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 inline BOOL buffer_is_dirty(struct wined3d_buffer *This)
81 return !!This->modified_areas;
84 static inline BOOL buffer_is_fully_dirty(struct wined3d_buffer *This)
88 for(i = 0; i < This->modified_areas; i++)
90 if (!This->maps[i].offset && This->maps[i].size == This->resource.size)
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)
101 if(!This->buffer_object) return;
104 GL_EXTCALL(glDeleteBuffersARB(1, &This->buffer_object));
105 checkGLcall("glDeleteBuffersARB");
107 This->buffer_object = 0;
111 wined3d_event_query_destroy(This->query);
114 This->flags &= ~WINED3D_BUFFER_APPLESYNC;
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)
120 GLenum error, gl_usage;
122 TRACE("Creating an OpenGL vertex buffer object for IWineD3DVertexBuffer %p Usage(%s)\n",
123 This, debug_d3dusage(This->resource.usage));
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)
132 while (glGetError() != GL_NO_ERROR);
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.
141 GL_EXTCALL(glGenBuffersARB(1, &This->buffer_object));
142 error = glGetError();
143 if (!This->buffer_object || error != GL_NO_ERROR)
145 ERR("Failed to create a VBO with error %s (%#x)\n", debug_glerror(error), error);
150 if (This->buffer_type_hint == GL_ELEMENT_ARRAY_BUFFER_ARB)
151 device_invalidate_state(This->resource.device, STATE_INDEXBUFFER);
152 GL_EXTCALL(glBindBufferARB(This->buffer_type_hint, This->buffer_object));
153 error = glGetError();
154 if (error != GL_NO_ERROR)
156 ERR("Failed to bind the VBO with error %s (%#x)\n", debug_glerror(error), error);
161 /* Don't use static, because dx apps tend to update the buffer
162 * quite often even if they specify 0 usage.
164 if(This->resource.usage & WINED3DUSAGE_DYNAMIC)
166 TRACE("Gl usage = GL_STREAM_DRAW_ARB\n");
167 gl_usage = GL_STREAM_DRAW_ARB;
169 if(gl_info->supported[APPLE_FLUSH_BUFFER_RANGE])
171 GL_EXTCALL(glBufferParameteriAPPLE(This->buffer_type_hint, GL_BUFFER_FLUSHING_UNMAP_APPLE, GL_FALSE));
172 checkGLcall("glBufferParameteriAPPLE(This->buffer_type_hint, GL_BUFFER_FLUSHING_UNMAP_APPLE, GL_FALSE)");
173 This->flags |= WINED3D_BUFFER_FLUSH;
175 GL_EXTCALL(glBufferParameteriAPPLE(This->buffer_type_hint, GL_BUFFER_SERIALIZED_MODIFY_APPLE, GL_FALSE));
176 checkGLcall("glBufferParameteriAPPLE(This->buffer_type_hint, GL_BUFFER_SERIALIZED_MODIFY_APPLE, GL_FALSE)");
177 This->flags |= WINED3D_BUFFER_APPLESYNC;
179 /* No setup is needed here for GL_ARB_map_buffer_range */
183 TRACE("Gl usage = GL_DYNAMIC_DRAW_ARB\n");
184 gl_usage = GL_DYNAMIC_DRAW_ARB;
187 /* Reserve memory for the buffer. The amount of data won't change
188 * so we are safe with calling glBufferData once and
189 * calling glBufferSubData on updates. Upload the actual data in case
190 * we're not double buffering, so we can release the heap mem afterwards
192 GL_EXTCALL(glBufferDataARB(This->buffer_type_hint, This->resource.size, This->resource.allocatedMemory, gl_usage));
193 error = glGetError();
195 if (error != GL_NO_ERROR)
197 ERR("glBufferDataARB failed with error %s (%#x)\n", debug_glerror(error), error);
201 This->buffer_object_size = This->resource.size;
202 This->buffer_object_usage = gl_usage;
204 if(This->flags & WINED3D_BUFFER_DOUBLEBUFFER)
206 if(!buffer_add_dirty_area(This, 0, 0))
208 ERR("buffer_add_dirty_area failed, this is not expected\n");
214 HeapFree(GetProcessHeap(), 0, This->resource.heapMemory);
215 This->resource.allocatedMemory = NULL;
216 This->resource.heapMemory = NULL;
222 /* Clean up all vbo init, but continue because we can work without a vbo :-) */
223 ERR("Failed to create a vertex buffer object. Continuing, but performance issues may occur\n");
224 delete_gl_buffer(This, gl_info);
225 buffer_clear_dirty_areas(This);
228 static BOOL buffer_process_converted_attribute(struct wined3d_buffer *This,
229 const enum wined3d_buffer_conversion_type conversion_type,
230 const struct wined3d_stream_info_element *attrib, DWORD *stride_this_run)
232 DWORD offset = This->resource.device->stateBlock->state.streams[attrib->stream_idx].offset;
238 /* Check for some valid situations which cause us pain. One is if the buffer is used for
239 * constant attributes(stride = 0), the other one is if the buffer is used on two streams
240 * with different strides. In the 2nd case we might have to drop conversion entirely,
241 * it is possible that the same bytes are once read as FLOAT2 and once as UBYTE4N.
