(GetProcessHeap(), 0, This);
77 return ref;
80 /* ****************************************************
81 IWineD3DVertexBuffer IWineD3DResource parts follow
82 **************************************************** */
83 static HRESULT WINAPI IWineD3DVertexBufferImpl_GetDevice(IWineD3DVertexBuffer *iface, IWineD3DDevice** ppDevice) {
84 return IWineD3DResourceImpl_GetDevice((IWineD3DResource *)iface, ppDevice);
87 static HRESULT WINAPI IWineD3DVertexBufferImpl_SetPrivateData(IWineD3DVertexBuffer *iface, REFGUID refguid, CONST void* pData, DWORD SizeOfData, DWORD Flags) {
88 return IWineD3DResourceImpl_SetPrivateData((IWineD3DResource *)iface, refguid, pData, SizeOfData, Flags);
91 static HRESULT WINAPI IWineD3DVertexBufferImpl_GetPrivateData(IWineD3DVertexBuffer *iface, REFGUID refguid, void* pData, DWORD* pSizeOfData) {
92 return IWineD3DResourceImpl_GetPrivateData((IWineD3DResource *)iface, refguid, pData, pSizeOfData);
95 static HRESULT WINAPI IWineD3DVertexBufferImpl_FreePrivateData(IWineD3DVertexBuffer *iface, REFGUID refguid) {
96 return IWineD3DResourceImpl_FreePrivateData((IWineD3DResource *)iface, refguid);
99 static DWORD WINAPI IWineD3DVertexBufferImpl_SetPriority(IWineD3DVertexBuffer *iface, DWORD PriorityNew) {
100 return IWineD3DResourceImpl_SetPriority((IWineD3DResource *)iface, PriorityNew);
103 static DWORD WINAPI IWineD3DVertexBufferImpl_GetPriority(IWineD3DVertexBuffer *iface) {
104 return IWineD3DResourceImpl_GetPriority((IWineD3DResource *)iface);
107 static inline void fixup_d3dcolor(DWORD *pos) {
108 DWORD srcColor = *pos;
110 /* Color conversion like in drawStridedSlow. watch out for little endianity
111 * If we want that stuff to work on big endian machines too we have to consider more things
113 * 0xff000000: Alpha mask
114 * 0x00ff0000: Blue mask
115 * 0x0000ff00: Green mask
116 * 0x000000ff: Red mask
118 *pos = 0;
119 *pos |= (srcColor & 0xff00ff00) ; /* Alpha Green */
120 *pos |= (srcColor & 0x00ff0000) >> 16; /* Red */
121 *pos |= (srcColor & 0x000000ff) << 16; /* Blue */
124 static inline void fixup_transformed_pos(float *p) {
125 float x, y, z, w;
127 /* rhw conversion like in drawStridedSlow */
128 if(p[3] == 1.0 || ((p[3] < eps) && (p[3] > -eps))) {
129 x = p[0];
130 y = p[1];
131 z = p[2];
132 w = 1.0;
133 } else {
134 w = 1.0 / p[3];
135 x = p[0] * w;
136 y = p[1] * w;
137 z = p[2] * w;
139 p[0] = x;
140 p[1] = y;
141 p[2] = z;
142 p[3] = w;
145 DWORD *find_conversion_shift(IWineD3DVertexBufferImpl *This, WineDirect3DVertexStridedData *strided, DWORD stride) {
146 DWORD *ret, i, shift, j, type;
147 DWORD orig_type_size;
149 if(!stride) {
150 TRACE("No shift\n");
151 return NULL;
154 This->conv_stride = stride;
155 ret = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(DWORD) * stride);
156 for(i = 0; i < MAX_ATTRIBS; i++) {
157 if(strided->u.input[i].VBO != This->vbo) continue;
159 type = strided->u.input[i].dwType;
160 if(type == WINED3DDECLTYPE_FLOAT16_2) {
161 shift = 4;
162 } else if(type == WINED3DDECLTYPE_FLOAT16_4) {
163 shift = 8;
164 /* Pre-shift the last 4 bytes in the FLOAT16_4 by 4 bytes - this makes FLOAT16_2 and FLOAT16_4 conversions
165 * compatible
167 for(j = 4; j < 8; j++) {
168 ret[(DWORD_PTR) strided->u.input[i].lpData + j ] += 4;
170 } else {
171 shift = 0;
173 This->conv_stride += shift;
175 if(shift) {
176 orig_type_size = WINED3D_ATR_TYPESIZE(type) * WINED3D_ATR_SIZE(type);
177 for(j = (DWORD_PTR) strided->u.input[i].lpData + orig_type_size; j < stride; j++) {
178 ret[j] += shift;
183 if(TRACE_ON(d3d)) {
184 TRACE("Dumping conversion shift:\n");
185 for(i = 0; i < stride; i++) {
186 TRACE("[%d]", ret[i]);
188 TRACE("\n");
190 return ret;
193 static inline BOOL process_converted_attribute(IWineD3DVertexBufferImpl *This,
194 const enum vbo_conversion_type conv_type,
195 const WineDirect3DStridedData *attrib,
196 DWORD *stride_this_run, const DWORD type) {
197 DWORD attrib_size;
198 BOOL ret = FALSE;
199 int i;
200 DWORD offset = This->resource.wineD3DDevice->stateBlock->streamOffset[attrib->streamNo];
201 DWORD_PTR data;
203 /* Check for some valid situations which cause us pain. One is if the buffer is used for
204 * constant attributes(stride = 0), the other one is if the buffer is used on two streams
205 * with different strides. In the 2nd case we might have to drop conversion entirely,
206 * it is possible that the same bytes are once read as FLOAT2 and once as UBYTE4N.
