wined3d: Set the nvidia_texture_shader.c GLINFO_LOCATION to *gl_info.
[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 GLINFO_LOCATION (*gl_info)
33
34 #define VB_MAXDECLCHANGES     100     /* After that number of decl changes we stop converting */
35 #define VB_RESETDECLCHANGE    1000    /* Reset the decl changecount after that number of draws */
36 #define VB_MAXFULLCONVERSIONS 5       /* Number of full conversions before we stop converting */
37 #define VB_RESETFULLCONVS     20      /* Reset full conversion counts after that number of draws */
38
39 static inline BOOL buffer_add_dirty_area(struct wined3d_buffer *This, UINT offset, UINT size)
40 {
41     if (!This->buffer_object) return TRUE;
42
43     if (This->maps_size <= This->modified_areas)
44     {
45         void *new = HeapReAlloc(GetProcessHeap(), 0, This->maps,
46                                 This->maps_size * 2 * sizeof(*This->maps));
47         if (!new)
48         {
49             ERR("Out of memory\n");
50             return FALSE;
51         }
52         else
53         {
54             This->maps = new;
55             This->maps_size *= 2;
56         }
57     }
58
59     if(offset > This->resource.size || offset + size > This->resource.size)
60     {
61         WARN("Invalid range dirtified, marking entire buffer dirty\n");
62         offset = 0;
63         size = This->resource.size;
64     }
65     else if(!offset && !size)
66     {
67         size = This->resource.size;
68     }
69
70     This->maps[This->modified_areas].offset = offset;
71     This->maps[This->modified_areas].size = size;
72     This->modified_areas++;
73     return TRUE;
74 }
75
76 static inline void buffer_clear_dirty_areas(struct wined3d_buffer *This)
77 {
78     This->modified_areas = 0;
79 }
80
81 static inline BOOL buffer_is_dirty(struct wined3d_buffer *This)
82 {
83     return This->modified_areas != 0;
84 }
85
86 static inline BOOL buffer_is_fully_dirty(struct wined3d_buffer *This)
87 {
88     unsigned int i;
89
90     for(i = 0; i < This->modified_areas; i++)
91     {
92         if(This->maps[i].offset == 0 && This->maps[i].size == This->resource.size)
93         {
94             return TRUE;
95         }
96     }
97     return FALSE;
98 }
99
100 /* Context activation is done by the caller */
101 static void delete_gl_buffer(struct wined3d_buffer *This, const struct wined3d_gl_info *gl_info)
102 {
103     if(!This->buffer_object) return;
104
105     ENTER_GL();
106     GL_EXTCALL(glDeleteBuffersARB(1, &This->buffer_object));
107     checkGLcall("glDeleteBuffersARB");
108     LEAVE_GL();
109     This->buffer_object = 0;
110
111     if(This->query)
112     {
113         wined3d_event_query_destroy(This->query);
114         This->query = NULL;
115     }
116     This->flags &= ~WINED3D_BUFFER_APPLESYNC;
117 }
118
119 /* Context activation is done by the caller. */
120 static void buffer_create_buffer_object(struct wined3d_buffer *This, const struct wined3d_gl_info *gl_info)
121 {
122     GLenum error, gl_usage;
123
124     TRACE("Creating an OpenGL vertex buffer object for IWineD3DVertexBuffer %p Usage(%s)\n",
125             This, debug_d3dusage(This->resource.usage));
126
127     ENTER_GL();
128
129     /* Make sure that the gl error is cleared. Do not use checkGLcall
130     * here because checkGLcall just prints a fixme and continues. However,
131     * if an error during VBO creation occurs we can fall back to non-vbo operation
132     * with full functionality(but performance loss)
133     */
134     while (glGetError() != GL_NO_ERROR);
135
136     /* Basically the FVF parameter passed to CreateVertexBuffer is no good
137      * It is the FVF set with IWineD3DDevice::SetFVF or the Vertex Declaration set with
138      * IWineD3DDevice::SetVertexDeclaration that decides how the vertices in the buffer
139      * look like. This means that on each DrawPrimitive call the vertex buffer has to be verified
140      * to check if the rhw and color values are in the correct format.
141      */
142
143     GL_EXTCALL(glGenBuffersARB(1, &This->buffer_object));
144     error = glGetError();
145     if (!This->buffer_object || error != GL_NO_ERROR)
146     {
147         ERR("Failed to create a VBO with error %s (%#x)\n", debug_glerror(error), error);
148         LEAVE_GL();
149         goto fail;
150     }
151
152     if(This->buffer_type_hint == GL_ELEMENT_ARRAY_BUFFER_ARB)
153     {
154         IWineD3DDeviceImpl_MarkStateDirty(This->resource.device, STATE_INDEXBUFFER);
155     }
156     GL_EXTCALL(glBindBufferARB(This->buffer_type_hint, This->buffer_object));
157     error = glGetError();
158     if (error != GL_NO_ERROR)
159     {
160         ERR("Failed to bind the VBO with error %s (%#x)\n", debug_glerror(error), error);
161         LEAVE_GL();
162         goto fail;
163     }
164
165     /* Don't use static, because dx apps tend to update the buffer
166      * quite often even if they specify 0 usage.
167      */
168     if(This->resource.usage & WINED3DUSAGE_DYNAMIC)
169     {
170         TRACE("Gl usage = GL_STREAM_DRAW_ARB\n");
171         gl_usage = GL_STREAM_DRAW_ARB;
172
173         if(gl_info->supported[APPLE_FLUSH_BUFFER_RANGE])
174         {
175             GL_EXTCALL(glBufferParameteriAPPLE(This->buffer_type_hint, GL_BUFFER_FLUSHING_UNMAP_APPLE, GL_FALSE));
176             checkGLcall("glBufferParameteriAPPLE(This->buffer_type_hint, GL_BUFFER_FLUSHING_UNMAP_APPLE, GL_FALSE)");
177             This->flags |= WINED3D_BUFFER_FLUSH;
178
179             GL_EXTCALL(glBufferParameteriAPPLE(This->buffer_type_hint, GL_BUFFER_SERIALIZED_MODIFY_APPLE, GL_FALSE));
180             checkGLcall("glBufferParameteriAPPLE(This->buffer_type_hint, GL_BUFFER_SERIALIZED_MODIFY_APPLE, GL_FALSE)");
181             This->flags |= WINED3D_BUFFER_APPLESYNC;
182         }
183         /* No setup is needed here for GL_ARB_map_buffer_range */
184     }
185     else
186     {
187         TRACE("Gl usage = GL_DYNAMIC_DRAW_ARB\n");
188         gl_usage = GL_DYNAMIC_DRAW_ARB;
189     }
190
191     /* Reserve memory for the buffer. The amount of data won't change
192      * so we are safe with calling glBufferData once and
193      * calling glBufferSubData on updates. Upload the actual data in case
194      * we're not double buffering, so we can release the heap mem afterwards
195      */
196     GL_EXTCALL(glBufferDataARB(This->buffer_type_hint, This->resource.size, This->resource.allocatedMemory, gl_usage));
197     error = glGetError();
198     LEAVE_GL();
199     if (error != GL_NO_ERROR)
200     {
201         ERR("glBufferDataARB failed with error %s (%#x)\n", debug_glerror(error), error);
202         goto fail;
203     }
204
205     This->buffer_object_size = This->resource.size;
206     This->buffer_object_usage = gl_usage;
207
208     if(This->flags & WINED3D_BUFFER_DOUBLEBUFFER)
209     {
210         if(!buffer_add_dirty_area(This, 0, 0))
211         {
212             ERR("buffer_add_dirty_area failed, this is not expected\n");
213             goto fail;
214         }
215     }
216     else
217     {
218         HeapFree(GetProcessHeap(), 0, This->resource.heapMemory);
219         This->resource.allocatedMemory = NULL;
220         This->resource.heapMemory = NULL;
221     }
222
223     return;
224
225 fail:
226     /* Clean up all vbo init, but continue because we can work without a vbo :-) */
227     ERR("Failed to create a vertex buffer object. Continuing, but performance issues may occur\n");
228     delete_gl_buffer(This, gl_info);
229     buffer_clear_dirty_areas(This);
230 }
231
232 static BOOL buffer_process_converted_attribute(struct wined3d_buffer *This,
233         const enum wined3d_buffer_conversion_type conversion_type,
234         const struct wined3d_stream_info_element *attrib, DWORD *stride_this_run)
235 {
236     DWORD attrib_size;
237     BOOL ret = FALSE;
238     unsigned int i;
239     DWORD offset = This->resource.device->stateBlock->streamOffset[attrib->stream_idx];
240     DWORD_PTR data;
241
242     /* Check for some valid situations which cause us pain. One is if the buffer is used for
243      * constant attributes(stride = 0), the other one is if the buffer is used on two streams
244      * with different strides. In the 2nd case we might have to drop conversion entirely,
245      * it is possible that the same bytes are once read as FLOAT2 and once as UBYTE4N.
