push 212f1dad91f15aefd8e676124180e0b86d7c9ee6
[wine/hacks.git] / dlls / wined3d / vertexbuffer.c
blobb0a6296b7315b32c51c9c92206b5bf12dd15b064
1 /*
2 * IWineD3DVertexBuffer Implementation
4 * Copyright 2002-2005 Jason Edmeades
5 * Raphael Junqueira
6 * Copyright 2004 Christian Costa
7 * Copyright 2007 Stefan Dösinger for CodeWeavers
9 * This library is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU Lesser General Public
11 * License as published by the Free Software Foundation; either
12 * version 2.1 of the License, or (at your option) any later version.
14 * This library is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 * Lesser General Public License for more details.
19 * You should have received a copy of the GNU Lesser General Public
20 * License along with this library; if not, write to the Free Software
21 * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301, USA
24 #include "config.h"
25 #include "wined3d_private.h"
27 WINE_DEFAULT_DEBUG_CHANNEL(d3d);
28 #define GLINFO_LOCATION This->resource.wineD3DDevice->adapter->gl_info
30 #define VB_MAXDECLCHANGES 100 /* After that number we stop converting */
31 #define VB_RESETDECLCHANGE 1000 /* Reset the changecount after that number of draws */
33 /* *******************************************
34 IWineD3DVertexBuffer IUnknown parts follow
35 ******************************************* */
36 static HRESULT WINAPI IWineD3DVertexBufferImpl_QueryInterface(IWineD3DVertexBuffer *iface, REFIID riid, LPVOID *ppobj)
38 IWineD3DVertexBufferImpl *This = (IWineD3DVertexBufferImpl *)iface;
39 TRACE("(%p)->(%s,%p)\n",This,debugstr_guid(riid),ppobj);
40 if (IsEqualGUID(riid, &IID_IUnknown)
41 || IsEqualGUID(riid, &IID_IWineD3DBase)
42 || IsEqualGUID(riid, &IID_IWineD3DResource)
43 || IsEqualGUID(riid, &IID_IWineD3DVertexBuffer)){
44 IUnknown_AddRef(iface);
45 *ppobj = This;
46 return S_OK;
48 *ppobj = NULL;
49 return E_NOINTERFACE;
52 static ULONG WINAPI IWineD3DVertexBufferImpl_AddRef(IWineD3DVertexBuffer *iface) {
53 IWineD3DVertexBufferImpl *This = (IWineD3DVertexBufferImpl *)iface;
54 ULONG ref = InterlockedIncrement(&This->resource.ref);
55 TRACE("(%p) : AddRef increasing from %d\n", This, ref - 1);
56 return ref;
59 static ULONG WINAPI IWineD3DVertexBufferImpl_Release(IWineD3DVertexBuffer *iface) {
60 IWineD3DVertexBufferImpl *This = (IWineD3DVertexBufferImpl *)iface;
61 ULONG ref = InterlockedDecrement(&This->resource.ref);
62 TRACE("(%p) : Releasing from %d\n", This, ref + 1);
63 if (ref == 0) {
65 if(This->vbo) {
66 IWineD3DDeviceImpl *device = This->resource.wineD3DDevice;
68 ActivateContext(device, device->lastActiveRenderTarget, CTXUSAGE_RESOURCELOAD);
69 ENTER_GL();
70 GL_EXTCALL(glDeleteBuffersARB(1, &This->vbo));
71 checkGLcall("glDeleteBuffersARB");
72 LEAVE_GL();
75 resource_cleanup((IWineD3DResource *)iface);
76 HeapFree(GetProcessHeap(), 0, This);
78 return ref;
81 /* ****************************************************
82 IWineD3DVertexBuffer IWineD3DResource parts follow
83 **************************************************** */
84 static HRESULT WINAPI IWineD3DVertexBufferImpl_GetDevice(IWineD3DVertexBuffer *iface, IWineD3DDevice** ppDevice) {
85 return resource_get_device((IWineD3DResource *)iface, ppDevice);
88 static HRESULT WINAPI IWineD3DVertexBufferImpl_SetPrivateData(IWineD3DVertexBuffer *iface, REFGUID refguid, CONST void* pData, DWORD SizeOfData, DWORD Flags) {
89 return resource_set_private_data((IWineD3DResource *)iface, refguid, pData, SizeOfData, Flags);
92 static HRESULT WINAPI IWineD3DVertexBufferImpl_GetPrivateData(IWineD3DVertexBuffer *iface, REFGUID refguid, void* pData, DWORD* pSizeOfData) {
