buffer.c@ 33252

Last change on this file since 33252 was 32622, checked in by vboxsync, 14 years ago
wddm: VBOXWDDM->VBOX_WITH_WDDM
Property svn:eol-style set to `native`
File size: 55.1 KB

Line
1	/*
2	* Copyright 2002-2005 Jason Edmeades
3	* Copyright 2002-2005 Raphael Junqueira
4	* Copyright 2004 Christian Costa
5	* Copyright 2005 Oliver Stieber
6	* Copyright 2007-2010 Stefan Dösinger for CodeWeavers
7	* Copyright 2009 Henri Verbeet for CodeWeavers
8	*
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.
13	*
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.
18	*
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
22	*
23	*/
24
25	/*
26	* Sun LGPL Disclaimer: For the avoidance of doubt, except that if any license choice
27	* other than GPL or LGPL is available it will apply instead, Sun elects to use only
28	* the Lesser General Public License version 2.1 (LGPLv2) at this time for any software where
29	* a choice of LGPL license versions is made available with the language indicating
30	* that LGPLv2 or any later version may be used, or where a choice of which version
31	* of the LGPL is applied is otherwise unspecified.
32	*/
33
34	#include "config.h"
35	#include "wine/port.h"
36
37	#include "wined3d_private.h"
38
39	WINE_DEFAULT_DEBUG_CHANNEL(d3d);
40
41	#define GLINFO_LOCATION This->resource.device->adapter->gl_info
42
43	#define VB_MAXDECLCHANGES 100 /* After that number of decl changes we stop converting */
44	#define VB_RESETDECLCHANGE 1000 /* Reset the decl changecount after that number of draws */
45	#define VB_MAXFULLCONVERSIONS 5 /* Number of full conversions before we stop converting */
46	#define VB_RESETFULLCONVS 20 /* Reset full conversion counts after that number of draws */
47
48	static inline BOOL buffer_add_dirty_area(struct wined3d_buffer *This, UINT offset, UINT size)
49	{
50	if (!This->buffer_object) return TRUE;
51
52	if (This->maps_size <= This->modified_areas)
53	{
54	void *new = HeapReAlloc(GetProcessHeap(), 0, This->maps,
55	This->maps_size * 2 * sizeof(*This->maps));
56	if (!new)
57	{
58	ERR("Out of memory\n");
59	return FALSE;
60	}
61	else
62	{
63	This->maps = new;
64	This->maps_size *= 2;
65	}
66	}
67
68	if(offset > This->resource.size \|\| offset + size > This->resource.size)
69	{
70	WARN("Invalid range dirtified, marking entire buffer dirty\n");
71	offset = 0;
72	size = This->resource.size;
73	}
74	else if(!offset && !size)
75	{
76	size = This->resource.size;
77	}
78
79	This->maps[This->modified_areas].offset = offset;
80	This->maps[This->modified_areas].size = size;
81	This->modified_areas++;
82	return TRUE;
83	}
84
85	static inline void buffer_clear_dirty_areas(struct wined3d_buffer *This)
86	{
87	This->modified_areas = 0;
88	}
89
90	static inline BOOL buffer_is_dirty(struct wined3d_buffer *This)
91	{
92	return This->modified_areas != 0;
93	}
94
95	static inline BOOL buffer_is_fully_dirty(struct wined3d_buffer *This)
96	{
97	unsigned int i;
98
99	for(i = 0; i < This->modified_areas; i++)
100	{
101	if(This->maps[i].offset == 0 && This->maps[i].size == This->resource.size)
102	{
103	return TRUE;
104	}
105	}
106	return FALSE;
107	}
108
109	/* Context activation is done by the caller */
110	static void delete_gl_buffer(struct wined3d_buffer *This)
111	{
112	if(!This->buffer_object) return;
113
114	ENTER_GL();
115	GL_EXTCALL(glDeleteBuffersARB(1, &This->buffer_object));
116	checkGLcall("glDeleteBuffersARB");
117	LEAVE_GL();
118	This->buffer_object = 0;
119
120	if(This->query)
121	{
122	wined3d_event_query_destroy(This->query);
123	This->query = NULL;
124	}
125	This->flags &= ~WINED3D_BUFFER_APPLESYNC;
126	}
127
128	/* Context activation is done by the caller. */
129	static void buffer_create_buffer_object(struct wined3d_buffer *This)
130	{
131	GLenum error, gl_usage;
132	const struct wined3d_gl_info *gl_info = &This->resource.device->adapter->gl_info;
133
134	TRACE("Creating an OpenGL vertex buffer object for IWineD3DVertexBuffer %p Usage(%s)\n",
135	This, debug_d3dusage(This->resource.usage));
136
137	ENTER_GL();
138
139	/* Make sure that the gl error is cleared. Do not use checkGLcall
140	* here because checkGLcall just prints a fixme and continues. However,
141	* if an error during VBO creation occurs we can fall back to non-vbo operation
142	* with full functionality(but performance loss)
143	*/
144	while (glGetError() != GL_NO_ERROR);
145
146	/* Basically the FVF parameter passed to CreateVertexBuffer is no good
147	* It is the FVF set with IWineD3DDevice::SetFVF or the Vertex Declaration set with
148	* IWineD3DDevice::SetVertexDeclaration that decides how the vertices in the buffer
149	* look like. This means that on each DrawPrimitive call the vertex buffer has to be verified
150	* to check if the rhw and color values are in the correct format.
151	*/
152
153	GL_EXTCALL(glGenBuffersARB(1, &This->buffer_object));
154	error = glGetError();
155	if (!This->buffer_object \|\| error != GL_NO_ERROR)
156	{
157	ERR("Failed to create a VBO with error %s (%#x)\n", debug_glerror(error), error);
158	LEAVE_GL();
159	goto fail;
160	}
161
162	if(This->buffer_type_hint == GL_ELEMENT_ARRAY_BUFFER_ARB)
163	{
164	IWineD3DDeviceImpl_MarkStateDirty(This->resource.device, STATE_INDEXBUFFER);
165	}
166	GL_EXTCALL(glBindBufferARB(This->buffer_type_hint, This->buffer_object));
167	error = glGetError();
168	if (error != GL_NO_ERROR)
169	{
170	ERR("Failed to bind the VBO with error %s (%#x)\n", debug_glerror(error), error);
171	LEAVE_GL();
172	goto fail;
173	}
174
175	/* Don't use static, because dx apps tend to update the buffer
176	* quite often even if they specify 0 usage.
