wined3d: Check the writemask before activating the oC0 optimization.
[wine] / dlls / wined3d / buffer.c
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 #include "config.h"
26 #include "wine/port.h"
27
28 #include "wined3d_private.h"
29
30 WINE_DEFAULT_DEBUG_CHANNEL(d3d);
31
32 #define VB_MAXDECLCHANGES     100     /* After that number of decl changes we stop converting */
33 #define VB_RESETDECLCHANGE    1000    /* Reset the decl changecount after that number of draws */
34 #define VB_MAXFULLCONVERSIONS 5       /* Number of full conversions before we stop converting */
35 #define VB_RESETFULLCONVS     20      /* Reset full conversion counts after that number of draws */
36
37 static inline BOOL buffer_add_dirty_area(struct wined3d_buffer *This, UINT offset, UINT size)
38 {
39     if (!This->buffer_object) return TRUE;
40
41     if (This->maps_size <= This->modified_areas)
42     {
43         void *new = HeapReAlloc(GetProcessHeap(), 0, This->maps,
44                                 This->maps_size * 2 * sizeof(*This->maps));
45         if (!new)
46         {
47             ERR("Out of memory\n");
48             return FALSE;
49         }
50         else
51         {
52             This->maps = new;
53             This->maps_size *= 2;
54         }
55     }
56
57     if(offset > This->resource.size || offset + size > This->resource.size)
58     {
59         WARN("Invalid range dirtified, marking entire buffer dirty\n");
60         offset = 0;
61         size = This->resource.size;
62     }
63     else if(!offset && !size)
64     {
65         size = This->resource.size;
66     }
67
68     This->maps[This->modified_areas].offset = offset;
69     This->maps[This->modified_areas].size = size;
70     This->modified_areas++;
71     return TRUE;
72 }
73
74 static inline void buffer_clear_dirty_areas(struct wined3d_buffer *This)
75 {
76     This->modified_areas = 0;
77 }
78
79 static inline BOOL buffer_is_dirty(struct wined3d_buffer *This)
80 {
81     return This->modified_areas != 0;
82 }
83
84 static inline BOOL buffer_is_fully_dirty(struct wined3d_buffer *This)
85 {
86     unsigned int i;
87
88     for(i = 0; i < This->modified_areas; i++)
89     {
90         if(This->maps[i].offset == 0 && This->maps[i].size == This->resource.size)
91         {
92             return TRUE;
93         }
94     }
95     return FALSE;
96 }
97
98 /* Context activation is done by the caller */
99 static void delete_gl_buffer(struct wined3d_buffer *This, const struct wined3d_gl_info *gl_info)
100 {
101     if(!This->buffer_object) return;
102
103     ENTER_GL();
104     GL_EXTCALL(glDeleteBuffersARB(1, &This->buffer_object));
105     checkGLcall("glDeleteBuffersARB");
106     LEAVE_GL();
107     This->buffer_object = 0;
108
109     if(This->query)
110     {
111         wined3d_event_query_destroy(This->query);
112         This->query = NULL;
113     }
114     This->flags &= ~WINED3D_BUFFER_APPLESYNC;
115 }
116
117 /* Context activation is done by the caller. */
118 static void buffer_create_buffer_object(struct wined3d_buffer *This, const struct wined3d_gl_info *gl_info)
119 {
120     GLenum error, gl_usage;
121
122     TRACE("Creating an OpenGL vertex buffer object for IWineD3DVertexBuffer %p Usage(%s)\n",
123             This, debug_d3dusage(This->resource.usage));
124
125     ENTER_GL();
126
127     /* Make sure that the gl error is cleared. Do not use checkGLcall
128     * here because checkGLcall just prints a fixme and continues. However,
129     * if an error during VBO creation occurs we can fall back to non-vbo operation
130     * with full functionality(but performance loss)
131     */
132     while (glGetError() != GL_NO_ERROR);
133
134     /* Basically the FVF parameter passed to CreateVertexBuffer is no good
135      * It is the FVF set with IWineD3DDevice::SetFVF or the Vertex Declaration set with
136      * IWineD3DDevice::SetVertexDeclaration that decides how the vertices in the buffer
137      * look like. This means that on each DrawPrimitive call the vertex buffer has to be verified
138      * to check if the rhw and color values are in the correct format.
139      */
140
141     GL_EXTCALL(glGenBuffersARB(1, &This->buffer_object));
142     error = glGetError();
143     if (!This->buffer_object || error != GL_NO_ERROR)
144     {
145         ERR("Failed to create a VBO with error %s (%#x)\n", debug_glerror(error), error);
146         LEAVE_GL();
147         goto fail;
148     }
149
150     if(This->buffer_type_hint == GL_ELEMENT_ARRAY_BUFFER_ARB)
151     {
152         IWineD3DDeviceImpl_MarkStateDirty(This->resource.device, STATE_INDEXBUFFER);
153     }
154     GL_EXTCALL(glBindBufferARB(This->buffer_type_hint, This->buffer_object));
155     error = glGetError();
156     if (error != GL_NO_ERROR)
157     {
158         ERR("Failed to bind the VBO with error %s (%#x)\n", debug_glerror(error), error);
159         LEAVE_GL();
160         goto fail;
161     }
162
163     /* Don't use static, because dx apps tend to update the buffer
164      * quite often even if they specify 0 usage.
165      */
166     if(This->resource.usage & WINED3DUSAGE_DYNAMIC)
167     {
168         TRACE("Gl usage = GL_STREAM_DRAW_ARB\n");
169         gl_usage = GL_STREAM_DRAW_ARB;
170
171         if(gl_info->supported[APPLE_FLUSH_BUFFER_RANGE])
172         {
173             GL_EXTCALL(glBufferParameteriAPPLE(This->buffer_type_hint, GL_BUFFER_FLUSHING_UNMAP_APPLE, GL_FALSE));
174             checkGLcall("glBufferParameteriAPPLE(This->buffer_type_hint, GL_BUFFER_FLUSHING_UNMAP_APPLE, GL_FALSE)");
175             This->flags |= WINED3D_BUFFER_FLUSH;
176
177             GL_EXTCALL(glBufferParameteriAPPLE(This->buffer_type_hint, GL_BUFFER_SERIALIZED_MODIFY_APPLE, GL_FALSE));
178             checkGLcall("glBufferParameteriAPPLE(This->buffer_type_hint, GL_BUFFER_SERIALIZED_MODIFY_APPLE, GL_FALSE)");
179             This->flags |= WINED3D_BUFFER_APPLESYNC;
180         }
181         /* No setup is needed here for GL_ARB_map_buffer_range */
182     }
183     else
184     {
185         TRACE("Gl usage = GL_DYNAMIC_DRAW_ARB\n");
186         gl_usage = GL_DYNAMIC_DRAW_ARB;
187     }
188
189     /* Reserve memory for the buffer. The amount of data won't change
190      * so we are safe with calling glBufferData once and
191      * calling glBufferSubData on updates. Upload the actual data in case
192      * we're not double buffering, so we can release the heap mem afterwards
193      */
194     GL_EXTCALL(glBufferDataARB(This->buffer_type_hint, This->resource.size, This->resource.allocatedMemory, gl_usage));
195     error = glGetError();
196     LEAVE_GL();
197     if (error != GL_NO_ERROR)
198     {
199         ERR("glBufferDataARB failed with error %s (%#x)\n", debug_glerror(error), error);
200         goto fail;
201     }
202
203     This->buffer_object_size = This->resource.size;
204     This->buffer_object_usage = gl_usage;
205
206     if(This->flags & WINED3D_BUFFER_DOUBLEBUFFER)
207     {
208         if(!buffer_add_dirty_area(This, 0, 0))
209         {
210             ERR("buffer_add_dirty_area failed, this is not expected\n");
211             goto fail;
212         }
213     }
214     else
215     {
216         HeapFree(GetProcessHeap(), 0, This->resource.heapMemory);
217         This->resource.allocatedMemory = NULL;
218         This->resource.heapMemory = NULL;
219     }
220
221     return;
222
223 fail:
224     /* Clean up all vbo init, but continue because we can work without a vbo :-) */
225     ERR("Failed to create a vertex buffer object. Continuing, but performance issues may occur\n");
226     delete_gl_buffer(This, gl_info);
227     buffer_clear_dirty_areas(This);
228 }
229
230 static BOOL buffer_process_converted_attribute(struct wined3d_buffer *This,
231         const enum wined3d_buffer_conversion_type conversion_type,
232         const struct wined3d_stream_info_element *attrib, DWORD *stride_this_run)
233 {
234     DWORD attrib_size;
235     BOOL ret = FALSE;
236     unsigned int i;
237     DWORD offset = This->resource.device->stateBlock->streamOffset[attrib->stream_idx];
238     DWORD_PTR data;
239
240     /* Check for some valid situations which cause us pain. One is if the buffer is used for
241      * constant attributes(stride = 0), the other one is if the buffer is used on two streams
242      * with different strides. In the 2nd case we might have to drop conversion entirely,
243      * it is possible that the same bytes are once read as FLOAT2 and once as UBYTE4N.
