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 2009-2010 Henri Verbeet for CodeWeavers
9 * This library is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU Lesser General Public
11 * License as published by the Free Software Foundation; either
12 * version 2.1 of the License, or (at your option) any later version.
14 * This library is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 * Lesser General Public License for more details.
19 * You should have received a copy of the GNU Lesser General Public
20 * License along with this library; if not, write to the Free Software
21 * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301, USA
26 #include "wine/port.h"
28 #include "wined3d_private.h"
30 WINE_DEFAULT_DEBUG_CHANNEL(d3d);
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 */
37 static inline BOOL buffer_add_dirty_area(struct wined3d_buffer *This, UINT offset, UINT size)
39 if (!This->buffer_object) return TRUE;
41 if (This->maps_size <= This->modified_areas)
43 void *new = HeapReAlloc(GetProcessHeap(), 0, This->maps,
44 This->maps_size * 2 * sizeof(*This->maps));
47 ERR("Out of memory\n");
57 if(offset > This->resource.size || offset + size > This->resource.size)
59 WARN("Invalid range dirtified, marking entire buffer dirty\n");
61 size = This->resource.size;
63 else if(!offset && !size)
65 size = This->resource.size;
68 This->maps[This->modified_areas].offset = offset;
69 This->maps[This->modified_areas].size = size;
70 This->modified_areas++;
74 static inline void buffer_clear_dirty_areas(struct wined3d_buffer *This)
76 This->modified_areas = 0;
79 static BOOL buffer_is_dirty(const struct wined3d_buffer *buffer)
81 return !!buffer->modified_areas;
84 static BOOL buffer_is_fully_dirty(const struct wined3d_buffer *buffer)
88 for (i = 0; i < buffer->modified_areas; ++i)
90 if (!buffer->maps[i].offset && buffer->maps[i].size == buffer->resource.size)
96 /* Context activation is done by the caller */
97 static void delete_gl_buffer(struct wined3d_buffer *This, const struct wined3d_gl_info *gl_info)
99 if(!This->buffer_object) return;
102 GL_EXTCALL(glDeleteBuffersARB(1, &This->buffer_object));
103 checkGLcall("glDeleteBuffersARB");
105 This->buffer_object = 0;
109 wined3d_event_query_destroy(This->query);
112 This->flags &= ~WINED3D_BUFFER_APPLESYNC;
115 /* Context activation is done by the caller. */
116 static void buffer_create_buffer_object(struct wined3d_buffer *This, const struct wined3d_gl_info *gl_info)
118 GLenum error, gl_usage;
120 TRACE("Creating an OpenGL vertex buffer object for wined3d_buffer %p with usage %s.\n",
121 This, debug_d3dusage(This->resource.usage));
125 /* Make sure that the gl error is cleared. Do not use checkGLcall
126 * here because checkGLcall just prints a fixme and continues. However,
127 * if an error during VBO creation occurs we can fall back to non-vbo operation
128 * with full functionality(but performance loss)
130 while (gl_info->gl_ops.gl.p_glGetError() != GL_NO_ERROR);
132 /* Basically the FVF parameter passed to CreateVertexBuffer is no good.
133 * The vertex declaration from the device determines how the data in the
134 * buffer is interpreted. This means that on each draw call the buffer has
135 * to be verified to check if the rhw and color values are in the correct
138 GL_EXTCALL(glGenBuffersARB(1, &This->buffer_object));
139 error = gl_info->gl_ops.gl.p_glGetError();
140 if (!This->buffer_object || error != GL_NO_ERROR)
142 ERR("Failed to create a VBO with error %s (%#x)\n", debug_glerror(error), error);
147 if (This->buffer_type_hint == GL_ELEMENT_ARRAY_BUFFER_ARB)
148 device_invalidate_state(This->resource.device, STATE_INDEXBUFFER);
149 GL_EXTCALL(glBindBufferARB(This->buffer_type_hint, This->buffer_object));
150 error = gl_info->gl_ops.gl.p_glGetError();
151 if (error != GL_NO_ERROR)
153 ERR("Failed to bind the VBO with error %s (%#x)\n", debug_glerror(error), error);
158 /* Don't use static, because dx apps tend to update the buffer
159 * quite often even if they specify 0 usage.
161 if(This->resource.usage & WINED3DUSAGE_DYNAMIC)
163 TRACE("Gl usage = GL_STREAM_DRAW_ARB\n");
164 gl_usage = GL_STREAM_DRAW_ARB;
166 if(gl_info->supported[APPLE_FLUSH_BUFFER_RANGE])
168 GL_EXTCALL(glBufferParameteriAPPLE(This->buffer_type_hint, GL_BUFFER_FLUSHING_UNMAP_APPLE, GL_FALSE));
169 checkGLcall("glBufferParameteriAPPLE(This->buffer_type_hint, GL_BUFFER_FLUSHING_UNMAP_APPLE, GL_FALSE)");
170 This->flags |= WINED3D_BUFFER_FLUSH;
172 GL_EXTCALL(glBufferParameteriAPPLE(This->buffer_type_hint, GL_BUFFER_SERIALIZED_MODIFY_APPLE, GL_FALSE));
173 checkGLcall("glBufferParameteriAPPLE(This->buffer_type_hint, GL_BUFFER_SERIALIZED_MODIFY_APPLE, GL_FALSE)");
174 This->flags |= WINED3D_BUFFER_APPLESYNC;
176 /* No setup is needed here for GL_ARB_map_buffer_range */
180 TRACE("Gl usage = GL_DYNAMIC_DRAW_ARB\n");
181 gl_usage = GL_DYNAMIC_DRAW_ARB;
184 /* Reserve memory for the buffer. The amount of data won't change
185 * so we are safe with calling glBufferData once and
186 * calling glBufferSubData on updates. Upload the actual data in case
187 * we're not double buffering, so we can release the heap mem afterwards
189 GL_EXTCALL(glBufferDataARB(This->buffer_type_hint, This->resource.size, This->resource.allocatedMemory, gl_usage));
190 error = gl_info->gl_ops.gl.p_glGetError();
192 if (error != GL_NO_ERROR)
194 ERR("glBufferDataARB failed with error %s (%#x)\n", debug_glerror(error), error);
198 This->buffer_object_size = This->resource.size;
199 This->buffer_object_usage = gl_usage;
201 if(This->flags & WINED3D_BUFFER_DOUBLEBUFFER)
203 if(!buffer_add_dirty_area(This, 0, 0))
205 ERR("buffer_add_dirty_area failed, this is not expected\n");
211 HeapFree(GetProcessHeap(), 0, This->resource.heapMemory);
212 This->resource.allocatedMemory = NULL;
213 This->resource.heapMemory = NULL;
219 /* Clean up all vbo init, but continue because we can work without a vbo :-) */
220 ERR("Failed to create a vertex buffer object. Continuing, but performance issues may occur\n");
221 delete_gl_buffer(This, gl_info);
222 buffer_clear_dirty_areas(This);
225 static BOOL buffer_process_converted_attribute(struct wined3d_buffer *This,
226 const enum wined3d_buffer_conversion_type conversion_type,
227 const struct wined3d_stream_info_element *attrib, DWORD *stride_this_run)
234 /* Check for some valid situations which cause us pain. One is if the buffer is used for
235 * constant attributes(stride = 0), the other one is if the buffer is used on two streams
236 * with different strides. In the 2nd case we might have to drop conversion entirely,
237 * it is possible that the same bytes are once read as FLOAT2 and once as UBYTE4N.