246      */
247     if (!attrib->stride)
248     {
249         FIXME("%s used with stride 0, let's hope we get the vertex stride from somewhere else\n",
250                 debug_d3dformat(attrib->format_desc->format));
251     }
252     else if(attrib->stride != *stride_this_run && *stride_this_run)
253     {
254         FIXME("Got two concurrent strides, %d and %d\n", attrib->stride, *stride_this_run);
255     }
256     else
257     {
258         *stride_this_run = attrib->stride;
259         if (This->stride != *stride_this_run)
260         {
261             /* We rely that this happens only on the first converted attribute that is found,
262              * if at all. See above check
263              */
264             TRACE("Reconverting because converted attributes occur, and the stride changed\n");
265             This->stride = *stride_this_run;
266             HeapFree(GetProcessHeap(), HEAP_ZERO_MEMORY, This->conversion_map);
267             This->conversion_map = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY,
268                     sizeof(*This->conversion_map) * This->stride);
269             ret = TRUE;
270         }
271     }
272
273     data = (((DWORD_PTR)attrib->data) + offset) % This->stride;
274     attrib_size = attrib->format_desc->component_count * attrib->format_desc->component_size;
275     for (i = 0; i < attrib_size; ++i)
276     {
277         DWORD_PTR idx = (data + i) % This->stride;
278         if (This->conversion_map[idx] != conversion_type)
279         {
280             TRACE("Byte %ld in vertex changed\n", idx);
281             TRACE("It was type %d, is %d now\n", This->conversion_map[idx], conversion_type);
282             ret = TRUE;
283             This->conversion_map[idx] = conversion_type;
284         }
285     }
286
287     return ret;
288 }
289
290 static BOOL buffer_check_attribute(struct wined3d_buffer *This, const struct wined3d_stream_info *si,
291         UINT attrib_idx, const BOOL check_d3dcolor, const BOOL is_ffp_position, const BOOL is_ffp_color,
292         DWORD *stride_this_run, BOOL *float16_used)
293 {
294     const struct wined3d_stream_info_element *attrib = &si->elements[attrib_idx];
295     IWineD3DDeviceImpl *device = This->resource.device;
296     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
297     BOOL ret = FALSE;
298     WINED3DFORMAT format;
299
300     /* Ignore attributes that do not have our vbo. After that check we can be sure that the attribute is
301      * there, on nonexistent attribs the vbo is 0.
302      */
303     if (!(si->use_map & (1 << attrib_idx))
304             || attrib->buffer_object != This->buffer_object)
305         return FALSE;
306
307     format = attrib->format_desc->format;
308     /* Look for newly appeared conversion */
309     if (!gl_info->supported[ARB_HALF_FLOAT_VERTEX]
310             && (format == WINED3DFMT_R16G16_FLOAT || format == WINED3DFMT_R16G16B16A16_FLOAT))
311     {
312         ret = buffer_process_converted_attribute(This, CONV_FLOAT16_2, attrib, stride_this_run);
313
314         if (is_ffp_position) FIXME("Test FLOAT16 fixed function processing positions\n");
315         else if (is_ffp_color) FIXME("test FLOAT16 fixed function processing colors\n");
316         *float16_used = TRUE;
317     }
318     else if (check_d3dcolor && format == WINED3DFMT_B8G8R8A8_UNORM)
319     {
320         ret = buffer_process_converted_attribute(This, CONV_D3DCOLOR, attrib, stride_this_run);
321
322         if (!is_ffp_color) FIXME("Test for non-color fixed function WINED3DFMT_B8G8R8A8_UNORM format\n");
323     }
324     else if (is_ffp_position && format == WINED3DFMT_R32G32B32A32_FLOAT)
325     {
326         ret = buffer_process_converted_attribute(This, CONV_POSITIONT, attrib, stride_this_run);
327     }
328     else if (This->conversion_map)
329     {
330         ret = buffer_process_converted_attribute(This, CONV_NONE, attrib, stride_this_run);
331     }
332
333     return ret;
334 }
335
336 static UINT *find_conversion_shift(struct wined3d_buffer *This,
337         const struct wined3d_stream_info *strided, UINT stride)
338 {
339     UINT *ret, i, j, shift, orig_type_size;
340
341     if (!stride)
342     {
343         TRACE("No shift\n");
344         return NULL;
345     }
346
347     This->conversion_stride = stride;
348     ret = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(DWORD) * stride);
349     for (i = 0; i < MAX_ATTRIBS; ++i)
350     {
351         WINED3DFORMAT format;
352
353         if (!(strided->use_map & (1 << i)) || strided->elements[i].buffer_object != This->buffer_object) continue;
354
355         format = strided->elements[i].format_desc->format;
356         if (format == WINED3DFMT_R16G16_FLOAT)
357         {
358             shift = 4;
359         }
360         else if (format == WINED3DFMT_R16G16B16A16_FLOAT)
361         {
362             shift = 8;
363             /* Pre-shift the last 4 bytes in the FLOAT16_4 by 4 bytes - this makes FLOAT16_2 and FLOAT16_4 conversions
364              * compatible
365              */
366             for (j = 4; j < 8; ++j)
367             {
368                 ret[(DWORD_PTR)strided->elements[i].data + j] += 4;
369             }
370         }
371         else
372         {
373             shift = 0;
374         }
375         This->conversion_stride += shift;
376
377         if (shift)
378         {
379             orig_type_size = strided->elements[i].format_desc->component_count
380                     * strided->elements[i].format_desc->component_size;
381             for (j = (DWORD_PTR)strided->elements[i].data + orig_type_size; j < stride; ++j)
382             {
383                 ret[j] += shift;
384             }
385         }
386     }
387
388     if (TRACE_ON(d3d))
389     {
390         TRACE("Dumping conversion shift:\n");
391         for (i = 0; i < stride; ++i)
392         {
393             TRACE("[%d]", ret[i]);
394         }
395         TRACE("\n");
396     }
397
398     return ret;
399 }
400
401 static BOOL buffer_find_decl(struct wined3d_buffer *This)
402 {
403     IWineD3DDeviceImpl *device = This->resource.device;
404     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
405     const struct wined3d_stream_info *si = &device->strided_streams;
406     UINT stride_this_run = 0;
407     BOOL float16_used = FALSE;
408     BOOL ret = FALSE;
409     unsigned int i;
410
411     /* In d3d7 the vertex buffer declaration NEVER changes because it is stored in the d3d7 vertex buffer.
412      * Once we have our declaration there is no need to look it up again. Index buffers also never need
413      * conversion, so once the (empty) conversion structure is created don't bother checking again
414      */
415     if (This->flags & WINED3D_BUFFER_HASDESC)
416     {
417         if(This->resource.usage & WINED3DUSAGE_STATICDECL) return FALSE;
418     }
419
420     TRACE("Finding vertex buffer conversion information\n");
421     /* Certain declaration types need some fixups before we can pass them to
422      * opengl. This means D3DCOLOR attributes with fixed function vertex
423      * processing, FLOAT4 POSITIONT with fixed function, and FLOAT16 if
424      * GL_ARB_half_float_vertex is not supported.
