wined3d: Recognize the SM4 itof opcode.
[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;
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 && 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 offset = This->resource.device->stateBlock->state.streams[attrib->stream_idx].offset;
235     DWORD attrib_size;
236     BOOL ret = FALSE;
237     unsigned int i;
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->id));
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->component_count * attrib->format->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     enum wined3d_format_id format;
296     BOOL ret = FALSE;
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->id;
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         enum wined3d_format_id 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->id;
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->component_count
378                     * strided->elements[i].format->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     const struct wined3d_state *state = &device->stateBlock->state;
405     UINT stride_this_run = 0;
406     BOOL float16_used = FALSE;
407     BOOL ret = FALSE;
408     unsigned int i;
409
410     /* In d3d7 the vertex buffer declaration NEVER changes because it is stored in the d3d7 vertex buffer.
411      * Once we have our declaration there is no need to look it up again. Index buffers also never need
412      * conversion, so once the (empty) conversion structure is created don't bother checking again
413      */
414     if (This->flags & WINED3D_BUFFER_HASDESC)
415     {
416         if(This->resource.usage & WINED3DUSAGE_STATICDECL) return FALSE;
417     }
418
419     TRACE("Finding vertex buffer conversion information\n");
420     /* Certain declaration types need some fixups before we can pass them to
421      * opengl. This means D3DCOLOR attributes with fixed function vertex
422      * processing, FLOAT4 POSITIONT with fixed function, and FLOAT16 if
423      * GL_ARB_half_float_vertex is not supported.
424      *
425      * Note for d3d8 and d3d9:
426      * The vertex buffer FVF doesn't help with finding them, we have to use
427      * the decoded vertex declaration and pick the things that concern the
428      * current buffer. A problem with this is that this can change between
429      * draws, so we have to validate the information and reprocess the buffer
430      * if it changes, and avoid false positives for performance reasons.
431      * WineD3D doesn't even know the vertex buffer any more, it is managed
432      * by the client libraries and passed to SetStreamSource and ProcessVertices
433      * as needed.
434      *
435      * We have to distinguish between vertex shaders and fixed function to
436      * pick the way we access the strided vertex information.
437      *
438      * This code sets up a per-byte array with the size of the detected
439      * stride of the arrays in the buffer. For each byte we have a field
440      * that marks the conversion needed on this byte. For example, the
441      * following declaration with fixed function vertex processing:
442      *
443      *      POSITIONT, FLOAT4
444      *      NORMAL, FLOAT3
445      *      DIFFUSE, FLOAT16_4
446      *      SPECULAR, D3DCOLOR
447      *
448      * Will result in
449      * {                 POSITIONT                    }{             NORMAL                }{    DIFFUSE          }{SPECULAR }
450      * [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]
451      *
452      * Where in this example map P means 4 component position conversion, 0
453      * means no conversion, F means FLOAT16_2 conversion and C means D3DCOLOR
454      * conversion (red / blue swizzle).
455      *
456      * If we're doing conversion and the stride changes we have to reconvert
457      * the whole buffer. Note that we do not mind if the semantic changes,
458      * we only care for the conversion type. So if the NORMAL is replaced
459      * with a TEXCOORD, nothing has to be done, or if the DIFFUSE is replaced
460      * with a D3DCOLOR BLENDWEIGHT we can happily dismiss the change. Some
461      * conversion types depend on the semantic as well, for example a FLOAT4
462      * texcoord needs no conversion while a FLOAT4 positiont needs one
463      */
464     if (use_vs(state))
465     {
466         TRACE("vshader\n");
467         /* If the current vertex declaration is marked for no half float conversion don't bother to
468          * analyse the strided streams in depth, just set them up for no conversion. Return decl changed
469          * if we used conversion before
470          */
471         if (!state->vertex_declaration->half_float_conv_needed)
472         {
473             if (This->conversion_map)
474             {
475                 TRACE("Now using shaders without conversion, but conversion used before\n");
476                 HeapFree(GetProcessHeap(), 0, This->conversion_map);
477                 HeapFree(GetProcessHeap(), 0, This->conversion_shift);
478                 This->conversion_map = NULL;
479                 This->stride = 0;
480                 This->conversion_shift = NULL;
481                 This->conversion_stride = 0;
482                 return TRUE;
483             }
484             else
485             {
486                 return FALSE;
487             }
488         }
489         for (i = 0; i < MAX_ATTRIBS; ++i)
490         {
491             ret = buffer_check_attribute(This, si, i, FALSE, FALSE, FALSE, &stride_this_run, &float16_used) || ret;
492         }
493
494         /* Recalculate the conversion shift map if the declaration has changed,
495          * and we're using float16 conversion or used it on the last run
496          */
497         if (ret && (float16_used || This->conversion_map))
498         {
499             HeapFree(GetProcessHeap(), 0, This->conversion_shift);
500             This->conversion_shift = find_conversion_shift(This, si, This->stride);
501         }
502     }
503     else
504     {
505         /* Fixed function is a bit trickier. We have to take care for D3DCOLOR types, FLOAT4 positions and of course
506          * FLOAT16s if not supported. Also, we can't iterate over the array, so use macros to generate code for all
507          * the attributes that our current fixed function pipeline implementation cares for.
