projects / mesa.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
history | raw | HEAD
freedreno: honor handle->offset
[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 /* XXX this should go away, needed for 'struct winsys_handle' */
47 #include "state_tracker/drm_driver.h"
48
49 static bool
50 pending(struct fd_resource *rsc, enum fd_resource_status status)
51 {
52 return (rsc->status & status) ||
53 (rsc->stencil && (rsc->stencil->status & status));
54 }
55
56 static void
57 fd_invalidate_resource(struct fd_context *ctx, struct pipe_resource *prsc)
58 {
59 int i;
60
61 /* Go through the entire state and see if the resource is bound
62 * anywhere. If it is, mark the relevant state as dirty. This is called on
63 * realloc_bo.
64 */
65
66 /* Constbufs */
67 for (i = 1; i < PIPE_MAX_CONSTANT_BUFFERS && !(ctx->dirty & FD_DIRTY_CONSTBUF); i++) {
68 if (ctx->constbuf[PIPE_SHADER_VERTEX].cb[i].buffer == prsc)
69 ctx->dirty |= FD_DIRTY_CONSTBUF;
70 if (ctx->constbuf[PIPE_SHADER_FRAGMENT].cb[i].buffer == prsc)
71 ctx->dirty |= FD_DIRTY_CONSTBUF;
72 }
73
74 /* VBOs */
75 for (i = 0; i < ctx->vtx.vertexbuf.count && !(ctx->dirty & FD_DIRTY_VTXBUF); i++) {
76 if (ctx->vtx.vertexbuf.vb[i].buffer == prsc)
77 ctx->dirty |= FD_DIRTY_VTXBUF;
78 }
79
80 /* Index buffer */
81 if (ctx->indexbuf.buffer == prsc)
82 ctx->dirty |= FD_DIRTY_INDEXBUF;
83
84 /* Textures */
85 for (i = 0; i < ctx->verttex.num_textures && !(ctx->dirty & FD_DIRTY_VERTTEX); i++) {
86 if (ctx->verttex.textures[i] && (ctx->verttex.textures[i]->texture == prsc))
87 ctx->dirty |= FD_DIRTY_VERTTEX;
88 }
89 for (i = 0; i < ctx->fragtex.num_textures && !(ctx->dirty & FD_DIRTY_FRAGTEX); i++) {
90 if (ctx->fragtex.textures[i] && (ctx->fragtex.textures[i]->texture == prsc))
91 ctx->dirty |= FD_DIRTY_FRAGTEX;
92 }
93 }
94
95 static void
96 realloc_bo(struct fd_resource *rsc, uint32_t size)
97 {
98 struct fd_screen *screen = fd_screen(rsc->base.b.screen);
99 uint32_t flags = DRM_FREEDRENO_GEM_CACHE_WCOMBINE |
100 DRM_FREEDRENO_GEM_TYPE_KMEM; /* TODO */
101
102 /* if we start using things other than write-combine,
103 * be sure to check for PIPE_RESOURCE_FLAG_MAP_COHERENT
104 */
105
106 if (rsc->bo)
107 fd_bo_del(rsc->bo);
108
109 rsc->bo = fd_bo_new(screen->dev, size, flags);
110 rsc->timestamp = 0;
111 rsc->status = 0;
112 rsc->pending_ctx = NULL;
113 list_delinit(&rsc->list);
114 util_range_set_empty(&rsc->valid_buffer_range);
115 }
116
117 static unsigned
118 fd_resource_layer_offset(struct fd_resource *rsc,
119 struct fd_resource_slice *slice,
120 unsigned layer)
121 {
122 if (rsc->layer_first)
123 return layer * rsc->layer_size;
124 else
125 return layer * slice->size0;
126 }
127
128 static void
129 fd_resource_flush_z32s8(struct fd_transfer *trans, const struct pipe_box *box)
130 {
131 struct fd_resource *rsc = fd_resource(trans->base.resource);
132 struct fd_resource_slice *slice = fd_resource_slice(rsc, trans->base.level);
133 struct fd_resource_slice *sslice = fd_resource_slice(rsc->stencil, trans->base.level);
134 enum pipe_format format = trans->base.resource->format;