241 FIXME("%s used with stride 0, let's hope we get the vertex stride from somewhere else\n",
242 debug_d3dformat(attrib->format->id));
244 else if(attrib->stride != *stride_this_run && *stride_this_run)
246 FIXME("Got two concurrent strides, %d and %d\n", attrib->stride, *stride_this_run);
250 *stride_this_run = attrib->stride;
251 if (This->stride != *stride_this_run)
253 /* We rely that this happens only on the first converted attribute that is found,
254 * if at all. See above check
256 TRACE("Reconverting because converted attributes occur, and the stride changed\n");
257 This->stride = *stride_this_run;
258 HeapFree(GetProcessHeap(), HEAP_ZERO_MEMORY, This->conversion_map);
259 This->conversion_map = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY,
260 sizeof(*This->conversion_map) * This->stride);
265 data = ((DWORD_PTR)attrib->data.addr) % This->stride;
266 attrib_size = attrib->format->component_count * attrib->format->component_size;
267 for (i = 0; i < attrib_size; ++i)
269 DWORD_PTR idx = (data + i) % This->stride;
270 if (This->conversion_map[idx] != conversion_type)
272 TRACE("Byte %ld in vertex changed\n", idx);
273 TRACE("It was type %d, is %d now\n", This->conversion_map[idx], conversion_type);
275 This->conversion_map[idx] = conversion_type;
282 static BOOL buffer_check_attribute(struct wined3d_buffer *This, const struct wined3d_stream_info *si,
283 UINT attrib_idx, const BOOL check_d3dcolor, const BOOL is_ffp_position, const BOOL is_ffp_color,
284 DWORD *stride_this_run)
286 const struct wined3d_stream_info_element *attrib = &si->elements[attrib_idx];
287 enum wined3d_format_id format;
290 /* Ignore attributes that do not have our vbo. After that check we can be sure that the attribute is
291 * there, on nonexistent attribs the vbo is 0.
293 if (!(si->use_map & (1 << attrib_idx))
294 || attrib->data.buffer_object != This->buffer_object)
297 format = attrib->format->id;
298 /* Look for newly appeared conversion */
299 if (check_d3dcolor && format == WINED3DFMT_B8G8R8A8_UNORM)
301 ret = buffer_process_converted_attribute(This, CONV_D3DCOLOR, attrib, stride_this_run);
303 if (!is_ffp_color) FIXME("Test for non-color fixed function WINED3DFMT_B8G8R8A8_UNORM format\n");
305 else if (is_ffp_position && si->position_transformed)
307 if (format != WINED3DFMT_R32G32B32A32_FLOAT)
309 FIXME("Unexpected format %s for transformed position.\n", debug_d3dformat(format));
313 ret = buffer_process_converted_attribute(This, CONV_POSITIONT, attrib, stride_this_run);
315 else if (This->conversion_map)
317 ret = buffer_process_converted_attribute(This, CONV_NONE, attrib, stride_this_run);
323 static BOOL buffer_find_decl(struct wined3d_buffer *This)
325 struct wined3d_device *device = This->resource.device;
326 const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
327 const struct wined3d_stream_info *si = &device->strided_streams;
328 const struct wined3d_state *state = &device->stateBlock->state;
329 UINT stride_this_run = 0;
331 BOOL support_d3dcolor = gl_info->supported[ARB_VERTEX_ARRAY_BGRA];
333 /* In d3d7 the vertex buffer declaration NEVER changes because it is stored in the d3d7 vertex buffer.
334 * Once we have our declaration there is no need to look it up again. Index buffers also never need
335 * conversion, so once the (empty) conversion structure is created don't bother checking again
337 if (This->flags & WINED3D_BUFFER_HASDESC)
339 if(This->resource.usage & WINED3DUSAGE_STATICDECL) return FALSE;
344 TRACE("Vertex shaders used, no VBO conversion is needed\n");
345 if(This->conversion_map)
347 HeapFree(GetProcessHeap(), 0, This->conversion_map);
348 This->conversion_map = NULL;
356 TRACE("Finding vertex buffer conversion information\n");
357 /* Certain declaration types need some fixups before we can pass them to
358 * opengl. This means D3DCOLOR attributes with fixed function vertex
359 * processing, FLOAT4 POSITIONT with fixed function, and FLOAT16 if
360 * GL_ARB_half_float_vertex is not supported.
362 * Note for d3d8 and d3d9:
363 * The vertex buffer FVF doesn't help with finding them, we have to use
364 * the decoded vertex declaration and pick the things that concern the
365 * current buffer. A problem with this is that this can change between
366 * draws, so we have to validate the information and reprocess the buffer
367 * if it changes, and avoid false positives for performance reasons.