245 FIXME("%s used with stride 0, let's hope we get the vertex stride from somewhere else\n",
246 debug_d3dformat(attrib->format->id));
248 else if(attrib->stride != *stride_this_run && *stride_this_run)
250 FIXME("Got two concurrent strides, %d and %d\n", attrib->stride, *stride_this_run);
254 *stride_this_run = attrib->stride;
255 if (This->stride != *stride_this_run)
257 /* We rely that this happens only on the first converted attribute that is found,
258 * if at all. See above check
260 TRACE("Reconverting because converted attributes occur, and the stride changed\n");
261 This->stride = *stride_this_run;
262 HeapFree(GetProcessHeap(), HEAP_ZERO_MEMORY, This->conversion_map);
263 This->conversion_map = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY,
264 sizeof(*This->conversion_map) * This->stride);
269 data = (((DWORD_PTR)attrib->data) + offset) % This->stride;
270 attrib_size = attrib->format->component_count * attrib->format->component_size;
271 for (i = 0; i < attrib_size; ++i)
273 DWORD_PTR idx = (data + i) % This->stride;
274 if (This->conversion_map[idx] != conversion_type)
276 TRACE("Byte %ld in vertex changed\n", idx);
277 TRACE("It was type %d, is %d now\n", This->conversion_map[idx], conversion_type);
279 This->conversion_map[idx] = conversion_type;
286 static BOOL buffer_check_attribute(struct wined3d_buffer *This, const struct wined3d_stream_info *si,
287 UINT attrib_idx, const BOOL check_d3dcolor, const BOOL is_ffp_position, const BOOL is_ffp_color,
288 DWORD *stride_this_run)
290 const struct wined3d_stream_info_element *attrib = &si->elements[attrib_idx];
291 enum wined3d_format_id format;
294 /* Ignore attributes that do not have our vbo. After that check we can be sure that the attribute is
295 * there, on nonexistent attribs the vbo is 0.
297 if (!(si->use_map & (1 << attrib_idx))
298 || attrib->buffer_object != This->buffer_object)
301 format = attrib->format->id;
302 /* Look for newly appeared conversion */
303 if (check_d3dcolor && format == WINED3DFMT_B8G8R8A8_UNORM)
305 ret = buffer_process_converted_attribute(This, CONV_D3DCOLOR, attrib, stride_this_run);
307 if (!is_ffp_color) FIXME("Test for non-color fixed function WINED3DFMT_B8G8R8A8_UNORM format\n");
309 else if (is_ffp_position && format == WINED3DFMT_R32G32B32A32_FLOAT)
311 ret = buffer_process_converted_attribute(This, CONV_POSITIONT, attrib, stride_this_run);
313 else if (This->conversion_map)
315 ret = buffer_process_converted_attribute(This, CONV_NONE, attrib, stride_this_run);
321 static BOOL buffer_find_decl(struct wined3d_buffer *This)
323 struct wined3d_device *device = This->resource.device;
324 const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
325 const struct wined3d_stream_info *si = &device->strided_streams;
326 const struct wined3d_state *state = &device->stateBlock->state;
327 UINT stride_this_run = 0;
329 BOOL support_d3dcolor = gl_info->supported[ARB_VERTEX_ARRAY_BGRA];
331 /* In d3d7 the vertex buffer declaration NEVER changes because it is stored in the d3d7 vertex buffer.
332 * Once we have our declaration there is no need to look it up again. Index buffers also never need
333 * conversion, so once the (empty) conversion structure is created don't bother checking again
335 if (This->flags & WINED3D_BUFFER_HASDESC)
337 if(This->resource.usage & WINED3DUSAGE_STATICDECL) return FALSE;
342 TRACE("Vertex shaders used, no VBO conversion is needed\n");
343 if(This->conversion_map)
345 HeapFree(GetProcessHeap(), 0, This->conversion_map);
346 This->conversion_map = NULL;
354 TRACE("Finding vertex buffer conversion information\n");
355 /* Certain declaration types need some fixups before we can pass them to
356 * opengl. This means D3DCOLOR attributes with fixed function vertex
357 * processing, FLOAT4 POSITIONT with fixed function, and FLOAT16 if
358 * GL_ARB_half_float_vertex is not supported.
360 * Note for d3d8 and d3d9:
361 * The vertex buffer FVF doesn't help with finding them, we have to use
362 * the decoded vertex declaration and pick the things that concern the
363 * current buffer. A problem with this is that this can change between
364 * draws, so we have to validate the information and reprocess the buffer
365 * if it changes, and avoid false positives for performance reasons.
366 * WineD3D doesn't even know the vertex buffer any more, it is managed
367 * by the client libraries and passed to SetStreamSource and ProcessVertices
370 * We have to distinguish between vertex shaders and fixed function to
371 * pick the way we access the strided vertex information.
373 * This code sets up a per-byte array with the size of the detected
374 * stride of the arrays in the buffer. For each byte we have a field
375 * that marks the conversion needed on this byte. For example, the
376 * following declaration with fixed function vertex processing:
384 * { POSITIONT }{ NORMAL }{ DIFFUSE }{SPECULAR }
385 * [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]
387 * Where in this example map P means 4 component position conversion, 0
388 * means no conversion, F means FLOAT16_2 conversion and C means D3DCOLOR
389 * conversion (red / blue swizzle).
