freedreno/a4xx: only align slices in non-layer_first textures
[mesa.git] / src / gallium / drivers / freedreno / freedreno_resource.c
1 /* -*- mode: C; c-file-style: "k&r"; tab-width 4; indent-tabs-mode: t; -*- */
2
3 /*
4 * Copyright (C) 2012 Rob Clark <robclark@freedesktop.org>
5 *
6 * Permission is hereby granted, free of charge, to any person obtaining a
7 * copy of this software and associated documentation files (the "Software"),
8 * to deal in the Software without restriction, including without limitation
9 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
10 * and/or sell copies of the Software, and to permit persons to whom the
11 * Software is furnished to do so, subject to the following conditions:
12 *
13 * The above copyright notice and this permission notice (including the next
14 * paragraph) shall be included in all copies or substantial portions of the
15 * Software.
16 *
17 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
18 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
19 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
20 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
21 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
22 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
23 * SOFTWARE.
24 *
25 * Authors:
26 * Rob Clark <robclark@freedesktop.org>
27 */
28
29 #include "util/u_format.h"
30 #include "util/u_format_rgtc.h"
31 #include "util/u_format_zs.h"
32 #include "util/u_inlines.h"
33 #include "util/u_transfer.h"
34 #include "util/u_string.h"
35 #include "util/u_surface.h"
36
37 #include "freedreno_resource.h"
38 #include "freedreno_screen.h"
39 #include "freedreno_surface.h"
40 #include "freedreno_context.h"
41 #include "freedreno_query_hw.h"
42 #include "freedreno_util.h"
43
44 #include <errno.h>
45
46
47 static bool
48 pending(struct fd_resource *rsc, enum fd_resource_status status)
49 {
50 return (rsc->status & status) ||
51 (rsc->stencil && (rsc->stencil->status & status));
52 }
53
54 static void
55 fd_invalidate_resource(struct fd_context *ctx, struct pipe_resource *prsc)
56 {
57 int i;
58
59 /* Go through the entire state and see if the resource is bound
60 * anywhere. If it is, mark the relevant state as dirty. This is called on
61 * realloc_bo.
62 */
63
64 /* Constbufs */
65 for (i = 1; i < PIPE_MAX_CONSTANT_BUFFERS && !(ctx->dirty & FD_DIRTY_CONSTBUF); i++) {
66 if (ctx->constbuf[PIPE_SHADER_VERTEX].cb[i].buffer == prsc)
67 ctx->dirty |= FD_DIRTY_CONSTBUF;
68 if (ctx->constbuf[PIPE_SHADER_FRAGMENT].cb[i].buffer == prsc)
69 ctx->dirty |= FD_DIRTY_CONSTBUF;
70 }
71
72 /* VBOs */
73 for (i = 0; i < ctx->vtx.vertexbuf.count && !(ctx->dirty & FD_DIRTY_VTXBUF); i++) {
74 if (ctx->vtx.vertexbuf.vb[i].buffer == prsc)
75 ctx->dirty |= FD_DIRTY_VTXBUF;
76 }
77
78 /* Index buffer */
79 if (ctx->indexbuf.buffer == prsc)
80 ctx->dirty |= FD_DIRTY_INDEXBUF;
81
82 /* Textures */
83 for (i = 0; i < ctx->verttex.num_textures && !(ctx->dirty & FD_DIRTY_VERTTEX); i++) {
84 if (ctx->verttex.textures[i] && (ctx->verttex.textures[i]->texture == prsc))
85 ctx->dirty |= FD_DIRTY_VERTTEX;
86 }
87 for (i = 0; i < ctx->fragtex.num_textures && !(ctx->dirty & FD_DIRTY_FRAGTEX); i++) {
88 if (ctx->fragtex.textures[i] && (ctx->fragtex.textures[i]->texture == prsc))
89 ctx->dirty |= FD_DIRTY_FRAGTEX;
90 }
91 }
92
93 static void
94 realloc_bo(struct fd_resource *rsc, uint32_t size)
95 {
96 struct fd_screen *screen = fd_screen(rsc->base.b.screen);
97 uint32_t flags = DRM_FREEDRENO_GEM_CACHE_WCOMBINE |
98 DRM_FREEDRENO_GEM_TYPE_KMEM; /* TODO */
99
100 /* if we start using things other than write-combine,
101 * be sure to check for PIPE_RESOURCE_FLAG_MAP_COHERENT
102 */
103
104 if (rsc->bo)
105 fd_bo_del(rsc->bo);
106
107 rsc->bo = fd_bo_new(screen->dev, size, flags);
108 rsc->timestamp = 0;
109 rsc->status = 0;
110 rsc->pending_ctx = NULL;
111 list_delinit(&rsc->list);
112 util_range_set_empty(&rsc->valid_buffer_range);
113 }
114
115 static unsigned
116 fd_resource_layer_offset(struct fd_resource *rsc,
117 struct fd_resource_slice *slice,
118 unsigned layer)
119 {
120 if (rsc->layer_first)
121 return layer * rsc->layer_size;
122 else
123 return layer * slice->size0;
124 }
125
126 static void
127 fd_resource_flush_z32s8(struct fd_transfer *trans, const struct pipe_box *box)
128 {
129 struct fd_resource *rsc = fd_resource(trans->base.resource);
130 struct fd_resource_slice *slice = fd_resource_slice(rsc, trans->base.level);
131 struct fd_resource_slice *sslice = fd_resource_slice(rsc->stencil, trans->base.level);
132 enum pipe_format format = trans->base.resource->format;
133
134 float *depth = fd_bo_map(rsc->bo) + slice->offset +
135 fd_resource_layer_offset(rsc, slice, trans->base.box.z) +
136 (trans->base.box.y + box->y) * slice->pitch * 4 + (trans->base.box.x + box->x) * 4;
137 uint8_t *stencil = fd_bo_map(rsc->stencil->bo) + sslice->offset +
138 fd_resource_layer_offset(rsc->stencil, sslice, trans->base.box.z) +
139 (trans->base.box.y + box->y) * sslice->pitch + trans->base.box.x + box->x;
140
141 if (format != PIPE_FORMAT_X32_S8X24_UINT)
142 util_format_z32_float_s8x24_uint_unpack_z_float(
143 depth, slice->pitch * 4,
144 trans->staging, trans->base.stride,
145 box->width, box->height);
146
147 util_format_z32_float_s8x24_uint_unpack_s_8uint(
148 stencil, sslice->pitch,
149 trans->staging, trans->base.stride,
150 box->width, box->height);
151 }
152
153 static void
154 fd_resource_flush_rgtc(struct fd_transfer *trans, const struct pipe_box *box)
155 {
156 struct fd_resource *rsc = fd_resource(trans->base.resource);
157 struct fd_resource_slice *slice = fd_resource_slice(rsc, trans->base.level);
158 enum pipe_format format = trans->base.resource->format;
159
160 uint8_t *data = fd_bo_map(rsc->bo) + slice->offset +
161 fd_resource_layer_offset(rsc, slice, trans->base.box.z) +
162 ((trans->base.box.y + box->y) * slice->pitch +
163 trans->base.box.x + box->x) * rsc->cpp;
164
165 uint8_t *source = trans->staging +
166 util_format_get_nblocksy(format, box->y) * trans->base.stride +
167 util_format_get_stride(format, box->x);
168
169 switch (format) {
170 case PIPE_FORMAT_RGTC1_UNORM:
171 case PIPE_FORMAT_RGTC1_SNORM:
172 case PIPE_FORMAT_LATC1_UNORM:
173 case PIPE_FORMAT_LATC1_SNORM:
174 util_format_rgtc1_unorm_unpack_rgba_8unorm(
175 data, slice->pitch * rsc->cpp,
176 source, trans->base.stride,
177 box->width, box->height);
178 break;
179 case PIPE_FORMAT_RGTC2_UNORM:
180 case PIPE_FORMAT_RGTC2_SNORM:
181 case PIPE_FORMAT_LATC2_UNORM:
182 case PIPE_FORMAT_LATC2_SNORM:
183 util_format_rgtc2_unorm_unpack_rgba_8unorm(
184 data, slice->pitch * rsc->cpp,
185 source, trans->base.stride,
186 box->width, box->height);
187 break;
188 default:
189 assert(!"Unexpected format\n");
190 break;
191 }
192 }
193
194 static void
195 fd_resource_flush(struct fd_transfer *trans, const struct pipe_box *box)
196 {
197 enum pipe_format format = trans->base.resource->format;
198
199 switch (format) {
200 case PIPE_FORMAT_Z32_FLOAT_S8X24_UINT:
201 case PIPE_FORMAT_X32_S8X24_UINT:
202 fd_resource_flush_z32s8(trans, box);
203 break;
204 case PIPE_FORMAT_RGTC1_UNORM:
205 case PIPE_FORMAT_RGTC1_SNORM:
206 case PIPE_FORMAT_RGTC2_UNORM:
207 case PIPE_FORMAT_RGTC2_SNORM:
208 case PIPE_FORMAT_LATC1_UNORM:
209 case PIPE_FORMAT_LATC1_SNORM:
210 case PIPE_FORMAT_LATC2_UNORM:
211 case PIPE_FORMAT_LATC2_SNORM:
212 fd_resource_flush_rgtc(trans, box);
213 break;
214 default:
215 assert(!"Unexpected staging transfer type");
216 break;
217 }
218 }
219
220 static void fd_resource_transfer_flush_region(struct pipe_context *pctx,
221 struct pipe_transfer *ptrans,
222 const struct pipe_box *box)
223 {
224 struct fd_resource *rsc = fd_resource(ptrans->resource);
225 struct fd_transfer *trans = fd_transfer(ptrans);
226
227 if (ptrans->resource->target == PIPE_BUFFER)
228 util_range_add(&rsc->valid_buffer_range,
229 ptrans->box.x + box->x,
230 ptrans->box.x + box->x + box->width);
231
232 if (trans->staging)
233 fd_resource_flush(trans, box);
234 }
235
236 static void
237 fd_resource_transfer_unmap(struct pipe_context *pctx,
238 struct pipe_transfer *ptrans)
239 {
240 struct fd_context *ctx = fd_context(pctx);
241 struct fd_resource *rsc = fd_resource(ptrans->resource);
242 struct fd_transfer *trans = fd_transfer(ptrans);
243
244 if (trans->staging && !(ptrans->usage & PIPE_TRANSFER_FLUSH_EXPLICIT)) {
245 struct pipe_box box;
246 u_box_2d(0, 0, ptrans->box.width, ptrans->box.height, &box);
247 fd_resource_flush(trans, &box);
248 }
249
250 if (!(ptrans->usage & PIPE_TRANSFER_UNSYNCHRONIZED)) {
251 fd_bo_cpu_fini(rsc->bo);
252 if (rsc->stencil)
253 fd_bo_cpu_fini(rsc->stencil->bo);
254 }
255
256 util_range_add(&rsc->valid_buffer_range,
257 ptrans->box.x,
258 ptrans->box.x + ptrans->box.width);
259
260 pipe_resource_reference(&ptrans->resource, NULL);
261 util_slab_free(&ctx->transfer_pool, ptrans);
262
263 if (trans->staging)
264 free(trans->staging);
265 }
266
267 static void *
268 fd_resource_transfer_map(struct pipe_context *pctx,
269 struct pipe_resource *prsc,
270 unsigned level, unsigned usage,
271 const struct pipe_box *box,
272 struct pipe_transfer **pptrans)
273 {
274 struct fd_context *ctx = fd_context(pctx);
275 struct fd_resource *rsc = fd_resource(prsc);
276 struct fd_resource_slice *slice = fd_resource_slice(rsc, level);
277 struct fd_transfer *trans;
278 struct pipe_transfer *ptrans;
279 enum pipe_format format = prsc->format;
280 uint32_t op = 0;
281 uint32_t offset;
282 char *buf;
283 int ret = 0;
284
285 DBG("prsc=%p, level=%u, usage=%x, box=%dx%d+%d,%d", prsc, level, usage,
286 box->width, box->height, box->x, box->y);
287
288 ptrans = util_slab_alloc(&ctx->transfer_pool);
289 if (!ptrans)
290 return NULL;
291
292 /* util_slab_alloc() doesn't zero: */
293 trans = fd_transfer(ptrans);
294 memset(trans, 0, sizeof(*trans));
295
296 pipe_resource_reference(&ptrans->resource, prsc);
297 ptrans->level = level;
298 ptrans->usage = usage;
299 ptrans->box = *box;
300 ptrans->stride = util_format_get_nblocksx(format, slice->pitch) * rsc->cpp;
301 ptrans->layer_stride = slice->size0;
302
303 if (usage & PIPE_TRANSFER_READ)
304 op |= DRM_FREEDRENO_PREP_READ;
305
306 if (usage & PIPE_TRANSFER_WRITE)
307 op |= DRM_FREEDRENO_PREP_WRITE;
308
309 if (usage & PIPE_TRANSFER_DISCARD_WHOLE_RESOURCE) {
310 realloc_bo(rsc, fd_bo_size(rsc->bo));
311 if (rsc->stencil)
312 realloc_bo(rsc->stencil, fd_bo_size(rsc->stencil->bo));
313 fd_invalidate_resource(ctx, prsc);
314 } else if ((usage & PIPE_TRANSFER_WRITE) &&
315 prsc->target == PIPE_BUFFER &&
316 !util_ranges_intersect(&rsc->valid_buffer_range,
317 box->x, box->x + box->width)) {
318 /* We are trying to write to a previously uninitialized range. No need
319 * to wait.
320 */
321 } else if (!(usage & PIPE_TRANSFER_UNSYNCHRONIZED)) {
322 /* If the GPU is writing to the resource, or if it is reading from the
323 * resource and we're trying to write to it, flush the renders.
324 */
325 if (((ptrans->usage & PIPE_TRANSFER_WRITE) &&
326 pending(rsc, FD_PENDING_READ | FD_PENDING_WRITE)) ||
327 pending(rsc, FD_PENDING_WRITE))
328 fd_context_render(pctx);
329
330 /* The GPU keeps track of how the various bo's are being used, and
331 * will wait if necessary for the proper operation to have
332 * completed.
333 */
334 ret = fd_bo_cpu_prep(rsc->bo, ctx->screen->pipe, op);
335 if (ret)
336 goto fail;
337 }
338
339 buf = fd_bo_map(rsc->bo);
340 if (!buf) {
341 fd_resource_transfer_unmap(pctx, ptrans);
342 return NULL;
343 }
344
345 offset = slice->offset +
346 box->y / util_format_get_blockheight(format) * ptrans->stride +
347 box->x / util_format_get_blockwidth(format) * rsc->cpp +
348 fd_resource_layer_offset(rsc, slice, box->z);
349
350 if (prsc->format == PIPE_FORMAT_Z32_FLOAT_S8X24_UINT ||
351 prsc->format == PIPE_FORMAT_X32_S8X24_UINT) {
352 assert(trans->base.box.depth == 1);
353
354 trans->base.stride = trans->base.box.width * rsc->cpp * 2;
355 trans->staging = malloc(trans->base.stride * trans->base.box.height);
356 if (!trans->staging)
357 goto fail;
358
359 /* if we're not discarding the whole range (or resource), we must copy
360 * the real data in.
361 */
362 if (!(usage & (PIPE_TRANSFER_DISCARD_WHOLE_RESOURCE |
363 PIPE_TRANSFER_DISCARD_RANGE))) {
364 struct fd_resource_slice *sslice =
365 fd_resource_slice(rsc->stencil, level);
366 void *sbuf = fd_bo_map(rsc->stencil->bo);
367 if (!sbuf)
368 goto fail;
369
370 float *depth = (float *)(buf + slice->offset +
371 fd_resource_layer_offset(rsc, slice, box->z) +
372 box->y * slice->pitch * 4 + box->x * 4);
373 uint8_t *stencil = sbuf + sslice->offset +
374 fd_resource_layer_offset(rsc->stencil, sslice, box->z) +
375 box->y * sslice->pitch + box->x;
376
377 if (format != PIPE_FORMAT_X32_S8X24_UINT)
378 util_format_z32_float_s8x24_uint_pack_z_float(
379 trans->staging, trans->base.stride,
380 depth, slice->pitch * 4,
381 box->width, box->height);
382
383 util_format_z32_float_s8x24_uint_pack_s_8uint(
384 trans->staging, trans->base.stride,
385 stencil, sslice->pitch,
386 box->width, box->height);
387 }
388
389 buf = trans->staging;
390 offset = 0;
391 } else if (rsc->internal_format != format &&
392 util_format_description(format)->layout == UTIL_FORMAT_LAYOUT_RGTC) {
393 assert(trans->base.box.depth == 1);
394
395 trans->base.stride = util_format_get_stride(
396 format, trans->base.box.width);
397 trans->staging = malloc(
398 util_format_get_2d_size(format, trans->base.stride,
399 trans->base.box.height));
400 if (!trans->staging)
401 goto fail;
402
403 /* if we're not discarding the whole range (or resource), we must copy
404 * the real data in.
405 */
406 if (!(usage & (PIPE_TRANSFER_DISCARD_WHOLE_RESOURCE |
407 PIPE_TRANSFER_DISCARD_RANGE))) {
408 uint8_t *rgba8 = (uint8_t *)buf + slice->offset +
409 fd_resource_layer_offset(rsc, slice, box->z) +
410 box->y * slice->pitch * rsc->cpp + box->x * rsc->cpp;
411
412 switch (format) {
413 case PIPE_FORMAT_RGTC1_UNORM:
414 case PIPE_FORMAT_RGTC1_SNORM:
415 case PIPE_FORMAT_LATC1_UNORM:
416 case PIPE_FORMAT_LATC1_SNORM:
417 util_format_rgtc1_unorm_pack_rgba_8unorm(
418 trans->staging, trans->base.stride,
419 rgba8, slice->pitch * rsc->cpp,
420 box->width, box->height);
421 break;
422 case PIPE_FORMAT_RGTC2_UNORM:
423 case PIPE_FORMAT_RGTC2_SNORM:
424 case PIPE_FORMAT_LATC2_UNORM:
425 case PIPE_FORMAT_LATC2_SNORM:
426 util_format_rgtc2_unorm_pack_rgba_8unorm(
427 trans->staging, trans->base.stride,
428 rgba8, slice->pitch * rsc->cpp,
429 box->width, box->height);
430 break;
431 default:
432 assert(!"Unexpected format");
433 break;
434 }
435 }
436
437 buf = trans->staging;
438 offset = 0;
439 }
440
441 *pptrans = ptrans;
442
443 return buf + offset;
444
445 fail:
446 fd_resource_transfer_unmap(pctx, ptrans);
447 return NULL;
448 }
449
450 static void
451 fd_resource_destroy(struct pipe_screen *pscreen,
452 struct pipe_resource *prsc)
453 {
454 struct fd_resource *rsc = fd_resource(prsc);
455 if (rsc->bo)
456 fd_bo_del(rsc->bo);
457 list_delinit(&rsc->list);
458 util_range_destroy(&rsc->valid_buffer_range);
459 FREE(rsc);
460 }
461
462 static boolean
463 fd_resource_get_handle(struct pipe_screen *pscreen,
464 struct pipe_resource *prsc,
465 struct winsys_handle *handle)
466 {
467 struct fd_resource *rsc = fd_resource(prsc);
468
469 return fd_screen_bo_get_handle(pscreen, rsc->bo,
470 rsc->slices[0].pitch * rsc->cpp, handle);
471 }
472
473
474 static const struct u_resource_vtbl fd_resource_vtbl = {
475 .resource_get_handle = fd_resource_get_handle,
476 .resource_destroy = fd_resource_destroy,
477 .transfer_map = fd_resource_transfer_map,
478 .transfer_flush_region = fd_resource_transfer_flush_region,
479 .transfer_unmap = fd_resource_transfer_unmap,
480 .transfer_inline_write = u_default_transfer_inline_write,
481 };
482
483 static uint32_t
484 setup_slices(struct fd_resource *rsc, uint32_t alignment, enum pipe_format format)
485 {
486 struct pipe_resource *prsc = &rsc->base.b;
487 uint32_t level, size = 0;
488 uint32_t width = prsc->width0;
489 uint32_t height = prsc->height0;
490 uint32_t depth = prsc->depth0;
491 /* in layer_first layout, the level (slice) contains just one
492 * layer (since in fact the layer contains the slices)
493 */
494 uint32_t layers_in_level = rsc->layer_first ? 1 : prsc->array_size;
495
496 for (level = 0; level <= prsc->last_level; level++) {
497 struct fd_resource_slice *slice = fd_resource_slice(rsc, level);
498 uint32_t blocks;
499
500 slice->pitch = width = align(width, 32);
501 slice->offset = size;
502 blocks = util_format_get_nblocks(format, width, height);
503 /* 1d array and 2d array textures must all have the same layer size
504 * for each miplevel on a3xx. 3d textures can have different layer
505 * sizes for high levels, but the hw auto-sizer is buggy (or at least
506 * different than what this code does), so as soon as the layer size
507 * range gets into range, we stop reducing it.
508 */
509 if (prsc->target == PIPE_TEXTURE_3D && (
510 level == 1 ||
511 (level > 1 && rsc->slices[level - 1].size0 > 0xf000)))
512 slice->size0 = align(blocks * rsc->cpp, alignment);
513 else if (level == 0 || rsc->layer_first || alignment == 1)
514 slice->size0 = align(blocks * rsc->cpp, alignment);
515 else
516 slice->size0 = rsc->slices[level - 1].size0;
517
518 size += slice->size0 * depth * layers_in_level;
519
520 width = u_minify(width, 1);
521 height = u_minify(height, 1);
522 depth = u_minify(depth, 1);
523 }
524
525 return size;
526 }
527
528 static uint32_t
529 slice_alignment(struct pipe_screen *pscreen, const struct pipe_resource *tmpl)
530 {
531 /* on a3xx, 2d array and 3d textures seem to want their
532 * layers aligned to page boundaries:
533 */
534 switch (tmpl->target) {
535 case PIPE_TEXTURE_3D:
536 case PIPE_TEXTURE_1D_ARRAY:
537 case PIPE_TEXTURE_2D_ARRAY:
538 return 4096;
539 default:
540 return 1;
541 }
542 }
543
544 /**
545 * Create a new texture object, using the given template info.
546 */
547 static struct pipe_resource *
548 fd_resource_create(struct pipe_screen *pscreen,
549 const struct pipe_resource *tmpl)
550 {
551 struct fd_resource *rsc = CALLOC_STRUCT(fd_resource);
552 struct pipe_resource *prsc = &rsc->base.b;
553 enum pipe_format format = tmpl->format;
554 uint32_t size, alignment;
555
556 DBG("target=%d, format=%s, %ux%ux%u, array_size=%u, last_level=%u, "
557 "nr_samples=%u, usage=%u, bind=%x, flags=%x",
558 tmpl->target, util_format_name(format),
559 tmpl->width0, tmpl->height0, tmpl->depth0,
560 tmpl->array_size, tmpl->last_level, tmpl->nr_samples,
561 tmpl->usage, tmpl->bind, tmpl->flags);
562
563 if (!rsc)
564 return NULL;
565
566 *prsc = *tmpl;
567
568 pipe_reference_init(&prsc->reference, 1);
569 list_inithead(&rsc->list);
570 prsc->screen = pscreen;
571
572 util_range_init(&rsc->valid_buffer_range);
573
574 rsc->base.vtbl = &fd_resource_vtbl;
575
576 if (format == PIPE_FORMAT_Z32_FLOAT_S8X24_UINT)
577 format = PIPE_FORMAT_Z32_FLOAT;
578 else if (fd_screen(pscreen)->gpu_id < 400 &&
579 util_format_description(format)->layout == UTIL_FORMAT_LAYOUT_RGTC)
580 format = PIPE_FORMAT_R8G8B8A8_UNORM;
581 rsc->internal_format = format;
582 rsc->cpp = util_format_get_blocksize(format);
583
584 assert(rsc->cpp);
585
586 alignment = slice_alignment(pscreen, tmpl);
587 if (is_a4xx(fd_screen(pscreen))) {
588 switch (tmpl->target) {
589 case PIPE_TEXTURE_3D:
590 rsc->layer_first = false;
591 break;
592 default:
593 rsc->layer_first = true;
594 alignment = 1;
595 break;
596 }
597 }
598
599 size = setup_slices(rsc, alignment, format);
600
601 if (rsc->layer_first) {
602 rsc->layer_size = align(size, 4096);
603 size = rsc->layer_size * prsc->array_size;
604 }
605
606 realloc_bo(rsc, size);
607 if (!rsc->bo)
608 goto fail;
609
610 /* There is no native Z32F_S8 sampling or rendering format, so this must
611 * be emulated via two separate textures. The depth texture still keeps
612 * its Z32F_S8 format though, and we also keep a reference to a separate
613 * S8 texture.
614 */
615 if (tmpl->format == PIPE_FORMAT_Z32_FLOAT_S8X24_UINT) {
616 struct pipe_resource stencil = *tmpl;
617 stencil.format = PIPE_FORMAT_S8_UINT;
618 rsc->stencil = fd_resource(fd_resource_create(pscreen, &stencil));
619 if (!rsc->stencil)
620 goto fail;
621 }
622
623 return prsc;
624 fail:
625 fd_resource_destroy(pscreen, prsc);
626 return NULL;
627 }
628
629 /**
630 * Create a texture from a winsys_handle. The handle is often created in
631 * another process by first creating a pipe texture and then calling
632 * resource_get_handle.
633 */
634 static struct pipe_resource *
635 fd_resource_from_handle(struct pipe_screen *pscreen,
636 const struct pipe_resource *tmpl,
637 struct winsys_handle *handle)
638 {
639 struct fd_resource *rsc = CALLOC_STRUCT(fd_resource);
640 struct fd_resource_slice *slice = &rsc->slices[0];
641 struct pipe_resource *prsc = &rsc->base.b;
642
643 DBG("target=%d, format=%s, %ux%ux%u, array_size=%u, last_level=%u, "
644 "nr_samples=%u, usage=%u, bind=%x, flags=%x",
645 tmpl->target, util_format_name(tmpl->format),
646 tmpl->width0, tmpl->height0, tmpl->depth0,
647 tmpl->array_size, tmpl->last_level, tmpl->nr_samples,
648 tmpl->usage, tmpl->bind, tmpl->flags);
649
650 if (!rsc)
651 return NULL;
652
653 *prsc = *tmpl;
654
655 pipe_reference_init(&prsc->reference, 1);
656 list_inithead(&rsc->list);
657 prsc->screen = pscreen;
658
659 util_range_init(&rsc->valid_buffer_range);
660
661 rsc->bo = fd_screen_bo_from_handle(pscreen, handle, &slice->pitch);
662 if (!rsc->bo)
663 goto fail;
664
665 rsc->base.vtbl = &fd_resource_vtbl;
666 rsc->cpp = util_format_get_blocksize(tmpl->format);
667 slice->pitch /= rsc->cpp;
668
669 assert(rsc->cpp);
670
671 return prsc;
672
673 fail:
674 fd_resource_destroy(pscreen, prsc);
675 return NULL;
676 }
677
678 static void fd_blitter_pipe_begin(struct fd_context *ctx, bool render_cond);
679 static void fd_blitter_pipe_end(struct fd_context *ctx);
680
681 /**
682 * _copy_region using pipe (3d engine)
683 */
684 static bool
685 fd_blitter_pipe_copy_region(struct fd_context *ctx,
686 struct pipe_resource *dst,
687 unsigned dst_level,
688 unsigned dstx, unsigned dsty, unsigned dstz,
689 struct pipe_resource *src,
690 unsigned src_level,
691 const struct pipe_box *src_box)
692 {
693 /* not until we allow rendertargets to be buffers */
694 if (dst->target == PIPE_BUFFER || src->target == PIPE_BUFFER)
695 return false;
696
697 if (!util_blitter_is_copy_supported(ctx->blitter, dst, src))
698 return false;
699
700 fd_blitter_pipe_begin(ctx, false);
701 util_blitter_copy_texture(ctx->blitter,
702 dst, dst_level, dstx, dsty, dstz,
703 src, src_level, src_box);
704 fd_blitter_pipe_end(ctx);
705
706 return true;
707 }
708
709 /**
710 * Copy a block of pixels from one resource to another.
711 * The resource must be of the same format.
712 * Resources with nr_samples > 1 are not allowed.
713 */
714 static void
715 fd_resource_copy_region(struct pipe_context *pctx,
716 struct pipe_resource *dst,
717 unsigned dst_level,
718 unsigned dstx, unsigned dsty, unsigned dstz,
719 struct pipe_resource *src,
720 unsigned src_level,
721 const struct pipe_box *src_box)
722 {
723 struct fd_context *ctx = fd_context(pctx);
724
725 /* TODO if we have 2d core, or other DMA engine that could be used
726 * for simple copies and reasonably easily synchronized with the 3d
727 * core, this is where we'd plug it in..
728 */
729
730 /* try blit on 3d pipe: */
731 if (fd_blitter_pipe_copy_region(ctx,
732 dst, dst_level, dstx, dsty, dstz,
733 src, src_level, src_box))
734 return;
735
736 /* else fallback to pure sw: */
737 util_resource_copy_region(pctx,
738 dst, dst_level, dstx, dsty, dstz,
739 src, src_level, src_box);
740 }
741
742 bool
743 fd_render_condition_check(struct pipe_context *pctx)
744 {
745 struct fd_context *ctx = fd_context(pctx);
746
747 if (!ctx->cond_query)
748 return true;
749
750 union pipe_query_result res = { 0 };
751 bool wait =
752 ctx->cond_mode != PIPE_RENDER_COND_NO_WAIT &&
753 ctx->cond_mode != PIPE_RENDER_COND_BY_REGION_NO_WAIT;
754
755 if (pctx->get_query_result(pctx, ctx->cond_query, wait, &res))
756 return (bool)res.u64 != ctx->cond_cond;
757
758 return true;
759 }
760
761 /**
762 * Optimal hardware path for blitting pixels.
763 * Scaling, format conversion, up- and downsampling (resolve) are allowed.
764 */
765 static void
766 fd_blit(struct pipe_context *pctx, const struct pipe_blit_info *blit_info)
767 {
768 struct fd_context *ctx = fd_context(pctx);
769 struct pipe_blit_info info = *blit_info;
770
771 if (info.src.resource->nr_samples > 1 &&
772 info.dst.resource->nr_samples <= 1 &&
773 !util_format_is_depth_or_stencil(info.src.resource->format) &&
774 !util_format_is_pure_integer(info.src.resource->format)) {
775 DBG("color resolve unimplemented");
776 return;
777 }
778
779 if (info.render_condition_enable && !fd_render_condition_check(pctx))
780 return;
781
782 if (util_try_blit_via_copy_region(pctx, &info)) {
783 return; /* done */
784 }
785
786 if (info.mask & PIPE_MASK_S) {
787 DBG("cannot blit stencil, skipping");
788 info.mask &= ~PIPE_MASK_S;
789 }
790
791 if (!util_blitter_is_blit_supported(ctx->blitter, &info)) {
792 DBG("blit unsupported %s -> %s",
793 util_format_short_name(info.src.resource->format),
794 util_format_short_name(info.dst.resource->format));
795 return;
796 }
797
798 fd_blitter_pipe_begin(ctx, info.render_condition_enable);
799 util_blitter_blit(ctx->blitter, &info);
800 fd_blitter_pipe_end(ctx);
801 }
802
803 static void
804 fd_blitter_pipe_begin(struct fd_context *ctx, bool render_cond)
805 {
806 util_blitter_save_vertex_buffer_slot(ctx->blitter, ctx->vtx.vertexbuf.vb);
807 util_blitter_save_vertex_elements(ctx->blitter, ctx->vtx.vtx);
808 util_blitter_save_vertex_shader(ctx->blitter, ctx->prog.vp);
809 util_blitter_save_so_targets(ctx->blitter, ctx->streamout.num_targets,
810 ctx->streamout.targets);
811 util_blitter_save_rasterizer(ctx->blitter, ctx->rasterizer);
812 util_blitter_save_viewport(ctx->blitter, &ctx->viewport);
813 util_blitter_save_scissor(ctx->blitter, &ctx->scissor);
814 util_blitter_save_fragment_shader(ctx->blitter, ctx->prog.fp);
815 util_blitter_save_blend(ctx->blitter, ctx->blend);
816 util_blitter_save_depth_stencil_alpha(ctx->blitter, ctx->zsa);
817 util_blitter_save_stencil_ref(ctx->blitter, &ctx->stencil_ref);
818 util_blitter_save_sample_mask(ctx->blitter, ctx->sample_mask);
819 util_blitter_save_framebuffer(ctx->blitter, &ctx->framebuffer);
820 util_blitter_save_fragment_sampler_states(ctx->blitter,
821 ctx->fragtex.num_samplers,
822 (void **)ctx->fragtex.samplers);
823 util_blitter_save_fragment_sampler_views(ctx->blitter,
824 ctx->fragtex.num_textures, ctx->fragtex.textures);
825 if (!render_cond)
826 util_blitter_save_render_condition(ctx->blitter,
827 ctx->cond_query, ctx->cond_cond, ctx->cond_mode);
828
829 fd_hw_query_set_stage(ctx, ctx->ring, FD_STAGE_BLIT);
830 }
831
832 static void
833 fd_blitter_pipe_end(struct fd_context *ctx)
834 {
835 fd_hw_query_set_stage(ctx, ctx->ring, FD_STAGE_NULL);
836 }
837
838 static void
839 fd_flush_resource(struct pipe_context *pctx, struct pipe_resource *prsc)
840 {
841 struct fd_resource *rsc = fd_resource(prsc);
842
843 if (pending(rsc, FD_PENDING_WRITE | FD_PENDING_READ))
844 fd_context_render(pctx);
845 }
846
847 void
848 fd_resource_screen_init(struct pipe_screen *pscreen)
849 {
850 pscreen->resource_create = fd_resource_create;
851 pscreen->resource_from_handle = fd_resource_from_handle;
852 pscreen->resource_get_handle = u_resource_get_handle_vtbl;
853 pscreen->resource_destroy = u_resource_destroy_vtbl;
854 }
855
856 void
857 fd_resource_context_init(struct pipe_context *pctx)
858 {
859 pctx->transfer_map = u_transfer_map_vtbl;
860 pctx->transfer_flush_region = u_transfer_flush_region_vtbl;
861 pctx->transfer_unmap = u_transfer_unmap_vtbl;
862 pctx->transfer_inline_write = u_transfer_inline_write_vtbl;
863 pctx->create_surface = fd_create_surface;
864 pctx->surface_destroy = fd_surface_destroy;
865 pctx->resource_copy_region = fd_resource_copy_region;
866 pctx->blit = fd_blit;
867 pctx->flush_resource = fd_flush_resource;
868 }