208 if(attrib->dwStride == 0) {
209 FIXME("%s used with stride 0, let's hope we get the vertex stride from somewhere else\n",
210 debug_d3ddecltype(type));
211 } else if(attrib->dwStride != *stride_this_run &&
212 *stride_this_run) {
213 FIXME("Got two concurrent strides, %d and %d\n", attrib->dwStride, *stride_this_run);
214 } else {
215 *stride_this_run = attrib->dwStride;
216 if(This->stride != *stride_this_run) {
217 /* We rely that this happens only on the first converted attribute that is found,
218 * if at all. See above check
220 TRACE("Reconverting because converted attributes occur, and the stride changed\n");
221 This->stride = *stride_this_run;
222 HeapFree(GetProcessHeap(), HEAP_ZERO_MEMORY, This->conv_map);
223 This->conv_map = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*This->conv_map) * This->stride);
224 ret = TRUE;
228 data = (((DWORD_PTR) attrib->lpData) + offset) % This->stride;
229 attrib_size = WINED3D_ATR_SIZE(type) * WINED3D_ATR_TYPESIZE(type);
230 for(i = 0; i < attrib_size; i++) {
231 if(This->conv_map[data + i] != conv_type) {
232 TRACE("Byte %ld in vertex changed\n", i + data);
233 TRACE("It was type %d, is %d now\n", This->conv_map[data + i], conv_type);
234 ret = TRUE;
235 This->conv_map[data + i] = conv_type;
238 return ret;
241 static inline BOOL check_attribute(IWineD3DVertexBufferImpl *This, const WineDirect3DStridedData *attrib,
242 const BOOL check_d3dcolor, const BOOL is_ffp_position, const BOOL is_ffp_color,
243 DWORD *stride_this_run, BOOL *float16_used) {
244 BOOL ret = FALSE;
245 DWORD type;
247 /* Ignore attributes that do not have our vbo. After that check we can be sure that the attribute is
248 * there, on nonexistent attribs the vbo is 0.
250 if(attrib->VBO != This->vbo) return FALSE;
252 type = attrib->dwType;
253 /* Look for newly appeared conversion */
254 if(!GL_SUPPORT(NV_HALF_FLOAT) && (
255 type == WINED3DDECLTYPE_FLOAT16_2 ||
256 type == WINED3DDECLTYPE_FLOAT16_4)) {
258 ret = process_converted_attribute(This, CONV_FLOAT16_2, attrib, stride_this_run, type);
260 if(is_ffp_position) {
261 FIXME("Test FLOAT16 fixed function processing positions\n");
262 } else if(is_ffp_color) {
263 FIXME("test FLOAT16 fixed function processing colors\n");
265 *float16_used = TRUE;
266 } else if(check_d3dcolor && type == WINED3DDECLTYPE_D3DCOLOR) {
268 ret = process_converted_attribute(This, CONV_D3DCOLOR, attrib, stride_this_run, WINED3DDECLTYPE_D3DCOLOR);
270 if(!is_ffp_color) {
271 FIXME("Test for non-color fixed function D3DCOLOR type\n");
273 } else if(is_ffp_position && type == WINED3DDECLTYPE_FLOAT4) {
274 ret = process_converted_attribute(This, CONV_POSITIONT, attrib, stride_this_run, WINED3DDECLTYPE_FLOAT4);
275 } else if(This->conv_map) {
276 ret = process_converted_attribute(This, CONV_NONE, attrib, stride_this_run, type);
278 return ret;
281 inline BOOL WINAPI IWineD3DVertexBufferImpl_FindDecl(IWineD3DVertexBufferImpl *This)
283 IWineD3DDeviceImpl *device = This->resource.wineD3DDevice;
284 BOOL ret = FALSE;
285 int i;
286 DWORD stride_this_run = 0;
287 BOOL float16_used = FALSE;
289 /* In d3d7 the vertex buffer declaration NEVER changes because it is stored in the d3d7 vertex buffer.