508          */
509         BOOL support_d3dcolor = gl_info->supported[ARB_VERTEX_ARRAY_BGRA];
510         ret = buffer_check_attribute(This, si, WINED3D_FFP_POSITION,
511                 TRUE, TRUE,  FALSE, &stride_this_run, &float16_used) || ret;
512         ret = buffer_check_attribute(This, si, WINED3D_FFP_NORMAL,
513                 TRUE, FALSE, FALSE, &stride_this_run, &float16_used) || ret;
514         ret = buffer_check_attribute(This, si, WINED3D_FFP_DIFFUSE,
515                 !support_d3dcolor, FALSE, TRUE,  &stride_this_run, &float16_used) || ret;
516         ret = buffer_check_attribute(This, si, WINED3D_FFP_SPECULAR,
517                 !support_d3dcolor, FALSE, TRUE,  &stride_this_run, &float16_used) || ret;
518         ret = buffer_check_attribute(This, si, WINED3D_FFP_TEXCOORD0,
519                 TRUE, FALSE, FALSE, &stride_this_run, &float16_used) || ret;
520         ret = buffer_check_attribute(This, si, WINED3D_FFP_TEXCOORD1,
521                 TRUE, FALSE, FALSE, &stride_this_run, &float16_used) || ret;
522         ret = buffer_check_attribute(This, si, WINED3D_FFP_TEXCOORD2,
523                 TRUE, FALSE, FALSE, &stride_this_run, &float16_used) || ret;
524         ret = buffer_check_attribute(This, si, WINED3D_FFP_TEXCOORD3,
525                 TRUE, FALSE, FALSE, &stride_this_run, &float16_used) || ret;
526         ret = buffer_check_attribute(This, si, WINED3D_FFP_TEXCOORD4,
527                 TRUE, FALSE, FALSE, &stride_this_run, &float16_used) || ret;
528         ret = buffer_check_attribute(This, si, WINED3D_FFP_TEXCOORD5,
529                 TRUE, FALSE, FALSE, &stride_this_run, &float16_used) || ret;
530         ret = buffer_check_attribute(This, si, WINED3D_FFP_TEXCOORD6,
531                 TRUE, FALSE, FALSE, &stride_this_run, &float16_used) || ret;
532         ret = buffer_check_attribute(This, si, WINED3D_FFP_TEXCOORD7,
533                 TRUE, FALSE, FALSE, &stride_this_run, &float16_used) || ret;
534
535         if (float16_used) FIXME("Float16 conversion used with fixed function vertex processing\n");
536     }
537
538     if (!stride_this_run && This->conversion_map)
539     {
540         /* Sanity test */
541         if (!ret) ERR("no converted attributes found, old conversion map exists, and no declaration change?\n");
542         HeapFree(GetProcessHeap(), 0, This->conversion_map);
543         This->conversion_map = NULL;
544         This->stride = 0;
545     }
546
547     if (ret) TRACE("Conversion information changed\n");
548
549     return ret;
550 }
551
552 /* Context activation is done by the caller. */
553 static void buffer_check_buffer_object_size(struct wined3d_buffer *This, const struct wined3d_gl_info *gl_info)
554 {
555     UINT size = This->conversion_stride ?
556             This->conversion_stride * (This->resource.size / This->stride) : This->resource.size;
557     if (This->buffer_object_size != size)
558     {
559         TRACE("Old size %u, creating new size %u\n", This->buffer_object_size, size);
560
561         if(This->buffer_type_hint == GL_ELEMENT_ARRAY_BUFFER_ARB)
562         {
563             IWineD3DDeviceImpl_MarkStateDirty(This->resource.device, STATE_INDEXBUFFER);
564         }
565
566         /* Rescue the data before resizing the buffer object if we do not have our backup copy */
567         if(!(This->flags & WINED3D_BUFFER_DOUBLEBUFFER))
568         {
569             buffer_get_sysmem(This, gl_info);
570         }
571
572         ENTER_GL();
573         GL_EXTCALL(glBindBufferARB(This->buffer_type_hint, This->buffer_object));
574         checkGLcall("glBindBufferARB");
575         GL_EXTCALL(glBufferDataARB(This->buffer_type_hint, size, NULL, This->buffer_object_usage));
576         This->buffer_object_size = size;
577         checkGLcall("glBufferDataARB");
578         LEAVE_GL();
579     }
580 }
581
582 static inline void fixup_d3dcolor(DWORD *dst_color)
583 {
584     DWORD src_color = *dst_color;
585
586     /* Color conversion like in drawStridedSlow. watch out for little endianity
587      * If we want that stuff to work on big endian machines too we have to consider more things
588      *
589      * 0xff000000: Alpha mask
590      * 0x00ff0000: Blue mask
591      * 0x0000ff00: Green mask
592      * 0x000000ff: Red mask
593      */
594     *dst_color = 0;
595     *dst_color |= (src_color & 0xff00ff00);         /* Alpha Green */
596     *dst_color |= (src_color & 0x00ff0000) >> 16;   /* Red */
597     *dst_color |= (src_color & 0x000000ff) << 16;   /* Blue */
598 }
599
600 static inline void fixup_transformed_pos(float *p)
601 {
602     /* rhw conversion like in position_float4(). */
603     if (p[3] != 1.0f && p[3] != 0.0f)
604     {
605         float w = 1.0f / p[3];
606         p[0] *= w;
607         p[1] *= w;
608         p[2] *= w;
609         p[3] = w;
610     }
611 }
612
613 /* Context activation is done by the caller. */
614 const BYTE *buffer_get_memory(IWineD3DBuffer *iface, const struct wined3d_gl_info *gl_info, GLuint *buffer_object)
615 {
616     struct wined3d_buffer *This = (struct wined3d_buffer *)iface;
617
618     *buffer_object = This->buffer_object;
619     if (!This->buffer_object)
620     {
621         if (This->flags & WINED3D_BUFFER_CREATEBO)
622         {
623             buffer_create_buffer_object(This, gl_info);
624             This->flags &= ~WINED3D_BUFFER_CREATEBO;
625             if (This->buffer_object)
626             {
627                 *buffer_object = This->buffer_object;
628                 return NULL;
629             }
630         }
631         return This->resource.allocatedMemory;
632     }
633     else
634     {
635         return NULL;
636     }
637 }
638
639 /* IUnknown methods */
640
