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 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 GLINFO_LOCATION This->resource.device->adapter->gl_info
  33
  34 #define VB_MAXDECLCHANGES     100     /* After that number of decl changes we stop converting */
  35 #define VB_RESETDECLCHANGE    1000    /* Reset the decl changecount after that number of draws */
  36 #define VB_MAXFULLCONVERSIONS 5       /* Number of full conversions before we stop converting */
  37 #define VB_RESETFULLCONVS     20      /* Reset full conversion counts after that number of draws */
  38
  39 static inline BOOL buffer_add_dirty_area(struct wined3d_buffer *This, UINT offset, UINT size)
  40 {
  41     if (!This->buffer_object) return TRUE;
  42
  43     if (This->maps_size <= This->modified_areas)
  44     {
  45         void *new = HeapReAlloc(GetProcessHeap(), 0, This->maps,
  46                                 This->maps_size * 2 * sizeof(*This->maps));
  47         if (!new)
  48         {
  49             ERR("Out of memory\n");
  50             return FALSE;
  51         }
  52         else
  53         {
  54             This->maps = new;
  55             This->maps_size *= 2;
  56         }
  57     }
  58
  59     if(offset > This->resource.size || offset + size > This->resource.size)
  60     {
  61         WARN("Invalid range dirtified, marking entire buffer dirty\n");
  62         offset = 0;
  63         size = This->resource.size;
  64     }
  65     else if(!offset && !size)
  66     {
  67         size = This->resource.size;
  68     }
  69
  70     This->maps[This->modified_areas].offset = offset;
  71     This->maps[This->modified_areas].size = size;
  72     This->modified_areas++;
  73     return TRUE;
  74 }
  75
  76 static inline void buffer_clear_dirty_areas(struct wined3d_buffer *This)
  77 {
  78     This->modified_areas = 0;
  79 }
  80
  81 static inline BOOL buffer_is_dirty(struct wined3d_buffer *This)
  82 {
  83     return This->modified_areas != 0;
  84 }
  85
  86 static inline BOOL buffer_is_fully_dirty(struct wined3d_buffer *This)
  87 {
  88     unsigned int i;
  89
  90     for(i = 0; i < This->modified_areas; i++)
  91     {
  92         if(This->maps[i].offset == 0 && This->maps[i].size == This->resource.size)
  93         {
  94             return TRUE;
  95         }
  96     }
  97     return FALSE;
  98 }
  99
 100 /* Context activation is done by the caller */
 101 static void delete_gl_buffer(struct wined3d_buffer *This)
 102 {
 103     if(!This->buffer_object) return;
 104
 105     ENTER_GL();
 106     GL_EXTCALL(glDeleteBuffersARB(1, &This->buffer_object));
 107     checkGLcall("glDeleteBuffersARB");
 108     LEAVE_GL();
 109     This->buffer_object = 0;
 110
 111     if(This->query)
 112     {
 113         wined3d_event_query_destroy(This->query);
 114         This->query = NULL;
 115     }
 116     This->flags &= ~WINED3D_BUFFER_APPLESYNC;
 117 }
 118
 119 /* Context activation is done by the caller. */
 120 static void buffer_create_buffer_object(struct wined3d_buffer *This)
 121 {
 122     GLenum error, gl_usage;
 123     const struct wined3d_gl_info *gl_info = &This->resource.device->adapter->gl_info;
 124
 125     TRACE("Creating an OpenGL vertex buffer object for IWineD3DVertexBuffer %p Usage(%s)\n",
 126             This, debug_d3dusage(This->resource.usage));
 127
 128     ENTER_GL();
 129
 130     /* Make sure that the gl error is cleared. Do not use checkGLcall
 131     * here because checkGLcall just prints a fixme and continues. However,
 132     * if an error during VBO creation occurs we can fall back to non-vbo operation
 133     * with full functionality(but performance loss)
 134     */
 135     while (glGetError() != GL_NO_ERROR);
 136
 137     /* Basically the FVF parameter passed to CreateVertexBuffer is no good
 138      * It is the FVF set with IWineD3DDevice::SetFVF or the Vertex Declaration set with
 139      * IWineD3DDevice::SetVertexDeclaration that decides how the vertices in the buffer
 140      * look like. This means that on each DrawPrimitive call the vertex buffer has to be verified
 141      * to check if the rhw and color values are in the correct format.
 142      */
 143
 144     GL_EXTCALL(glGenBuffersARB(1, &This->buffer_object));
 145     error = glGetError();
 146     if (!This->buffer_object || error != GL_NO_ERROR)
 147     {
 148         ERR("Failed to create a VBO with error %s (%#x)\n", debug_glerror(error), error);
 149         LEAVE_GL();
 150         goto fail;
 151     }
 152
 153     if(This->buffer_type_hint == GL_ELEMENT_ARRAY_BUFFER_ARB)
 154     {
 155         IWineD3DDeviceImpl_MarkStateDirty(This->resource.device, STATE_INDEXBUFFER);
 156     }
 157     GL_EXTCALL(glBindBufferARB(This->buffer_type_hint, This->buffer_object));
 158     error = glGetError();
 159     if (error != GL_NO_ERROR)
 160     {
 161         ERR("Failed to bind the VBO with error %s (%#x)\n", debug_glerror(error), error);
 162         LEAVE_GL();
 163         goto fail;
 164     }
 165
 166     /* Don't use static, because dx apps tend to update the buffer
 167      * quite often even if they specify 0 usage.
 168      */
 169     if(This->resource.usage & WINED3DUSAGE_DYNAMIC)
 170     {
 171         TRACE("Gl usage = GL_STREAM_DRAW_ARB\n");
 172         gl_usage = GL_STREAM_DRAW_ARB;
 173
 174         if(gl_info->supported[APPLE_FLUSH_BUFFER_RANGE])
 175         {
 176             GL_EXTCALL(glBufferParameteriAPPLE(This->buffer_type_hint, GL_BUFFER_FLUSHING_UNMAP_APPLE, GL_FALSE));
 177             checkGLcall("glBufferParameteriAPPLE(This->buffer_type_hint, GL_BUFFER_FLUSHING_UNMAP_APPLE, GL_FALSE)");
 178             This->flags |= WINED3D_BUFFER_FLUSH;
 179
 180             GL_EXTCALL(glBufferParameteriAPPLE(This->buffer_type_hint, GL_BUFFER_SERIALIZED_MODIFY_APPLE, GL_FALSE));
 181             checkGLcall("glBufferParameteriAPPLE(This->buffer_type_hint, GL_BUFFER_SERIALIZED_MODIFY_APPLE, GL_FALSE)");
 182             This->flags |= WINED3D_BUFFER_APPLESYNC;
 183         }
 184         /* No setup is needed here for GL_ARB_map_buffer_range */
 185     }
 186     else
 187     {
 188         TRACE("Gl usage = GL_DYNAMIC_DRAW_ARB\n");
 189         gl_usage = GL_DYNAMIC_DRAW_ARB;
 190     }
 191
 192     /* Reserve memory for the buffer. The amount of data won't change
 193      * so we are safe with calling glBufferData once and
 194      * calling glBufferSubData on updates. Upload the actual data in case
 195      * we're not double buffering, so we can release the heap mem afterwards
 196      */
 197     GL_EXTCALL(glBufferDataARB(This->buffer_type_hint, This->resource.size, This->resource.allocatedMemory, gl_usage));
 198     error = glGetError();
 199     LEAVE_GL();
 200     if (error != GL_NO_ERROR)
 201     {
 202         ERR("glBufferDataARB failed with error %s (%#x)\n", debug_glerror(error), error);
 203         goto fail;
 204     }
 205
 206     This->buffer_object_size = This->resource.size;
 207     This->buffer_object_usage = gl_usage;
 208
 209     if(This->flags & WINED3D_BUFFER_DOUBLEBUFFER)
 210     {
 211         if(!buffer_add_dirty_area(This, 0, 0))
 212         {
 213             ERR("buffer_add_dirty_area failed, this is not expected\n");
 214             goto fail;
 215         }
 216     }
 217     else
 218     {
 219         HeapFree(GetProcessHeap(), 0, This->resource.heapMemory);
 220         This->resource.allocatedMemory = NULL;
 221         This->resource.heapMemory = NULL;
 222     }
 223
 224     return;
 225
 226 fail:
 227     /* Clean up all vbo init, but continue because we can work without a vbo :-) */
 228     ERR("Failed to create a vertex buffer object. Continuing, but performance issues may occur\n");
 229     delete_gl_buffer(This);
 230     buffer_clear_dirty_areas(This);
 231 }
 232
 233 static BOOL buffer_process_converted_attribute(struct wined3d_buffer *This,
 234         const enum wined3d_buffer_conversion_type conversion_type,
 235         const struct wined3d_stream_info_element *attrib, DWORD *stride_this_run)
 236 {
 237     DWORD attrib_size;
 238     BOOL ret = FALSE;
 239     unsigned int i;
 240     DWORD offset = This->resource.device->stateBlock->streamOffset[attrib->stream_idx];
 241     DWORD_PTR data;
 242
 243     /* Check for some valid situations which cause us pain. One is if the buffer is used for
 244      * constant attributes(stride = 0), the other one is if the buffer is used on two streams
 245      * with different strides. In the 2nd case we might have to drop conversion entirely,
 246      * it is possible that the same bytes are once read as FLOAT2 and once as UBYTE4N.