177	*/
178	if(This->resource.usage & WINED3DUSAGE_DYNAMIC)
179	{
180	TRACE("Gl usage = GL_STREAM_DRAW_ARB\n");
181	gl_usage = GL_STREAM_DRAW_ARB;
182
183	if(gl_info->supported[APPLE_FLUSH_BUFFER_RANGE])
184	{
185	GL_EXTCALL(glBufferParameteriAPPLE(This->buffer_type_hint, GL_BUFFER_FLUSHING_UNMAP_APPLE, GL_FALSE));
186	checkGLcall("glBufferParameteriAPPLE(This->buffer_type_hint, GL_BUFFER_FLUSHING_UNMAP_APPLE, GL_FALSE)");
187	This->flags \|= WINED3D_BUFFER_FLUSH;
188
189	GL_EXTCALL(glBufferParameteriAPPLE(This->buffer_type_hint, GL_BUFFER_SERIALIZED_MODIFY_APPLE, GL_FALSE));
190	checkGLcall("glBufferParameteriAPPLE(This->buffer_type_hint, GL_BUFFER_SERIALIZED_MODIFY_APPLE, GL_FALSE)");
191	This->flags \|= WINED3D_BUFFER_APPLESYNC;
192	}
193	/* No setup is needed here for GL_ARB_map_buffer_range */
194	}
195	else
196	{
197	TRACE("Gl usage = GL_DYNAMIC_DRAW_ARB\n");
198	gl_usage = GL_DYNAMIC_DRAW_ARB;
199	}
200
201	/* Reserve memory for the buffer. The amount of data won't change
202	* so we are safe with calling glBufferData once and
203	* calling glBufferSubData on updates. Upload the actual data in case
204	* we're not double buffering, so we can release the heap mem afterwards
205	*/
206	GL_EXTCALL(glBufferDataARB(This->buffer_type_hint, This->resource.size, This->resource.allocatedMemory, gl_usage));
207	error = glGetError();
208	LEAVE_GL();
209	if (error != GL_NO_ERROR)
210	{
211	ERR("glBufferDataARB failed with error %s (%#x)\n", debug_glerror(error), error);
212	goto fail;
213	}
214
215	This->buffer_object_size = This->resource.size;
216	This->buffer_object_usage = gl_usage;
217
218	if(This->flags & WINED3D_BUFFER_DOUBLEBUFFER)
219	{
220	if(!buffer_add_dirty_area(This, 0, 0))
221	{
222	ERR("buffer_add_dirty_area failed, this is not expected\n");
223	goto fail;
224	}
225	}
226	else
227	{
228	HeapFree(GetProcessHeap(), 0, This->resource.heapMemory);
229	This->resource.allocatedMemory = NULL;
230	This->resource.heapMemory = NULL;
231	}
232
233	return;
234
235	fail:
236	/* Clean up all vbo init, but continue because we can work without a vbo :-) */
237	ERR("Failed to create a vertex buffer object. Continuing, but performance issues may occur\n");
238	delete_gl_buffer(This);
239	buffer_clear_dirty_areas(This);
240	}
241
242	static BOOL buffer_process_converted_attribute(struct wined3d_buffer *This,
243	const enum wined3d_buffer_conversion_type conversion_type,
244	const struct wined3d_stream_info_element attrib, DWORD stride_this_run)
245	{
246	DWORD attrib_size;
247	BOOL ret = FALSE;
248	unsigned int i;
249	DWORD offset = This->resource.device->stateBlock->streamOffset[attrib->stream_idx];
250	DWORD_PTR data;
251
252	/* Check for some valid situations which cause us pain. One is if the buffer is used for
253	* constant attributes(stride = 0), the other one is if the buffer is used on two streams
254	* with different strides. In the 2nd case we might have to drop conversion entirely,
255	* it is possible that the same bytes are once read as FLOAT2 and once as UBYTE4N.
256	*/
257	if (!attrib->stride)
258	{
259	FIXME("%s used with stride 0, let's hope we get the vertex stride from somewhere else\n",
260	debug_d3dformat(attrib->format_desc->format));
261	}
262	else if(attrib->stride != stride_this_run && stride_this_run)
263	{
264	FIXME("Got two concurrent strides, %d and %d\n", attrib->stride, *stride_this_run);
265	}
266	else
267	{
268	*stride_this_run = attrib->stride;
269	if (This->stride != *stride_this_run)
270	{
271	/* We rely that this happens only on the first converted attribute that is found,
272	* if at all. See above check
273	*/
274	TRACE("Reconverting because converted attributes occur, and the stride changed\n");
275	This->stride = *stride_this_run;
276	HeapFree(GetProcessHeap(), HEAP_ZERO_MEMORY, This->conversion_map);
277	This->conversion_map = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY,
278	sizeof(This->conversion_map) This->stride);
279	ret = TRUE;
280	}
281	}
282
283	data = (((DWORD_PTR)attrib->data) + offset) % This->stride;
284	attrib_size = attrib->format_desc->component_count * attrib->format_desc->component_size;
285	for (i = 0; i < attrib_size; ++i)
286	{
287	DWORD_PTR idx = (data + i) % This->stride;
288	if (This->conversion_map[idx] != conversion_type)
289	{
290	TRACE("Byte %ld in vertex changed\n", idx);
291	TRACE("It was type %d, is %d now\n", This->conversion_map[idx], conversion_type);
292	ret = TRUE;
293	This->conversion_map[idx] = conversion_type;
294	}
295	}
296
297	return ret;
298	}
299
300	static BOOL buffer_check_attribute(struct wined3d_buffer This, const struct wined3d_stream_info si,
301	UINT attrib_idx, const BOOL check_d3dcolor, const BOOL is_ffp_position, const BOOL is_ffp_color,
302	DWORD stride_this_run, BOOL float16_used)
303	{
304	const struct wined3d_stream_info_element *attrib = &si->elements[attrib_idx];
305	IWineD3DDeviceImpl *device = This->resource.device;
306	const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
307	BOOL ret = FALSE;
308	WINED3DFORMAT format;
309
310	/* Ignore attributes that do not have our vbo. After that check we can be sure that the attribute is
311	* there, on nonexistent attribs the vbo is 0.
312	*/
313	if (!(si->use_map & (1 << attrib_idx))
314	\|\| attrib->buffer_object != This->buffer_object)
315	return FALSE;
316
317	format = attrib->format_desc->format;
318	/* Look for newly appeared conversion */
319	if (!gl_info->supported[ARB_HALF_FLOAT_VERTEX]
320	&& (format == WINED3DFMT_R16G16_FLOAT \|\| format == WINED3DFMT_R16G16B16A16_FLOAT))
321	{
322	ret = buffer_process_converted_attribute(This, CONV_FLOAT16_2, attrib, stride_this_run);
323
324	if (is_ffp_position) FIXME("Test FLOAT16 fixed function processing positions\n");
325	else if (is_ffp_color) FIXME("test FLOAT16 fixed function processing colors\n");
326	*float16_used = TRUE;
327	}
328	else if (check_d3dcolor && format == WINED3DFMT_B8G8R8A8_UNORM)
329	{
330	ret = buffer_process_converted_attribute(This, CONV_D3DCOLOR, attrib, stride_this_run);
331
332	if (!is_ffp_color) FIXME("Test for non-color fixed function WINED3DFMT_B8G8R8A8_UNORM format\n");
333	}
334	else if (is_ffp_position && format == WINED3DFMT_R32G32B32A32_FLOAT)
335	{
336	ret = buffer_process_converted_attribute(This, CONV_POSITIONT, attrib, stride_this_run);
337	}
338	else if (This->conversion_map)
339	{
340	ret = buffer_process_converted_attribute(This, CONV_NONE, attrib, stride_this_run);
341	}
342
343	return ret;
344	}
345
346	static UINT find_conversion_shift(struct wined3d_buffer This,
347	const struct wined3d_stream_info *strided, UINT stride)
348	{
349	UINT *ret, i, j, shift, orig_type_size;
350
351	if (!stride)
352	{
353	TRACE("No shift\n");
354	return NULL;
355	}
356
357	This->conversion_stride = stride;
358	ret = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(DWORD) * stride);
359	for (i = 0; i < MAX_ATTRIBS; ++i)
360	{
361	WINED3DFORMAT format;
362
363	if (!(strided->use_map & (1 << i)) \|\| strided->elements[i].buffer_object != This->buffer_object) continue;
364
365	format = strided->elements[i].format_desc->format;
366	if (format == WINED3DFMT_R16G16_FLOAT)
367	{
368	shift = 4;
369	}
370	else if (format == WINED3DFMT_R16G16B16A16_FLOAT)
371	{
372	shift = 8;
373	/* Pre-shift the last 4 bytes in the FLOAT16_4 by 4 bytes - this makes FLOAT16_2 and FLOAT16_4 conversions
374	* compatible
375	*/
376	for (j = 4; j < 8; ++j)
377	{
378	ret[(DWORD_PTR)strided->elements[i].data + j] += 4;
379	}
380	}
381	else
382	{
383	shift = 0;
384	}
385	This->conversion_stride += shift;
386
387	if (shift)
388	{
389	orig_type_size = strided->elements[i].format_desc->component_count
390	* strided->elements[i].format_desc->component_size;
391	for (j = (DWORD_PTR)strided->elements[i].data + orig_type_size; j < stride; ++j)
392	{
393	ret[j] += shift;
394	}
395	}
396	}
397
398	if (TRACE_ON(d3d))
399	{
400	TRACE("Dumping conversion shift:\n");
401	for (i = 0; i < stride; ++i)
402	{
403	TRACE("[%d]", ret[i]);
404	}
405	TRACE("\n");
406	}
407
408	return ret;
409	}
410
411	static BOOL buffer_find_decl(struct wined3d_buffer *This)
412	{
413	IWineD3DDeviceImpl *device = This->resource.device;
414	const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
415	const struct wined3d_stream_info *si = &device->strided_streams;
416	UINT stride_this_run = 0;
417	BOOL float16_used = FALSE;
418	BOOL ret = FALSE;
419	unsigned int i;
420
421	/* In d3d7 the vertex buffer declaration NEVER changes because it is stored in the d3d7 vertex buffer.