368 * WineD3D doesn't even know the vertex buffer any more, it is managed
369 * by the client libraries and passed to SetStreamSource and ProcessVertices
372 * We have to distinguish between vertex shaders and fixed function to
373 * pick the way we access the strided vertex information.
375 * This code sets up a per-byte array with the size of the detected
376 * stride of the arrays in the buffer. For each byte we have a field
377 * that marks the conversion needed on this byte. For example, the
378 * following declaration with fixed function vertex processing:
386 * { POSITIONT }{ NORMAL }{ DIFFUSE }{SPECULAR }
387 * [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]
389 * Where in this example map P means 4 component position conversion, 0
390 * means no conversion, F means FLOAT16_2 conversion and C means D3DCOLOR
391 * conversion (red / blue swizzle).
393 * If we're doing conversion and the stride changes we have to reconvert
394 * the whole buffer. Note that we do not mind if the semantic changes,
395 * we only care for the conversion type. So if the NORMAL is replaced
396 * with a TEXCOORD, nothing has to be done, or if the DIFFUSE is replaced
397 * with a D3DCOLOR BLENDWEIGHT we can happily dismiss the change. Some
398 * conversion types depend on the semantic as well, for example a FLOAT4
399 * texcoord needs no conversion while a FLOAT4 positiont needs one
402 ret = buffer_check_attribute(This, si, WINED3D_FFP_POSITION,
403 TRUE, TRUE, FALSE, &stride_this_run) || ret;
404 ret = buffer_check_attribute(This, si, WINED3D_FFP_NORMAL,
405 TRUE, FALSE, FALSE, &stride_this_run) || ret;
406 ret = buffer_check_attribute(This, si, WINED3D_FFP_DIFFUSE,
407 !support_d3dcolor, FALSE, TRUE, &stride_this_run) || ret;
408 ret = buffer_check_attribute(This, si, WINED3D_FFP_SPECULAR,
409 !support_d3dcolor, FALSE, TRUE, &stride_this_run) || ret;
410 ret = buffer_check_attribute(This, si, WINED3D_FFP_TEXCOORD0,
411 TRUE, FALSE, FALSE, &stride_this_run) || ret;
412 ret = buffer_check_attribute(This, si, WINED3D_FFP_TEXCOORD1,
413 TRUE, FALSE, FALSE, &stride_this_run) || ret;
414 ret = buffer_check_attribute(This, si, WINED3D_FFP_TEXCOORD2,
415 TRUE, FALSE, FALSE, &stride_this_run) || ret;
416 ret = buffer_check_attribute(This, si, WINED3D_FFP_TEXCOORD3,
417 TRUE, FALSE, FALSE, &stride_this_run) || ret;
418 ret = buffer_check_attribute(This, si, WINED3D_FFP_TEXCOORD4,
419 TRUE, FALSE, FALSE, &stride_this_run) || ret;
420 ret = buffer_check_attribute(This, si, WINED3D_FFP_TEXCOORD5,
421 TRUE, FALSE, FALSE, &stride_this_run) || ret;
422 ret = buffer_check_attribute(This, si, WINED3D_FFP_TEXCOORD6,
423 TRUE, FALSE, FALSE, &stride_this_run) || ret;
424 ret = buffer_check_attribute(This, si, WINED3D_FFP_TEXCOORD7,
425 TRUE, FALSE, FALSE, &stride_this_run) || ret;
427 if (!stride_this_run && This->conversion_map)
430 if (!ret) ERR("no converted attributes found, old conversion map exists, and no declaration change?\n");
431 HeapFree(GetProcessHeap(), 0, This->conversion_map);
432 This->conversion_map = NULL;
436 if (ret) TRACE("Conversion information changed\n");
441 static inline void fixup_d3dcolor(DWORD *dst_color)
443 DWORD src_color = *dst_color;
445 /* Color conversion like in drawStridedSlow. watch out for little endianity
446 * If we want that stuff to work on big endian machines too we have to consider more things
448 * 0xff000000: Alpha mask
449 * 0x00ff0000: Blue mask
450 * 0x0000ff00: Green mask
451 * 0x000000ff: Red mask
454 *dst_color |= (src_color & 0xff00ff00); /* Alpha Green */
455 *dst_color |= (src_color & 0x00ff0000) >> 16; /* Red */
456 *dst_color |= (src_color & 0x000000ff) << 16; /* Blue */
459 static inline void fixup_transformed_pos(float *p)
461 /* rhw conversion like in position_float4(). */
462 if (p[3] != 1.0f && p[3] != 0.0f)
464 float w = 1.0f / p[3];
472 /* Context activation is done by the caller. */
473 void buffer_get_memory(struct wined3d_buffer *buffer, const struct wined3d_gl_info *gl_info,
474 struct wined3d_bo_address *data)
476 data->buffer_object = buffer->buffer_object;
477 if (!buffer->buffer_object)
479 if ((buffer->flags & WINED3D_BUFFER_CREATEBO) && !buffer->resource.map_count)
481 buffer_create_buffer_object(buffer, gl_info);
482 buffer->flags &= ~WINED3D_BUFFER_CREATEBO;
483 if (buffer->buffer_object)
485 data->buffer_object = buffer->buffer_object;
490 data->addr = buffer->resource.allocatedMemory;
498 ULONG CDECL wined3d_buffer_incref(struct wined3d_buffer *buffer)
500 ULONG refcount = InterlockedIncrement(&buffer->resource.ref);
502 TRACE("%p increasing refcount to %u.\n", buffer, refcount);
507 /* Context activation is done by the caller. */
508 BYTE *buffer_get_sysmem(struct wined3d_buffer *This, const struct wined3d_gl_info *gl_info)
510 /* AllocatedMemory exists if the buffer is double buffered or has no buffer object at all */
511 if(This->resource.allocatedMemory) return This->resource.allocatedMemory;
513 This->resource.heapMemory = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, This->resource.size + RESOURCE_ALIGNMENT);