 247      */
 248     if (!attrib->stride)
 249     {
 250         FIXME("%s used with stride 0, let's hope we get the vertex stride from somewhere else\n",
 251                 debug_d3dformat(attrib->format_desc->format));
 252     }
 253     else if(attrib->stride != *stride_this_run && *stride_this_run)
 254     {
 255         FIXME("Got two concurrent strides, %d and %d\n", attrib->stride, *stride_this_run);
 256     }
 257     else
 258     {
 259         *stride_this_run = attrib->stride;
 260         if (This->stride != *stride_this_run)
 261         {
 262             /* We rely that this happens only on the first converted attribute that is found,
 263              * if at all. See above check
 264              */
 265             TRACE("Reconverting because converted attributes occur, and the stride changed\n");
 266             This->stride = *stride_this_run;
 267             HeapFree(GetProcessHeap(), HEAP_ZERO_MEMORY, This->conversion_map);
 268             This->conversion_map = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY,
 269                     sizeof(*This->conversion_map) * This->stride);
 270             ret = TRUE;
 271         }
 272     }
 273
 274     data = (((DWORD_PTR)attrib->data) + offset) % This->stride;
 275     attrib_size = attrib->format_desc->component_count * attrib->format_desc->component_size;
 276     for (i = 0; i < attrib_size; ++i)
 277     {
 278         DWORD_PTR idx = (data + i) % This->stride;
 279         if (This->conversion_map[idx] != conversion_type)
 280         {
 281             TRACE("Byte %ld in vertex changed\n", idx);
 282             TRACE("It was type %d, is %d now\n", This->conversion_map[idx], conversion_type);
 283             ret = TRUE;
 284             This->conversion_map[idx] = conversion_type;
 285         }
 286     }
 287
 288     return ret;
 289 }
 290
 291 static BOOL buffer_check_attribute(struct wined3d_buffer *This, const struct wined3d_stream_info *si,
 292         UINT attrib_idx, const BOOL check_d3dcolor, const BOOL is_ffp_position, const BOOL is_ffp_color,
 293         DWORD *stride_this_run, BOOL *float16_used)
 294 {
 295     const struct wined3d_stream_info_element *attrib = &si->elements[attrib_idx];
 296     IWineD3DDeviceImpl *device = This->resource.device;
 297     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
 298     BOOL ret = FALSE;
 299     WINED3DFORMAT format;
 300
 301     /* Ignore attributes that do not have our vbo. After that check we can be sure that the attribute is
 302      * there, on nonexistent attribs the vbo is 0.
 303      */
 304     if (!(si->use_map & (1 << attrib_idx))
 305             || attrib->buffer_object != This->buffer_object)
 306         return FALSE;
 307
 308     format = attrib->format_desc->format;
 309     /* Look for newly appeared conversion */
 310     if (!gl_info->supported[ARB_HALF_FLOAT_VERTEX]
 311             && (format == WINED3DFMT_R16G16_FLOAT || format == WINED3DFMT_R16G16B16A16_FLOAT))
 312     {
 313         ret = buffer_process_converted_attribute(This, CONV_FLOAT16_2, attrib, stride_this_run);
 314
 315         if (is_ffp_position) FIXME("Test FLOAT16 fixed function processing positions\n");
 316         else if (is_ffp_color) FIXME("test FLOAT16 fixed function processing colors\n");
 317         *float16_used = TRUE;
 318     }
 319     else if (check_d3dcolor && format == WINED3DFMT_B8G8R8A8_UNORM)
 320     {
 321         ret = buffer_process_converted_attribute(This, CONV_D3DCOLOR, attrib, stride_this_run);
 322
 323         if (!is_ffp_color) FIXME("Test for non-color fixed function WINED3DFMT_B8G8R8A8_UNORM format\n");
 324     }
 325     else if (is_ffp_position && format == WINED3DFMT_R32G32B32A32_FLOAT)
 326     {
 327         ret = buffer_process_converted_attribute(This, CONV_POSITIONT, attrib, stride_this_run);
 328     }
 329     else if (This->conversion_map)
 330     {
 331         ret = buffer_process_converted_attribute(This, CONV_NONE, attrib, stride_this_run);
 332     }
 333
 334     return ret;
 335 }
 336
 337 static UINT *find_conversion_shift(struct wined3d_buffer *This,
 338         const struct wined3d_stream_info *strided, UINT stride)
 339 {
 340     UINT *ret, i, j, shift, orig_type_size;
 341
 342     if (!stride)
 343     {
 344         TRACE("No shift\n");
 345         return NULL;
 346     }
 347
 348     This->conversion_stride = stride;
 349     ret = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(DWORD) * stride);
 350     for (i = 0; i < MAX_ATTRIBS; ++i)
 351     {
 352         WINED3DFORMAT format;
 353
 354         if (!(strided->use_map & (1 << i)) || strided->elements[i].buffer_object != This->buffer_object) continue;
 355
 356         format = strided->elements[i].format_desc->format;
 357         if (format == WINED3DFMT_R16G16_FLOAT)
 358         {
 359             shift = 4;
 360         }
 361         else if (format == WINED3DFMT_R16G16B16A16_FLOAT)
 362         {
 363             shift = 8;
 364             /* Pre-shift the last 4 bytes in the FLOAT16_4 by 4 bytes - this makes FLOAT16_2 and FLOAT16_4 conversions
 365              * compatible
 366              */
 367             for (j = 4; j < 8; ++j)
 368             {
 369                 ret[(DWORD_PTR)strided->elements[i].data + j] += 4;
 370             }
 371         }
 372         else
 373         {
 374             shift = 0;
 375         }
 376         This->conversion_stride += shift;
 377
 378         if (shift)
 379         {
 380             orig_type_size = strided->elements[i].format_desc->component_count
 381                     * strided->elements[i].format_desc->component_size;
 382             for (j = (DWORD_PTR)strided->elements[i].data + orig_type_size; j < stride; ++j)
 383             {
 384                 ret[j] += shift;
 385             }
 386         }
 387     }
 388
 389     if (TRACE_ON(d3d))
 390     {
 391         TRACE("Dumping conversion shift:\n");
 392         for (i = 0; i < stride; ++i)
 393         {
 394             TRACE("[%d]", ret[i]);
 395         }
 396         TRACE("\n");
 397     }
 398
 399     return ret;
 400 }
 401
 402 static BOOL buffer_find_decl(struct wined3d_buffer *This)
 403 {
 404     IWineD3DDeviceImpl *device = This->resource.device;
 405     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
 406     const struct wined3d_stream_info *si = &device->strided_streams;
 407     UINT stride_this_run = 0;
 408     BOOL float16_used = FALSE;
 409     BOOL ret = FALSE;
 410     unsigned int i;
 411
 412     /* In d3d7 the vertex buffer declaration NEVER changes because it is stored in the d3d7 vertex buffer.