93 return resource_get_private_data((IWineD3DResource *)iface, refguid, pData, pSizeOfData);
96 static HRESULT WINAPI IWineD3DVertexBufferImpl_FreePrivateData(IWineD3DVertexBuffer *iface, REFGUID refguid) {
97 return resource_free_private_data((IWineD3DResource *)iface, refguid);
100 static DWORD WINAPI IWineD3DVertexBufferImpl_SetPriority(IWineD3DVertexBuffer *iface, DWORD PriorityNew) {
101 return resource_set_priority((IWineD3DResource *)iface, PriorityNew);
104 static DWORD WINAPI IWineD3DVertexBufferImpl_GetPriority(IWineD3DVertexBuffer *iface) {
105 return resource_get_priority((IWineD3DResource *)iface);
108 static inline void fixup_d3dcolor(DWORD *pos) {
109 DWORD srcColor = *pos;
111 /* Color conversion like in drawStridedSlow. watch out for little endianity
112 * If we want that stuff to work on big endian machines too we have to consider more things
114 * 0xff000000: Alpha mask
115 * 0x00ff0000: Blue mask
116 * 0x0000ff00: Green mask
117 * 0x000000ff: Red mask
119 *pos = 0;
120 *pos |= (srcColor & 0xff00ff00) ; /* Alpha Green */
121 *pos |= (srcColor & 0x00ff0000) >> 16; /* Red */
122 *pos |= (srcColor & 0x000000ff) << 16; /* Blue */
125 static inline void fixup_transformed_pos(float *p) {
126 float x, y, z, w;
128 /* rhw conversion like in drawStridedSlow */
129 if(p[3] == 1.0 || ((p[3] < eps) && (p[3] > -eps))) {
130 x = p[0];
131 y = p[1];
132 z = p[2];
133 w = 1.0;
134 } else {
135 w = 1.0 / p[3];
136 x = p[0] * w;
137 y = p[1] * w;
138 z = p[2] * w;
140 p[0] = x;
141 p[1] = y;
142 p[2] = z;
143 p[3] = w;
146 static DWORD *find_conversion_shift(IWineD3DVertexBufferImpl *This, const WineDirect3DVertexStridedData *strided, DWORD stride)
148 DWORD *ret, i, shift, j, type;
149 DWORD orig_type_size;
151 if(!stride) {
152 TRACE("No shift\n");
153 return NULL;
156 This->conv_stride = stride;
157 ret = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(DWORD) * stride);
158 for(i = 0; i < MAX_ATTRIBS; i++) {
159 if(strided->u.input[i].VBO != This->vbo) continue;
161 type = strided->u.input[i].dwType;
162 if(type == WINED3DDECLTYPE_FLOAT16_2) {
163 shift = 4;
164 } else if(type == WINED3DDECLTYPE_FLOAT16_4) {
165 shift = 8;
166 /* Pre-shift the last 4 bytes in the FLOAT16_4 by 4 bytes - this makes FLOAT16_2 and FLOAT16_4 conversions
167 * compatible
169 for(j = 4; j < 8; j++) {
170 ret[(DWORD_PTR) strided->u.input[i].lpData + j ] += 4;
172 } else {
173 shift = 0;
175 This->conv_stride += shift;
177 if(shift) {
178 orig_type_size = WINED3D_ATR_TYPESIZE(type) * WINED3D_ATR_SIZE(type);
179 for(j = (DWORD_PTR) strided->u.input[i].lpData + orig_type_size; j < stride; j++) {
180 ret[j] += shift;
185 if(TRACE_ON(d3d)) {
186 TRACE("Dumping conversion shift:\n");
187 for(i = 0; i < stride; i++) {
188 TRACE("[%d]", ret[i]);
190 TRACE("\n");
192 return ret;
195 static inline BOOL process_converted_attribute(IWineD3DVertexBufferImpl *This,
196 const enum vbo_conversion_type conv_type,
197 const WineDirect3DStridedData *attrib,
198 DWORD *stride_this_run, const DWORD type) {
199 DWORD attrib_size;
200 BOOL ret = FALSE;
201 int i;
202 DWORD offset = This->resource.wineD3DDevice->stateBlock->streamOffset[attrib->streamNo];
203 DWORD_PTR data;
205 /* Check for some valid situations which cause us pain. One is if the buffer is used for
206 * constant attributes(stride = 0), the other one is if the buffer is used on two streams
207 * with different strides. In the 2nd case we might have to drop conversion entirely,
208 * it is possible that the same bytes are once read as FLOAT2 and once as UBYTE4N.