514 This->resource.allocatedMemory = (BYTE *)(((ULONG_PTR)This->resource.heapMemory + (RESOURCE_ALIGNMENT - 1)) & ~(RESOURCE_ALIGNMENT - 1));
516 if (This->buffer_type_hint == GL_ELEMENT_ARRAY_BUFFER_ARB)
517 device_invalidate_state(This->resource.device, STATE_INDEXBUFFER);
520 GL_EXTCALL(glBindBufferARB(This->buffer_type_hint, This->buffer_object));
521 GL_EXTCALL(glGetBufferSubDataARB(This->buffer_type_hint, 0, This->resource.size, This->resource.allocatedMemory));
523 This->flags |= WINED3D_BUFFER_DOUBLEBUFFER;
525 return This->resource.allocatedMemory;
528 /* Do not call while under the GL lock. */
529 static void buffer_unload(struct wined3d_resource *resource)
531 struct wined3d_buffer *buffer = buffer_from_resource(resource);
533 TRACE("buffer %p.\n", buffer);
535 if (buffer->buffer_object)
537 struct wined3d_device *device = resource->device;
538 struct wined3d_context *context;
540 context = context_acquire(device, NULL);
542 /* Download the buffer, but don't permanently enable double buffering */
543 if (!(buffer->flags & WINED3D_BUFFER_DOUBLEBUFFER))
545 buffer_get_sysmem(buffer, context->gl_info);
546 buffer->flags &= ~WINED3D_BUFFER_DOUBLEBUFFER;
549 delete_gl_buffer(buffer, context->gl_info);
550 buffer->flags |= WINED3D_BUFFER_CREATEBO; /* Recreate the buffer object next load */
551 buffer_clear_dirty_areas(buffer);
553 context_release(context);
555 HeapFree(GetProcessHeap(), 0, buffer->conversion_map);
556 buffer->conversion_map = NULL;
558 buffer->conversion_stride = 0;
559 buffer->flags &= ~WINED3D_BUFFER_HASDESC;
562 resource_unload(resource);
565 /* Do not call while under the GL lock. */
566 ULONG CDECL wined3d_buffer_decref(struct wined3d_buffer *buffer)
568 ULONG refcount = InterlockedDecrement(&buffer->resource.ref);
570 TRACE("%p decreasing refcount to %u.\n", buffer, refcount);
574 buffer_unload(&buffer->resource);
575 resource_cleanup(&buffer->resource);
576 buffer->resource.parent_ops->wined3d_object_destroyed(buffer->resource.parent);
577 HeapFree(GetProcessHeap(), 0, buffer->maps);
578 HeapFree(GetProcessHeap(), 0, buffer);
584 void * CDECL wined3d_buffer_get_parent(const struct wined3d_buffer *buffer)
586 TRACE("buffer %p.\n", buffer);
588 return buffer->resource.parent;
591 DWORD CDECL wined3d_buffer_set_priority(struct wined3d_buffer *buffer, DWORD priority)
593 return resource_set_priority(&buffer->resource, priority);
596 DWORD CDECL wined3d_buffer_get_priority(const struct wined3d_buffer *buffer)
598 return resource_get_priority(&buffer->resource);
601 /* The caller provides a context and binds the buffer */
602 static void buffer_sync_apple(struct wined3d_buffer *This, DWORD flags, const struct wined3d_gl_info *gl_info)
604 enum wined3d_event_query_result ret;
606 /* No fencing needs to be done if the app promises not to overwrite
608 if (flags & WINED3D_MAP_NOOVERWRITE)
611 if (flags & WINED3D_MAP_DISCARD)
614 GL_EXTCALL(glBufferDataARB(This->buffer_type_hint, This->resource.size, NULL, This->buffer_object_usage));
615 checkGLcall("glBufferDataARB\n");
622 TRACE("Creating event query for buffer %p\n", This);
624 if (!wined3d_event_query_supported(gl_info))
626 FIXME("Event queries not supported, dropping async buffer locks.\n");
630 This->query = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*This->query));
633 ERR("Failed to allocate event query memory, dropping async buffer locks.\n");
637 /* Since we don't know about old draws a glFinish is needed once */
638 gl_info->gl_ops.gl.p_glFinish();
641 TRACE("Synchronizing buffer %p\n", This);
642 ret = wined3d_event_query_finish(This->query, This->resource.device);
645 case WINED3D_EVENT_QUERY_NOT_STARTED:
646 case WINED3D_EVENT_QUERY_OK:
650 case WINED3D_EVENT_QUERY_WRONG_THREAD:
651 WARN("Cannot synchronize buffer lock due to a thread conflict\n");
655 ERR("wined3d_event_query_finish returned %u, dropping async buffer locks\n", ret);
662 wined3d_event_query_destroy(This->query);
666 gl_info->gl_ops.gl.p_glFinish();
668 GL_EXTCALL(glBufferParameteriAPPLE(This->buffer_type_hint, GL_BUFFER_SERIALIZED_MODIFY_APPLE, GL_TRUE));
669 checkGLcall("glBufferParameteriAPPLE(This->buffer_type_hint, GL_BUFFER_SERIALIZED_MODIFY_APPLE, GL_TRUE)");
671 This->flags &= ~WINED3D_BUFFER_APPLESYNC;
674 /* The caller provides a GL context */
675 static void buffer_direct_upload(struct wined3d_buffer *This, const struct wined3d_gl_info *gl_info, DWORD flags)
678 UINT start = 0, len = 0;
682 /* This potentially invalidates the element array buffer binding, but the
683 * caller always takes care of this. */
684 GL_EXTCALL(glBindBufferARB(This->buffer_type_hint, This->buffer_object));
685 checkGLcall("glBindBufferARB");
686 if (gl_info->supported[ARB_MAP_BUFFER_RANGE])
689 mapflags = GL_MAP_WRITE_BIT | GL_MAP_FLUSH_EXPLICIT_BIT;
690 if (flags & WINED3D_BUFFER_DISCARD)
691 mapflags |= GL_MAP_INVALIDATE_BUFFER_BIT;
692 if (flags & WINED3D_BUFFER_NOSYNC)
693 mapflags |= GL_MAP_UNSYNCHRONIZED_BIT;
694 map = GL_EXTCALL(glMapBufferRange(This->buffer_type_hint, 0,
695 This->resource.size, mapflags));
696 checkGLcall("glMapBufferRange");
700 if (This->flags & WINED3D_BUFFER_APPLESYNC)