 413      * Once we have our declaration there is no need to look it up again. Index buffers also never need
 414      * conversion, so once the (empty) conversion structure is created don't bother checking again
 415      */
 416     if (This->flags & WINED3D_BUFFER_HASDESC)
 417     {
 418         if(This->resource.usage & WINED3DUSAGE_STATICDECL) return FALSE;
 419     }
 420
 421     TRACE("Finding vertex buffer conversion information\n");
 422     /* Certain declaration types need some fixups before we can pass them to
 423      * opengl. This means D3DCOLOR attributes with fixed function vertex
 424      * processing, FLOAT4 POSITIONT with fixed function, and FLOAT16 if
 425      * GL_ARB_half_float_vertex is not supported.
 426      *
 427      * Note for d3d8 and d3d9:
 428      * The vertex buffer FVF doesn't help with finding them, we have to use
 429      * the decoded vertex declaration and pick the things that concern the
 430      * current buffer. A problem with this is that this can change between
 431      * draws, so we have to validate the information and reprocess the buffer
 432      * if it changes, and avoid false positives for performance reasons.
 433      * WineD3D doesn't even know the vertex buffer any more, it is managed
 434      * by the client libraries and passed to SetStreamSource and ProcessVertices
 435      * as needed.
 436      *
 437      * We have to distinguish between vertex shaders and fixed function to
 438      * pick the way we access the strided vertex information.
 439      *
 440      * This code sets up a per-byte array with the size of the detected
 441      * stride of the arrays in the buffer. For each byte we have a field
 442      * that marks the conversion needed on this byte. For example, the
 443      * following declaration with fixed function vertex processing:
 444      *
 445      *      POSITIONT, FLOAT4
 446      *      NORMAL, FLOAT3
 447      *      DIFFUSE, FLOAT16_4
 448      *      SPECULAR, D3DCOLOR
 449      *
 450      * Will result in
 451      * {                 POSITIONT                    }{             NORMAL                }{    DIFFUSE          }{SPECULAR }
 452      * [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]
 453      *
 454      * Where in this example map P means 4 component position conversion, 0
 455      * means no conversion, F means FLOAT16_2 conversion and C means D3DCOLOR
 456      * conversion (red / blue swizzle).
 457      *
 458      * If we're doing conversion and the stride changes we have to reconvert
 459      * the whole buffer. Note that we do not mind if the semantic changes,
 460      * we only care for the conversion type. So if the NORMAL is replaced
 461      * with a TEXCOORD, nothing has to be done, or if the DIFFUSE is replaced
 462      * with a D3DCOLOR BLENDWEIGHT we can happily dismiss the change. Some
 463      * conversion types depend on the semantic as well, for example a FLOAT4
 464      * texcoord needs no conversion while a FLOAT4 positiont needs one
 465      */
 466     if (use_vs(device->stateBlock))
 467     {
 468         TRACE("vshader\n");
 469         /* If the current vertex declaration is marked for no half float conversion don't bother to
 470          * analyse the strided streams in depth, just set them up for no conversion. Return decl changed
 471          * if we used conversion before
 472          */
 473         if (!((IWineD3DVertexDeclarationImpl *) device->stateBlock->vertexDecl)->half_float_conv_needed)
 474         {
 475             if (This->conversion_map)
 476             {
 477                 TRACE("Now using shaders without conversion, but conversion used before\n");
 478                 HeapFree(GetProcessHeap(), 0, This->conversion_map);
 479                 HeapFree(GetProcessHeap(), 0, This->conversion_shift);
 480                 This->conversion_map = NULL;
 481                 This->stride = 0;
 482                 This->conversion_shift = NULL;
 483                 This->conversion_stride = 0;
 484                 return TRUE;
 485             }
 486             else
 487             {
 488                 return FALSE;
 489             }
 490         }
 491         for (i = 0; i < MAX_ATTRIBS; ++i)
 492         {
 493             ret = buffer_check_attribute(This, si, i, FALSE, FALSE, FALSE, &stride_this_run, &float16_used) || ret;
 494         }
 495
 496         /* Recalculate the conversion shift map if the declaration has changed,
 497          * and we're using float16 conversion or used it on the last run
 498          */
 499         if (ret && (float16_used || This->conversion_map))
 500         {
 501             HeapFree(GetProcessHeap(), 0, This->conversion_shift);
 502             This->conversion_shift = find_conversion_shift(This, si, This->stride);
 503         }
 504     }
 505     else
 506     {
 507         /* Fixed function is a bit trickier. We have to take care for D3DCOLOR types, FLOAT4 positions and of course
 508          * FLOAT16s if not supported. Also, we can't iterate over the array, so use macros to generate code for all
 509          * the attributes that our current fixed function pipeline implementation cares for.
 510          */
 511         BOOL support_d3dcolor = gl_info->supported[ARB_VERTEX_ARRAY_BGRA];
 512         ret = buffer_check_attribute(This, si, WINED3D_FFP_POSITION,
 513                 TRUE, TRUE,  FALSE, &stride_this_run, &float16_used) || ret;
 514         ret = buffer_check_attribute(This, si, WINED3D_FFP_NORMAL,
 515                 TRUE, FALSE, FALSE, &stride_this_run, &float16_used) || ret;
 516         ret = buffer_check_attribute(This, si, WINED3D_FFP_DIFFUSE,
 517                 !support_d3dcolor, FALSE, TRUE,  &stride_this_run, &float16_used) || ret;
 518         ret = buffer_check_attribute(This, si, WINED3D_FFP_SPECULAR,
 519                 !support_d3dcolor, FALSE, TRUE,  &stride_this_run, &float16_used) || ret;
 520         ret = buffer_check_attribute(This, si, WINED3D_FFP_TEXCOORD0,
 521                 TRUE, FALSE, FALSE, &stride_this_run, &float16_used) || ret;
 522         ret = buffer_check_attribute(This, si, WINED3D_FFP_TEXCOORD1,
 523                 TRUE, FALSE, FALSE, &stride_this_run, &float16_used) || ret;
 524         ret = buffer_check_attribute(This, si, WINED3D_FFP_TEXCOORD2,
 525                 TRUE, FALSE, FALSE, &stride_this_run, &float16_used) || ret;
 526         ret = buffer_check_attribute(This, si, WINED3D_FFP_TEXCOORD3,
 527                 TRUE, FALSE, FALSE, &stride_this_run, &float16_used) || ret;
 528         ret = buffer_check_attribute(This, si, WINED3D_FFP_TEXCOORD4,
 529                 TRUE, FALSE, FALSE, &stride_this_run, &float16_used) || ret;
 530         ret = buffer_check_attribute(This, si, WINED3D_FFP_TEXCOORD5,
 531                 TRUE, FALSE, FALSE, &stride_this_run, &float16_used) || ret;
 532         ret = buffer_check_attribute(This, si, WINED3D_FFP_TEXCOORD6,
 533                 TRUE, FALSE, FALSE, &stride_this_run, &float16_used) || ret;
 534         ret = buffer_check_attribute(This, si, WINED3D_FFP_TEXCOORD7,
 535                 TRUE, FALSE, FALSE, &stride_this_run, &float16_used) || ret;
 536
 537         if (float16_used) FIXME("Float16 conversion used with fixed function vertex processing\n");
 538     }
 539
 540     if (stride_this_run == 0 && This->conversion_map)
 541     {
 542         /* Sanity test */
 543         if (!ret) ERR("no converted attributes found, old conversion map exists, and no declaration change?\n");
 544         HeapFree(GetProcessHeap(), 0, This->conversion_map);
 545         This->conversion_map = NULL;
 546         This->stride = 0;
 547     }
 548
 549     if (ret) TRACE("Conversion information changed\n");
 550
 551     return ret;
 552 }
 553
 554 /* Context activation is done by the caller. */
 555 static void buffer_check_buffer_object_size(struct wined3d_buffer *This)