244      */
245     if (!attrib->stride)
246     {
247         FIXME("%s used with stride 0, let's hope we get the vertex stride from somewhere else\n",
248                 debug_d3dformat(attrib->format_desc->format));
249     }
250     else if(attrib->stride != *stride_this_run && *stride_this_run)
251     {
252         FIXME("Got two concurrent strides, %d and %d\n", attrib->stride, *stride_this_run);
253     }
254     else
255     {
256         *stride_this_run = attrib->stride;
257         if (This->stride != *stride_this_run)
258         {
259             /* We rely that this happens only on the first converted attribute that is found,
260              * if at all. See above check
261              */
262             TRACE("Reconverting because converted attributes occur, and the stride changed\n");
263             This->stride = *stride_this_run;
264             HeapFree(GetProcessHeap(), HEAP_ZERO_MEMORY, This->conversion_map);
265             This->conversion_map = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY,
266                     sizeof(*This->conversion_map) * This->stride);
267             ret = TRUE;
268         }
269     }
270
271     data = (((DWORD_PTR)attrib->data) + offset) % This->stride;
272     attrib_size = attrib->format_desc->component_count * attrib->format_desc->component_size;
273     for (i = 0; i < attrib_size; ++i)
274     {
275         DWORD_PTR idx = (data + i) % This->stride;
276         if (This->conversion_map[idx] != conversion_type)
277         {
278             TRACE("Byte %ld in vertex changed\n", idx);
279             TRACE("It was type %d, is %d now\n", This->conversion_map[idx], conversion_type);
280             ret = TRUE;
281             This->conversion_map[idx] = conversion_type;
282         }
283     }
284
285     return ret;
286 }
287
288 static BOOL buffer_check_attribute(struct wined3d_buffer *This, const struct wined3d_stream_info *si,
289         UINT attrib_idx, const BOOL check_d3dcolor, const BOOL is_ffp_position, const BOOL is_ffp_color,
290         DWORD *stride_this_run, BOOL *float16_used)
291 {
292     const struct wined3d_stream_info_element *attrib = &si->elements[attrib_idx];
293     IWineD3DDeviceImpl *device = This->resource.device;
294     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
295     BOOL ret = FALSE;
296     WINED3DFORMAT format;
297
298     /* Ignore attributes that do not have our vbo. After that check we can be sure that the attribute is
299      * there, on nonexistent attribs the vbo is 0.
300      */
301     if (!(si->use_map & (1 << attrib_idx))
302             || attrib->buffer_object != This->buffer_object)
303         return FALSE;
304
305     format = attrib->format_desc->format;
306     /* Look for newly appeared conversion */
307     if (!gl_info->supported[ARB_HALF_FLOAT_VERTEX]
308             && (format == WINED3DFMT_R16G16_FLOAT || format == WINED3DFMT_R16G16B16A16_FLOAT))
309     {
310         ret = buffer_process_converted_attribute(This, CONV_FLOAT16_2, attrib, stride_this_run);
311
312         if (is_ffp_position) FIXME("Test FLOAT16 fixed function processing positions\n");
313         else if (is_ffp_color) FIXME("test FLOAT16 fixed function processing colors\n");
314         *float16_used = TRUE;
315     }
316     else if (check_d3dcolor && format == WINED3DFMT_B8G8R8A8_UNORM)
317     {
318         ret = buffer_process_converted_attribute(This, CONV_D3DCOLOR, attrib, stride_this_run);
319
320         if (!is_ffp_color) FIXME("Test for non-color fixed function WINED3DFMT_B8G8R8A8_UNORM format\n");
321     }
322     else if (is_ffp_position && format == WINED3DFMT_R32G32B32A32_FLOAT)
323     {
324         ret = buffer_process_converted_attribute(This, CONV_POSITIONT, attrib, stride_this_run);
325     }
326     else if (This->conversion_map)
327     {
328         ret = buffer_process_converted_attribute(This, CONV_NONE, attrib, stride_this_run);
329     }
330
331     return ret;
332 }
333
334 static UINT *find_conversion_shift(struct wined3d_buffer *This,
335         const struct wined3d_stream_info *strided, UINT stride)
336 {
337     UINT *ret, i, j, shift, orig_type_size;
338
339     if (!stride)
340     {
341         TRACE("No shift\n");
342         return NULL;
343     }
344
345     This->conversion_stride = stride;
346     ret = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(DWORD) * stride);
347     for (i = 0; i < MAX_ATTRIBS; ++i)
348     {
349         WINED3DFORMAT format;
350
351         if (!(strided->use_map & (1 << i)) || strided->elements[i].buffer_object != This->buffer_object) continue;
352
353         format = strided->elements[i].format_desc->format;
354         if (format == WINED3DFMT_R16G16_FLOAT)
355         {
356             shift = 4;
357         }
358         else if (format == WINED3DFMT_R16G16B16A16_FLOAT)
359         {
360             shift = 8;
361             /* Pre-shift the last 4 bytes in the FLOAT16_4 by 4 bytes - this makes FLOAT16_2 and FLOAT16_4 conversions
362              * compatible
363              */
364             for (j = 4; j < 8; ++j)
365             {
366                 ret[(DWORD_PTR)strided->elements[i].data + j] += 4;
367             }
368         }
369         else
370         {
371             shift = 0;
372         }
373         This->conversion_stride += shift;
374
375         if (shift)
376         {
377             orig_type_size = strided->elements[i].format_desc->component_count
378                     * strided->elements[i].format_desc->component_size;
379             for (j = (DWORD_PTR)strided->elements[i].data + orig_type_size; j < stride; ++j)
380             {
381                 ret[j] += shift;
382             }
383         }
384     }
385
386     if (TRACE_ON(d3d))
387     {
388         TRACE("Dumping conversion shift:\n");
389         for (i = 0; i < stride; ++i)
390         {
391             TRACE("[%d]", ret[i]);
392         }
393         TRACE("\n");
394     }
395
396     return ret;
397 }
398
399 static BOOL buffer_find_decl(struct wined3d_buffer *This)
400 {
401     IWineD3DDeviceImpl *device = This->resource.device;
402     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
403     const struct wined3d_stream_info *si = &device->strided_streams;
404     UINT stride_this_run = 0;
405     BOOL float16_used = FALSE;
406     BOOL ret = FALSE;
407     unsigned int i;
408
409     /* In d3d7 the vertex buffer declaration NEVER changes because it is stored in the d3d7 vertex buffer.
410      * Once we have our declaration there is no need to look it up again. Index buffers also never need
411      * conversion, so once the (empty) conversion structure is created don't bother checking again
412      */
413     if (This->flags & WINED3D_BUFFER_HASDESC)
414     {
415         if(This->resource.usage & WINED3DUSAGE_STATICDECL) return FALSE;
416     }
417
418     TRACE("Finding vertex buffer conversion information\n");
419     /* Certain declaration types need some fixups before we can pass them to
420      * opengl. This means D3DCOLOR attributes with fixed function vertex
421      * processing, FLOAT4 POSITIONT with fixed function, and FLOAT16 if
422      * GL_ARB_half_float_vertex is not supported.