703 if (flags & WINED3D_BUFFER_DISCARD)
704 syncflags |= WINED3D_MAP_DISCARD;
705 if (flags & WINED3D_BUFFER_NOSYNC)
706 syncflags |= WINED3D_MAP_NOOVERWRITE;
708 buffer_sync_apple(This, syncflags, gl_info);
711 map = GL_EXTCALL(glMapBufferARB(This->buffer_type_hint, GL_WRITE_ONLY_ARB));
712 checkGLcall("glMapBufferARB");
717 ERR("Failed to map opengl buffer\n");
721 while (This->modified_areas)
723 This->modified_areas--;
724 start = This->maps[This->modified_areas].offset;
725 len = This->maps[This->modified_areas].size;
727 memcpy(map + start, This->resource.allocatedMemory + start, len);
729 if (gl_info->supported[ARB_MAP_BUFFER_RANGE])
731 GL_EXTCALL(glFlushMappedBufferRange(This->buffer_type_hint, start, len));
732 checkGLcall("glFlushMappedBufferRange");
734 else if (This->flags & WINED3D_BUFFER_FLUSH)
736 GL_EXTCALL(glFlushMappedBufferRangeAPPLE(This->buffer_type_hint, start, len));
737 checkGLcall("glFlushMappedBufferRangeAPPLE");
740 GL_EXTCALL(glUnmapBufferARB(This->buffer_type_hint));
741 checkGLcall("glUnmapBufferARB");
746 /* Do not call while under the GL lock. */
747 void CDECL wined3d_buffer_preload(struct wined3d_buffer *buffer)
749 DWORD flags = buffer->flags & (WINED3D_BUFFER_NOSYNC | WINED3D_BUFFER_DISCARD);
750 struct wined3d_device *device = buffer->resource.device;
751 UINT start = 0, end = 0, len = 0, vertices;
752 const struct wined3d_gl_info *gl_info;
753 struct wined3d_context *context;
754 BOOL decl_changed = FALSE;
758 TRACE("buffer %p.\n", buffer);
760 if (buffer->resource.map_count)
762 WARN("Buffer is mapped, skipping preload.\n");
766 buffer->flags &= ~(WINED3D_BUFFER_NOSYNC | WINED3D_BUFFER_DISCARD);
768 if (!buffer->buffer_object)
770 /* TODO: Make converting independent from VBOs */
771 if (buffer->flags & WINED3D_BUFFER_CREATEBO)
773 context = context_acquire(device, NULL);
774 buffer_create_buffer_object(buffer, context->gl_info);
775 context_release(context);
776 buffer->flags &= ~WINED3D_BUFFER_CREATEBO;
780 /* Not doing any conversion */
785 /* Reading the declaration makes only sense if the stateblock is finalized and the buffer bound to a stream */
786 if (device->isInDraw && buffer->resource.bind_count > 0)
788 decl_changed = buffer_find_decl(buffer);
789 buffer->flags |= WINED3D_BUFFER_HASDESC;
792 if (!decl_changed && !(buffer->flags & WINED3D_BUFFER_HASDESC && buffer_is_dirty(buffer)))
794 ++buffer->draw_count;
795 if (buffer->draw_count > VB_RESETDECLCHANGE)
796 buffer->decl_change_count = 0;
797 if (buffer->draw_count > VB_RESETFULLCONVS)
798 buffer->full_conversion_count = 0;
802 /* If applications change the declaration over and over, reconverting all the time is a huge
803 * performance hit. So count the declaration changes and release the VBO if there are too many
804 * of them (and thus stop converting)
808 ++buffer->decl_change_count;
809 buffer->draw_count = 0;
811 if (buffer->decl_change_count > VB_MAXDECLCHANGES
812 || (buffer->conversion_map && (buffer->resource.usage & WINED3DUSAGE_DYNAMIC)))
814 FIXME("Too many declaration changes or converting dynamic buffer, stopping converting\n");
816 buffer_unload(&buffer->resource);
817 buffer->flags &= ~WINED3D_BUFFER_CREATEBO;
819 /* The stream source state handler might have read the memory of
820 * the vertex buffer already and got the memory in the vbo which
821 * is not valid any longer. Dirtify the stream source to force a
822 * reload. This happens only once per changed vertexbuffer and
823 * should occur rather rarely. */
824 device_invalidate_state(device, STATE_STREAMSRC);
828 /* The declaration changed, reload the whole buffer */
829 WARN("Reloading buffer because of decl change\n");
830 buffer_clear_dirty_areas(buffer);
831 if (!buffer_add_dirty_area(buffer, 0, 0))
833 ERR("buffer_add_dirty_area failed, this is not expected\n");
836 /* Avoid unfenced updates, we might overwrite more areas of the buffer than the application
837 * cleared for unsynchronized updates
843 /* However, it is perfectly fine to change the declaration every now and then. We don't want a game that
844 * changes it every minute drop the VBO after VB_MAX_DECL_CHANGES minutes. So count draws without
845 * decl changes and reset the decl change count after a specific number of them
847 if (buffer->conversion_map && buffer_is_fully_dirty(buffer))
849 ++buffer->full_conversion_count;
850 if (buffer->full_conversion_count > VB_MAXFULLCONVERSIONS)
852 FIXME("Too many full buffer conversions, stopping converting.\n");
853 buffer_unload(&buffer->resource);
854 buffer->flags &= ~WINED3D_BUFFER_CREATEBO;
855 if (buffer->resource.bind_count)
856 device_invalidate_state(device, STATE_STREAMSRC);
862 ++buffer->draw_count;
863 if (buffer->draw_count > VB_RESETDECLCHANGE)
864 buffer->decl_change_count = 0;
865 if (buffer->draw_count > VB_RESETFULLCONVS)
866 buffer->full_conversion_count = 0;
870 if (buffer->buffer_type_hint == GL_ELEMENT_ARRAY_BUFFER_ARB)
871 device_invalidate_state(device, STATE_INDEXBUFFER);
873 if (!buffer->conversion_map)
875 /* That means that there is nothing to fixup. Just upload from
876 * buffer->resource.allocatedMemory directly into the vbo. Do not