 556 {
 557     UINT size = This->conversion_stride ?
 558             This->conversion_stride * (This->resource.size / This->stride) : This->resource.size;
 559     if (This->buffer_object_size != size)
 560     {
 561         TRACE("Old size %u, creating new size %u\n", This->buffer_object_size, size);
 562
 563         if(This->buffer_type_hint == GL_ELEMENT_ARRAY_BUFFER_ARB)
 564         {
 565             IWineD3DDeviceImpl_MarkStateDirty(This->resource.device, STATE_INDEXBUFFER);
 566         }
 567
 568         /* Rescue the data before resizing the buffer object if we do not have our backup copy */
 569         if(!(This->flags & WINED3D_BUFFER_DOUBLEBUFFER))
 570         {
 571             buffer_get_sysmem(This);
 572         }
 573
 574         ENTER_GL();
 575         GL_EXTCALL(glBindBufferARB(This->buffer_type_hint, This->buffer_object));
 576         checkGLcall("glBindBufferARB");
 577         GL_EXTCALL(glBufferDataARB(This->buffer_type_hint, size, NULL, This->buffer_object_usage));
 578         This->buffer_object_size = size;
 579         checkGLcall("glBufferDataARB");
 580         LEAVE_GL();
 581     }
 582 }
 583
 584 static inline void fixup_d3dcolor(DWORD *dst_color)
 585 {
 586     DWORD src_color = *dst_color;
 587
 588     /* Color conversion like in drawStridedSlow. watch out for little endianity
 589      * If we want that stuff to work on big endian machines too we have to consider more things
 590      *
 591      * 0xff000000: Alpha mask
 592      * 0x00ff0000: Blue mask
 593      * 0x0000ff00: Green mask
 594      * 0x000000ff: Red mask
 595      */
 596     *dst_color = 0;
 597     *dst_color |= (src_color & 0xff00ff00);         /* Alpha Green */
 598     *dst_color |= (src_color & 0x00ff0000) >> 16;   /* Red */
 599     *dst_color |= (src_color & 0x000000ff) << 16;   /* Blue */
 600 }
 601
 602 static inline void fixup_transformed_pos(float *p)
 603 {
 604     /* rhw conversion like in position_float4(). */
 605     if (p[3] != 1.0f && p[3] != 0.0f)
 606     {
 607         float w = 1.0f / p[3];
 608         p[0] *= w;
 609         p[1] *= w;
 610         p[2] *= w;
 611         p[3] = w;
 612     }
 613 }
 614
 615 /* Context activation is done by the caller. */
 616 const BYTE *buffer_get_memory(IWineD3DBuffer *iface, GLuint *buffer_object)
 617 {
 618     struct wined3d_buffer *This = (struct wined3d_buffer *)iface;
 619
 620     *buffer_object = This->buffer_object;
 621     if (!This->buffer_object)
 622     {
 623         if (This->flags & WINED3D_BUFFER_CREATEBO)
 624         {
 625             buffer_create_buffer_object(This);
 626             This->flags &= ~WINED3D_BUFFER_CREATEBO;
 627             if (This->buffer_object)
 628             {
 629                 *buffer_object = This->buffer_object;
 630                 return NULL;
 631             }
 632         }
 633         return This->resource.allocatedMemory;
 634     }
 635     else
 636     {
 637         return NULL;
 638     }
 639 }
 640
 641 /* IUnknown methods */
 642
 643 static HRESULT STDMETHODCALLTYPE buffer_QueryInterface(IWineD3DBuffer *iface,
 644         REFIID riid, void **object)
 645 {
 646     TRACE("iface %p, riid %s, object %p\n", iface, debugstr_guid(riid), object);
 647
 648     if (IsEqualGUID(riid, &IID_IWineD3DBuffer)
 649             || IsEqualGUID(riid, &IID_IWineD3DResource)
 650             || IsEqualGUID(riid, &IID_IWineD3DBase)
 651             || IsEqualGUID(riid, &IID_IUnknown))
 652     {
 653         IUnknown_AddRef(iface);
 654         *object = iface;
 655         return S_OK;
 656     }
 657
 658     WARN("%s not implemented, returning E_NOINTERFACE\n", debugstr_guid(riid));
 659
 660     *object = NULL;
 661
 662     return E_NOINTERFACE;
 663 }
 664
 665 static ULONG STDMETHODCALLTYPE buffer_AddRef(IWineD3DBuffer *iface)
 666 {
 667     struct wined3d_buffer *This = (struct wined3d_buffer *)iface;
 668     ULONG refcount = InterlockedIncrement(&This->resource.ref);
 669
 670     TRACE("%p increasing refcount to %u\n", This, refcount);
 671
 672     return refcount;
 673 }
 674
 675 /* Context activation is done by the caller. */
 676 BYTE *buffer_get_sysmem(struct wined3d_buffer *This)
 677 {
 678     /* AllocatedMemory exists if the buffer is double buffered or has no buffer object at all */
 679     if(This->resource.allocatedMemory) return This->resource.allocatedMemory;
 680
 681     This->resource.heapMemory = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, This->resource.size + RESOURCE_ALIGNMENT);
 682     This->resource.allocatedMemory = (BYTE *)(((ULONG_PTR)This->resource.heapMemory + (RESOURCE_ALIGNMENT - 1)) & ~(RESOURCE_ALIGNMENT - 1));
 683     ENTER_GL();
 684     GL_EXTCALL(glBindBufferARB(This->buffer_type_hint, This->buffer_object));
 685     GL_EXTCALL(glGetBufferSubDataARB(This->buffer_type_hint, 0, This->resource.size, This->resource.allocatedMemory));
 686     LEAVE_GL();
 687     This->flags |= WINED3D_BUFFER_DOUBLEBUFFER;
 688
 689     return This->resource.allocatedMemory;
 690 }
 691
 692 static void STDMETHODCALLTYPE buffer_UnLoad(IWineD3DBuffer *iface)
 693 {
 694     struct wined3d_buffer *This = (struct wined3d_buffer *)iface;
 695
 696     TRACE("iface %p\n", iface);
 697
 698     if (This->buffer_object)
 699     {
 700         IWineD3DDeviceImpl *device = This->resource.device;
 701         struct wined3d_context *context;
 702
 703         context = context_acquire(device, NULL, CTXUSAGE_RESOURCELOAD);
 704
 705         /* Download the buffer, but don't permanently enable double buffering */
 706         if(!(This->flags & WINED3D_BUFFER_DOUBLEBUFFER))
 707         {
 708             buffer_get_sysmem(This);
 709             This->flags &= ~WINED3D_BUFFER_DOUBLEBUFFER;
 710         }
 711
 712         delete_gl_buffer(This);
 713         This->flags |= WINED3D_BUFFER_CREATEBO; /* Recreate the buffer object next load */
 714         buffer_clear_dirty_areas(This);
 715
 716         context_release(context);
 717
 718         HeapFree(GetProcessHeap(), 0, This->conversion_shift);
 719         This->conversion_shift = NULL;
 720         HeapFree(GetProcessHeap(), 0, This->conversion_map);
 721         This->conversion_map = NULL;
 722         This->stride = 0;
 723         This->conversion_stride = 0;
 724         This->flags &= ~WINED3D_BUFFER_HASDESC;
 725     }
 726 }
 727
 728 static ULONG STDMETHODCALLTYPE buffer_Release(IWineD3DBuffer *iface)
 729 {
 730     struct wined3d_buffer *This = (struct wined3d_buffer *)iface;
 731     ULONG refcount = InterlockedDecrement(&This->resource.ref);
 732
 733     TRACE("%p decreasing refcount to %u\n", This, refcount);
 734
 735     if (!refcount)
 736     {
 737         buffer_UnLoad(iface);
 738         resource_cleanup((IWineD3DResource *)iface);
 739         This->resource.parent_ops->wined3d_object_destroyed(This->resource.parent);
 740         HeapFree(GetProcessHeap(), 0, This->maps);
 741         HeapFree(GetProcessHeap(), 0, This);
 742     }
 743
 744     return refcount;
 745 }
 746
 747 /* IWineD3DBase methods */
 748
 749 static HRESULT STDMETHODCALLTYPE buffer_GetParent(IWineD3DBuffer *iface, IUnknown **parent)
 750 {
 751     return resource_get_parent((IWineD3DResource *)iface, parent);
 752 }
 753
 754 /* IWineD3DResource methods */
 755
 756 static HRESULT STDMETHODCALLTYPE buffer_SetPrivateData(IWineD3DBuffer *iface,
 757         REFGUID guid, const void *data, DWORD data_size, DWORD flags)
 758 {
 759     return resource_set_private_data((IWineD3DResource *)iface, guid, data, data_size, flags);
 760 }
 761
 762 static HRESULT STDMETHODCALLTYPE buffer_GetPrivateData(IWineD3DBuffer *iface,
 763         REFGUID guid, void *data, DWORD *data_size)
 764 {
 765     return resource_get_private_data((IWineD3DResource *)iface, guid, data, data_size);