423      *
424      * Note for d3d8 and d3d9:
425      * The vertex buffer FVF doesn't help with finding them, we have to use
426      * the decoded vertex declaration and pick the things that concern the
427      * current buffer. A problem with this is that this can change between
428      * draws, so we have to validate the information and reprocess the buffer
429      * if it changes, and avoid false positives for performance reasons.
430      * WineD3D doesn't even know the vertex buffer any more, it is managed
431      * by the client libraries and passed to SetStreamSource and ProcessVertices
432      * as needed.
433      *
434      * We have to distinguish between vertex shaders and fixed function to
435      * pick the way we access the strided vertex information.
436      *
437      * This code sets up a per-byte array with the size of the detected
438      * stride of the arrays in the buffer. For each byte we have a field
439      * that marks the conversion needed on this byte. For example, the
440      * following declaration with fixed function vertex processing:
441      *
442      *      POSITIONT, FLOAT4
443      *      NORMAL, FLOAT3
444      *      DIFFUSE, FLOAT16_4
445      *      SPECULAR, D3DCOLOR
446      *
447      * Will result in
448      * {                 POSITIONT                    }{             NORMAL                }{    DIFFUSE          }{SPECULAR }
449      * [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]
450      *
451      * Where in this example map P means 4 component position conversion, 0
452      * means no conversion, F means FLOAT16_2 conversion and C means D3DCOLOR
453      * conversion (red / blue swizzle).
454      *
455      * If we're doing conversion and the stride changes we have to reconvert
456      * the whole buffer. Note that we do not mind if the semantic changes,
457      * we only care for the conversion type. So if the NORMAL is replaced
458      * with a TEXCOORD, nothing has to be done, or if the DIFFUSE is replaced
459      * with a D3DCOLOR BLENDWEIGHT we can happily dismiss the change. Some
460      * conversion types depend on the semantic as well, for example a FLOAT4
461      * texcoord needs no conversion while a FLOAT4 positiont needs one
462      */
463     if (use_vs(device->stateBlock))
464     {
465         TRACE("vshader\n");
466         /* If the current vertex declaration is marked for no half float conversion don't bother to
467          * analyse the strided streams in depth, just set them up for no conversion. Return decl changed
468          * if we used conversion before
469          */
470         if (!((IWineD3DVertexDeclarationImpl *) device->stateBlock->vertexDecl)->half_float_conv_needed)
471         {
472             if (This->conversion_map)
473             {
474                 TRACE("Now using shaders without conversion, but conversion used before\n");
475                 HeapFree(GetProcessHeap(), 0, This->conversion_map);
476                 HeapFree(GetProcessHeap(), 0, This->conversion_shift);
477                 This->conversion_map = NULL;
478                 This->stride = 0;
479                 This->conversion_shift = NULL;
480                 This->conversion_stride = 0;
481                 return TRUE;
482             }
483             else
484             {
485                 return FALSE;
486             }
487         }
488         for (i = 0; i < MAX_ATTRIBS; ++i)
489         {
490             ret = buffer_check_attribute(This, si, i, FALSE, FALSE, FALSE, &stride_this_run, &float16_used) || ret;
491         }
492
493         /* Recalculate the conversion shift map if the declaration has changed,
494          * and we're using float16 conversion or used it on the last run
495          */
496         if (ret && (float16_used || This->conversion_map))
497         {
498             HeapFree(GetProcessHeap(), 0, This->conversion_shift);
499             This->conversion_shift = find_conversion_shift(This, si, This->stride);
500         }
501     }
502     else
503     {
504         /* Fixed function is a bit trickier. We have to take care for D3DCOLOR types, FLOAT4 positions and of course
505          * FLOAT16s if not supported. Also, we can't iterate over the array, so use macros to generate code for all
506          * the attributes that our current fixed function pipeline implementation cares for.
507          */
508         BOOL support_d3dcolor = gl_info->supported[ARB_VERTEX_ARRAY_BGRA];
509         ret = buffer_check_attribute(This, si, WINED3D_FFP_POSITION,
510                 TRUE, TRUE,  FALSE, &stride_this_run, &float16_used) || ret;
511         ret = buffer_check_attribute(This, si, WINED3D_FFP_NORMAL,
512                 TRUE, FALSE, FALSE, &stride_this_run, &float16_used) || ret;
513         ret = buffer_check_attribute(This, si, WINED3D_FFP_DIFFUSE,
514                 !support_d3dcolor, FALSE, TRUE,  &stride_this_run, &float16_used) || ret;
515         ret = buffer_check_attribute(This, si, WINED3D_FFP_SPECULAR,
516                 !support_d3dcolor, FALSE, TRUE,  &stride_this_run, &float16_used) || ret;
517         ret = buffer_check_attribute(This, si, WINED3D_FFP_TEXCOORD0,
518                 TRUE, FALSE, FALSE, &stride_this_run, &float16_used) || ret;
519         ret = buffer_check_attribute(This, si, WINED3D_FFP_TEXCOORD1,
520                 TRUE, FALSE, FALSE, &stride_this_run, &float16_used) || ret;
521         ret = buffer_check_attribute(This, si, WINED3D_FFP_TEXCOORD2,
522                 TRUE, FALSE, FALSE, &stride_this_run, &float16_used) || ret;
523         ret = buffer_check_attribute(This, si, WINED3D_FFP_TEXCOORD3,
524                 TRUE, FALSE, FALSE, &stride_this_run, &float16_used) || ret;
525         ret = buffer_check_attribute(This, si, WINED3D_FFP_TEXCOORD4,
526                 TRUE, FALSE, FALSE, &stride_this_run, &float16_used) || ret;
527         ret = buffer_check_attribute(This, si, WINED3D_FFP_TEXCOORD5,
528                 TRUE, FALSE, FALSE, &stride_this_run, &float16_used) || ret;
529         ret = buffer_check_attribute(This, si, WINED3D_FFP_TEXCOORD6,
530                 TRUE, FALSE, FALSE, &stride_this_run, &float16_used) || ret;
531         ret = buffer_check_attribute(This, si, WINED3D_FFP_TEXCOORD7,
532                 TRUE, FALSE, FALSE, &stride_this_run, &float16_used) || ret;
533
534         if (float16_used) FIXME("Float16 conversion used with fixed function vertex processing\n");
535     }
536
537     if (stride_this_run == 0 && This->conversion_map)
538     {
539         /* Sanity test */
540         if (!ret) ERR("no converted attributes found, old conversion map exists, and no declaration change?\n");
541         HeapFree(GetProcessHeap(), 0, This->conversion_map);
542         This->conversion_map = NULL;
543         This->stride = 0;
544     }
545
546     if (ret) TRACE("Conversion information changed\n");
547
548     return ret;
549 }
550
551 /* Context activation is done by the caller. */
552 static void buffer_check_buffer_object_size(struct wined3d_buffer *This, const struct wined3d_gl_info *gl_info)