391 * If we're doing conversion and the stride changes we have to reconvert
392 * the whole buffer. Note that we do not mind if the semantic changes,
393 * we only care for the conversion type. So if the NORMAL is replaced
394 * with a TEXCOORD, nothing has to be done, or if the DIFFUSE is replaced
395 * with a D3DCOLOR BLENDWEIGHT we can happily dismiss the change. Some
396 * conversion types depend on the semantic as well, for example a FLOAT4
397 * texcoord needs no conversion while a FLOAT4 positiont needs one
400 ret = buffer_check_attribute(This, si, WINED3D_FFP_POSITION,
401 TRUE, TRUE, FALSE, &stride_this_run) || ret;
402 ret = buffer_check_attribute(This, si, WINED3D_FFP_NORMAL,
403 TRUE, FALSE, FALSE, &stride_this_run) || ret;
404 ret = buffer_check_attribute(This, si, WINED3D_FFP_DIFFUSE,
405 !support_d3dcolor, FALSE, TRUE, &stride_this_run) || ret;
406 ret = buffer_check_attribute(This, si, WINED3D_FFP_SPECULAR,
407 !support_d3dcolor, FALSE, TRUE, &stride_this_run) || ret;
408 ret = buffer_check_attribute(This, si, WINED3D_FFP_TEXCOORD0,
409 TRUE, FALSE, FALSE, &stride_this_run) || ret;
410 ret = buffer_check_attribute(This, si, WINED3D_FFP_TEXCOORD1,
411 TRUE, FALSE, FALSE, &stride_this_run) || ret;
412 ret = buffer_check_attribute(This, si, WINED3D_FFP_TEXCOORD2,
413 TRUE, FALSE, FALSE, &stride_this_run) || ret;
414 ret = buffer_check_attribute(This, si, WINED3D_FFP_TEXCOORD3,
415 TRUE, FALSE, FALSE, &stride_this_run) || ret;
416 ret = buffer_check_attribute(This, si, WINED3D_FFP_TEXCOORD4,
417 TRUE, FALSE, FALSE, &stride_this_run) || ret;
418 ret = buffer_check_attribute(This, si, WINED3D_FFP_TEXCOORD5,
419 TRUE, FALSE, FALSE, &stride_this_run) || ret;
420 ret = buffer_check_attribute(This, si, WINED3D_FFP_TEXCOORD6,
421 TRUE, FALSE, FALSE, &stride_this_run) || ret;
422 ret = buffer_check_attribute(This, si, WINED3D_FFP_TEXCOORD7,
423 TRUE, FALSE, FALSE, &stride_this_run) || ret;
425 if (!stride_this_run && This->conversion_map)
428 if (!ret) ERR("no converted attributes found, old conversion map exists, and no declaration change?\n");
429 HeapFree(GetProcessHeap(), 0, This->conversion_map);
430 This->conversion_map = NULL;
434 if (ret) TRACE("Conversion information changed\n");
439 static inline void fixup_d3dcolor(DWORD *dst_color)
441 DWORD src_color = *dst_color;
443 /* Color conversion like in drawStridedSlow. watch out for little endianity
444 * If we want that stuff to work on big endian machines too we have to consider more things
446 * 0xff000000: Alpha mask
447 * 0x00ff0000: Blue mask
448 * 0x0000ff00: Green mask
449 * 0x000000ff: Red mask
452 *dst_color |= (src_color & 0xff00ff00); /* Alpha Green */
453 *dst_color |= (src_color & 0x00ff0000) >> 16; /* Red */
454 *dst_color |= (src_color & 0x000000ff) << 16; /* Blue */
457 static inline void fixup_transformed_pos(float *p)
459 /* rhw conversion like in position_float4(). */
460 if (p[3] != 1.0f && p[3] != 0.0f)
462 float w = 1.0f / p[3];
470 /* Context activation is done by the caller. */
471 const BYTE *buffer_get_memory(struct wined3d_buffer *buffer,
472 const struct wined3d_gl_info *gl_info, GLuint *buffer_object)
474 *buffer_object = buffer->buffer_object;
475 if (!buffer->buffer_object)
477 if (buffer->flags & WINED3D_BUFFER_CREATEBO)
479 buffer_create_buffer_object(buffer, gl_info);
480 buffer->flags &= ~WINED3D_BUFFER_CREATEBO;
481 if (buffer->buffer_object)
483 *buffer_object = buffer->buffer_object;
487 return buffer->resource.allocatedMemory;
495 ULONG CDECL wined3d_buffer_incref(struct wined3d_buffer *buffer)
497 ULONG refcount = InterlockedIncrement(&buffer->resource.ref);
499 TRACE("%p increasing refcount to %u.\n", buffer, refcount);
504 /* Context activation is done by the caller. */
505 BYTE *buffer_get_sysmem(struct wined3d_buffer *This, const struct wined3d_gl_info *gl_info)
507 /* AllocatedMemory exists if the buffer is double buffered or has no buffer object at all */
508 if(This->resource.allocatedMemory) return This->resource.allocatedMemory;
510 This->resource.heapMemory = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, This->resource.size + RESOURCE_ALIGNMENT);
511 This->resource.allocatedMemory = (BYTE *)(((ULONG_PTR)This->resource.heapMemory + (RESOURCE_ALIGNMENT - 1)) & ~(RESOURCE_ALIGNMENT - 1));
513 if (This->buffer_type_hint == GL_ELEMENT_ARRAY_BUFFER_ARB)
514 device_invalidate_state(This->resource.device, STATE_INDEXBUFFER);
517 GL_EXTCALL(glBindBufferARB(This->buffer_type_hint, This->buffer_object));
518 GL_EXTCALL(glGetBufferSubDataARB(This->buffer_type_hint, 0, This->resource.size, This->resource.allocatedMemory));
520 This->flags |= WINED3D_BUFFER_DOUBLEBUFFER;
522 return This->resource.allocatedMemory;
525 /* Do not call while under the GL lock. */