290 * Once we have our declaration there is no need to look it up again.
292 if(((IWineD3DImpl *)device->wineD3D)->dxVersion == 7 && This->Flags & VBFLAG_HASDESC) {
293 return FALSE;
296 TRACE("Finding vertex buffer conversion information\n");
297 /* Certain declaration types need some fixups before we can pass them to opengl. This means D3DCOLOR attributes with fixed
298 * function vertex processing, FLOAT4 POSITIONT with fixed function, and FLOAT16 if GL_NV_half_float is not supported.
300 * The vertex buffer FVF doesn't help with finding them, we have to use the decoded vertex declaration and pick the things
301 * that concern the current buffer. A problem with this is that this can change between draws, so we have to validate
302 * the information and reprocess the buffer if it changes, and avoid false positives for performance reasons.
304 * We have to distinguish between vertex shaders and fixed function to pick the way we access the
305 * strided vertex information.
307 * This code sets up a per-byte array with the size of the detected stride of the arrays in the
308 * buffer. For each byte we have a field that marks the conversion needed on this byte. For example,
309 * the following declaration with fixed function vertex processing:
311 * POSITIONT, FLOAT4
312 * NORMAL, FLOAT3
313 * DIFFUSE, FLOAT16_4
314 * SPECULAR, D3DCOLOR
316 * Will result in
317 * { POSITIONT }{ NORMAL }{ DIFFUSE }{SPECULAR }
318 * [P][P][P][P][P][P][P][P][P][P][P][P][P][P][P][P][0][0][0][0][0][0][0][0][0][0][0][0][F][F][F][F][F][F][F][F][C][C][C][C]
320 * Where in this example map P means 4 component position conversion, 0 means no conversion, F means FLOAT16_2 conversion
321 * and C means D3DCOLOR conversion(red / blue swizzle).
323 * If we're doing conversion and the stride changes we have to reconvert the whole buffer. Note that we do not mind if the
324 * semantic changes, we only care for the conversion type. So if the NORMAL is replaced with a TEXCOORD, nothing has to be
325 * done, or if the DIFFUSE is replaced with a D3DCOLOR BLENDWEIGHT we can happily dismiss the change. Some conversion types
326 * depend on the semantic as well, for example a FLOAT4 texcoord needs no conversion while a FLOAT4 positiont needs one
328 if(use_vs(device)) {
329 TRACE("vhsader\n");
330 /* If the current vertex declaration is marked for no half float conversion don't bother to
331 * analyse the strided streams in depth, just set them up for no conversion. Return decl changed
332 * if we used conversion before
334 if(!((IWineD3DVertexDeclarationImpl *) device->stateBlock->vertexDecl)->half_float_conv_needed) {
335 if(This->conv_map) {
336 TRACE("Now using shaders without conversion, but conversion used before\n");
337 HeapFree(GetProcessHeap(), 0, This->conv_map);
338 HeapFree(GetProcessHeap(), 0, This->conv_shift);
339 This->conv_map = NULL;
340 This->stride = 0;
341 This->conv_shift = NULL;
342 This->conv_stride = 0;
343 return TRUE;
344 } else {
345 return FALSE;
348 for(i = 0; i < MAX_ATTRIBS; i++) {
349 ret = check_attribute(This, &device->strided_streams.u.input[i], FALSE, FALSE, FALSE, &stride_this_run, &float16_used) || ret;
352 /* Recalculate the conversion shift map if the declaration has changed,
353 * and we're using float16 conversion or used it on the last run
355 if(ret && (float16_used || This->conv_map)) {
356 HeapFree(GetProcessHeap(), 0, This->conv_shift);
357 This->conv_shift = find_conversion_shift(This, &device->strided_streams, This->stride);
359 } else {
360 /* Fixed function is a bit trickier. We have to take care for D3DCOLOR types, FLOAT4 positions and of course
361 * FLOAT16s if not supported. Also, we can't iterate over the array, so use macros to generate code for all
362 * the attributes that our current fixed function pipeline implementation cares for.