 766 }
 767
 768 static HRESULT STDMETHODCALLTYPE buffer_FreePrivateData(IWineD3DBuffer *iface, REFGUID guid)
 769 {
 770     return resource_free_private_data((IWineD3DResource *)iface, guid);
 771 }
 772
 773 static DWORD STDMETHODCALLTYPE buffer_SetPriority(IWineD3DBuffer *iface, DWORD priority)
 774 {
 775     return resource_set_priority((IWineD3DResource *)iface, priority);
 776 }
 777
 778 static DWORD STDMETHODCALLTYPE buffer_GetPriority(IWineD3DBuffer *iface)
 779 {
 780     return resource_get_priority((IWineD3DResource *)iface);
 781 }
 782
 783 /* The caller provides a context and binds the buffer */
 784 static void buffer_sync_apple(struct wined3d_buffer *This, DWORD flags, const struct wined3d_gl_info *gl_info)
 785 {
 786     enum wined3d_event_query_result ret;
 787
 788     /* No fencing needs to be done if the app promises not to overwrite
 789      * existing data */
 790     if(flags & WINED3DLOCK_NOOVERWRITE) return;
 791     if(flags & WINED3DLOCK_DISCARD)
 792     {
 793         ENTER_GL();
 794         GL_EXTCALL(glBufferDataARB(This->buffer_type_hint, This->resource.size, NULL, This->buffer_object_usage));
 795         checkGLcall("glBufferDataARB\n");
 796         LEAVE_GL();
 797         return;
 798     }
 799
 800     if(!This->query)
 801     {
 802         TRACE("Creating event query for buffer %p\n", This);
 803
 804         if (!wined3d_event_query_supported(gl_info))
 805         {
 806             FIXME("Event queries not supported, dropping async buffer locks.\n");
 807             goto drop_query;
 808         }
 809
 810         This->query = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*This->query));
 811         if (!This->query)
 812         {
 813             ERR("Failed to allocate event query memory, dropping async buffer locks.\n");
 814             goto drop_query;
 815         }
 816
 817         /* Since we don't know about old draws a glFinish is needed once */
 818         wglFinish();
 819         return;
 820     }
 821     TRACE("Synchronizing buffer %p\n", This);
 822     ret = wined3d_event_query_finish(This->query, This->resource.device);
 823     switch(ret)
 824     {
 825         case WINED3D_EVENT_QUERY_NOT_STARTED:
 826         case WINED3D_EVENT_QUERY_OK:
 827             /* All done */
 828             return;
 829
 830         case WINED3D_EVENT_QUERY_WRONG_THREAD:
 831             WARN("Cannot synchronize buffer lock due to a thread conflict\n");
 832             goto drop_query;
 833
 834         default:
 835             ERR("wined3d_event_query_finish returned %u, dropping async buffer locks\n", ret);
 836             goto drop_query;
 837     }
 838
 839 drop_query:
 840     if(This->query)
 841     {
 842         wined3d_event_query_destroy(This->query);
 843         This->query = NULL;
 844     }
 845
 846     wglFinish();
 847     ENTER_GL();
 848     GL_EXTCALL(glBufferParameteriAPPLE(This->buffer_type_hint, GL_BUFFER_SERIALIZED_MODIFY_APPLE, GL_TRUE));
 849     checkGLcall("glBufferParameteriAPPLE(This->buffer_type_hint, GL_BUFFER_SERIALIZED_MODIFY_APPLE, GL_TRUE)");
 850     LEAVE_GL();
 851     This->flags &= ~WINED3D_BUFFER_APPLESYNC;
 852 }
 853
 854 /* The caller provides a GL context */
 855 static void buffer_direct_upload(struct wined3d_buffer *This, const struct wined3d_gl_info *gl_info, DWORD flags)
 856 {
 857         BYTE *map;
 858         UINT start = 0, len = 0;
 859
 860         ENTER_GL();
 861         GL_EXTCALL(glBindBufferARB(This->buffer_type_hint, This->buffer_object));
 862         checkGLcall("glBindBufferARB");
 863         if (gl_info->supported[ARB_MAP_BUFFER_RANGE])
 864         {
 865             GLbitfield mapflags;
 866             mapflags = GL_MAP_WRITE_BIT | GL_MAP_FLUSH_EXPLICIT_BIT;
 867             if (flags & WINED3D_BUFFER_DISCARD)
 868             {
 869                 mapflags |= GL_MAP_UNSYNCHRONIZED_BIT | GL_MAP_INVALIDATE_BUFFER_BIT;
 870             }
 871             else if (flags & WINED3D_BUFFER_NOSYNC)
 872             {
 873                 mapflags |= GL_MAP_UNSYNCHRONIZED_BIT;
 874             }
 875             map = GL_EXTCALL(glMapBufferRange(This->buffer_type_hint, 0,
 876                                               This->resource.size, mapflags));
 877             checkGLcall("glMapBufferRange");
 878         }
 879         else
 880         {
 881             if (This->flags & WINED3D_BUFFER_APPLESYNC)
 882             {
 883                 DWORD syncflags = 0;
 884                 if (flags & WINED3D_BUFFER_DISCARD) syncflags |= WINED3DLOCK_DISCARD;
 885                 if (flags & WINED3D_BUFFER_NOSYNC) syncflags |= WINED3DLOCK_NOOVERWRITE;
 886                 LEAVE_GL();
 887                 buffer_sync_apple(This, syncflags, gl_info);
 888                 ENTER_GL();
 889             }
 890             map = GL_EXTCALL(glMapBufferARB(This->buffer_type_hint, GL_WRITE_ONLY_ARB));
 891             checkGLcall("glMapBufferARB");
 892         }
 893         if (!map)
 894         {
 895             LEAVE_GL();
 896             ERR("Failed to map opengl buffer\n");
 897             return;
 898         }
 899
 900         while(This->modified_areas)
 901         {
 902             This->modified_areas--;
 903             start = This->maps[This->modified_areas].offset;
 904             len = This->maps[This->modified_areas].size;
 905
 906             memcpy(map + start, This->resource.allocatedMemory + start, len);
 907
 908             if (gl_info->supported[ARB_MAP_BUFFER_RANGE])
 909             {
 910                 GL_EXTCALL(glFlushMappedBufferRange(This->buffer_type_hint, start, len));
 911                 checkGLcall("glFlushMappedBufferRange");
 912             }
 913             else if (This->flags & WINED3D_BUFFER_FLUSH)
 914             {
 915                 GL_EXTCALL(glFlushMappedBufferRangeAPPLE(This->buffer_type_hint, start, len));
 916                 checkGLcall("glFlushMappedBufferRangeAPPLE");
 917             }
 918         }
 919         GL_EXTCALL(glUnmapBufferARB(This->buffer_type_hint));
 920         checkGLcall("glUnmapBufferARB");
 921         LEAVE_GL();
 922 }
 923
 924 static void STDMETHODCALLTYPE buffer_PreLoad(IWineD3DBuffer *iface)
 925 {
 926     struct wined3d_buffer *This = (struct wined3d_buffer *)iface;
 927     IWineD3DDeviceImpl *device = This->resource.device;
 928     UINT start = 0, end = 0, len = 0, vertices;
 929     struct wined3d_context *context;
 930     BOOL decl_changed = FALSE;
 931     unsigned int i, j;
 932     BYTE *data;
 933     DWORD flags = This->flags & (WINED3D_BUFFER_NOSYNC | WINED3D_BUFFER_DISCARD);
 934
 935     TRACE("iface %p\n", iface);
 936     This->flags &= ~(WINED3D_BUFFER_NOSYNC | WINED3D_BUFFER_DISCARD);
 937
 938     context = context_acquire(device, NULL, CTXUSAGE_RESOURCELOAD);
 939
 940     if (!This->buffer_object)
 941     {
 942         /* TODO: Make converting independent from VBOs */
 943         if (This->flags & WINED3D_BUFFER_CREATEBO)
 944         {
 945             buffer_create_buffer_object(This);
 946             This->flags &= ~WINED3D_BUFFER_CREATEBO;
 947         }
 948         else
 949         {
 950             /* Not doing any conversion */
 951             goto end;
 952         }
 953     }
 954
 955     /* Reading the declaration makes only sense if the stateblock is finalized and the buffer bound to a stream */
 956     if (device->isInDraw && This->bind_count > 0)
 957     {
 958         decl_changed = buffer_find_decl(This);
 959         This->flags |= WINED3D_BUFFER_HASDESC;
 960     }
 961
 962     if (!decl_changed && !(This->flags & WINED3D_BUFFER_HASDESC && buffer_is_dirty(This)))
 963     {
 964         context_release(context);
 965         ++This->draw_count;
 966         if (This->draw_count > VB_RESETDECLCHANGE) This->decl_change_count = 0;