422	* Once we have our declaration there is no need to look it up again. Index buffers also never need
423	* conversion, so once the (empty) conversion structure is created don't bother checking again
424	*/
425	if (This->flags & WINED3D_BUFFER_HASDESC)
426	{
427	if(This->resource.usage & WINED3DUSAGE_STATICDECL) return FALSE;
428	}
429
430	TRACE("Finding vertex buffer conversion information\n");
431	/* Certain declaration types need some fixups before we can pass them to
432	* opengl. This means D3DCOLOR attributes with fixed function vertex
433	* processing, FLOAT4 POSITIONT with fixed function, and FLOAT16 if
434	* GL_ARB_half_float_vertex is not supported.
435	*
436	* Note for d3d8 and d3d9:
437	* The vertex buffer FVF doesn't help with finding them, we have to use
438	* the decoded vertex declaration and pick the things that concern the
439	* current buffer. A problem with this is that this can change between
440	* draws, so we have to validate the information and reprocess the buffer
441	* if it changes, and avoid false positives for performance reasons.
442	* WineD3D doesn't even know the vertex buffer any more, it is managed
443	* by the client libraries and passed to SetStreamSource and ProcessVertices
444	* as needed.
445	*
446	* We have to distinguish between vertex shaders and fixed function to
447	* pick the way we access the strided vertex information.
448	*
449	* This code sets up a per-byte array with the size of the detected
450	* stride of the arrays in the buffer. For each byte we have a field
451	* that marks the conversion needed on this byte. For example, the
452	* following declaration with fixed function vertex processing:
453	*
454	* POSITIONT, FLOAT4
455	* NORMAL, FLOAT3
456	* DIFFUSE, FLOAT16_4
457	* SPECULAR, D3DCOLOR
458	*
459	* Will result in
460	* { POSITIONT }{ NORMAL }{ DIFFUSE }{SPECULAR }
461	* [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]
462	*
463	* Where in this example map P means 4 component position conversion, 0
464	* means no conversion, F means FLOAT16_2 conversion and C means D3DCOLOR
465	* conversion (red / blue swizzle).
466	*
467	* If we're doing conversion and the stride changes we have to reconvert
468	* the whole buffer. Note that we do not mind if the semantic changes,
469	* we only care for the conversion type. So if the NORMAL is replaced
470	* with a TEXCOORD, nothing has to be done, or if the DIFFUSE is replaced
471	* with a D3DCOLOR BLENDWEIGHT we can happily dismiss the change. Some
472	* conversion types depend on the semantic as well, for example a FLOAT4
473	* texcoord needs no conversion while a FLOAT4 positiont needs one
474	*/
475	if (use_vs(device->stateBlock))
476	{
477	TRACE("vshader\n");
478	/* If the current vertex declaration is marked for no half float conversion don't bother to
479	* analyse the strided streams in depth, just set them up for no conversion. Return decl changed
480	* if we used conversion before
481	*/
482	if (!((IWineD3DVertexDeclarationImpl *) device->stateBlock->vertexDecl)->half_float_conv_needed)
483	{
484	if (This->conversion_map)
485	{
486	TRACE("Now using shaders without conversion, but conversion used before\n");
487	HeapFree(GetProcessHeap(), 0, This->conversion_map);
488	HeapFree(GetProcessHeap(), 0, This->conversion_shift);
489	This->conversion_map = NULL;
490	This->stride = 0;
491	This->conversion_shift = NULL;
492	This->conversion_stride = 0;
493	return TRUE;
494	}
495	else
496	{
497	return FALSE;
498	}
499	}
500	for (i = 0; i < MAX_ATTRIBS; ++i)
501	{
502	ret = buffer_check_attribute(This, si, i, FALSE, FALSE, FALSE, &stride_this_run, &float16_used) \|\| ret;
503	}
504
505	/* Recalculate the conversion shift map if the declaration has changed,
506	* and we're using float16 conversion or used it on the last run
507	*/
508	if (ret && (float16_used \|\| This->conversion_map))
509	{
510	HeapFree(GetProcessHeap(), 0, This->conversion_shift);
511	This->conversion_shift = find_conversion_shift(This, si, This->stride);
512	}
513	}
514	else
515	{
516	/* Fixed function is a bit trickier. We have to take care for D3DCOLOR types, FLOAT4 positions and of course
517	* FLOAT16s if not supported. Also, we can't iterate over the array, so use macros to generate code for all
518	* the attributes that our current fixed function pipeline implementation cares for.
519	*/
520	BOOL support_d3dcolor = gl_info->supported[ARB_VERTEX_ARRAY_BGRA];
521	ret = buffer_check_attribute(This, si, WINED3D_FFP_POSITION,
522	TRUE, TRUE, FALSE, &stride_this_run, &float16_used) \|\| ret;
523	ret = buffer_check_attribute(This, si, WINED3D_FFP_NORMAL,
524	TRUE, FALSE, FALSE, &stride_this_run, &float16_used) \|\| ret;
525	ret = buffer_check_attribute(This, si, WINED3D_FFP_DIFFUSE,
526	!support_d3dcolor, FALSE, TRUE, &stride_this_run, &float16_used) \|\| ret;
527	ret = buffer_check_attribute(This, si, WINED3D_FFP_SPECULAR,
528	!support_d3dcolor, FALSE, TRUE, &stride_this_run, &float16_used) \|\| ret;
529	ret = buffer_check_attribute(This, si, WINED3D_FFP_TEXCOORD0,
530	TRUE, FALSE, FALSE, &stride_this_run, &float16_used) \|\| ret;
531	ret = buffer_check_attribute(This, si, WINED3D_FFP_TEXCOORD1,
532	TRUE, FALSE, FALSE, &stride_this_run, &float16_used) \|\| ret;
533	ret = buffer_check_attribute(This, si, WINED3D_FFP_TEXCOORD2,
534	TRUE, FALSE, FALSE, &stride_this_run, &float16_used) \|\| ret;
535	ret = buffer_check_attribute(This, si, WINED3D_FFP_TEXCOORD3,
536	TRUE, FALSE, FALSE, &stride_this_run, &float16_used) \|\| ret;
537	ret = buffer_check_attribute(This, si, WINED3D_FFP_TEXCOORD4,
538	TRUE, FALSE, FALSE, &stride_this_run, &float16_used) \|\| ret;
539	ret = buffer_check_attribute(This, si, WINED3D_FFP_TEXCOORD5,
540	TRUE, FALSE, FALSE, &stride_this_run, &float16_used) \|\| ret;
541	ret = buffer_check_attribute(This, si, WINED3D_FFP_TEXCOORD6,
542	TRUE, FALSE, FALSE, &stride_this_run, &float16_used) \|\| ret;
543	ret = buffer_check_attribute(This, si, WINED3D_FFP_TEXCOORD7,
544	TRUE, FALSE, FALSE, &stride_this_run, &float16_used) \|\| ret;
545
546	if (float16_used) FIXME("Float16 conversion used with fixed function vertex processing\n");
547	}
548
549	if (stride_this_run == 0 && This->conversion_map)
550	{
551	/* Sanity test */
552	if (!ret) ERR("no converted attributes found, old conversion map exists, and no declaration change?\n");
553	HeapFree(GetProcessHeap(), 0, This->conversion_map);
554	This->conversion_map = NULL;
555	This->stride = 0;
556	}
557
558	if (ret) TRACE("Conversion information changed\n");
559
560	return ret;
561	}
562
563	/* Context activation is done by the caller. */
564	static void buffer_check_buffer_object_size(struct wined3d_buffer *This)