425      *
426      * Note for d3d8 and d3d9:
427      * The vertex buffer FVF doesn't help with finding them, we have to use
428      * the decoded vertex declaration and pick the things that concern the
429      * current buffer. A problem with this is that this can change between
430      * draws, so we have to validate the information and reprocess the buffer
431      * if it changes, and avoid false positives for performance reasons.
432      * WineD3D doesn't even know the vertex buffer any more, it is managed
433      * by the client libraries and passed to SetStreamSource and ProcessVertices
434      * as needed.
435      *
436      * We have to distinguish between vertex shaders and fixed function to
437      * pick the way we access the strided vertex information.
438      *
439      * This code sets up a per-byte array with the size of the detected
440      * stride of the arrays in the buffer. For each byte we have a field
441      * that marks the conversion needed on this byte. For example, the
442      * following declaration with fixed function vertex processing:
443      *
444      *      POSITIONT, FLOAT4
445      *      NORMAL, FLOAT3
446      *      DIFFUSE, FLOAT16_4
447      *      SPECULAR, D3DCOLOR
448      *
449      * Will result in
450      * {                 POSITIONT                    }{             NORMAL                }{    DIFFUSE          }{SPECULAR }
451      * [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]
452      *
453      * Where in this example map P means 4 component position conversion, 0
454      * means no conversion, F means FLOAT16_2 conversion and C means D3DCOLOR
455      * conversion (red / blue swizzle).
456      *
457      * If we're doing conversion and the stride changes we have to reconvert
458      * the whole buffer. Note that we do not mind if the semantic changes,
459      * we only care for the conversion type. So if the NORMAL is replaced
460      * with a TEXCOORD, nothing has to be done, or if the DIFFUSE is replaced
461      * with a D3DCOLOR BLENDWEIGHT we can happily dismiss the change. Some
462      * conversion types depend on the semantic as well, for example a FLOAT4
463      * texcoord needs no conversion while a FLOAT4 positiont needs one
464      */
465     if (use_vs(device->stateBlock))
466     {
467         TRACE("vshader\n");
468         /* If the current vertex declaration is marked for no half float conversion don't bother to
469          * analyse the strided streams in depth, just set them up for no conversion. Return decl changed
470          * if we used conversion before
471          */
472         if (!((IWineD3DVertexDeclarationImpl *) device->stateBlock->vertexDecl)->half_float_conv_needed)
473         {
474             if (This->conversion_map)
475             {
476                 TRACE("Now using shaders without conversion, but conversion used before\n");
477                 HeapFree(GetProcessHeap(), 0, This->conversion_map);
478                 HeapFree(GetProcessHeap(), 0, This->conversion_shift);
479                 This->conversion_map = NULL;
480                 This->stride = 0;
481                 This->conversion_shift = NULL;
482                 This->conversion_stride = 0;
483                 return TRUE;
484             }
485             else
486             {
487                 return FALSE;
488             }
489         }
490         for (i = 0; i < MAX_ATTRIBS; ++i)
491         {
492             ret = buffer_check_attribute(This, si, i, FALSE, FALSE, FALSE, &stride_this_run, &float16_used) || ret;
493         }
494
495         /* Recalculate the conversion shift map if the declaration has changed,
496          * and we're using float16 conversion or used it on the last run
497          */
498         if (ret && (float16_used || This->conversion_map))
499         {
500             HeapFree(GetProcessHeap(), 0, This->conversion_shift);
501             This->conversion_shift = find_conversion_shift(This, si, This->stride);
502         }
503     }
504     else
505     {
506         /* Fixed function is a bit trickier. We have to take care for D3DCOLOR types, FLOAT4 positions and of course
507          * FLOAT16s if not supported. Also, we can't iterate over the array, so use macros to generate code for all
508          * the attributes that our current fixed function pipeline implementation cares for.
509          */
510         BOOL support_d3dcolor = gl_info->supported[ARB_VERTEX_ARRAY_BGRA];
511         ret = buffer_check_attribute(This, si, WINED3D_FFP_POSITION,
512                 TRUE, TRUE,  FALSE, &stride_this_run, &float16_used) || ret;
513         ret = buffer_check_attribute(This, si, WINED3D_FFP_NORMAL,
514                 TRUE, FALSE, FALSE, &stride_this_run, &float16_used) || ret;
515         ret = buffer_check_attribute(This, si, WINED3D_FFP_DIFFUSE,
516                 !support_d3dcolor, FALSE, TRUE,  &stride_this_run, &float16_used) || ret;
517         ret = buffer_check_attribute(This, si, WINED3D_FFP_SPECULAR,
518                 !support_d3dcolor, FALSE, TRUE,  &stride_this_run, &float16_used) || ret;
519         ret = buffer_check_attribute(This, si, WINED3D_FFP_TEXCOORD0,
520                 TRUE, FALSE, FALSE, &stride_this_run, &float16_used) || ret;
521         ret = buffer_check_attribute(This, si, WINED3D_FFP_TEXCOORD1,
522                 TRUE, FALSE, FALSE, &stride_this_run, &float16_used) || ret;
523         ret = buffer_check_attribute(This, si, WINED3D_FFP_TEXCOORD2,
524                 TRUE, FALSE, FALSE, &stride_this_run, &float16_used) || ret;
525         ret = buffer_check_attribute(This, si, WINED3D_FFP_TEXCOORD3,
526                 TRUE, FALSE, FALSE, &stride_this_run, &float16_used) || ret;
527         ret = buffer_check_attribute(This, si, WINED3D_FFP_TEXCOORD4,
528                 TRUE, FALSE, FALSE, &stride_this_run, &float16_used) || ret;
529         ret = buffer_check_attribute(This, si, WINED3D_FFP_TEXCOORD5,
530                 TRUE, FALSE, FALSE, &stride_this_run, &float16_used) || ret;
531         ret = buffer_check_attribute(This, si, WINED3D_FFP_TEXCOORD6,
532                 TRUE, FALSE, FALSE, &stride_this_run, &float16_used) || ret;
533         ret = buffer_check_attribute(This, si, WINED3D_FFP_TEXCOORD7,
534                 TRUE, FALSE, FALSE, &stride_this_run, &float16_used) || ret;
535
536         if (float16_used) FIXME("Float16 conversion used with fixed function vertex processing\n");
537     }
538
539     if (stride_this_run == 0 && This->conversion_map)
540     {
541         /* Sanity test */
542         if (!ret) ERR("no converted attributes found, old conversion map exists, and no declaration change?\n");
543         HeapFree(GetProcessHeap(), 0, This->conversion_map);
544         This->conversion_map = NULL;
545         This->stride = 0;
546     }
547
548     if (ret) TRACE("Conversion information changed\n");
549
550     return ret;
551 }
552
553 /* Context activation is done by the caller. */
554 static void buffer_check_buffer_object_size(struct wined3d_buffer *This, const struct wined3d_gl_info *gl_info)