364 ret = check_attribute(This, &device->strided_streams.u.s.position, TRUE, TRUE, FALSE, &stride_this_run, &float16_used) || ret;
365 ret = check_attribute(This, &device->strided_streams.u.s.normal, TRUE, FALSE, FALSE, &stride_this_run, &float16_used) || ret;
366 ret = check_attribute(This, &device->strided_streams.u.s.diffuse, TRUE, FALSE, TRUE, &stride_this_run, &float16_used) || ret;
367 ret = check_attribute(This, &device->strided_streams.u.s.specular, TRUE, FALSE, TRUE, &stride_this_run, &float16_used) || ret;
368 ret = check_attribute(This, &device->strided_streams.u.s.texCoords[0], TRUE, FALSE, FALSE, &stride_this_run, &float16_used) || ret;
369 ret = check_attribute(This, &device->strided_streams.u.s.texCoords[1], TRUE, FALSE, FALSE, &stride_this_run, &float16_used) || ret;
370 ret = check_attribute(This, &device->strided_streams.u.s.texCoords[2], TRUE, FALSE, FALSE, &stride_this_run, &float16_used) || ret;
371 ret = check_attribute(This, &device->strided_streams.u.s.texCoords[3], TRUE, FALSE, FALSE, &stride_this_run, &float16_used) || ret;
372 ret = check_attribute(This, &device->strided_streams.u.s.texCoords[4], TRUE, FALSE, FALSE, &stride_this_run, &float16_used) || ret;
373 ret = check_attribute(This, &device->strided_streams.u.s.texCoords[5], TRUE, FALSE, FALSE, &stride_this_run, &float16_used) || ret;
374 ret = check_attribute(This, &device->strided_streams.u.s.texCoords[6], TRUE, FALSE, FALSE, &stride_this_run, &float16_used) || ret;
375 ret = check_attribute(This, &device->strided_streams.u.s.texCoords[7], TRUE, FALSE, FALSE, &stride_this_run, &float16_used) || ret;
377 if(float16_used) FIXME("Float16 conversion used with fixed function vertex processing\n");
380 if(stride_this_run == 0 && This->conv_map) {
381 /* Sanity test */
382 if(ret == FALSE) {
383 ERR("no converted attributes found, old conversion map exists, and no declaration change???\n");
385 HeapFree(GetProcessHeap(), 0, This->conv_map);
386 This->conv_map = NULL;
387 This->stride = 0;
389 This->Flags |= VBFLAG_HASDESC;
391 if(ret) TRACE("Conversion information changed\n");
392 return ret;
395 static void check_vbo_size(IWineD3DVertexBufferImpl *This) {
396 DWORD size = This->conv_stride ? This->conv_stride * (This->resource.size / This->stride) : This->resource.size;
397 if(This->vbo_size != size) {
398 TRACE("Old size %d, creating new size %d\n", This->vbo_size, size);
399 ENTER_GL();
400 GL_EXTCALL(glBindBufferARB(GL_ARRAY_BUFFER_ARB, This->vbo));
401 checkGLcall("glBindBufferARB");
402 GL_EXTCALL(glBufferDataARB(GL_ARRAY_BUFFER_ARB, size, NULL, This->vbo_usage));
403 This->vbo_size = size;
404 checkGLcall("glBufferDataARB");
405 LEAVE_GL();
409 static void CreateVBO(IWineD3DVertexBufferImpl *This) {
410 GLenum error, glUsage;
411 DWORD vboUsage = This->resource.usage;
412 IWineD3DDeviceImpl *device = This->resource.wineD3DDevice;
414 TRACE("Creating an OpenGL vertex buffer object for IWineD3DVertexBuffer %p Usage(%s)\n", This, debug_d3dusage(vboUsage));
416 /* Make sure that a context is there. Needed in a multithreaded environment. Otherwise this call is a nop */
417 ActivateContext(device, device->lastActiveRenderTarget, CTXUSAGE_RESOURCELOAD);
418 ENTER_GL();
420 /* Make sure that the gl error is cleared. Do not use checkGLcall
421 * here because checkGLcall just prints a fixme and continues. However,
422 * if an error during VBO creation occurs we can fall back to non-vbo operation
423 * with full functionality(but performance loss)
425 while(glGetError() != GL_NO_ERROR);
427 /* Basically the FVF parameter passed to CreateVertexBuffer is no good
428 * It is the FVF set with IWineD3DDevice::SetFVF or the Vertex Declaration set with
429 * IWineD3DDevice::SetVertexDeclaration that decides how the vertices in the buffer
430 * look like. This means that on each DrawPrimitive call the vertex buffer has to be verified
431 * to check if the rhw and color values are in the correct format.