 967         if (This->draw_count > VB_RESETFULLCONVS) This->full_conversion_count = 0;
 968         return;
 969     }
 970
 971     /* If applications change the declaration over and over, reconverting all the time is a huge
 972      * performance hit. So count the declaration changes and release the VBO if there are too many
 973      * of them (and thus stop converting)
 974      */
 975     if (decl_changed)
 976     {
 977         ++This->decl_change_count;
 978         This->draw_count = 0;
 979
 980         if (This->decl_change_count > VB_MAXDECLCHANGES ||
 981             (This->conversion_map && (This->resource.usage & WINED3DUSAGE_DYNAMIC)))
 982         {
 983             FIXME("Too many declaration changes or converting dynamic buffer, stopping converting\n");
 984
 985             IWineD3DBuffer_UnLoad(iface);
 986             This->flags &= ~WINED3D_BUFFER_CREATEBO;
 987
 988             /* The stream source state handler might have read the memory of the vertex buffer already
 989              * and got the memory in the vbo which is not valid any longer. Dirtify the stream source
 990              * to force a reload. This happens only once per changed vertexbuffer and should occur rather
 991              * rarely
 992              */
 993             IWineD3DDeviceImpl_MarkStateDirty(device, STATE_STREAMSRC);
 994             goto end;
 995         }
 996         buffer_check_buffer_object_size(This);
 997
 998         /* The declaration changed, reload the whole buffer */
 999         WARN("Reloading buffer because of decl change\n");
1000         buffer_clear_dirty_areas(This);
1001         if(!buffer_add_dirty_area(This, 0, 0))
1002         {
1003             ERR("buffer_add_dirty_area failed, this is not expected\n");
1004             return;
1005         }
1006         /* Avoid unfenced updates, we might overwrite more areas of the buffer than the application
1007          * cleared for unsynchronized updates
1008          */
1009         flags = 0;
1010     }
1011     else
1012     {
1013         /* However, it is perfectly fine to change the declaration every now and then. We don't want a game that
1014          * changes it every minute drop the VBO after VB_MAX_DECL_CHANGES minutes. So count draws without
1015          * decl changes and reset the decl change count after a specific number of them
1016          */
1017         if(buffer_is_fully_dirty(This))
1018         {
1019             ++This->full_conversion_count;
1020             if(This->full_conversion_count > VB_MAXFULLCONVERSIONS)
1021             {
1022                 FIXME("Too many full buffer conversions, stopping converting\n");
1023                 IWineD3DBuffer_UnLoad(iface);
1024                 This->flags &= ~WINED3D_BUFFER_CREATEBO;
1025                 IWineD3DDeviceImpl_MarkStateDirty(device, STATE_STREAMSRC);
1026                 goto end;
1027             }
1028         }
1029         else
1030         {
1031             ++This->draw_count;
1032             if (This->draw_count > VB_RESETDECLCHANGE) This->decl_change_count = 0;
1033             if (This->draw_count > VB_RESETFULLCONVS) This->full_conversion_count = 0;
1034         }
1035     }
1036
1037     if(This->buffer_type_hint == GL_ELEMENT_ARRAY_BUFFER_ARB)
1038     {
1039         IWineD3DDeviceImpl_MarkStateDirty(This->resource.device, STATE_INDEXBUFFER);
1040     }
1041
1042     if (!This->conversion_map)
1043     {
1044         /* That means that there is nothing to fixup. Just upload from This->resource.allocatedMemory
1045          * directly into the vbo. Do not free the system memory copy because drawPrimitive may need it if
1046          * the stride is 0, for instancing emulation, vertex blending emulation or shader emulation.
1047          */
1048         TRACE("No conversion needed\n");
1049
1050         /* Nothing to do because we locked directly into the vbo */
1051         if (!(This->flags & WINED3D_BUFFER_DOUBLEBUFFER))
1052         {
1053             context_release(context);
1054             return;
1055         }
1056
1057         buffer_direct_upload(This, context->gl_info, flags);
1058
1059         context_release(context);
1060         return;
1061     }
1062
1063     if(!(This->flags & WINED3D_BUFFER_DOUBLEBUFFER))
1064     {
1065         buffer_get_sysmem(This);
1066     }
1067
1068     /* Now for each vertex in the buffer that needs conversion */
1069     vertices = This->resource.size / This->stride;
1070
1071     if (This->conversion_shift)
1072     {
1073         TRACE("Shifted conversion\n");
1074         data = HeapAlloc(GetProcessHeap(), 0, vertices * This->conversion_stride);
1075
1076         start = 0;
1077         len = This->resource.size;
1078         end = start + len;
1079
1080         if (This->maps[0].offset || This->maps[0].size != This->resource.size)
1081         {
1082             FIXME("Implement partial buffer load with shifted conversion\n");
1083         }
1084
1085         for (i = start / This->stride; i < min((end / This->stride) + 1, vertices); ++i)
1086         {
1087             for (j = 0; j < This->stride; ++j)
1088             {
1089                 switch(This->conversion_map[j])
1090                 {
1091                     case CONV_NONE:
1092                         data[This->conversion_stride * i + j + This->conversion_shift[j]]
1093                                 = This->resource.allocatedMemory[This->stride * i + j];
1094                         break;
1095
1096                     case CONV_FLOAT16_2:
1097                     {
1098                         float *out = (float *)(&data[This->conversion_stride * i + j + This->conversion_shift[j]]);
1099                         const WORD *in = (WORD *)(&This->resource.allocatedMemory[i * This->stride + j]);
1100
1101                         out[1] = float_16_to_32(in + 1);
1102                         out[0] = float_16_to_32(in + 0);
1103                         j += 3;    /* Skip 3 additional bytes,as a FLOAT16_2 has 4 bytes */
1104                         break;
1105                     }
1106
1107                     default:
1108                         FIXME("Unimplemented conversion %d in shifted conversion\n", This->conversion_map[j]);
1109                         break;
1110                 }
1111             }
1112         }
1113
1114         ENTER_GL();
1115         GL_EXTCALL(glBindBufferARB(This->buffer_type_hint, This->buffer_object));
1116         checkGLcall("glBindBufferARB");
1117         GL_EXTCALL(glBufferSubDataARB(This->buffer_type_hint, 0, vertices * This->conversion_stride, data));
1118         checkGLcall("glBufferSubDataARB");
1119         LEAVE_GL();
1120     }
1121     else
1122     {
1123         data = HeapAlloc(GetProcessHeap(), 0, This->resource.size);
1124
1125         while(This->modified_areas)
1126         {
1127             This->modified_areas--;
1128             start = This->maps[This->modified_areas].offset;
1129             len = This->maps[This->modified_areas].size;
1130             end = start + len;
1131
1132             memcpy(data + start, This->resource.allocatedMemory + start, end - start);
1133             for (i = start / This->stride; i < min((end / This->stride) + 1, vertices); ++i)
1134             {
1135                 for (j = 0; j < This->stride; ++j)
1136                 {
1137                     switch(This->conversion_map[j])
1138                     {
1139                         case CONV_NONE:
1140                             /* Done already */
1141                             j += 3;
1142                             break;
1143                         case CONV_D3DCOLOR:
1144                             fixup_d3dcolor((DWORD *) (data + i * This->stride + j));
1145                             j += 3;
1146                             break;