555 {
556     UINT size = This->conversion_stride ?
557             This->conversion_stride * (This->resource.size / This->stride) : This->resource.size;
558     if (This->buffer_object_size != size)
559     {
560         TRACE("Old size %u, creating new size %u\n", This->buffer_object_size, size);
561
562         if(This->buffer_type_hint == GL_ELEMENT_ARRAY_BUFFER_ARB)
563         {
564             IWineD3DDeviceImpl_MarkStateDirty(This->resource.device, STATE_INDEXBUFFER);
565         }
566
567         /* Rescue the data before resizing the buffer object if we do not have our backup copy */
568         if(!(This->flags & WINED3D_BUFFER_DOUBLEBUFFER))
569         {
570             buffer_get_sysmem(This, gl_info);
571         }
572
573         ENTER_GL();
574         GL_EXTCALL(glBindBufferARB(This->buffer_type_hint, This->buffer_object));
575         checkGLcall("glBindBufferARB");
576         GL_EXTCALL(glBufferDataARB(This->buffer_type_hint, size, NULL, This->buffer_object_usage));
577         This->buffer_object_size = size;
578         checkGLcall("glBufferDataARB");
579         LEAVE_GL();
580     }
581 }
582
583 static inline void fixup_d3dcolor(DWORD *dst_color)
584 {
585     DWORD src_color = *dst_color;
586
587     /* Color conversion like in drawStridedSlow. watch out for little endianity
588      * If we want that stuff to work on big endian machines too we have to consider more things
589      *
590      * 0xff000000: Alpha mask
591      * 0x00ff0000: Blue mask
592      * 0x0000ff00: Green mask
593      * 0x000000ff: Red mask
594      */
595     *dst_color = 0;
596     *dst_color |= (src_color & 0xff00ff00);         /* Alpha Green */
597     *dst_color |= (src_color & 0x00ff0000) >> 16;   /* Red */
598     *dst_color |= (src_color & 0x000000ff) << 16;   /* Blue */
599 }
600
601 static inline void fixup_transformed_pos(float *p)
602 {
603     /* rhw conversion like in position_float4(). */
604     if (p[3] != 1.0f && p[3] != 0.0f)
605     {
606         float w = 1.0f / p[3];
607         p[0] *= w;
608         p[1] *= w;
609         p[2] *= w;
610         p[3] = w;
611     }
612 }
613
614 /* Context activation is done by the caller. */
615 const BYTE *buffer_get_memory(IWineD3DBuffer *iface, const struct wined3d_gl_info *gl_info, GLuint *buffer_object)
616 {
617     struct wined3d_buffer *This = (struct wined3d_buffer *)iface;
618
619     *buffer_object = This->buffer_object;
620     if (!This->buffer_object)
621     {
622         if (This->flags & WINED3D_BUFFER_CREATEBO)
623         {
624             buffer_create_buffer_object(This, gl_info);
625             This->flags &= ~WINED3D_BUFFER_CREATEBO;
626             if (This->buffer_object)
627             {
628                 *buffer_object = This->buffer_object;
629                 return NULL;
630             }
631         }
632         return This->resource.allocatedMemory;
633     }
634     else
635     {
636         return NULL;
637     }
638 }
639
640 /* IUnknown methods */
641
642 static HRESULT STDMETHODCALLTYPE buffer_QueryInterface(IWineD3DBuffer *iface,
643         REFIID riid, void **object)
644 {
645     TRACE("iface %p, riid %s, object %p\n", iface, debugstr_guid(riid), object);
646
647     if (IsEqualGUID(riid, &IID_IWineD3DBuffer)
648             || IsEqualGUID(riid, &IID_IWineD3DResource)
649             || IsEqualGUID(riid, &IID_IWineD3DBase)
650             || IsEqualGUID(riid, &IID_IUnknown))
651     {
652         IUnknown_AddRef(iface);
653         *object = iface;
654         return S_OK;
655     }
656
657     WARN("%s not implemented, returning E_NOINTERFACE\n", debugstr_guid(riid));
658
659     *object = NULL;
660
661     return E_NOINTERFACE;
662 }
663
664 static ULONG STDMETHODCALLTYPE buffer_AddRef(IWineD3DBuffer *iface)
665 {
666     struct wined3d_buffer *This = (struct wined3d_buffer *)iface;
667     ULONG refcount = InterlockedIncrement(&This->resource.ref);
668
669     TRACE("%p increasing refcount to %u\n", This, refcount);
670
671     return refcount;
672 }
673
674 /* Context activation is done by the caller. */
675 BYTE *buffer_get_sysmem(struct wined3d_buffer *This, const struct wined3d_gl_info *gl_info)
676 {
677     /* AllocatedMemory exists if the buffer is double buffered or has no buffer object at all */
678     if(This->resource.allocatedMemory) return This->resource.allocatedMemory;
679
680     This->resource.heapMemory = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, This->resource.size + RESOURCE_ALIGNMENT);
681     This->resource.allocatedMemory = (BYTE *)(((ULONG_PTR)This->resource.heapMemory + (RESOURCE_ALIGNMENT - 1)) & ~(RESOURCE_ALIGNMENT - 1));
682     ENTER_GL();
683     GL_EXTCALL(glBindBufferARB(This->buffer_type_hint, This->buffer_object));
684     GL_EXTCALL(glGetBufferSubDataARB(This->buffer_type_hint, 0, This->resource.size, This->resource.allocatedMemory));
685     LEAVE_GL();
686     This->flags |= WINED3D_BUFFER_DOUBLEBUFFER;
687
688     return This->resource.allocatedMemory;
689 }
690
691 static void STDMETHODCALLTYPE buffer_UnLoad(IWineD3DBuffer *iface)
692 {
693     struct wined3d_buffer *This = (struct wined3d_buffer *)iface;
694
695     TRACE("iface %p\n", iface);
696
697     if (This->buffer_object)
698     {
699         IWineD3DDeviceImpl *device = This->resource.device;
700         struct wined3d_context *context;
701
702         context = context_acquire(device, NULL);
703
704         /* Download the buffer, but don't permanently enable double buffering */
705         if(!(This->flags & WINED3D_BUFFER_DOUBLEBUFFER))
706         {
707             buffer_get_sysmem(This, context->gl_info);
708             This->flags &= ~WINED3D_BUFFER_DOUBLEBUFFER;
709         }
710
711         delete_gl_buffer(This, context->gl_info);
712         This->flags |= WINED3D_BUFFER_CREATEBO; /* Recreate the buffer object next load */
713         buffer_clear_dirty_areas(This);
714
715         context_release(context);
716
717         HeapFree(GetProcessHeap(), 0, This->conversion_shift);
718         This->conversion_shift = NULL;
719         HeapFree(GetProcessHeap(), 0, This->conversion_map);
720         This->conversion_map = NULL;
721         This->stride = 0;
722         This->conversion_stride = 0;
723         This->flags &= ~WINED3D_BUFFER_HASDESC;
724     }
725 }
726
727 static ULONG STDMETHODCALLTYPE buffer_Release(IWineD3DBuffer *iface)
728 {
729     struct wined3d_buffer *This = (struct wined3d_buffer *)iface;
730     ULONG refcount = InterlockedDecrement(&This->resource.ref);
731
732     TRACE("%p decreasing refcount to %u\n", This, refcount);
733
734     if (!refcount)
735     {
736         buffer_UnLoad(iface);
737         resource_cleanup((IWineD3DResource *)iface);
738         This->resource.parent_ops->wined3d_object_destroyed(This->resource.parent);
739         HeapFree(GetProcessHeap(), 0, This->maps);
740         HeapFree(GetProcessHeap(), 0, This);
741     }
742
743     return refcount;
744 }
745
746 /* IWineD3DBase methods */
747
748 static HRESULT STDMETHODCALLTYPE buffer_GetParent(IWineD3DBuffer *iface, IUnknown **parent)
749 {
750     return resource_get_parent((IWineD3DResource *)iface, parent);
751 }
752
753 /* IWineD3DResource methods */
754
755 static HRESULT STDMETHODCALLTYPE buffer_SetPrivateData(IWineD3DBuffer *iface,
756         REFGUID guid, const void *data, DWORD data_size, DWORD flags)
757 {
758     return resource_set_private_data((IWineD3DResource *)iface, guid, data, data_size, flags);
759 }
760
761 static HRESULT STDMETHODCALLTYPE buffer_GetPrivateData(IWineD3DBuffer *iface,
762         REFGUID guid, void *data, DWORD *data_size)
763 {
764     return resource_get_private_data((IWineD3DResource *)iface, guid, data, data_size);
765 }
766