565	{
566	UINT size = This->conversion_stride ?
567	This->conversion_stride * (This->resource.size / This->stride) : This->resource.size;
568	if (This->buffer_object_size != size)
569	{
570	TRACE("Old size %u, creating new size %u\n", This->buffer_object_size, size);
571
572	if(This->buffer_type_hint == GL_ELEMENT_ARRAY_BUFFER_ARB)
573	{
574	IWineD3DDeviceImpl_MarkStateDirty(This->resource.device, STATE_INDEXBUFFER);
575	}
576
577	/* Rescue the data before resizing the buffer object if we do not have our backup copy */
578	if(!(This->flags & WINED3D_BUFFER_DOUBLEBUFFER))
579	{
580	buffer_get_sysmem(This);
581	}
582
583	ENTER_GL();
584	GL_EXTCALL(glBindBufferARB(This->buffer_type_hint, This->buffer_object));
585	checkGLcall("glBindBufferARB");
586	GL_EXTCALL(glBufferDataARB(This->buffer_type_hint, size, NULL, This->buffer_object_usage));
587	This->buffer_object_size = size;
588	checkGLcall("glBufferDataARB");
589	LEAVE_GL();
590	}
591	}
592
593	static inline void fixup_d3dcolor(DWORD *dst_color)
594	{
595	DWORD src_color = *dst_color;
596
597	/* Color conversion like in drawStridedSlow. watch out for little endianity
598	* If we want that stuff to work on big endian machines too we have to consider more things
599	*
600	* 0xff000000: Alpha mask
601	* 0x00ff0000: Blue mask
602	* 0x0000ff00: Green mask
603	* 0x000000ff: Red mask
604	*/
605	*dst_color = 0;
606	dst_color \|= (src_color & 0xff00ff00); / Alpha Green */
607	dst_color \|= (src_color & 0x00ff0000) >> 16; / Red */
608	dst_color \|= (src_color & 0x000000ff) << 16; / Blue */
609	}
610
611	static inline void fixup_transformed_pos(float *p)
612	{
613	/* rhw conversion like in position_float4(). */
614	if (p[3] != 1.0f && p[3] != 0.0f)
615	{
616	float w = 1.0f / p[3];
617	p[0] *= w;
618	p[1] *= w;
619	p[2] *= w;
620	p[3] = w;
621	}
622	}
623
624	/* Context activation is done by the caller. */
625	const BYTE buffer_get_memory(IWineD3DBuffer iface, GLuint *buffer_object)
626	{
627	struct wined3d_buffer This = (struct wined3d_buffer )iface;
628
629	*buffer_object = This->buffer_object;
630	if (!This->buffer_object)
631	{
632	if (This->flags & WINED3D_BUFFER_CREATEBO)
633	{
634	buffer_create_buffer_object(This);
635	This->flags &= ~WINED3D_BUFFER_CREATEBO;
636	if (This->buffer_object)
637	{
638	*buffer_object = This->buffer_object;
639	return NULL;
640	}
641	}
642	return This->resource.allocatedMemory;
643	}
644	else
645	{
646	return NULL;
647	}
648	}
649
650	/* IUnknown methods */
651
652	static HRESULT STDMETHODCALLTYPE buffer_QueryInterface(IWineD3DBuffer *iface,
653	REFIID riid, void **object)
654	{
655	TRACE("iface %p, riid %s, object %p\n", iface, debugstr_guid(riid), object);
656
657	if (IsEqualGUID(riid, &IID_IWineD3DBuffer)
658	\|\| IsEqualGUID(riid, &IID_IWineD3DResource)
659	\|\| IsEqualGUID(riid, &IID_IWineD3DBase)
660	\|\| IsEqualGUID(riid, &IID_IUnknown))
661	{
662	IUnknown_AddRef(iface);
663	*object = iface;
664	return S_OK;
665	}
666
667	WARN("%s not implemented, returning E_NOINTERFACE\n", debugstr_guid(riid));
668
669	*object = NULL;
670
671	return E_NOINTERFACE;
672	}
673
674	static ULONG STDMETHODCALLTYPE buffer_AddRef(IWineD3DBuffer *iface)
675	{
676	struct wined3d_buffer This = (struct wined3d_buffer )iface;
677	ULONG refcount = InterlockedIncrement(&This->resource.ref);
678
679	TRACE("%p increasing refcount to %u\n", This, refcount);
680
681	return refcount;
682	}
683
684	/* Context activation is done by the caller. */
685	BYTE buffer_get_sysmem(struct wined3d_buffer This)
686	{
687	/* AllocatedMemory exists if the buffer is double buffered or has no buffer object at all */
688	if(This->resource.allocatedMemory) return This->resource.allocatedMemory;
689
690	This->resource.heapMemory = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, This->resource.size + RESOURCE_ALIGNMENT);
691	This->resource.allocatedMemory = (BYTE *)(((ULONG_PTR)This->resource.heapMemory + (RESOURCE_ALIGNMENT - 1)) & ~(RESOURCE_ALIGNMENT - 1));
692	ENTER_GL();
693	GL_EXTCALL(glBindBufferARB(This->buffer_type_hint, This->buffer_object));
694	GL_EXTCALL(glGetBufferSubDataARB(This->buffer_type_hint, 0, This->resource.size, This->resource.allocatedMemory));
695	LEAVE_GL();
696	This->flags \|= WINED3D_BUFFER_DOUBLEBUFFER;
697
698	return This->resource.allocatedMemory;
699	}
700
701	static void STDMETHODCALLTYPE buffer_UnLoad(IWineD3DBuffer *iface)
702	{
703	struct wined3d_buffer This = (struct wined3d_buffer )iface;
704
705	TRACE("iface %p\n", iface);
706
707	if (This->buffer_object)
708	{
709	IWineD3DDeviceImpl *device = This->resource.device;
710	struct wined3d_context *context;
711
712	context = context_acquire(device, NULL, CTXUSAGE_RESOURCELOAD);
713
714	/* Download the buffer, but don't permanently enable double buffering */
715	if(!(This->flags & WINED3D_BUFFER_DOUBLEBUFFER))
716	{
717	buffer_get_sysmem(This);
718	This->flags &= ~WINED3D_BUFFER_DOUBLEBUFFER;
719	}
720
721	delete_gl_buffer(This);
722	This->flags \|= WINED3D_BUFFER_CREATEBO; /* Recreate the buffer object next load */
723	buffer_clear_dirty_areas(This);
724
725	context_release(context);
726
727	HeapFree(GetProcessHeap(), 0, This->conversion_shift);
728	This->conversion_shift = NULL;
729	HeapFree(GetProcessHeap(), 0, This->conversion_map);
730	This->conversion_map = NULL;
731	This->stride = 0;
732	This->conversion_stride = 0;
733	This->flags &= ~WINED3D_BUFFER_HASDESC;
734	}
735	}
736
737	static ULONG STDMETHODCALLTYPE buffer_Release(IWineD3DBuffer *iface)
738	{
739	struct wined3d_buffer This = (struct wined3d_buffer )iface;
740	ULONG refcount = InterlockedDecrement(&This->resource.ref);
741
742	TRACE("%p decreasing refcount to %u\n", This, refcount);
743
744	if (!refcount)
745	{
746	buffer_UnLoad(iface);
747	resource_cleanup((IWineD3DResource *)iface);
748	This->resource.parent_ops->wined3d_object_destroyed(This->resource.parent);
749	HeapFree(GetProcessHeap(), 0, This->maps);
750	HeapFree(GetProcessHeap(), 0, This);
751	}
752
753	return refcount;
754	}
755
756	/* IWineD3DBase methods */
757
758	static HRESULT STDMETHODCALLTYPE buffer_GetParent(IWineD3DBuffer iface, IUnknown *parent)
759	{
760	return resource_get_parent((IWineD3DResource *)iface, parent);
761	}
762
763	/* IWineD3DResource methods */
764
