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 wined3d_buffer %p with 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 * The vertex declaration from the device determines how the data in the
136 * buffer is interpreted. This means that on each draw call the buffer has
137 * to be verified to check if the rhw and color values are in the correct
140 GL_EXTCALL(glGenBuffersARB(1, &This->buffer_object));
141 error = glGetError();
142 if (!This->buffer_object || error != GL_NO_ERROR)
144 ERR("Failed to create a VBO with error %s (%#x)\n", debug_glerror(error), error);
149 if (This->buffer_type_hint == GL_ELEMENT_ARRAY_BUFFER_ARB)
150 device_invalidate_state(This->resource.device, STATE_INDEXBUFFER);
151 GL_EXTCALL(glBindBufferARB(This->buffer_type_hint, This->buffer_object));
152 error = glGetError();
153 if (error != GL_NO_ERROR)
155 ERR("Failed to bind the VBO with error %s (%#x)\n", debug_glerror(error), error);
160 /* Don't use static, because dx apps tend to update the buffer
161 * quite often even if they specify 0 usage.
163 if(This->resource.usage & WINED3DUSAGE_DYNAMIC)
165 TRACE("Gl usage = GL_STREAM_DRAW_ARB\n");
166 gl_usage = GL_STREAM_DRAW_ARB;
168 if(gl_info->supported[APPLE_FLUSH_BUFFER_RANGE])
170 GL_EXTCALL(glBufferParameteriAPPLE(This->buffer_type_hint, GL_BUFFER_FLUSHING_UNMAP_APPLE, GL_FALSE));
171 checkGLcall("glBufferParameteriAPPLE(This->buffer_type_hint, GL_BUFFER_FLUSHING_UNMAP_APPLE, GL_FALSE)");
172 This->flags |= WINED3D_BUFFER_FLUSH;
174 GL_EXTCALL(glBufferParameteriAPPLE(This->buffer_type_hint, GL_BUFFER_SERIALIZED_MODIFY_APPLE, GL_FALSE));
175 checkGLcall("glBufferParameteriAPPLE(This->buffer_type_hint, GL_BUFFER_SERIALIZED_MODIFY_APPLE, GL_FALSE)");
176 This->flags |= WINED3D_BUFFER_APPLESYNC;
178 /* No setup is needed here for GL_ARB_map_buffer_range */
182 TRACE("Gl usage = GL_DYNAMIC_DRAW_ARB\n");
183 gl_usage = GL_DYNAMIC_DRAW_ARB;
186 /* Reserve memory for the buffer. The amount of data won't change
187 * so we are safe with calling glBufferData once and
188 * calling glBufferSubData on updates. Upload the actual data in case
189 * we're not double buffering, so we can release the heap mem afterwards
191 GL_EXTCALL(glBufferDataARB(This->buffer_type_hint, This->resource.size, This->resource.allocatedMemory, gl_usage));
192 error = glGetError();
194 if (error != GL_NO_ERROR)
196 ERR("glBufferDataARB failed with error %s (%#x)\n", debug_glerror(error), error);
200 This->buffer_object_size = This->resource.size;
201 This->buffer_object_usage = gl_usage;
203 if(This->flags & WINED3D_BUFFER_DOUBLEBUFFER)
205 if(!buffer_add_dirty_area(This, 0, 0))
207 ERR("buffer_add_dirty_area failed, this is not expected\n");
213 HeapFree(GetProcessHeap(), 0, This->resource.heapMemory);
214 This->resource.allocatedMemory = NULL;
215 This->resource.heapMemory = NULL;
221 /* Clean up all vbo init, but continue because we can work without a vbo :-) */
222 ERR("Failed to create a vertex buffer object. Continuing, but performance issues may occur\n");
223 delete_gl_buffer(This, gl_info);
224 buffer_clear_dirty_areas(This);
227 static BOOL buffer_process_converted_attribute(struct wined3d_buffer *This,
228 const enum wined3d_buffer_conversion_type conversion_type,
229 const struct wined3d_stream_info_element *attrib, DWORD *stride_this_run)
236 /* Check for some valid situations which cause us pain. One is if the buffer is used for
237 * constant attributes(stride = 0), the other one is if the buffer is used on two streams
238 * with different strides. In the 2nd case we might have to drop conversion entirely,
239 * it is possible that the same bytes are once read as FLOAT2 and once as UBYTE4N.
243 FIXME("%s used with stride 0, let's hope we get the vertex stride from somewhere else\n",
244 debug_d3dformat(attrib->format->id));
246 else if(attrib->stride != *stride_this_run && *stride_this_run)
248 FIXME("Got two concurrent strides, %d and %d\n", attrib->stride, *stride_this_run);
252 *stride_this_run = attrib->stride;
253 if (This->stride != *stride_this_run)
255 /* We rely that this happens only on the first converted attribute that is found,
256 * if at all. See above check
258 TRACE("Reconverting because converted attributes occur, and the stride changed\n");
259 This->stride = *stride_this_run;
260 HeapFree(GetProcessHeap(), HEAP_ZERO_MEMORY, This->conversion_map);
261 This->conversion_map = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY,
262 sizeof(*This->conversion_map) * This->stride);
267 data = ((DWORD_PTR)attrib->data.addr) % This->stride;
268 attrib_size = attrib->format->component_count * attrib->format->component_size;
269 for (i = 0; i < attrib_size; ++i)
271 DWORD_PTR idx = (data + i) % This->stride;
272 if (This->conversion_map[idx] != conversion_type)
274 TRACE("Byte %ld in vertex changed\n", idx);
275 TRACE("It was type %d, is %d now\n", This->conversion_map[idx], conversion_type);
277 This->conversion_map[idx] = conversion_type;
284 static BOOL buffer_check_attribute(struct wined3d_buffer *This, const struct wined3d_stream_info *si,
285 UINT attrib_idx, const BOOL check_d3dcolor, const BOOL is_ffp_position, const BOOL is_ffp_color,
286 DWORD *stride_this_run)
288 const struct wined3d_stream_info_element *attrib = &si->elements[attrib_idx];
289 enum wined3d_format_id format;
292 /* Ignore attributes that do not have our vbo. After that check we can be sure that the attribute is
293 * there, on nonexistent attribs the vbo is 0.