210 if(attrib->dwStride == 0) {
211 FIXME("%s used with stride 0, let's hope we get the vertex stride from somewhere else\n",
212 debug_d3ddecltype(type));
213 } else if(attrib->dwStride != *stride_this_run &&
214 *stride_this_run) {
215 FIXME("Got two concurrent strides, %d and %d\n", attrib->dwStride, *stride_this_run);
216 } else {
217 *stride_this_run = attrib->dwStride;
218 if(This->stride != *stride_this_run) {
219 /* We rely that this happens only on the first converted attribute that is found,
220 * if at all. See above check
222 TRACE("Reconverting because converted attributes occur, and the stride changed\n");
223 This->stride = *stride_this_run;
224 HeapFree(GetProcessHeap(), HEAP_ZERO_MEMORY, This->conv_map);
225 This->conv_map = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*This->conv_map) * This->stride);
226 ret = TRUE;
230 data = (((DWORD_PTR) attrib->lpData) + offset) % This->stride;
231 attrib_size = WINED3D_ATR_SIZE(type) * WINED3D_ATR_TYPESIZE(type);
232 for(i = 0; i < attrib_size; i++) {
233 if(This->conv_map[data + i] != conv_type) {
234 TRACE("Byte %ld in vertex changed\n", i + data);
235 TRACE("It was type %d, is %d now\n", This->conv_map[data + i], conv_type);
236 ret = TRUE;
237 This->conv_map[data + i] = conv_type;
240 return ret;
243 static inline BOOL check_attribute(IWineD3DVertexBufferImpl *This, const WineDirect3DStridedData *attrib,
244 const BOOL check_d3dcolor, const BOOL is_ffp_position, const BOOL is_ffp_color,
245 DWORD *stride_this_run, BOOL *float16_used) {
246 BOOL ret = FALSE;
247 DWORD type;
249 /* Ignore attributes that do not have our vbo. After that check we can be sure that the attribute is
250 * there, on nonexistent attribs the vbo is 0.
252 if(attrib->VBO != This->vbo) return FALSE;
254 type = attrib->dwType;
255 /* Look for newly appeared conversion */
256 if(!GL_SUPPORT(NV_HALF_FLOAT) && (
257 type == WINED3DDECLTYPE_FLOAT16_2 ||
258 type == WINED3DDECLTYPE_FLOAT16_4)) {
260 ret = process_converted_attribute(This, CONV_FLOAT16_2, attrib, stride_this_run, type);
262 if(is_ffp_position) {
263 FIXME("Test FLOAT16 fixed function processing positions\n");
264 } else if(is_ffp_color) {
265 FIXME("test FLOAT16 fixed function processing colors\n");
267 *float16_used = TRUE;
268 } else if(check_d3dcolor && type == WINED3DDECLTYPE_D3DCOLOR) {
270 ret = process_converted_attribute(This, CONV_D3DCOLOR, attrib, stride_this_run, WINED3DDECLTYPE_D3DCOLOR);
272 if(!is_ffp_color) {
273 FIXME("Test for non-color fixed function D3DCOLOR type\n");
275 } else if(is_ffp_position && type == WINED3DDECLTYPE_FLOAT4) {
276 ret = process_converted_attribute(This, CONV_POSITIONT, attrib, stride_this_run, WINED3DDECLTYPE_FLOAT4);
277 } else if(This->conv_map) {
278 ret = process_converted_attribute(This, CONV_NONE, attrib, stride_this_run, type);
280 return ret;
283 static inline BOOL IWineD3DVertexBufferImpl_FindDecl(IWineD3DVertexBufferImpl *This)
285 IWineD3DDeviceImpl *device = This->resource.wineD3DDevice;
286 BOOL ret = FALSE;
287 int i;
288 DWORD stride_this_run = 0;
289 BOOL float16_used = FALSE;
291 /* In d3d7 the vertex buffer declaration NEVER changes because it is stored in the d3d7 vertex buffer.
292 * Once we have our declaration there is no need to look it up again.
294 if(((IWineD3DImpl *)device->wineD3D)->dxVersion == 7 && This->Flags & VBFLAG_HASDESC) {
295 return FALSE;
298 TRACE("Finding vertex buffer conversion information\n");
299 /* Certain declaration types need some fixups before we can pass them to opengl. This means D3DCOLOR attributes with fixed
300 * function vertex processing, FLOAT4 POSITIONT with fixed function, and FLOAT16 if GL_NV_half_float is not supported.
302 * The vertex buffer FVF doesn't help with finding them, we have to use the decoded vertex declaration and pick the things
303 * that concern the current buffer. A problem with this is that this can change between draws, so we have to validate
304 * the information and reprocess the buffer if it changes, and avoid false positives for performance reasons.
306 * We have to distinguish between vertex shaders and fixed function to pick the way we access the
307 * strided vertex information.
309 * This code sets up a per-byte array with the size of the detected stride of the arrays in the
310 * buffer. For each byte we have a field that marks the conversion needed on this byte. For example,
311 * the following declaration with fixed function vertex processing:
313 * POSITIONT, FLOAT4
314 * NORMAL, FLOAT3
315 * DIFFUSE, FLOAT16_4
316 * SPECULAR, D3DCOLOR
318 * Will result in
319 * { POSITIONT }{ NORMAL }{ DIFFUSE }{SPECULAR }
320 * [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]
322 * Where in this example map P means 4 component position conversion, 0 means no conversion, F means FLOAT16_2 conversion
323 * and C means D3DCOLOR conversion(red / blue swizzle).