1147
1148                         case CONV_POSITIONT:
1149                             fixup_transformed_pos((float *) (data + i * This->stride + j));
1150                             j += 15;
1151                             break;
1152
1153                         case CONV_FLOAT16_2:
1154                             ERR("Did not expect FLOAT16 conversion in unshifted conversion\n");
1155                         default:
1156                             FIXME("Unimplemented conversion %d in shifted conversion\n", This->conversion_map[j]);
1157                     }
1158                 }
1159             }
1160
1161             ENTER_GL();
1162             GL_EXTCALL(glBindBufferARB(This->buffer_type_hint, This->buffer_object));
1163             checkGLcall("glBindBufferARB");
1164             GL_EXTCALL(glBufferSubDataARB(This->buffer_type_hint, start, len, data + start));
1165             checkGLcall("glBufferSubDataARB");
1166             LEAVE_GL();
1167         }
1168     }
1169
1170     HeapFree(GetProcessHeap(), 0, data);
1171
1172 end:
1173     context_release(context);
1174 }
1175
1176 static WINED3DRESOURCETYPE STDMETHODCALLTYPE buffer_GetType(IWineD3DBuffer *iface)
1177 {
1178     return resource_get_type((IWineD3DResource *)iface);
1179 }
1180
1181 /* IWineD3DBuffer methods */
1182
1183 static DWORD buffer_sanitize_flags(struct wined3d_buffer *buffer, DWORD flags)
1184 {
1185     /* Not all flags make sense together, but Windows never returns an error. Catch the
1186      * cases that could cause issues */
1187     if(flags & WINED3DLOCK_READONLY)
1188     {
1189         if(flags & WINED3DLOCK_DISCARD)
1190         {
1191             WARN("WINED3DLOCK_READONLY combined with WINED3DLOCK_DISCARD, ignoring flags\n");
1192             return 0;
1193         }
1194         if(flags & WINED3DLOCK_NOOVERWRITE)
1195         {
1196             WARN("WINED3DLOCK_READONLY combined with WINED3DLOCK_NOOVERWRITE, ignoring flags\n");
1197             return 0;
1198         }
1199     }
1200     else if((flags & (WINED3DLOCK_DISCARD | WINED3DLOCK_NOOVERWRITE)) == (WINED3DLOCK_DISCARD | WINED3DLOCK_NOOVERWRITE))
1201     {
1202         WARN("WINED3DLOCK_DISCARD and WINED3DLOCK_NOOVERWRITE used together, ignoring\n");
1203         return 0;
1204     }
1205     else if (flags & (WINED3DLOCK_DISCARD | WINED3DLOCK_NOOVERWRITE) && !(buffer->resource.usage & WINED3DUSAGE_DYNAMIC))
1206     {
1207         WARN("DISCARD or NOOVERWRITE lock on non-dynamic buffer, ignoring\n");
1208         return 0;
1209     }
1210
1211     return flags;
1212 }
1213
1214 static GLbitfield buffer_gl_map_flags(DWORD d3d_flags)
1215 {
1216     GLbitfield ret = 0;
1217
1218     if (!(d3d_flags & WINED3DLOCK_READONLY)) ret = GL_MAP_WRITE_BIT | GL_MAP_FLUSH_EXPLICIT_BIT;
1219
1220     if (d3d_flags & (WINED3DLOCK_DISCARD | WINED3DLOCK_NOOVERWRITE))
1221     {
1222         if(d3d_flags & WINED3DLOCK_DISCARD) ret |= GL_MAP_INVALIDATE_BUFFER_BIT;
1223         ret |= GL_MAP_UNSYNCHRONIZED_BIT;
1224     }
1225     else
1226     {
1227         ret |= GL_MAP_READ_BIT;
1228     }
1229
1230     return ret;
1231 }
1232
1233 static HRESULT STDMETHODCALLTYPE buffer_Map(IWineD3DBuffer *iface, UINT offset, UINT size, BYTE **data, DWORD flags)
1234 {
1235     struct wined3d_buffer *This = (struct wined3d_buffer *)iface;
1236     LONG count;
1237     BOOL dirty = buffer_is_dirty(This);
1238
1239     TRACE("iface %p, offset %u, size %u, data %p, flags %#x\n", iface, offset, size, data, flags);
1240
1241     flags = buffer_sanitize_flags(This, flags);
1242     if (!(flags & WINED3DLOCK_READONLY))
1243     {
1244         if (!buffer_add_dirty_area(This, offset, size)) return E_OUTOFMEMORY;
1245     }
1246
1247     count = InterlockedIncrement(&This->lock_count);
1248
1249     if (This->buffer_object)
1250     {
1251         if(!(This->flags & WINED3D_BUFFER_DOUBLEBUFFER))
1252         {
1253             if(count == 1)
1254             {
1255                 IWineD3DDeviceImpl *device = This->resource.device;
1256                 struct wined3d_context *context;
1257                 const struct wined3d_gl_info *gl_info;
1258
1259                 if(This->buffer_type_hint == GL_ELEMENT_ARRAY_BUFFER_ARB)
1260                 {
1261                     IWineD3DDeviceImpl_MarkStateDirty(This->resource.device, STATE_INDEXBUFFER);
1262                 }
1263
1264                 context = context_acquire(device, NULL, CTXUSAGE_RESOURCELOAD);
1265                 gl_info = context->gl_info;
1266                 ENTER_GL();
1267                 GL_EXTCALL(glBindBufferARB(This->buffer_type_hint, This->buffer_object));
1268
1269                 if (gl_info->supported[ARB_MAP_BUFFER_RANGE])
1270                 {
1271                     GLbitfield mapflags = buffer_gl_map_flags(flags);
1272                     This->resource.allocatedMemory = GL_EXTCALL(glMapBufferRange(This->buffer_type_hint, 0,
1273                                                                                 This->resource.size, mapflags));
1274                     checkGLcall("glMapBufferRange");
1275                 }
1276                 else
1277                 {
1278                     if(This->flags & WINED3D_BUFFER_APPLESYNC)
1279                     {
1280                         LEAVE_GL();
1281                         buffer_sync_apple(This, flags, gl_info);
1282                         ENTER_GL();
1283                     }
1284                     This->resource.allocatedMemory = GL_EXTCALL(glMapBufferARB(This->buffer_type_hint, GL_READ_WRITE_ARB));
1285                     checkGLcall("glMapBufferARB");
1286                 }
1287                 LEAVE_GL();
1288
1289                 if (((DWORD_PTR) This->resource.allocatedMemory) & (RESOURCE_ALIGNMENT - 1))
1290                 {
1291                     WARN("Pointer %p is not %u byte aligned, falling back to double buffered operation\n",
1292                         This->resource.allocatedMemory, RESOURCE_ALIGNMENT);
1293
1294                     ENTER_GL();
1295                     GL_EXTCALL(glUnmapBufferARB(This->buffer_type_hint));
1296                     checkGLcall("glUnmapBufferARB");
1297                     LEAVE_GL();
1298                     This->resource.allocatedMemory = NULL;
1299
1300                     buffer_get_sysmem(This);
1301                     TRACE("New pointer is %p\n", This->resource.allocatedMemory);
1302                 }
1303                 context_release(context);
1304             }
1305         }
1306         else
1307         {
1308             if (dirty)
1309             {
1310                 if (This->flags & WINED3D_BUFFER_NOSYNC && !(flags & WINED3DLOCK_NOOVERWRITE))
1311                 {
1312                     This->flags &= ~WINED3D_BUFFER_NOSYNC;
1313                 }
1314             }
1315             else if(flags & WINED3DLOCK_NOOVERWRITE)
1316             {
1317                 This->flags |= WINED3D_BUFFER_NOSYNC;
1318             }
1319
1320             if (flags & WINED3DLOCK_DISCARD)
1321             {
1322                 This->flags |= WINED3D_BUFFER_DISCARD;
1323             }
1324         }
1325     }
1326
1327     *data = This->resource.allocatedMemory + offset;
1328
1329     TRACE("Returning memory at %p (base %p, offset %u)\n", *data, This->resource.allocatedMemory, offset);
1330     /* TODO: check Flags compatibility with This->currentDesc.Usage (see MSDN) */
1331
1332     return WINED3D_OK;
1333 }
1334
1335 static HRESULT STDMETHODCALLTYPE buffer_Unmap(IWineD3DBuffer *iface)
1336 {
1337     struct wined3d_buffer *This = (struct wined3d_buffer *)iface;
1338     ULONG i;
1339
1340     TRACE("(%p)\n", This);
1341
1342     /* In the case that the number of Unmap calls > the
1343      * number of Map calls, d3d returns always D3D_OK.