434 GL_EXTCALL(glGenBuffersARB(1, &This->vbo));
435 error = glGetError();
436 if(This->vbo == 0 || error != GL_NO_ERROR) {
437 WARN("Failed to create a VBO with error %s (%#x)\n", debug_glerror(error), error);
438 goto error;
441 GL_EXTCALL(glBindBufferARB(GL_ARRAY_BUFFER_ARB, This->vbo));
442 error = glGetError();
443 if(error != GL_NO_ERROR) {
444 WARN("Failed to bind the VBO with error %s (%#x)\n", debug_glerror(error), error);
445 goto error;
448 /* Don't use static, because dx apps tend to update the buffer
449 * quite often even if they specify 0 usage. Because we always keep the local copy
450 * we never read from the vbo and can create a write only opengl buffer.
452 switch(vboUsage & (WINED3DUSAGE_WRITEONLY | WINED3DUSAGE_DYNAMIC) ) {
453 case WINED3DUSAGE_WRITEONLY | WINED3DUSAGE_DYNAMIC:
454 case WINED3DUSAGE_DYNAMIC:
455 TRACE("Gl usage = GL_STREAM_DRAW\n");
456 glUsage = GL_STREAM_DRAW_ARB;
457 break;
458 case WINED3DUSAGE_WRITEONLY:
459 default:
460 TRACE("Gl usage = GL_DYNAMIC_DRAW\n");
461 glUsage = GL_DYNAMIC_DRAW_ARB;
462 break;
465 /* Reserve memory for the buffer. The amount of data won't change
466 * so we are safe with calling glBufferData once with a NULL ptr and
467 * calling glBufferSubData on updates
469 GL_EXTCALL(glBufferDataARB(GL_ARRAY_BUFFER_ARB, This->resource.size, NULL, glUsage));
470 error = glGetError();
471 if(error != GL_NO_ERROR) {
472 WARN("glBufferDataARB failed with error %s (%#x)\n", debug_glerror(error), error);
473 goto error;
475 This->vbo_size = This->resource.size;
476 This->vbo_usage = glUsage;
478 LEAVE_GL();
480 return;
481 error:
482 /* Clean up all vbo init, but continue because we can work without a vbo :-) */
483 FIXME("Failed to create a vertex buffer object. Continuing, but performance issues can occur\n");
484 if(This->vbo) GL_EXTCALL(glDeleteBuffersARB(1, &This->vbo));
485 This->vbo = 0;
486 LEAVE_GL();
487 return;
490 static void WINAPI IWineD3DVertexBufferImpl_PreLoad(IWineD3DVertexBuffer *iface) {
491 IWineD3DVertexBufferImpl *This = (IWineD3DVertexBufferImpl *) iface;
492 IWineD3DDeviceImpl *device = This->resource.wineD3DDevice;
493 BYTE *data;
494 UINT start = 0, end = 0, vertices;
495 BOOL declChanged = FALSE;
496 int i, j;
497 TRACE("(%p)->()\n", This);
499 if(This->Flags & VBFLAG_LOAD) {
500 return; /* Already doing that stuff */
503 if(!This->vbo) {
504 /* TODO: Make converting independent from VBOs */
505 if(This->Flags & VBFLAG_CREATEVBO) {
506 CreateVBO(This);
507 This->Flags &= ~VBFLAG_CREATEVBO;
508 } else {
509 return; /* Not doing any conversion */
513 /* Reading the declaration makes only sense if the stateblock is finalized and the buffer bound to a stream */
514 if(device->isInDraw && This->bindCount > 0) {
515 declChanged = IWineD3DVertexBufferImpl_FindDecl(This);
516 } else if(This->Flags & VBFLAG_HASDESC) {
517 /* Reuse the declaration stored in the buffer. It will most likely not change, and if it does
518 * the stream source state handler will call PreLoad again and the change will be caught
520 } else {
521 /* Cannot get a declaration, and no declaration is stored in the buffer. It is pointless to preload
522 * now. When the buffer is used, PreLoad will be called by the stream source state handler and a valid
523 * declaration for the buffer can be found
525 return;