553 {
554     UINT size = This->conversion_stride ?
555             This->conversion_stride * (This->resource.size / This->stride) : This->resource.size;
556     if (This->buffer_object_size != size)
557     {
558         TRACE("Old size %u, creating new size %u\n", This->buffer_object_size, size);
559
560         if(This->buffer_type_hint == GL_ELEMENT_ARRAY_BUFFER_ARB)
561         {
562             IWineD3DDeviceImpl_MarkStateDirty(This->resource.device, STATE_INDEXBUFFER);
563         }
564
565         /* Rescue the data before resizing the buffer object if we do not have our backup copy */
566         if(!(This->flags & WINED3D_BUFFER_DOUBLEBUFFER))
567         {
568             buffer_get_sysmem(This, gl_info);
569         }
570
571         ENTER_GL();
572         GL_EXTCALL(glBindBufferARB(This->buffer_type_hint, This->buffer_object));
573         checkGLcall("glBindBufferARB");
574         GL_EXTCALL(glBufferDataARB(This->buffer_type_hint, size, NULL, This->buffer_object_usage));
575         This->buffer_object_size = size;
576         checkGLcall("glBufferDataARB");
577         LEAVE_GL();
578     }
579 }
580
581 static inline void fixup_d3dcolor(DWORD *dst_color)
582 {
583     DWORD src_color = *dst_color;
584
585     /* Color conversion like in drawStridedSlow. watch out for little endianity
586      * If we want that stuff to work on big endian machines too we have to consider more things
587      *
588      * 0xff000000: Alpha mask
589      * 0x00ff0000: Blue mask
590      * 0x0000ff00: Green mask
591      * 0x000000ff: Red mask
592      */
593     *dst_color = 0;
594     *dst_color |= (src_color & 0xff00ff00);         /* Alpha Green */
595     *dst_color |= (src_color & 0x00ff0000) >> 16;   /* Red */
596     *dst_color |= (src_color & 0x000000ff) << 16;   /* Blue */
597 }
598
599 static inline void fixup_transformed_pos(float *p)
600 {
601     /* rhw conversion like in position_float4(). */
602     if (p[3] != 1.0f && p[3] != 0.0f)
603     {
604         float w = 1.0f / p[3];
605         p[0] *= w;
606         p[1] *= w;
607         p[2] *= w;
608         p[3] = w;
609     }
610 }
611
612 /* Context activation is done by the caller. */
613 const BYTE *buffer_get_memory(IWineD3DBuffer *iface, const struct wined3d_gl_info *gl_info, GLuint *buffer_object)
614 {
615     struct wined3d_buffer *This = (struct wined3d_buffer *)iface;
616
617     *buffer_object = This->buffer_object;
618     if (!This->buffer_object)
619     {
620         if (This->flags & WINED3D_BUFFER_CREATEBO)
621         {
622             buffer_create_buffer_object(This, gl_info);
623             This->flags &= ~WINED3D_BUFFER_CREATEBO;
624             if (This->buffer_object)
625             {
626                 *buffer_object = This->buffer_object;
627                 return NULL;
628             }
629         }
630         return This->resource.allocatedMemory;
631     }
632     else
633     {
634         return NULL;
635     }
636 }
637
638 /* IUnknown methods */
639
640 static HRESULT STDMETHODCALLTYPE buffer_QueryInterface(IWineD3DBuffer *iface,
641         REFIID riid, void **object)
642 {
643     TRACE("iface %p, riid %s, object %p\n", iface, debugstr_guid(riid), object);
644
645     if (IsEqualGUID(riid, &IID_IWineD3DBuffer)
646             || IsEqualGUID(riid, &IID_IWineD3DResource)
647             || IsEqualGUID(riid, &IID_IWineD3DBase)
648             || IsEqualGUID(riid, &IID_IUnknown))
649     {
650         IUnknown_AddRef(iface);
651         *object = iface;
652         return S_OK;
653     }
654
655     WARN("%s not implemented, returning E_NOINTERFACE\n", debugstr_guid(riid));
656
657     *object = NULL;
658
659     return E_NOINTERFACE;
660 }
661
662 static ULONG STDMETHODCALLTYPE buffer_AddRef(IWineD3DBuffer *iface)
663 {
664     struct wined3d_buffer *This = (struct wined3d_buffer *)iface;
665     ULONG refcount = InterlockedIncrement(&This->resource.ref);
666
667     TRACE("%p increasing refcount to %u\n", This, refcount);
668
669     return refcount;
670 }
671
672 /* Context activation is done by the caller. */
673 BYTE *buffer_get_sysmem(struct wined3d_buffer *This, const struct wined3d_gl_info *gl_info)
674 {
675     /* AllocatedMemory exists if the buffer is double buffered or has no buffer object at all */
676     if(This->resource.allocatedMemory) return This->resource.allocatedMemory;
677
678     This->resource.heapMemory = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, This->resource.size + RESOURCE_ALIGNMENT);
679     This->resource.allocatedMemory = (BYTE *)(((ULONG_PTR)This->resource.heapMemory + (RESOURCE_ALIGNMENT - 1)) & ~(RESOURCE_ALIGNMENT - 1));
680     ENTER_GL();
681     GL_EXTCALL(glBindBufferARB(This->buffer_type_hint, This->buffer_object));
682     GL_EXTCALL(glGetBufferSubDataARB(This->buffer_type_hint, 0, This->resource.size, This->resource.allocatedMemory));
683     LEAVE_GL();
684     This->flags |= WINED3D_BUFFER_DOUBLEBUFFER;
685
686     return This->resource.allocatedMemory;
687 }
688
689 static void STDMETHODCALLTYPE buffer_UnLoad(IWineD3DBuffer *iface)
690 {
691     struct wined3d_buffer *This = (struct wined3d_buffer *)iface;
692
693     TRACE("iface %p\n", iface);
694
695     if (This->buffer_object)
696     {
697         IWineD3DDeviceImpl *device = This->resource.device;
698         struct wined3d_context *context;
699
700         context = context_acquire(device, NULL);
701
702         /* Download the buffer, but don't permanently enable double buffering */
703         if(!(This->flags & WINED3D_BUFFER_DOUBLEBUFFER))
704         {
705             buffer_get_sysmem(This, context->gl_info);
706             This->flags &= ~WINED3D_BUFFER_DOUBLEBUFFER;
707         }
708
709         delete_gl_buffer(This, context->gl_info);
710         This->flags |= WINED3D_BUFFER_CREATEBO; /* Recreate the buffer object next load */
711         buffer_clear_dirty_areas(This);
712
713         context_release(context);
714
715         HeapFree(GetProcessHeap(), 0, This->conversion_shift);
716         This->conversion_shift = NULL;
717         HeapFree(GetProcessHeap(), 0, This->conversion_map);
718         This->conversion_map = NULL;
719         This->stride = 0;
720         This->conversion_stride = 0;
721         This->flags &= ~WINED3D_BUFFER_HASDESC;
722     }
723 }
724
725 static ULONG STDMETHODCALLTYPE buffer_Release(IWineD3DBuffer *iface)
726 {
727     struct wined3d_buffer *This = (struct wined3d_buffer *)iface;
728     ULONG refcount = InterlockedDecrement(&This->resource.ref);
729
730     TRACE("%p decreasing refcount to %u\n", This, refcount);
731
732     if (!refcount)
733     {
734         buffer_UnLoad(iface);
735         resource_cleanup((IWineD3DResource *)iface);
736         This->resource.parent_ops->wined3d_object_destroyed(This->resource.parent);
737         HeapFree(GetProcessHeap(), 0, This->maps);
738         HeapFree(GetProcessHeap(), 0, This);
739     }
740
741     return refcount;
742 }
743
744 /* IWineD3DBase methods */
745
746 static HRESULT STDMETHODCALLTYPE buffer_GetParent(IWineD3DBuffer *iface, IUnknown **parent)
747 {
748     return resource_get_parent((IWineD3DResource *)iface, parent);
749 }
750
751 /* IWineD3DResource methods */
752
753 static HRESULT STDMETHODCALLTYPE buffer_SetPrivateData(IWineD3DBuffer *iface,
754         REFGUID guid, const void *data, DWORD data_size, DWORD flags)
755 {
756     return resource_set_private_data((IWineD3DResource *)iface, guid, data, data_size, flags);
757 }
758
759 static HRESULT STDMETHODCALLTYPE buffer_GetPrivateData(IWineD3DBuffer *iface,
760         REFGUID guid, void *data, DWORD *data_size)
761 {
762     return resource_get_private_data((IWineD3DResource *)iface, guid, data, data_size);
763 }
764
