projects / mesa.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
history | raw | HEAD
freedreno: per-generation OUT_IB packet
[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 goto fail;
342
343 offset = slice->offset +
344 box->y / util_format_get_blockheight(format) * ptrans->stride +
345 box->x / util_format_get_blockwidth(format) * rsc->cpp +
346 fd_resource_layer_offset(rsc, slice, box->z);
347
348 if (prsc->format == PIPE_FORMAT_Z32_FLOAT_S8X24_UINT ||
349 prsc->format == PIPE_FORMAT_X32_S8X24_UINT) {
350 assert(trans->base.box.depth == 1);
351
352 trans->base.stride = trans->base.box.width * rsc->cpp * 2;
353 trans->staging = malloc(trans->base.stride * trans->base.box.height);
354 if (!trans->staging)
355 goto fail;
356
357 /* if we're not discarding the whole range (or resource), we must copy
358 * the real data in.
359 */
360 if (!(usage & (PIPE_TRANSFER_DISCARD_WHOLE_RESOURCE |
361 PIPE_TRANSFER_DISCARD_RANGE))) {
362 struct fd_resource_slice *sslice =
363 fd_resource_slice(rsc->stencil, level);
364 void *sbuf = fd_bo_map(rsc->stencil->bo);
365 if (!sbuf)
366 goto fail;
367
368 float *depth = (float *)(buf + slice->offset +
369 fd_resource_layer_offset(rsc, slice, box->z) +
370 box->y * slice->pitch * 4 + box->x * 4);
371 uint8_t *stencil = sbuf + sslice->offset +
372 fd_resource_layer_offset(rsc->stencil, sslice, box->z) +
373 box->y * sslice->pitch + box->x;
374
375 if (format != PIPE_FORMAT_X32_S8X24_UINT)
376 util_format_z32_float_s8x24_uint_pack_z_float(
377 trans->staging, trans->base.stride,
378 depth, slice->pitch * 4,
379 box->width, box->height);
380
381 util_format_z32_float_s8x24_uint_pack_s_8uint(
382 trans->staging, trans->base.stride,
383 stencil, sslice->pitch,
384 box->width, box->height);
385 }
386
387 buf = trans->staging;
388 offset = 0;
389 } else if (rsc->internal_format != format &&
390 util_format_description(format)->layout == UTIL_FORMAT_LAYOUT_RGTC) {
391 assert(trans->base.box.depth == 1);
392
393 trans->base.stride = util_format_get_stride(
394 format, trans->base.box.width);
395 trans->staging = malloc(
396 util_format_get_2d_size(format, trans->base.stride,
397 trans->base.box.height));
398 if (!trans->staging)
399 goto fail;
400
401 /* if we're not discarding the whole range (or resource), we must copy
402 * the real data in.
403 */
404 if (!(usage & (PIPE_TRANSFER_DISCARD_WHOLE_RESOURCE |
405 PIPE_TRANSFER_DISCARD_RANGE))) {
406 uint8_t *rgba8 = (uint8_t *)buf + slice->offset +
407 fd_resource_layer_offset(rsc, slice, box->z) +
408 box->y * slice->pitch * rsc->cpp + box->x * rsc->cpp;
409
410 switch (format) {
411 case PIPE_FORMAT_RGTC1_UNORM:
412 case PIPE_FORMAT_RGTC1_SNORM:
413 case PIPE_FORMAT_LATC1_UNORM:
414 case PIPE_FORMAT_LATC1_SNORM:
415 util_format_rgtc1_unorm_pack_rgba_8unorm(
416 trans->staging, trans->base.stride,
417 rgba8, slice->pitch * rsc->cpp,
418 box->width, box->height);
419 break;
420 case PIPE_FORMAT_RGTC2_UNORM:
421 case PIPE_FORMAT_RGTC2_SNORM:
422 case PIPE_FORMAT_LATC2_UNORM:
423 case PIPE_FORMAT_LATC2_SNORM:
424 util_format_rgtc2_unorm_pack_rgba_8unorm(
425 trans->staging, trans->base.stride,
426 rgba8, slice->pitch * rsc->cpp,
427 box->width, box->height);
428 break;
429 default:
430 assert(!"Unexpected format");
431 break;
432 }
433 }
434
435 buf = trans->staging;
436 offset = 0;
437 }
438
439 *pptrans = ptrans;
440
441 return buf + offset;
442
443 fail:
444 fd_resource_transfer_unmap(pctx, ptrans);
445 return NULL;
446 }
447
448 static void
449 fd_resource_destroy(struct pipe_screen *pscreen,
450 struct pipe_resource *prsc)
451 {
452 struct fd_resource *rsc = fd_resource(prsc);
453 if (rsc->bo)
454 fd_bo_del(rsc->bo);
455 list_delinit(&rsc->list);
456 util_range_destroy(&rsc->valid_buffer_range);
457 FREE(rsc);
458 }
459
460 static boolean
461 fd_resource_get_handle(struct pipe_screen *pscreen,
462 struct pipe_resource *prsc,
463 struct winsys_handle *handle)
464 {
465 struct fd_resource *rsc = fd_resource(prsc);
466
467 return fd_screen_bo_get_handle(pscreen, rsc->bo,
468 rsc->slices[0].pitch * rsc->cpp, handle);
469 }
470
471
472 static const struct u_resource_vtbl fd_resource_vtbl = {
473 .resource_get_handle = fd_resource_get_handle,
474 .resource_destroy = fd_resource_destroy,
475 .transfer_map = fd_resource_transfer_map,
476 .transfer_flush_region = fd_resource_transfer_flush_region,
477 .transfer_unmap = fd_resource_transfer_unmap,
478 .transfer_inline_write = u_default_transfer_inline_write,
479 };
480
481 static uint32_t
482 setup_slices(struct fd_resource *rsc, uint32_t alignment, enum pipe_format format)
483 {
484 struct pipe_resource *prsc = &rsc->base.b;
485 enum util_format_layout layout = util_format_description(format)->layout;
486 uint32_t level, size = 0;
487 uint32_t width = prsc->width0;
488 uint32_t height = prsc->height0;
489 uint32_t depth = prsc->depth0;
490 /* in layer_first layout, the level (slice) contains just one
491 * layer (since in fact the layer contains the slices)
492 */
493 uint32_t layers_in_level = rsc->layer_first ? 1 : prsc->array_size;
494
495 for (level = 0; level <= prsc->last_level; level++) {
496 struct fd_resource_slice *slice = fd_resource_slice(rsc, level);
497 uint32_t blocks;
498
499 if (layout == UTIL_FORMAT_LAYOUT_ASTC)
500 slice->pitch = width =
501 util_align_npot(width, 32 * util_format_get_blockwidth(format));
502 else
503 slice->pitch = width = align(width, 32);
504 slice->offset = size;
505 blocks = util_format_get_nblocks(format, width, height);
506 /* 1d array and 2d array textures must all have the same layer size
507 * for each miplevel on a3xx. 3d textures can have different layer
508 * sizes for high levels, but the hw auto-sizer is buggy (or at least
509 * different than what this code does), so as soon as the layer size
510 * range gets into range, we stop reducing it.
511 */
512 if (prsc->target == PIPE_TEXTURE_3D && (
513 level == 1 ||
514 (level > 1 && rsc->slices[level - 1].size0 > 0xf000)))
515 slice->size0 = align(blocks * rsc->cpp, alignment);
516 else if (level == 0 || rsc->layer_first || alignment == 1)
517 slice->size0 = align(blocks * rsc->cpp, alignment);
518 else
519 slice->size0 = rsc->slices[level - 1].size0;
520
521 size += slice->size0 * depth * layers_in_level;
522
523 width = u_minify(width, 1);
524 height = u_minify(height, 1);
525 depth = u_minify(depth, 1);
526 }
527
528 return size;
529 }
530
531 static uint32_t
532 slice_alignment(struct pipe_screen *pscreen, const struct pipe_resource *tmpl)
533 {
534 /* on a3xx, 2d array and 3d textures seem to want their
535 * layers aligned to page boundaries:
536 */
537 switch (tmpl->target) {
538 case PIPE_TEXTURE_3D:
539 case PIPE_TEXTURE_1D_ARRAY:
540 case PIPE_TEXTURE_2D_ARRAY:
541 return 4096;
542 default:
543 return 1;
544 }
545 }
546
547 /**
548 * Create a new texture object, using the given template info.
549 */
550 static struct pipe_resource *
551 fd_resource_create(struct pipe_screen *pscreen,
552 const struct pipe_resource *tmpl)
553 {
554 struct fd_resource *rsc = CALLOC_STRUCT(fd_resource);
555 struct pipe_resource *prsc = &rsc->base.b;
556 enum pipe_format format = tmpl->format;
557 uint32_t size, alignment;
558
559 DBG("target=%d, format=%s, %ux%ux%u, array_size=%u, last_level=%u, "
560 "nr_samples=%u, usage=%u, bind=%x, flags=%x",
561 tmpl->target, util_format_name(format),
562 tmpl->width0, tmpl->height0, tmpl->depth0,
563 tmpl->array_size, tmpl->last_level, tmpl->nr_samples,
564 tmpl->usage, tmpl->bind, tmpl->flags);
565
566 if (!rsc)
567 return NULL;
568
569 *prsc = *tmpl;
570
571 pipe_reference_init(&prsc->reference, 1);
572 list_inithead(&rsc->list);
573 prsc->screen = pscreen;
574
575 util_range_init(&rsc->valid_buffer_range);
576
577 rsc->base.vtbl = &fd_resource_vtbl;
578
579 if (format == PIPE_FORMAT_Z32_FLOAT_S8X24_UINT)
580 format = PIPE_FORMAT_Z32_FLOAT;
581 else if (fd_screen(pscreen)->gpu_id < 400 &&
582 util_format_description(format)->layout == UTIL_FORMAT_LAYOUT_RGTC)
583 format = PIPE_FORMAT_R8G8B8A8_UNORM;
584 rsc->internal_format = format;
585 rsc->cpp = util_format_get_blocksize(format);
586
587 assert(rsc->cpp);
588
589 alignment = slice_alignment(pscreen, tmpl);
590 if (is_a4xx(fd_screen(pscreen))) {
591 switch (tmpl->target) {
592 case PIPE_TEXTURE_3D:
593 rsc->layer_first = false;
594 break;
595 default:
596 rsc->layer_first = true;
597 alignment = 1;
598 break;
599 }
600 }
601
602 size = setup_slices(rsc, alignment, format);
603
604 if (rsc->layer_first) {
605 rsc->layer_size = align(size, 4096);
606 size = rsc->layer_size * prsc->array_size;
607 }
608
609 realloc_bo(rsc, size);
610 if (!rsc->bo)
611 goto fail;
612
613 /* There is no native Z32F_S8 sampling or rendering format, so this must
614 * be emulated via two separate textures. The depth texture still keeps
615 * its Z32F_S8 format though, and we also keep a reference to a separate
616 * S8 texture.
617 */
618 if (tmpl->format == PIPE_FORMAT_Z32_FLOAT_S8X24_UINT) {
619 struct pipe_resource stencil = *tmpl;
620 stencil.format = PIPE_FORMAT_S8_UINT;
621 rsc->stencil = fd_resource(fd_resource_create(pscreen, &stencil));
622 if (!rsc->stencil)
623 goto fail;
624 }
625
626 return prsc;
627 fail:
628 fd_resource_destroy(pscreen, prsc);
629 return NULL;
630 }
631
632 /**
633 * Create a texture from a winsys_handle. The handle is often created in
634 * another process by first creating a pipe texture and then calling
635 * resource_get_handle.
636 */
637 static struct pipe_resource *
638 fd_resource_from_handle(struct pipe_screen *pscreen,
639 const struct pipe_resource *tmpl,
640 struct winsys_handle *handle)
641 {
642 struct fd_resource *rsc = CALLOC_STRUCT(fd_resource);
643 struct fd_resource_slice *slice = &rsc->slices[0];
644 struct pipe_resource *prsc = &rsc->base.b;
645
646 DBG("target=%d, format=%s, %ux%ux%u, array_size=%u, last_level=%u, "
647 "nr_samples=%u, usage=%u, bind=%x, flags=%x",
648 tmpl->target, util_format_name(tmpl->format),
649 tmpl->width0, tmpl->height0, tmpl->depth0,
650 tmpl->array_size, tmpl->last_level, tmpl->nr_samples,
651 tmpl->usage, tmpl->bind, tmpl->flags);
652
653 if (!rsc)
654 return NULL;
655
656 *prsc = *tmpl;
657
658 pipe_reference_init(&prsc->reference, 1);
659 list_inithead(&rsc->list);
660 prsc->screen = pscreen;
661
662 util_range_init(&rsc->valid_buffer_range);
663
664 rsc->bo = fd_screen_bo_from_handle(pscreen, handle, &slice->pitch);
665 if (!rsc->bo)
666 goto fail;
667
668 rsc->base.vtbl = &fd_resource_vtbl;
669 rsc->cpp = util_format_get_blocksize(tmpl->format);
670 slice->pitch /= rsc->cpp;
671
672 assert(rsc->cpp);
673
674 return prsc;
675
676 fail:
677 fd_resource_destroy(pscreen, prsc);
678 return NULL;
679 }
680
681 static void fd_blitter_pipe_begin(struct fd_context *ctx, bool render_cond);
682 static void fd_blitter_pipe_end(struct fd_context *ctx);
683
684 /**
685 * _copy_region using pipe (3d engine)
686 */
687 static bool
688 fd_blitter_pipe_copy_region(struct fd_context *ctx,
689 struct pipe_resource *dst,
690 unsigned dst_level,
691 unsigned dstx, unsigned dsty, unsigned dstz,
692 struct pipe_resource *src,
693 unsigned src_level,
694 const struct pipe_box *src_box)
695 {
696 /* not until we allow rendertargets to be buffers */
697 if (dst->target == PIPE_BUFFER || src->target == PIPE_BUFFER)
698 return false;
699
700 if (!util_blitter_is_copy_supported(ctx->blitter, dst, src))
701 return false;
702
703 fd_blitter_pipe_begin(ctx, false);
704 util_blitter_copy_texture(ctx->blitter,
705 dst, dst_level, dstx, dsty, dstz,
706 src, src_level, src_box);
707 fd_blitter_pipe_end(ctx);
708
709 return true;
710 }
711
712 /**
713 * Copy a block of pixels from one resource to another.
714 * The resource must be of the same format.
715 * Resources with nr_samples > 1 are not allowed.
716 */
717 static void
718 fd_resource_copy_region(struct pipe_context *pctx,
719 struct pipe_resource *dst,
720 unsigned dst_level,
721 unsigned dstx, unsigned dsty, unsigned dstz,
722 struct pipe_resource *src,
723 unsigned src_level,
724 const struct pipe_box *src_box)
725 {
726 struct fd_context *ctx = fd_context(pctx);
727
728 /* TODO if we have 2d core, or other DMA engine that could be used
729 * for simple copies and reasonably easily synchronized with the 3d
730 * core, this is where we'd plug it in..
731 */
732
733 /* try blit on 3d pipe: */
734 if (fd_blitter_pipe_copy_region(ctx,
735 dst, dst_level, dstx, dsty, dstz,
736 src, src_level, src_box))
737 return;
738
739 /* else fallback to pure sw: */
740 util_resource_copy_region(pctx,
741 dst, dst_level, dstx, dsty, dstz,
742 src, src_level, src_box);
743 }
744
745 bool
746 fd_render_condition_check(struct pipe_context *pctx)
747 {
748 struct fd_context *ctx = fd_context(pctx);
749
750 if (!ctx->cond_query)
751 return true;
752
753 union pipe_query_result res = { 0 };
754 bool wait =
755 ctx->cond_mode != PIPE_RENDER_COND_NO_WAIT &&
756 ctx->cond_mode != PIPE_RENDER_COND_BY_REGION_NO_WAIT;
757
758 if (pctx->get_query_result(pctx, ctx->cond_query, wait, &res))
759 return (bool)res.u64 != ctx->cond_cond;
760
761 return true;
762 }
763
764 /**
765 * Optimal hardware path for blitting pixels.
766 * Scaling, format conversion, up- and downsampling (resolve) are allowed.
767 */
768 static void
769 fd_blit(struct pipe_context *pctx, const struct pipe_blit_info *blit_info)
770 {
771 struct fd_context *ctx = fd_context(pctx);
772 struct pipe_blit_info info = *blit_info;
773
774 if (info.src.resource->nr_samples > 1 &&
775 info.dst.resource->nr_samples <= 1 &&
776 !util_format_is_depth_or_stencil(info.src.resource->format) &&
777 !util_format_is_pure_integer(info.src.resource->format)) {
778 DBG("color resolve unimplemented");
779 return;
780 }
781
782 if (info.render_condition_enable && !fd_render_condition_check(pctx))
783 return;
784
785 if (util_try_blit_via_copy_region(pctx, &info)) {
786 return; /* done */
787 }
788
789 if (info.mask & PIPE_MASK_S) {
790 DBG("cannot blit stencil, skipping");
791 info.mask &= ~PIPE_MASK_S;
792 }
793
794 if (!util_blitter_is_blit_supported(ctx->blitter, &info)) {
795 DBG("blit unsupported %s -> %s",
796 util_format_short_name(info.src.resource->format),
797 util_format_short_name(info.dst.resource->format));
798 return;
799 }
800
801 fd_blitter_pipe_begin(ctx, info.render_condition_enable);
802 util_blitter_blit(ctx->blitter, &info);
803 fd_blitter_pipe_end(ctx);
804 }
805
806 static void
807 fd_blitter_pipe_begin(struct fd_context *ctx, bool render_cond)
808 {
809 util_blitter_save_vertex_buffer_slot(ctx->blitter, ctx->vtx.vertexbuf.vb);
810 util_blitter_save_vertex_elements(ctx->blitter, ctx->vtx.vtx);
811 util_blitter_save_vertex_shader(ctx->blitter, ctx->prog.vp);
812 util_blitter_save_so_targets(ctx->blitter, ctx->streamout.num_targets,
813 ctx->streamout.targets);
814 util_blitter_save_rasterizer(ctx->blitter, ctx->rasterizer);
815 util_blitter_save_viewport(ctx->blitter, &ctx->viewport);
816 util_blitter_save_scissor(ctx->blitter, &ctx->scissor);
817 util_blitter_save_fragment_shader(ctx->blitter, ctx->prog.fp);
818 util_blitter_save_blend(ctx->blitter, ctx->blend);
819 util_blitter_save_depth_stencil_alpha(ctx->blitter, ctx->zsa);
820 util_blitter_save_stencil_ref(ctx->blitter, &ctx->stencil_ref);
821 util_blitter_save_sample_mask(ctx->blitter, ctx->sample_mask);
822 util_blitter_save_framebuffer(ctx->blitter, &ctx->framebuffer);
823 util_blitter_save_fragment_sampler_states(ctx->blitter,
824 ctx->fragtex.num_samplers,
825 (void **)ctx->fragtex.samplers);
826 util_blitter_save_fragment_sampler_views(ctx->blitter,
827 ctx->fragtex.num_textures, ctx->fragtex.textures);
828 if (!render_cond)
829 util_blitter_save_render_condition(ctx->blitter,
830 ctx->cond_query, ctx->cond_cond, ctx->cond_mode);
831
832 fd_hw_query_set_stage(ctx, ctx->ring, FD_STAGE_BLIT);
833 }
834
835 static void
836 fd_blitter_pipe_end(struct fd_context *ctx)
837 {
838 fd_hw_query_set_stage(ctx, ctx->ring, FD_STAGE_NULL);
839 }
840
841 static void
842 fd_flush_resource(struct pipe_context *pctx, struct pipe_resource *prsc)
843 {
844 struct fd_resource *rsc = fd_resource(prsc);
845
846 if (pending(rsc, FD_PENDING_WRITE | FD_PENDING_READ))
847 fd_context_render(pctx);
848 }
849
850 void
851 fd_resource_screen_init(struct pipe_screen *pscreen)
852 {
853 pscreen->resource_create = fd_resource_create;
854 pscreen->resource_from_handle = fd_resource_from_handle;
855 pscreen->resource_get_handle = u_resource_get_handle_vtbl;
856 pscreen->resource_destroy = u_resource_destroy_vtbl;
857 }
858
859 void
860 fd_resource_context_init(struct pipe_context *pctx)
861 {
862 pctx->transfer_map = u_transfer_map_vtbl;
863 pctx->transfer_flush_region = u_transfer_flush_region_vtbl;
864 pctx->transfer_unmap = u_transfer_unmap_vtbl;
865 pctx->transfer_inline_write = u_transfer_inline_write_vtbl;
866 pctx->create_surface = fd_create_surface;
867 pctx->surface_destroy = fd_surface_destroy;
868 pctx->resource_copy_region = fd_resource_copy_region;
869 pctx->blit = fd_blit;
870 pctx->flush_resource = fd_flush_resource;
871 }