135
136 float *depth = fd_bo_map(rsc->bo) + slice->offset +
137 fd_resource_layer_offset(rsc, slice, trans->base.box.z) +
138 (trans->base.box.y + box->y) * slice->pitch * 4 + (trans->base.box.x + box->x) * 4;
139 uint8_t *stencil = fd_bo_map(rsc->stencil->bo) + sslice->offset +
140 fd_resource_layer_offset(rsc->stencil, sslice, trans->base.box.z) +
141 (trans->base.box.y + box->y) * sslice->pitch + trans->base.box.x + box->x;
142
143 if (format != PIPE_FORMAT_X32_S8X24_UINT)
144 util_format_z32_float_s8x24_uint_unpack_z_float(
145 depth, slice->pitch * 4,
146 trans->staging, trans->base.stride,
147 box->width, box->height);
148
149 util_format_z32_float_s8x24_uint_unpack_s_8uint(
150 stencil, sslice->pitch,
151 trans->staging, trans->base.stride,
152 box->width, box->height);
153 }
154
155 static void
156 fd_resource_flush_rgtc(struct fd_transfer *trans, const struct pipe_box *box)
157 {
158 struct fd_resource *rsc = fd_resource(trans->base.resource);
159 struct fd_resource_slice *slice = fd_resource_slice(rsc, trans->base.level);
160 enum pipe_format format = trans->base.resource->format;
161
162 uint8_t *data = fd_bo_map(rsc->bo) + slice->offset +
163 fd_resource_layer_offset(rsc, slice, trans->base.box.z) +
164 ((trans->base.box.y + box->y) * slice->pitch +
165 trans->base.box.x + box->x) * rsc->cpp;
166
167 uint8_t *source = trans->staging +
168 util_format_get_nblocksy(format, box->y) * trans->base.stride +
169 util_format_get_stride(format, box->x);
170
171 switch (format) {
172 case PIPE_FORMAT_RGTC1_UNORM:
173 case PIPE_FORMAT_RGTC1_SNORM:
174 case PIPE_FORMAT_LATC1_UNORM:
175 case PIPE_FORMAT_LATC1_SNORM:
176 util_format_rgtc1_unorm_unpack_rgba_8unorm(
177 data, slice->pitch * rsc->cpp,
178 source, trans->base.stride,
179 box->width, box->height);
180 break;
181 case PIPE_FORMAT_RGTC2_UNORM:
182 case PIPE_FORMAT_RGTC2_SNORM:
183 case PIPE_FORMAT_LATC2_UNORM:
184 case PIPE_FORMAT_LATC2_SNORM:
185 util_format_rgtc2_unorm_unpack_rgba_8unorm(
186 data, slice->pitch * rsc->cpp,
187 source, trans->base.stride,
188 box->width, box->height);
189 break;
190 default:
191 assert(!"Unexpected format\n");
192 break;
193 }
194 }
195
196 static void
197 fd_resource_flush(struct fd_transfer *trans, const struct pipe_box *box)
198 {
199 enum pipe_format format = trans->base.resource->format;
200
201 switch (format) {
202 case PIPE_FORMAT_Z32_FLOAT_S8X24_UINT:
203 case PIPE_FORMAT_X32_S8X24_UINT:
204 fd_resource_flush_z32s8(trans, box);
205 break;
206 case PIPE_FORMAT_RGTC1_UNORM:
207 case PIPE_FORMAT_RGTC1_SNORM:
208 case PIPE_FORMAT_RGTC2_UNORM:
209 case PIPE_FORMAT_RGTC2_SNORM:
210 case PIPE_FORMAT_LATC1_UNORM:
211 case PIPE_FORMAT_LATC1_SNORM:
212 case PIPE_FORMAT_LATC2_UNORM:
213 case PIPE_FORMAT_LATC2_SNORM:
214 fd_resource_flush_rgtc(trans, box);
215 break;
216 default:
217 assert(!"Unexpected staging transfer type");
218 break;
219 }
220 }
221
222 static void fd_resource_transfer_flush_region(struct pipe_context *pctx,
223 struct pipe_transfer *ptrans,
224 const struct pipe_box *box)
225 {
226 struct fd_resource *rsc = fd_resource(ptrans->resource);
227 struct fd_transfer *trans = fd_transfer(ptrans);
228
229 if (ptrans->resource->target == PIPE_BUFFER)
230 util_range_add(&rsc->valid_buffer_range,
231 ptrans->box.x + box->x,
232 ptrans->box.x + box->x + box->width);
233
234 if (trans->staging)
235 fd_resource_flush(trans, box);
236 }
237
238 static void
239 fd_resource_transfer_unmap(struct pipe_context *pctx,
240 struct pipe_transfer *ptrans)
241 {
242 struct fd_context *ctx = fd_context(pctx);
243 struct fd_resource *rsc = fd_resource(ptrans->resource);
244 struct fd_transfer *trans = fd_transfer(ptrans);
245
246 if (trans->staging && !(ptrans->usage & PIPE_TRANSFER_FLUSH_EXPLICIT)) {
247 struct pipe_box box;
248 u_box_2d(0, 0, ptrans->box.width, ptrans->box.height, &box);
249 fd_resource_flush(trans, &box);
250 }
251
252 if (!(ptrans->usage & PIPE_TRANSFER_UNSYNCHRONIZED)) {
253 fd_bo_cpu_fini(rsc->bo);
254 if (rsc->stencil)
255 fd_bo_cpu_fini(rsc->stencil->bo);
256 }
257
258 util_range_add(&rsc->valid_buffer_range,
259 ptrans->box.x,
260 ptrans->box.x + ptrans->box.width);
261
262 pipe_resource_reference(&ptrans->resource, NULL);
263 util_slab_free(&ctx->transfer_pool, ptrans);
264
265 if (trans->staging)
266 free(trans->staging);
267 }
268
269 static void *
270 fd_resource_transfer_map(struct pipe_context *pctx,
271 struct pipe_resource *prsc,
272 unsigned level, unsigned usage,
273 const struct pipe_box *box,
274 struct pipe_transfer **pptrans)
275 {
276 struct fd_context *ctx = fd_context(pctx);
277 struct fd_resource *rsc = fd_resource(prsc);
278 struct fd_resource_slice *slice = fd_resource_slice(rsc, level);
279 struct fd_transfer *trans;
280 struct pipe_transfer *ptrans;
281 enum pipe_format format = prsc->format;
282 uint32_t op = 0;
283 uint32_t offset;
284 char *buf;
285 int ret = 0;
286
287 DBG("prsc=%p, level=%u, usage=%x, box=%dx%d+%d,%d", prsc, level, usage,
288 box->width, box->height, box->x, box->y);
289
290 ptrans = util_slab_alloc(&ctx->transfer_pool);
291 if (!ptrans)
292 return NULL;
293
294 /* util_slab_alloc() doesn't zero: */
295 trans = fd_transfer(ptrans);
296 memset(trans, 0, sizeof(*trans));
297
298 pipe_resource_reference(&ptrans->resource, prsc);
299 ptrans->level = level;
300 ptrans->usage = usage;
301 ptrans->box = *box;
302 ptrans->stride = util_format_get_nblocksx(format, slice->pitch) * rsc->cpp;
303 ptrans->layer_stride = slice->size0;
304
305 if (usage & PIPE_TRANSFER_READ)
306 op |= DRM_FREEDRENO_PREP_READ;
307
308 if (usage & PIPE_TRANSFER_WRITE)
309 op |= DRM_FREEDRENO_PREP_WRITE;
310
311 if (usage & PIPE_TRANSFER_DISCARD_WHOLE_RESOURCE) {
312 realloc_bo(rsc, fd_bo_size(rsc->bo));
313 if (rsc->stencil)
314 realloc_bo(rsc->stencil, fd_bo_size(rsc->stencil->bo));
315 fd_invalidate_resource(ctx, prsc);
316 } else if ((usage & PIPE_TRANSFER_WRITE) &&
317 prsc->target == PIPE_BUFFER &&
318 !util_ranges_intersect(&rsc->valid_buffer_range,
319 box->x, box->x + box->width)) {
320 /* We are trying to write to a previously uninitialized range. No need
321 * to wait.
322 */
323 } else if (!(usage & PIPE_TRANSFER_UNSYNCHRONIZED)) {
324 /* If the GPU is writing to the resource, or if it is reading from the
325 * resource and we're trying to write to it, flush the renders.
326 */
327 if (((ptrans->usage & PIPE_TRANSFER_WRITE) &&
328 pending(rsc, FD_PENDING_READ | FD_PENDING_WRITE)) ||
329 pending(rsc, FD_PENDING_WRITE))
330 fd_context_render(pctx);
331
332 /* The GPU keeps track of how the various bo's are being used, and
333 * will wait if necessary for the proper operation to have
334 * completed.
335 */
336 ret = fd_bo_cpu_prep(rsc->bo, ctx->screen->pipe, op);
337 if (ret)
338 goto fail;
339 }
340
341 buf = fd_bo_map(rsc->bo);
342 if (!buf)
343 goto fail;
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 enum util_format_layout layout = util_format_description(format)->layout;
488 uint32_t level, size = 0;
489 uint32_t width = prsc->width0;
490 uint32_t height = prsc->height0;
491 uint32_t depth = prsc->depth0;
492 /* in layer_first layout, the level (slice) contains just one
493 * layer (since in fact the layer contains the slices)
494 */
495 uint32_t layers_in_level = rsc->layer_first ? 1 : prsc->array_size;
496
497 for (level = 0; level <= prsc->last_level; level++) {
498 struct fd_resource_slice *slice = fd_resource_slice(rsc, level);
499 uint32_t blocks;
500
501 if (layout == UTIL_FORMAT_LAYOUT_ASTC)
502 slice->pitch = width =
503 util_align_npot(width, 32 * util_format_get_blockwidth(format));
504 else
505 slice->pitch = width = align(width, 32);
506 slice->offset = size;
507 blocks = util_format_get_nblocks(format, width, height);
508 /* 1d array and 2d array textures must all have the same layer size
509 * for each miplevel on a3xx. 3d textures can have different layer
510 * sizes for high levels, but the hw auto-sizer is buggy (or at least
511 * different than what this code does), so as soon as the layer size
512 * range gets into range, we stop reducing it.
513 */
514 if (prsc->target == PIPE_TEXTURE_3D && (
515 level == 1 ||
516 (level > 1 && rsc->slices[level - 1].size0 > 0xf000)))
517 slice->size0 = align(blocks * rsc->cpp, alignment);
518 else if (level == 0 || rsc->layer_first || alignment == 1)
519 slice->size0 = align(blocks * rsc->cpp, alignment);
520 else
521 slice->size0 = rsc->slices[level - 1].size0;
522
523 size += slice->size0 * depth * layers_in_level;
524
525 width = u_minify(width, 1);
526 height = u_minify(height, 1);
527 depth = u_minify(depth, 1);
528 }
529
530 return size;
531 }
532
533 static uint32_t
534 slice_alignment(struct pipe_screen *pscreen, const struct pipe_resource *tmpl)
535 {
536 /* on a3xx, 2d array and 3d textures seem to want their
537 * layers aligned to page boundaries:
538 */
539 switch (tmpl->target) {
540 case PIPE_TEXTURE_3D:
541 case PIPE_TEXTURE_1D_ARRAY:
542 case PIPE_TEXTURE_2D_ARRAY:
543 return 4096;
544 default:
545 return 1;
546 }
547 }
548
549 /**
550 * Create a new texture object, using the given template info.
551 */
552 static struct pipe_resource *
553 fd_resource_create(struct pipe_screen *pscreen,
554 const struct pipe_resource *tmpl)
555 {
556 struct fd_resource *rsc = CALLOC_STRUCT(fd_resource);
557 struct pipe_resource *prsc = &rsc->base.b;
558 enum pipe_format format = tmpl->format;
559 uint32_t size, alignment;
560
561 DBG("target=%d, format=%s, %ux%ux%u, array_size=%u, last_level=%u, "
562 "nr_samples=%u, usage=%u, bind=%x, flags=%x",
563 tmpl->target, util_format_name(format),
564 tmpl->width0, tmpl->height0, tmpl->depth0,
565 tmpl->array_size, tmpl->last_level, tmpl->nr_samples,
566 tmpl->usage, tmpl->bind, tmpl->flags);
567
568 if (!rsc)
569 return NULL;
570
571 *prsc = *tmpl;
572
573 pipe_reference_init(&prsc->reference, 1);
574 list_inithead(&rsc->list);
575 prsc->screen = pscreen;
576
577 util_range_init(&rsc->valid_buffer_range);
578
579 rsc->base.vtbl = &fd_resource_vtbl;
580
581 if (format == PIPE_FORMAT_Z32_FLOAT_S8X24_UINT)
582 format = PIPE_FORMAT_Z32_FLOAT;
583 else if (fd_screen(pscreen)->gpu_id < 400 &&
584 util_format_description(format)->layout == UTIL_FORMAT_LAYOUT_RGTC)
585 format = PIPE_FORMAT_R8G8B8A8_UNORM;
586 rsc->internal_format = format;
587 rsc->cpp = util_format_get_blocksize(format);
588
589 assert(rsc->cpp);
590
591 alignment = slice_alignment(pscreen, tmpl);
592 if (is_a4xx(fd_screen(pscreen))) {
593 switch (tmpl->target) {
594 case PIPE_TEXTURE_3D:
595 rsc->layer_first = false;
596 break;
597 default:
598 rsc->layer_first = true;
599 alignment = 1;
600 break;
601 }
602 }
603
604 size = setup_slices(rsc, alignment, format);
605
606 if (rsc->layer_first) {
607 rsc->layer_size = align(size, 4096);
608 size = rsc->layer_size * prsc->array_size;
609 }
610
611 realloc_bo(rsc, size);
612 if (!rsc->bo)
613 goto fail;
614
615 /* There is no native Z32F_S8 sampling or rendering format, so this must
616 * be emulated via two separate textures. The depth texture still keeps
617 * its Z32F_S8 format though, and we also keep a reference to a separate
618 * S8 texture.
619 */
620 if (tmpl->format == PIPE_FORMAT_Z32_FLOAT_S8X24_UINT) {
621 struct pipe_resource stencil = *tmpl;
622 stencil.format = PIPE_FORMAT_S8_UINT;
623 rsc->stencil = fd_resource(fd_resource_create(pscreen, &stencil));
624 if (!rsc->stencil)
625 goto fail;
626 }
627
628 return prsc;
629 fail:
630 fd_resource_destroy(pscreen, prsc);
631 return NULL;
632 }
633
634 /**
635 * Create a texture from a winsys_handle. The handle is often created in
636 * another process by first creating a pipe texture and then calling
637 * resource_get_handle.
638 */
639 static struct pipe_resource *
640 fd_resource_from_handle(struct pipe_screen *pscreen,
641 const struct pipe_resource *tmpl,
642 struct winsys_handle *handle, unsigned usage)
643 {
644 struct fd_resource *rsc = CALLOC_STRUCT(fd_resource);
645 struct fd_resource_slice *slice = &rsc->slices[0];
646 struct pipe_resource *prsc = &rsc->base.b;
647
648 DBG("target=%d, format=%s, %ux%ux%u, array_size=%u, last_level=%u, "
649 "nr_samples=%u, usage=%u, bind=%x, flags=%x",
650 tmpl->target, util_format_name(tmpl->format),
651 tmpl->width0, tmpl->height0, tmpl->depth0,
652 tmpl->array_size, tmpl->last_level, tmpl->nr_samples,
653 tmpl->usage, tmpl->bind, tmpl->flags);
654
655 if (!rsc)
656 return NULL;
657
658 *prsc = *tmpl;
659
660 pipe_reference_init(&prsc->reference, 1);
661 list_inithead(&rsc->list);
662 prsc->screen = pscreen;
663
664 util_range_init(&rsc->valid_buffer_range);
665
666 rsc->bo = fd_screen_bo_from_handle(pscreen, handle, &slice->pitch);
667 if (!rsc->bo)
668 goto fail;
669
670 rsc->base.vtbl = &fd_resource_vtbl;
671 rsc->cpp = util_format_get_blocksize(tmpl->format);
672 slice->pitch /= rsc->cpp;
673 slice->offset = handle->offset;
674
675 assert(rsc->cpp);
676
677 return prsc;
678
679 fail:
680 fd_resource_destroy(pscreen, prsc);
681 return NULL;
682 }
683
684 static void fd_blitter_pipe_begin(struct fd_context *ctx, bool render_cond);
685 static void fd_blitter_pipe_end(struct fd_context *ctx);
686
687 /**
688 * _copy_region using pipe (3d engine)
689 */
690 static bool
691 fd_blitter_pipe_copy_region(struct fd_context *ctx,
692 struct pipe_resource *dst,
693 unsigned dst_level,
694 unsigned dstx, unsigned dsty, unsigned dstz,
695 struct pipe_resource *src,
696 unsigned src_level,
697 const struct pipe_box *src_box)
698 {
699 /* not until we allow rendertargets to be buffers */
700 if (dst->target == PIPE_BUFFER || src->target == PIPE_BUFFER)
701 return false;
702
703 if (!util_blitter_is_copy_supported(ctx->blitter, dst, src))
704 return false;
705
706 fd_blitter_pipe_begin(ctx, false);
707 util_blitter_copy_texture(ctx->blitter,
708 dst, dst_level, dstx, dsty, dstz,
709 src, src_level, src_box);
710 fd_blitter_pipe_end(ctx);
711
712 return true;
713 }
714
715 /**
716 * Copy a block of pixels from one resource to another.
717 * The resource must be of the same format.
718 * Resources with nr_samples > 1 are not allowed.
719 */
720 static void
721 fd_resource_copy_region(struct pipe_context *pctx,
722 struct pipe_resource *dst,
723 unsigned dst_level,
724 unsigned dstx, unsigned dsty, unsigned dstz,
725 struct pipe_resource *src,
726 unsigned src_level,
727 const struct pipe_box *src_box)
728 {
729 struct fd_context *ctx = fd_context(pctx);
730
731 /* TODO if we have 2d core, or other DMA engine that could be used
732 * for simple copies and reasonably easily synchronized with the 3d
733 * core, this is where we'd plug it in..
734 */
735
736 /* try blit on 3d pipe: */
737 if (fd_blitter_pipe_copy_region(ctx,
738 dst, dst_level, dstx, dsty, dstz,
739 src, src_level, src_box))
740 return;
741
742 /* else fallback to pure sw: */
743 util_resource_copy_region(pctx,
744 dst, dst_level, dstx, dsty, dstz,
745 src, src_level, src_box);
746 }
747
748 bool
749 fd_render_condition_check(struct pipe_context *pctx)
750 {
751 struct fd_context *ctx = fd_context(pctx);
752
753 if (!ctx->cond_query)
754 return true;
755
756 union pipe_query_result res = { 0 };
757 bool wait =
758 ctx->cond_mode != PIPE_RENDER_COND_NO_WAIT &&
759 ctx->cond_mode != PIPE_RENDER_COND_BY_REGION_NO_WAIT;
760
761 if (pctx->get_query_result(pctx, ctx->cond_query, wait, &res))
762 return (bool)res.u64 != ctx->cond_cond;
763
764 return true;
765 }
766
767 /**
768 * Optimal hardware path for blitting pixels.
769 * Scaling, format conversion, up- and downsampling (resolve) are allowed.
770 */
771 static void
772 fd_blit(struct pipe_context *pctx, const struct pipe_blit_info *blit_info)
773 {
774 struct fd_context *ctx = fd_context(pctx);
775 struct pipe_blit_info info = *blit_info;
776
777 if (info.src.resource->nr_samples > 1 &&
778 info.dst.resource->nr_samples <= 1 &&
779 !util_format_is_depth_or_stencil(info.src.resource->format) &&
780 !util_format_is_pure_integer(info.src.resource->format)) {
781 DBG("color resolve unimplemented");
782 return;
783 }
784
785 if (info.render_condition_enable && !fd_render_condition_check(pctx))
786 return;
787
788 if (util_try_blit_via_copy_region(pctx, &info)) {
789 return; /* done */
790 }
791
792 if (info.mask & PIPE_MASK_S) {
793 DBG("cannot blit stencil, skipping");
794 info.mask &= ~PIPE_MASK_S;
795 }
796
797 if (!util_blitter_is_blit_supported(ctx->blitter, &info)) {
798 DBG("blit unsupported %s -> %s",
799 util_format_short_name(info.src.resource->format),
800 util_format_short_name(info.dst.resource->format));
801 return;
802 }
803
804 fd_blitter_pipe_begin(ctx, info.render_condition_enable);
805 util_blitter_blit(ctx->blitter, &info);
806 fd_blitter_pipe_end(ctx);
807 }
808
809 static void
810 fd_blitter_pipe_begin(struct fd_context *ctx, bool render_cond)
811 {
812 util_blitter_save_vertex_buffer_slot(ctx->blitter, ctx->vtx.vertexbuf.vb);
813 util_blitter_save_vertex_elements(ctx->blitter, ctx->vtx.vtx);
814 util_blitter_save_vertex_shader(ctx->blitter, ctx->prog.vp);
815 util_blitter_save_so_targets(ctx->blitter, ctx->streamout.num_targets,
816 ctx->streamout.targets);
817 util_blitter_save_rasterizer(ctx->blitter, ctx->rasterizer);
818 util_blitter_save_viewport(ctx->blitter, &ctx->viewport);
819 util_blitter_save_scissor(ctx->blitter, &ctx->scissor);
820 util_blitter_save_fragment_shader(ctx->blitter, ctx->prog.fp);
821 util_blitter_save_blend(ctx->blitter, ctx->blend);
822 util_blitter_save_depth_stencil_alpha(ctx->blitter, ctx->zsa);
823 util_blitter_save_stencil_ref(ctx->blitter, &ctx->stencil_ref);
824 util_blitter_save_sample_mask(ctx->blitter, ctx->sample_mask);
825 util_blitter_save_framebuffer(ctx->blitter, &ctx->framebuffer);
826 util_blitter_save_fragment_sampler_states(ctx->blitter,
827 ctx->fragtex.num_samplers,
828 (void **)ctx->fragtex.samplers);
829 util_blitter_save_fragment_sampler_views(ctx->blitter,
830 ctx->fragtex.num_textures, ctx->fragtex.textures);
831 if (!render_cond)
832 util_blitter_save_render_condition(ctx->blitter,
833 ctx->cond_query, ctx->cond_cond, ctx->cond_mode);
834
835 fd_hw_query_set_stage(ctx, ctx->ring, FD_STAGE_BLIT);
836 }
837
838 static void
839 fd_blitter_pipe_end(struct fd_context *ctx)
840 {
841 fd_hw_query_set_stage(ctx, ctx->ring, FD_STAGE_NULL);
842 }
843
844 static void
845 fd_flush_resource(struct pipe_context *pctx, struct pipe_resource *prsc)
846 {
847 struct fd_resource *rsc = fd_resource(prsc);
848
849 if (pending(rsc, FD_PENDING_WRITE | FD_PENDING_READ))
850 fd_context_render(pctx);
851 }
852
853 void
854 fd_resource_screen_init(struct pipe_screen *pscreen)
855 {
856 pscreen->resource_create = fd_resource_create;
857 pscreen->resource_from_handle = fd_resource_from_handle;
858 pscreen->resource_get_handle = u_resource_get_handle_vtbl;
859 pscreen->resource_destroy = u_resource_destroy_vtbl;
860 }
861
862 void
863 fd_resource_context_init(struct pipe_context *pctx)
864 {
865 pctx->transfer_map = u_transfer_map_vtbl;
866 pctx->transfer_flush_region = u_transfer_flush_region_vtbl;
867 pctx->transfer_unmap = u_transfer_unmap_vtbl;
868 pctx->transfer_inline_write = u_transfer_inline_write_vtbl;
869 pctx->create_surface = fd_create_surface;
870 pctx->surface_destroy = fd_surface_destroy;
871 pctx->resource_copy_region = fd_resource_copy_region;
872 pctx->blit = fd_blit;
873 pctx->flush_resource = fd_flush_resource;
874 }