765 static HRESULT STDMETHODCALLTYPE buffer_FreePrivateData(IWineD3DBuffer *iface, REFGUID guid)
766 {
767     return resource_free_private_data((IWineD3DResource *)iface, guid);
768 }
769
770 static DWORD STDMETHODCALLTYPE buffer_SetPriority(IWineD3DBuffer *iface, DWORD priority)
771 {
772     return resource_set_priority((IWineD3DResource *)iface, priority);
773 }
774
775 static DWORD STDMETHODCALLTYPE buffer_GetPriority(IWineD3DBuffer *iface)
776 {
777     return resource_get_priority((IWineD3DResource *)iface);
778 }
779
780 /* The caller provides a context and binds the buffer */
781 static void buffer_sync_apple(struct wined3d_buffer *This, DWORD flags, const struct wined3d_gl_info *gl_info)
782 {
783     enum wined3d_event_query_result ret;
784
785     /* No fencing needs to be done if the app promises not to overwrite
786      * existing data */
787     if(flags & WINED3DLOCK_NOOVERWRITE) return;
788     if(flags & WINED3DLOCK_DISCARD)
789     {
790         ENTER_GL();
791         GL_EXTCALL(glBufferDataARB(This->buffer_type_hint, This->resource.size, NULL, This->buffer_object_usage));
792         checkGLcall("glBufferDataARB\n");
793         LEAVE_GL();
794         return;
795     }
796
797     if(!This->query)
798     {
799         TRACE("Creating event query for buffer %p\n", This);
800
801         if (!wined3d_event_query_supported(gl_info))
802         {
803             FIXME("Event queries not supported, dropping async buffer locks.\n");
804             goto drop_query;
805         }
806
807         This->query = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*This->query));
808         if (!This->query)
809         {
810             ERR("Failed to allocate event query memory, dropping async buffer locks.\n");
811             goto drop_query;
812         }
813
814         /* Since we don't know about old draws a glFinish is needed once */
815         wglFinish();
816         return;
817     }
818     TRACE("Synchronizing buffer %p\n", This);
819     ret = wined3d_event_query_finish(This->query, This->resource.device);
820     switch(ret)
821     {
822         case WINED3D_EVENT_QUERY_NOT_STARTED:
823         case WINED3D_EVENT_QUERY_OK:
824             /* All done */
825             return;
826
827         case WINED3D_EVENT_QUERY_WRONG_THREAD:
828             WARN("Cannot synchronize buffer lock due to a thread conflict\n");
829             goto drop_query;
830
831         default:
832             ERR("wined3d_event_query_finish returned %u, dropping async buffer locks\n", ret);
833             goto drop_query;
834     }
835
836 drop_query:
837     if(This->query)
838     {
839         wined3d_event_query_destroy(This->query);
840         This->query = NULL;
841     }
842
843     wglFinish();
844     ENTER_GL();
845     GL_EXTCALL(glBufferParameteriAPPLE(This->buffer_type_hint, GL_BUFFER_SERIALIZED_MODIFY_APPLE, GL_TRUE));
846     checkGLcall("glBufferParameteriAPPLE(This->buffer_type_hint, GL_BUFFER_SERIALIZED_MODIFY_APPLE, GL_TRUE)");
847     LEAVE_GL();
848     This->flags &= ~WINED3D_BUFFER_APPLESYNC;
849 }
850
851 /* The caller provides a GL context */
852 static void buffer_direct_upload(struct wined3d_buffer *This, const struct wined3d_gl_info *gl_info, DWORD flags)
853 {
854         BYTE *map;
855         UINT start = 0, len = 0;
856
857         ENTER_GL();
858         GL_EXTCALL(glBindBufferARB(This->buffer_type_hint, This->buffer_object));
859         checkGLcall("glBindBufferARB");
860         if (gl_info->supported[ARB_MAP_BUFFER_RANGE])
861         {
862             GLbitfield mapflags;
863             mapflags = GL_MAP_WRITE_BIT | GL_MAP_FLUSH_EXPLICIT_BIT;
864             if (flags & WINED3D_BUFFER_DISCARD)
865             {
866                 mapflags |= GL_MAP_UNSYNCHRONIZED_BIT | GL_MAP_INVALIDATE_BUFFER_BIT;
867             }
868             else if (flags & WINED3D_BUFFER_NOSYNC)
869             {
870                 mapflags |= GL_MAP_UNSYNCHRONIZED_BIT;
871             }
872             map = GL_EXTCALL(glMapBufferRange(This->buffer_type_hint, 0,
873                                               This->resource.size, mapflags));
874             checkGLcall("glMapBufferRange");
875         }
876         else
877         {
878             if (This->flags & WINED3D_BUFFER_APPLESYNC)
879             {
880                 DWORD syncflags = 0;
881                 if (flags & WINED3D_BUFFER_DISCARD) syncflags |= WINED3DLOCK_DISCARD;
882                 if (flags & WINED3D_BUFFER_NOSYNC) syncflags |= WINED3DLOCK_NOOVERWRITE;
883                 LEAVE_GL();
884                 buffer_sync_apple(This, syncflags, gl_info);
885                 ENTER_GL();
886             }
887             map = GL_EXTCALL(glMapBufferARB(This->buffer_type_hint, GL_WRITE_ONLY_ARB));
888             checkGLcall("glMapBufferARB");
889         }
890         if (!map)
891         {
892             LEAVE_GL();
893             ERR("Failed to map opengl buffer\n");
894             return;
895         }
896
897         while(This->modified_areas)
898         {
899             This->modified_areas--;
900             start = This->maps[This->modified_areas].offset;
901             len = This->maps[This->modified_areas].size;
902
903             memcpy(map + start, This->resource.allocatedMemory + start, len);
904
905             if (gl_info->supported[ARB_MAP_BUFFER_RANGE])
906             {
907                 GL_EXTCALL(glFlushMappedBufferRange(This->buffer_type_hint, start, len));
908                 checkGLcall("glFlushMappedBufferRange");
909             }
910             else if (This->flags & WINED3D_BUFFER_FLUSH)
911             {
912                 GL_EXTCALL(glFlushMappedBufferRangeAPPLE(This->buffer_type_hint, start, len));
913                 checkGLcall("glFlushMappedBufferRangeAPPLE");
914             }
915         }
916         GL_EXTCALL(glUnmapBufferARB(This->buffer_type_hint));
917         checkGLcall("glUnmapBufferARB");
918         LEAVE_GL();
919 }
920
921 static void STDMETHODCALLTYPE buffer_PreLoad(IWineD3DBuffer *iface)
922 {
923     struct wined3d_buffer *This = (struct wined3d_buffer *)iface;
924     IWineD3DDeviceImpl *device = This->resource.device;
925     UINT start = 0, end = 0, len = 0, vertices;
926     const struct wined3d_gl_info *gl_info;
927     struct wined3d_context *context;
928     BOOL decl_changed = FALSE;
929     unsigned int i, j;
930     BYTE *data;
931     DWORD flags = This->flags & (WINED3D_BUFFER_NOSYNC | WINED3D_BUFFER_DISCARD);
932
933     TRACE("iface %p\n", iface);
934     This->flags &= ~(WINED3D_BUFFER_NOSYNC | WINED3D_BUFFER_DISCARD);
935
936     context = context_acquire(device, NULL);
937     gl_info = context->gl_info;
938
939     if (!This->buffer_object)
940     {
941         /* TODO: Make converting independent from VBOs */
942         if (This->flags & WINED3D_BUFFER_CREATEBO)
943         {
944             buffer_create_buffer_object(This, gl_info);
945             This->flags &= ~WINED3D_BUFFER_CREATEBO;
946         }
947         else
948         {
949             /* Not doing any conversion */
950             goto end;
951         }
952     }
953
954     /* Reading the declaration makes only sense if the stateblock is finalized and the buffer bound to a stream */
955     if (device->isInDraw && This->bind_count > 0)
956     {
957         decl_changed = buffer_find_decl(This);
958         This->flags |= WINED3D_BUFFER_HASDESC;
959     }
960
961     if (!decl_changed && !(This->flags & WINED3D_BUFFER_HASDESC && buffer_is_dirty(This)))
962     {
963         context_release(context);
964         ++This->draw_count;
965         if (This->draw_count > VB_RESETDECLCHANGE) This->decl_change_count = 0;
966         if (This->draw_count > VB_RESETFULLCONVS) This->full_conversion_count = 0;