526 static void buffer_unload(struct wined3d_resource *resource)
528 struct wined3d_buffer *buffer = buffer_from_resource(resource);
530 TRACE("buffer %p.\n", buffer);
532 if (buffer->buffer_object)
534 struct wined3d_device *device = resource->device;
535 struct wined3d_context *context;
537 context = context_acquire(device, NULL);
539 /* Download the buffer, but don't permanently enable double buffering */
540 if (!(buffer->flags & WINED3D_BUFFER_DOUBLEBUFFER))
542 buffer_get_sysmem(buffer, context->gl_info);
543 buffer->flags &= ~WINED3D_BUFFER_DOUBLEBUFFER;
546 delete_gl_buffer(buffer, context->gl_info);
547 buffer->flags |= WINED3D_BUFFER_CREATEBO; /* Recreate the buffer object next load */
548 buffer_clear_dirty_areas(buffer);
550 context_release(context);
552 HeapFree(GetProcessHeap(), 0, buffer->conversion_map);
553 buffer->conversion_map = NULL;
555 buffer->conversion_stride = 0;
556 buffer->flags &= ~WINED3D_BUFFER_HASDESC;
559 resource_unload(resource);
562 /* Do not call while under the GL lock. */
563 ULONG CDECL wined3d_buffer_decref(struct wined3d_buffer *buffer)
565 ULONG refcount = InterlockedDecrement(&buffer->resource.ref);
567 TRACE("%p decreasing refcount to %u.\n", buffer, refcount);
571 buffer_unload(&buffer->resource);
572 resource_cleanup(&buffer->resource);
573 buffer->resource.parent_ops->wined3d_object_destroyed(buffer->resource.parent);
574 HeapFree(GetProcessHeap(), 0, buffer->maps);
575 HeapFree(GetProcessHeap(), 0, buffer);
581 void * CDECL wined3d_buffer_get_parent(const struct wined3d_buffer *buffer)
583 TRACE("buffer %p.\n", buffer);
585 return buffer->resource.parent;
588 DWORD CDECL wined3d_buffer_set_priority(struct wined3d_buffer *buffer, DWORD priority)
590 return resource_set_priority(&buffer->resource, priority);
593 DWORD CDECL wined3d_buffer_get_priority(const struct wined3d_buffer *buffer)
595 return resource_get_priority(&buffer->resource);
598 /* The caller provides a context and binds the buffer */
599 static void buffer_sync_apple(struct wined3d_buffer *This, DWORD flags, const struct wined3d_gl_info *gl_info)
601 enum wined3d_event_query_result ret;
603 /* No fencing needs to be done if the app promises not to overwrite
605 if(flags & WINED3DLOCK_NOOVERWRITE) return;
606 if(flags & WINED3DLOCK_DISCARD)
609 GL_EXTCALL(glBufferDataARB(This->buffer_type_hint, This->resource.size, NULL, This->buffer_object_usage));
610 checkGLcall("glBufferDataARB\n");
617 TRACE("Creating event query for buffer %p\n", This);
619 if (!wined3d_event_query_supported(gl_info))
621 FIXME("Event queries not supported, dropping async buffer locks.\n");
625 This->query = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*This->query));
628 ERR("Failed to allocate event query memory, dropping async buffer locks.\n");
632 /* Since we don't know about old draws a glFinish is needed once */
636 TRACE("Synchronizing buffer %p\n", This);
637 ret = wined3d_event_query_finish(This->query, This->resource.device);
640 case WINED3D_EVENT_QUERY_NOT_STARTED:
641 case WINED3D_EVENT_QUERY_OK:
645 case WINED3D_EVENT_QUERY_WRONG_THREAD:
646 WARN("Cannot synchronize buffer lock due to a thread conflict\n");
650 ERR("wined3d_event_query_finish returned %u, dropping async buffer locks\n", ret);
657 wined3d_event_query_destroy(This->query);
663 GL_EXTCALL(glBufferParameteriAPPLE(This->buffer_type_hint, GL_BUFFER_SERIALIZED_MODIFY_APPLE, GL_TRUE));
664 checkGLcall("glBufferParameteriAPPLE(This->buffer_type_hint, GL_BUFFER_SERIALIZED_MODIFY_APPLE, GL_TRUE)");
666 This->flags &= ~WINED3D_BUFFER_APPLESYNC;
669 /* The caller provides a GL context */
670 static void buffer_direct_upload(struct wined3d_buffer *This, const struct wined3d_gl_info *gl_info, DWORD flags)
673 UINT start = 0, len = 0;
677 /* This potentially invalidates the element array buffer binding, but the
678 * caller always takes care of this. */
679 GL_EXTCALL(glBindBufferARB(This->buffer_type_hint, This->buffer_object));
680 checkGLcall("glBindBufferARB");
681 if (gl_info->supported[ARB_MAP_BUFFER_RANGE])
684 mapflags = GL_MAP_WRITE_BIT | GL_MAP_FLUSH_EXPLICIT_BIT;
685 if (flags & WINED3D_BUFFER_DISCARD)
687 mapflags |= GL_MAP_UNSYNCHRONIZED_BIT | GL_MAP_INVALIDATE_BUFFER_BIT;
689 else if (flags & WINED3D_BUFFER_NOSYNC)
691 mapflags |= GL_MAP_UNSYNCHRONIZED_BIT;
693 map = GL_EXTCALL(glMapBufferRange(This->buffer_type_hint, 0,
694 This->resource.size, mapflags));
695 checkGLcall("glMapBufferRange");
699 if (This->flags & WINED3D_BUFFER_APPLESYNC)
702 if (flags & WINED3D_BUFFER_DISCARD) syncflags |= WINED3DLOCK_DISCARD;
703 if (flags & WINED3D_BUFFER_NOSYNC) syncflags |= WINED3DLOCK_NOOVERWRITE;
705 buffer_sync_apple(This, syncflags, gl_info);
708 map = GL_EXTCALL(glMapBufferARB(This->buffer_type_hint, GL_WRITE_ONLY_ARB));
709 checkGLcall("glMapBufferARB");