641 static HRESULT STDMETHODCALLTYPE buffer_QueryInterface(IWineD3DBuffer *iface,
642         REFIID riid, void **object)
643 {
644     TRACE("iface %p, riid %s, object %p\n", iface, debugstr_guid(riid), object);
645
646     if (IsEqualGUID(riid, &IID_IWineD3DBuffer)
647             || IsEqualGUID(riid, &IID_IWineD3DResource)
648             || IsEqualGUID(riid, &IID_IWineD3DBase)
649             || IsEqualGUID(riid, &IID_IUnknown))
650     {
651         IUnknown_AddRef(iface);
652         *object = iface;
653         return S_OK;
654     }
655
656     WARN("%s not implemented, returning E_NOINTERFACE\n", debugstr_guid(riid));
657
658     *object = NULL;
659
660     return E_NOINTERFACE;
661 }
662
663 static ULONG STDMETHODCALLTYPE buffer_AddRef(IWineD3DBuffer *iface)
664 {
665     struct wined3d_buffer *This = (struct wined3d_buffer *)iface;
666     ULONG refcount = InterlockedIncrement(&This->resource.ref);
667
668     TRACE("%p increasing refcount to %u\n", This, refcount);
669
670     return refcount;
671 }
672
673 /* Context activation is done by the caller. */
674 BYTE *buffer_get_sysmem(struct wined3d_buffer *This, const struct wined3d_gl_info *gl_info)
675 {
676     /* AllocatedMemory exists if the buffer is double buffered or has no buffer object at all */
677     if(This->resource.allocatedMemory) return This->resource.allocatedMemory;
678
679     This->resource.heapMemory = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, This->resource.size + RESOURCE_ALIGNMENT);
680     This->resource.allocatedMemory = (BYTE *)(((ULONG_PTR)This->resource.heapMemory + (RESOURCE_ALIGNMENT - 1)) & ~(RESOURCE_ALIGNMENT - 1));
681     ENTER_GL();
682     GL_EXTCALL(glBindBufferARB(This->buffer_type_hint, This->buffer_object));
683     GL_EXTCALL(glGetBufferSubDataARB(This->buffer_type_hint, 0, This->resource.size, This->resource.allocatedMemory));
684     LEAVE_GL();
685     This->flags |= WINED3D_BUFFER_DOUBLEBUFFER;
686
687     return This->resource.allocatedMemory;
688 }
689
690 /* Do not call while under the GL lock. */
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     resource_unload((IWineD3DResourceImpl *)This);
727 }
728
729 /* Do not call while under the GL lock. */
730 static ULONG STDMETHODCALLTYPE buffer_Release(IWineD3DBuffer *iface)
731 {
732     struct wined3d_buffer *This = (struct wined3d_buffer *)iface;
733     ULONG refcount = InterlockedDecrement(&This->resource.ref);
734
735     TRACE("%p decreasing refcount to %u\n", This, refcount);
736
737     if (!refcount)
738     {
739         buffer_UnLoad(iface);
740         resource_cleanup((IWineD3DResource *)iface);
741         This->resource.parent_ops->wined3d_object_destroyed(This->resource.parent);
742         HeapFree(GetProcessHeap(), 0, This->maps);
743         HeapFree(GetProcessHeap(), 0, This);
744     }
745
746     return refcount;
747 }
748
749 /* IWineD3DBase methods */
750 static void * STDMETHODCALLTYPE buffer_GetParent(IWineD3DBuffer *iface)
751 {
752     TRACE("iface %p.\n", iface);
753
754     return ((struct wined3d_buffer *)iface)->resource.parent;
755 }
756
757 /* IWineD3DResource methods */
758
759 static HRESULT STDMETHODCALLTYPE buffer_SetPrivateData(IWineD3DBuffer *iface,
760         REFGUID guid, const void *data, DWORD data_size, DWORD flags)
761 {
762     return resource_set_private_data((IWineD3DResource *)iface, guid, data, data_size, flags);
763 }
764
765 static HRESULT STDMETHODCALLTYPE buffer_GetPrivateData(IWineD3DBuffer *iface,
766         REFGUID guid, void *data, DWORD *data_size)
767 {
768     return resource_get_private_data((IWineD3DResource *)iface, guid, data, data_size);
769 }
770
771 static HRESULT STDMETHODCALLTYPE buffer_FreePrivateData(IWineD3DBuffer *iface, REFGUID guid)
772 {
773     return resource_free_private_data((IWineD3DResource *)iface, guid);
774 }
775
776 static DWORD STDMETHODCALLTYPE buffer_SetPriority(IWineD3DBuffer *iface, DWORD priority)
777 {
778     return resource_set_priority((IWineD3DResource *)iface, priority);
779 }
780
781 static DWORD STDMETHODCALLTYPE buffer_GetPriority(IWineD3DBuffer *iface)
782 {
783     return resource_get_priority((IWineD3DResource *)iface);
784 }
785
786 /* The caller provides a context and binds the buffer */
787 static void buffer_sync_apple(struct wined3d_buffer *This, DWORD flags, const struct wined3d_gl_info *gl_info)
788 {
789     enum wined3d_event_query_result ret;
790
791     /* No fencing needs to be done if the app promises not to overwrite
792      * existing data */
793     if(flags & WINED3DLOCK_NOOVERWRITE) return;
794     if(flags & WINED3DLOCK_DISCARD)
795     {
796         ENTER_GL();
797         GL_EXTCALL(glBufferDataARB(This->buffer_type_hint, This->resource.size, NULL, This->buffer_object_usage));
798         checkGLcall("glBufferDataARB\n");
799         LEAVE_GL();
800         return;
801     }
802
803     if(!This->query)
804     {
805         TRACE("Creating event query for buffer %p\n", This);
806
807         if (!wined3d_event_query_supported(gl_info))
808         {
809             FIXME("Event queries not supported, dropping async buffer locks.\n");
810             goto drop_query;
811         }
812
813         This->query = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*This->query));
814         if (!This->query)
815         {
816             ERR("Failed to allocate event query memory, dropping async buffer locks.\n");
817             goto drop_query;
818         }
819
820         /* Since we don't know about old draws a glFinish is needed once */
821         wglFinish();
822         return;
823     }
824     TRACE("Synchronizing buffer %p\n", This);
825     ret = wined3d_event_query_finish(This->query, This->resource.device);
826     switch(ret)