765	static HRESULT STDMETHODCALLTYPE buffer_SetPrivateData(IWineD3DBuffer *iface,
766	REFGUID guid, const void *data, DWORD data_size, DWORD flags)
767	{
768	return resource_set_private_data((IWineD3DResource *)iface, guid, data, data_size, flags);
769	}
770
771	static HRESULT STDMETHODCALLTYPE buffer_GetPrivateData(IWineD3DBuffer *iface,
772	REFGUID guid, void data, DWORD data_size)
773	{
774	return resource_get_private_data((IWineD3DResource *)iface, guid, data, data_size);
775	}
776
777	static HRESULT STDMETHODCALLTYPE buffer_FreePrivateData(IWineD3DBuffer *iface, REFGUID guid)
778	{
779	return resource_free_private_data((IWineD3DResource *)iface, guid);
780	}
781
782	static DWORD STDMETHODCALLTYPE buffer_SetPriority(IWineD3DBuffer *iface, DWORD priority)
783	{
784	return resource_set_priority((IWineD3DResource *)iface, priority);
785	}
786
787	static DWORD STDMETHODCALLTYPE buffer_GetPriority(IWineD3DBuffer *iface)
788	{
789	return resource_get_priority((IWineD3DResource *)iface);
790	}
791
792	/* The caller provides a context and binds the buffer */
793	static void buffer_sync_apple(struct wined3d_buffer This, DWORD flags, const struct wined3d_gl_info gl_info)
794	{
795	enum wined3d_event_query_result ret;
796
797	/* No fencing needs to be done if the app promises not to overwrite
798	* existing data */
799	if(flags & WINED3DLOCK_NOOVERWRITE) return;
800	if(flags & WINED3DLOCK_DISCARD)
801	{
802	ENTER_GL();
803	GL_EXTCALL(glBufferDataARB(This->buffer_type_hint, This->resource.size, NULL, This->buffer_object_usage));
804	checkGLcall("glBufferDataARB\n");
805	LEAVE_GL();
806	return;
807	}
808
809	if(!This->query)
810	{
811	TRACE("Creating event query for buffer %p\n", This);
812
813	if (!wined3d_event_query_supported(gl_info))
814	{
815	FIXME("Event queries not supported, dropping async buffer locks.\n");
816	goto drop_query;
817	}
818
819	This->query = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*This->query));
820	if (!This->query)
821	{
822	ERR("Failed to allocate event query memory, dropping async buffer locks.\n");
823	goto drop_query;
824	}
825
826	/* Since we don't know about old draws a glFinish is needed once */
827	wglFinish();
828	return;
829	}
830	TRACE("Synchronizing buffer %p\n", This);
831	ret = wined3d_event_query_finish(This->query, This->resource.device);
832	switch(ret)
833	{
834	case WINED3D_EVENT_QUERY_NOT_STARTED:
835	case WINED3D_EVENT_QUERY_OK:
836	/* All done */
837	return;
838
839	case WINED3D_EVENT_QUERY_WRONG_THREAD:
840	WARN("Cannot synchronize buffer lock due to a thread conflict\n");
841	goto drop_query;
842
843	default:
844	ERR("wined3d_event_query_finish returned %u, dropping async buffer locks\n", ret);
845	goto drop_query;
846	}
847
848	drop_query:
849	if(This->query)
850	{
851	wined3d_event_query_destroy(This->query);
852	This->query = NULL;
853	}
854
855	wglFinish();
856	ENTER_GL();
857	GL_EXTCALL(glBufferParameteriAPPLE(This->buffer_type_hint, GL_BUFFER_SERIALIZED_MODIFY_APPLE, GL_TRUE));
858	checkGLcall("glBufferParameteriAPPLE(This->buffer_type_hint, GL_BUFFER_SERIALIZED_MODIFY_APPLE, GL_TRUE)");
859	LEAVE_GL();
860	This->flags &= ~WINED3D_BUFFER_APPLESYNC;
861	}
862
863	/* The caller provides a GL context */
864	static void buffer_direct_upload(struct wined3d_buffer This, const struct wined3d_gl_info gl_info, DWORD flags)
865	{
866	BYTE *map;
867	UINT start = 0, len = 0;
868
869	ENTER_GL();
870	GL_EXTCALL(glBindBufferARB(This->buffer_type_hint, This->buffer_object));
871	checkGLcall("glBindBufferARB");
872	if (gl_info->supported[ARB_MAP_BUFFER_RANGE])
873	{
874	GLbitfield mapflags;
875	mapflags = GL_MAP_WRITE_BIT \| GL_MAP_FLUSH_EXPLICIT_BIT;
876	if (flags & WINED3D_BUFFER_DISCARD)
877	{
878	mapflags \|= GL_MAP_UNSYNCHRONIZED_BIT \| GL_MAP_INVALIDATE_BUFFER_BIT;
879	}
880	else if (flags & WINED3D_BUFFER_NOSYNC)
881	{
882	mapflags \|= GL_MAP_UNSYNCHRONIZED_BIT;
883	}
884	map = GL_EXTCALL(glMapBufferRange(This->buffer_type_hint, 0,
885	This->resource.size, mapflags));
886	checkGLcall("glMapBufferRange");
887	}
888	else
889	{
890	if (This->flags & WINED3D_BUFFER_APPLESYNC)
891	{
892	DWORD syncflags = 0;
893	if (flags & WINED3D_BUFFER_DISCARD) syncflags \|= WINED3DLOCK_DISCARD;
894	if (flags & WINED3D_BUFFER_NOSYNC) syncflags \|= WINED3DLOCK_NOOVERWRITE;
895	LEAVE_GL();
896	buffer_sync_apple(This, syncflags, gl_info);
897	ENTER_GL();
898	}
899	map = GL_EXTCALL(glMapBufferARB(This->buffer_type_hint, GL_WRITE_ONLY_ARB));
900	checkGLcall("glMapBufferARB");
901	}
902	if (!map)
903	{
904	LEAVE_GL();
905	ERR("Failed to map opengl buffer\n");
906	return;
907	}
908
909	while(This->modified_areas)
910	{
911	This->modified_areas--;
912	start = This->maps[This->modified_areas].offset;
913	len = This->maps[This->modified_areas].size;
914
915	memcpy(map + start, This->resource.allocatedMemory + start, len);
916
917	if (gl_info->supported[ARB_MAP_BUFFER_RANGE])
918	{
919	GL_EXTCALL(glFlushMappedBufferRange(This->buffer_type_hint, start, len));
920	checkGLcall("glFlushMappedBufferRange");
921	}
922	else if (This->flags & WINED3D_BUFFER_FLUSH)
923	{
924	GL_EXTCALL(glFlushMappedBufferRangeAPPLE(This->buffer_type_hint, start, len));
925	checkGLcall("glFlushMappedBufferRangeAPPLE");
926	}
927	}
928	GL_EXTCALL(glUnmapBufferARB(This->buffer_type_hint));
929	checkGLcall("glUnmapBufferARB");
930	LEAVE_GL();
931	}
932
933	static void STDMETHODCALLTYPE buffer_PreLoad(IWineD3DBuffer *iface)
934	{
935	struct wined3d_buffer This = (struct wined3d_buffer )iface;
936	IWineD3DDeviceImpl *device = This->resource.device;
937	UINT start = 0, end = 0, len = 0, vertices;
938	struct wined3d_context *context;
939	BOOL decl_changed = FALSE;
940	unsigned int i, j;
941	BYTE *data;
942	DWORD flags = This->flags & (WINED3D_BUFFER_NOSYNC \| WINED3D_BUFFER_DISCARD);
943
944	TRACE("iface %p\n", iface);
945	This->flags &= ~(WINED3D_BUFFER_NOSYNC \| WINED3D_BUFFER_DISCARD);
946
947	context = context_acquire(device, NULL, CTXUSAGE_RESOURCELOAD);
948
949	if (!This->buffer_object)
950	{
951	/* TODO: Make converting independent from VBOs */
952	if (This->flags & WINED3D_BUFFER_CREATEBO)
953	{
954	buffer_create_buffer_object(This);
955	This->flags &= ~WINED3D_BUFFER_CREATEBO;
956	}
957	else
958	{
959	/* Not doing any conversion */
960	goto end;
961	}