967         return;
968     }
969
970     /* If applications change the declaration over and over, reconverting all the time is a huge
971      * performance hit. So count the declaration changes and release the VBO if there are too many
972      * of them (and thus stop converting)
973      */
974     if (decl_changed)
975     {
976         ++This->decl_change_count;
977         This->draw_count = 0;
978
979         if (This->decl_change_count > VB_MAXDECLCHANGES ||
980             (This->conversion_map && (This->resource.usage & WINED3DUSAGE_DYNAMIC)))
981         {
982             FIXME("Too many declaration changes or converting dynamic buffer, stopping converting\n");
983
984             IWineD3DBuffer_UnLoad(iface);
985             This->flags &= ~WINED3D_BUFFER_CREATEBO;
986
987             /* The stream source state handler might have read the memory of the vertex buffer already
988              * and got the memory in the vbo which is not valid any longer. Dirtify the stream source
989              * to force a reload. This happens only once per changed vertexbuffer and should occur rather
990              * rarely
991              */
992             IWineD3DDeviceImpl_MarkStateDirty(device, STATE_STREAMSRC);
993             goto end;
994         }
995         buffer_check_buffer_object_size(This, gl_info);
996
997         /* The declaration changed, reload the whole buffer */
998         WARN("Reloading buffer because of decl change\n");
999         buffer_clear_dirty_areas(This);
1000         if(!buffer_add_dirty_area(This, 0, 0))
1001         {
1002             ERR("buffer_add_dirty_area failed, this is not expected\n");
1003             goto end;
1004         }
1005         /* Avoid unfenced updates, we might overwrite more areas of the buffer than the application
1006          * cleared for unsynchronized updates
1007          */
1008         flags = 0;
1009     }
1010     else
1011     {
1012         /* However, it is perfectly fine to change the declaration every now and then. We don't want a game that
1013          * changes it every minute drop the VBO after VB_MAX_DECL_CHANGES minutes. So count draws without
1014          * decl changes and reset the decl change count after a specific number of them
1015          */
1016         if(buffer_is_fully_dirty(This))
1017         {
1018             ++This->full_conversion_count;
1019             if(This->full_conversion_count > VB_MAXFULLCONVERSIONS)
1020             {
1021                 FIXME("Too many full buffer conversions, stopping converting\n");
1022                 IWineD3DBuffer_UnLoad(iface);
1023                 This->flags &= ~WINED3D_BUFFER_CREATEBO;
1024                 IWineD3DDeviceImpl_MarkStateDirty(device, STATE_STREAMSRC);
1025                 goto end;
1026             }
1027         }
1028         else
1029         {
1030             ++This->draw_count;
1031             if (This->draw_count > VB_RESETDECLCHANGE) This->decl_change_count = 0;
1032             if (This->draw_count > VB_RESETFULLCONVS) This->full_conversion_count = 0;
1033         }
1034     }
1035
1036     if(This->buffer_type_hint == GL_ELEMENT_ARRAY_BUFFER_ARB)
1037     {
1038         IWineD3DDeviceImpl_MarkStateDirty(This->resource.device, STATE_INDEXBUFFER);
1039     }
1040
1041     if (!This->conversion_map)
1042     {
1043         /* That means that there is nothing to fixup. Just upload from This->resource.allocatedMemory
1044          * directly into the vbo. Do not free the system memory copy because drawPrimitive may need it if
1045          * the stride is 0, for instancing emulation, vertex blending emulation or shader emulation.
1046          */
1047         TRACE("No conversion needed\n");
1048
1049         /* Nothing to do because we locked directly into the vbo */
1050         if (!(This->flags & WINED3D_BUFFER_DOUBLEBUFFER))
1051         {
1052             context_release(context);
1053             return;
1054         }
1055
1056         buffer_direct_upload(This, context->gl_info, flags);
1057
1058         context_release(context);
1059         return;
1060     }
1061
1062     if(!(This->flags & WINED3D_BUFFER_DOUBLEBUFFER))
1063     {
1064         buffer_get_sysmem(This, gl_info);
1065     }
1066
1067     /* Now for each vertex in the buffer that needs conversion */
1068     vertices = This->resource.size / This->stride;
1069
1070     if (This->conversion_shift)
1071     {
1072         TRACE("Shifted conversion\n");
1073         data = HeapAlloc(GetProcessHeap(), 0, vertices * This->conversion_stride);
1074
1075         start = 0;
1076         len = This->resource.size;
1077         end = start + len;
1078
1079         if (This->maps[0].offset || This->maps[0].size != This->resource.size)
1080         {
1081             FIXME("Implement partial buffer load with shifted conversion\n");
1082         }
1083
1084         for (i = start / This->stride; i < min((end / This->stride) + 1, vertices); ++i)
1085         {
1086             for (j = 0; j < This->stride; ++j)
1087             {
1088                 switch(This->conversion_map[j])
1089                 {
1090                     case CONV_NONE:
1091                         data[This->conversion_stride * i + j + This->conversion_shift[j]]
1092                                 = This->resource.allocatedMemory[This->stride * i + j];
1093                         break;
1094
1095                     case CONV_FLOAT16_2:
1096                     {
1097                         float *out = (float *)(&data[This->conversion_stride * i + j + This->conversion_shift[j]]);
1098                         const WORD *in = (WORD *)(&This->resource.allocatedMemory[i * This->stride + j]);
1099
1100                         out[1] = float_16_to_32(in + 1);
1101                         out[0] = float_16_to_32(in + 0);
1102                         j += 3;    /* Skip 3 additional bytes,as a FLOAT16_2 has 4 bytes */
1103                         break;
1104                     }
1105
1106                     default:
1107                         FIXME("Unimplemented conversion %d in shifted conversion\n", This->conversion_map[j]);
1108                         break;
1109                 }
1110             }
1111         }
1112
1113         ENTER_GL();
1114         GL_EXTCALL(glBindBufferARB(This->buffer_type_hint, This->buffer_object));
1115         checkGLcall("glBindBufferARB");
1116         GL_EXTCALL(glBufferSubDataARB(This->buffer_type_hint, 0, vertices * This->conversion_stride, data));
1117         checkGLcall("glBufferSubDataARB");
1118         LEAVE_GL();
1119     }
1120     else
1121     {
1122         data = HeapAlloc(GetProcessHeap(), 0, This->resource.size);
1123
1124         while(This->modified_areas)
1125         {
1126             This->modified_areas--;
1127             start = This->maps[This->modified_areas].offset;
1128             len = This->maps[This->modified_areas].size;
1129             end = start + len;
1130
1131             memcpy(data + start, This->resource.allocatedMemory + start, end - start);
1132             for (i = start / This->stride; i < min((end / This->stride) + 1, vertices); ++i)
1133             {
1134                 for (j = 0; j < This->stride; ++j)
1135                 {
1136                     switch(This->conversion_map[j])
1137                     {
1138                         case CONV_NONE:
1139                             /* Done already */
1140                             j += 3;
1141                             break;
1142                         case CONV_D3DCOLOR:
1143                             fixup_d3dcolor((DWORD *) (data + i * This->stride + j));
1144                             j += 3;
1145                             break;
1146
1147                         case CONV_POSITIONT:
1148                             fixup_transformed_pos((float *) (data + i * This->stride + j));
1149                             j += 15;
1150                             break;
1151
1152                         case CONV_FLOAT16_2:
1153                             ERR("Did not expect FLOAT16 conversion in unshifted conversion\n");
1154                         default:
1155                             FIXME("Unimplemented conversion %d in shifted conversion\n", This->conversion_map[j]);
1156                     }
1157                 }
1158             }
1159
1160             ENTER_GL();
1161             GL_EXTCALL(glBindBufferARB(This->buffer_type_hint, This->buffer_object));
1162             checkGLcall("glBindBufferARB");
1163             GL_EXTCALL(glBufferSubDataARB(This->buffer_type_hint, start, len, data + start));
1164             checkGLcall("glBufferSubDataARB");
1165             LEAVE_GL();
1166         }
1167     }
1168
1169     HeapFree(GetProcessHeap(), 0, data);
1170
1171 end:
1172     context_release(context);
1173 }
1174
1175 static WINED3DRESOURCETYPE STDMETHODCALLTYPE buffer_GetType(IWineD3DBuffer *iface)
1176 {
1177     return resource_get_type((IWineD3DResource *)iface);
1178 }
1179
1180 /* IWineD3DBuffer methods */
1181
1182 static DWORD buffer_sanitize_flags(struct wined3d_buffer *buffer, DWORD flags)
1183 {
1184     /* Not all flags make sense together, but Windows never returns an error. Catch the
1185      * cases that could cause issues */
1186     if(flags & WINED3DLOCK_READONLY)
1187     {
1188         if(flags & WINED3DLOCK_DISCARD)
1189         {
1190             WARN("WINED3DLOCK_READONLY combined with WINED3DLOCK_DISCARD, ignoring flags\n");
1191             return 0;
1192         }
1193         if(flags & WINED3DLOCK_NOOVERWRITE)
1194         {
1195             WARN("WINED3DLOCK_READONLY combined with WINED3DLOCK_NOOVERWRITE, ignoring flags\n");
1196             return 0;
1197         }
1198     }
1199     else if((flags & (WINED3DLOCK_DISCARD | WINED3DLOCK_NOOVERWRITE)) == (WINED3DLOCK_DISCARD | WINED3DLOCK_NOOVERWRITE))
1200     {
1201         WARN("WINED3DLOCK_DISCARD and WINED3DLOCK_NOOVERWRITE used together, ignoring\n");
1202         return 0;
1203     }
1204     else if (flags & (WINED3DLOCK_DISCARD | WINED3DLOCK_NOOVERWRITE) && !(buffer->resource.usage & WINED3DUSAGE_DYNAMIC))
1205     {
1206         WARN("DISCARD or NOOVERWRITE lock on non-dynamic buffer, ignoring\n");
1207         return 0;
1208     }
1209
1210     return flags;
1211 }
1212
1213 static GLbitfield buffer_gl_map_flags(DWORD d3d_flags)
1214 {
1215     GLbitfield ret = 0;
1216
1217     if (!(d3d_flags & WINED3DLOCK_READONLY)) ret = GL_MAP_WRITE_BIT | GL_MAP_FLUSH_EXPLICIT_BIT;
1218
1219     if (d3d_flags & (WINED3DLOCK_DISCARD | WINED3DLOCK_NOOVERWRITE))
1220     {
1221         if(d3d_flags & WINED3DLOCK_DISCARD) ret |= GL_MAP_INVALIDATE_BUFFER_BIT;
1222         ret |= GL_MAP_UNSYNCHRONIZED_BIT;
1223     }
1224     else
1225     {
1226         ret |= GL_MAP_READ_BIT;
1227     }
1228
1229     return ret;
1230 }
1231
1232 static HRESULT STDMETHODCALLTYPE buffer_Map(IWineD3DBuffer *iface, UINT offset, UINT size, BYTE **data, DWORD flags)
1233 {
1234     struct wined3d_buffer *This = (struct wined3d_buffer *)iface;
1235     LONG count;
1236     BOOL dirty = buffer_is_dirty(This);
1237
1238     TRACE("iface %p, offset %u, size %u, data %p, flags %#x\n", iface, offset, size, data, flags);
1239
1240     flags = buffer_sanitize_flags(This, flags);
1241     if (!(flags & WINED3DLOCK_READONLY))
1242     {
1243         if (!buffer_add_dirty_area(This, offset, size)) return E_OUTOFMEMORY;
1244     }
1245
1246     count = InterlockedIncrement(&This->lock_count);
1247
1248     if (This->buffer_object)
1249     {
1250         if(!(This->flags & WINED3D_BUFFER_DOUBLEBUFFER))
1251         {
1252             if(count == 1)
1253             {
1254                 IWineD3DDeviceImpl *device = This->resource.device;
1255                 struct wined3d_context *context;
1256                 const struct wined3d_gl_info *gl_info;
1257
1258                 if(This->buffer_type_hint == GL_ELEMENT_ARRAY_BUFFER_ARB)
1259                 {
1260                     IWineD3DDeviceImpl_MarkStateDirty(This->resource.device, STATE_INDEXBUFFER);
1261                 }
1262
1263                 context = context_acquire(device, NULL);
1264                 gl_info = context->gl_info;
1265                 ENTER_GL();
1266                 GL_EXTCALL(glBindBufferARB(This->buffer_type_hint, This->buffer_object));
1267
1268                 if (gl_info->supported[ARB_MAP_BUFFER_RANGE])
1269                 {
1270                     GLbitfield mapflags = buffer_gl_map_flags(flags);
1271                     This->resource.allocatedMemory = GL_EXTCALL(glMapBufferRange(This->buffer_type_hint, 0,
1272                                                                                 This->resource.size, mapflags));
1273                     checkGLcall("glMapBufferRange");
1274                 }
1275                 else
1276                 {
1277                     if(This->flags & WINED3D_BUFFER_APPLESYNC)
1278                     {
1279                         LEAVE_GL();
1280                         buffer_sync_apple(This, flags, gl_info);
1281                         ENTER_GL();
1282                     }
1283                     This->resource.allocatedMemory = GL_EXTCALL(glMapBufferARB(This->buffer_type_hint, GL_READ_WRITE_ARB));
1284                     checkGLcall("glMapBufferARB");
1285                 }
1286                 LEAVE_GL();
1287
1288                 if (((DWORD_PTR) This->resource.allocatedMemory) & (RESOURCE_ALIGNMENT - 1))
1289                 {
1290                     WARN("Pointer %p is not %u byte aligned, falling back to double buffered operation\n",
1291                         This->resource.allocatedMemory, RESOURCE_ALIGNMENT);
1292
1293                     ENTER_GL();
1294                     GL_EXTCALL(glUnmapBufferARB(This->buffer_type_hint));
1295                     checkGLcall("glUnmapBufferARB");
1296                     LEAVE_GL();
1297                     This->resource.allocatedMemory = NULL;
1298
1299                     buffer_get_sysmem(This, gl_info);
1300                     TRACE("New pointer is %p\n", This->resource.allocatedMemory);
1301                 }
1302                 context_release(context);
1303             }
1304         }
1305         else
1306         {
1307             if (dirty)
1308             {
1309                 if (This->flags & WINED3D_BUFFER_NOSYNC && !(flags & WINED3DLOCK_NOOVERWRITE))
1310                 {
1311                     This->flags &= ~WINED3D_BUFFER_NOSYNC;
1312                 }
1313             }
1314             else if(flags & WINED3DLOCK_NOOVERWRITE)
1315             {
1316                 This->flags |= WINED3D_BUFFER_NOSYNC;
1317             }
1318
1319             if (flags & WINED3DLOCK_DISCARD)
1320             {
1321                 This->flags |= WINED3D_BUFFER_DISCARD;
1322             }
1323         }
1324     }
1325
1326     *data = This->resource.allocatedMemory + offset;
1327
1328     TRACE("Returning memory at %p (base %p, offset %u)\n", *data, This->resource.allocatedMemory, offset);
1329     /* TODO: check Flags compatibility with This->currentDesc.Usage (see MSDN) */
1330
1331     return WINED3D_OK;
1332 }
1333
1334 static HRESULT STDMETHODCALLTYPE buffer_Unmap(IWineD3DBuffer *iface)
1335 {
1336     struct wined3d_buffer *This = (struct wined3d_buffer *)iface;
1337     ULONG i;
1338
1339     TRACE("(%p)\n", This);
1340
1341     /* In the case that the number of Unmap calls > the
1342      * number of Map calls, d3d returns always D3D_OK.