767 static HRESULT STDMETHODCALLTYPE buffer_FreePrivateData(IWineD3DBuffer *iface, REFGUID guid)
768 {
769     return resource_free_private_data((IWineD3DResource *)iface, guid);
770 }
771
772 static DWORD STDMETHODCALLTYPE buffer_SetPriority(IWineD3DBuffer *iface, DWORD priority)
773 {
774     return resource_set_priority((IWineD3DResource *)iface, priority);
775 }
776
777 static DWORD STDMETHODCALLTYPE buffer_GetPriority(IWineD3DBuffer *iface)
778 {
779     return resource_get_priority((IWineD3DResource *)iface);
780 }
781
782 /* The caller provides a context and binds the buffer */
783 static void buffer_sync_apple(struct wined3d_buffer *This, DWORD flags, const struct wined3d_gl_info *gl_info)
784 {
785     enum wined3d_event_query_result ret;
786
787     /* No fencing needs to be done if the app promises not to overwrite
788      * existing data */
789     if(flags & WINED3DLOCK_NOOVERWRITE) return;
790     if(flags & WINED3DLOCK_DISCARD)
791     {
792         ENTER_GL();
793         GL_EXTCALL(glBufferDataARB(This->buffer_type_hint, This->resource.size, NULL, This->buffer_object_usage));
794         checkGLcall("glBufferDataARB\n");
795         LEAVE_GL();
796         return;
797     }
798
799     if(!This->query)
800     {
801         TRACE("Creating event query for buffer %p\n", This);
802
803         if (!wined3d_event_query_supported(gl_info))
804         {
805             FIXME("Event queries not supported, dropping async buffer locks.\n");
806             goto drop_query;
807         }
808
809         This->query = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*This->query));
810         if (!This->query)
811         {
812             ERR("Failed to allocate event query memory, dropping async buffer locks.\n");
813             goto drop_query;
814         }
815
816         /* Since we don't know about old draws a glFinish is needed once */
817         wglFinish();
818         return;
819     }
820     TRACE("Synchronizing buffer %p\n", This);
821     ret = wined3d_event_query_finish(This->query, This->resource.device);
822     switch(ret)
823     {
824         case WINED3D_EVENT_QUERY_NOT_STARTED:
825         case WINED3D_EVENT_QUERY_OK:
826             /* All done */
827             return;
828
829         case WINED3D_EVENT_QUERY_WRONG_THREAD:
830             WARN("Cannot synchronize buffer lock due to a thread conflict\n");
831             goto drop_query;
832
833         default:
834             ERR("wined3d_event_query_finish returned %u, dropping async buffer locks\n", ret);
835             goto drop_query;
836     }
837
838 drop_query:
839     if(This->query)
840     {
841         wined3d_event_query_destroy(This->query);
842         This->query = NULL;
843     }
844
845     wglFinish();
846     ENTER_GL();
847     GL_EXTCALL(glBufferParameteriAPPLE(This->buffer_type_hint, GL_BUFFER_SERIALIZED_MODIFY_APPLE, GL_TRUE));
848     checkGLcall("glBufferParameteriAPPLE(This->buffer_type_hint, GL_BUFFER_SERIALIZED_MODIFY_APPLE, GL_TRUE)");
849     LEAVE_GL();
850     This->flags &= ~WINED3D_BUFFER_APPLESYNC;
851 }
852
853 /* The caller provides a GL context */
854 static void buffer_direct_upload(struct wined3d_buffer *This, const struct wined3d_gl_info *gl_info, DWORD flags)
855 {
856         BYTE *map;
857         UINT start = 0, len = 0;
858
859         ENTER_GL();
860         GL_EXTCALL(glBindBufferARB(This->buffer_type_hint, This->buffer_object));
861         checkGLcall("glBindBufferARB");
862         if (gl_info->supported[ARB_MAP_BUFFER_RANGE])
863         {
864             GLbitfield mapflags;
865             mapflags = GL_MAP_WRITE_BIT | GL_MAP_FLUSH_EXPLICIT_BIT;
866             if (flags & WINED3D_BUFFER_DISCARD)
867             {
868                 mapflags |= GL_MAP_UNSYNCHRONIZED_BIT | GL_MAP_INVALIDATE_BUFFER_BIT;
869             }
870             else if (flags & WINED3D_BUFFER_NOSYNC)
871             {
872                 mapflags |= GL_MAP_UNSYNCHRONIZED_BIT;
873             }
874             map = GL_EXTCALL(glMapBufferRange(This->buffer_type_hint, 0,
875                                               This->resource.size, mapflags));
876             checkGLcall("glMapBufferRange");
877         }
878         else
879         {
880             if (This->flags & WINED3D_BUFFER_APPLESYNC)
881             {
882                 DWORD syncflags = 0;
883                 if (flags & WINED3D_BUFFER_DISCARD) syncflags |= WINED3DLOCK_DISCARD;
884                 if (flags & WINED3D_BUFFER_NOSYNC) syncflags |= WINED3DLOCK_NOOVERWRITE;
885                 LEAVE_GL();
886                 buffer_sync_apple(This, syncflags, gl_info);
887                 ENTER_GL();
888             }
889             map = GL_EXTCALL(glMapBufferARB(This->buffer_type_hint, GL_WRITE_ONLY_ARB));
890             checkGLcall("glMapBufferARB");
891         }
892         if (!map)
893         {
894             LEAVE_GL();
895             ERR("Failed to map opengl buffer\n");
896             return;
897         }
898
899         while(This->modified_areas)
900         {
901             This->modified_areas--;
902             start = This->maps[This->modified_areas].offset;
903             len = This->maps[This->modified_areas].size;
904
905             memcpy(map + start, This->resource.allocatedMemory + start, len);
906
907             if (gl_info->supported[ARB_MAP_BUFFER_RANGE])
908             {
909                 GL_EXTCALL(glFlushMappedBufferRange(This->buffer_type_hint, start, len));
910                 checkGLcall("glFlushMappedBufferRange");
911             }
912             else if (This->flags & WINED3D_BUFFER_FLUSH)
913             {
914                 GL_EXTCALL(glFlushMappedBufferRangeAPPLE(This->buffer_type_hint, start, len));
915                 checkGLcall("glFlushMappedBufferRangeAPPLE");
916             }
917         }
918         GL_EXTCALL(glUnmapBufferARB(This->buffer_type_hint));
919         checkGLcall("glUnmapBufferARB");
920         LEAVE_GL();
921 }
922
923 static void STDMETHODCALLTYPE buffer_PreLoad(IWineD3DBuffer *iface)
924 {
925     struct wined3d_buffer *This = (struct wined3d_buffer *)iface;
926     IWineD3DDeviceImpl *device = This->resource.device;
927     UINT start = 0, end = 0, len = 0, vertices;
928     const struct wined3d_gl_info *gl_info;
929     struct wined3d_context *context;
930     BOOL decl_changed = FALSE;
931     unsigned int i, j;
932     BYTE *data;
933     DWORD flags = This->flags & (WINED3D_BUFFER_NOSYNC | WINED3D_BUFFER_DISCARD);
934
935     TRACE("iface %p\n", iface);
936     This->flags &= ~(WINED3D_BUFFER_NOSYNC | WINED3D_BUFFER_DISCARD);
937
938     context = context_acquire(device, NULL);
939     gl_info = context->gl_info;
940
941     if (!This->buffer_object)
942     {
943         /* TODO: Make converting independent from VBOs */
944         if (This->flags & WINED3D_BUFFER_CREATEBO)
945         {
946             buffer_create_buffer_object(This, gl_info);
947             This->flags &= ~WINED3D_BUFFER_CREATEBO;
948         }
949         else
950         {
951             /* Not doing any conversion */
952             goto end;
953         }
954     }
955
956     /* Reading the declaration makes only sense if the stateblock is finalized and the buffer bound to a stream */
957     if (device->isInDraw && This->bind_count > 0)
958     {
959         decl_changed = buffer_find_decl(This);
960         This->flags |= WINED3D_BUFFER_HASDESC;
961     }
962
963     if (!decl_changed && !(This->flags & WINED3D_BUFFER_HASDESC && buffer_is_dirty(This)))
964     {
965         context_release(context);
966         ++This->draw_count;
967         if (This->draw_count > VB_RESETDECLCHANGE) This->decl_change_count = 0;
968         if (This->draw_count > VB_RESETFULLCONVS) This->full_conversion_count = 0;