827     {
828         case WINED3D_EVENT_QUERY_NOT_STARTED:
829         case WINED3D_EVENT_QUERY_OK:
830             /* All done */
831             return;
832
833         case WINED3D_EVENT_QUERY_WRONG_THREAD:
834             WARN("Cannot synchronize buffer lock due to a thread conflict\n");
835             goto drop_query;
836
837         default:
838             ERR("wined3d_event_query_finish returned %u, dropping async buffer locks\n", ret);
839             goto drop_query;
840     }
841
842 drop_query:
843     if(This->query)
844     {
845         wined3d_event_query_destroy(This->query);
846         This->query = NULL;
847     }
848
849     wglFinish();
850     ENTER_GL();
851     GL_EXTCALL(glBufferParameteriAPPLE(This->buffer_type_hint, GL_BUFFER_SERIALIZED_MODIFY_APPLE, GL_TRUE));
852     checkGLcall("glBufferParameteriAPPLE(This->buffer_type_hint, GL_BUFFER_SERIALIZED_MODIFY_APPLE, GL_TRUE)");
853     LEAVE_GL();
854     This->flags &= ~WINED3D_BUFFER_APPLESYNC;
855 }
856
857 /* The caller provides a GL context */
858 static void buffer_direct_upload(struct wined3d_buffer *This, const struct wined3d_gl_info *gl_info, DWORD flags)
859 {
860         BYTE *map;
861         UINT start = 0, len = 0;
862
863         ENTER_GL();
864         GL_EXTCALL(glBindBufferARB(This->buffer_type_hint, This->buffer_object));
865         checkGLcall("glBindBufferARB");
866         if (gl_info->supported[ARB_MAP_BUFFER_RANGE])
867         {
868             GLbitfield mapflags;
869             mapflags = GL_MAP_WRITE_BIT | GL_MAP_FLUSH_EXPLICIT_BIT;
870             if (flags & WINED3D_BUFFER_DISCARD)
871             {
872                 mapflags |= GL_MAP_UNSYNCHRONIZED_BIT | GL_MAP_INVALIDATE_BUFFER_BIT;
873             }
874             else if (flags & WINED3D_BUFFER_NOSYNC)
875             {
876                 mapflags |= GL_MAP_UNSYNCHRONIZED_BIT;
877             }
878             map = GL_EXTCALL(glMapBufferRange(This->buffer_type_hint, 0,
879                                               This->resource.size, mapflags));
880             checkGLcall("glMapBufferRange");
881         }
882         else
883         {
884             if (This->flags & WINED3D_BUFFER_APPLESYNC)
885             {
886                 DWORD syncflags = 0;
887                 if (flags & WINED3D_BUFFER_DISCARD) syncflags |= WINED3DLOCK_DISCARD;
888                 if (flags & WINED3D_BUFFER_NOSYNC) syncflags |= WINED3DLOCK_NOOVERWRITE;
889                 LEAVE_GL();
890                 buffer_sync_apple(This, syncflags, gl_info);
891                 ENTER_GL();
892             }
893             map = GL_EXTCALL(glMapBufferARB(This->buffer_type_hint, GL_WRITE_ONLY_ARB));
894             checkGLcall("glMapBufferARB");
895         }
896         if (!map)
897         {
898             LEAVE_GL();
899             ERR("Failed to map opengl buffer\n");
900             return;
901         }
902
903         while(This->modified_areas)
904         {
905             This->modified_areas--;
906             start = This->maps[This->modified_areas].offset;
907             len = This->maps[This->modified_areas].size;
908
909             memcpy(map + start, This->resource.allocatedMemory + start, len);
910
911             if (gl_info->supported[ARB_MAP_BUFFER_RANGE])
912             {
913                 GL_EXTCALL(glFlushMappedBufferRange(This->buffer_type_hint, start, len));
914                 checkGLcall("glFlushMappedBufferRange");
915             }
916             else if (This->flags & WINED3D_BUFFER_FLUSH)
917             {
918                 GL_EXTCALL(glFlushMappedBufferRangeAPPLE(This->buffer_type_hint, start, len));
919                 checkGLcall("glFlushMappedBufferRangeAPPLE");
920             }
921         }
922         GL_EXTCALL(glUnmapBufferARB(This->buffer_type_hint));
923         checkGLcall("glUnmapBufferARB");
924         LEAVE_GL();
925 }
926
927 /* Do not call while under the GL lock. */
928 static void STDMETHODCALLTYPE buffer_PreLoad(IWineD3DBuffer *iface)
929 {
930     struct wined3d_buffer *This = (struct wined3d_buffer *)iface;
931     IWineD3DDeviceImpl *device = This->resource.device;
932     UINT start = 0, end = 0, len = 0, vertices;
933     const struct wined3d_gl_info *gl_info;
934     struct wined3d_context *context;
935     BOOL decl_changed = FALSE;
936     unsigned int i, j;
937     BYTE *data;
938     DWORD flags = This->flags & (WINED3D_BUFFER_NOSYNC | WINED3D_BUFFER_DISCARD);
939
940     TRACE("iface %p\n", iface);
941     This->flags &= ~(WINED3D_BUFFER_NOSYNC | WINED3D_BUFFER_DISCARD);
942
943     context = context_acquire(device, NULL);
944     gl_info = context->gl_info;
945
946     if (!This->buffer_object)
947     {
948         /* TODO: Make converting independent from VBOs */
949         if (This->flags & WINED3D_BUFFER_CREATEBO)
950         {
951             buffer_create_buffer_object(This, gl_info);
952             This->flags &= ~WINED3D_BUFFER_CREATEBO;
953         }
954         else
955         {
956             /* Not doing any conversion */
957             goto end;
958         }
959     }
960
961     /* Reading the declaration makes only sense if the stateblock is finalized and the buffer bound to a stream */
962     if (device->isInDraw && This->bind_count > 0)
963     {
964         decl_changed = buffer_find_decl(This);
965         This->flags |= WINED3D_BUFFER_HASDESC;
966     }
967
968     if (!decl_changed && !(This->flags & WINED3D_BUFFER_HASDESC && buffer_is_dirty(This)))
969     {
970         context_release(context);
971         ++This->draw_count;
972         if (This->draw_count > VB_RESETDECLCHANGE) This->decl_change_count = 0;
973         if (This->draw_count > VB_RESETFULLCONVS) This->full_conversion_count = 0;