528 /* If applications change the declaration over and over, reconverting all the time is a huge
529 * performance hit. So count the declaration changes and release the VBO if there are too many
530 * of them (and thus stop converting)
532 if(declChanged) {
533 This->declChanges++;
534 This->draws = 0;
536 if(This->declChanges > VB_MAXDECLCHANGES) {
537 FIXME("Too many declaration changes, stopping converting\n");
538 ActivateContext(device, device->lastActiveRenderTarget, CTXUSAGE_RESOURCELOAD);
539 ENTER_GL();
540 GL_EXTCALL(glDeleteBuffersARB(1, &This->vbo));
541 checkGLcall("glDeleteBuffersARB");
542 LEAVE_GL();
543 This->vbo = 0;
544 HeapFree(GetProcessHeap(), 0, This->conv_shift);
546 /* The stream source state handler might have read the memory of the vertex buffer already
547 * and got the memory in the vbo which is not valid any longer. Dirtify the stream source
548 * to force a reload. This happens only once per changed vertexbuffer and should occur rather
549 * rarely
551 IWineD3DDeviceImpl_MarkStateDirty(device, STATE_STREAMSRC);
553 return;
555 check_vbo_size(This);
556 } else {
557 /* However, it is perfectly fine to change the declaration every now and then. We don't want a game that
558 * changes it every minute drop the VBO after VB_MAX_DECL_CHANGES minutes. So count draws without
559 * decl changes and reset the decl change count after a specific number of them
561 This->draws++;
562 if(This->draws > VB_RESETDECLCHANGE) This->declChanges = 0;
565 if(declChanged) {
566 /* The declaration changed, reload the whole buffer */
567 WARN("Reloading buffer because of decl change\n");
568 start = 0;
569 end = This->resource.size;
570 } else if(This->Flags & VBFLAG_DIRTY) {
571 /* No decl change, but dirty data, reload the changed stuff */
572 if(This->conv_shift) {
573 if(This->dirtystart != 0 || This->dirtyend != 0) {
574 FIXME("Implement partial buffer loading with shifted conversion\n");
577 start = This->dirtystart;
578 end = This->dirtyend;
579 } else {
580 /* Desc not changed, buffer not dirty, nothing to do :-) */
581 return;
584 /* Mark the buffer clean */
585 This->Flags &= ~VBFLAG_DIRTY;
586 This->dirtystart = 0;
587 This->dirtyend = 0;
589 if(!This->conv_map) {
590 /* That means that there is nothing to fixup. Just upload from This->resource.allocatedMemory
591 * directly into the vbo. Do not free the system memory copy because drawPrimitive may need it if
592 * the stride is 0, for instancing emulation, vertex blending emulation or shader emulation.
594 TRACE("No conversion needed\n");
596 if(!device->isInDraw) {
597 ActivateContext(device, device->lastActiveRenderTarget, CTXUSAGE_RESOURCELOAD);
599 ENTER_GL();
600 GL_EXTCALL(glBindBufferARB(GL_ARRAY_BUFFER_ARB, This->vbo));
601 checkGLcall("glBindBufferARB");
602 GL_EXTCALL(glBufferSubDataARB(GL_ARRAY_BUFFER_ARB, start, end-start, This->resource.allocatedMemory + start));
603 checkGLcall("glBufferSubDataARB");
604 LEAVE_GL();
605 return;
608 /* Now for each vertex in the buffer that needs conversion */
609 vertices = This->resource.size / This->stride;
611 if(This->conv_shift) {
612 TRACE("Shifted conversion\n");
613 data = HeapAlloc(GetProcessHeap(), 0, vertices * This->conv_stride);
615 for(i = start / This->stride; i < min((end / This->stride) + 1, vertices); i++) {
616 for(j = 0; j < This->stride; j++) {