969         return;
970     }
971
972     /* If applications change the declaration over and over, reconverting all the time is a huge
973      * performance hit. So count the declaration changes and release the VBO if there are too many
974      * of them (and thus stop converting)
975      */
976     if (decl_changed)
977     {
978         ++This->decl_change_count;
979         This->draw_count = 0;
980
981         if (This->decl_change_count > VB_MAXDECLCHANGES ||
982             (This->conversion_map && (This->resource.usage & WINED3DUSAGE_DYNAMIC)))
983         {
984             FIXME("Too many declaration changes or converting dynamic buffer, stopping converting\n");
985
986             IWineD3DBuffer_UnLoad(iface);
987             This->flags &= ~WINED3D_BUFFER_CREATEBO;
988
989             /* The stream source state handler might have read the memory of the vertex buffer already
990              * and got the memory in the vbo which is not valid any longer. Dirtify the stream source
991              * to force a reload. This happens only once per changed vertexbuffer and should occur rather
992              * rarely
993              */
994             IWineD3DDeviceImpl_MarkStateDirty(device, STATE_STREAMSRC);
995             goto end;
996         }
997         buffer_check_buffer_object_size(This, gl_info);
998
999         /* The declaration changed, reload the whole buffer */
1000         WARN("Reloading buffer because of decl change\n");
1001         buffer_clear_dirty_areas(This);
1002         if(!buffer_add_dirty_area(This, 0, 0))
1003         {
1004             ERR("buffer_add_dirty_area failed, this is not expected\n");
1005             return;
1006         }
1007         /* Avoid unfenced updates, we might overwrite more areas of the buffer than the application
1008          * cleared for unsynchronized updates
1009          */
1010         flags = 0;
1011     }
1012     else
1013     {
1014         /* However, it is perfectly fine to change the declaration every now and then. We don't want a game that
1015          * changes it every minute drop the VBO after VB_MAX_DECL_CHANGES minutes. So count draws without
1016          * decl changes and reset the decl change count after a specific number of them
1017          */
1018         if(buffer_is_fully_dirty(This))
1019         {
1020             ++This->full_conversion_count;
1021             if(This->full_conversion_count > VB_MAXFULLCONVERSIONS)
1022             {
1023                 FIXME("Too many full buffer conversions, stopping converting\n");
1024                 IWineD3DBuffer_UnLoad(iface);
1025                 This->flags &= ~WINED3D_BUFFER_CREATEBO;
1026                 IWineD3DDeviceImpl_MarkStateDirty(device, STATE_STREAMSRC);
1027                 goto end;
1028             }
1029         }
1030         else
1031         {
1032             ++This->draw_count;
1033             if (This->draw_count > VB_RESETDECLCHANGE) This->decl_change_count = 0;
1034             if (This->draw_count > VB_RESETFULLCONVS) This->full_conversion_count = 0;
1035         }
1036     }
1037
1038     if(This->buffer_type_hint == GL_ELEMENT_ARRAY_BUFFER_ARB)
1039     {
1040         IWineD3DDeviceImpl_MarkStateDirty(This->resource.device, STATE_INDEXBUFFER);
1041     }
1042
1043     if (!This->conversion_map)
1044     {
1045         /* That means that there is nothing to fixup. Just upload from This->resource.allocatedMemory
1046          * directly into the vbo. Do not free the system memory copy because drawPrimitive may need it if
1047          * the stride is 0, for instancing emulation, vertex blending emulation or shader emulation.
1048          */
1049         TRACE("No conversion needed\n");
1050
1051         /* Nothing to do because we locked directly into the vbo */
1052         if (!(This->flags & WINED3D_BUFFER_DOUBLEBUFFER))
1053         {
1054             context_release(context);
1055             return;
1056         }
1057
1058         buffer_direct_upload(This, context->gl_info, flags);
1059
1060         context_release(context);
1061         return;
1062     }
1063
1064     if(!(This->flags & WINED3D_BUFFER_DOUBLEBUFFER))
1065     {
1066         buffer_get_sysmem(This, gl_info);
1067     }
1068
1069     /* Now for each vertex in the buffer that needs conversion */
1070     vertices = This->resource.size / This->stride;
1071
1072     if (This->conversion_shift)
1073     {
1074         TRACE("Shifted conversion\n");
1075         data = HeapAlloc(GetProcessHeap(), 0, vertices * This->conversion_stride);
1076
1077         start = 0;
1078         len = This->resource.size;
1079         end = start + len;
1080
1081         if (This->maps[0].offset || This->maps[0].size != This->resource.size)
1082         {
1083             FIXME("Implement partial buffer load with shifted conversion\n");
1084         }
1085
1086         for (i = start / This->stride; i < min((end / This->stride) + 1, vertices); ++i)
1087         {
1088             for (j = 0; j < This->stride; ++j)
1089             {
1090                 switch(This->conversion_map[j])
1091                 {
1092                     case CONV_NONE:
1093                         data[This->conversion_stride * i + j + This->conversion_shift[j]]
1094                                 = This->resource.allocatedMemory[This->stride * i + j];
1095                         break;
1096
1097                     case CONV_FLOAT16_2:
1098                     {
1099                         float *out = (float *)(&data[This->conversion_stride * i + j + This->conversion_shift[j]]);
1100                         const WORD *in = (WORD *)(&This->resource.allocatedMemory[i * This->stride + j]);
1101
1102                         out[1] = float_16_to_32(in + 1);
1103                         out[0] = float_16_to_32(in + 0);
1104                         j += 3;    /* Skip 3 additional bytes,as a FLOAT16_2 has 4 bytes */
1105                         break;
1106                     }
1107
1108                     default:
1109                         FIXME("Unimplemented conversion %d in shifted conversion\n", This->conversion_map[j]);
1110                         break;
1111                 }
1112             }
1113         }
1114
1115         ENTER_GL();
1116         GL_EXTCALL(glBindBufferARB(This->buffer_type_hint, This->buffer_object));
1117         checkGLcall("glBindBufferARB");
1118         GL_EXTCALL(glBufferSubDataARB(This->buffer_type_hint, 0, vertices * This->conversion_stride, data));
1119         checkGLcall("glBufferSubDataARB");
1120         LEAVE_GL();
1121     }
1122     else
1123     {
1124         data = HeapAlloc(GetProcessHeap(), 0, This->resource.size);
1125
1126         while(This->modified_areas)
1127         {
1128             This->modified_areas--;
1129             start = This->maps[This->modified_areas].offset;
1130             len = This->maps[This->modified_areas].size;
1131             end = start + len;
1132
1133             memcpy(data + start, This->resource.allocatedMemory + start, end - start);
1134             for (i = start / This->stride; i < min((end / This->stride) + 1, vertices); ++i)
1135             {
1136                 for (j = 0; j < This->stride; ++j)
1137                 {
1138                     switch(This->conversion_map[j])
1139                     {
1140                         case CONV_NONE:
1141                             /* Done already */
1142                             j += 3;
1143                             break;
1144                         case CONV_D3DCOLOR:
1145                             fixup_d3dcolor((DWORD *) (data + i * This->stride + j));
1146                             j += 3;
1147                             break;
1148
1149                         case CONV_POSITIONT:
1150                             fixup_transformed_pos((float *) (data + i * This->stride + j));
1151                             j += 15;
1152                             break;
1153
1154                         case CONV_FLOAT16_2:
1155                             ERR("Did not expect FLOAT16 conversion in unshifted conversion\n");
1156                         default:
1157                             FIXME("Unimplemented conversion %d in shifted conversion\n", This->conversion_map[j]);
1158                     }
1159                 }
1160             }
1161
1162             ENTER_GL();
1163             GL_EXTCALL(glBindBufferARB(This->buffer_type_hint, This->buffer_object));
1164             checkGLcall("glBindBufferARB");
1165             GL_EXTCALL(glBufferSubDataARB(This->buffer_type_hint, start, len, data + start));
1166             checkGLcall("glBufferSubDataARB");
1167             LEAVE_GL();
1168         }
1169     }
1170
1171     HeapFree(GetProcessHeap(), 0, data);
1172
1173 end:
1174     context_release(context);
1175 }
1176
1177 static WINED3DRESOURCETYPE STDMETHODCALLTYPE buffer_GetType(IWineD3DBuffer *iface)
1178 {
1179     return resource_get_type((IWineD3DResource *)iface);
1180 }
1181
1182 /* IWineD3DBuffer methods */
1183
1184 static DWORD buffer_sanitize_flags(struct wined3d_buffer *buffer, DWORD flags)
1185 {
1186     /* Not all flags make sense together, but Windows never returns an error. Catch the
1187      * cases that could cause issues */
1188     if(flags & WINED3DLOCK_READONLY)
1189     {
1190         if(flags & WINED3DLOCK_DISCARD)
1191         {
1192             WARN("WINED3DLOCK_READONLY combined with WINED3DLOCK_DISCARD, ignoring flags\n");
1193             return 0;
1194         }
1195         if(flags & WINED3DLOCK_NOOVERWRITE)
1196         {
1197             WARN("WINED3DLOCK_READONLY combined with WINED3DLOCK_NOOVERWRITE, ignoring flags\n");
1198             return 0;
1199         }
1200     }
1201     else if((flags & (WINED3DLOCK_DISCARD | WINED3DLOCK_NOOVERWRITE)) == (WINED3DLOCK_DISCARD | WINED3DLOCK_NOOVERWRITE))
1202     {
1203         WARN("WINED3DLOCK_DISCARD and WINED3DLOCK_NOOVERWRITE used together, ignoring\n");
1204         return 0;
1205     }
1206     else if (flags & (WINED3DLOCK_DISCARD | WINED3DLOCK_NOOVERWRITE) && !(buffer->resource.usage & WINED3DUSAGE_DYNAMIC))
1207     {
1208         WARN("DISCARD or NOOVERWRITE lock on non-dynamic buffer, ignoring\n");
1209         return 0;
1210     }
1211
1212     return flags;
1213 }
1214
1215 static GLbitfield buffer_gl_map_flags(DWORD d3d_flags)
1216 {
1217     GLbitfield ret = 0;
1218
1219     if (!(d3d_flags & WINED3DLOCK_READONLY)) ret = GL_MAP_WRITE_BIT | GL_MAP_FLUSH_EXPLICIT_BIT;
1220
1221     if (d3d_flags & (WINED3DLOCK_DISCARD | WINED3DLOCK_NOOVERWRITE))
1222     {
1223         if(d3d_flags & WINED3DLOCK_DISCARD) ret |= GL_MAP_INVALIDATE_BUFFER_BIT;
1224         ret |= GL_MAP_UNSYNCHRONIZED_BIT;
1225     }
1226     else
1227     {
1228         ret |= GL_MAP_READ_BIT;
1229     }
1230
1231     return ret;
1232 }
1233
1234 static HRESULT STDMETHODCALLTYPE buffer_Map(IWineD3DBuffer *iface, UINT offset, UINT size, BYTE **data, DWORD flags)
1235 {
1236     struct wined3d_buffer *This = (struct wined3d_buffer *)iface;
1237     LONG count;
1238     BOOL dirty = buffer_is_dirty(This);
1239
1240     TRACE("iface %p, offset %u, size %u, data %p, flags %#x\n", iface, offset, size, data, flags);
1241
1242     flags = buffer_sanitize_flags(This, flags);
1243     if (!(flags & WINED3DLOCK_READONLY))
1244     {
1245         if (!buffer_add_dirty_area(This, offset, size)) return E_OUTOFMEMORY;
1246     }
1247
1248     count = InterlockedIncrement(&This->lock_count);
1249
1250     if (This->buffer_object)
1251     {
1252         if(!(This->flags & WINED3D_BUFFER_DOUBLEBUFFER))
1253         {
1254             if(count == 1)
1255             {
1256                 IWineD3DDeviceImpl *device = This->resource.device;
1257                 struct wined3d_context *context;
1258                 const struct wined3d_gl_info *gl_info;
1259
1260                 if(This->buffer_type_hint == GL_ELEMENT_ARRAY_BUFFER_ARB)
1261                 {
1262                     IWineD3DDeviceImpl_MarkStateDirty(This->resource.device, STATE_INDEXBUFFER);
1263                 }
1264
1265                 context = context_acquire(device, NULL);
1266                 gl_info = context->gl_info;
1267                 ENTER_GL();
1268                 GL_EXTCALL(glBindBufferARB(This->buffer_type_hint, This->buffer_object));
1269
1270                 if (gl_info->supported[ARB_MAP_BUFFER_RANGE])
1271                 {
1272                     GLbitfield mapflags = buffer_gl_map_flags(flags);
1273                     This->resource.allocatedMemory = GL_EXTCALL(glMapBufferRange(This->buffer_type_hint, 0,
1274                                                                                 This->resource.size, mapflags));
1275                     checkGLcall("glMapBufferRange");
1276                 }
1277                 else
1278                 {
1279                     if(This->flags & WINED3D_BUFFER_APPLESYNC)
1280                     {
1281                         LEAVE_GL();
1282                         buffer_sync_apple(This, flags, gl_info);
1283                         ENTER_GL();
1284                     }
1285                     This->resource.allocatedMemory = GL_EXTCALL(glMapBufferARB(This->buffer_type_hint, GL_READ_WRITE_ARB));
1286                     checkGLcall("glMapBufferARB");
1287                 }
1288                 LEAVE_GL();
1289
1290                 if (((DWORD_PTR) This->resource.allocatedMemory) & (RESOURCE_ALIGNMENT - 1))
1291                 {
1292                     WARN("Pointer %p is not %u byte aligned, falling back to double buffered operation\n",
1293                         This->resource.allocatedMemory, RESOURCE_ALIGNMENT);
1294
1295                     ENTER_GL();
1296                     GL_EXTCALL(glUnmapBufferARB(This->buffer_type_hint));
1297                     checkGLcall("glUnmapBufferARB");
1298                     LEAVE_GL();
1299                     This->resource.allocatedMemory = NULL;
1300
1301                     buffer_get_sysmem(This, gl_info);
1302                     TRACE("New pointer is %p\n", This->resource.allocatedMemory);
1303                 }
1304                 context_release(context);
1305             }
1306         }
1307         else
1308         {
1309             if (dirty)
1310             {
1311                 if (This->flags & WINED3D_BUFFER_NOSYNC && !(flags & WINED3DLOCK_NOOVERWRITE))
1312                 {
1313                     This->flags &= ~WINED3D_BUFFER_NOSYNC;
1314                 }
1315             }
1316             else if(flags & WINED3DLOCK_NOOVERWRITE)
1317             {
1318                 This->flags |= WINED3D_BUFFER_NOSYNC;
1319             }
1320
1321             if (flags & WINED3DLOCK_DISCARD)
1322             {
1323                 This->flags |= WINED3D_BUFFER_DISCARD;
1324             }
1325         }
1326     }
1327
1328     *data = This->resource.allocatedMemory + offset;
1329
1330     TRACE("Returning memory at %p (base %p, offset %u)\n", *data, This->resource.allocatedMemory, offset);
1331     /* TODO: check Flags compatibility with This->currentDesc.Usage (see MSDN) */
1332
1333     return WINED3D_OK;
1334 }
1335
1336 static HRESULT STDMETHODCALLTYPE buffer_Unmap(IWineD3DBuffer *iface)
1337 {
1338     struct wined3d_buffer *This = (struct wined3d_buffer *)iface;
1339     ULONG i;
1340
1341     TRACE("(%p)\n", This);
1342
1343     /* In the case that the number of Unmap calls > the
1344      * number of Map calls, d3d returns always D3D_OK.