714 ERR("Failed to map opengl buffer\n");
718 while (This->modified_areas)
720 This->modified_areas--;
721 start = This->maps[This->modified_areas].offset;
722 len = This->maps[This->modified_areas].size;
724 memcpy(map + start, This->resource.allocatedMemory + start, len);
726 if (gl_info->supported[ARB_MAP_BUFFER_RANGE])
728 GL_EXTCALL(glFlushMappedBufferRange(This->buffer_type_hint, start, len));
729 checkGLcall("glFlushMappedBufferRange");
731 else if (This->flags & WINED3D_BUFFER_FLUSH)
733 GL_EXTCALL(glFlushMappedBufferRangeAPPLE(This->buffer_type_hint, start, len));
734 checkGLcall("glFlushMappedBufferRangeAPPLE");
737 GL_EXTCALL(glUnmapBufferARB(This->buffer_type_hint));
738 checkGLcall("glUnmapBufferARB");
743 /* Do not call while under the GL lock. */
744 void CDECL wined3d_buffer_preload(struct wined3d_buffer *buffer)
746 DWORD flags = buffer->flags & (WINED3D_BUFFER_NOSYNC | WINED3D_BUFFER_DISCARD);
747 struct wined3d_device *device = buffer->resource.device;
748 UINT start = 0, end = 0, len = 0, vertices;
749 const struct wined3d_gl_info *gl_info;
750 struct wined3d_context *context;
751 BOOL decl_changed = FALSE;
755 TRACE("buffer %p.\n", buffer);
757 buffer->flags &= ~(WINED3D_BUFFER_NOSYNC | WINED3D_BUFFER_DISCARD);
759 if (!buffer->buffer_object)
761 /* TODO: Make converting independent from VBOs */
762 if (buffer->flags & WINED3D_BUFFER_CREATEBO)
764 context = context_acquire(device, NULL);
765 buffer_create_buffer_object(buffer, context->gl_info);
766 context_release(context);
767 buffer->flags &= ~WINED3D_BUFFER_CREATEBO;
771 /* Not doing any conversion */
776 /* Reading the declaration makes only sense if the stateblock is finalized and the buffer bound to a stream */
777 if (device->isInDraw && buffer->bind_count > 0)
779 decl_changed = buffer_find_decl(buffer);
780 buffer->flags |= WINED3D_BUFFER_HASDESC;
783 if (!decl_changed && !(buffer->flags & WINED3D_BUFFER_HASDESC && buffer_is_dirty(buffer)))
785 ++buffer->draw_count;
786 if (buffer->draw_count > VB_RESETDECLCHANGE)
787 buffer->decl_change_count = 0;
788 if (buffer->draw_count > VB_RESETFULLCONVS)
789 buffer->full_conversion_count = 0;
793 /* If applications change the declaration over and over, reconverting all the time is a huge
794 * performance hit. So count the declaration changes and release the VBO if there are too many
795 * of them (and thus stop converting)
799 ++buffer->decl_change_count;
800 buffer->draw_count = 0;
802 if (buffer->decl_change_count > VB_MAXDECLCHANGES
803 || (buffer->conversion_map && (buffer->resource.usage & WINED3DUSAGE_DYNAMIC)))
805 FIXME("Too many declaration changes or converting dynamic buffer, stopping converting\n");
807 buffer_unload(&buffer->resource);
808 buffer->flags &= ~WINED3D_BUFFER_CREATEBO;
810 /* The stream source state handler might have read the memory of
811 * the vertex buffer already and got the memory in the vbo which
812 * is not valid any longer. Dirtify the stream source to force a
813 * reload. This happens only once per changed vertexbuffer and
814 * should occur rather rarely. */
815 device_invalidate_state(device, STATE_STREAMSRC);
819 /* The declaration changed, reload the whole buffer */
820 WARN("Reloading buffer because of decl change\n");
821 buffer_clear_dirty_areas(buffer);
822 if (!buffer_add_dirty_area(buffer, 0, 0))
824 ERR("buffer_add_dirty_area failed, this is not expected\n");
827 /* Avoid unfenced updates, we might overwrite more areas of the buffer than the application
828 * cleared for unsynchronized updates
834 /* However, it is perfectly fine to change the declaration every now and then. We don't want a game that
835 * changes it every minute drop the VBO after VB_MAX_DECL_CHANGES minutes. So count draws without
836 * decl changes and reset the decl change count after a specific number of them
838 if (buffer->conversion_map && buffer_is_fully_dirty(buffer))
840 ++buffer->full_conversion_count;
841 if (buffer->full_conversion_count > VB_MAXFULLCONVERSIONS)
843 FIXME("Too many full buffer conversions, stopping converting.\n");
844 buffer_unload(&buffer->resource);
845 buffer->flags &= ~WINED3D_BUFFER_CREATEBO;
846 if (buffer->bind_count)
847 device_invalidate_state(device, STATE_STREAMSRC);
853 ++buffer->draw_count;
854 if (buffer->draw_count > VB_RESETDECLCHANGE)
855 buffer->decl_change_count = 0;
856 if (buffer->draw_count > VB_RESETFULLCONVS)
857 buffer->full_conversion_count = 0;
861 if (buffer->buffer_type_hint == GL_ELEMENT_ARRAY_BUFFER_ARB)
862 device_invalidate_state(device, STATE_INDEXBUFFER);
864 if (!buffer->conversion_map)
866 /* That means that there is nothing to fixup. Just upload from
867 * buffer->resource.allocatedMemory directly into the vbo. Do not
868 * free the system memory copy because drawPrimitive may need it if