974         return;
975     }
976
977     /* If applications change the declaration over and over, reconverting all the time is a huge
978      * performance hit. So count the declaration changes and release the VBO if there are too many
979      * of them (and thus stop converting)
980      */
981     if (decl_changed)
982     {
983         ++This->decl_change_count;
984         This->draw_count = 0;
985
986         if (This->decl_change_count > VB_MAXDECLCHANGES ||
987             (This->conversion_map && (This->resource.usage & WINED3DUSAGE_DYNAMIC)))
988         {
989             FIXME("Too many declaration changes or converting dynamic buffer, stopping converting\n");
990
991             IWineD3DBuffer_UnLoad(iface);
992             This->flags &= ~WINED3D_BUFFER_CREATEBO;
993
994             /* The stream source state handler might have read the memory of the vertex buffer already
995              * and got the memory in the vbo which is not valid any longer. Dirtify the stream source
996              * to force a reload. This happens only once per changed vertexbuffer and should occur rather
997              * rarely
998              */
999             IWineD3DDeviceImpl_MarkStateDirty(device, STATE_STREAMSRC);
1000             goto end;
1001         }
1002         buffer_check_buffer_object_size(This, gl_info);
1003
1004         /* The declaration changed, reload the whole buffer */
1005         WARN("Reloading buffer because of decl change\n");
1006         buffer_clear_dirty_areas(This);
1007         if(!buffer_add_dirty_area(This, 0, 0))
1008         {
1009             ERR("buffer_add_dirty_area failed, this is not expected\n");
1010             goto end;
1011         }
1012         /* Avoid unfenced updates, we might overwrite more areas of the buffer than the application
1013          * cleared for unsynchronized updates
1014          */
1015         flags = 0;
1016     }
1017     else
1018     {
1019         /* However, it is perfectly fine to change the declaration every now and then. We don't want a game that
1020          * changes it every minute drop the VBO after VB_MAX_DECL_CHANGES minutes. So count draws without
1021          * decl changes and reset the decl change count after a specific number of them
1022          */
1023         if(buffer_is_fully_dirty(This))
1024         {
1025             ++This->full_conversion_count;
1026             if(This->full_conversion_count > VB_MAXFULLCONVERSIONS)
1027             {
1028                 FIXME("Too many full buffer conversions, stopping converting\n");
1029                 IWineD3DBuffer_UnLoad(iface);
1030                 This->flags &= ~WINED3D_BUFFER_CREATEBO;
1031                 IWineD3DDeviceImpl_MarkStateDirty(device, STATE_STREAMSRC);
1032                 goto end;
1033             }
1034         }
1035         else
1036         {
1037             ++This->draw_count;
1038             if (This->draw_count > VB_RESETDECLCHANGE) This->decl_change_count = 0;
1039             if (This->draw_count > VB_RESETFULLCONVS) This->full_conversion_count = 0;
1040         }
1041     }
1042
1043     if(This->buffer_type_hint == GL_ELEMENT_ARRAY_BUFFER_ARB)
1044     {
1045         IWineD3DDeviceImpl_MarkStateDirty(This->resource.device, STATE_INDEXBUFFER);
1046     }
1047
1048     if (!This->conversion_map)
1049     {
1050         /* That means that there is nothing to fixup. Just upload from This->resource.allocatedMemory
1051          * directly into the vbo. Do not free the system memory copy because drawPrimitive may need it if
1052          * the stride is 0, for instancing emulation, vertex blending emulation or shader emulation.
1053          */
1054         TRACE("No conversion needed\n");
1055
1056         /* Nothing to do because we locked directly into the vbo */
1057         if (!(This->flags & WINED3D_BUFFER_DOUBLEBUFFER))
1058         {
1059             context_release(context);
1060             return;
1061         }
1062
1063         buffer_direct_upload(This, context->gl_info, flags);
1064
1065         context_release(context);
1066         return;
1067     }
1068
1069     if(!(This->flags & WINED3D_BUFFER_DOUBLEBUFFER))
1070     {
1071         buffer_get_sysmem(This, gl_info);
1072     }
1073
1074     /* Now for each vertex in the buffer that needs conversion */
1075     vertices = This->resource.size / This->stride;
1076
1077     if (This->conversion_shift)
1078     {
1079         TRACE("Shifted conversion\n");
1080         data = HeapAlloc(GetProcessHeap(), 0, vertices * This->conversion_stride);
1081
1082         start = 0;
1083         len = This->resource.size;
1084         end = start + len;
1085
1086         if (This->maps[0].offset || This->maps[0].size != This->resource.size)
1087         {
1088             FIXME("Implement partial buffer load with shifted conversion\n");
1089         }
1090
1091         for (i = start / This->stride; i < min((end / This->stride) + 1, vertices); ++i)
1092         {
1093             for (j = 0; j < This->stride; ++j)
1094             {
1095                 switch(This->conversion_map[j])
1096                 {
1097                     case CONV_NONE:
1098                         data[This->conversion_stride * i + j + This->conversion_shift[j]]
1099                                 = This->resource.allocatedMemory[This->stride * i + j];
1100                         break;
1101
1102                     case CONV_FLOAT16_2:
1103                     {
1104                         float *out = (float *)(&data[This->conversion_stride * i + j + This->conversion_shift[j]]);
1105                         const WORD *in = (WORD *)(&This->resource.allocatedMemory[i * This->stride + j]);
1106