617 switch(This->conv_map[j]) {
618 case CONV_NONE:
619 data[This->conv_stride * i + j + This->conv_shift[j]] = This->resource.allocatedMemory[This->stride * i + j];
620 break;
622 case CONV_FLOAT16_2:
624 float *out = (float *) (&data[This->conv_stride * i + j + This->conv_shift[j]]);
625 WORD *in = (WORD *) (&This->resource.allocatedMemory[i * This->stride + j]);
627 out[1] = float_16_to_32(in + 1);
628 out[0] = float_16_to_32(in + 0);
629 j += 3; /* Skip 3 additional bytes,as a FLOAT16_2 has 4 bytes */
630 break;
633 default:
634 FIXME("Unimplemented conversion %d in shifted conversion\n", This->conv_map[j]);
639 ENTER_GL();
640 GL_EXTCALL(glBindBufferARB(GL_ARRAY_BUFFER_ARB, This->vbo));
641 checkGLcall("glBindBufferARB");
642 GL_EXTCALL(glBufferSubDataARB(GL_ARRAY_BUFFER_ARB, 0, vertices * This->conv_stride, data));
643 checkGLcall("glBufferSubDataARB");
644 LEAVE_GL();
646 } else {
647 data = HeapAlloc(GetProcessHeap(), 0, This->resource.size);
648 memcpy(data + start, This->resource.allocatedMemory + start, end - start);
649 for(i = start / This->stride; i < min((end / This->stride) + 1, vertices); i++) {
650 for(j = 0; j < This->stride; j++) {
651 switch(This->conv_map[j]) {
652 case CONV_NONE:
653 /* Done already */
654 j += 3;
655 break;
656 case CONV_D3DCOLOR:
657 fixup_d3dcolor((DWORD *) (data + i * This->stride + j));
658 j += 3;
659 break;
661 case CONV_POSITIONT:
662 fixup_transformed_pos((float *) (data + i * This->stride + j));
663 j += 15;
664 break;
666 case CONV_FLOAT16_2:
667 ERR("Did not expect FLOAT16 conversion in unshifted conversion\n");
668 default:
669 FIXME("Unimplemented conversion %d in shifted conversion\n", This->conv_map[j]);
674 ENTER_GL();
675 GL_EXTCALL(glBindBufferARB(GL_ARRAY_BUFFER_ARB, This->vbo));
676 checkGLcall("glBindBufferARB");
677 GL_EXTCALL(glBufferSubDataARB(GL_ARRAY_BUFFER_ARB, start, end - start, data + start));
678 checkGLcall("glBufferSubDataARB");
679 LEAVE_GL();
682 HeapFree(GetProcessHeap(), 0, data);
685 static void WINAPI IWineD3DVertexBufferImpl_UnLoad(IWineD3DVertexBuffer *iface) {
686 IWineD3DVertexBufferImpl *This = (IWineD3DVertexBufferImpl *) iface;
687 IWineD3DDeviceImpl *device = This->resource.wineD3DDevice;
688 TRACE("(%p)\n", This);
690 /* This is easy: The whole content is shadowed in This->resource.allocatedMemory,
691 * so we only have to destroy the vbo. Only do it if we have a vbo, which implies
692 * that vbos are supported
694 if(This->vbo) {
695 ActivateContext(device, device->lastActiveRenderTarget, CTXUSAGE_RESOURCELOAD);
696 ENTER_GL();
697 GL_EXTCALL(glDeleteBuffersARB(1, &This->vbo));
698 checkGLcall("glDeleteBuffersARB");
699 LEAVE_GL();
700 This->vbo = 0;
701 This->Flags |= VBFLAG_CREATEVBO; /* Recreate the VBO next load */
705 static WINED3DRESOURCETYPE WINAPI IWineD3DVertexBufferImpl_GetType(IWineD3DVertexBuffer *iface) {
706 return IWineD3DResourceImpl_GetType((IWineD3DResource *)iface);
709 static HRESULT WINAPI IWineD3DVertexBufferImpl_GetParent(IWineD3DVertexBuffer *iface, IUnknown **pParent) {
710 return IWineD3DResourceImpl_GetParent((IWineD3DResource *)iface, pParent);
713 /* ******************************************************
714 IWineD3DVertexBuffer IWineD3DVertexBuffer parts follow
715 ****************************************************** */