325 * If we're doing conversion and the stride changes we have to reconvert the whole buffer. Note that we do not mind if the
326 * semantic changes, we only care for the conversion type. So if the NORMAL is replaced with a TEXCOORD, nothing has to be
327 * done, or if the DIFFUSE is replaced with a D3DCOLOR BLENDWEIGHT we can happily dismiss the change. Some conversion types
328 * depend on the semantic as well, for example a FLOAT4 texcoord needs no conversion while a FLOAT4 positiont needs one
330 if (use_vs(device->stateBlock))
332 TRACE("vshader\n");
333 /* If the current vertex declaration is marked for no half float conversion don't bother to
334 * analyse the strided streams in depth, just set them up for no conversion. Return decl changed
335 * if we used conversion before
337 if(!((IWineD3DVertexDeclarationImpl *) device->stateBlock->vertexDecl)->half_float_conv_needed) {
338 if(This->conv_map) {
339 TRACE("Now using shaders without conversion, but conversion used before\n");
340 HeapFree(GetProcessHeap(), 0, This->conv_map);
341 HeapFree(GetProcessHeap(), 0, This->conv_shift);
342 This->conv_map = NULL;
343 This->stride = 0;
344 This->conv_shift = NULL;
345 This->conv_stride = 0;
346 return TRUE;
347 } else {
348 return FALSE;
351 for(i = 0; i < MAX_ATTRIBS; i++) {
352 ret = check_attribute(This, &device->strided_streams.u.input[i], FALSE, FALSE, FALSE, &stride_this_run, &float16_used) || ret;
355 /* Recalculate the conversion shift map if the declaration has changed,
356 * and we're using float16 conversion or used it on the last run
358 if(ret && (float16_used || This->conv_map)) {
359 HeapFree(GetProcessHeap(), 0, This->conv_shift);
360 This->conv_shift = find_conversion_shift(This, &device->strided_streams, This->stride);
362 } else {
363 /* Fixed function is a bit trickier. We have to take care for D3DCOLOR types, FLOAT4 positions and of course
364 * FLOAT16s if not supported. Also, we can't iterate over the array, so use macros to generate code for all
365 * the attributes that our current fixed function pipeline implementation cares for.
367 ret = check_attribute(This, &device->strided_streams.u.s.position, TRUE, TRUE, FALSE, &stride_this_run, &float16_used) || ret;
368 ret = check_attribute(This, &device->strided_streams.u.s.normal, TRUE, FALSE, FALSE, &stride_this_run, &float16_used) || ret;
369 ret = check_attribute(This, &device->strided_streams.u.s.diffuse, TRUE, FALSE, TRUE, &stride_this_run, &float16_used) || ret;
370 ret = check_attribute(This, &device->strided_streams.u.s.specular, TRUE, FALSE, TRUE, &stride_this_run, &float16_used) || ret;
371 ret = check_attribute(This, &device->strided_streams.u.s.texCoords[0], TRUE, FALSE, FALSE, &stride_this_run, &float16_used) || ret;
372 ret = check_attribute(This, &device->strided_streams.u.s.texCoords[1], TRUE, FALSE, FALSE, &stride_this_run, &float16_used) || ret;
373 ret = check_attribute(This, &device->strided_streams.u.s.texCoords[2], TRUE, FALSE, FALSE, &stride_this_run, &float16_used) || ret;
374 ret = check_attribute(This, &device->strided_streams.u.s.texCoords[3], TRUE, FALSE, FALSE, &stride_this_run, &float16_used) || ret;
375 ret = check_attribute(This, &device->strided_streams.u.s.texCoords[4], TRUE, FALSE, FALSE, &stride_this_run, &float16_used) || ret;
376 ret = check_attribute(This, &device->strided_streams.u.s.texCoords[5], TRUE, FALSE, FALSE, &stride_this_run, &float16_used) || ret;
377 ret = check_attribute(This, &device->strided_streams.u.s.texCoords[6], TRUE, FALSE, FALSE, &stride_this_run, &float16_used) || ret;
378 ret = check_attribute(This, &device->strided_streams.u.s.texCoords[7], TRUE, FALSE, FALSE, &stride_this_run, &float16_used) || ret;
380 if(float16_used) FIXME("Float16 conversion used with fixed function vertex processing\n");
383 if(stride_this_run == 0 && This->conv_map) {
384 /* Sanity test */
385 if(ret == FALSE) {
386 ERR("no converted attributes found, old conversion map exists, and no declaration change???\n");