869 * the stride is 0, for instancing emulation, vertex blending
870 * emulation or shader emulation. */
871 TRACE("No conversion needed.\n");
873 /* Nothing to do because we locked directly into the vbo */
874 if (!(buffer->flags & WINED3D_BUFFER_DOUBLEBUFFER))
879 context = context_acquire(device, NULL);
880 buffer_direct_upload(buffer, context->gl_info, flags);
882 context_release(context);
886 context = context_acquire(device, NULL);
887 gl_info = context->gl_info;
889 if(!(buffer->flags & WINED3D_BUFFER_DOUBLEBUFFER))
891 buffer_get_sysmem(buffer, gl_info);
894 /* Now for each vertex in the buffer that needs conversion */
895 vertices = buffer->resource.size / buffer->stride;
897 data = HeapAlloc(GetProcessHeap(), 0, buffer->resource.size);
899 while(buffer->modified_areas)
901 buffer->modified_areas--;
902 start = buffer->maps[buffer->modified_areas].offset;
903 len = buffer->maps[buffer->modified_areas].size;
906 memcpy(data + start, buffer->resource.allocatedMemory + start, end - start);
907 for (i = start / buffer->stride; i < min((end / buffer->stride) + 1, vertices); ++i)
909 for (j = 0; j < buffer->stride; ++j)
911 switch (buffer->conversion_map[j])
918 fixup_d3dcolor((DWORD *) (data + i * buffer->stride + j));
923 fixup_transformed_pos((float *) (data + i * buffer->stride + j));
927 FIXME("Unimplemented conversion %d in shifted conversion\n", buffer->conversion_map[j]);
933 GL_EXTCALL(glBindBufferARB(buffer->buffer_type_hint, buffer->buffer_object));
934 checkGLcall("glBindBufferARB");
935 GL_EXTCALL(glBufferSubDataARB(buffer->buffer_type_hint, start, len, data + start));
936 checkGLcall("glBufferSubDataARB");
940 HeapFree(GetProcessHeap(), 0, data);
941 context_release(context);
944 static DWORD buffer_sanitize_flags(struct wined3d_buffer *buffer, DWORD flags)
946 /* Not all flags make sense together, but Windows never returns an error. Catch the
947 * cases that could cause issues */
948 if(flags & WINED3DLOCK_READONLY)
950 if(flags & WINED3DLOCK_DISCARD)
952 WARN("WINED3DLOCK_READONLY combined with WINED3DLOCK_DISCARD, ignoring flags\n");
955 if(flags & WINED3DLOCK_NOOVERWRITE)
957 WARN("WINED3DLOCK_READONLY combined with WINED3DLOCK_NOOVERWRITE, ignoring flags\n");
961 else if((flags & (WINED3DLOCK_DISCARD | WINED3DLOCK_NOOVERWRITE)) == (WINED3DLOCK_DISCARD | WINED3DLOCK_NOOVERWRITE))
963 WARN("WINED3DLOCK_DISCARD and WINED3DLOCK_NOOVERWRITE used together, ignoring\n");
966 else if (flags & (WINED3DLOCK_DISCARD | WINED3DLOCK_NOOVERWRITE) && !(buffer->resource.usage & WINED3DUSAGE_DYNAMIC))
968 WARN("DISCARD or NOOVERWRITE lock on non-dynamic buffer, ignoring\n");
975 static GLbitfield buffer_gl_map_flags(DWORD d3d_flags)
979 if (!(d3d_flags & WINED3DLOCK_READONLY)) ret = GL_MAP_WRITE_BIT | GL_MAP_FLUSH_EXPLICIT_BIT;
981 if (d3d_flags & (WINED3DLOCK_DISCARD | WINED3DLOCK_NOOVERWRITE))
983 if(d3d_flags & WINED3DLOCK_DISCARD) ret |= GL_MAP_INVALIDATE_BUFFER_BIT;
984 ret |= GL_MAP_UNSYNCHRONIZED_BIT;
988 ret |= GL_MAP_READ_BIT;
994 struct wined3d_resource * CDECL wined3d_buffer_get_resource(struct wined3d_buffer *buffer)
996 TRACE("buffer %p.\n", buffer);
998 return &buffer->resource;
1001 HRESULT CDECL wined3d_buffer_map(struct wined3d_buffer *buffer, UINT offset, UINT size, BYTE **data, DWORD flags)
1003 BOOL dirty = buffer_is_dirty(buffer);
1006 TRACE("buffer %p, offset %u, size %u, data %p, flags %#x\n", buffer, offset, size, data, flags);
1008 flags = buffer_sanitize_flags(buffer, flags);
1009 if (!(flags & WINED3DLOCK_READONLY))
1011 if (!buffer_add_dirty_area(buffer, offset, size)) return E_OUTOFMEMORY;
1014 count = InterlockedIncrement(&buffer->lock_count);
1016 if (buffer->buffer_object)
1018 if (!(buffer->flags & WINED3D_BUFFER_DOUBLEBUFFER))
1022 struct wined3d_device *device = buffer->resource.device;
1023 struct wined3d_context *context;
1024 const struct wined3d_gl_info *gl_info;
1026 if (buffer->buffer_type_hint == GL_ELEMENT_ARRAY_BUFFER_ARB)
1027 device_invalidate_state(device, STATE_INDEXBUFFER);
1029 context = context_acquire(device, NULL);
1030 gl_info = context->gl_info;
1032 GL_EXTCALL(glBindBufferARB(buffer->buffer_type_hint, buffer->buffer_object));
1034 if (gl_info->supported[ARB_MAP_BUFFER_RANGE])
1036 GLbitfield mapflags = buffer_gl_map_flags(flags);
1037 buffer->resource.allocatedMemory = GL_EXTCALL(glMapBufferRange(buffer->buffer_type_hint,
1038 0, buffer->resource.size, mapflags));
1039 checkGLcall("glMapBufferRange");
1043 if (buffer->flags & WINED3D_BUFFER_APPLESYNC)
1046 buffer_sync_apple(buffer, flags, gl_info);
1049 buffer->resource.allocatedMemory = GL_EXTCALL(glMapBufferARB(buffer->buffer_type_hint,
1050 GL_READ_WRITE_ARB));
1051 checkGLcall("glMapBufferARB");
1055 if (((DWORD_PTR)buffer->resource.allocatedMemory) & (RESOURCE_ALIGNMENT - 1))
