1/*
2 * Copyright 2011 Joakim Sindholt <opensource@zhasha.com>
3 * Copyright 2015 Patrick Rudolph <siro@das-labor.org>
4 *
5 * Permission is hereby granted, free of charge, to any person obtaining a
6 * copy of this software and associated documentation files (the "Software"),
7 * to deal in the Software without restriction, including without limitation
8 * on the rights to use, copy, modify, merge, publish, distribute, sub
9 * license, and/or sell copies of the Software, and to permit persons to whom
10 * the Software is furnished to do so, subject to the following conditions:
11 *
12 * The above copyright notice and this permission notice (including the next
13 * paragraph) shall be included in all copies or substantial portions of the
14 * Software.
15 *
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
20 * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
21 * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
22 * USE OR OTHER DEALINGS IN THE SOFTWARE. */
23
24#include "buffer9.h"
25#include "device9.h"
26#include "indexbuffer9.h"
27#include "nine_buffer_upload.h"
28#include "nine_helpers.h"
29#include "nine_pipe.h"
30
31#include "pipe/p_screen.h"
32#include "pipe/p_context.h"
33#include "pipe/p_state.h"
34#include "pipe/p_defines.h"
35#include "pipe/p_format.h"
36#include "util/u_box.h"
37#include "util/u_inlines.h"
38
39#define DBG_CHANNEL (DBG_INDEXBUFFER|DBG_VERTEXBUFFER)
40
41HRESULT
42NineBuffer9_ctor( struct NineBuffer9 *This,
43                        struct NineUnknownParams *pParams,
44                        D3DRESOURCETYPE Type,
45                        DWORD Usage,
46                        UINT Size,
47                        D3DPOOL Pool )
48{
49    struct pipe_resource *info = &This->base.info;
50    HRESULT hr;
51
52    DBG("This=%p Size=0x%x Usage=%x Pool=%u\n", This, Size, Usage, Pool);
53
54    user_assert(Pool != D3DPOOL_SCRATCH, D3DERR_INVALIDCALL);
55
56    This->maps = MALLOC(sizeof(struct NineTransfer));
57    if (!This->maps)
58        return E_OUTOFMEMORY;
59    This->nlocks = 0;
60    This->nmaps = 0;
61    This->maxmaps = 1;
62    This->size = Size;
63
64    info->screen = pParams->device->screen;
65    info->target = PIPE_BUFFER;
66    info->format = PIPE_FORMAT_R8_UNORM;
67    info->width0 = Size;
68    info->flags = 0;
69
70    /* Note: WRITEONLY is just tip for resource placement, the resource
71     * can still be read (but slower). */
72    info->bind = (Type == D3DRTYPE_INDEXBUFFER) ? PIPE_BIND_INDEX_BUFFER : PIPE_BIND_VERTEX_BUFFER;
73
74    /* Software vertex processing:
75     * If the device is full software vertex processing,
76     * then the buffer is supposed to be used only for sw processing.
77     * For mixed vertex processing, buffers with D3DUSAGE_SOFTWAREPROCESSING
78     * can be used for both sw and hw processing.
79     * These buffers are expected to be stored in RAM.
80     * Apps expect locking the full buffer with no flags, then
81     * render a a few primitive, then locking again, etc
82     * to be a fast pattern. Only the SYSTEMMEM DYNAMIC path
83     * will give that pattern ok performance in our case.
84     * An alternative would be when sw processing is detected to
85     * convert Draw* calls to Draw*Up calls. */
86    if (Usage & D3DUSAGE_SOFTWAREPROCESSING ||
87        pParams->device->params.BehaviorFlags & D3DCREATE_SOFTWARE_VERTEXPROCESSING) {
88        Pool = D3DPOOL_SYSTEMMEM;
89        Usage |= D3DUSAGE_DYNAMIC;
90        /* Note: the application cannot retrieve Pool and Usage */
91    }
92
93    /* Always use the DYNAMIC path for SYSTEMMEM.
94     * If the app uses the vertex buffer is a dynamic fashion,
95     * this is going to be very significantly faster that way.
96     * If the app uses the vertex buffer in a static fashion,
97     * instead of being filled all at once, the buffer will be filled
98     * little per little, until it is fully filled, thus the perf hit
99     * will be very small. */
100    if (Pool == D3DPOOL_SYSTEMMEM)
101        Usage |= D3DUSAGE_DYNAMIC;
102
103    /* It is hard to find clear information on where to place the buffer in
104     * memory depending on the flag.
105     * MSDN: resources are static, except for those with DYNAMIC, thus why you
106     *   can only use DISCARD on them.
107     * ATI doc: The driver has the liberty it wants for having things static
108     *   or not.
109     *   MANAGED: Ram + uploads to Vram copy at unlock (msdn and nvidia doc say
110     *   at first draw call using the buffer)
111     *   DEFAULT + Usage = 0 => System memory backing for easy read access
112     *   (That doc is very unclear on the details, like whether some copies to
113     *   vram copy are involved or not).
114     *   DEFAULT + WRITEONLY => Vram
115     *   DEFAULT + WRITEONLY + DYNAMIC => Either Vram buffer or GTT_WC, depending on what the driver wants.
116     *   SYSTEMMEM: Same as MANAGED, but handled by the driver instead of the runtime (which means
117     *   some small behavior differences between vendors). Implementing exactly as MANAGED should
118     *   be fine.
119     */
120    if (Pool == D3DPOOL_SYSTEMMEM && Usage & D3DUSAGE_DYNAMIC)
121        info->usage = PIPE_USAGE_STREAM;
122    else if (Pool != D3DPOOL_DEFAULT)
123        info->usage = PIPE_USAGE_DEFAULT;
124    else if (Usage & D3DUSAGE_DYNAMIC && Usage & D3DUSAGE_WRITEONLY)
125        info->usage = PIPE_USAGE_STREAM;
126    else if (Usage & D3DUSAGE_WRITEONLY)
127        info->usage = PIPE_USAGE_DEFAULT;
128    /* For the remaining two, PIPE_USAGE_STAGING would probably be
129     * a good fit according to the doc. However it seems rather a mistake
130     * from apps to use these (mistakes that do really happen). Try
131     * to put the flags that are the best compromise between the real
132     * behaviour and what buggy apps should get for better performance. */
133    else if (Usage & D3DUSAGE_DYNAMIC)
134        info->usage = PIPE_USAGE_STREAM;
135    else
136        info->usage = PIPE_USAGE_DYNAMIC;
137
138    /* When Writeonly is not set, we don't want to enable the
139     * optimizations */
140    This->discard_nooverwrite_only = !!(Usage & D3DUSAGE_WRITEONLY) &&
141                                     pParams->device->buffer_upload;
142    /* if (pDesc->Usage & D3DUSAGE_DONOTCLIP) { } */
143    /* if (pDesc->Usage & D3DUSAGE_NONSECURE) { } */
144    /* if (pDesc->Usage & D3DUSAGE_NPATCHES) { } */
145    /* if (pDesc->Usage & D3DUSAGE_POINTS) { } */
146    /* if (pDesc->Usage & D3DUSAGE_RTPATCHES) { } */
147    /* if (pDesc->Usage & D3DUSAGE_TEXTAPI) { } */
148
149    info->height0 = 1;
150    info->depth0 = 1;
151    info->array_size = 1;
152    info->last_level = 0;
153    info->nr_samples = 0;
154    info->nr_storage_samples = 0;
155
156    hr = NineResource9_ctor(&This->base, pParams, NULL, TRUE,
157                            Type, Pool, Usage);
158
159    if (FAILED(hr))
160        return hr;
161
162    if (Pool != D3DPOOL_DEFAULT) {
163        This->managed.data = align_calloc(
164            nine_format_get_level_alloc_size(This->base.info.format,
165                                             Size, 1, 0), 32);
166        if (!This->managed.data)
167            return E_OUTOFMEMORY;
168        This->managed.dirty = TRUE;
169        u_box_1d(0, Size, &This->managed.dirty_box);
170        u_box_1d(0, 0, &This->managed.valid_region);
171        u_box_1d(0, 0, &This->managed.required_valid_region);
172        u_box_1d(0, 0, &This->managed.filled_region);
173        This->managed.can_unsynchronized = true;
174        This->managed.num_worker_thread_syncs = 0;
175        list_inithead(&This->managed.list);
176        list_inithead(&This->managed.list2);
177        list_add(&This->managed.list2, &pParams->device->managed_buffers);
178    }
179
180    return D3D_OK;
181}
182
183void
184NineBuffer9_dtor( struct NineBuffer9 *This )
185{
186    DBG("This=%p\n", This);
187
188    if (This->maps) {
189        while (This->nlocks) {
190            NineBuffer9_Unlock(This);
191        }
192        assert(!This->nmaps);
193        FREE(This->maps);
194    }
195
196    if (This->base.pool != D3DPOOL_DEFAULT) {
197        if (This->managed.data)
198            align_free(This->managed.data);
199        if (list_is_linked(&This->managed.list))
200            list_del(&This->managed.list);
201        if (list_is_linked(&This->managed.list2))
202            list_del(&This->managed.list2);
203    }
204
205    if (This->buf)
206        nine_upload_release_buffer(This->base.base.device->buffer_upload, This->buf);
207
208    NineResource9_dtor(&This->base);
209}
210
211struct pipe_resource *
212NineBuffer9_GetResource( struct NineBuffer9 *This, unsigned *offset )
213{
214    if (This->buf)
215        return nine_upload_buffer_resource_and_offset(This->buf, offset);
216    *offset = 0;
217    return NineResource9_GetResource(&This->base);
218}
219
220static void
221NineBuffer9_RebindIfRequired( struct NineBuffer9 *This,
222                              struct NineDevice9 *device,
223                              struct pipe_resource *resource,
224                              unsigned offset )
225{
226    int i;
227
228    if (!This->bind_count)
229        return;
230    for (i = 0; i < device->caps.MaxStreams; i++) {
231        if (device->state.stream[i] == (struct NineVertexBuffer9 *)This)
232            nine_context_set_stream_source_apply(device, i,
233                                                 resource,
234                                                 device->state.vtxbuf[i].buffer_offset + offset,
235                                                 device->state.vtxbuf[i].stride);
236    }
237    if (device->state.idxbuf == (struct NineIndexBuffer9 *)This)
238        nine_context_set_indices_apply(device, resource,
239                                       ((struct NineIndexBuffer9 *)This)->index_size,
240                                       offset);
241}
242
243HRESULT NINE_WINAPI
244NineBuffer9_Lock( struct NineBuffer9 *This,
245                        UINT OffsetToLock,
246                        UINT SizeToLock,
247                        void **ppbData,
248                        DWORD Flags )
249{
250    struct NineDevice9 *device = This->base.base.device;
251    struct pipe_box box;
252    struct pipe_context *pipe;
253    void *data;
254    unsigned usage;
255
256    DBG("This=%p(pipe=%p) OffsetToLock=0x%x, SizeToLock=0x%x, Flags=0x%x\n",
257        This, This->base.resource,
258        OffsetToLock, SizeToLock, Flags);
259
260    user_assert(ppbData, E_POINTER);
261
262    if (SizeToLock == 0) {
263        SizeToLock = This->size - OffsetToLock;
264        user_warn(OffsetToLock != 0);
265    }
266
267    /* Write out of bound seems to have to be taken into account for these.
268     * TODO: Do more tests (is it only at buffer first lock ? etc).
269     * Since these buffers are supposed to be locked once and never
270     * writen again (MANAGED or DYNAMIC is used for the other uses cases),
271     * performance should be unaffected. */
272    if (!(This->base.usage & D3DUSAGE_DYNAMIC) && This->base.pool == D3DPOOL_DEFAULT)
273        SizeToLock = This->size - OffsetToLock;
274
275    SizeToLock = MIN2(SizeToLock, This->size - OffsetToLock); /* Do not read or track out of the buffer */
276    u_box_1d(OffsetToLock, SizeToLock, &box);
277
278    if (This->base.pool != D3DPOOL_DEFAULT) {
279        /* MANAGED: READONLY doesn't dirty the buffer, nor
280         * wait the upload in the worker thread
281         * SYSTEMMEM: AMD/NVidia: All locks dirty the full buffer. Not on Intel
282         * For Nvidia, SYSTEMMEM behaves are if there is no worker thread.
283         * On AMD, READONLY and NOOVERWRITE do dirty the buffer, but do not sync the previous uploads
284         * in the worker thread. On Intel only NOOVERWRITE has that effect.
285         * We implement the AMD behaviour. */
286        if (This->base.pool == D3DPOOL_MANAGED) {
287            if (!(Flags & D3DLOCK_READONLY)) {
288                if (!This->managed.dirty) {
289                    assert(list_is_empty(&This->managed.list));
290                    This->managed.dirty = TRUE;
291                    This->managed.dirty_box = box;
292                    /* Flush if regions pending to be uploaded would be dirtied */
293                    if (p_atomic_read(&This->managed.pending_upload)) {
294                        u_box_intersect_1d(&box, &box, &This->managed.upload_pending_regions);
295                        if (box.width != 0)
296                            nine_csmt_process(This->base.base.device);
297                    }
298                } else
299                    u_box_union_1d(&This->managed.dirty_box, &This->managed.dirty_box, &box);
300                /* Tests trying to draw while the buffer is locked show that
301                 * SYSTEMMEM/MANAGED buffers are made dirty at Lock time */
302                BASEBUF_REGISTER_UPDATE(This);
303            }
304        } else {
305            if (!(Flags & (D3DLOCK_READONLY|D3DLOCK_NOOVERWRITE)) &&
306                p_atomic_read(&This->managed.pending_upload)) {
307                This->managed.num_worker_thread_syncs++;
308                /* If we sync too often, pick the vertex_uploader path */
309                if (This->managed.num_worker_thread_syncs >= 3)
310                    This->managed.can_unsynchronized = false;
311                nine_csmt_process(This->base.base.device);
312                /* Note: AS DISCARD is not relevant for SYSTEMMEM,
313                 * NOOVERWRITE might have a similar meaning as what is
314                 * in D3D7 doc. Basically that data from previous draws
315                 * OF THIS FRAME are unaffected. As we flush csmt in Present(),
316                 * we should be correct. In some parts of the doc, the notion
317                 * of frame is implied to be related to Begin/EndScene(),
318                 * but tests show NOOVERWRITE after EndScene() doesn't flush
319                 * the csmt thread. */
320            }
321            This->managed.dirty = true;
322            u_box_1d(0, This->size, &This->managed.dirty_box); /* systemmem non-dynamic */
323            u_box_1d(0, 0, &This->managed.valid_region); /* systemmem dynamic */
324            BASEBUF_REGISTER_UPDATE(This);
325        }
326
327        *ppbData = (int8_t *)This->managed.data + OffsetToLock;
328        DBG("returning pointer %p\n", *ppbData);
329        This->nlocks++;
330        return D3D_OK;
331    }
332
333    /* Driver ddi doc: READONLY is never passed to the device. So it can only
334     * have effect on things handled by the driver (MANAGED pool for example).
335     * Msdn doc: DISCARD and NOOVERWRITE are only for DYNAMIC.
336     * ATI doc: You can use DISCARD and NOOVERWRITE without DYNAMIC.
337     * Msdn doc: D3DLOCK_DONOTWAIT is not among the valid flags for buffers.
338     * Our tests: On win 7 nvidia, D3DLOCK_DONOTWAIT does return
339     * D3DERR_WASSTILLDRAWING if the resource is in use, except for DYNAMIC.
340     * Our tests: some apps do use both DISCARD and NOOVERWRITE at the same
341     * time. On windows it seems to return different pointer in some conditions,
342     * creation flags and drivers. However these tests indicate having
343     * NOOVERWRITE win is a valid behaviour (NVidia).
344     */
345
346    /* Have NOOVERWRITE win over DISCARD. This is allowed (see above) and
347     * it prevents overconsuming buffers if apps do use both at the same time. */
348    if ((Flags & (D3DLOCK_DISCARD | D3DLOCK_NOOVERWRITE)) == (D3DLOCK_DISCARD | D3DLOCK_NOOVERWRITE))
349        Flags &= ~D3DLOCK_DISCARD;
350
351    if (Flags & D3DLOCK_DISCARD)
352        usage = PIPE_MAP_WRITE | PIPE_MAP_DISCARD_WHOLE_RESOURCE;
353    else if (Flags & D3DLOCK_NOOVERWRITE)
354        usage = PIPE_MAP_WRITE | PIPE_MAP_UNSYNCHRONIZED;
355    else
356        /* Do not ask for READ if writeonly and default pool (should be safe enough,
357         * as the doc says app shouldn't expect reading to work with writeonly). */
358        usage = (This->base.usage & D3DUSAGE_WRITEONLY) ?
359            PIPE_MAP_WRITE :
360            PIPE_MAP_READ_WRITE;
361    if (Flags & D3DLOCK_DONOTWAIT && !(This->base.usage & D3DUSAGE_DYNAMIC))
362        usage |= PIPE_MAP_DONTBLOCK;
363
364    This->discard_nooverwrite_only &= !!(Flags & (D3DLOCK_DISCARD | D3DLOCK_NOOVERWRITE));
365
366    if (This->nmaps == This->maxmaps) {
367        struct NineTransfer *newmaps =
368            REALLOC(This->maps, sizeof(struct NineTransfer)*This->maxmaps,
369                    sizeof(struct NineTransfer)*(This->maxmaps << 1));
370        if (newmaps == NULL)
371            return E_OUTOFMEMORY;
372
373        This->maxmaps <<= 1;
374        This->maps = newmaps;
375    }
376
377    if (This->buf && !This->discard_nooverwrite_only) {
378        struct pipe_box src_box;
379        unsigned offset;
380        struct pipe_resource *src_res;
381        DBG("Disabling nine_subbuffer for a buffer having"
382            "used a nine_subbuffer buffer\n");
383        /* Copy buffer content to the buffer resource, which
384         * we will now use.
385         * Note: The behaviour may be different from what is expected
386         * with double lock. However applications can't really make expectations
387         * about double locks, and don't really use them, so that's ok. */
388        src_res = nine_upload_buffer_resource_and_offset(This->buf, &offset);
389        u_box_1d(offset, This->size, &src_box);
390
391        pipe = NineDevice9_GetPipe(device);
392        pipe->resource_copy_region(pipe, This->base.resource, 0, 0, 0, 0,
393                                   src_res, 0, &src_box);
394        /* Release previous resource */
395        if (This->nmaps >= 1)
396            This->maps[This->nmaps-1].should_destroy_buf = true;
397        else
398            nine_upload_release_buffer(device->buffer_upload, This->buf);
399        This->buf = NULL;
400        /* Rebind buffer */
401        NineBuffer9_RebindIfRequired(This, device, This->base.resource, 0);
402    }
403
404    This->maps[This->nmaps].transfer = NULL;
405    This->maps[This->nmaps].is_pipe_secondary = false;
406    This->maps[This->nmaps].buf = NULL;
407    This->maps[This->nmaps].should_destroy_buf = false;
408
409    if (This->discard_nooverwrite_only) {
410        if (This->buf && (Flags & D3DLOCK_DISCARD)) {
411            /* Release previous buffer */
412            if (This->nmaps >= 1)
413                This->maps[This->nmaps-1].should_destroy_buf = true;
414            else
415                nine_upload_release_buffer(device->buffer_upload, This->buf);
416            This->buf = NULL;
417        }
418
419        if (!This->buf) {
420            unsigned offset;
421            struct pipe_resource *res;
422            This->buf = nine_upload_create_buffer(device->buffer_upload, This->base.info.width0);
423            res = nine_upload_buffer_resource_and_offset(This->buf, &offset);
424            NineBuffer9_RebindIfRequired(This, device, res, offset);
425        }
426
427        if (This->buf) {
428            This->maps[This->nmaps].buf = This->buf;
429            This->nmaps++;
430            This->nlocks++;
431            DBG("Returning %p\n", nine_upload_buffer_get_map(This->buf) + OffsetToLock);
432            *ppbData = nine_upload_buffer_get_map(This->buf) + OffsetToLock;
433            return D3D_OK;
434        } else {
435            /* Fallback to normal path, and don't try again */
436            This->discard_nooverwrite_only = false;
437        }
438    }
439
440    /* Previous mappings may need pending commands to write to the
441     * buffer (staging buffer for example). Before a NOOVERWRITE,
442     * we thus need a finish, to guarantee any upload is finished.
443     * Note for discard_nooverwrite_only we don't need to do this
444     * check as neither discard nor nooverwrite have issues there */
445    if (This->need_sync_if_nooverwrite && !(Flags & D3DLOCK_DISCARD) &&
446        (Flags & D3DLOCK_NOOVERWRITE)) {
447        struct pipe_screen *screen = NineDevice9_GetScreen(device);
448        struct pipe_fence_handle *fence = NULL;
449
450        pipe = NineDevice9_GetPipe(device);
451        pipe->flush(pipe, &fence, 0);
452        (void) screen->fence_finish(screen, NULL, fence, PIPE_TIMEOUT_INFINITE);
453        screen->fence_reference(screen, &fence, NULL);
454    }
455    This->need_sync_if_nooverwrite = !(Flags & (D3DLOCK_DISCARD | D3DLOCK_NOOVERWRITE));
456
457    /* When csmt is active, we want to avoid stalls as much as possible,
458     * and thus we want to create a new resource on discard and map it
459     * with the secondary pipe, instead of waiting on the main pipe. */
460    if (Flags & D3DLOCK_DISCARD && device->csmt_active) {
461        struct pipe_screen *screen = NineDevice9_GetScreen(device);
462        struct pipe_resource *new_res = nine_resource_create_with_retry(device, screen, &This->base.info);
463        if (new_res) {
464            /* Use the new resource */
465            pipe_resource_reference(&This->base.resource, new_res);
466            pipe_resource_reference(&new_res, NULL);
467            usage = PIPE_MAP_WRITE | PIPE_MAP_UNSYNCHRONIZED;
468            NineBuffer9_RebindIfRequired(This, device, This->base.resource, 0);
469            This->maps[This->nmaps].is_pipe_secondary = TRUE;
470        }
471    } else if (Flags & D3DLOCK_NOOVERWRITE && device->csmt_active)
472        This->maps[This->nmaps].is_pipe_secondary = TRUE;
473
474    if (This->maps[This->nmaps].is_pipe_secondary)
475        pipe = device->pipe_secondary;
476    else
477        pipe = NineDevice9_GetPipe(device);
478
479    data = pipe->buffer_map(pipe, This->base.resource, 0,
480                              usage, &box, &This->maps[This->nmaps].transfer);
481
482    if (!data) {
483        DBG("pipe::buffer_map failed\n"
484            " usage = %x\n"
485            " box.x = %u\n"
486            " box.width = %u\n",
487            usage, box.x, box.width);
488
489        if (Flags & D3DLOCK_DONOTWAIT)
490            return D3DERR_WASSTILLDRAWING;
491        return D3DERR_INVALIDCALL;
492    }
493
494    DBG("returning pointer %p\n", data);
495    This->nmaps++;
496    This->nlocks++;
497    *ppbData = data;
498
499    return D3D_OK;
500}
501
502HRESULT NINE_WINAPI
503NineBuffer9_Unlock( struct NineBuffer9 *This )
504{
505    struct NineDevice9 *device = This->base.base.device;
506    struct pipe_context *pipe;
507    int i;
508    DBG("This=%p\n", This);
509
510    user_assert(This->nlocks > 0, D3DERR_INVALIDCALL);
511    This->nlocks--;
512    if (This->nlocks > 0)
513        return D3D_OK; /* Pending unlocks. Wait all unlocks before unmapping */
514
515    if (This->base.pool == D3DPOOL_DEFAULT) {
516        for (i = 0; i < This->nmaps; i++) {
517            if (!This->maps[i].buf) {
518                pipe = This->maps[i].is_pipe_secondary ?
519                    device->pipe_secondary :
520                    nine_context_get_pipe_acquire(device);
521                pipe->buffer_unmap(pipe, This->maps[i].transfer);
522                /* We need to flush in case the driver does implicit copies */
523                if (This->maps[i].is_pipe_secondary)
524                    pipe->flush(pipe, NULL, 0);
525                else
526                    nine_context_get_pipe_release(device);
527            } else if (This->maps[i].should_destroy_buf)
528                nine_upload_release_buffer(device->buffer_upload, This->maps[i].buf);
529        }
530        This->nmaps = 0;
531    }
532    return D3D_OK;
533}
534
535void
536NineBuffer9_SetDirty( struct NineBuffer9 *This )
537{
538    assert(This->base.pool != D3DPOOL_DEFAULT);
539
540    This->managed.dirty = TRUE;
541    u_box_1d(0, This->size, &This->managed.dirty_box);
542    BASEBUF_REGISTER_UPDATE(This);
543}
544
545/* Try to remove b from a, supposed to include b */
546static void u_box_try_remove_region_1d(struct pipe_box *dst,
547                                       const struct pipe_box *a,
548                                       const struct pipe_box *b)
549{
550    int x, width;
551    if (a->x == b->x) {
552        x = a->x + b->width;
553        width = a->width - b->width;
554    } else if ((a->x + a->width) == (b->x + b->width)) {
555        x = a->x;
556        width = a->width - b->width;
557    } else {
558        x = a->x;
559        width = a->width;
560    }
561    dst->x = x;
562    dst->width = width;
563}
564
565void
566NineBuffer9_Upload( struct NineBuffer9 *This )
567{
568    struct NineDevice9 *device = This->base.base.device;
569    unsigned upload_flags = 0;
570    struct pipe_box box_upload;
571
572    assert(This->base.pool != D3DPOOL_DEFAULT && This->managed.dirty);
573
574    if (This->base.pool == D3DPOOL_SYSTEMMEM && This->base.usage & D3DUSAGE_DYNAMIC) {
575        struct pipe_box region_already_valid;
576        struct pipe_box conflicting_region;
577        struct pipe_box *valid_region = &This->managed.valid_region;
578        struct pipe_box *required_valid_region = &This->managed.required_valid_region;
579        struct pipe_box *filled_region = &This->managed.filled_region;
580        /* Try to upload SYSTEMMEM DYNAMIC in an efficient fashion.
581         * Unlike non-dynamic for which we upload the whole dirty region, try to
582         * only upload the data needed for the draw. The draw call preparation
583         * fills This->managed.required_valid_region for that */
584        u_box_intersect_1d(&region_already_valid,
585                           valid_region,
586                           required_valid_region);
587        /* If the required valid region is already valid, nothing to do */
588        if (region_already_valid.x == required_valid_region->x &&
589            region_already_valid.width == required_valid_region->width) {
590            /* Rebind if the region happens to be valid in the original buffer
591             * but we have since used vertex_uploader */
592            if (!This->managed.can_unsynchronized)
593                NineBuffer9_RebindIfRequired(This, device, This->base.resource, 0);
594            u_box_1d(0, 0, required_valid_region);
595            return;
596        }
597        /* (Try to) Remove valid areas from the region to upload */
598        u_box_try_remove_region_1d(&box_upload,
599                                   required_valid_region,
600                                   &region_already_valid);
601        assert(box_upload.width > 0);
602        /* To maintain correctly the valid region, as we will do union later with
603         * box_upload, we must ensure box_upload is consecutive with valid_region */
604        if (box_upload.x > valid_region->x + valid_region->width && valid_region->width > 0) {
605            box_upload.width = box_upload.x + box_upload.width - (valid_region->x + valid_region->width);
606            box_upload.x = valid_region->x + valid_region->width;
607        } else if (box_upload.x + box_upload.width < valid_region->x && valid_region->width > 0) {
608            box_upload.width = valid_region->x - box_upload.x;
609        }
610        /* There is conflict if some areas, that are not valid but are filled for previous draw calls,
611         * intersect with the region we plan to upload. Note by construction valid_region IS
612         * included in filled_region, thus so is region_already_valid. */
613        u_box_intersect_1d(&conflicting_region, &box_upload, filled_region);
614        /* As box_upload could still contain region_already_valid, check the intersection
615         * doesn't happen to be exactly region_already_valid (it cannot be smaller, see above) */
616        if (This->managed.can_unsynchronized && (conflicting_region.width == 0 ||
617            (conflicting_region.x == region_already_valid.x &&
618             conflicting_region.width == region_already_valid.width))) {
619            /* No conflicts. */
620            upload_flags |= PIPE_MAP_UNSYNCHRONIZED;
621        } else {
622            /* We cannot use PIPE_MAP_UNSYNCHRONIZED. We must choose between no flag and DISCARD.
623             * Criterias to discard:
624             * . Most of the resource was filled (but some apps do allocate a big buffer
625             * to only use a small part in a round fashion)
626             * . The region to upload is very small compared to the filled region and
627             * at the start of the buffer (hints at round usage starting again)
628             * . The region to upload is very big compared to the required region
629             * . We have not discarded yet this frame
630             * If the buffer use pattern seems to sync the worker thread too often,
631             * revert to the vertex_uploader */
632            if (This->managed.num_worker_thread_syncs < 3 &&
633                (filled_region->width > (This->size / 2) ||
634                 (10 * box_upload.width < filled_region->width &&
635                  box_upload.x < (filled_region->x + filled_region->width)/2) ||
636                 box_upload.width > 2 * required_valid_region->width ||
637                 This->managed.frame_count_last_discard != device->frame_count)) {
638                /* Avoid DISCARDING too much by discarding only if most of the buffer
639                 * has been used */
640                DBG_FLAG(DBG_INDEXBUFFER|DBG_VERTEXBUFFER,
641             "Uploading %p DISCARD: valid %d %d, filled %d %d, required %d %d, box_upload %d %d, required already_valid %d %d, conficting %d %d\n",
642             This, valid_region->x, valid_region->width, filled_region->x, filled_region->width,
643             required_valid_region->x, required_valid_region->width, box_upload.x, box_upload.width,
644             region_already_valid.x, region_already_valid.width, conflicting_region.x, conflicting_region.width
645                );
646                upload_flags |= PIPE_MAP_DISCARD_WHOLE_RESOURCE;
647                u_box_1d(0, 0, filled_region);
648                u_box_1d(0, 0, valid_region);
649                box_upload = This->managed.required_valid_region;
650                /* Rebind the buffer if we used intermediate alternative buffer */
651                if (!This->managed.can_unsynchronized)
652                    NineBuffer9_RebindIfRequired(This, device, This->base.resource, 0);
653                This->managed.can_unsynchronized = true;
654                This->managed.frame_count_last_discard = device->frame_count;
655            } else {
656                /* Once we use without UNSYNCHRONIZED, we cannot use it anymore.
657                 * Use a different buffer. */
658                unsigned buffer_offset = 0;
659                struct pipe_resource *resource = NULL;
660                This->managed.can_unsynchronized = false;
661                u_upload_data(device->vertex_uploader,
662                    required_valid_region->x,
663                    required_valid_region->width,
664                    64,
665                    This->managed.data + required_valid_region->x,
666                    &buffer_offset,
667                    &resource);
668                buffer_offset -= required_valid_region->x;
669                u_upload_unmap(device->vertex_uploader);
670                if (resource) {
671                    NineBuffer9_RebindIfRequired(This, device, resource, buffer_offset);
672                    /* Note: This only works because for these types of buffers this function
673                     * is called before every draw call. Else it wouldn't work when the app
674                     * rebinds buffers. In addition it needs this function to be called only
675                     * once per buffers even if bound several times, which we do. */
676                    u_box_1d(0, 0, required_valid_region);
677                    pipe_resource_reference(&resource, NULL);
678                    return;
679                }
680            }
681        }
682
683        u_box_union_1d(filled_region,
684                       filled_region,
685                       &box_upload);
686        u_box_union_1d(valid_region,
687                       valid_region,
688                       &box_upload);
689        u_box_1d(0, 0, required_valid_region);
690    } else
691        box_upload = This->managed.dirty_box;
692
693    if (box_upload.x == 0 && box_upload.width == This->size) {
694        upload_flags |= PIPE_MAP_DISCARD_WHOLE_RESOURCE;
695    }
696
697    if (This->managed.pending_upload) {
698        u_box_union_1d(&This->managed.upload_pending_regions,
699                       &This->managed.upload_pending_regions,
700                       &box_upload);
701    } else {
702        This->managed.upload_pending_regions = box_upload;
703    }
704
705    DBG_FLAG(DBG_INDEXBUFFER|DBG_VERTEXBUFFER,
706             "Uploading %p, offset=%d, size=%d, Flags=0x%x\n",
707             This, box_upload.x, box_upload.width, upload_flags);
708    nine_context_range_upload(device, &This->managed.pending_upload,
709                              (struct NineUnknown *)This,
710                              This->base.resource,
711                              box_upload.x,
712                              box_upload.width,
713                              upload_flags,
714                              (int8_t *)This->managed.data + box_upload.x);
715    This->managed.dirty = FALSE;
716}
717