1107                         out[1] = float_16_to_32(in + 1);
1108                         out[0] = float_16_to_32(in + 0);
1109                         j += 3;    /* Skip 3 additional bytes,as a FLOAT16_2 has 4 bytes */
1110                         break;
1111                     }
1112
1113                     default:
1114                         FIXME("Unimplemented conversion %d in shifted conversion\n", This->conversion_map[j]);
1115                         break;
1116                 }
1117             }
1118         }
1119
1120         ENTER_GL();
1121         GL_EXTCALL(glBindBufferARB(This->buffer_type_hint, This->buffer_object));
1122         checkGLcall("glBindBufferARB");
1123         GL_EXTCALL(glBufferSubDataARB(This->buffer_type_hint, 0, vertices * This->conversion_stride, data));
1124         checkGLcall("glBufferSubDataARB");
1125         LEAVE_GL();
1126     }
1127     else
1128     {
1129         data = HeapAlloc(GetProcessHeap(), 0, This->resource.size);
1130
1131         while(This->modified_areas)
1132         {
1133             This->modified_areas--;
1134             start = This->maps[This->modified_areas].offset;
1135             len = This->maps[This->modified_areas].size;
1136             end = start + len;
1137
1138             memcpy(data + start, This->resource.allocatedMemory + start, end - start);
1139             for (i = start / This->stride; i < min((end / This->stride) + 1, vertices); ++i)
1140             {
1141                 for (j = 0; j < This->stride; ++j)
1142                 {
1143                     switch(This->conversion_map[j])
1144                     {
1145                         case CONV_NONE:
1146                             /* Done already */
1147                             j += 3;
1148                             break;
1149                         case CONV_D3DCOLOR:
1150                             fixup_d3dcolor((DWORD *) (data + i * This->stride + j));
1151                             j += 3;
1152                             break;
1153
1154                         case CONV_POSITIONT:
1155                             fixup_transformed_pos((float *) (data + i * This->stride + j));
1156                             j += 15;
1157                             break;
1158
1159                         case CONV_FLOAT16_2:
1160                             ERR("Did not expect FLOAT16 conversion in unshifted conversion\n");
1161                         default:
1162                             FIXME("Unimplemented conversion %d in shifted conversion\n", This->conversion_map[j]);
1163                     }
1164                 }
1165             }
1166
1167             ENTER_GL();
1168             GL_EXTCALL(glBindBufferARB(This->buffer_type_hint, This->buffer_object));
1169             checkGLcall("glBindBufferARB");
1170             GL_EXTCALL(glBufferSubDataARB(This->buffer_type_hint, start, len, data + start));
1171             checkGLcall("glBufferSubDataARB");
1172             LEAVE_GL();
1173         }
1174     }
1175
1176     HeapFree(GetProcessHeap(), 0, data);
1177
1178 end:
1179     context_release(context);
1180 }
1181
1182 static WINED3DRESOURCETYPE STDMETHODCALLTYPE buffer_GetType(IWineD3DBuffer *iface)
1183 {
1184     return resource_get_type((IWineD3DResource *)iface);
1185 }
1186
1187 /* IWineD3DBuffer methods */
1188
1189 static DWORD buffer_sanitize_flags(struct wined3d_buffer *buffer, DWORD flags)
1190 {
1191     /* Not all flags make sense together, but Windows never returns an error. Catch the
1192      * cases that could cause issues */
1193     if(flags & WINED3DLOCK_READONLY)
1194     {
1195         if(flags & WINED3DLOCK_DISCARD)
1196         {
1197             WARN("WINED3DLOCK_READONLY combined with WINED3DLOCK_DISCARD, ignoring flags\n");
1198             return 0;
1199         }
1200         if(flags & WINED3DLOCK_NOOVERWRITE)
1201         {
1202             WARN("WINED3DLOCK_READONLY combined with WINED3DLOCK_NOOVERWRITE, ignoring flags\n");
1203             return 0;
1204         }
1205     }
1206     else if((flags & (WINED3DLOCK_DISCARD | WINED3DLOCK_NOOVERWRITE)) == (WINED3DLOCK_DISCARD | WINED3DLOCK_NOOVERWRITE))
1207     {
1208         WARN("WINED3DLOCK_DISCARD and WINED3DLOCK_NOOVERWRITE used together, ignoring\n");
1209         return 0;
1210     }
1211     else if (flags & (WINED3DLOCK_DISCARD | WINED3DLOCK_NOOVERWRITE) && !(buffer->resource.usage & WINED3DUSAGE_DYNAMIC))
1212     {
1213         WARN("DISCARD or NOOVERWRITE lock on non-dynamic buffer, ignoring\n");
1214         return 0;
1215     }
1216
1217     return flags;
1218 }
1219
1220 static GLbitfield buffer_gl_map_flags(DWORD d3d_flags)
1221 {
1222     GLbitfield ret = 0;
1223
1224     if (!(d3d_flags & WINED3DLOCK_READONLY)) ret = GL_MAP_WRITE_BIT | GL_MAP_FLUSH_EXPLICIT_BIT;
1225
1226     if (d3d_flags & (WINED3DLOCK_DISCARD | WINED3DLOCK_NOOVERWRITE))
1227     {
1228         if(d3d_flags & WINED3DLOCK_DISCARD) ret |= GL_MAP_INVALIDATE_BUFFER_BIT;
1229         ret |= GL_MAP_UNSYNCHRONIZED_BIT;
1230     }
1231     else
1232     {
1233         ret |= GL_MAP_READ_BIT;
1234     }
1235
1236     return ret;
1237 }
1238
1239 static HRESULT STDMETHODCALLTYPE buffer_Map(IWineD3DBuffer *iface, UINT offset, UINT size, BYTE **data, DWORD flags)
1240 {
1241     struct wined3d_buffer *This = (struct wined3d_buffer *)iface;
1242     LONG count;
1243     BOOL dirty = buffer_is_dirty(This);
1244
1245     TRACE("iface %p, offset %u, size %u, data %p, flags %#x\n", iface, offset, size, data, flags);
1246
1247     flags = buffer_sanitize_flags(This, flags);
1248     if (!(flags & WINED3DLOCK_READONLY))
1249     {
1250         if (!buffer_add_dirty_area(This, offset, size)) return E_OUTOFMEMORY;
1251     }
1252