1057 WARN("Pointer %p is not %u byte aligned.\n", buffer->resource.allocatedMemory, RESOURCE_ALIGNMENT);
1060 GL_EXTCALL(glUnmapBufferARB(buffer->buffer_type_hint));
1061 checkGLcall("glUnmapBufferARB");
1063 buffer->resource.allocatedMemory = NULL;
1065 if (buffer->resource.usage & WINED3DUSAGE_DYNAMIC)
1067 /* The extra copy is more expensive than not using VBOs at
1068 * all on the Nvidia Linux driver, which is the only driver
1069 * that returns unaligned pointers
1071 TRACE("Dynamic buffer, dropping VBO\n");
1072 buffer_unload(&buffer->resource);
1073 buffer->flags &= ~WINED3D_BUFFER_CREATEBO;
1074 if (buffer->bind_count)
1075 device_invalidate_state(device, STATE_STREAMSRC);
1079 TRACE("Falling back to doublebuffered operation\n");
1080 buffer_get_sysmem(buffer, gl_info);
1082 TRACE("New pointer is %p.\n", buffer->resource.allocatedMemory);
1084 context_release(context);
1091 if (buffer->flags & WINED3D_BUFFER_NOSYNC && !(flags & WINED3DLOCK_NOOVERWRITE))
1093 buffer->flags &= ~WINED3D_BUFFER_NOSYNC;
1096 else if(flags & WINED3DLOCK_NOOVERWRITE)
1098 buffer->flags |= WINED3D_BUFFER_NOSYNC;
1101 if (flags & WINED3DLOCK_DISCARD)
1103 buffer->flags |= WINED3D_BUFFER_DISCARD;
1108 *data = buffer->resource.allocatedMemory + offset;
1110 TRACE("Returning memory at %p (base %p, offset %u).\n", *data, buffer->resource.allocatedMemory, offset);
1111 /* TODO: check Flags compatibility with buffer->currentDesc.Usage (see MSDN) */
1116 void CDECL wined3d_buffer_unmap(struct wined3d_buffer *buffer)
1120 TRACE("buffer %p.\n", buffer);
1122 /* In the case that the number of Unmap calls > the
1123 * number of Map calls, d3d returns always D3D_OK.
1124 * This is also needed to prevent Map from returning garbage on
1125 * the next call (this will happen if the lock_count is < 0). */
1126 if (!buffer->lock_count)
1128 WARN("Unmap called without a previous map call.\n");
1132 if (InterlockedDecrement(&buffer->lock_count))
1134 /* Delay loading the buffer until everything is unlocked */
1135 TRACE("Ignoring unmap.\n");
1139 if (!(buffer->flags & WINED3D_BUFFER_DOUBLEBUFFER) && buffer->buffer_object)
1141 struct wined3d_device *device = buffer->resource.device;
1142 const struct wined3d_gl_info *gl_info;
1143 struct wined3d_context *context;
1145 if (buffer->buffer_type_hint == GL_ELEMENT_ARRAY_BUFFER_ARB)
1146 device_invalidate_state(device, STATE_INDEXBUFFER);
1148 context = context_acquire(device, NULL);
1149 gl_info = context->gl_info;
1151 GL_EXTCALL(glBindBufferARB(buffer->buffer_type_hint, buffer->buffer_object));
1153 if (gl_info->supported[ARB_MAP_BUFFER_RANGE])
1155 for (i = 0; i < buffer->modified_areas; ++i)
1157 GL_EXTCALL(glFlushMappedBufferRange(buffer->buffer_type_hint,
1158 buffer->maps[i].offset, buffer->maps[i].size));
1159 checkGLcall("glFlushMappedBufferRange");
1162 else if (buffer->flags & WINED3D_BUFFER_FLUSH)
1164 for (i = 0; i < buffer->modified_areas; ++i)
1166 GL_EXTCALL(glFlushMappedBufferRangeAPPLE(buffer->buffer_type_hint,
1167 buffer->maps[i].offset, buffer->maps[i].size));
1168 checkGLcall("glFlushMappedBufferRangeAPPLE");
1172 GL_EXTCALL(glUnmapBufferARB(buffer->buffer_type_hint));
1174 context_release(context);
1176 buffer->resource.allocatedMemory = NULL;
1177 buffer_clear_dirty_areas(buffer);
1179 else if (buffer->flags & WINED3D_BUFFER_HASDESC)
1181 wined3d_buffer_preload(buffer);
1185 static const struct wined3d_resource_ops buffer_resource_ops =
1190 static HRESULT buffer_init(struct wined3d_buffer *buffer, struct wined3d_device *device,
1191 UINT size, DWORD usage, enum wined3d_format_id format_id, WINED3DPOOL pool, GLenum bind_hint,
1192 const char *data, void *parent, const struct wined3d_parent_ops *parent_ops)
1194 const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
1195 const struct wined3d_format *format = wined3d_get_format(gl_info, format_id);
1197 BOOL dynamic_buffer_ok;
1201 WARN("Size 0 requested, returning WINED3DERR_INVALIDCALL\n");
1202 return WINED3DERR_INVALIDCALL;
1205 hr = resource_init(&buffer->resource, device, WINED3DRTYPE_BUFFER, format,
1206 WINED3DMULTISAMPLE_NONE, 0, usage, pool, size, 1, 1, size,
1207 parent, parent_ops, &buffer_resource_ops);
1210 WARN("Failed to initialize resource, hr %#x\n", hr);
1213 buffer->buffer_type_hint = bind_hint;
1215 TRACE("size %#x, usage %#x, format %s, memory @ %p, iface @ %p.\n", buffer->resource.size, buffer->resource.usage,
1216 debug_d3dformat(buffer->resource.format->id), buffer->resource.allocatedMemory, buffer);
1218 dynamic_buffer_ok = gl_info->supported[APPLE_FLUSH_BUFFER_RANGE] || gl_info->supported[ARB_MAP_BUFFER_RANGE];
1220 /* Observations show that drawStridedSlow is faster on dynamic VBs than converting +
1221 * drawStridedFast (half-life 2 and others).