1345      * This is also needed to prevent Map from returning garbage on
1346      * the next call (this will happen if the lock_count is < 0). */
1347     if(This->lock_count == 0)
1348     {
1349         TRACE("Unmap called without a previous Map call!\n");
1350         return WINED3D_OK;
1351     }
1352
1353     if (InterlockedDecrement(&This->lock_count))
1354     {
1355         /* Delay loading the buffer until everything is unlocked */
1356         TRACE("Ignoring unlock\n");
1357         return WINED3D_OK;
1358     }
1359
1360     if(!(This->flags & WINED3D_BUFFER_DOUBLEBUFFER) && This->buffer_object)
1361     {
1362         IWineD3DDeviceImpl *device = This->resource.device;
1363         const struct wined3d_gl_info *gl_info;
1364         struct wined3d_context *context;
1365
1366         if(This->buffer_type_hint == GL_ELEMENT_ARRAY_BUFFER_ARB)
1367         {
1368             IWineD3DDeviceImpl_MarkStateDirty(This->resource.device, STATE_INDEXBUFFER);
1369         }
1370
1371         context = context_acquire(device, NULL);
1372         gl_info = context->gl_info;
1373         ENTER_GL();
1374         GL_EXTCALL(glBindBufferARB(This->buffer_type_hint, This->buffer_object));
1375
1376         if (gl_info->supported[ARB_MAP_BUFFER_RANGE])
1377         {
1378             for(i = 0; i < This->modified_areas; i++)
1379             {
1380                 GL_EXTCALL(glFlushMappedBufferRange(This->buffer_type_hint,
1381                                                     This->maps[i].offset,
1382                                                     This->maps[i].size));
1383                 checkGLcall("glFlushMappedBufferRange");
1384             }
1385         }
1386         else if (This->flags & WINED3D_BUFFER_FLUSH)
1387         {
1388             for(i = 0; i < This->modified_areas; i++)
1389             {
1390                 GL_EXTCALL(glFlushMappedBufferRangeAPPLE(This->buffer_type_hint,
1391                                                          This->maps[i].offset,
1392                                                          This->maps[i].size));
1393                 checkGLcall("glFlushMappedBufferRangeAPPLE");
1394             }
1395         }
1396
1397         GL_EXTCALL(glUnmapBufferARB(This->buffer_type_hint));
1398         LEAVE_GL();
1399         context_release(context);
1400
1401         This->resource.allocatedMemory = NULL;
1402         buffer_clear_dirty_areas(This);
1403     }
1404     else if (This->flags & WINED3D_BUFFER_HASDESC)
1405     {
1406         buffer_PreLoad(iface);
1407     }
1408
1409     return WINED3D_OK;
1410 }
1411
1412 static HRESULT STDMETHODCALLTYPE buffer_GetDesc(IWineD3DBuffer *iface, WINED3DBUFFER_DESC *desc)
1413 {
1414     struct wined3d_buffer *This = (struct wined3d_buffer *)iface;
1415
1416     TRACE("(%p)\n", This);
1417
1418     desc->Type = This->resource.resourceType;
1419     desc->Usage = This->resource.usage;
1420     desc->Pool = This->resource.pool;
1421     desc->Size = This->resource.size;
1422
1423     return WINED3D_OK;
1424 }
1425
1426 static const struct IWineD3DBufferVtbl wined3d_buffer_vtbl =
1427 {
1428     /* IUnknown methods */
1429     buffer_QueryInterface,
1430     buffer_AddRef,
1431     buffer_Release,
1432     /* IWineD3DBase methods */
1433     buffer_GetParent,
1434     /* IWineD3DResource methods */
1435     buffer_SetPrivateData,
1436     buffer_GetPrivateData,
1437     buffer_FreePrivateData,
1438     buffer_SetPriority,
1439     buffer_GetPriority,
1440     buffer_PreLoad,
1441     buffer_UnLoad,
1442     buffer_GetType,
1443     /* IWineD3DBuffer methods */
1444     buffer_Map,
1445     buffer_Unmap,
1446     buffer_GetDesc,
1447 };
1448
1449 HRESULT buffer_init(struct wined3d_buffer *buffer, IWineD3DDeviceImpl *device,
1450         UINT size, DWORD usage, WINED3DFORMAT format, WINED3DPOOL pool, GLenum bind_hint,
1451         const char *data, IUnknown *parent, const struct wined3d_parent_ops *parent_ops)
1452 {
1453     const struct wined3d_format_desc *format_desc = getFormatDescEntry(format, &device->adapter->gl_info);
1454     HRESULT hr;
1455     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
1456     BOOL dynamic_buffer_ok;
1457
1458     if (!size)
1459     {
1460         WARN("Size 0 requested, returning WINED3DERR_INVALIDCALL\n");
1461         return WINED3DERR_INVALIDCALL;
1462     }
1463
1464     buffer->vtbl = &wined3d_buffer_vtbl;
1465
1466     hr = resource_init((IWineD3DResource *)buffer, WINED3DRTYPE_BUFFER,
1467             device, size, usage, format_desc, pool, parent, parent_ops);
1468     if (FAILED(hr))
1469     {
1470         WARN("Failed to initialize resource, hr %#x\n", hr);
1471         return hr;
1472     }
1473     buffer->buffer_type_hint = bind_hint;
1474
1475     TRACE("size %#x, usage %#x, format %s, memory @ %p, iface @ %p.\n", buffer->resource.size, buffer->resource.usage,
1476             debug_d3dformat(buffer->resource.format_desc->format), buffer->resource.allocatedMemory, buffer);
1477
1478     /* GL_ARB_map_buffer_range is disabled for now due to numerous bugs and no gains */
1479     dynamic_buffer_ok = gl_info->supported[APPLE_FLUSH_BUFFER_RANGE];
1480
1481     /* Observations show that drawStridedSlow is faster on dynamic VBs than converting +
1482      * drawStridedFast (half-life 2 and others).
1483      *
1484      * Basically converting the vertices in the buffer is quite expensive, and observations
1485      * show that drawStridedSlow is faster than converting + uploading + drawStridedFast.
1486      * Therefore do not create a VBO for WINED3DUSAGE_DYNAMIC buffers.
1487      */
1488     if (!gl_info->supported[ARB_VERTEX_BUFFER_OBJECT])
1489     {
1490         TRACE("Not creating a vbo because GL_ARB_vertex_buffer is not supported\n");
1491     }
1492     else if(buffer->resource.pool == WINED3DPOOL_SYSTEMMEM)
1493     {
1494         TRACE("Not creating a vbo because the vertex buffer is in system memory\n");
1495     }
1496     else if(!dynamic_buffer_ok && (buffer->resource.usage & WINED3DUSAGE_DYNAMIC))
1497     {
1498         TRACE("Not creating a vbo because the buffer has dynamic usage and no GL support\n");
1499     }
1500     else
1501     {
1502         buffer->flags |= WINED3D_BUFFER_CREATEBO;
1503     }
1504
1505     if (data)
1506     {
1507         BYTE *ptr;
1508
1509         hr = IWineD3DBuffer_Map((IWineD3DBuffer *)buffer, 0, size, &ptr, 0);
1510         if (FAILED(hr))
1511         {
1512             ERR("Failed to map buffer, hr %#x\n", hr);
1513             buffer_UnLoad((IWineD3DBuffer *)buffer);
1514             resource_cleanup((IWineD3DResource *)buffer);
1515             return hr;
1516         }
1517
1518         memcpy(ptr, data, size);
1519
1520         hr = IWineD3DBuffer_Unmap((IWineD3DBuffer *)buffer);
1521         if (FAILED(hr))
1522         {
1523             ERR("Failed to unmap buffer, hr %#x\n", hr);
1524             buffer_UnLoad((IWineD3DBuffer *)buffer);
1525             resource_cleanup((IWineD3DResource *)buffer);
1526             return hr;
1527         }
1528     }
1529
1530     buffer->maps = HeapAlloc(GetProcessHeap(), 0, sizeof(*buffer->maps));
1531     if (!buffer->maps)
1532     {
1533         ERR("Out of memory\n");
1534         buffer_UnLoad((IWineD3DBuffer *)buffer);
1535         resource_cleanup((IWineD3DResource *)buffer);
1536         return E_OUTOFMEMORY;
1537     }
1538     buffer->maps_size = 1;
1539
1540     return WINED3D_OK;
1541 }