1253     count = InterlockedIncrement(&This->lock_count);
1254
1255     if (This->buffer_object)
1256     {
1257         if(!(This->flags & WINED3D_BUFFER_DOUBLEBUFFER))
1258         {
1259             if(count == 1)
1260             {
1261                 IWineD3DDeviceImpl *device = This->resource.device;
1262                 struct wined3d_context *context;
1263                 const struct wined3d_gl_info *gl_info;
1264
1265                 if(This->buffer_type_hint == GL_ELEMENT_ARRAY_BUFFER_ARB)
1266                 {
1267                     IWineD3DDeviceImpl_MarkStateDirty(This->resource.device, STATE_INDEXBUFFER);
1268                 }
1269
1270                 context = context_acquire(device, NULL);
1271                 gl_info = context->gl_info;
1272                 ENTER_GL();
1273                 GL_EXTCALL(glBindBufferARB(This->buffer_type_hint, This->buffer_object));
1274
1275                 if (gl_info->supported[ARB_MAP_BUFFER_RANGE])
1276                 {
1277                     GLbitfield mapflags = buffer_gl_map_flags(flags);
1278                     This->resource.allocatedMemory = GL_EXTCALL(glMapBufferRange(This->buffer_type_hint, 0,
1279                                                                                 This->resource.size, mapflags));
1280                     checkGLcall("glMapBufferRange");
1281                 }
1282                 else
1283                 {
1284                     if(This->flags & WINED3D_BUFFER_APPLESYNC)
1285                     {
1286                         LEAVE_GL();
1287                         buffer_sync_apple(This, flags, gl_info);
1288                         ENTER_GL();
1289                     }
1290                     This->resource.allocatedMemory = GL_EXTCALL(glMapBufferARB(This->buffer_type_hint, GL_READ_WRITE_ARB));
1291                     checkGLcall("glMapBufferARB");
1292                 }
1293                 LEAVE_GL();
1294
1295                 if (((DWORD_PTR) This->resource.allocatedMemory) & (RESOURCE_ALIGNMENT - 1))
1296                 {
1297                     WARN("Pointer %p is not %u byte aligned, falling back to double buffered operation\n",
1298                         This->resource.allocatedMemory, RESOURCE_ALIGNMENT);
1299
1300                     ENTER_GL();
1301                     GL_EXTCALL(glUnmapBufferARB(This->buffer_type_hint));
1302                     checkGLcall("glUnmapBufferARB");
1303                     LEAVE_GL();
1304                     This->resource.allocatedMemory = NULL;
1305
1306                     buffer_get_sysmem(This, gl_info);
1307                     TRACE("New pointer is %p\n", This->resource.allocatedMemory);
1308                 }
1309                 context_release(context);
1310             }
1311         }
1312         else
1313         {
1314             if (dirty)
1315             {
1316                 if (This->flags & WINED3D_BUFFER_NOSYNC && !(flags & WINED3DLOCK_NOOVERWRITE))
1317                 {
1318                     This->flags &= ~WINED3D_BUFFER_NOSYNC;
1319                 }
1320             }
1321             else if(flags & WINED3DLOCK_NOOVERWRITE)
1322             {
1323                 This->flags |= WINED3D_BUFFER_NOSYNC;
1324             }
1325
1326             if (flags & WINED3DLOCK_DISCARD)
1327             {
1328                 This->flags |= WINED3D_BUFFER_DISCARD;
1329             }
1330         }
1331     }
1332
1333     *data = This->resource.allocatedMemory + offset;
1334
1335     TRACE("Returning memory at %p (base %p, offset %u)\n", *data, This->resource.allocatedMemory, offset);
1336     /* TODO: check Flags compatibility with This->currentDesc.Usage (see MSDN) */
1337
1338     return WINED3D_OK;
1339 }
1340
1341 static HRESULT STDMETHODCALLTYPE buffer_Unmap(IWineD3DBuffer *iface)
1342 {
1343     struct wined3d_buffer *This = (struct wined3d_buffer *)iface;
1344     ULONG i;
1345
1346     TRACE("(%p)\n", This);
1347
1348     /* In the case that the number of Unmap calls > the
1349      * number of Map calls, d3d returns always D3D_OK.
1350      * This is also needed to prevent Map from returning garbage on
1351      * the next call (this will happen if the lock_count is < 0). */
1352     if (!This->lock_count)
1353     {
1354         TRACE("Unmap called without a previous Map call!\n");
1355         return WINED3D_OK;
1356     }
1357
1358     if (InterlockedDecrement(&This->lock_count))
1359     {
1360         /* Delay loading the buffer until everything is unlocked */
1361         TRACE("Ignoring unlock\n");
1362         return WINED3D_OK;
1363     }
1364
1365     if(!(This->flags & WINED3D_BUFFER_DOUBLEBUFFER) && This->buffer_object)
1366     {
1367         IWineD3DDeviceImpl *device = This->resource.device;
1368         const struct wined3d_gl_info *gl_info;
1369         struct wined3d_context *context;
1370
1371         if(This->buffer_type_hint == GL_ELEMENT_ARRAY_BUFFER_ARB)
1372         {
1373             IWineD3DDeviceImpl_MarkStateDirty(This->resource.device, STATE_INDEXBUFFER);
1374         }
1375
1376         context = context_acquire(device, NULL);
1377         gl_info = context->gl_info;
1378         ENTER_GL();
1379         GL_EXTCALL(glBindBufferARB(This->buffer_type_hint, This->buffer_object));
1380
1381         if (gl_info->supported[ARB_MAP_BUFFER_RANGE])
1382         {
1383             for(i = 0; i < This->modified_areas; i++)
1384             {
1385                 GL_EXTCALL(glFlushMappedBufferRange(This->buffer_type_hint,
1386                                                     This->maps[i].offset,
1387                                                     This->maps[i].size));
1388                 checkGLcall("glFlushMappedBufferRange");