1343      * This is also needed to prevent Map from returning garbage on
1344      * the next call (this will happen if the lock_count is < 0). */
1345     if(This->lock_count == 0)
1346     {
1347         TRACE("Unmap called without a previous Map call!\n");
1348         return WINED3D_OK;
1349     }
1350
1351     if (InterlockedDecrement(&This->lock_count))
1352     {
1353         /* Delay loading the buffer until everything is unlocked */
1354         TRACE("Ignoring unlock\n");
1355         return WINED3D_OK;
1356     }
1357
1358     if(!(This->flags & WINED3D_BUFFER_DOUBLEBUFFER) && This->buffer_object)
1359     {
1360         IWineD3DDeviceImpl *device = This->resource.device;
1361         const struct wined3d_gl_info *gl_info;
1362         struct wined3d_context *context;
1363
1364         if(This->buffer_type_hint == GL_ELEMENT_ARRAY_BUFFER_ARB)
1365         {
1366             IWineD3DDeviceImpl_MarkStateDirty(This->resource.device, STATE_INDEXBUFFER);
1367         }
1368
1369         context = context_acquire(device, NULL);
1370         gl_info = context->gl_info;
1371         ENTER_GL();
1372         GL_EXTCALL(glBindBufferARB(This->buffer_type_hint, This->buffer_object));
1373
1374         if (gl_info->supported[ARB_MAP_BUFFER_RANGE])
1375         {
1376             for(i = 0; i < This->modified_areas; i++)
1377             {
1378                 GL_EXTCALL(glFlushMappedBufferRange(This->buffer_type_hint,
1379                                                     This->maps[i].offset,
1380                                                     This->maps[i].size));
1381                 checkGLcall("glFlushMappedBufferRange");
1382             }
1383         }
1384         else if (This->flags & WINED3D_BUFFER_FLUSH)
1385         {
1386             for(i = 0; i < This->modified_areas; i++)
1387             {
1388                 GL_EXTCALL(glFlushMappedBufferRangeAPPLE(This->buffer_type_hint,
1389                                                          This->maps[i].offset,
1390                                                          This->maps[i].size));
1391                 checkGLcall("glFlushMappedBufferRangeAPPLE");
1392             }
1393         }
1394
1395         GL_EXTCALL(glUnmapBufferARB(This->buffer_type_hint));
1396         LEAVE_GL();
1397         context_release(context);
1398
1399         This->resource.allocatedMemory = NULL;
1400         buffer_clear_dirty_areas(This);
1401     }
1402     else if (This->flags & WINED3D_BUFFER_HASDESC)
1403     {
1404         buffer_PreLoad(iface);
1405     }
1406
1407     return WINED3D_OK;
1408 }
1409
1410 static HRESULT STDMETHODCALLTYPE buffer_GetDesc(IWineD3DBuffer *iface, WINED3DBUFFER_DESC *desc)
1411 {
1412     struct wined3d_buffer *This = (struct wined3d_buffer *)iface;
1413
1414     TRACE("(%p)\n", This);
1415
1416     desc->Type = This->resource.resourceType;
1417     desc->Usage = This->resource.usage;
1418     desc->Pool = This->resource.pool;
1419     desc->Size = This->resource.size;
1420
1421     return WINED3D_OK;
1422 }
1423
1424 static const struct IWineD3DBufferVtbl wined3d_buffer_vtbl =
1425 {
1426     /* IUnknown methods */
1427     buffer_QueryInterface,
1428     buffer_AddRef,
1429     buffer_Release,
1430     /* IWineD3DBase methods */
1431     buffer_GetParent,
1432     /* IWineD3DResource methods */
1433     buffer_SetPrivateData,
1434     buffer_GetPrivateData,
1435     buffer_FreePrivateData,
1436     buffer_SetPriority,
1437     buffer_GetPriority,
1438     buffer_PreLoad,
1439     buffer_UnLoad,
1440     buffer_GetType,
1441     /* IWineD3DBuffer methods */
1442     buffer_Map,
1443     buffer_Unmap,
1444     buffer_GetDesc,
1445 };
1446
1447 HRESULT buffer_init(struct wined3d_buffer *buffer, IWineD3DDeviceImpl *device,
1448         UINT size, DWORD usage, WINED3DFORMAT format, WINED3DPOOL pool, GLenum bind_hint,
1449         const char *data, IUnknown *parent, const struct wined3d_parent_ops *parent_ops)
1450 {
1451     const struct wined3d_format_desc *format_desc = getFormatDescEntry(format, &device->adapter->gl_info);
1452     HRESULT hr;
1453     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
1454     BOOL dynamic_buffer_ok;
1455
1456     if (!size)
1457     {
1458         WARN("Size 0 requested, returning WINED3DERR_INVALIDCALL\n");
1459         return WINED3DERR_INVALIDCALL;
1460     }
1461
1462     buffer->vtbl = &wined3d_buffer_vtbl;
1463
1464     hr = resource_init((IWineD3DResource *)buffer, WINED3DRTYPE_BUFFER,
1465             device, size, usage, format_desc, pool, parent, parent_ops);
1466     if (FAILED(hr))
1467     {
1468         WARN("Failed to initialize resource, hr %#x\n", hr);
1469         return hr;
1470     }
1471     buffer->buffer_type_hint = bind_hint;
1472
1473     TRACE("size %#x, usage %#x, format %s, memory @ %p, iface @ %p.\n", buffer->resource.size, buffer->resource.usage,
1474             debug_d3dformat(buffer->resource.format_desc->format), buffer->resource.allocatedMemory, buffer);
1475
1476     /* GL_ARB_map_buffer_range is disabled for now due to numerous bugs and no gains */
1477     dynamic_buffer_ok = gl_info->supported[APPLE_FLUSH_BUFFER_RANGE];
1478
1479     /* Observations show that drawStridedSlow is faster on dynamic VBs than converting +
1480      * drawStridedFast (half-life 2 and others).
1481      *
1482      * Basically converting the vertices in the buffer is quite expensive, and observations
1483      * show that drawStridedSlow is faster than converting + uploading + drawStridedFast.
1484      * Therefore do not create a VBO for WINED3DUSAGE_DYNAMIC buffers.
1485      */
1486     if (!gl_info->supported[ARB_VERTEX_BUFFER_OBJECT])
1487     {
1488         TRACE("Not creating a vbo because GL_ARB_vertex_buffer is not supported\n");
1489     }
1490     else if(buffer->resource.pool == WINED3DPOOL_SYSTEMMEM)
1491     {
1492         TRACE("Not creating a vbo because the vertex buffer is in system memory\n");
1493     }
1494     else if(!dynamic_buffer_ok && (buffer->resource.usage & WINED3DUSAGE_DYNAMIC))
1495     {
1496         TRACE("Not creating a vbo because the buffer has dynamic usage and no GL support\n");
1497     }
1498     else
1499     {
1500         buffer->flags |= WINED3D_BUFFER_CREATEBO;
1501     }
1502
1503     if (data)
1504     {
1505         BYTE *ptr;
1506
1507         hr = IWineD3DBuffer_Map((IWineD3DBuffer *)buffer, 0, size, &ptr, 0);
1508         if (FAILED(hr))
1509         {
1510             ERR("Failed to map buffer, hr %#x\n", hr);
1511             buffer_UnLoad((IWineD3DBuffer *)buffer);
1512             resource_cleanup((IWineD3DResource *)buffer);
1513             return hr;
1514         }
1515
1516         memcpy(ptr, data, size);
1517
1518         hr = IWineD3DBuffer_Unmap((IWineD3DBuffer *)buffer);
1519         if (FAILED(hr))
1520         {
1521             ERR("Failed to unmap buffer, hr %#x\n", hr);
1522             buffer_UnLoad((IWineD3DBuffer *)buffer);
1523             resource_cleanup((IWineD3DResource *)buffer);
1524             return hr;
1525         }
1526     }
1527
1528     buffer->maps = HeapAlloc(GetProcessHeap(), 0, sizeof(*buffer->maps));
1529     if (!buffer->maps)
1530     {
1531         ERR("Out of memory\n");
1532         buffer_UnLoad((IWineD3DBuffer *)buffer);
1533         resource_cleanup((IWineD3DResource *)buffer);
1534         return E_OUTOFMEMORY;
1535     }
1536     buffer->maps_size = 1;
1537
1538     return WINED3D_OK;
1539 }