388 HeapFree(GetProcessHeap(), 0, This->conv_map);
389 This->conv_map = NULL;
390 This->stride = 0;
392 This->Flags |= VBFLAG_HASDESC;
394 if(ret) TRACE("Conversion information changed\n");
395 return ret;
398 static void check_vbo_size(IWineD3DVertexBufferImpl *This) {
399 DWORD size = This->conv_stride ? This->conv_stride * (This->resource.size / This->stride) : This->resource.size;
400 if(This->vbo_size != size) {
401 TRACE("Old size %d, creating new size %d\n", This->vbo_size, size);
402 ENTER_GL();
403 GL_EXTCALL(glBindBufferARB(GL_ARRAY_BUFFER_ARB, This->vbo));
404 checkGLcall("glBindBufferARB");
405 GL_EXTCALL(glBufferDataARB(GL_ARRAY_BUFFER_ARB, size, NULL, This->vbo_usage));
406 This->vbo_size = size;
407 checkGLcall("glBufferDataARB");
408 LEAVE_GL();
412 static void CreateVBO(IWineD3DVertexBufferImpl *This) {
413 GLenum error, glUsage;
414 DWORD vboUsage = This->resource.usage;
415 IWineD3DDeviceImpl *device = This->resource.wineD3DDevice;
417 TRACE("Creating an OpenGL vertex buffer object for IWineD3DVertexBuffer %p Usage(%s)\n", This, debug_d3dusage(vboUsage));
419 /* Make sure that a context is there. Needed in a multithreaded environment. Otherwise this call is a nop */
420 ActivateContext(device, device->lastActiveRenderTarget, CTXUSAGE_RESOURCELOAD);
421 ENTER_GL();
423 /* Make sure that the gl error is cleared. Do not use checkGLcall
424 * here because checkGLcall just prints a fixme and continues. However,
425 * if an error during VBO creation occurs we can fall back to non-vbo operation
426 * with full functionality(but performance loss)
428 while(glGetError() != GL_NO_ERROR);
430 /* Basically the FVF parameter passed to CreateVertexBuffer is no good
431 * It is the FVF set with IWineD3DDevice::SetFVF or the Vertex Declaration set with
432 * IWineD3DDevice::SetVertexDeclaration that decides how the vertices in the buffer
433 * look like. This means that on each DrawPrimitive call the vertex buffer has to be verified
434 * to check if the rhw and color values are in the correct format.
437 GL_EXTCALL(glGenBuffersARB(1, &This->vbo));
438 error = glGetError();
439 if(This->vbo == 0 || error != GL_NO_ERROR) {
440 WARN("Failed to create a VBO with error %s (%#x)\n", debug_glerror(error), error);
441 goto error;
444 GL_EXTCALL(glBindBufferARB(GL_ARRAY_BUFFER_ARB, This->vbo));
445 error = glGetError();
446 if(error != GL_NO_ERROR) {
447 WARN("Failed to bind the VBO with error %s (%#x)\n", debug_glerror(error), error);
448 goto error;
451 /* Don't use static, because dx apps tend to update the buffer
452 * quite often even if they specify 0 usage. Because we always keep the local copy
453 * we never read from the vbo and can create a write only opengl buffer.
455 switch(vboUsage & (WINED3DUSAGE_WRITEONLY | WINED3DUSAGE_DYNAMIC) ) {
456 case WINED3DUSAGE_WRITEONLY | WINED3DUSAGE_DYNAMIC:
457 case WINED3DUSAGE_DYNAMIC:
458 TRACE("Gl usage = GL_STREAM_DRAW\n");
459 glUsage = GL_STREAM_DRAW_ARB;
460 break;
461 case WINED3DUSAGE_WRITEONLY:
462 default:
463 TRACE("Gl usage = GL_DYNAMIC_DRAW\n");
464 glUsage = GL_DYNAMIC_DRAW_ARB;
465 break;
468 /* Reserve memory for the buffer. The amount of data won't change
469 * so we are safe with calling glBufferData once with a NULL ptr and
470 * calling glBufferSubData on updates
472 GL_EXTCALL(glBufferDataARB(GL_ARRAY_BUFFER_ARB, This->resource.size, NULL, glUsage));
473 error = glGetError();
474 if(error != GL_NO_ERROR) {
475 WARN("glBufferDataARB failed with error %s (%#x)\n", debug_glerror(error), error);
476 goto error;
478 This->vbo_size = This->resource.size;
479 This->vbo_usage = glUsage;
480 This->dirtystart = 0;
481 This->dirtyend = This->resource.size;
482 This->Flags |= VBFLAG_DIRTY;
484 LEAVE_GL();
486 return;
487 error:
488 /* Clean up all vbo init, but continue because we can work without a vbo :-) */
489 FIXME("Failed to create a vertex buffer object. Continuing, but performance issues can occur\n");