877 * free the system memory copy because drawPrimitive may need it if
878 * the stride is 0, for instancing emulation, vertex blending
879 * emulation or shader emulation. */
880 TRACE("No conversion needed.\n");
882 /* Nothing to do because we locked directly into the vbo */
883 if (!(buffer->flags & WINED3D_BUFFER_DOUBLEBUFFER))
888 context = context_acquire(device, NULL);
889 buffer_direct_upload(buffer, context->gl_info, flags);
891 context_release(context);
895 context = context_acquire(device, NULL);
896 gl_info = context->gl_info;
898 if(!(buffer->flags & WINED3D_BUFFER_DOUBLEBUFFER))
900 buffer_get_sysmem(buffer, gl_info);
903 /* Now for each vertex in the buffer that needs conversion */
904 vertices = buffer->resource.size / buffer->stride;
906 data = HeapAlloc(GetProcessHeap(), 0, buffer->resource.size);
908 while(buffer->modified_areas)
910 buffer->modified_areas--;
911 start = buffer->maps[buffer->modified_areas].offset;
912 len = buffer->maps[buffer->modified_areas].size;
915 memcpy(data + start, buffer->resource.allocatedMemory + start, end - start);
916 for (i = start / buffer->stride; i < min((end / buffer->stride) + 1, vertices); ++i)
918 for (j = 0; j < buffer->stride; ++j)
920 switch (buffer->conversion_map[j])
927 fixup_d3dcolor((DWORD *) (data + i * buffer->stride + j));
932 fixup_transformed_pos((float *) (data + i * buffer->stride + j));
936 FIXME("Unimplemented conversion %d in shifted conversion\n", buffer->conversion_map[j]);
942 GL_EXTCALL(glBindBufferARB(buffer->buffer_type_hint, buffer->buffer_object));
943 checkGLcall("glBindBufferARB");
944 GL_EXTCALL(glBufferSubDataARB(buffer->buffer_type_hint, start, len, data + start));
945 checkGLcall("glBufferSubDataARB");
949 HeapFree(GetProcessHeap(), 0, data);
950 context_release(context);
953 static DWORD buffer_sanitize_flags(const struct wined3d_buffer *buffer, DWORD flags)
955 /* Not all flags make sense together, but Windows never returns an error.
956 * Catch the cases that could cause issues. */
957 if (flags & WINED3D_MAP_READONLY)
959 if (flags & WINED3D_MAP_DISCARD)
961 WARN("WINED3D_MAP_READONLY combined with WINED3D_MAP_DISCARD, ignoring flags.\n");
964 if (flags & WINED3D_MAP_NOOVERWRITE)
966 WARN("WINED3D_MAP_READONLY combined with WINED3D_MAP_NOOVERWRITE, ignoring flags.\n");
970 else if ((flags & (WINED3D_MAP_DISCARD | WINED3D_MAP_NOOVERWRITE))
971 == (WINED3D_MAP_DISCARD | WINED3D_MAP_NOOVERWRITE))
973 WARN("WINED3D_MAP_DISCARD and WINED3D_MAP_NOOVERWRITE used together, ignoring.\n");
976 else if (flags & (WINED3D_MAP_DISCARD | WINED3D_MAP_NOOVERWRITE)
977 && !(buffer->resource.usage & WINED3DUSAGE_DYNAMIC))
979 WARN("DISCARD or NOOVERWRITE map on non-dynamic buffer, ignoring.\n");
986 static GLbitfield buffer_gl_map_flags(DWORD d3d_flags)
990 if (!(d3d_flags & WINED3D_MAP_READONLY))
991 ret |= GL_MAP_WRITE_BIT | GL_MAP_FLUSH_EXPLICIT_BIT;
992 if (!(d3d_flags & (WINED3D_MAP_DISCARD | WINED3D_MAP_NOOVERWRITE)))
993 ret |= GL_MAP_READ_BIT;
995 if (d3d_flags & WINED3D_MAP_DISCARD)
996 ret |= GL_MAP_INVALIDATE_BUFFER_BIT;
997 if (d3d_flags & WINED3D_MAP_NOOVERWRITE)
998 ret |= GL_MAP_UNSYNCHRONIZED_BIT;
1003 struct wined3d_resource * CDECL wined3d_buffer_get_resource(struct wined3d_buffer *buffer)
1005 TRACE("buffer %p.\n", buffer);
1007 return &buffer->resource;
1010 HRESULT CDECL wined3d_buffer_map(struct wined3d_buffer *buffer, UINT offset, UINT size, BYTE **data, DWORD flags)
1012 BOOL dirty = buffer_is_dirty(buffer);
1015 TRACE("buffer %p, offset %u, size %u, data %p, flags %#x\n", buffer, offset, size, data, flags);
1017 flags = buffer_sanitize_flags(buffer, flags);
1018 if (!(flags & WINED3D_MAP_READONLY))
1020 if (flags & WINED3D_MAP_DISCARD)
1022 /* DISCARD invalidates the entire buffer, regardless of the
1023 * specified offset and size. Some applications also depend on the
1024 * entire buffer being uploaded in that case. Two such
1025 * applications are Port Royale and Darkstar One. */
1026 if (!buffer_add_dirty_area(buffer, 0, 0))
1027 return E_OUTOFMEMORY;
1031 if (!buffer_add_dirty_area(buffer, offset, size))
1032 return E_OUTOFMEMORY;
1036 count = ++buffer->resource.map_count;
1038 if (buffer->buffer_object)
1040 if (!(buffer->flags & WINED3D_BUFFER_DOUBLEBUFFER))
1044 struct wined3d_device *device = buffer->resource.device;
1045 struct wined3d_context *context;
1046 const struct wined3d_gl_info *gl_info;
1048 context = context_acquire(device, NULL);
1049 gl_info = context->gl_info;
1053 if (buffer->buffer_type_hint == GL_ELEMENT_ARRAY_BUFFER_ARB)
1054 context_invalidate_state(context, STATE_INDEXBUFFER);
1055 GL_EXTCALL(glBindBufferARB(buffer->buffer_type_hint, buffer->buffer_object));
1057 if (gl_info->supported[ARB_MAP_BUFFER_RANGE])
1059 GLbitfield mapflags = buffer_gl_map_flags(flags);
1060 buffer->resource.allocatedMemory = GL_EXTCALL(glMapBufferRange(buffer->buffer_type_hint,
1061 0, buffer->resource.size, mapflags));
1062 checkGLcall("glMapBufferRange");
1066 if (buffer->flags & WINED3D_BUFFER_APPLESYNC)
1069 buffer_sync_apple(buffer, flags, gl_info);