1389             }
1390         }
1391         else if (This->flags & WINED3D_BUFFER_FLUSH)
1392         {
1393             for(i = 0; i < This->modified_areas; i++)
1394             {
1395                 GL_EXTCALL(glFlushMappedBufferRangeAPPLE(This->buffer_type_hint,
1396                                                          This->maps[i].offset,
1397                                                          This->maps[i].size));
1398                 checkGLcall("glFlushMappedBufferRangeAPPLE");
1399             }
1400         }
1401
1402         GL_EXTCALL(glUnmapBufferARB(This->buffer_type_hint));
1403         LEAVE_GL();
1404         context_release(context);
1405
1406         This->resource.allocatedMemory = NULL;
1407         buffer_clear_dirty_areas(This);
1408     }
1409     else if (This->flags & WINED3D_BUFFER_HASDESC)
1410     {
1411         buffer_PreLoad(iface);
1412     }
1413
1414     return WINED3D_OK;
1415 }
1416
1417 static void STDMETHODCALLTYPE buffer_GetDesc(IWineD3DBuffer *iface, WINED3DBUFFER_DESC *desc)
1418 {
1419     struct wined3d_buffer *This = (struct wined3d_buffer *)iface;
1420
1421     TRACE("(%p)\n", This);
1422
1423     desc->Type = This->resource.resourceType;
1424     desc->Usage = This->resource.usage;
1425     desc->Pool = This->resource.pool;
1426     desc->Size = This->resource.size;
1427 }
1428
1429 static const struct IWineD3DBufferVtbl wined3d_buffer_vtbl =
1430 {
1431     /* IUnknown methods */
1432     buffer_QueryInterface,
1433     buffer_AddRef,
1434     buffer_Release,
1435     /* IWineD3DBase methods */
1436     buffer_GetParent,
1437     /* IWineD3DResource methods */
1438     buffer_SetPrivateData,
1439     buffer_GetPrivateData,
1440     buffer_FreePrivateData,
1441     buffer_SetPriority,
1442     buffer_GetPriority,
1443     buffer_PreLoad,
1444     buffer_UnLoad,
1445     buffer_GetType,
1446     /* IWineD3DBuffer methods */
1447     buffer_Map,
1448     buffer_Unmap,
1449     buffer_GetDesc,
1450 };
1451
1452 HRESULT buffer_init(struct wined3d_buffer *buffer, IWineD3DDeviceImpl *device,
1453         UINT size, DWORD usage, enum wined3d_format_id format_id, WINED3DPOOL pool, GLenum bind_hint,
1454         const char *data, void *parent, const struct wined3d_parent_ops *parent_ops)
1455 {
1456     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
1457     const struct wined3d_format *format = wined3d_get_format(gl_info, format_id);
1458     HRESULT hr;
1459     BOOL dynamic_buffer_ok;
1460
1461     if (!size)
1462     {
1463         WARN("Size 0 requested, returning WINED3DERR_INVALIDCALL\n");
1464         return WINED3DERR_INVALIDCALL;
1465     }
1466
1467     buffer->vtbl = &wined3d_buffer_vtbl;
1468
1469     hr = resource_init((IWineD3DResource *)buffer, WINED3DRTYPE_BUFFER,
1470             device, size, usage, format, pool, parent, parent_ops);
1471     if (FAILED(hr))
1472     {
1473         WARN("Failed to initialize resource, hr %#x\n", hr);
1474         return hr;
1475     }
1476     buffer->buffer_type_hint = bind_hint;
1477
1478     TRACE("size %#x, usage %#x, format %s, memory @ %p, iface @ %p.\n", buffer->resource.size, buffer->resource.usage,
1479             debug_d3dformat(buffer->resource.format->id), buffer->resource.allocatedMemory, buffer);
1480
1481     /* GL_ARB_map_buffer_range is disabled for now due to numerous bugs and no gains */
1482     dynamic_buffer_ok = gl_info->supported[APPLE_FLUSH_BUFFER_RANGE];
1483
1484     /* Observations show that drawStridedSlow is faster on dynamic VBs than converting +
1485      * drawStridedFast (half-life 2 and others).
1486      *
1487      * Basically converting the vertices in the buffer is quite expensive, and observations
1488      * show that drawStridedSlow is faster than converting + uploading + drawStridedFast.
1489      * Therefore do not create a VBO for WINED3DUSAGE_DYNAMIC buffers.
1490      */
1491     if (!gl_info->supported[ARB_VERTEX_BUFFER_OBJECT])
1492     {
1493         TRACE("Not creating a vbo because GL_ARB_vertex_buffer is not supported\n");
1494     }
1495     else if(buffer->resource.pool == WINED3DPOOL_SYSTEMMEM)
1496     {
1497         TRACE("Not creating a vbo because the vertex buffer is in system memory\n");
1498     }
1499     else if(!dynamic_buffer_ok && (buffer->resource.usage & WINED3DUSAGE_DYNAMIC))
1500     {
1501         TRACE("Not creating a vbo because the buffer has dynamic usage and no GL support\n");
1502     }
1503     else
1504     {
1505         buffer->flags |= WINED3D_BUFFER_CREATEBO;
1506     }
1507
1508     if (data)
1509     {
1510         BYTE *ptr;
1511
1512         hr = IWineD3DBuffer_Map((IWineD3DBuffer *)buffer, 0, size, &ptr, 0);
1513         if (FAILED(hr))
1514         {
1515             ERR("Failed to map buffer, hr %#x\n", hr);
1516             buffer_UnLoad((IWineD3DBuffer *)buffer);
1517             resource_cleanup((IWineD3DResource *)buffer);
1518             return hr;
1519         }
1520
1521         memcpy(ptr, data, size);
1522
1523         hr = IWineD3DBuffer_Unmap((IWineD3DBuffer *)buffer);
1524         if (FAILED(hr))
1525         {
1526             ERR("Failed to unmap buffer, hr %#x\n", hr);
1527             buffer_UnLoad((IWineD3DBuffer *)buffer);
1528             resource_cleanup((IWineD3DResource *)buffer);
1529             return hr;
1530         }
1531     }
1532
1533     buffer->maps = HeapAlloc(GetProcessHeap(), 0, sizeof(*buffer->maps));
1534     if (!buffer->maps)
1535     {
1536         ERR("Out of memory\n");
1537         buffer_UnLoad((IWineD3DBuffer *)buffer);
1538         resource_cleanup((IWineD3DResource *)buffer);
1539         return E_OUTOFMEMORY;
1540     }
1541     buffer->maps_size = 1;
1542
1543     return WINED3D_OK;
1544 }