295 if (!(si->use_map & (1 << attrib_idx))
296 || attrib->data.buffer_object != This->buffer_object)
299 format = attrib->format->id;
300 /* Look for newly appeared conversion */
301 if (check_d3dcolor && format == WINED3DFMT_B8G8R8A8_UNORM)
303 ret = buffer_process_converted_attribute(This, CONV_D3DCOLOR, attrib, stride_this_run);
305 if (!is_ffp_color) FIXME("Test for non-color fixed function WINED3DFMT_B8G8R8A8_UNORM format\n");
307 else if (is_ffp_position && format == WINED3DFMT_R32G32B32A32_FLOAT)
309 ret = buffer_process_converted_attribute(This, CONV_POSITIONT, attrib, stride_this_run);
311 else if (This->conversion_map)
313 ret = buffer_process_converted_attribute(This, CONV_NONE, attrib, stride_this_run);
319 static BOOL buffer_find_decl(struct wined3d_buffer *This)
321 struct wined3d_device *device = This->resource.device;
322 const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
323 const struct wined3d_stream_info *si = &device->strided_streams;
324 const struct wined3d_state *state = &device->stateBlock->state;
325 UINT stride_this_run = 0;
327 BOOL support_d3dcolor = gl_info->supported[ARB_VERTEX_ARRAY_BGRA];
329 /* In d3d7 the vertex buffer declaration NEVER changes because it is stored in the d3d7 vertex buffer.
330 * Once we have our declaration there is no need to look it up again. Index buffers also never need
331 * conversion, so once the (empty) conversion structure is created don't bother checking again
333 if (This->flags & WINED3D_BUFFER_HASDESC)
335 if(This->resource.usage & WINED3DUSAGE_STATICDECL) return FALSE;
340 TRACE("Vertex shaders used, no VBO conversion is needed\n");
341 if(This->conversion_map)
343 HeapFree(GetProcessHeap(), 0, This->conversion_map);
344 This->conversion_map = NULL;
352 TRACE("Finding vertex buffer conversion information\n");
353 /* Certain declaration types need some fixups before we can pass them to
354 * opengl. This means D3DCOLOR attributes with fixed function vertex
355 * processing, FLOAT4 POSITIONT with fixed function, and FLOAT16 if
356 * GL_ARB_half_float_vertex is not supported.
358 * Note for d3d8 and d3d9:
359 * The vertex buffer FVF doesn't help with finding them, we have to use
360 * the decoded vertex declaration and pick the things that concern the
361 * current buffer. A problem with this is that this can change between
362 * draws, so we have to validate the information and reprocess the buffer
363 * if it changes, and avoid false positives for performance reasons.
364 * WineD3D doesn't even know the vertex buffer any more, it is managed
365 * by the client libraries and passed to SetStreamSource and ProcessVertices
368 * We have to distinguish between vertex shaders and fixed function to
369 * pick the way we access the strided vertex information.
371 * This code sets up a per-byte array with the size of the detected
372 * stride of the arrays in the buffer. For each byte we have a field
373 * that marks the conversion needed on this byte. For example, the
374 * following declaration with fixed function vertex processing:
382 * { POSITIONT }{ NORMAL }{ DIFFUSE }{SPECULAR }
383 * [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]
385 * Where in this example map P means 4 component position conversion, 0
386 * means no conversion, F means FLOAT16_2 conversion and C means D3DCOLOR
387 * conversion (red / blue swizzle).