1072 buffer->resource.allocatedMemory = GL_EXTCALL(glMapBufferARB(buffer->buffer_type_hint,
1073 GL_READ_WRITE_ARB));
1074 checkGLcall("glMapBufferARB");
1078 if (((DWORD_PTR)buffer->resource.allocatedMemory) & (RESOURCE_ALIGNMENT - 1))
1080 WARN("Pointer %p is not %u byte aligned.\n", buffer->resource.allocatedMemory, RESOURCE_ALIGNMENT);
1083 GL_EXTCALL(glUnmapBufferARB(buffer->buffer_type_hint));
1084 checkGLcall("glUnmapBufferARB");
1086 buffer->resource.allocatedMemory = NULL;
1088 if (buffer->resource.usage & WINED3DUSAGE_DYNAMIC)
1090 /* The extra copy is more expensive than not using VBOs at
1091 * all on the Nvidia Linux driver, which is the only driver
1092 * that returns unaligned pointers
1094 TRACE("Dynamic buffer, dropping VBO\n");
1095 buffer_unload(&buffer->resource);
1096 buffer->flags &= ~WINED3D_BUFFER_CREATEBO;
1097 if (buffer->resource.bind_count)
1098 device_invalidate_state(device, STATE_STREAMSRC);
1102 TRACE("Falling back to doublebuffered operation\n");
1103 buffer_get_sysmem(buffer, gl_info);
1105 TRACE("New pointer is %p.\n", buffer->resource.allocatedMemory);
1107 context_release(context);
1114 if (buffer->flags & WINED3D_BUFFER_NOSYNC && !(flags & WINED3D_MAP_NOOVERWRITE))
1116 buffer->flags &= ~WINED3D_BUFFER_NOSYNC;
1119 else if(flags & WINED3D_MAP_NOOVERWRITE)
1121 buffer->flags |= WINED3D_BUFFER_NOSYNC;
1124 if (flags & WINED3D_MAP_DISCARD)
1126 buffer->flags |= WINED3D_BUFFER_DISCARD;
1131 *data = buffer->resource.allocatedMemory + offset;
1133 TRACE("Returning memory at %p (base %p, offset %u).\n", *data, buffer->resource.allocatedMemory, offset);
1134 /* TODO: check Flags compatibility with buffer->currentDesc.Usage (see MSDN) */
1139 void CDECL wined3d_buffer_unmap(struct wined3d_buffer *buffer)
1143 TRACE("buffer %p.\n", buffer);
1145 /* In the case that the number of Unmap calls > the
1146 * number of Map calls, d3d returns always D3D_OK.
1147 * This is also needed to prevent Map from returning garbage on
1148 * the next call (this will happen if the lock_count is < 0). */
1149 if (!buffer->resource.map_count)
1151 WARN("Unmap called without a previous map call.\n");
1155 if (--buffer->resource.map_count)
1157 /* Delay loading the buffer until everything is unlocked */
1158 TRACE("Ignoring unmap.\n");
1162 if (!(buffer->flags & WINED3D_BUFFER_DOUBLEBUFFER) && buffer->buffer_object)
1164 struct wined3d_device *device = buffer->resource.device;
1165 const struct wined3d_gl_info *gl_info;
1166 struct wined3d_context *context;
1168 context = context_acquire(device, NULL);
1169 gl_info = context->gl_info;
1173 if (buffer->buffer_type_hint == GL_ELEMENT_ARRAY_BUFFER_ARB)
1174 context_invalidate_state(context, STATE_INDEXBUFFER);
1175 GL_EXTCALL(glBindBufferARB(buffer->buffer_type_hint, buffer->buffer_object));
1177 if (gl_info->supported[ARB_MAP_BUFFER_RANGE])
1179 for (i = 0; i < buffer->modified_areas; ++i)
1181 GL_EXTCALL(glFlushMappedBufferRange(buffer->buffer_type_hint,
1182 buffer->maps[i].offset, buffer->maps[i].size));
1183 checkGLcall("glFlushMappedBufferRange");
1186 else if (buffer->flags & WINED3D_BUFFER_FLUSH)
1188 for (i = 0; i < buffer->modified_areas; ++i)
1190 GL_EXTCALL(glFlushMappedBufferRangeAPPLE(buffer->buffer_type_hint,
1191 buffer->maps[i].offset, buffer->maps[i].size));
1192 checkGLcall("glFlushMappedBufferRangeAPPLE");
1196 GL_EXTCALL(glUnmapBufferARB(buffer->buffer_type_hint));
1198 if (wined3d_settings.strict_draw_ordering)
1199 gl_info->gl_ops.gl.p_glFlush(); /* Flush to ensure ordering across contexts. */
1200 context_release(context);
1202 buffer->resource.allocatedMemory = NULL;
1203 buffer_clear_dirty_areas(buffer);
1205 else if (buffer->flags & WINED3D_BUFFER_HASDESC)
1207 wined3d_buffer_preload(buffer);
1211 static const struct wined3d_resource_ops buffer_resource_ops =
1216 static HRESULT buffer_init(struct wined3d_buffer *buffer, struct wined3d_device *device,
1217 UINT size, DWORD usage, enum wined3d_format_id format_id, enum wined3d_pool pool, GLenum bind_hint,
1218 const char *data, void *parent, const struct wined3d_parent_ops *parent_ops)
1220 const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
1221 const struct wined3d_format *format = wined3d_get_format(gl_info, format_id);
1223 BOOL dynamic_buffer_ok;
1227 WARN("Size 0 requested, returning WINED3DERR_INVALIDCALL\n");
1228 return WINED3DERR_INVALIDCALL;
1231 hr = resource_init(&buffer->resource, device, WINED3D_RTYPE_BUFFER, format,
1232 WINED3D_MULTISAMPLE_NONE, 0, usage, pool, size, 1, 1, size,
1233 parent, parent_ops, &buffer_resource_ops);
1236 WARN("Failed to initialize resource, hr %#x\n", hr);
1239 buffer->buffer_type_hint = bind_hint;
1241 TRACE("size %#x, usage %#x, format %s, memory @ %p, iface @ %p.\n", buffer->resource.size, buffer->resource.usage,
1242 debug_d3dformat(buffer->resource.format->id), buffer->resource.allocatedMemory, buffer);
1244 dynamic_buffer_ok = gl_info->supported[APPLE_FLUSH_BUFFER_RANGE] || gl_info->supported[ARB_MAP_BUFFER_RANGE];
1246 /* Observations show that drawStridedSlow is faster on dynamic VBs than converting +
1247 * drawStridedFast (half-life 2 and others).