490 if(This->vbo) GL_EXTCALL(glDeleteBuffersARB(1, &This->vbo));
491 This->vbo = 0;
492 LEAVE_GL();
493 return;
496 static void WINAPI IWineD3DVertexBufferImpl_PreLoad(IWineD3DVertexBuffer *iface) {
497 IWineD3DVertexBufferImpl *This = (IWineD3DVertexBufferImpl *) iface;
498 IWineD3DDeviceImpl *device = This->resource.wineD3DDevice;
499 BYTE *data;
500 UINT start = 0, end = 0, vertices;
501 BOOL declChanged = FALSE;
502 int i, j;
503 TRACE("(%p)->()\n", This);
505 if(!This->vbo) {
506 /* TODO: Make converting independent from VBOs */
507 if(This->Flags & VBFLAG_CREATEVBO) {
508 CreateVBO(This);
509 This->Flags &= ~VBFLAG_CREATEVBO;
510 } else {
511 return; /* Not doing any conversion */
515 /* Reading the declaration makes only sense if the stateblock is finalized and the buffer bound to a stream */
516 if(device->isInDraw && This->bindCount > 0) {
517 declChanged = IWineD3DVertexBufferImpl_FindDecl(This);
518 } else if(This->Flags & VBFLAG_HASDESC) {
519 /* Reuse the declaration stored in the buffer. It will most likely not change, and if it does
520 * the stream source state handler will call PreLoad again and the change will be caught
522 } else {
523 /* Cannot get a declaration, and no declaration is stored in the buffer. It is pointless to preload
524 * now. When the buffer is used, PreLoad will be called by the stream source state handler and a valid
525 * declaration for the buffer can be found
527 return;
530 /* If applications change the declaration over and over, reconverting all the time is a huge
531 * performance hit. So count the declaration changes and release the VBO if there are too many
532 * of them (and thus stop converting)
534 if(declChanged) {
535 This->declChanges++;
536 This->draws = 0;
538 if(This->declChanges > VB_MAXDECLCHANGES) {
539 FIXME("Too many declaration changes, stopping converting\n");
540 ActivateContext(device, device->lastActiveRenderTarget, CTXUSAGE_RESOURCELOAD);
541 ENTER_GL();
542 GL_EXTCALL(glDeleteBuffersARB(1, &This->vbo));
543 checkGLcall("glDeleteBuffersARB");
544 LEAVE_GL();
545 This->vbo = 0;
546 HeapFree(GetProcessHeap(), 0, This->conv_shift);
548 /* The stream source state handler might have read the memory of the vertex buffer already
549 * and got the memory in the vbo which is not valid any longer. Dirtify the stream source
550 * to force a reload. This happens only once per changed vertexbuffer and should occur rather
551 * rarely
553 IWineD3DDeviceImpl_MarkStateDirty(device, STATE_STREAMSRC);
555 return;
557 check_vbo_size(This);
558 } else {
559 /* However, it is perfectly fine to change the declaration every now and then. We don't want a game that
560 * changes it every minute drop the VBO after VB_MAX_DECL_CHANGES minutes. So count draws without
561 * decl changes and reset the decl change count after a specific number of them
563 This->draws++;
564 if(This->draws > VB_RESETDECLCHANGE) This->declChanges = 0;
567 if(declChanged) {
568 /* The declaration changed, reload the whole buffer */
569 WARN("Reloading buffer because of decl change\n");
570 start = 0;
571 end = This->resource.size;
572 } else if(This->Flags & VBFLAG_DIRTY) {
573 /* No decl change, but dirty data, reload the changed stuff */
574 if(This->conv_shift) {
575 if(This->dirtystart != 0 || This->dirtyend != 0) {
576 FIXME("Implement partial buffer loading with shifted conversion\n");
579 start = This->dirtystart;
580 end = This->dirtyend;
581 } else {
582 /* Desc not changed, buffer not dirty, nothing to do :-) */
583 return;
586 /* Mark the buffer clean */
587 This->Flags &= ~VBFLAG_DIRTY;
588 This->dirtystart = 0;
589 This->dirtyend = 0;
591 if(!This->conv_map) {
592 /* That means that there is nothing to fixup. Just upload from This->resource.allocatedMemory
593 * directly into the vbo. Do not free the system memory copy because drawPrimitive may need it if
594 * the stride is 0, for instancing emulation, vertex blending emulation or shader emulation.