962	}
963
964	/* Reading the declaration makes only sense if the stateblock is finalized and the buffer bound to a stream */
965	if (device->isInDraw && This->bind_count > 0)
966	{
967	decl_changed = buffer_find_decl(This);
968	This->flags \|= WINED3D_BUFFER_HASDESC;
969	}
970
971	if (!decl_changed && !(This->flags & WINED3D_BUFFER_HASDESC && buffer_is_dirty(This)))
972	{
973	context_release(context);
974	++This->draw_count;
975	if (This->draw_count > VB_RESETDECLCHANGE) This->decl_change_count = 0;
976	if (This->draw_count > VB_RESETFULLCONVS) This->full_conversion_count = 0;
977	return;
978	}
979
980	/* If applications change the declaration over and over, reconverting all the time is a huge
981	* performance hit. So count the declaration changes and release the VBO if there are too many
982	* of them (and thus stop converting)
983	*/
984	if (decl_changed)
985	{
986	++This->decl_change_count;
987	This->draw_count = 0;
988
989	if (This->decl_change_count > VB_MAXDECLCHANGES \|\|
990	(This->conversion_map && (This->resource.usage & WINED3DUSAGE_DYNAMIC)))
991	{
992	FIXME("Too many declaration changes or converting dynamic buffer, stopping converting\n");
993
994	IWineD3DBuffer_UnLoad(iface);
995	This->flags &= ~WINED3D_BUFFER_CREATEBO;
996
997	/* The stream source state handler might have read the memory of the vertex buffer already
998	* and got the memory in the vbo which is not valid any longer. Dirtify the stream source
999	* to force a reload. This happens only once per changed vertexbuffer and should occur rather
1000	* rarely
1001	*/
1002	IWineD3DDeviceImpl_MarkStateDirty(device, STATE_STREAMSRC);
1003	goto end;
1004	}
1005	buffer_check_buffer_object_size(This);
1006
1007	/* The declaration changed, reload the whole buffer */
1008	WARN("Reloading buffer because of decl change\n");
1009	buffer_clear_dirty_areas(This);
1010	if(!buffer_add_dirty_area(This, 0, 0))
1011	{
1012	ERR("buffer_add_dirty_area failed, this is not expected\n");
1013	return;
1014	}
1015	/* Avoid unfenced updates, we might overwrite more areas of the buffer than the application
1016	* cleared for unsynchronized updates
1017	*/
1018	flags = 0;
1019	}
1020	else
1021	{
1022	/* However, it is perfectly fine to change the declaration every now and then. We don't want a game that
1023	* changes it every minute drop the VBO after VB_MAX_DECL_CHANGES minutes. So count draws without
1024	* decl changes and reset the decl change count after a specific number of them
1025	*/
1026	if(buffer_is_fully_dirty(This))
1027	{
1028	++This->full_conversion_count;
1029	if(This->full_conversion_count > VB_MAXFULLCONVERSIONS)
1030	{
1031	FIXME("Too many full buffer conversions, stopping converting\n");
1032	IWineD3DBuffer_UnLoad(iface);
1033	This->flags &= ~WINED3D_BUFFER_CREATEBO;
1034	IWineD3DDeviceImpl_MarkStateDirty(device, STATE_STREAMSRC);
1035	goto end;
1036	}
1037	}
1038	else
1039	{
1040	++This->draw_count;
1041	if (This->draw_count > VB_RESETDECLCHANGE) This->decl_change_count = 0;
1042	if (This->draw_count > VB_RESETFULLCONVS) This->full_conversion_count = 0;
1043	}
1044	}
1045
1046	if(This->buffer_type_hint == GL_ELEMENT_ARRAY_BUFFER_ARB)
1047	{
1048	IWineD3DDeviceImpl_MarkStateDirty(This->resource.device, STATE_INDEXBUFFER);
1049	}
1050
1051	if (!This->conversion_map)
1052	{
1053	/* That means that there is nothing to fixup. Just upload from This->resource.allocatedMemory
1054	* directly into the vbo. Do not free the system memory copy because drawPrimitive may need it if
1055	* the stride is 0, for instancing emulation, vertex blending emulation or shader emulation.
1056	*/
1057	TRACE("No conversion needed\n");
1058
1059	/* Nothing to do because we locked directly into the vbo */
1060	if (!(This->flags & WINED3D_BUFFER_DOUBLEBUFFER))
1061	{
1062	context_release(context);
1063	return;
1064	}
1065
1066	buffer_direct_upload(This, context->gl_info, flags);
1067
1068	context_release(context);
1069	return;
1070	}
1071
1072	if(!(This->flags & WINED3D_BUFFER_DOUBLEBUFFER))
1073	{
1074	buffer_get_sysmem(This);
1075	}
1076
1077	/* Now for each vertex in the buffer that needs conversion */
1078	vertices = This->resource.size / This->stride;
1079
1080	if (This->conversion_shift)
1081	{
1082	TRACE("Shifted conversion\n");
1083	data = HeapAlloc(GetProcessHeap(), 0, vertices * This->conversion_stride);
1084
1085	start = 0;
1086	len = This->resource.size;
1087	end = start + len;
1088
1089	if (This->maps[0].offset \|\| This->maps[0].size != This->resource.size)
1090	{
1091	FIXME("Implement partial buffer load with shifted conversion\n");
1092	}
1093
1094	for (i = start / This->stride; i < min((end / This->stride) + 1, vertices); ++i)
1095	{
1096	for (j = 0; j < This->stride; ++j)
1097	{
1098	switch(This->conversion_map[j])
1099	{
1100	case CONV_NONE:
1101	data[This->conversion_stride * i + j + This->conversion_shift[j]]
1102	= This->resource.allocatedMemory[This->stride * i + j];
1103	break;
1104
1105	case CONV_FLOAT16_2:
1106	{
1107	float out = (float )(&data[This->conversion_stride * i + j + This->conversion_shift[j]]);
1108	const WORD in = (WORD )(&This->resource.allocatedMemory[i * This->stride + j]);
1109
1110	out[1] = float_16_to_32(in + 1);
1111	out[0] = float_16_to_32(in + 0);
1112	j += 3; /* Skip 3 additional bytes,as a FLOAT16_2 has 4 bytes */
1113	break;
1114	}
1115
1116	default:
1117	FIXME("Unimplemented conversion %d in shifted conversion\n", This->conversion_map[j]);
1118	break;
1119	}
1120	}
1121	}
1122
1123	ENTER_GL();
1124	GL_EXTCALL(glBindBufferARB(This->buffer_type_hint, This->buffer_object));
1125	checkGLcall("glBindBufferARB");
1126	GL_EXTCALL(glBufferSubDataARB(This->buffer_type_hint, 0, vertices * This->conversion_stride, data));
1127	checkGLcall("glBufferSubDataARB");
1128	LEAVE_GL();
1129	}
1130	else
1131	{
1132	data = HeapAlloc(GetProcessHeap(), 0, This->resource.size);
1133
1134	while(This->modified_areas)
1135	{
1136	This->modified_areas--;
1137	start = This->maps[This->modified_areas].offset;
1138	len = This->maps[This->modified_areas].size;
1139	end = start + len;
1140
1141	memcpy(data + start, This->resource.allocatedMemory + start, end - start);
1142	for (i = start / This->stride; i < min((end / This->stride) + 1, vertices); ++i)
1143	{