1223 * Basically converting the vertices in the buffer is quite expensive, and observations
1224 * show that drawStridedSlow is faster than converting + uploading + drawStridedFast.
1225 * Therefore do not create a VBO for WINED3DUSAGE_DYNAMIC buffers.
1227 if (!gl_info->supported[ARB_VERTEX_BUFFER_OBJECT])
1229 TRACE("Not creating a vbo because GL_ARB_vertex_buffer is not supported\n");
1231 else if(buffer->resource.pool == WINED3DPOOL_SYSTEMMEM)
1233 TRACE("Not creating a vbo because the vertex buffer is in system memory\n");
1235 else if(!dynamic_buffer_ok && (buffer->resource.usage & WINED3DUSAGE_DYNAMIC))
1237 TRACE("Not creating a vbo because the buffer has dynamic usage and no GL support\n");
1241 buffer->flags |= WINED3D_BUFFER_CREATEBO;
1248 hr = wined3d_buffer_map(buffer, 0, size, &ptr, 0);
1251 ERR("Failed to map buffer, hr %#x\n", hr);
1252 buffer_unload(&buffer->resource);
1253 resource_cleanup(&buffer->resource);
1257 memcpy(ptr, data, size);
1259 wined3d_buffer_unmap(buffer);
1262 buffer->maps = HeapAlloc(GetProcessHeap(), 0, sizeof(*buffer->maps));
1265 ERR("Out of memory\n");
1266 buffer_unload(&buffer->resource);
1267 resource_cleanup(&buffer->resource);
1268 return E_OUTOFMEMORY;
1270 buffer->maps_size = 1;
1275 HRESULT CDECL wined3d_buffer_create(struct wined3d_device *device, struct wined3d_buffer_desc *desc, const void *data,
1276 void *parent, const struct wined3d_parent_ops *parent_ops, struct wined3d_buffer **buffer)
1278 struct wined3d_buffer *object;
1281 TRACE("device %p, desc %p, data %p, parent %p, buffer %p\n", device, desc, data, parent, buffer);
1283 object = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*object));
1286 ERR("Failed to allocate memory\n");
1287 return E_OUTOFMEMORY;
1290 FIXME("Ignoring access flags (pool)\n");
1292 hr = buffer_init(object, device, desc->byte_width, desc->usage, WINED3DFMT_UNKNOWN,
1293 WINED3DPOOL_MANAGED, GL_ARRAY_BUFFER_ARB, data, parent, parent_ops);
1296 WARN("Failed to initialize buffer, hr %#x.\n", hr);
1297 HeapFree(GetProcessHeap(), 0, object);
1300 object->desc = *desc;
1302 TRACE("Created buffer %p.\n", object);
1309 HRESULT CDECL wined3d_buffer_create_vb(struct wined3d_device *device, UINT size, DWORD usage, WINED3DPOOL pool,
1310 void *parent, const struct wined3d_parent_ops *parent_ops, struct wined3d_buffer **buffer)
1312 struct wined3d_buffer *object;
1315 TRACE("device %p, size %u, usage %#x, pool %#x, parent %p, parent_ops %p, buffer %p.\n",
1316 device, size, usage, pool, parent, parent_ops, buffer);
1318 if (pool == WINED3DPOOL_SCRATCH)
1320 /* The d3d9 tests shows that this is not allowed. It doesn't make much
1321 * sense anyway, SCRATCH buffers wouldn't be usable anywhere. */
1322 WARN("Vertex buffer in WINED3DPOOL_SCRATCH requested, returning WINED3DERR_INVALIDCALL.\n");
1324 return WINED3DERR_INVALIDCALL;
1327 object = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*object));
1330 ERR("Out of memory\n");
1332 return WINED3DERR_OUTOFVIDEOMEMORY;
1335 hr = buffer_init(object, device, size, usage, WINED3DFMT_VERTEXDATA,
1336 pool, GL_ARRAY_BUFFER_ARB, NULL, parent, parent_ops);
1339 WARN("Failed to initialize buffer, hr %#x.\n", hr);
1340 HeapFree(GetProcessHeap(), 0, object);
1344 TRACE("Created buffer %p.\n", object);
1350 HRESULT CDECL wined3d_buffer_create_ib(struct wined3d_device *device, UINT size, DWORD usage, WINED3DPOOL pool,
1351 void *parent, const struct wined3d_parent_ops *parent_ops, struct wined3d_buffer **buffer)
1353 struct wined3d_buffer *object;
1356 TRACE("device %p, size %u, usage %#x, pool %#x, parent %p, parent_ops %p, buffer %p.\n",
1357 device, size, usage, pool, parent, parent_ops, buffer);
1359 object = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*object));
1362 ERR("Out of memory\n");
1364 return WINED3DERR_OUTOFVIDEOMEMORY;
1367 hr = buffer_init(object, device, size, usage | WINED3DUSAGE_STATICDECL,
1368 WINED3DFMT_UNKNOWN, pool, GL_ELEMENT_ARRAY_BUFFER_ARB, NULL,
1369 parent, parent_ops);
1372 WARN("Failed to initialize buffer, hr %#x\n", hr);
1373 HeapFree(GetProcessHeap(), 0, object);
1377 TRACE("Created buffer %p.\n", object);