596 TRACE("No conversion needed\n");
598 if(!device->isInDraw) {
599 ActivateContext(device, device->lastActiveRenderTarget, CTXUSAGE_RESOURCELOAD);
601 ENTER_GL();
602 GL_EXTCALL(glBindBufferARB(GL_ARRAY_BUFFER_ARB, This->vbo));
603 checkGLcall("glBindBufferARB");
604 GL_EXTCALL(glBufferSubDataARB(GL_ARRAY_BUFFER_ARB, start, end-start, This->resource.allocatedMemory + start));
605 checkGLcall("glBufferSubDataARB");
606 LEAVE_GL();
607 return;
610 /* Now for each vertex in the buffer that needs conversion */
611 vertices = This->resource.size / This->stride;
613 if(This->conv_shift) {
614 TRACE("Shifted conversion\n");
615 data = HeapAlloc(GetProcessHeap(), 0, vertices * This->conv_stride);
617 for(i = start / This->stride; i < min((end / This->stride) + 1, vertices); i++) {
618 for(j = 0; j < This->stride; j++) {
619 switch(This->conv_map[j]) {
620 case CONV_NONE:
621 data[This->conv_stride * i + j + This->conv_shift[j]] = This->resource.allocatedMemory[This->stride * i + j];
622 break;
624 case CONV_FLOAT16_2:
626 float *out = (float *) (&data[This->conv_stride * i + j + This->conv_shift[j]]);
627 const WORD *in = (WORD *) (&This->resource.allocatedMemory[i * This->stride + j]);
629 out[1] = float_16_to_32(in + 1);
630 out[0] = float_16_to_32(in + 0);
631 j += 3; /* Skip 3 additional bytes,as a FLOAT16_2 has 4 bytes */
632 break;
635 default:
636 FIXME("Unimplemented conversion %d in shifted conversion\n", This->conv_map[j]);
641 ENTER_GL();
642 GL_EXTCALL(glBindBufferARB(GL_ARRAY_BUFFER_ARB, This->vbo));
643 checkGLcall("glBindBufferARB");
644 GL_EXTCALL(glBufferSubDataARB(GL_ARRAY_BUFFER_ARB, 0, vertices * This->conv_stride, data));
645 checkGLcall("glBufferSubDataARB");
646 LEAVE_GL();
648 } else {
649 data = HeapAlloc(GetProcessHeap(), 0, This->resource.size);
650 memcpy(data + start, This->resource.allocatedMemory + start, end - start);
651 for(i = start / This->stride; i < min((end / This->stride) + 1, vertices); i++) {
652 for(j = 0; j < This->stride; j++) {
653 switch(This->conv_map[j]) {
654 case CONV_NONE:
655 /* Done already */
656 j += 3;
657 break;
658 case CONV_D3DCOLOR:
659 fixup_d3dcolor((DWORD *) (data + i * This->stride + j));
660 j += 3;
661 break;
663 case CONV_POSITIONT:
664 fixup_transformed_pos((float *) (data + i * This->stride + j));
665 j += 15;
666 break;
668 case CONV_FLOAT16_2:
669 ERR("Did not expect FLOAT16 conversion in unshifted conversion\n");
670 default:
671 FIXME("Unimplemented conversion %d in shifted conversion\n", This->conv_map[j]);
676 ENTER_GL();
677 GL_EXTCALL(glBindBufferARB(GL_ARRAY_BUFFER_ARB, This->vbo));
678 checkGLcall("glBindBufferARB");
679 GL_EXTCALL(glBufferSubDataARB(GL_ARRAY_BUFFER_ARB, start, end - start, data + start));
680 checkGLcall("glBufferSubDataARB");
681 LEAVE_GL();
684 HeapFree(GetProcessHeap(), 0, data);
687 static void WINAPI IWineD3DVertexBufferImpl_UnLoad(IWineD3DVertexBuffer *iface) {
688 IWineD3DVertexBufferImpl *This = (IWineD3DVertexBufferImpl *) iface;
689 IWineD3DDeviceImpl *device = This->resource.wineD3DDevice;
690 TRACE("(%p)\n", This);
692 /* This is easy: The whole content is shadowed in This->resource.allocatedMemory,
693 * so we only have to destroy the vbo. Only do it if we have a vbo, which implies
694 * that vbos are supported
696 if(This->vbo) {
697 ActivateContext(device, device->lastActiveRenderTarget, CTXUSAGE_RESOURCELOAD);
698 ENTER_GL();
699 GL_EXTCALL(glDeleteBuffersARB(1, &This->vbo));
700 checkGLcall("glDeleteBuffersARB");
701 LEAVE_GL();
702 This->vbo = 0;
703 This->Flags |= VBFLAG_CREATEVBO; /* Recreate the VBO next load */
707 static WINED3DRESOURCETYPE WINAPI IWineD3DVertexBufferImpl_GetType(IWineD3DVertexBuffer *iface) {
708 return resource_get_type((IWineD3DResource *)iface);