716 static HRESULT WINAPI IWineD3DVertexBufferImpl_Lock(IWineD3DVertexBuffer *iface, UINT OffsetToLock, UINT SizeToLock, BYTE** ppbData, DWORD Flags) {
717 IWineD3DVertexBufferImpl *This = (IWineD3DVertexBufferImpl *)iface;
718 BYTE *data;
719 TRACE("(%p)->%d, %d, %p, %08x\n", This, OffsetToLock, SizeToLock, ppbData, Flags);
721 InterlockedIncrement(&This->lockcount);
723 if(This->Flags & VBFLAG_DIRTY) {
724 if(This->dirtystart > OffsetToLock) This->dirtystart = OffsetToLock;
725 if(SizeToLock) {
726 if(This->dirtyend < OffsetToLock + SizeToLock) This->dirtyend = OffsetToLock + SizeToLock;
727 } else {
728 This->dirtyend = This->resource.size;
730 } else {
731 This->dirtystart = OffsetToLock;
732 if(SizeToLock)
733 This->dirtyend = OffsetToLock + SizeToLock;
734 else
735 This->dirtyend = This->resource.size;
738 data = This->resource.allocatedMemory;
739 This->Flags |= VBFLAG_DIRTY;
740 *ppbData = data + OffsetToLock;
742 TRACE("(%p) : returning memory of %p (base:%p,offset:%u)\n", This, data + OffsetToLock, data, OffsetToLock);
743 /* TODO: check Flags compatibility with This->currentDesc.Usage (see MSDN) */
744 return WINED3D_OK;
746 HRESULT WINAPI IWineD3DVertexBufferImpl_Unlock(IWineD3DVertexBuffer *iface) {
747 IWineD3DVertexBufferImpl *This = (IWineD3DVertexBufferImpl *) iface;
748 LONG lockcount;
749 TRACE("(%p)\n", This);
751 lockcount = InterlockedDecrement(&This->lockcount);
752 if(lockcount > 0) {
753 /* Delay loading the buffer until everything is unlocked */
754 TRACE("Ignoring the unlock\n");
755 return WINED3D_OK;
758 if(This->Flags & VBFLAG_HASDESC) {
759 IWineD3DVertexBufferImpl_PreLoad(iface);
761 return WINED3D_OK;
763 static HRESULT WINAPI IWineD3DVertexBufferImpl_GetDesc(IWineD3DVertexBuffer *iface, WINED3DVERTEXBUFFER_DESC *pDesc) {
764 IWineD3DVertexBufferImpl *This = (IWineD3DVertexBufferImpl *)iface;
766 TRACE("(%p)\n", This);
767 pDesc->Format = This->resource.format;
768 pDesc->Type = This->resource.resourceType;
769 pDesc->Usage = This->resource.usage;
770 pDesc->Pool = This->resource.pool;
771 pDesc->Size = This->resource.size;
772 pDesc->FVF = This->fvf;
773 return WINED3D_OK;
776 const IWineD3DVertexBufferVtbl IWineD3DVertexBuffer_Vtbl =
778 /* IUnknown */
779 IWineD3DVertexBufferImpl_QueryInterface,
780 IWineD3DVertexBufferImpl_AddRef,
781 IWineD3DVertexBufferImpl_Release,
782 /* IWineD3DResource */
783 IWineD3DVertexBufferImpl_GetParent,
784 IWineD3DVertexBufferImpl_GetDevice,
785 IWineD3DVertexBufferImpl_SetPrivateData,
786 IWineD3DVertexBufferImpl_GetPrivateData,
787 IWineD3DVertexBufferImpl_FreePrivateData,
788 IWineD3DVertexBufferImpl_SetPriority,
789 IWineD3DVertexBufferImpl_GetPriority,
790 IWineD3DVertexBufferImpl_PreLoad,
791 IWineD3DVertexBufferImpl_UnLoad,
792 IWineD3DVertexBufferImpl_GetType,
793 /* IWineD3DVertexBuffer */
794 IWineD3DVertexBufferImpl_Lock,
795 IWineD3DVertexBufferImpl_Unlock,
796 IWineD3DVertexBufferImpl_GetDesc
799 BYTE* WINAPI IWineD3DVertexBufferImpl_GetMemory(IWineD3DVertexBuffer* iface, DWORD iOffset, GLint *vbo) {
800 IWineD3DVertexBufferImpl *This = (IWineD3DVertexBufferImpl *)iface;
802 *vbo = This->vbo;
803 if(This->vbo == 0) {
804 return This->resource.allocatedMemory + iOffset;
805 } else {
806 return (BYTE *) iOffset;
810 HRESULT WINAPI IWineD3DVertexBufferImpl_ReleaseMemory(IWineD3DVertexBuffer* iface) {
811 return WINED3D_OK;