389 * If we're doing conversion and the stride changes we have to reconvert
390 * the whole buffer. Note that we do not mind if the semantic changes,
391 * we only care for the conversion type. So if the NORMAL is replaced
392 * with a TEXCOORD, nothing has to be done, or if the DIFFUSE is replaced
393 * with a D3DCOLOR BLENDWEIGHT we can happily dismiss the change. Some
394 * conversion types depend on the semantic as well, for example a FLOAT4
395 * texcoord needs no conversion while a FLOAT4 positiont needs one
398 ret = buffer_check_attribute(This, si, WINED3D_FFP_POSITION,
399 TRUE, TRUE, FALSE, &stride_this_run) || ret;
400 ret = buffer_check_attribute(This, si, WINED3D_FFP_NORMAL,
401 TRUE, FALSE, FALSE, &stride_this_run) || ret;
402 ret = buffer_check_attribute(This, si, WINED3D_FFP_DIFFUSE,
403 !support_d3dcolor, FALSE, TRUE, &stride_this_run) || ret;
404 ret = buffer_check_attribute(This, si, WINED3D_FFP_SPECULAR,
405 !support_d3dcolor, FALSE, TRUE, &stride_this_run) || ret;
406 ret = buffer_check_attribute(This, si, WINED3D_FFP_TEXCOORD0,
407 TRUE, FALSE, FALSE, &stride_this_run) || ret;
408 ret = buffer_check_attribute(This, si, WINED3D_FFP_TEXCOORD1,
409 TRUE, FALSE, FALSE, &stride_this_run) || ret;
410 ret = buffer_check_attribute(This, si, WINED3D_FFP_TEXCOORD2,
411 TRUE, FALSE, FALSE, &stride_this_run) || ret;
412 ret = buffer_check_attribute(This, si, WINED3D_FFP_TEXCOORD3,
413 TRUE, FALSE, FALSE, &stride_this_run) || ret;
414 ret = buffer_check_attribute(This, si, WINED3D_FFP_TEXCOORD4,
415 TRUE, FALSE, FALSE, &stride_this_run) || ret;
416 ret = buffer_check_attribute(This, si, WINED3D_FFP_TEXCOORD5,
417 TRUE, FALSE, FALSE, &stride_this_run) || ret;
418 ret = buffer_check_attribute(This, si, WINED3D_FFP_TEXCOORD6,
419 TRUE, FALSE, FALSE, &stride_this_run) || ret;
420 ret = buffer_check_attribute(This, si, WINED3D_FFP_TEXCOORD7,
421 TRUE, FALSE, FALSE, &stride_this_run) || ret;
423 if (!stride_this_run && This->conversion_map)
426 if (!ret) ERR("no converted attributes found, old conversion map exists, and no declaration change?\n");
427 HeapFree(GetProcessHeap(), 0, This->conversion_map);
428 This->conversion_map = NULL;
432 if (ret) TRACE("Conversion information changed\n");
437 static inline void fixup_d3dcolor(DWORD *dst_color)
439 DWORD src_color = *dst_color;
441 /* Color conversion like in drawStridedSlow. watch out for little endianity
442 * If we want that stuff to work on big endian machines too we have to consider more things
444 * 0xff000000: Alpha mask
445 * 0x00ff0000: Blue mask
446 * 0x0000ff00: Green mask
447 * 0x000000ff: Red mask
450 *dst_color |= (src_color & 0xff00ff00); /* Alpha Green */
451 *dst_color |= (src_color & 0x00ff0000) >> 16; /* Red */
452 *dst_color |= (src_color & 0x000000ff) << 16; /* Blue */
455 static inline void fixup_transformed_pos(float *p)
457 /* rhw conversion like in position_float4(). */
458 if (p[3] != 1.0f && p[3] != 0.0f)
460 float w = 1.0f / p[3];
468 /* Context activation is done by the caller. */
469 void buffer_get_memory(struct wined3d_buffer *buffer, const struct wined3d_gl_info *gl_info,
470 struct wined3d_bo_address *data)
472 data->buffer_object = buffer->buffer_object;
473 if (!buffer->buffer_object)
475 if (buffer->flags & WINED3D_BUFFER_CREATEBO)
477 buffer_create_buffer_object(buffer, gl_info);
478 buffer->flags &= ~WINED3D_BUFFER_CREATEBO;
479 if (buffer->buffer_object)
481 data->buffer_object = buffer->buffer_object;
486 data->addr = buffer->resource.allocatedMemory;
494 ULONG CDECL wined3d_buffer_incref(struct wined3d_buffer *buffer)
496 ULONG refcount = InterlockedIncrement(&buffer->resource.ref);
498 TRACE("%p increasing refcount to %u.\n", buffer, refcount);
503 /* Context activation is done by the caller. */
504 BYTE *buffer_get_sysmem(struct wined3d_buffer *This, const struct wined3d_gl_info *gl_info)
506 /* AllocatedMemory exists if the buffer is double buffered or has no buffer object at all */
507 if(This->resource.allocatedMemory) return This->resource.allocatedMemory;
509 This->resource.heapMemory = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, This->resource.size + RESOURCE_ALIGNMENT);
510 This->resource.allocatedMemory = (BYTE *)(((ULONG_PTR)This->resource.heapMemory + (RESOURCE_ALIGNMENT - 1)) & ~(RESOURCE_ALIGNMENT - 1));
512 if (This->buffer_type_hint == GL_ELEMENT_ARRAY_BUFFER_ARB)
513 device_invalidate_state(This->resource.device, STATE_INDEXBUFFER);
516 GL_EXTCALL(glBindBufferARB(This->buffer_type_hint, This->buffer_object));
517 GL_EXTCALL(glGetBufferSubDataARB(This->buffer_type_hint, 0, This->resource.size, This->resource.allocatedMemory));
519 This->flags |= WINED3D_BUFFER_DOUBLEBUFFER;
521 return This->resource.allocatedMemory;
524 /* Do not call while under the GL lock. */
525 static void buffer_unload(struct wined3d_resource *resource)