1144	for (j = 0; j < This->stride; ++j)
1145	{
1146	switch(This->conversion_map[j])
1147	{
1148	case CONV_NONE:
1149	/* Done already */
1150	j += 3;
1151	break;
1152	case CONV_D3DCOLOR:
1153	fixup_d3dcolor((DWORD ) (data + i This->stride + j));
1154	j += 3;
1155	break;
1156
1157	case CONV_POSITIONT:
1158	fixup_transformed_pos((float ) (data + i This->stride + j));
1159	j += 15;
1160	break;
1161
1162	case CONV_FLOAT16_2:
1163	ERR("Did not expect FLOAT16 conversion in unshifted conversion\n");
1164	default:
1165	FIXME("Unimplemented conversion %d in shifted conversion\n", This->conversion_map[j]);
1166	}
1167	}
1168	}
1169
1170	ENTER_GL();
1171	GL_EXTCALL(glBindBufferARB(This->buffer_type_hint, This->buffer_object));
1172	checkGLcall("glBindBufferARB");
1173	GL_EXTCALL(glBufferSubDataARB(This->buffer_type_hint, start, len, data + start));
1174	checkGLcall("glBufferSubDataARB");
1175	LEAVE_GL();
1176	}
1177	}
1178
1179	HeapFree(GetProcessHeap(), 0, data);
1180
1181	end:
1182	context_release(context);
1183	}
1184
1185	static WINED3DRESOURCETYPE STDMETHODCALLTYPE buffer_GetType(IWineD3DBuffer *iface)
1186	{
1187	return resource_get_type((IWineD3DResource *)iface);
1188	}
1189
1190	/* IWineD3DBuffer methods */
1191
1192	static DWORD buffer_sanitize_flags(struct wined3d_buffer *buffer, DWORD flags)
1193	{
1194	/* Not all flags make sense together, but Windows never returns an error. Catch the
1195	* cases that could cause issues */
1196	if(flags & WINED3DLOCK_READONLY)
1197	{
1198	if(flags & WINED3DLOCK_DISCARD)
1199	{
1200	WARN("WINED3DLOCK_READONLY combined with WINED3DLOCK_DISCARD, ignoring flags\n");
1201	return 0;
1202	}
1203	if(flags & WINED3DLOCK_NOOVERWRITE)
1204	{
1205	WARN("WINED3DLOCK_READONLY combined with WINED3DLOCK_NOOVERWRITE, ignoring flags\n");
1206	return 0;
1207	}
1208	}
1209	else if((flags & (WINED3DLOCK_DISCARD \| WINED3DLOCK_NOOVERWRITE)) == (WINED3DLOCK_DISCARD \| WINED3DLOCK_NOOVERWRITE))
1210	{
1211	WARN("WINED3DLOCK_DISCARD and WINED3DLOCK_NOOVERWRITE used together, ignoring\n");
1212	return 0;
1213	}
1214	else if (flags & (WINED3DLOCK_DISCARD \| WINED3DLOCK_NOOVERWRITE) && !(buffer->resource.usage & WINED3DUSAGE_DYNAMIC))
1215	{
1216	WARN("DISCARD or NOOVERWRITE lock on non-dynamic buffer, ignoring\n");
1217	return 0;
1218	}
1219
1220	return flags;
1221	}
1222
1223	static GLbitfield buffer_gl_map_flags(DWORD d3d_flags)
1224	{
1225	GLbitfield ret = 0;
1226
1227	if (!(d3d_flags & WINED3DLOCK_READONLY)) ret = GL_MAP_WRITE_BIT \| GL_MAP_FLUSH_EXPLICIT_BIT;
1228
1229	if (d3d_flags & (WINED3DLOCK_DISCARD \| WINED3DLOCK_NOOVERWRITE))
1230	{
1231	if(d3d_flags & WINED3DLOCK_DISCARD) ret \|= GL_MAP_INVALIDATE_BUFFER_BIT;
1232	ret \|= GL_MAP_UNSYNCHRONIZED_BIT;
1233	}
1234	else
1235	{
1236	ret \|= GL_MAP_READ_BIT;
1237	}
1238
1239	return ret;
1240	}
1241
1242	static HRESULT STDMETHODCALLTYPE buffer_Map(IWineD3DBuffer iface, UINT offset, UINT size, BYTE *data, DWORD flags)
1243	{
1244	struct wined3d_buffer This = (struct wined3d_buffer )iface;
1245	LONG count;
1246	BOOL dirty = buffer_is_dirty(This);
1247
1248	TRACE("iface %p, offset %u, size %u, data %p, flags %#x\n", iface, offset, size, data, flags);
1249
1250	flags = buffer_sanitize_flags(This, flags);
1251	if (!(flags & WINED3DLOCK_READONLY))
1252	{
1253	if (!buffer_add_dirty_area(This, offset, size)) return E_OUTOFMEMORY;
1254	}
1255
1256	count = InterlockedIncrement(&This->lock_count);
1257
1258	if (This->buffer_object)
1259	{
1260	if(!(This->flags & WINED3D_BUFFER_DOUBLEBUFFER))
1261	{
1262	if(count == 1)
1263	{
1264	IWineD3DDeviceImpl *device = This->resource.device;
1265	struct wined3d_context *context;
1266	const struct wined3d_gl_info *gl_info;
1267
1268	if(This->buffer_type_hint == GL_ELEMENT_ARRAY_BUFFER_ARB)
1269	{
1270	IWineD3DDeviceImpl_MarkStateDirty(This->resource.device, STATE_INDEXBUFFER);
1271	}
1272
1273	context = context_acquire(device, NULL, CTXUSAGE_RESOURCELOAD);
1274	gl_info = context->gl_info;
1275	ENTER_GL();
1276	GL_EXTCALL(glBindBufferARB(This->buffer_type_hint, This->buffer_object));
1277
1278	if (gl_info->supported[ARB_MAP_BUFFER_RANGE])
1279	{
1280	GLbitfield mapflags = buffer_gl_map_flags(flags);
1281	This->resource.allocatedMemory = GL_EXTCALL(glMapBufferRange(This->buffer_type_hint, 0,
1282	This->resource.size, mapflags));
1283	checkGLcall("glMapBufferRange");
1284	}
1285	else
1286	{
1287	if(This->flags & WINED3D_BUFFER_APPLESYNC)
1288	{
1289	LEAVE_GL();
1290	buffer_sync_apple(This, flags, gl_info);
1291	ENTER_GL();
1292	}
1293	This->resource.allocatedMemory = GL_EXTCALL(glMapBufferARB(This->buffer_type_hint, GL_READ_WRITE_ARB));
1294	checkGLcall("glMapBufferARB");
1295	}
1296	LEAVE_GL();
1297
1298	if (((DWORD_PTR) This->resource.allocatedMemory) & (RESOURCE_ALIGNMENT - 1))
1299	{
1300	WARN("Pointer %p is not %u byte aligned, falling back to double buffered operation\n",
1301	This->resource.allocatedMemory, RESOURCE_ALIGNMENT);
1302
1303	ENTER_GL();
1304	GL_EXTCALL(glUnmapBufferARB(This->buffer_type_hint));
1305	checkGLcall("glUnmapBufferARB");
1306	LEAVE_GL();
1307	This->resource.allocatedMemory = NULL;
1308
1309	buffer_get_sysmem(This);
1310	TRACE("New pointer is %p\n", This->resource.allocatedMemory);
1311	}
1312	context_release(context);
1313	}
1314	}
1315	else
1316	{
1317	if (dirty)
1318	{
1319	if (This->flags & WINED3D_BUFFER_NOSYNC && !(flags & WINED3DLOCK_NOOVERWRITE))
1320	{
1321	This->flags &= ~WINED3D_BUFFER_NOSYNC;
1322	}
1323	}
1324	else if(flags & WINED3DLOCK_NOOVERWRITE)
1325	{
1326	This->flags \|= WINED3D_BUFFER_NOSYNC;
1327	}
1328
1329	if (flags & WINED3DLOCK_DISCARD)
1330	{
1331	This->flags \|= WINED3D_BUFFER_DISCARD;
1332	}
1333	}
1334	}
1335
1336	*data = This->resource.allocatedMemory + offset;
1337
1338	TRACE("Returning memory at %p (base %p, offset %u)\n", *data, This->resource.allocatedMemory, offset);
1339	/* TODO: check Flags compatibility with This->currentDesc.Usage (see MSDN) */
1340
1341	return WINED3D_OK;
1342	}
1343
1344	static HRESULT STDMETHODCALLTYPE buffer_Unmap(IWineD3DBuffer *iface)
1345	{
1346	struct wined3d_buffer This = (struct wined3d_buffer )iface;
1347	ULONG i;
1348
1349	TRACE("(%p)\n", This);
1350
1351	/* In the case that the number of Unmap calls > the
1352	* number of Map calls, d3d returns always D3D_OK.