1344      * This is also needed to prevent Map from returning garbage on
1345      * the next call (this will happen if the lock_count is < 0). */
1346     if(This->lock_count == 0)
1347     {
1348         TRACE("Unmap called without a previous Map call!\n");
1349         return WINED3D_OK;
1350     }
1351
1352     if (InterlockedDecrement(&This->lock_count))
1353     {
1354         /* Delay loading the buffer until everything is unlocked */
1355         TRACE("Ignoring unlock\n");
1356         return WINED3D_OK;
1357     }
1358
1359     if(!(This->flags & WINED3D_BUFFER_DOUBLEBUFFER) && This->buffer_object)
1360     {
1361         IWineD3DDeviceImpl *device = This->resource.device;
1362         const struct wined3d_gl_info *gl_info;
1363         struct wined3d_context *context;
1364
1365         if(This->buffer_type_hint == GL_ELEMENT_ARRAY_BUFFER_ARB)
1366         {
1367             IWineD3DDeviceImpl_MarkStateDirty(This->resource.device, STATE_INDEXBUFFER);
1368         }
1369
1370         context = context_acquire(device, NULL, CTXUSAGE_RESOURCELOAD);
1371         gl_info = context->gl_info;
1372         ENTER_GL();
1373         GL_EXTCALL(glBindBufferARB(This->buffer_type_hint, This->buffer_object));
1374
1375         if (gl_info->supported[ARB_MAP_BUFFER_RANGE])
1376         {
1377             for(i = 0; i < This->modified_areas; i++)
1378             {
1379                 GL_EXTCALL(glFlushMappedBufferRange(This->buffer_type_hint,
1380                                                     This->maps[i].offset,
1381                                                     This->maps[i].size));
1382                 checkGLcall("glFlushMappedBufferRange");
1383             }
1384         }
1385         else if (This->flags & WINED3D_BUFFER_FLUSH)
1386         {
1387             for(i = 0; i < This->modified_areas; i++)
1388             {
1389                 GL_EXTCALL(glFlushMappedBufferRangeAPPLE(This->buffer_type_hint,
1390                                                          This->maps[i].offset,
1391                                                          This->maps[i].size));
1392                 checkGLcall("glFlushMappedBufferRangeAPPLE");
1393             }
1394         }
1395
1396         GL_EXTCALL(glUnmapBufferARB(This->buffer_type_hint));
1397         LEAVE_GL();
1398         context_release(context);
1399
1400         This->resource.allocatedMemory = NULL;
1401         buffer_clear_dirty_areas(This);
1402     }
1403     else if (This->flags & WINED3D_BUFFER_HASDESC)
1404     {
1405         buffer_PreLoad(iface);
1406     }
1407
1408     return WINED3D_OK;
1409 }
1410
1411 static HRESULT STDMETHODCALLTYPE buffer_GetDesc(IWineD3DBuffer *iface, WINED3DBUFFER_DESC *desc)
1412 {
1413     struct wined3d_buffer *This = (struct wined3d_buffer *)iface;
1414
1415     TRACE("(%p)\n", This);
1416
1417     desc->Type = This->resource.resourceType;
1418     desc->Usage = This->resource.usage;
1419     desc->Pool = This->resource.pool;
1420     desc->Size = This->resource.size;
1421
1422     return WINED3D_OK;
1423 }
1424
1425 static const struct IWineD3DBufferVtbl wined3d_buffer_vtbl =
1426 {
1427     /* IUnknown methods */
1428     buffer_QueryInterface,
1429     buffer_AddRef,
1430     buffer_Release,
1431     /* IWineD3DBase methods */
1432     buffer_GetParent,
1433     /* IWineD3DResource methods */
1434     buffer_SetPrivateData,
1435     buffer_GetPrivateData,
1436     buffer_FreePrivateData,
1437     buffer_SetPriority,
1438     buffer_GetPriority,
1439     buffer_PreLoad,
1440     buffer_UnLoad,
1441     buffer_GetType,
1442     /* IWineD3DBuffer methods */
1443     buffer_Map,
1444     buffer_Unmap,
1445     buffer_GetDesc,
1446 };
1447
1448 HRESULT buffer_init(struct wined3d_buffer *buffer, IWineD3DDeviceImpl *device,
1449         UINT size, DWORD usage, WINED3DFORMAT format, WINED3DPOOL pool, GLenum bind_hint,
1450         const char *data, IUnknown *parent, const struct wined3d_parent_ops *parent_ops)
1451 {
1452     const struct wined3d_format_desc *format_desc = getFormatDescEntry(format, &device->adapter->gl_info);
1453     HRESULT hr;
1454     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
1455     BOOL dynamic_buffer_ok;
1456
1457     if (!size)
1458     {
1459         WARN("Size 0 requested, returning WINED3DERR_INVALIDCALL\n");
1460         return WINED3DERR_INVALIDCALL;
1461     }
1462
1463     buffer->vtbl = &wined3d_buffer_vtbl;
1464
1465     hr = resource_init((IWineD3DResource *)buffer, WINED3DRTYPE_BUFFER,
1466             device, size, usage, format_desc, pool, parent, parent_ops);
1467     if (FAILED(hr))
1468     {
1469         WARN("Failed to initialize resource, hr %#x\n", hr);
1470         return hr;
1471     }
1472     buffer->buffer_type_hint = bind_hint;
1473
1474     TRACE("size %#x, usage %#x, format %s, memory @ %p, iface @ %p.\n", buffer->resource.size, buffer->resource.usage,
1475             debug_d3dformat(buffer->resource.format_desc->format), buffer->resource.allocatedMemory, buffer);
1476
1477     dynamic_buffer_ok = gl_info->supported[APPLE_FLUSH_BUFFER_RANGE] || gl_info->supported[ARB_MAP_BUFFER_RANGE];
1478
1479     /* Observations show that drawStridedSlow is faster on dynamic VBs than converting +
1480      * drawStridedFast (half-life 2 and others).
1481      *
1482      * Basically converting the vertices in the buffer is quite expensive, and observations
1483      * show that drawStridedSlow is faster than converting + uploading + drawStridedFast.
1484      * Therefore do not create a VBO for WINED3DUSAGE_DYNAMIC buffers.
1485      */
1486     if (!gl_info->supported[ARB_VERTEX_BUFFER_OBJECT])
1487     {
1488         TRACE("Not creating a vbo because GL_ARB_vertex_buffer is not supported\n");
1489     }
1490     else if(buffer->resource.pool == WINED3DPOOL_SYSTEMMEM)
1491     {
1492         TRACE("Not creating a vbo because the vertex buffer is in system memory\n");
1493     }
1494     else if(!dynamic_buffer_ok && (buffer->resource.usage & WINED3DUSAGE_DYNAMIC))
1495     {
1496         TRACE("Not creating a vbo because the buffer has dynamic usage and no GL support\n");
1497     }
1498     else
1499     {
1500         buffer->flags |= WINED3D_BUFFER_CREATEBO;
1501     }
1502
1503     if (data)
1504     {
1505         BYTE *ptr;
1506
1507         hr = IWineD3DBuffer_Map((IWineD3DBuffer *)buffer, 0, size, &ptr, 0);
1508         if (FAILED(hr))
1509         {
1510             ERR("Failed to map buffer, hr %#x\n", hr);
1511             buffer_UnLoad((IWineD3DBuffer *)buffer);
1512             resource_cleanup((IWineD3DResource *)buffer);
1513             return hr;
1514         }
1515
1516         memcpy(ptr, data, size);
1517
1518         hr = IWineD3DBuffer_Unmap((IWineD3DBuffer *)buffer);
1519         if (FAILED(hr))
1520         {
1521             ERR("Failed to unmap buffer, hr %#x\n", hr);
1522             buffer_UnLoad((IWineD3DBuffer *)buffer);
1523             resource_cleanup((IWineD3DResource *)buffer);
1524             return hr;
1525         }
1526     }
1527
1528     buffer->maps = HeapAlloc(GetProcessHeap(), 0, sizeof(*buffer->maps));
1529     if (!buffer->maps)
1530     {
1531         ERR("Out of memory\n");
1532         buffer_UnLoad((IWineD3DBuffer *)buffer);
1533         resource_cleanup((IWineD3DResource *)buffer);
1534         return E_OUTOFMEMORY;
1535     }
1536     buffer->maps_size = 1;
1537
1538     return WINED3D_OK;
1539 }