711 static HRESULT WINAPI IWineD3DVertexBufferImpl_GetParent(IWineD3DVertexBuffer *iface, IUnknown **pParent) {
712 return resource_get_parent((IWineD3DResource *)iface, pParent);
715 /* ******************************************************
716 IWineD3DVertexBuffer IWineD3DVertexBuffer parts follow
717 ****************************************************** */
718 static HRESULT WINAPI IWineD3DVertexBufferImpl_Lock(IWineD3DVertexBuffer *iface, UINT OffsetToLock, UINT SizeToLock, BYTE** ppbData, DWORD Flags) {
719 IWineD3DVertexBufferImpl *This = (IWineD3DVertexBufferImpl *)iface;
720 BYTE *data;
721 TRACE("(%p)->%d, %d, %p, %08x\n", This, OffsetToLock, SizeToLock, ppbData, Flags);
723 InterlockedIncrement(&This->lockcount);
725 if(This->Flags & VBFLAG_DIRTY) {
726 if(This->dirtystart > OffsetToLock) This->dirtystart = OffsetToLock;
727 if(SizeToLock) {
728 if(This->dirtyend < OffsetToLock + SizeToLock) This->dirtyend = OffsetToLock + SizeToLock;
729 } else {
730 This->dirtyend = This->resource.size;
732 } else {
733 This->dirtystart = OffsetToLock;
734 if(SizeToLock)
735 This->dirtyend = OffsetToLock + SizeToLock;
736 else
737 This->dirtyend = This->resource.size;
740 data = This->resource.allocatedMemory;
741 This->Flags |= VBFLAG_DIRTY;
742 *ppbData = data + OffsetToLock;
744 TRACE("(%p) : returning memory of %p (base:%p,offset:%u)\n", This, data + OffsetToLock, data, OffsetToLock);
745 /* TODO: check Flags compatibility with This->currentDesc.Usage (see MSDN) */
746 return WINED3D_OK;
748 static HRESULT WINAPI IWineD3DVertexBufferImpl_Unlock(IWineD3DVertexBuffer *iface) {
749 IWineD3DVertexBufferImpl *This = (IWineD3DVertexBufferImpl *) iface;
750 LONG lockcount;
751 TRACE("(%p)\n", This);
753 lockcount = InterlockedDecrement(&This->lockcount);
754 if(lockcount > 0) {
755 /* Delay loading the buffer until everything is unlocked */
756 TRACE("Ignoring the unlock\n");
757 return WINED3D_OK;
760 if(This->Flags & VBFLAG_HASDESC) {
761 IWineD3DVertexBufferImpl_PreLoad(iface);
763 return WINED3D_OK;
765 static HRESULT WINAPI IWineD3DVertexBufferImpl_GetDesc(IWineD3DVertexBuffer *iface, WINED3DVERTEXBUFFER_DESC *pDesc) {
766 IWineD3DVertexBufferImpl *This = (IWineD3DVertexBufferImpl *)iface;
768 TRACE("(%p)\n", This);
769 pDesc->Format = This->resource.format;
770 pDesc->Type = This->resource.resourceType;
771 pDesc->Usage = This->resource.usage;
772 pDesc->Pool = This->resource.pool;
773 pDesc->Size = This->resource.size;
774 pDesc->FVF = This->fvf;
775 return WINED3D_OK;
778 const IWineD3DVertexBufferVtbl IWineD3DVertexBuffer_Vtbl =
780 /* IUnknown */
781 IWineD3DVertexBufferImpl_QueryInterface,
782 IWineD3DVertexBufferImpl_AddRef,
783 IWineD3DVertexBufferImpl_Release,
784 /* IWineD3DResource */
785 IWineD3DVertexBufferImpl_GetParent,
786 IWineD3DVertexBufferImpl_GetDevice,
787 IWineD3DVertexBufferImpl_SetPrivateData,
788 IWineD3DVertexBufferImpl_GetPrivateData,
789 IWineD3DVertexBufferImpl_FreePrivateData,
790 IWineD3DVertexBufferImpl_SetPriority,
791 IWineD3DVertexBufferImpl_GetPriority,
792 IWineD3DVertexBufferImpl_PreLoad,
793 IWineD3DVertexBufferImpl_UnLoad,
794 IWineD3DVertexBufferImpl_GetType,
795 /* IWineD3DVertexBuffer */
796 IWineD3DVertexBufferImpl_Lock,
797 IWineD3DVertexBufferImpl_Unlock,
798 IWineD3DVertexBufferImpl_GetDesc
801 const BYTE* IWineD3DVertexBufferImpl_GetMemory(IWineD3DVertexBuffer* iface, DWORD iOffset, GLint *vbo)
803 IWineD3DVertexBufferImpl *This = (IWineD3DVertexBufferImpl *)iface;
805 *vbo = This->vbo;
806 if(This->vbo == 0) {
807 if(This->Flags & VBFLAG_CREATEVBO) {
808 CreateVBO(This);
809 This->Flags &= ~VBFLAG_CREATEVBO;
810 if(This->vbo) {
811 *vbo = This->vbo;
812 return (const BYTE *)iOffset;
815 return This->resource.allocatedMemory + iOffset;
816 } else {
817 return (const BYTE *)iOffset;