527 struct wined3d_buffer *buffer = buffer_from_resource(resource);
529 TRACE("buffer %p.\n", buffer);
531 if (buffer->buffer_object)
533 struct wined3d_device *device = resource->device;
534 struct wined3d_context *context;
536 context = context_acquire(device, NULL);
538 /* Download the buffer, but don't permanently enable double buffering */
539 if (!(buffer->flags & WINED3D_BUFFER_DOUBLEBUFFER))
541 buffer_get_sysmem(buffer, context->gl_info);
542 buffer->flags &= ~WINED3D_BUFFER_DOUBLEBUFFER;
545 delete_gl_buffer(buffer, context->gl_info);
546 buffer->flags |= WINED3D_BUFFER_CREATEBO; /* Recreate the buffer object next load */
547 buffer_clear_dirty_areas(buffer);
549 context_release(context);
551 HeapFree(GetProcessHeap(), 0, buffer->conversion_map);
552 buffer->conversion_map = NULL;
554 buffer->conversion_stride = 0;
555 buffer->flags &= ~WINED3D_BUFFER_HASDESC;
558 resource_unload(resource);
561 /* Do not call while under the GL lock. */
562 ULONG CDECL wined3d_buffer_decref(struct wined3d_buffer *buffer)
564 ULONG refcount = InterlockedDecrement(&buffer->resource.ref);
566 TRACE("%p decreasing refcount to %u.\n", buffer, refcount);
570 buffer_unload(&buffer->resource);
571 resource_cleanup(&buffer->resource);
572 buffer->resource.parent_ops->wined3d_object_destroyed(buffer->resource.parent);
573 HeapFree(GetProcessHeap(), 0, buffer->maps);
574 HeapFree(GetProcessHeap(), 0, buffer);
580 void * CDECL wined3d_buffer_get_parent(const struct wined3d_buffer *buffer)
582 TRACE("buffer %p.\n", buffer);
584 return buffer->resource.parent;
587 DWORD CDECL wined3d_buffer_set_priority(struct wined3d_buffer *buffer, DWORD priority)
589 return resource_set_priority(&buffer->resource, priority);
592 DWORD CDECL wined3d_buffer_get_priority(const struct wined3d_buffer *buffer)
594 return resource_get_priority(&buffer->resource);
597 /* The caller provides a context and binds the buffer */
598 static void buffer_sync_apple(struct wined3d_buffer *This, DWORD flags, const struct wined3d_gl_info *gl_info)
600 enum wined3d_event_query_result ret;
602 /* No fencing needs to be done if the app promises not to overwrite
604 if(flags & WINED3DLOCK_NOOVERWRITE) return;
605 if(flags & WINED3DLOCK_DISCARD)
608 GL_EXTCALL(glBufferDataARB(This->buffer_type_hint, This->resource.size, NULL, This->buffer_object_usage));
609 checkGLcall("glBufferDataARB\n");
616 TRACE("Creating event query for buffer %p\n", This);
618 if (!wined3d_event_query_supported(gl_info))
620 FIXME("Event queries not supported, dropping async buffer locks.\n");
624 This->query = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*This->query));
627 ERR("Failed to allocate event query memory, dropping async buffer locks.\n");
631 /* Since we don't know about old draws a glFinish is needed once */
635 TRACE("Synchronizing buffer %p\n", This);
636 ret = wined3d_event_query_finish(This->query, This->resource.device);
639 case WINED3D_EVENT_QUERY_NOT_STARTED:
640 case WINED3D_EVENT_QUERY_OK:
644 case WINED3D_EVENT_QUERY_WRONG_THREAD:
645 WARN("Cannot synchronize buffer lock due to a thread conflict\n");
649 ERR("wined3d_event_query_finish returned %u, dropping async buffer locks\n", ret);
656 wined3d_event_query_destroy(This->query);
662 GL_EXTCALL(glBufferParameteriAPPLE(This->buffer_type_hint, GL_BUFFER_SERIALIZED_MODIFY_APPLE, GL_TRUE));
663 checkGLcall("glBufferParameteriAPPLE(This->buffer_type_hint, GL_BUFFER_SERIALIZED_MODIFY_APPLE, GL_TRUE)");
665 This->flags &= ~WINED3D_BUFFER_APPLESYNC;
668 /* The caller provides a GL context */
669 static void buffer_direct_upload(struct wined3d_buffer *This, const struct wined3d_gl_info *gl_info, DWORD flags)
672 UINT start = 0, len = 0;
676 /* This potentially invalidates the element array buffer binding, but the
677 * caller always takes care of this. */
678 GL_EXTCALL(glBindBufferARB(This->buffer_type_hint, This->buffer_object));
679 checkGLcall("glBindBufferARB");
680 if (gl_info->supported[ARB_MAP_BUFFER_RANGE])
683 mapflags = GL_MAP_WRITE_BIT | GL_MAP_FLUSH_EXPLICIT_BIT;
684 if (flags & WINED3D_BUFFER_DISCARD)
685 mapflags |= GL_MAP_INVALIDATE_BUFFER_BIT;
686 if (flags & WINED3D_BUFFER_NOSYNC)
687 mapflags |= GL_MAP_UNSYNCHRONIZED_BIT;
688 map = GL_EXTCALL(glMapBufferRange(This->buffer_type_hint, 0,
689 This->resource.size, mapflags));
690 checkGLcall("glMapBufferRange");
694 if (This->flags & WINED3D_BUFFER_APPLESYNC)
697 if (flags & WINED3D_BUFFER_DISCARD) syncflags |= WINED3DLOCK_DISCARD;
698 if (flags & WINED3D_BUFFER_NOSYNC) syncflags |= WINED3DLOCK_NOOVERWRITE;
700 buffer_sync_apple(This, syncflags, gl_info);
703 map = GL_EXTCALL(glMapBufferARB(This->buffer_type_hint, GL_WRITE_ONLY_ARB));
704 checkGLcall("glMapBufferARB");
709 ERR("Failed to map opengl buffer\n");
713 while (This->modified_areas)
715 This->modified_areas--;
716 start = This->maps[This->modified_areas].offset;
717 len = This->maps[This->modified_areas].size;
719 memcpy(map + start, This->resource.allocatedMemory + start, len);
721 if (gl_info->supported[ARB_MAP_BUFFER_RANGE])
723 GL_EXTCALL(glFlushMappedBufferRange(This->buffer_type_hint, start, len));
724 checkGLcall("glFlushMappedBufferRange");
726 else if (This->flags & WINED3D_BUFFER_FLUSH)
728 GL_EXTCALL(glFlushMappedBufferRangeAPPLE(This->buffer_type_hint, start, len));
729 checkGLcall("glFlushMappedBufferRangeAPPLE");
732 GL_EXTCALL(glUnmapBufferARB(This->buffer_type_hint));
733 checkGLcall("glUnmapBufferARB");
738 /* Do not call while under the GL lock. */
739 void CDECL wined3d_buffer_preload(struct wined3d_buffer *buffer)
741 DWORD flags = buffer->flags & (WINED3D_BUFFER_NOSYNC | WINED3D_BUFFER_DISCARD);
742 struct wined3d_device *device = buffer->resource.device;
743 UINT start = 0, end = 0, len = 0, vertices;
744 const struct wined3d_gl_info *gl_info;
745 struct wined3d_context *context;
746 BOOL decl_changed = FALSE;
750 TRACE("buffer %p.\n", buffer);
752 buffer->flags &= ~(WINED3D_BUFFER_NOSYNC | WINED3D_BUFFER_DISCARD);
754 if (!buffer->buffer_object)
756 /* TODO: Make converting independent from VBOs */
757 if (buffer->flags & WINED3D_BUFFER_CREATEBO)
759 context = context_acquire(device, NULL);
760 buffer_create_buffer_object(buffer, context->gl_info);
761 context_release(context);
762 buffer->flags &= ~WINED3D_BUFFER_CREATEBO;
766 /* Not doing any conversion */
771 /* Reading the declaration makes only sense if the stateblock is finalized and the buffer bound to a stream */
772 if (device->isInDraw && buffer->bind_count > 0)
774 decl_changed = buffer_find_decl(buffer);
775 buffer->flags |= WINED3D_BUFFER_HASDESC;
778 if (!decl_changed && !(buffer->flags & WINED3D_BUFFER_HASDESC && buffer_is_dirty(buffer)))
780 ++buffer->draw_count;
781 if (buffer->draw_count > VB_RESETDECLCHANGE)
782 buffer->decl_change_count = 0;
783 if (buffer->draw_count > VB_RESETFULLCONVS)
784 buffer->full_conversion_count = 0;
788 /* If applications change the declaration over and over, reconverting all the time is a huge
789 * performance hit. So count the declaration changes and release the VBO if there are too many
790 * of them (and thus stop converting)
794 ++buffer->decl_change_count;
795 buffer->draw_count = 0;
797 if (buffer->decl_change_count > VB_MAXDECLCHANGES
798 || (buffer->conversion_map && (buffer->resource.usage & WINED3DUSAGE_DYNAMIC)))
800 FIXME("Too many declaration changes or converting dynamic buffer, stopping converting\n");
802 buffer_unload(&buffer->resource);
803 buffer->flags &= ~WINED3D_BUFFER_CREATEBO;
805 /* The stream source state handler might have read the memory of
806 * the vertex buffer already and got the memory in the vbo which
807 * is not valid any longer. Dirtify the stream source to force a
808 * reload. This happens only once per changed vertexbuffer and
809 * should occur rather rarely. */
810 device_invalidate_state(device, STATE_STREAMSRC);
814 /* The declaration changed, reload the whole buffer */
815 WARN("Reloading buffer because of decl change\n");
816 buffer_clear_dirty_areas(buffer);
817 if (!buffer_add_dirty_area(buffer, 0, 0))
819 ERR("buffer_add_dirty_area failed, this is not expected\n");
822 /* Avoid unfenced updates, we might overwrite more areas of the buffer than the application
823 * cleared for unsynchronized updates
829 /* However, it is perfectly fine to change the declaration every now and then. We don't want a game that
830 * changes it every minute drop the VBO after VB_MAX_DECL_CHANGES minutes. So count draws without
831 * decl changes and reset the decl change count after a specific number of them
833 if (buffer->conversion_map && buffer_is_fully_dirty(buffer))
835 ++buffer->full_conversion_count;
836 if (buffer->full_conversion_count > VB_MAXFULLCONVERSIONS)
838 FIXME("Too many full buffer conversions, stopping converting.\n");
839 buffer_unload(&buffer->resource);
840 buffer->flags &= ~WINED3D_BUFFER_CREATEBO;
841 if (buffer->bind_count)
842 device_invalidate_state(device, STATE_STREAMSRC);
848 ++buffer->draw_count;
849 if (buffer->draw_count > VB_RESETDECLCHANGE)
850 buffer->decl_change_count = 0;
851 if (buffer->draw_count > VB_RESETFULLCONVS)
852 buffer->full_conversion_count = 0;
856 if (buffer->buffer_type_hint == GL_ELEMENT_ARRAY_BUFFER_ARB)
857 device_invalidate_state(device, STATE_INDEXBUFFER);
859 if (!buffer->conversion_map)
861 /* That means that there is nothing to fixup. Just upload from
862 * buffer->resource.allocatedMemory directly into the vbo. Do not
863 * free the system memory copy because drawPrimitive may need it if
864 * the stride is 0, for instancing emulation, vertex blending
865 * emulation or shader emulation. */
866 TRACE("No conversion needed.\n");
868 /* Nothing to do because we locked directly into the vbo */
869 if (!(buffer->flags & WINED3D_BUFFER_DOUBLEBUFFER))
874 context = context_acquire(device, NULL);
875 buffer_direct_upload(buffer, context->gl_info, flags);
877 context_release(context);
881 context = context_acquire(device, NULL);
882 gl_info = context->gl_info;
884 if(!(buffer->flags & WINED3D_BUFFER_DOUBLEBUFFER))
886 buffer_get_sysmem(buffer, gl_info);
889 /* Now for each vertex in the buffer that needs conversion */
890 vertices = buffer->resource.size / buffer->stride;
892 data = HeapAlloc(GetProcessHeap(), 0, buffer->resource.size);
894 while(buffer->modified_areas)
896 buffer->modified_areas--;
897 start = buffer->maps[buffer->modified_areas].offset;
898 len = buffer->maps[buffer->modified_areas].size;
901 memcpy(data + start, buffer->resource.allocatedMemory + start, end - start);
902 for (i = start / buffer->stride; i < min((end / buffer->stride) + 1, vertices); ++i)
904 for (j = 0; j < buffer->stride; ++j)
906 switch (buffer->conversion_map[j])
913 fixup_d3dcolor((DWORD *) (data + i * buffer->stride + j));
918 fixup_transformed_pos((float *) (data + i * buffer->stride + j));
922 FIXME("Unimplemented conversion %d in shifted conversion\n", buffer->conversion_map[j]);
928 GL_EXTCALL(glBindBufferARB(buffer->buffer_type_hint, buffer->buffer_object));
929 checkGLcall("glBindBufferARB");
930 GL_EXTCALL(glBufferSubDataARB(buffer->buffer_type_hint, start, len, data + start));
931 checkGLcall("glBufferSubDataARB");
935 HeapFree(GetProcessHeap(), 0, data);
936 context_release(context);
939 static DWORD buffer_sanitize_flags(struct wined3d_buffer *buffer, DWORD flags)
941 /* Not all flags make sense together, but Windows never returns an error. Catch the
942 * cases that could cause issues */
943 if(flags & WINED3DLOCK_READONLY)
945 if(flags & WINED3DLOCK_DISCARD)
947 WARN("WINED3DLOCK_READONLY combined with WINED3DLOCK_DISCARD, ignoring flags\n");
950 if(flags & WINED3DLOCK_NOOVERWRITE)
952 WARN("WINED3DLOCK_READONLY combined with WINED3DLOCK_NOOVERWRITE, ignoring flags\n");
956 else if((flags & (WINED3DLOCK_DISCARD | WINED3DLOCK_NOOVERWRITE)) == (WINED3DLOCK_DISCARD | WINED3DLOCK_NOOVERWRITE))
958 WARN("WINED3DLOCK_DISCARD and WINED3DLOCK_NOOVERWRITE used together, ignoring\n");
961 else if (flags & (WINED3DLOCK_DISCARD | WINED3DLOCK_NOOVERWRITE) && !(buffer->resource.usage & WINED3DUSAGE_DYNAMIC))
963 WARN("DISCARD or NOOVERWRITE lock on non-dynamic buffer, ignoring\n");
970 static GLbitfield buffer_gl_map_flags(DWORD d3d_flags)
974 if (!(d3d_flags & WINED3DLOCK_READONLY))
975 ret |= GL_MAP_WRITE_BIT | GL_MAP_FLUSH_EXPLICIT_BIT;
976 if (!(d3d_flags & (WINED3DLOCK_DISCARD | WINED3DLOCK_NOOVERWRITE)))
977 ret |= GL_MAP_READ_BIT;
979 if (d3d_flags & WINED3DLOCK_DISCARD)
980 ret |= GL_MAP_INVALIDATE_BUFFER_BIT;
981 if (d3d_flags & WINED3DLOCK_NOOVERWRITE)
982 ret |= GL_MAP_UNSYNCHRONIZED_BIT;
987 struct wined3d_resource * CDECL wined3d_buffer_get_resource(struct wined3d_buffer *buffer)
989 TRACE("buffer %p.\n", buffer);
991 return &buffer->resource;
994 HRESULT CDECL wined3d_buffer_map(struct wined3d_buffer *buffer, UINT offset, UINT size, BYTE **data, DWORD flags)
996 BOOL dirty = buffer_is_dirty(buffer);
999 TRACE("buffer %p, offset %u, size %u, data %p, flags %#x\n", buffer, offset, size, data, flags);
1001 flags = buffer_sanitize_flags(buffer, flags);
1002 if (!(flags & WINED3DLOCK_READONLY))
1004 if (!buffer_add_dirty_area(buffer, offset, size)) return E_OUTOFMEMORY;
1007 count = InterlockedIncrement(&buffer->lock_count);
1009 if (buffer->buffer_object)
1011 if (!(buffer->flags & WINED3D_BUFFER_DOUBLEBUFFER))
1015 struct wined3d_device *device = buffer->resource.device;
1016 struct wined3d_context *context;
1017 const struct wined3d_gl_info *gl_info;
1019 context = context_acquire(device, NULL);
1020 gl_info = context->gl_info;
1024 if (buffer->buffer_type_hint == GL_ELEMENT_ARRAY_BUFFER_ARB)
1025 context_invalidate_state(context, STATE_INDEXBUFFER);
1026 GL_EXTCALL(glBindBufferARB(buffer->buffer_type_hint, buffer->buffer_object));
1028 if (gl_info->supported[ARB_MAP_BUFFER_RANGE])
1030 GLbitfield mapflags = buffer_gl_map_flags(flags);
1031 buffer->resource.allocatedMemory = GL_EXTCALL(glMapBufferRange(buffer->buffer_type_hint,
1032 0, buffer->resource.size, mapflags));
1033 checkGLcall("glMapBufferRange");
1037 if (buffer->flags & WINED3D_BUFFER_APPLESYNC)
1040 buffer_sync_apple(buffer, flags, gl_info);
1043 buffer->resource.allocatedMemory = GL_EXTCALL(glMapBufferARB(buffer->buffer_type_hint,
1044 GL_READ_WRITE_ARB));
1045 checkGLcall("glMapBufferARB");
1049 if (((DWORD_PTR)buffer->resource.allocatedMemory) & (RESOURCE_ALIGNMENT - 1))
1051 WARN("Pointer %p is not %u byte aligned.\n", buffer->resource.allocatedMemory, RESOURCE_ALIGNMENT);
1054 GL_EXTCALL(glUnmapBufferARB(buffer->buffer_type_hint));
1055 checkGLcall("glUnmapBufferARB");
1057 buffer->resource.allocatedMemory = NULL;
1059 if (buffer->resource.usage & WINED3DUSAGE_DYNAMIC)
1061 /* The extra copy is more expensive than not using VBOs at
1062 * all on the Nvidia Linux driver, which is the only driver
1063 * that returns unaligned pointers
1065 TRACE("Dynamic buffer, dropping VBO\n");
1066 buffer_unload(&buffer->resource);
1067 buffer->flags &= ~WINED3D_BUFFER_CREATEBO;
1068 if (buffer->bind_count)
1069 device_invalidate_state(device, STATE_STREAMSRC);
1073 TRACE("Falling back to doublebuffered operation\n");
1074 buffer_get_sysmem(buffer, gl_info);
1076 TRACE("New pointer is %p.\n", buffer->resource.allocatedMemory);
1078 context_release(context);
1085 if (buffer->flags & WINED3D_BUFFER_NOSYNC && !(flags & WINED3DLOCK_NOOVERWRITE))
1087 buffer->flags &= ~WINED3D_BUFFER_NOSYNC;
1090 else if(flags & WINED3DLOCK_NOOVERWRITE)
1092 buffer->flags |= WINED3D_BUFFER_NOSYNC;
1095 if (flags & WINED3DLOCK_DISCARD)
1097 buffer->flags |= WINED3D_BUFFER_DISCARD;
1102 *data = buffer->resource.allocatedMemory + offset;
1104 TRACE("Returning memory at %p (base %p, offset %u).\n", *data, buffer->resource.allocatedMemory, offset);
1105 /* TODO: check Flags compatibility with buffer->currentDesc.Usage (see MSDN) */
1110 void CDECL wined3d_buffer_unmap(struct wined3d_buffer *buffer)
1114 TRACE("buffer %p.\n", buffer);
1116 /* In the case that the number of Unmap calls > the
1117 * number of Map calls, d3d returns always D3D_OK.
1118 * This is also needed to prevent Map from returning garbage on
1119 * the next call (this will happen if the lock_count is < 0). */
1120 if (!buffer->lock_count)
1122 WARN("Unmap called without a previous map call.\n");
1126 if (InterlockedDecrement(&buffer->lock_count))
1128 /* Delay loading the buffer until everything is unlocked */
1129 TRACE("Ignoring unmap.\n");
1133 if (!(buffer->flags & WINED3D_BUFFER_DOUBLEBUFFER) && buffer->buffer_object)
1135 struct wined3d_device *device = buffer->resource.device;
1136 const struct wined3d_gl_info *gl_info;
1137 struct wined3d_context *context;
1139 context = context_acquire(device, NULL);
1140 gl_info = context->gl_info;
1144 if (buffer->buffer_type_hint == GL_ELEMENT_ARRAY_BUFFER_ARB)
1145 context_invalidate_state(context, STATE_INDEXBUFFER);
1146 GL_EXTCALL(glBindBufferARB(buffer->buffer_type_hint, buffer->buffer_object));
1148 if (gl_info->supported[ARB_MAP_BUFFER_RANGE])
1150 for (i = 0; i < buffer->modified_areas; ++i)
1152 GL_EXTCALL(glFlushMappedBufferRange(buffer->buffer_type_hint,
1153 buffer->maps[i].offset, buffer->maps[i].size));
1154 checkGLcall("glFlushMappedBufferRange");
1157 else if (buffer->flags & WINED3D_BUFFER_FLUSH)
1159 for (i = 0; i < buffer->modified_areas; ++i)
1161 GL_EXTCALL(glFlushMappedBufferRangeAPPLE(buffer->buffer_type_hint,
1162 buffer->maps[i].offset, buffer->maps[i].size));
1163 checkGLcall("glFlushMappedBufferRangeAPPLE");
1167 GL_EXTCALL(glUnmapBufferARB(buffer->buffer_type_hint));
1169 if (wined3d_settings.strict_draw_ordering) wglFlush(); /* Flush to ensure ordering across contexts. */
1170 context_release(context);
1172 buffer->resource.allocatedMemory = NULL;
1173 buffer_clear_dirty_areas(buffer);
1175 else if (buffer->flags & WINED3D_BUFFER_HASDESC)
1177 wined3d_buffer_preload(buffer);
1181 static const struct wined3d_resource_ops buffer_resource_ops =
1186 static HRESULT buffer_init(struct wined3d_buffer *buffer, struct wined3d_device *device,
1187 UINT size, DWORD usage, enum wined3d_format_id format_id, WINED3DPOOL pool, GLenum bind_hint,
1188 const char *data, void *parent, const struct wined3d_parent_ops *parent_ops)
1190 const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
1191 const struct wined3d_format *format = wined3d_get_format(gl_info, format_id);
1193 BOOL dynamic_buffer_ok;
1197 WARN("Size 0 requested, returning WINED3DERR_INVALIDCALL\n");
1198 return WINED3DERR_INVALIDCALL;
1201 hr = resource_init(&buffer->resource, device, WINED3DRTYPE_BUFFER, format,
1202 WINED3DMULTISAMPLE_NONE, 0, usage, pool, size, 1, 1, size,
1203 parent, parent_ops, &buffer_resource_ops);
1206 WARN("Failed to initialize resource, hr %#x\n", hr);
1209 buffer->buffer_type_hint = bind_hint;
1211 TRACE("size %#x, usage %#x, format %s, memory @ %p, iface @ %p.\n", buffer->resource.size, buffer->resource.usage,
1212 debug_d3dformat(buffer->resource.format->id), buffer->resource.allocatedMemory, buffer);
1214 dynamic_buffer_ok = gl_info->supported[APPLE_FLUSH_BUFFER_RANGE] || gl_info->supported[ARB_MAP_BUFFER_RANGE];
1216 /* Observations show that drawStridedSlow is faster on dynamic VBs than converting +
1217 * drawStridedFast (half-life 2 and others).
1219 * Basically converting the vertices in the buffer is quite expensive, and observations
1220 * show that drawStridedSlow is faster than converting + uploading + drawStridedFast.
1221 * Therefore do not create a VBO for WINED3DUSAGE_DYNAMIC buffers.
1223 if (!gl_info->supported[ARB_VERTEX_BUFFER_OBJECT])
1225 TRACE("Not creating a vbo because GL_ARB_vertex_buffer is not supported\n");
1227 else if(buffer->resource.pool == WINED3DPOOL_SYSTEMMEM)
1229 TRACE("Not creating a vbo because the vertex buffer is in system memory\n");
1231 else if(!dynamic_buffer_ok && (buffer->resource.usage & WINED3DUSAGE_DYNAMIC))
1233 TRACE("Not creating a vbo because the buffer has dynamic usage and no GL support\n");
1237 buffer->flags |= WINED3D_BUFFER_CREATEBO;
1244 hr = wined3d_buffer_map(buffer, 0, size, &ptr, 0);
1247 ERR("Failed to map buffer, hr %#x\n", hr);
1248 buffer_unload(&buffer->resource);
1249 resource_cleanup(&buffer->resource);
1253 memcpy(ptr, data, size);
1255 wined3d_buffer_unmap(buffer);
1258 buffer->maps = HeapAlloc(GetProcessHeap(), 0, sizeof(*buffer->maps));
1261 ERR("Out of memory\n");
1262 buffer_unload(&buffer->resource);
1263 resource_cleanup(&buffer->resource);
1264 return E_OUTOFMEMORY;
1266 buffer->maps_size = 1;
1271 HRESULT CDECL wined3d_buffer_create(struct wined3d_device *device, struct wined3d_buffer_desc *desc, const void *data,
1272 void *parent, const struct wined3d_parent_ops *parent_ops, struct wined3d_buffer **buffer)
1274 struct wined3d_buffer *object;
1277 TRACE("device %p, desc %p, data %p, parent %p, buffer %p\n", device, desc, data, parent, buffer);
1279 object = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*object));
1282 ERR("Failed to allocate memory\n");
1283 return E_OUTOFMEMORY;
1286 FIXME("Ignoring access flags (pool)\n");
1288 hr = buffer_init(object, device, desc->byte_width, desc->usage, WINED3DFMT_UNKNOWN,
1289 WINED3DPOOL_MANAGED, GL_ARRAY_BUFFER_ARB, data, parent, parent_ops);
1292 WARN("Failed to initialize buffer, hr %#x.\n", hr);
1293 HeapFree(GetProcessHeap(), 0, object);
1296 object->desc = *desc;
1298 TRACE("Created buffer %p.\n", object);
1305 HRESULT CDECL wined3d_buffer_create_vb(struct wined3d_device *device, UINT size, DWORD usage, WINED3DPOOL pool,
1306 void *parent, const struct wined3d_parent_ops *parent_ops, struct wined3d_buffer **buffer)
1308 struct wined3d_buffer *object;
1311 TRACE("device %p, size %u, usage %#x, pool %#x, parent %p, parent_ops %p, buffer %p.\n",
1312 device, size, usage, pool, parent, parent_ops, buffer);
1314 if (pool == WINED3DPOOL_SCRATCH)
1316 /* The d3d9 tests shows that this is not allowed. It doesn't make much
1317 * sense anyway, SCRATCH buffers wouldn't be usable anywhere. */
1318 WARN("Vertex buffer in WINED3DPOOL_SCRATCH requested, returning WINED3DERR_INVALIDCALL.\n");
1320 return WINED3DERR_INVALIDCALL;
1323 object = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*object));
1326 ERR("Out of memory\n");
1328 return WINED3DERR_OUTOFVIDEOMEMORY;
1331 hr = buffer_init(object, device, size, usage, WINED3DFMT_VERTEXDATA,
1332 pool, GL_ARRAY_BUFFER_ARB, NULL, parent, parent_ops);
1335 WARN("Failed to initialize buffer, hr %#x.\n", hr);
1336 HeapFree(GetProcessHeap(), 0, object);
1340 TRACE("Created buffer %p.\n", object);
1346 HRESULT CDECL wined3d_buffer_create_ib(struct wined3d_device *device, UINT size, DWORD usage, WINED3DPOOL pool,
1347 void *parent, const struct wined3d_parent_ops *parent_ops, struct wined3d_buffer **buffer)
1349 struct wined3d_buffer *object;
1352 TRACE("device %p, size %u, usage %#x, pool %#x, parent %p, parent_ops %p, buffer %p.\n",
1353 device, size, usage, pool, parent, parent_ops, buffer);
1355 object = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*object));
1358 ERR("Out of memory\n");
1360 return WINED3DERR_OUTOFVIDEOMEMORY;
1363 hr = buffer_init(object, device, size, usage | WINED3DUSAGE_STATICDECL,
1364 WINED3DFMT_UNKNOWN, pool, GL_ELEMENT_ARRAY_BUFFER_ARB, NULL,
1365 parent, parent_ops);
1368 WARN("Failed to initialize buffer, hr %#x\n", hr);
1369 HeapFree(GetProcessHeap(), 0, object);
1373 TRACE("Created buffer %p.\n", object);