1249 * Basically converting the vertices in the buffer is quite expensive, and observations
1250 * show that drawStridedSlow is faster than converting + uploading + drawStridedFast.
1251 * Therefore do not create a VBO for WINED3DUSAGE_DYNAMIC buffers.
1253 if (!gl_info->supported[ARB_VERTEX_BUFFER_OBJECT])
1255 TRACE("Not creating a vbo because GL_ARB_vertex_buffer is not supported\n");
1257 else if(buffer->resource.pool == WINED3D_POOL_SYSTEM_MEM)
1259 TRACE("Not creating a vbo because the vertex buffer is in system memory\n");
1261 else if(!dynamic_buffer_ok && (buffer->resource.usage & WINED3DUSAGE_DYNAMIC))
1263 TRACE("Not creating a vbo because the buffer has dynamic usage and no GL support\n");
1267 buffer->flags |= WINED3D_BUFFER_CREATEBO;
1274 hr = wined3d_buffer_map(buffer, 0, size, &ptr, 0);
1277 ERR("Failed to map buffer, hr %#x\n", hr);
1278 buffer_unload(&buffer->resource);
1279 resource_cleanup(&buffer->resource);
1283 memcpy(ptr, data, size);
1285 wined3d_buffer_unmap(buffer);
1288 buffer->maps = HeapAlloc(GetProcessHeap(), 0, sizeof(*buffer->maps));
1291 ERR("Out of memory\n");
1292 buffer_unload(&buffer->resource);
1293 resource_cleanup(&buffer->resource);
1294 return E_OUTOFMEMORY;
1296 buffer->maps_size = 1;
1301 HRESULT CDECL wined3d_buffer_create(struct wined3d_device *device, struct wined3d_buffer_desc *desc, const void *data,
1302 void *parent, const struct wined3d_parent_ops *parent_ops, struct wined3d_buffer **buffer)
1304 struct wined3d_buffer *object;
1307 TRACE("device %p, desc %p, data %p, parent %p, buffer %p\n", device, desc, data, parent, buffer);
1309 object = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*object));
1312 ERR("Failed to allocate memory\n");
1313 return E_OUTOFMEMORY;
1316 FIXME("Ignoring access flags (pool)\n");
1318 hr = buffer_init(object, device, desc->byte_width, desc->usage, WINED3DFMT_UNKNOWN,
1319 WINED3D_POOL_MANAGED, GL_ARRAY_BUFFER_ARB, data, parent, parent_ops);
1322 WARN("Failed to initialize buffer, hr %#x.\n", hr);
1323 HeapFree(GetProcessHeap(), 0, object);
1326 object->desc = *desc;
1328 TRACE("Created buffer %p.\n", object);
1335 HRESULT CDECL wined3d_buffer_create_vb(struct wined3d_device *device, UINT size, DWORD usage, enum wined3d_pool pool,
1336 void *parent, const struct wined3d_parent_ops *parent_ops, struct wined3d_buffer **buffer)
1338 struct wined3d_buffer *object;
1341 TRACE("device %p, size %u, usage %#x, pool %#x, parent %p, parent_ops %p, buffer %p.\n",
1342 device, size, usage, pool, parent, parent_ops, buffer);
1344 if (pool == WINED3D_POOL_SCRATCH)
1346 /* The d3d9 tests shows that this is not allowed. It doesn't make much
1347 * sense anyway, SCRATCH buffers wouldn't be usable anywhere. */
1348 WARN("Vertex buffer in WINED3D_POOL_SCRATCH requested, returning WINED3DERR_INVALIDCALL.\n");
1350 return WINED3DERR_INVALIDCALL;
1353 object = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*object));
1356 ERR("Out of memory\n");
1358 return WINED3DERR_OUTOFVIDEOMEMORY;
1361 hr = buffer_init(object, device, size, usage, WINED3DFMT_VERTEXDATA,
1362 pool, GL_ARRAY_BUFFER_ARB, NULL, parent, parent_ops);
1365 WARN("Failed to initialize buffer, hr %#x.\n", hr);
1366 HeapFree(GetProcessHeap(), 0, object);
1370 TRACE("Created buffer %p.\n", object);
1376 HRESULT CDECL wined3d_buffer_create_ib(struct wined3d_device *device, UINT size, DWORD usage, enum wined3d_pool pool,
1377 void *parent, const struct wined3d_parent_ops *parent_ops, struct wined3d_buffer **buffer)
1379 struct wined3d_buffer *object;
1382 TRACE("device %p, size %u, usage %#x, pool %#x, parent %p, parent_ops %p, buffer %p.\n",
1383 device, size, usage, pool, parent, parent_ops, buffer);
1385 object = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*object));
1388 ERR("Out of memory\n");
1390 return WINED3DERR_OUTOFVIDEOMEMORY;
1393 hr = buffer_init(object, device, size, usage | WINED3DUSAGE_STATICDECL,
1394 WINED3DFMT_UNKNOWN, pool, GL_ELEMENT_ARRAY_BUFFER_ARB, NULL,
1395 parent, parent_ops);
1398 WARN("Failed to initialize buffer, hr %#x\n", hr);
1399 HeapFree(GetProcessHeap(), 0, object);
1403 TRACE("Created buffer %p.\n", object);