1353	* This is also needed to prevent Map from returning garbage on
1354	* the next call (this will happen if the lock_count is < 0). */
1355	if(This->lock_count == 0)
1356	{
1357	TRACE("Unmap called without a previous Map call!\n");
1358	return WINED3D_OK;
1359	}
1360
1361	if (InterlockedDecrement(&This->lock_count))
1362	{
1363	/* Delay loading the buffer until everything is unlocked */
1364	TRACE("Ignoring unlock\n");
1365	return WINED3D_OK;
1366	}
1367
1368	if(!(This->flags & WINED3D_BUFFER_DOUBLEBUFFER) && This->buffer_object)
1369	{
1370	IWineD3DDeviceImpl *device = This->resource.device;
1371	const struct wined3d_gl_info *gl_info;
1372	struct wined3d_context *context;
1373
1374	if(This->buffer_type_hint == GL_ELEMENT_ARRAY_BUFFER_ARB)
1375	{
1376	IWineD3DDeviceImpl_MarkStateDirty(This->resource.device, STATE_INDEXBUFFER);
1377	}
1378
1379	context = context_acquire(device, NULL, CTXUSAGE_RESOURCELOAD);
1380	gl_info = context->gl_info;
1381	ENTER_GL();
1382	GL_EXTCALL(glBindBufferARB(This->buffer_type_hint, This->buffer_object));
1383
1384	if (gl_info->supported[ARB_MAP_BUFFER_RANGE])
1385	{
1386	for(i = 0; i < This->modified_areas; i++)
1387	{
1388	GL_EXTCALL(glFlushMappedBufferRange(This->buffer_type_hint,
1389	This->maps[i].offset,
1390	This->maps[i].size));
1391	checkGLcall("glFlushMappedBufferRange");
1392	}
1393	}
1394	else if (This->flags & WINED3D_BUFFER_FLUSH)
1395	{
1396	for(i = 0; i < This->modified_areas; i++)
1397	{
1398	GL_EXTCALL(glFlushMappedBufferRangeAPPLE(This->buffer_type_hint,
1399	This->maps[i].offset,
1400	This->maps[i].size));
1401	checkGLcall("glFlushMappedBufferRangeAPPLE");
1402	}
1403	}
1404
1405	GL_EXTCALL(glUnmapBufferARB(This->buffer_type_hint));
1406	LEAVE_GL();
1407	context_release(context);
1408
1409	This->resource.allocatedMemory = NULL;
1410	buffer_clear_dirty_areas(This);
1411	}
1412	else if (This->flags & WINED3D_BUFFER_HASDESC)
1413	{
1414	buffer_PreLoad(iface);
1415	}
1416
1417	return WINED3D_OK;
1418	}
1419
1420	static HRESULT STDMETHODCALLTYPE buffer_GetDesc(IWineD3DBuffer iface, WINED3DBUFFER_DESC desc)
1421	{
1422	struct wined3d_buffer This = (struct wined3d_buffer )iface;
1423
1424	TRACE("(%p)\n", This);
1425
1426	desc->Type = This->resource.resourceType;
1427	desc->Usage = This->resource.usage;
1428	desc->Pool = This->resource.pool;
1429	desc->Size = This->resource.size;
1430
1431	return WINED3D_OK;
1432	}
1433
1434	static const struct IWineD3DBufferVtbl wined3d_buffer_vtbl =
1435	{
1436	/* IUnknown methods */
1437	buffer_QueryInterface,
1438	buffer_AddRef,
1439	buffer_Release,
1440	/* IWineD3DBase methods */
1441	buffer_GetParent,
1442	/* IWineD3DResource methods */
1443	buffer_SetPrivateData,
1444	buffer_GetPrivateData,
1445	buffer_FreePrivateData,
1446	buffer_SetPriority,
1447	buffer_GetPriority,
1448	buffer_PreLoad,
1449	buffer_UnLoad,
1450	buffer_GetType,
1451	/* IWineD3DBuffer methods */
1452	buffer_Map,
1453	buffer_Unmap,
1454	buffer_GetDesc,
1455	};
1456
1457	HRESULT buffer_init(struct wined3d_buffer buffer, IWineD3DDeviceImpl device,
1458	UINT size, DWORD usage, WINED3DFORMAT format, WINED3DPOOL pool, GLenum bind_hint,
1459	const char data, IUnknown parent, const struct wined3d_parent_ops *parent_ops)
1460	{
1461	const struct wined3d_format_desc *format_desc = getFormatDescEntry(format, &device->adapter->gl_info);
1462	HRESULT hr;
1463	const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
1464	BOOL dynamic_buffer_ok;
1465
1466	if (!size)
1467	{
1468	WARN("Size 0 requested, returning WINED3DERR_INVALIDCALL\n");
1469	return WINED3DERR_INVALIDCALL;
1470	}
1471
1472	buffer->vtbl = &wined3d_buffer_vtbl;
1473
1474	hr = resource_init((IWineD3DResource *)buffer, WINED3DRTYPE_BUFFER,
1475	device, size, usage, format_desc, pool, parent, parent_ops
1476	#ifdef VBOX_WITH_WDDM
1477	, NULL, NULL /* <- no need this info here so far */
1478	#endif
1479	);
1480	if (FAILED(hr))
1481	{
1482	WARN("Failed to initialize resource, hr %#x\n", hr);
1483	return hr;
1484	}
1485	buffer->buffer_type_hint = bind_hint;
1486
1487	TRACE("size %#x, usage %#x, format %s, memory @ %p, iface @ %p.\n", buffer->resource.size, buffer->resource.usage,
1488	debug_d3dformat(buffer->resource.format_desc->format), buffer->resource.allocatedMemory, buffer);
1489
1490	dynamic_buffer_ok = gl_info->supported[APPLE_FLUSH_BUFFER_RANGE] \|\| gl_info->supported[ARB_MAP_BUFFER_RANGE];
1491
1492	/* Observations show that drawStridedSlow is faster on dynamic VBs than converting +
1493	* drawStridedFast (half-life 2 and others).
1494	*
1495	* Basically converting the vertices in the buffer is quite expensive, and observations
1496	* show that drawStridedSlow is faster than converting + uploading + drawStridedFast.
1497	* Therefore do not create a VBO for WINED3DUSAGE_DYNAMIC buffers.
1498	*/
1499	if (!gl_info->supported[ARB_VERTEX_BUFFER_OBJECT])
1500	{
1501	TRACE("Not creating a vbo because GL_ARB_vertex_buffer is not supported\n");
1502	}
1503	else if(buffer->resource.pool == WINED3DPOOL_SYSTEMMEM)
1504	{
1505	TRACE("Not creating a vbo because the vertex buffer is in system memory\n");
1506	}
1507	else if(!dynamic_buffer_ok && (buffer->resource.usage & WINED3DUSAGE_DYNAMIC))
1508	{
1509	TRACE("Not creating a vbo because the buffer has dynamic usage and no GL support\n");
1510	}
1511	else
1512	{
1513	buffer->flags \|= WINED3D_BUFFER_CREATEBO;
1514	}
1515
1516	if (data)
1517	{
1518	BYTE *ptr;
1519
1520	hr = IWineD3DBuffer_Map((IWineD3DBuffer *)buffer, 0, size, &ptr, 0);
1521	if (FAILED(hr))
1522	{
1523	ERR("Failed to map buffer, hr %#x\n", hr);
1524	buffer_UnLoad((IWineD3DBuffer *)buffer);
1525	resource_cleanup((IWineD3DResource *)buffer);
1526	return hr;
1527	}
1528
1529	memcpy(ptr, data, size);
1530
1531	hr = IWineD3DBuffer_Unmap((IWineD3DBuffer *)buffer);
1532	if (FAILED(hr))
1533	{
1534	ERR("Failed to unmap buffer, hr %#x\n", hr);
1535	buffer_UnLoad((IWineD3DBuffer *)buffer);
1536	resource_cleanup((IWineD3DResource *)buffer);
1537	return hr;
1538	}
1539	}
1540
1541	buffer->maps = HeapAlloc(GetProcessHeap(), 0, sizeof(*buffer->maps));
1542	if (!buffer->maps)
1543	{
1544	ERR("Out of memory\n");
1545	buffer_UnLoad((IWineD3DBuffer *)buffer);
1546	resource_cleanup((IWineD3DResource *)buffer);
1547	return E_OUTOFMEMORY;
1548	}
1549	buffer->maps_size = 1;
1550
1551	return WINED3D_OK;
1552	}

Note: See TracBrowser for help on using the repository browser.

source: vbox/trunk/src/VBox/Additions/WINNT/Graphics/Wine/wined3d/buffer.c@ 33252

Download in other formats: