git.oblomov.eu Git - wine/blob - dlls/wined3d/buffer.c

   1 /*
   2  * Copyright 2002-2005 Jason Edmeades
   3  * Copyright 2002-2005 Raphael Junqueira
   4  * Copyright 2004 Christian Costa
   5  * Copyright 2005 Oliver Stieber
   6  * Copyright 2007-2010 Stefan Dösinger for CodeWeavers
   7  * Copyright 2009-2010 Henri Verbeet for CodeWeavers
   8  *
   9  * This library is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public
  11  * License as published by the Free Software Foundation; either
  12  * version 2.1 of the License, or (at your option) any later version.
  13  *
  14  * This library is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with this library; if not, write to the Free Software
  21  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301, USA
  22  *
  23  */
  24
  25 #include "config.h"
  26 #include "wine/port.h"
  27
  28 #include "wined3d_private.h"
  29
  30 WINE_DEFAULT_DEBUG_CHANNEL(d3d);
  31
  32 #define VB_MAXDECLCHANGES     100     /* After that number of decl changes we stop converting */
  33 #define VB_RESETDECLCHANGE    1000    /* Reset the decl changecount after that number of draws */
  34 #define VB_MAXFULLCONVERSIONS 5       /* Number of full conversions before we stop converting */
  35 #define VB_RESETFULLCONVS     20      /* Reset full conversion counts after that number of draws */
  36
  37 static inline BOOL buffer_add_dirty_area(struct wined3d_buffer *This, UINT offset, UINT size)
  38 {
  39     if (!This->buffer_object) return TRUE;
  40
  41     if (This->maps_size <= This->modified_areas)
  42     {
  43         void *new = HeapReAlloc(GetProcessHeap(), 0, This->maps,
  44                                 This->maps_size * 2 * sizeof(*This->maps));
  45         if (!new)
  46         {
  47             ERR("Out of memory\n");
  48             return FALSE;
  49         }
  50         else
  51         {
  52             This->maps = new;
  53             This->maps_size *= 2;
  54         }
  55     }
  56
  57     if(offset > This->resource.size || offset + size > This->resource.size)
  58     {
  59         WARN("Invalid range dirtified, marking entire buffer dirty\n");
  60         offset = 0;
  61         size = This->resource.size;
  62     }
  63     else if(!offset && !size)
  64     {
  65         size = This->resource.size;
  66     }
  67
  68     This->maps[This->modified_areas].offset = offset;
  69     This->maps[This->modified_areas].size = size;
  70     This->modified_areas++;
  71     return TRUE;
  72 }
  73
  74 static inline void buffer_clear_dirty_areas(struct wined3d_buffer *This)
  75 {
  76     This->modified_areas = 0;
  77 }
  78
  79 static inline BOOL buffer_is_dirty(struct wined3d_buffer *This)
  80 {
  81     return !!This->modified_areas;
  82 }
  83
  84 static inline BOOL buffer_is_fully_dirty(struct wined3d_buffer *This)
  85 {
  86     unsigned int i;
  87
  88     for(i = 0; i < This->modified_areas; i++)
  89     {
  90         if (!This->maps[i].offset && This->maps[i].size == This->resource.size)
  91         {
  92             return TRUE;
  93         }
  94     }
  95     return FALSE;
  96 }
  97
  98 /* Context activation is done by the caller */
  99 static void delete_gl_buffer(struct wined3d_buffer *This, const struct wined3d_gl_info *gl_info)
 100 {
 101     if(!This->buffer_object) return;
 102
 103     ENTER_GL();
 104     GL_EXTCALL(glDeleteBuffersARB(1, &This->buffer_object));
 105     checkGLcall("glDeleteBuffersARB");
 106     LEAVE_GL();
 107     This->buffer_object = 0;
 108
 109     if(This->query)
 110     {
 111         wined3d_event_query_destroy(This->query);
 112         This->query = NULL;
 113     }
 114     This->flags &= ~WINED3D_BUFFER_APPLESYNC;
 115 }
 116
 117 /* Context activation is done by the caller. */
 118 static void buffer_create_buffer_object(struct wined3d_buffer *This, const struct wined3d_gl_info *gl_info)
 119 {
 120     GLenum error, gl_usage;
 121
 122     TRACE("Creating an OpenGL vertex buffer object for IWineD3DVertexBuffer %p Usage(%s)\n",
 123             This, debug_d3dusage(This->resource.usage));
 124
 125     ENTER_GL();
 126
 127     /* Make sure that the gl error is cleared. Do not use checkGLcall
 128     * here because checkGLcall just prints a fixme and continues. However,
 129     * if an error during VBO creation occurs we can fall back to non-vbo operation
 130     * with full functionality(but performance loss)
 131     */
 132     while (glGetError() != GL_NO_ERROR);
 133
 134     /* Basically the FVF parameter passed to CreateVertexBuffer is no good
 135      * It is the FVF set with IWineD3DDevice::SetFVF or the Vertex Declaration set with
 136      * IWineD3DDevice::SetVertexDeclaration that decides how the vertices in the buffer
 137      * look like. This means that on each DrawPrimitive call the vertex buffer has to be verified
 138      * to check if the rhw and color values are in the correct format.
 139      */
 140
 141     GL_EXTCALL(glGenBuffersARB(1, &This->buffer_object));
 142     error = glGetError();
 143     if (!This->buffer_object || error != GL_NO_ERROR)
 144     {
 145         ERR("Failed to create a VBO with error %s (%#x)\n", debug_glerror(error), error);
 146         LEAVE_GL();
 147         goto fail;
 148     }
 149
 150     if(This->buffer_type_hint == GL_ELEMENT_ARRAY_BUFFER_ARB)
 151     {
 152         IWineD3DDeviceImpl_MarkStateDirty(This->resource.device, STATE_INDEXBUFFER);
 153     }
 154     GL_EXTCALL(glBindBufferARB(This->buffer_type_hint, This->buffer_object));
 155     error = glGetError();
 156     if (error != GL_NO_ERROR)
 157     {
 158         ERR("Failed to bind the VBO with error %s (%#x)\n", debug_glerror(error), error);
 159         LEAVE_GL();
 160         goto fail;
 161     }
 162
 163     /* Don't use static, because dx apps tend to update the buffer
 164      * quite often even if they specify 0 usage.
 165      */
 166     if(This->resource.usage & WINED3DUSAGE_DYNAMIC)
 167     {
 168         TRACE("Gl usage = GL_STREAM_DRAW_ARB\n");
 169         gl_usage = GL_STREAM_DRAW_ARB;
 170
 171         if(gl_info->supported[APPLE_FLUSH_BUFFER_RANGE])
 172         {
 173             GL_EXTCALL(glBufferParameteriAPPLE(This->buffer_type_hint, GL_BUFFER_FLUSHING_UNMAP_APPLE, GL_FALSE));
 174             checkGLcall("glBufferParameteriAPPLE(This->buffer_type_hint, GL_BUFFER_FLUSHING_UNMAP_APPLE, GL_FALSE)");
 175             This->flags |= WINED3D_BUFFER_FLUSH;
 176
 177             GL_EXTCALL(glBufferParameteriAPPLE(This->buffer_type_hint, GL_BUFFER_SERIALIZED_MODIFY_APPLE, GL_FALSE));
 178             checkGLcall("glBufferParameteriAPPLE(This->buffer_type_hint, GL_BUFFER_SERIALIZED_MODIFY_APPLE, GL_FALSE)");
 179             This->flags |= WINED3D_BUFFER_APPLESYNC;
 180         }
 181         /* No setup is needed here for GL_ARB_map_buffer_range */
 182     }
 183     else
 184     {
 185         TRACE("Gl usage = GL_DYNAMIC_DRAW_ARB\n");
 186         gl_usage = GL_DYNAMIC_DRAW_ARB;
 187     }
 188
 189     /* Reserve memory for the buffer. The amount of data won't change
 190      * so we are safe with calling glBufferData once and
 191      * calling glBufferSubData on updates. Upload the actual data in case
 192      * we're not double buffering, so we can release the heap mem afterwards
 193      */
 194     GL_EXTCALL(glBufferDataARB(This->buffer_type_hint, This->resource.size, This->resource.allocatedMemory, gl_usage));
 195     error = glGetError();
 196     LEAVE_GL();
 197     if (error != GL_NO_ERROR)
 198     {
 199         ERR("glBufferDataARB failed with error %s (%#x)\n", debug_glerror(error), error);
 200         goto fail;
 201     }
 202
 203     This->buffer_object_size = This->resource.size;
 204     This->buffer_object_usage = gl_usage;
 205
 206     if(This->flags & WINED3D_BUFFER_DOUBLEBUFFER)
 207     {
 208         if(!buffer_add_dirty_area(This, 0, 0))
 209         {
 210             ERR("buffer_add_dirty_area failed, this is not expected\n");
 211             goto fail;
 212         }
 213     }
 214     else
 215     {
 216         HeapFree(GetProcessHeap(), 0, This->resource.heapMemory);
 217         This->resource.allocatedMemory = NULL;
 218         This->resource.heapMemory = NULL;
 219     }
 220
 221     return;
 222
 223 fail:
 224     /* Clean up all vbo init, but continue because we can work without a vbo :-) */
 225     ERR("Failed to create a vertex buffer object. Continuing, but performance issues may occur\n");
 226     delete_gl_buffer(This, gl_info);
 227     buffer_clear_dirty_areas(This);
 228 }
 229
 230 static BOOL buffer_process_converted_attribute(struct wined3d_buffer *This,
 231         const enum wined3d_buffer_conversion_type conversion_type,
 232         const struct wined3d_stream_info_element *attrib, DWORD *stride_this_run)
 233 {
 234     DWORD offset = This->resource.device->stateBlock->state.streams[attrib->stream_idx].offset;
 235     DWORD attrib_size;
 236     BOOL ret = FALSE;
 237     unsigned int i;
 238     DWORD_PTR data;
 239
 240     /* Check for some valid situations which cause us pain. One is if the buffer is used for
 241      * constant attributes(stride = 0), the other one is if the buffer is used on two streams
 242      * with different strides. In the 2nd case we might have to drop conversion entirely,
 243      * it is possible that the same bytes are once read as FLOAT2 and once as UBYTE4N.
 244      */
 245     if (!attrib->stride)
 246     {
 247         FIXME("%s used with stride 0, let's hope we get the vertex stride from somewhere else\n",
 248                 debug_d3dformat(attrib->format->id));
 249     }
 250     else if(attrib->stride != *stride_this_run && *stride_this_run)
 251     {
 252         FIXME("Got two concurrent strides, %d and %d\n", attrib->stride, *stride_this_run);
 253     }
 254     else
 255     {
 256         *stride_this_run = attrib->stride;
 257         if (This->stride != *stride_this_run)
 258         {
 259             /* We rely that this happens only on the first converted attribute that is found,
 260              * if at all. See above check
 261              */
 262             TRACE("Reconverting because converted attributes occur, and the stride changed\n");
 263             This->stride = *stride_this_run;
 264             HeapFree(GetProcessHeap(), HEAP_ZERO_MEMORY, This->conversion_map);
 265             This->conversion_map = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY,
 266                     sizeof(*This->conversion_map) * This->stride);
 267             ret = TRUE;
 268         }
 269     }
 270
 271     data = (((DWORD_PTR)attrib->data) + offset) % This->stride;
 272     attrib_size = attrib->format->component_count * attrib->format->component_size;
 273     for (i = 0; i < attrib_size; ++i)
 274     {
 275         DWORD_PTR idx = (data + i) % This->stride;
 276         if (This->conversion_map[idx] != conversion_type)
 277         {
 278             TRACE("Byte %ld in vertex changed\n", idx);
 279             TRACE("It was type %d, is %d now\n", This->conversion_map[idx], conversion_type);
 280             ret = TRUE;
 281             This->conversion_map[idx] = conversion_type;
 282         }
 283     }
 284
 285     return ret;
 286 }
 287
 288 static BOOL buffer_check_attribute(struct wined3d_buffer *This, const struct wined3d_stream_info *si,
 289         UINT attrib_idx, const BOOL check_d3dcolor, const BOOL is_ffp_position, const BOOL is_ffp_color,
 290         DWORD *stride_this_run, BOOL *float16_used)
 291 {
 292     const struct wined3d_stream_info_element *attrib = &si->elements[attrib_idx];
 293     IWineD3DDeviceImpl *device = This->resource.device;
 294     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
 295     enum wined3d_format_id format;
 296     BOOL ret = FALSE;
 297
 298     /* Ignore attributes that do not have our vbo. After that check we can be sure that the attribute is
 299      * there, on nonexistent attribs the vbo is 0.
 300      */
 301     if (!(si->use_map & (1 << attrib_idx))
 302             || attrib->buffer_object != This->buffer_object)
 303         return FALSE;
 304
 305     format = attrib->format->id;
 306     /* Look for newly appeared conversion */
 307     if (!gl_info->supported[ARB_HALF_FLOAT_VERTEX]
 308             && (format == WINED3DFMT_R16G16_FLOAT || format == WINED3DFMT_R16G16B16A16_FLOAT))
 309     {
 310         ret = buffer_process_converted_attribute(This, CONV_FLOAT16_2, attrib, stride_this_run);
 311
 312         if (is_ffp_position) FIXME("Test FLOAT16 fixed function processing positions\n");
 313         else if (is_ffp_color) FIXME("test FLOAT16 fixed function processing colors\n");
 314         *float16_used = TRUE;
 315     }
 316     else if (check_d3dcolor && format == WINED3DFMT_B8G8R8A8_UNORM)
 317     {
 318         ret = buffer_process_converted_attribute(This, CONV_D3DCOLOR, attrib, stride_this_run);
 319
 320         if (!is_ffp_color) FIXME("Test for non-color fixed function WINED3DFMT_B8G8R8A8_UNORM format\n");
 321     }
 322     else if (is_ffp_position && format == WINED3DFMT_R32G32B32A32_FLOAT)
 323     {
 324         ret = buffer_process_converted_attribute(This, CONV_POSITIONT, attrib, stride_this_run);
 325     }
 326     else if (This->conversion_map)
 327     {
 328         ret = buffer_process_converted_attribute(This, CONV_NONE, attrib, stride_this_run);
 329     }
 330
 331     return ret;
 332 }
 333
 334 static UINT *find_conversion_shift(struct wined3d_buffer *This,
 335         const struct wined3d_stream_info *strided, UINT stride)
 336 {
 337     UINT *ret, i, j, shift, orig_type_size;
 338
 339     if (!stride)
 340     {
 341         TRACE("No shift\n");
 342         return NULL;
 343     }
 344
 345     This->conversion_stride = stride;
 346     ret = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(DWORD) * stride);
 347     for (i = 0; i < MAX_ATTRIBS; ++i)
 348     {
 349         enum wined3d_format_id format;
 350
 351         if (!(strided->use_map & (1 << i)) || strided->elements[i].buffer_object != This->buffer_object) continue;
 352
 353         format = strided->elements[i].format->id;
 354         if (format == WINED3DFMT_R16G16_FLOAT)
 355         {
 356             shift = 4;
 357         }
 358         else if (format == WINED3DFMT_R16G16B16A16_FLOAT)
 359         {
 360             shift = 8;
 361             /* Pre-shift the last 4 bytes in the FLOAT16_4 by 4 bytes - this makes FLOAT16_2 and FLOAT16_4 conversions
 362              * compatible
 363              */
 364             for (j = 4; j < 8; ++j)
 365             {
 366                 ret[(DWORD_PTR)strided->elements[i].data + j] += 4;
 367             }
 368         }
 369         else
 370         {
 371             shift = 0;
 372         }
 373         This->conversion_stride += shift;
 374
 375         if (shift)
 376         {
 377             orig_type_size = strided->elements[i].format->component_count
 378                     * strided->elements[i].format->component_size;
 379             for (j = (DWORD_PTR)strided->elements[i].data + orig_type_size; j < stride; ++j)
 380             {
 381                 ret[j] += shift;
 382             }
 383         }
 384     }
 385
 386     if (TRACE_ON(d3d))
 387     {
 388         TRACE("Dumping conversion shift:\n");
 389         for (i = 0; i < stride; ++i)
 390         {
 391             TRACE("[%d]", ret[i]);
 392         }
 393         TRACE("\n");
 394     }
 395
 396     return ret;
 397 }
 398
 399 static BOOL buffer_find_decl(struct wined3d_buffer *This)
 400 {
 401     IWineD3DDeviceImpl *device = This->resource.device;
 402     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
 403     const struct wined3d_stream_info *si = &device->strided_streams;
 404     const struct wined3d_state *state = &device->stateBlock->state;
 405     UINT stride_this_run = 0;
 406     BOOL float16_used = FALSE;
 407     BOOL ret = FALSE;
 408     unsigned int i;
 409
 410     /* In d3d7 the vertex buffer declaration NEVER changes because it is stored in the d3d7 vertex buffer.
 411      * Once we have our declaration there is no need to look it up again. Index buffers also never need
 412      * conversion, so once the (empty) conversion structure is created don't bother checking again
 413      */
 414     if (This->flags & WINED3D_BUFFER_HASDESC)
 415     {
 416         if(This->resource.usage & WINED3DUSAGE_STATICDECL) return FALSE;
 417     }
 418
 419     TRACE("Finding vertex buffer conversion information\n");
 420     /* Certain declaration types need some fixups before we can pass them to
 421      * opengl. This means D3DCOLOR attributes with fixed function vertex
 422      * processing, FLOAT4 POSITIONT with fixed function, and FLOAT16 if
 423      * GL_ARB_half_float_vertex is not supported.
 424      *
 425      * Note for d3d8 and d3d9:
 426      * The vertex buffer FVF doesn't help with finding them, we have to use
 427      * the decoded vertex declaration and pick the things that concern the
 428      * current buffer. A problem with this is that this can change between
 429      * draws, so we have to validate the information and reprocess the buffer
 430      * if it changes, and avoid false positives for performance reasons.
 431      * WineD3D doesn't even know the vertex buffer any more, it is managed
 432      * by the client libraries and passed to SetStreamSource and ProcessVertices
 433      * as needed.
 434      *
 435      * We have to distinguish between vertex shaders and fixed function to
 436      * pick the way we access the strided vertex information.
 437      *
 438      * This code sets up a per-byte array with the size of the detected
 439      * stride of the arrays in the buffer. For each byte we have a field
 440      * that marks the conversion needed on this byte. For example, the
 441      * following declaration with fixed function vertex processing:
 442      *
 443      *      POSITIONT, FLOAT4
 444      *      NORMAL, FLOAT3
 445      *      DIFFUSE, FLOAT16_4
 446      *      SPECULAR, D3DCOLOR
 447      *
 448      * Will result in
 449      * {                 POSITIONT                    }{             NORMAL                }{    DIFFUSE          }{SPECULAR }
 450      * [P][P][P][P][P][P][P][P][P][P][P][P][P][P][P][P][0][0][0][0][0][0][0][0][0][0][0][0][F][F][F][F][F][F][F][F][C][C][C][C]
 451      *
 452      * Where in this example map P means 4 component position conversion, 0
 453      * means no conversion, F means FLOAT16_2 conversion and C means D3DCOLOR
 454      * conversion (red / blue swizzle).
 455      *
 456      * If we're doing conversion and the stride changes we have to reconvert
 457      * the whole buffer. Note that we do not mind if the semantic changes,
 458      * we only care for the conversion type. So if the NORMAL is replaced
 459      * with a TEXCOORD, nothing has to be done, or if the DIFFUSE is replaced
 460      * with a D3DCOLOR BLENDWEIGHT we can happily dismiss the change. Some
 461      * conversion types depend on the semantic as well, for example a FLOAT4
 462      * texcoord needs no conversion while a FLOAT4 positiont needs one
 463      */
 464     if (use_vs(state))
 465     {
 466         TRACE("vshader\n");
 467         /* If the current vertex declaration is marked for no half float conversion don't bother to
 468          * analyse the strided streams in depth, just set them up for no conversion. Return decl changed
 469          * if we used conversion before
 470          */
 471         if (!state->vertex_declaration->half_float_conv_needed)
 472         {
 473             if (This->conversion_map)
 474             {
 475                 TRACE("Now using shaders without conversion, but conversion used before\n");
 476                 HeapFree(GetProcessHeap(), 0, This->conversion_map);
 477                 HeapFree(GetProcessHeap(), 0, This->conversion_shift);
 478                 This->conversion_map = NULL;
 479                 This->stride = 0;
 480                 This->conversion_shift = NULL;
 481                 This->conversion_stride = 0;
 482                 return TRUE;
 483             }
 484             else
 485             {
 486                 return FALSE;
 487             }
 488         }
 489         for (i = 0; i < MAX_ATTRIBS; ++i)
 490         {
 491             ret = buffer_check_attribute(This, si, i, FALSE, FALSE, FALSE, &stride_this_run, &float16_used) || ret;
 492         }
 493
 494         /* Recalculate the conversion shift map if the declaration has changed,
 495          * and we're using float16 conversion or used it on the last run
 496          */
 497         if (ret && (float16_used || This->conversion_map))
 498         {
 499             HeapFree(GetProcessHeap(), 0, This->conversion_shift);
 500             This->conversion_shift = find_conversion_shift(This, si, This->stride);
 501         }
 502     }
 503     else
 504     {
 505         /* Fixed function is a bit trickier. We have to take care for D3DCOLOR types, FLOAT4 positions and of course
 506          * FLOAT16s if not supported. Also, we can't iterate over the array, so use macros to generate code for all
 507          * the attributes that our current fixed function pipeline implementation cares for.
 508          */
 509         BOOL support_d3dcolor = gl_info->supported[ARB_VERTEX_ARRAY_BGRA];
 510         ret = buffer_check_attribute(This, si, WINED3D_FFP_POSITION,
 511                 TRUE, TRUE,  FALSE, &stride_this_run, &float16_used) || ret;
 512         ret = buffer_check_attribute(This, si, WINED3D_FFP_NORMAL,
 513                 TRUE, FALSE, FALSE, &stride_this_run, &float16_used) || ret;
 514         ret = buffer_check_attribute(This, si, WINED3D_FFP_DIFFUSE,
 515                 !support_d3dcolor, FALSE, TRUE,  &stride_this_run, &float16_used) || ret;
 516         ret = buffer_check_attribute(This, si, WINED3D_FFP_SPECULAR,
 517                 !support_d3dcolor, FALSE, TRUE,  &stride_this_run, &float16_used) || ret;
 518         ret = buffer_check_attribute(This, si, WINED3D_FFP_TEXCOORD0,
 519                 TRUE, FALSE, FALSE, &stride_this_run, &float16_used) || ret;
 520         ret = buffer_check_attribute(This, si, WINED3D_FFP_TEXCOORD1,
 521                 TRUE, FALSE, FALSE, &stride_this_run, &float16_used) || ret;
 522         ret = buffer_check_attribute(This, si, WINED3D_FFP_TEXCOORD2,
 523                 TRUE, FALSE, FALSE, &stride_this_run, &float16_used) || ret;
 524         ret = buffer_check_attribute(This, si, WINED3D_FFP_TEXCOORD3,
 525                 TRUE, FALSE, FALSE, &stride_this_run, &float16_used) || ret;
 526         ret = buffer_check_attribute(This, si, WINED3D_FFP_TEXCOORD4,
 527                 TRUE, FALSE, FALSE, &stride_this_run, &float16_used) || ret;
 528         ret = buffer_check_attribute(This, si, WINED3D_FFP_TEXCOORD5,
 529                 TRUE, FALSE, FALSE, &stride_this_run, &float16_used) || ret;
 530         ret = buffer_check_attribute(This, si, WINED3D_FFP_TEXCOORD6,
 531                 TRUE, FALSE, FALSE, &stride_this_run, &float16_used) || ret;
 532         ret = buffer_check_attribute(This, si, WINED3D_FFP_TEXCOORD7,
 533                 TRUE, FALSE, FALSE, &stride_this_run, &float16_used) || ret;
 534
 535         if (float16_used) FIXME("Float16 conversion used with fixed function vertex processing\n");
 536     }
 537
 538     if (!stride_this_run && This->conversion_map)
 539     {
 540         /* Sanity test */
 541         if (!ret) ERR("no converted attributes found, old conversion map exists, and no declaration change?\n");
 542         HeapFree(GetProcessHeap(), 0, This->conversion_map);
 543         This->conversion_map = NULL;
 544         This->stride = 0;
 545     }
 546
 547     if (ret) TRACE("Conversion information changed\n");
 548
 549     return ret;
 550 }
 551
 552 /* Context activation is done by the caller. */
 553 static void buffer_check_buffer_object_size(struct wined3d_buffer *This, const struct wined3d_gl_info *gl_info)
 554 {
 555     UINT size = This->conversion_stride ?
 556             This->conversion_stride * (This->resource.size / This->stride) : This->resource.size;
 557     if (This->buffer_object_size != size)
 558     {
 559         TRACE("Old size %u, creating new size %u\n", This->buffer_object_size, size);
 560
 561         if(This->buffer_type_hint == GL_ELEMENT_ARRAY_BUFFER_ARB)
 562         {
 563             IWineD3DDeviceImpl_MarkStateDirty(This->resource.device, STATE_INDEXBUFFER);
 564         }
 565
 566         /* Rescue the data before resizing the buffer object if we do not have our backup copy */
 567         if(!(This->flags & WINED3D_BUFFER_DOUBLEBUFFER))
 568         {
 569             buffer_get_sysmem(This, gl_info);
 570         }
 571
 572         ENTER_GL();
 573         GL_EXTCALL(glBindBufferARB(This->buffer_type_hint, This->buffer_object));
 574         checkGLcall("glBindBufferARB");
 575         GL_EXTCALL(glBufferDataARB(This->buffer_type_hint, size, NULL, This->buffer_object_usage));
 576         This->buffer_object_size = size;
 577         checkGLcall("glBufferDataARB");
 578         LEAVE_GL();
 579     }
 580 }
 581
 582 static inline void fixup_d3dcolor(DWORD *dst_color)
 583 {
 584     DWORD src_color = *dst_color;
 585
 586     /* Color conversion like in drawStridedSlow. watch out for little endianity
 587      * If we want that stuff to work on big endian machines too we have to consider more things
 588      *
 589      * 0xff000000: Alpha mask
 590      * 0x00ff0000: Blue mask
 591      * 0x0000ff00: Green mask
 592      * 0x000000ff: Red mask
 593      */
 594     *dst_color = 0;
 595     *dst_color |= (src_color & 0xff00ff00);         /* Alpha Green */
 596     *dst_color |= (src_color & 0x00ff0000) >> 16;   /* Red */
 597     *dst_color |= (src_color & 0x000000ff) << 16;   /* Blue */
 598 }
 599
 600 static inline void fixup_transformed_pos(float *p)
 601 {
 602     /* rhw conversion like in position_float4(). */
 603     if (p[3] != 1.0f && p[3] != 0.0f)
 604     {
 605         float w = 1.0f / p[3];
 606         p[0] *= w;
 607         p[1] *= w;
 608         p[2] *= w;
 609         p[3] = w;
 610     }
 611 }
 612
 613 /* Context activation is done by the caller. */
 614 const BYTE *buffer_get_memory(struct wined3d_buffer *buffer,
 615         const struct wined3d_gl_info *gl_info, GLuint *buffer_object)
 616 {
 617     *buffer_object = buffer->buffer_object;
 618     if (!buffer->buffer_object)
 619     {
 620         if (buffer->flags & WINED3D_BUFFER_CREATEBO)
 621         {
 622             buffer_create_buffer_object(buffer, gl_info);
 623             buffer->flags &= ~WINED3D_BUFFER_CREATEBO;
 624             if (buffer->buffer_object)
 625             {
 626                 *buffer_object = buffer->buffer_object;
 627                 return NULL;
 628             }
 629         }
 630         return buffer->resource.allocatedMemory;
 631     }
 632     else
 633     {
 634         return NULL;
 635     }
 636 }
 637
 638 /* IUnknown methods */
 639
 640 static HRESULT STDMETHODCALLTYPE buffer_QueryInterface(IWineD3DBuffer *iface,
 641         REFIID riid, void **object)
 642 {
 643     TRACE("iface %p, riid %s, object %p\n", iface, debugstr_guid(riid), object);
 644
 645     if (IsEqualGUID(riid, &IID_IWineD3DBuffer)
 646             || IsEqualGUID(riid, &IID_IWineD3DResource)
 647             || IsEqualGUID(riid, &IID_IWineD3DBase)
 648             || IsEqualGUID(riid, &IID_IUnknown))
 649     {
 650         IUnknown_AddRef(iface);
 651         *object = iface;
 652         return S_OK;
 653     }
 654
 655     WARN("%s not implemented, returning E_NOINTERFACE\n", debugstr_guid(riid));
 656
 657     *object = NULL;
 658
 659     return E_NOINTERFACE;
 660 }
 661
 662 static ULONG STDMETHODCALLTYPE buffer_AddRef(IWineD3DBuffer *iface)
 663 {
 664     struct wined3d_buffer *This = (struct wined3d_buffer *)iface;
 665     ULONG refcount = InterlockedIncrement(&This->resource.ref);
 666
 667     TRACE("%p increasing refcount to %u\n", This, refcount);
 668
 669     return refcount;
 670 }
 671
 672 /* Context activation is done by the caller. */
 673 BYTE *buffer_get_sysmem(struct wined3d_buffer *This, const struct wined3d_gl_info *gl_info)
 674 {
 675     /* AllocatedMemory exists if the buffer is double buffered or has no buffer object at all */
 676     if(This->resource.allocatedMemory) return This->resource.allocatedMemory;
 677
 678     This->resource.heapMemory = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, This->resource.size + RESOURCE_ALIGNMENT);
 679     This->resource.allocatedMemory = (BYTE *)(((ULONG_PTR)This->resource.heapMemory + (RESOURCE_ALIGNMENT - 1)) & ~(RESOURCE_ALIGNMENT - 1));
 680
 681     if (This->buffer_type_hint == GL_ELEMENT_ARRAY_BUFFER_ARB)
 682     {
 683         IWineD3DDeviceImpl_MarkStateDirty(This->resource.device, STATE_INDEXBUFFER);
 684     }
 685
 686     ENTER_GL();
 687     GL_EXTCALL(glBindBufferARB(This->buffer_type_hint, This->buffer_object));
 688     GL_EXTCALL(glGetBufferSubDataARB(This->buffer_type_hint, 0, This->resource.size, This->resource.allocatedMemory));
 689     LEAVE_GL();
 690     This->flags |= WINED3D_BUFFER_DOUBLEBUFFER;
 691
 692     return This->resource.allocatedMemory;
 693 }
 694
 695 /* Do not call while under the GL lock. */
 696 static void STDMETHODCALLTYPE buffer_UnLoad(IWineD3DBuffer *iface)
 697 {
 698     struct wined3d_buffer *This = (struct wined3d_buffer *)iface;
 699
 700     TRACE("iface %p\n", iface);
 701
 702     if (This->buffer_object)
 703     {
 704         IWineD3DDeviceImpl *device = This->resource.device;
 705         struct wined3d_context *context;
 706
 707         context = context_acquire(device, NULL);
 708
 709         /* Download the buffer, but don't permanently enable double buffering */
 710         if(!(This->flags & WINED3D_BUFFER_DOUBLEBUFFER))
 711         {
 712             buffer_get_sysmem(This, context->gl_info);
 713             This->flags &= ~WINED3D_BUFFER_DOUBLEBUFFER;
 714         }
 715
 716         delete_gl_buffer(This, context->gl_info);
 717         This->flags |= WINED3D_BUFFER_CREATEBO; /* Recreate the buffer object next load */
 718         buffer_clear_dirty_areas(This);
 719
 720         context_release(context);
 721
 722         HeapFree(GetProcessHeap(), 0, This->conversion_shift);
 723         This->conversion_shift = NULL;
 724         HeapFree(GetProcessHeap(), 0, This->conversion_map);
 725         This->conversion_map = NULL;
 726         This->stride = 0;
 727         This->conversion_stride = 0;
 728         This->flags &= ~WINED3D_BUFFER_HASDESC;
 729     }
 730
 731     resource_unload((IWineD3DResourceImpl *)This);
 732 }
 733
 734 /* Do not call while under the GL lock. */
 735 static ULONG STDMETHODCALLTYPE buffer_Release(IWineD3DBuffer *iface)
 736 {
 737     struct wined3d_buffer *This = (struct wined3d_buffer *)iface;
 738     ULONG refcount = InterlockedDecrement(&This->resource.ref);
 739
 740     TRACE("%p decreasing refcount to %u\n", This, refcount);
 741
 742     if (!refcount)
 743     {
 744         buffer_UnLoad(iface);
 745         resource_cleanup((IWineD3DResource *)iface);
 746         This->resource.parent_ops->wined3d_object_destroyed(This->resource.parent);
 747         HeapFree(GetProcessHeap(), 0, This->maps);
 748         HeapFree(GetProcessHeap(), 0, This);
 749     }
 750
 751     return refcount;
 752 }
 753
 754 /* IWineD3DBase methods */
 755 static void * STDMETHODCALLTYPE buffer_GetParent(IWineD3DBuffer *iface)
 756 {
 757     TRACE("iface %p.\n", iface);
 758
 759     return ((struct wined3d_buffer *)iface)->resource.parent;
 760 }
 761
 762 /* IWineD3DResource methods */
 763
 764 static HRESULT STDMETHODCALLTYPE buffer_SetPrivateData(IWineD3DBuffer *iface,
 765         REFGUID guid, const void *data, DWORD data_size, DWORD flags)
 766 {
 767     return resource_set_private_data((IWineD3DResource *)iface, guid, data, data_size, flags);
 768 }
 769
 770 static HRESULT STDMETHODCALLTYPE buffer_GetPrivateData(IWineD3DBuffer *iface,
 771         REFGUID guid, void *data, DWORD *data_size)
 772 {
 773     return resource_get_private_data((IWineD3DResource *)iface, guid, data, data_size);
 774 }
 775
 776 static HRESULT STDMETHODCALLTYPE buffer_FreePrivateData(IWineD3DBuffer *iface, REFGUID guid)
 777 {
 778     return resource_free_private_data((IWineD3DResource *)iface, guid);
 779 }
 780
 781 static DWORD STDMETHODCALLTYPE buffer_SetPriority(IWineD3DBuffer *iface, DWORD priority)
 782 {
 783     return resource_set_priority((IWineD3DResource *)iface, priority);
 784 }
 785
 786 static DWORD STDMETHODCALLTYPE buffer_GetPriority(IWineD3DBuffer *iface)
 787 {
 788     return resource_get_priority((IWineD3DResource *)iface);
 789 }
 790
 791 /* The caller provides a context and binds the buffer */
 792 static void buffer_sync_apple(struct wined3d_buffer *This, DWORD flags, const struct wined3d_gl_info *gl_info)
 793 {
 794     enum wined3d_event_query_result ret;
 795
 796     /* No fencing needs to be done if the app promises not to overwrite
 797      * existing data */
 798     if(flags & WINED3DLOCK_NOOVERWRITE) return;
 799     if(flags & WINED3DLOCK_DISCARD)
 800     {
 801         ENTER_GL();
 802         GL_EXTCALL(glBufferDataARB(This->buffer_type_hint, This->resource.size, NULL, This->buffer_object_usage));
 803         checkGLcall("glBufferDataARB\n");
 804         LEAVE_GL();
 805         return;
 806     }
 807
 808     if(!This->query)
 809     {
 810         TRACE("Creating event query for buffer %p\n", This);
 811
 812         if (!wined3d_event_query_supported(gl_info))
 813         {
 814             FIXME("Event queries not supported, dropping async buffer locks.\n");
 815             goto drop_query;
 816         }
 817
 818         This->query = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*This->query));
 819         if (!This->query)
 820         {
 821             ERR("Failed to allocate event query memory, dropping async buffer locks.\n");
 822             goto drop_query;
 823         }
 824
 825         /* Since we don't know about old draws a glFinish is needed once */
 826         wglFinish();
 827         return;
 828     }
 829     TRACE("Synchronizing buffer %p\n", This);
 830     ret = wined3d_event_query_finish(This->query, This->resource.device);
 831     switch(ret)
 832     {
 833         case WINED3D_EVENT_QUERY_NOT_STARTED:
 834         case WINED3D_EVENT_QUERY_OK:
 835             /* All done */
 836             return;
 837
 838         case WINED3D_EVENT_QUERY_WRONG_THREAD:
 839             WARN("Cannot synchronize buffer lock due to a thread conflict\n");
 840             goto drop_query;
 841
 842         default:
 843             ERR("wined3d_event_query_finish returned %u, dropping async buffer locks\n", ret);
 844             goto drop_query;
 845     }
 846
 847 drop_query:
 848     if(This->query)
 849     {
 850         wined3d_event_query_destroy(This->query);
 851         This->query = NULL;
 852     }
 853
 854     wglFinish();
 855     ENTER_GL();
 856     GL_EXTCALL(glBufferParameteriAPPLE(This->buffer_type_hint, GL_BUFFER_SERIALIZED_MODIFY_APPLE, GL_TRUE));
 857     checkGLcall("glBufferParameteriAPPLE(This->buffer_type_hint, GL_BUFFER_SERIALIZED_MODIFY_APPLE, GL_TRUE)");
 858     LEAVE_GL();
 859     This->flags &= ~WINED3D_BUFFER_APPLESYNC;
 860 }
 861
 862 /* The caller provides a GL context */
 863 static void buffer_direct_upload(struct wined3d_buffer *This, const struct wined3d_gl_info *gl_info, DWORD flags)
 864 {
 865     BYTE *map;
 866     UINT start = 0, len = 0;
 867
 868     ENTER_GL();
 869
 870     /* This potentially invalidates the element array buffer binding, but the
 871      * caller always takes care of this. */
 872     GL_EXTCALL(glBindBufferARB(This->buffer_type_hint, This->buffer_object));
 873     checkGLcall("glBindBufferARB");
 874     if (gl_info->supported[ARB_MAP_BUFFER_RANGE])
 875     {
 876         GLbitfield mapflags;
 877         mapflags = GL_MAP_WRITE_BIT | GL_MAP_FLUSH_EXPLICIT_BIT;
 878         if (flags & WINED3D_BUFFER_DISCARD)
 879         {
 880             mapflags |= GL_MAP_UNSYNCHRONIZED_BIT | GL_MAP_INVALIDATE_BUFFER_BIT;
 881         }
 882         else if (flags & WINED3D_BUFFER_NOSYNC)
 883         {
 884             mapflags |= GL_MAP_UNSYNCHRONIZED_BIT;
 885         }
 886         map = GL_EXTCALL(glMapBufferRange(This->buffer_type_hint, 0,
 887                     This->resource.size, mapflags));
 888         checkGLcall("glMapBufferRange");
 889     }
 890     else
 891     {
 892         if (This->flags & WINED3D_BUFFER_APPLESYNC)
 893         {
 894             DWORD syncflags = 0;
 895             if (flags & WINED3D_BUFFER_DISCARD) syncflags |= WINED3DLOCK_DISCARD;
 896             if (flags & WINED3D_BUFFER_NOSYNC) syncflags |= WINED3DLOCK_NOOVERWRITE;
 897             LEAVE_GL();
 898             buffer_sync_apple(This, syncflags, gl_info);
 899             ENTER_GL();
 900         }
 901         map = GL_EXTCALL(glMapBufferARB(This->buffer_type_hint, GL_WRITE_ONLY_ARB));
 902         checkGLcall("glMapBufferARB");
 903     }
 904     if (!map)
 905     {
 906         LEAVE_GL();
 907         ERR("Failed to map opengl buffer\n");
 908         return;
 909     }
 910
 911     while (This->modified_areas)
 912     {
 913         This->modified_areas--;
 914         start = This->maps[This->modified_areas].offset;
 915         len = This->maps[This->modified_areas].size;
 916
 917         memcpy(map + start, This->resource.allocatedMemory + start, len);
 918
 919         if (gl_info->supported[ARB_MAP_BUFFER_RANGE])
 920         {
 921             GL_EXTCALL(glFlushMappedBufferRange(This->buffer_type_hint, start, len));
 922             checkGLcall("glFlushMappedBufferRange");
 923         }
 924         else if (This->flags & WINED3D_BUFFER_FLUSH)
 925         {
 926             GL_EXTCALL(glFlushMappedBufferRangeAPPLE(This->buffer_type_hint, start, len));
 927             checkGLcall("glFlushMappedBufferRangeAPPLE");
 928         }
 929     }
 930     GL_EXTCALL(glUnmapBufferARB(This->buffer_type_hint));
 931     checkGLcall("glUnmapBufferARB");
 932
 933     LEAVE_GL();
 934 }
 935
 936 /* Do not call while under the GL lock. */
 937 static void STDMETHODCALLTYPE buffer_PreLoad(IWineD3DBuffer *iface)
 938 {
 939     struct wined3d_buffer *This = (struct wined3d_buffer *)iface;
 940     IWineD3DDeviceImpl *device = This->resource.device;
 941     UINT start = 0, end = 0, len = 0, vertices;
 942     const struct wined3d_gl_info *gl_info;
 943     struct wined3d_context *context;
 944     BOOL decl_changed = FALSE;
 945     unsigned int i, j;
 946     BYTE *data;
 947     DWORD flags = This->flags & (WINED3D_BUFFER_NOSYNC | WINED3D_BUFFER_DISCARD);
 948
 949     TRACE("iface %p\n", iface);
 950     This->flags &= ~(WINED3D_BUFFER_NOSYNC | WINED3D_BUFFER_DISCARD);
 951
 952     context = context_acquire(device, NULL);
 953     gl_info = context->gl_info;
 954
 955     if (!This->buffer_object)
 956     {
 957         /* TODO: Make converting independent from VBOs */
 958         if (This->flags & WINED3D_BUFFER_CREATEBO)
 959         {
 960             buffer_create_buffer_object(This, gl_info);
 961             This->flags &= ~WINED3D_BUFFER_CREATEBO;
 962         }
 963         else
 964         {
 965             /* Not doing any conversion */
 966             goto end;
 967         }
 968     }
 969
 970     /* Reading the declaration makes only sense if the stateblock is finalized and the buffer bound to a stream */
 971     if (device->isInDraw && This->bind_count > 0)
 972     {
 973         decl_changed = buffer_find_decl(This);
 974         This->flags |= WINED3D_BUFFER_HASDESC;
 975     }
 976
 977     if (!decl_changed && !(This->flags & WINED3D_BUFFER_HASDESC && buffer_is_dirty(This)))
 978     {
 979         context_release(context);
 980         ++This->draw_count;
 981         if (This->draw_count > VB_RESETDECLCHANGE) This->decl_change_count = 0;
 982         if (This->draw_count > VB_RESETFULLCONVS) This->full_conversion_count = 0;
 983         return;
 984     }
 985
 986     /* If applications change the declaration over and over, reconverting all the time is a huge
 987      * performance hit. So count the declaration changes and release the VBO if there are too many
 988      * of them (and thus stop converting)
 989      */
 990     if (decl_changed)
 991     {
 992         ++This->decl_change_count;
 993         This->draw_count = 0;
 994
 995         if (This->decl_change_count > VB_MAXDECLCHANGES ||
 996             (This->conversion_map && (This->resource.usage & WINED3DUSAGE_DYNAMIC)))
 997         {
 998             FIXME("Too many declaration changes or converting dynamic buffer, stopping converting\n");
 999
1000             IWineD3DBuffer_UnLoad(iface);
1001             This->flags &= ~WINED3D_BUFFER_CREATEBO;
1002
1003             /* The stream source state handler might have read the memory of the vertex buffer already
1004              * and got the memory in the vbo which is not valid any longer. Dirtify the stream source
1005              * to force a reload. This happens only once per changed vertexbuffer and should occur rather
1006              * rarely
1007              */
1008             IWineD3DDeviceImpl_MarkStateDirty(device, STATE_STREAMSRC);
1009             goto end;
1010         }
1011         buffer_check_buffer_object_size(This, gl_info);
1012
1013         /* The declaration changed, reload the whole buffer */
1014         WARN("Reloading buffer because of decl change\n");
1015         buffer_clear_dirty_areas(This);
1016         if(!buffer_add_dirty_area(This, 0, 0))
1017         {
1018             ERR("buffer_add_dirty_area failed, this is not expected\n");
1019             goto end;
1020         }
1021         /* Avoid unfenced updates, we might overwrite more areas of the buffer than the application
1022          * cleared for unsynchronized updates
1023          */
1024         flags = 0;
1025     }
1026     else
1027     {
1028         /* However, it is perfectly fine to change the declaration every now and then. We don't want a game that
1029          * changes it every minute drop the VBO after VB_MAX_DECL_CHANGES minutes. So count draws without
1030          * decl changes and reset the decl change count after a specific number of them
1031          */
1032         if(buffer_is_fully_dirty(This))
1033         {
1034             ++This->full_conversion_count;
1035             if(This->full_conversion_count > VB_MAXFULLCONVERSIONS)
1036             {
1037                 FIXME("Too many full buffer conversions, stopping converting\n");
1038                 IWineD3DBuffer_UnLoad(iface);
1039                 This->flags &= ~WINED3D_BUFFER_CREATEBO;
1040                 IWineD3DDeviceImpl_MarkStateDirty(device, STATE_STREAMSRC);
1041                 goto end;
1042             }
1043         }
1044         else
1045         {
1046             ++This->draw_count;
1047             if (This->draw_count > VB_RESETDECLCHANGE) This->decl_change_count = 0;
1048             if (This->draw_count > VB_RESETFULLCONVS) This->full_conversion_count = 0;
1049         }
1050     }
1051
1052     if(This->buffer_type_hint == GL_ELEMENT_ARRAY_BUFFER_ARB)
1053     {
1054         IWineD3DDeviceImpl_MarkStateDirty(This->resource.device, STATE_INDEXBUFFER);
1055     }
1056
1057     if (!This->conversion_map)
1058     {
1059         /* That means that there is nothing to fixup. Just upload from This->resource.allocatedMemory
1060          * directly into the vbo. Do not free the system memory copy because drawPrimitive may need it if
1061          * the stride is 0, for instancing emulation, vertex blending emulation or shader emulation.
1062          */
1063         TRACE("No conversion needed\n");
1064
1065         /* Nothing to do because we locked directly into the vbo */
1066         if (!(This->flags & WINED3D_BUFFER_DOUBLEBUFFER))
1067         {
1068             context_release(context);
1069             return;
1070         }
1071
1072         buffer_direct_upload(This, context->gl_info, flags);
1073
1074         context_release(context);
1075         return;
1076     }
1077
1078     if(!(This->flags & WINED3D_BUFFER_DOUBLEBUFFER))
1079     {
1080         buffer_get_sysmem(This, gl_info);
1081     }
1082
1083     /* Now for each vertex in the buffer that needs conversion */
1084     vertices = This->resource.size / This->stride;
1085
1086     if (This->conversion_shift)
1087     {
1088         TRACE("Shifted conversion\n");
1089         data = HeapAlloc(GetProcessHeap(), 0, vertices * This->conversion_stride);
1090
1091         start = 0;
1092         len = This->resource.size;
1093         end = start + len;
1094
1095         if (This->maps[0].offset || This->maps[0].size != This->resource.size)
1096         {
1097             FIXME("Implement partial buffer load with shifted conversion\n");
1098         }
1099
1100         for (i = start / This->stride; i < min((end / This->stride) + 1, vertices); ++i)
1101         {
1102             for (j = 0; j < This->stride; ++j)
1103             {
1104                 switch(This->conversion_map[j])
1105                 {
1106                     case CONV_NONE:
1107                         data[This->conversion_stride * i + j + This->conversion_shift[j]]
1108                                 = This->resource.allocatedMemory[This->stride * i + j];
1109                         break;
1110
1111                     case CONV_FLOAT16_2:
1112                     {
1113                         float *out = (float *)(&data[This->conversion_stride * i + j + This->conversion_shift[j]]);
1114                         const WORD *in = (WORD *)(&This->resource.allocatedMemory[i * This->stride + j]);
1115
1116                         out[1] = float_16_to_32(in + 1);
1117                         out[0] = float_16_to_32(in + 0);
1118                         j += 3;    /* Skip 3 additional bytes,as a FLOAT16_2 has 4 bytes */
1119                         break;
1120                     }
1121
1122                     default:
1123                         FIXME("Unimplemented conversion %d in shifted conversion\n", This->conversion_map[j]);
1124                         break;
1125                 }
1126             }
1127         }
1128
1129         ENTER_GL();
1130         GL_EXTCALL(glBindBufferARB(This->buffer_type_hint, This->buffer_object));
1131         checkGLcall("glBindBufferARB");
1132         GL_EXTCALL(glBufferSubDataARB(This->buffer_type_hint, 0, vertices * This->conversion_stride, data));
1133         checkGLcall("glBufferSubDataARB");
1134         LEAVE_GL();
1135     }
1136     else
1137     {
1138         data = HeapAlloc(GetProcessHeap(), 0, This->resource.size);
1139
1140         while(This->modified_areas)
1141         {
1142             This->modified_areas--;
1143             start = This->maps[This->modified_areas].offset;
1144             len = This->maps[This->modified_areas].size;
1145             end = start + len;
1146
1147             memcpy(data + start, This->resource.allocatedMemory + start, end - start);
1148             for (i = start / This->stride; i < min((end / This->stride) + 1, vertices); ++i)
1149             {
1150                 for (j = 0; j < This->stride; ++j)
1151                 {
1152                     switch(This->conversion_map[j])
1153                     {
1154                         case CONV_NONE:
1155                             /* Done already */
1156                             j += 3;
1157                             break;
1158                         case CONV_D3DCOLOR:
1159                             fixup_d3dcolor((DWORD *) (data + i * This->stride + j));
1160                             j += 3;
1161                             break;
1162
1163                         case CONV_POSITIONT:
1164                             fixup_transformed_pos((float *) (data + i * This->stride + j));
1165                             j += 15;
1166                             break;
1167
1168                         case CONV_FLOAT16_2:
1169                             ERR("Did not expect FLOAT16 conversion in unshifted conversion\n");
1170                         default:
1171                             FIXME("Unimplemented conversion %d in shifted conversion\n", This->conversion_map[j]);
1172                     }
1173                 }
1174             }
1175
1176             ENTER_GL();
1177             GL_EXTCALL(glBindBufferARB(This->buffer_type_hint, This->buffer_object));
1178             checkGLcall("glBindBufferARB");
1179             GL_EXTCALL(glBufferSubDataARB(This->buffer_type_hint, start, len, data + start));
1180             checkGLcall("glBufferSubDataARB");
1181             LEAVE_GL();
1182         }
1183     }
1184
1185     HeapFree(GetProcessHeap(), 0, data);
1186
1187 end:
1188     context_release(context);
1189 }
1190
1191 static WINED3DRESOURCETYPE STDMETHODCALLTYPE buffer_GetType(IWineD3DBuffer *iface)
1192 {
1193     return resource_get_type((IWineD3DResource *)iface);
1194 }
1195
1196 /* IWineD3DBuffer methods */
1197
1198 static DWORD buffer_sanitize_flags(struct wined3d_buffer *buffer, DWORD flags)
1199 {
1200     /* Not all flags make sense together, but Windows never returns an error. Catch the
1201      * cases that could cause issues */
1202     if(flags & WINED3DLOCK_READONLY)
1203     {
1204         if(flags & WINED3DLOCK_DISCARD)
1205         {
1206             WARN("WINED3DLOCK_READONLY combined with WINED3DLOCK_DISCARD, ignoring flags\n");
1207             return 0;
1208         }
1209         if(flags & WINED3DLOCK_NOOVERWRITE)
1210         {
1211             WARN("WINED3DLOCK_READONLY combined with WINED3DLOCK_NOOVERWRITE, ignoring flags\n");
1212             return 0;
1213         }
1214     }
1215     else if((flags & (WINED3DLOCK_DISCARD | WINED3DLOCK_NOOVERWRITE)) == (WINED3DLOCK_DISCARD | WINED3DLOCK_NOOVERWRITE))
1216     {
1217         WARN("WINED3DLOCK_DISCARD and WINED3DLOCK_NOOVERWRITE used together, ignoring\n");
1218         return 0;
1219     }
1220     else if (flags & (WINED3DLOCK_DISCARD | WINED3DLOCK_NOOVERWRITE) && !(buffer->resource.usage & WINED3DUSAGE_DYNAMIC))
1221     {
1222         WARN("DISCARD or NOOVERWRITE lock on non-dynamic buffer, ignoring\n");
1223         return 0;
1224     }
1225
1226     return flags;
1227 }
1228
1229 static GLbitfield buffer_gl_map_flags(DWORD d3d_flags)
1230 {
1231     GLbitfield ret = 0;
1232
1233     if (!(d3d_flags & WINED3DLOCK_READONLY)) ret = GL_MAP_WRITE_BIT | GL_MAP_FLUSH_EXPLICIT_BIT;
1234
1235     if (d3d_flags & (WINED3DLOCK_DISCARD | WINED3DLOCK_NOOVERWRITE))
1236     {
1237         if(d3d_flags & WINED3DLOCK_DISCARD) ret |= GL_MAP_INVALIDATE_BUFFER_BIT;
1238         ret |= GL_MAP_UNSYNCHRONIZED_BIT;
1239     }
1240     else
1241     {
1242         ret |= GL_MAP_READ_BIT;
1243     }
1244
1245     return ret;
1246 }
1247
1248 static HRESULT STDMETHODCALLTYPE buffer_Map(IWineD3DBuffer *iface, UINT offset, UINT size, BYTE **data, DWORD flags)
1249 {
1250     struct wined3d_buffer *This = (struct wined3d_buffer *)iface;
1251     LONG count;
1252     BOOL dirty = buffer_is_dirty(This);
1253
1254     TRACE("iface %p, offset %u, size %u, data %p, flags %#x\n", iface, offset, size, data, flags);
1255
1256     flags = buffer_sanitize_flags(This, flags);
1257     if (!(flags & WINED3DLOCK_READONLY))
1258     {
1259         if (!buffer_add_dirty_area(This, offset, size)) return E_OUTOFMEMORY;
1260     }
1261
1262     count = InterlockedIncrement(&This->lock_count);
1263
1264     if (This->buffer_object)
1265     {
1266         if(!(This->flags & WINED3D_BUFFER_DOUBLEBUFFER))
1267         {
1268             if(count == 1)
1269             {
1270                 IWineD3DDeviceImpl *device = This->resource.device;
1271                 struct wined3d_context *context;
1272                 const struct wined3d_gl_info *gl_info;
1273
1274                 if(This->buffer_type_hint == GL_ELEMENT_ARRAY_BUFFER_ARB)
1275                 {
1276                     IWineD3DDeviceImpl_MarkStateDirty(This->resource.device, STATE_INDEXBUFFER);
1277                 }
1278
1279                 context = context_acquire(device, NULL);
1280                 gl_info = context->gl_info;
1281                 ENTER_GL();
1282                 GL_EXTCALL(glBindBufferARB(This->buffer_type_hint, This->buffer_object));
1283
1284                 if (gl_info->supported[ARB_MAP_BUFFER_RANGE])
1285                 {
1286                     GLbitfield mapflags = buffer_gl_map_flags(flags);
1287                     This->resource.allocatedMemory = GL_EXTCALL(glMapBufferRange(This->buffer_type_hint, 0,
1288                                                                                 This->resource.size, mapflags));
1289                     checkGLcall("glMapBufferRange");
1290                 }
1291                 else
1292                 {
1293                     if(This->flags & WINED3D_BUFFER_APPLESYNC)
1294                     {
1295                         LEAVE_GL();
1296                         buffer_sync_apple(This, flags, gl_info);
1297                         ENTER_GL();
1298                     }
1299                     This->resource.allocatedMemory = GL_EXTCALL(glMapBufferARB(This->buffer_type_hint, GL_READ_WRITE_ARB));
1300                     checkGLcall("glMapBufferARB");
1301                 }
1302                 LEAVE_GL();
1303
1304                 if (((DWORD_PTR) This->resource.allocatedMemory) & (RESOURCE_ALIGNMENT - 1))
1305                 {
1306                     WARN("Pointer %p is not %u byte aligned, falling back to double buffered operation\n",
1307                         This->resource.allocatedMemory, RESOURCE_ALIGNMENT);
1308
1309                     ENTER_GL();
1310                     GL_EXTCALL(glUnmapBufferARB(This->buffer_type_hint));
1311                     checkGLcall("glUnmapBufferARB");
1312                     LEAVE_GL();
1313                     This->resource.allocatedMemory = NULL;
1314
1315                     buffer_get_sysmem(This, gl_info);
1316                     TRACE("New pointer is %p\n", This->resource.allocatedMemory);
1317                 }
1318                 context_release(context);
1319             }
1320         }
1321         else
1322         {
1323             if (dirty)
1324             {
1325                 if (This->flags & WINED3D_BUFFER_NOSYNC && !(flags & WINED3DLOCK_NOOVERWRITE))
1326                 {
1327                     This->flags &= ~WINED3D_BUFFER_NOSYNC;
1328                 }
1329             }
1330             else if(flags & WINED3DLOCK_NOOVERWRITE)
1331             {
1332                 This->flags |= WINED3D_BUFFER_NOSYNC;
1333             }
1334
1335             if (flags & WINED3DLOCK_DISCARD)
1336             {
1337                 This->flags |= WINED3D_BUFFER_DISCARD;
1338             }
1339         }
1340     }
1341
1342     *data = This->resource.allocatedMemory + offset;
1343
1344     TRACE("Returning memory at %p (base %p, offset %u)\n", *data, This->resource.allocatedMemory, offset);
1345     /* TODO: check Flags compatibility with This->currentDesc.Usage (see MSDN) */
1346
1347     return WINED3D_OK;
1348 }
1349
1350 static void STDMETHODCALLTYPE buffer_Unmap(IWineD3DBuffer *iface)
1351 {
1352     struct wined3d_buffer *This = (struct wined3d_buffer *)iface;
1353     ULONG i;
1354
1355     TRACE("(%p)\n", This);
1356
1357     /* In the case that the number of Unmap calls > the
1358      * number of Map calls, d3d returns always D3D_OK.
1359      * This is also needed to prevent Map from returning garbage on
1360      * the next call (this will happen if the lock_count is < 0). */
1361     if (!This->lock_count)
1362     {
1363         WARN("Unmap called without a previous Map call.\n");
1364         return;
1365     }
1366
1367     if (InterlockedDecrement(&This->lock_count))
1368     {
1369         /* Delay loading the buffer until everything is unlocked */
1370         TRACE("Ignoring unlock\n");
1371         return;
1372     }
1373
1374     if(!(This->flags & WINED3D_BUFFER_DOUBLEBUFFER) && This->buffer_object)
1375     {
1376         IWineD3DDeviceImpl *device = This->resource.device;
1377         const struct wined3d_gl_info *gl_info;
1378         struct wined3d_context *context;
1379
1380         if(This->buffer_type_hint == GL_ELEMENT_ARRAY_BUFFER_ARB)
1381         {
1382             IWineD3DDeviceImpl_MarkStateDirty(This->resource.device, STATE_INDEXBUFFER);
1383         }
1384
1385         context = context_acquire(device, NULL);
1386         gl_info = context->gl_info;
1387         ENTER_GL();
1388         GL_EXTCALL(glBindBufferARB(This->buffer_type_hint, This->buffer_object));
1389
1390         if (gl_info->supported[ARB_MAP_BUFFER_RANGE])
1391         {
1392             for(i = 0; i < This->modified_areas; i++)
1393             {
1394                 GL_EXTCALL(glFlushMappedBufferRange(This->buffer_type_hint,
1395                                                     This->maps[i].offset,
1396                                                     This->maps[i].size));
1397                 checkGLcall("glFlushMappedBufferRange");
1398             }
1399         }
1400         else if (This->flags & WINED3D_BUFFER_FLUSH)
1401         {
1402             for(i = 0; i < This->modified_areas; i++)
1403             {
1404                 GL_EXTCALL(glFlushMappedBufferRangeAPPLE(This->buffer_type_hint,
1405                                                          This->maps[i].offset,
1406                                                          This->maps[i].size));
1407                 checkGLcall("glFlushMappedBufferRangeAPPLE");
1408             }
1409         }
1410
1411         GL_EXTCALL(glUnmapBufferARB(This->buffer_type_hint));
1412         LEAVE_GL();
1413         context_release(context);
1414
1415         This->resource.allocatedMemory = NULL;
1416         buffer_clear_dirty_areas(This);
1417     }
1418     else if (This->flags & WINED3D_BUFFER_HASDESC)
1419     {
1420         buffer_PreLoad(iface);
1421     }
1422 }
1423
1424 static void STDMETHODCALLTYPE buffer_GetDesc(IWineD3DBuffer *iface, WINED3DBUFFER_DESC *desc)
1425 {
1426     struct wined3d_buffer *This = (struct wined3d_buffer *)iface;
1427
1428     TRACE("(%p)\n", This);
1429
1430     desc->Type = This->resource.resourceType;
1431     desc->Usage = This->resource.usage;
1432     desc->Pool = This->resource.pool;
1433     desc->Size = This->resource.size;
1434 }
1435
1436 static const struct IWineD3DBufferVtbl wined3d_buffer_vtbl =
1437 {
1438     /* IUnknown methods */
1439     buffer_QueryInterface,
1440     buffer_AddRef,
1441     buffer_Release,
1442     /* IWineD3DBase methods */
1443     buffer_GetParent,
1444     /* IWineD3DResource methods */
1445     buffer_SetPrivateData,
1446     buffer_GetPrivateData,
1447     buffer_FreePrivateData,
1448     buffer_SetPriority,
1449     buffer_GetPriority,
1450     buffer_PreLoad,
1451     buffer_UnLoad,
1452     buffer_GetType,
1453     /* IWineD3DBuffer methods */
1454     buffer_Map,
1455     buffer_Unmap,
1456     buffer_GetDesc,
1457 };
1458
1459 HRESULT buffer_init(struct wined3d_buffer *buffer, IWineD3DDeviceImpl *device,
1460         UINT size, DWORD usage, enum wined3d_format_id format_id, WINED3DPOOL pool, GLenum bind_hint,
1461         const char *data, void *parent, const struct wined3d_parent_ops *parent_ops)
1462 {
1463     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
1464     const struct wined3d_format *format = wined3d_get_format(gl_info, format_id);
1465     HRESULT hr;
1466     BOOL dynamic_buffer_ok;
1467
1468     if (!size)
1469     {
1470         WARN("Size 0 requested, returning WINED3DERR_INVALIDCALL\n");
1471         return WINED3DERR_INVALIDCALL;
1472     }
1473
1474     buffer->vtbl = &wined3d_buffer_vtbl;
1475
1476     hr = resource_init((IWineD3DResource *)buffer, WINED3DRTYPE_BUFFER,
1477             device, size, usage, format, pool, parent, parent_ops);
1478     if (FAILED(hr))
1479     {
1480         WARN("Failed to initialize resource, hr %#x\n", hr);
1481         return hr;
1482     }
1483     buffer->buffer_type_hint = bind_hint;
1484
1485     TRACE("size %#x, usage %#x, format %s, memory @ %p, iface @ %p.\n", buffer->resource.size, buffer->resource.usage,
1486             debug_d3dformat(buffer->resource.format->id), buffer->resource.allocatedMemory, buffer);
1487
1488     /* GL_ARB_map_buffer_range is disabled for now due to numerous bugs and no gains */
1489     dynamic_buffer_ok = gl_info->supported[APPLE_FLUSH_BUFFER_RANGE];
1490
1491     /* Observations show that drawStridedSlow is faster on dynamic VBs than converting +
1492      * drawStridedFast (half-life 2 and others).
1493      *
1494      * Basically converting the vertices in the buffer is quite expensive, and observations
1495      * show that drawStridedSlow is faster than converting + uploading + drawStridedFast.
1496      * Therefore do not create a VBO for WINED3DUSAGE_DYNAMIC buffers.
1497      */
1498     if (!gl_info->supported[ARB_VERTEX_BUFFER_OBJECT])
1499     {
1500         TRACE("Not creating a vbo because GL_ARB_vertex_buffer is not supported\n");
1501     }
1502     else if(buffer->resource.pool == WINED3DPOOL_SYSTEMMEM)
1503     {
1504         TRACE("Not creating a vbo because the vertex buffer is in system memory\n");
1505     }
1506     else if(!dynamic_buffer_ok && (buffer->resource.usage & WINED3DUSAGE_DYNAMIC))
1507     {
1508         TRACE("Not creating a vbo because the buffer has dynamic usage and no GL support\n");
1509     }
1510     else
1511     {
1512         buffer->flags |= WINED3D_BUFFER_CREATEBO;
1513     }
1514
1515     if (data)
1516     {
1517         BYTE *ptr;
1518
1519         hr = IWineD3DBuffer_Map((IWineD3DBuffer *)buffer, 0, size, &ptr, 0);
1520         if (FAILED(hr))
1521         {
1522             ERR("Failed to map buffer, hr %#x\n", hr);
1523             buffer_UnLoad((IWineD3DBuffer *)buffer);
1524             resource_cleanup((IWineD3DResource *)buffer);
1525             return hr;
1526         }
1527
1528         memcpy(ptr, data, size);
1529
1530         IWineD3DBuffer_Unmap((IWineD3DBuffer *)buffer);
1531     }
1532
1533     buffer->maps = HeapAlloc(GetProcessHeap(), 0, sizeof(*buffer->maps));
1534     if (!buffer->maps)
1535     {
1536         ERR("Out of memory\n");
1537         buffer_UnLoad((IWineD3DBuffer *)buffer);
1538         resource_cleanup((IWineD3DResource *)buffer);
1539         return E_OUTOFMEMORY;
1540     }
1541     buffer->maps_size = 1;
1542
1543     return WINED3D_OK;
1544 }