freedreno/a3xx: use the same layer size for all slices
[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_inlines.h"
31 #include "util/u_transfer.h"
32 #include "util/u_string.h"
33 #include "util/u_surface.h"
34
35 #include "freedreno_resource.h"
36 #include "freedreno_screen.h"
37 #include "freedreno_surface.h"
38 #include "freedreno_context.h"
39 #include "freedreno_query_hw.h"
40 #include "freedreno_util.h"
41
42 #include <errno.h>
43
44 static void
45 realloc_bo(struct fd_resource *rsc, uint32_t size)
46 {
47 struct fd_screen *screen = fd_screen(rsc->base.b.screen);
48 uint32_t flags = DRM_FREEDRENO_GEM_CACHE_WCOMBINE |
49 DRM_FREEDRENO_GEM_TYPE_KMEM; /* TODO */
50
51 /* if we start using things other than write-combine,
52 * be sure to check for PIPE_RESOURCE_FLAG_MAP_COHERENT
53 */
54
55 if (rsc->bo)
56 fd_bo_del(rsc->bo);
57
58 rsc->bo = fd_bo_new(screen->dev, size, flags);
59 rsc->timestamp = 0;
60 rsc->dirty = false;
61 }
62
63 static void fd_resource_transfer_flush_region(struct pipe_context *pctx,
64 struct pipe_transfer *ptrans,
65 const struct pipe_box *box)
66 {
67 struct fd_context *ctx = fd_context(pctx);
68 struct fd_resource *rsc = fd_resource(ptrans->resource);
69
70 if (rsc->dirty)
71 fd_context_render(pctx);
72
73 if (rsc->timestamp) {
74 fd_pipe_wait(ctx->screen->pipe, rsc->timestamp);
75 rsc->timestamp = 0;
76 }
77 }
78
79 static void
80 fd_resource_transfer_unmap(struct pipe_context *pctx,
81 struct pipe_transfer *ptrans)
82 {
83 struct fd_context *ctx = fd_context(pctx);
84 struct fd_resource *rsc = fd_resource(ptrans->resource);
85 if (!(ptrans->usage & PIPE_TRANSFER_UNSYNCHRONIZED))
86 fd_bo_cpu_fini(rsc->bo);
87 pipe_resource_reference(&ptrans->resource, NULL);
88 util_slab_free(&ctx->transfer_pool, ptrans);
89 }
90
91 static void *
92 fd_resource_transfer_map(struct pipe_context *pctx,
93 struct pipe_resource *prsc,
94 unsigned level, unsigned usage,
95 const struct pipe_box *box,
96 struct pipe_transfer **pptrans)
97 {
98 struct fd_context *ctx = fd_context(pctx);
99 struct fd_resource *rsc = fd_resource(prsc);
100 struct fd_resource_slice *slice = fd_resource_slice(rsc, level);
101 struct pipe_transfer *ptrans;
102 enum pipe_format format = prsc->format;
103 uint32_t op = 0;
104 uint32_t offset;
105 char *buf;
106 int ret = 0;
107
108 DBG("prsc=%p, level=%u, usage=%x", prsc, level, usage);
109
110 ptrans = util_slab_alloc(&ctx->transfer_pool);
111 if (!ptrans)
112 return NULL;
113
114 /* util_slab_alloc() doesn't zero: */
115 memset(ptrans, 0, sizeof(*ptrans));
116
117 pipe_resource_reference(&ptrans->resource, prsc);
118 ptrans->level = level;
119 ptrans->usage = usage;
120 ptrans->box = *box;
121 ptrans->stride = slice->pitch * rsc->cpp;
122 ptrans->layer_stride = slice->size0;
123
124 if (usage & PIPE_TRANSFER_READ)
125 op |= DRM_FREEDRENO_PREP_READ;
126
127 if (usage & PIPE_TRANSFER_WRITE)
128 op |= DRM_FREEDRENO_PREP_WRITE;
129
130 /* some state trackers (at least XA) don't do this.. */
131 if (!(usage & (PIPE_TRANSFER_FLUSH_EXPLICIT | PIPE_TRANSFER_DISCARD_WHOLE_RESOURCE)))
132 fd_resource_transfer_flush_region(pctx, ptrans, box);
133
134 if (usage & PIPE_TRANSFER_DISCARD_WHOLE_RESOURCE) {
135 realloc_bo(rsc, fd_bo_size(rsc->bo));
136 } else if (!(usage & PIPE_TRANSFER_UNSYNCHRONIZED)) {
137 ret = fd_bo_cpu_prep(rsc->bo, ctx->screen->pipe, op);
138 if (ret)
139 goto fail;
140 }
141
142 buf = fd_bo_map(rsc->bo);
143 if (!buf) {
144 fd_resource_transfer_unmap(pctx, ptrans);
145 return NULL;
146 }
147
148 *pptrans = ptrans;
149
150 if (rsc->layer_first) {
151 offset = slice->offset +
152 box->y / util_format_get_blockheight(format) * ptrans->stride +
153 box->x / util_format_get_blockwidth(format) * rsc->cpp +
154 box->z * rsc->layer_size;
155 } else {
156 offset = slice->offset +
157 box->y / util_format_get_blockheight(format) * ptrans->stride +
158 box->x / util_format_get_blockwidth(format) * rsc->cpp +
159 box->z * slice->size0;
160 }
161
162 return buf + offset;
163
164 fail:
165 fd_resource_transfer_unmap(pctx, ptrans);
166 return NULL;
167 }
168
169 static void
170 fd_resource_destroy(struct pipe_screen *pscreen,
171 struct pipe_resource *prsc)
172 {
173 struct fd_resource *rsc = fd_resource(prsc);
174 if (rsc->bo)
175 fd_bo_del(rsc->bo);
176 FREE(rsc);
177 }
178
179 static boolean
180 fd_resource_get_handle(struct pipe_screen *pscreen,
181 struct pipe_resource *prsc,
182 struct winsys_handle *handle)
183 {
184 struct fd_resource *rsc = fd_resource(prsc);
185
186 return fd_screen_bo_get_handle(pscreen, rsc->bo,
187 rsc->slices[0].pitch * rsc->cpp, handle);
188 }
189
190
191 static const struct u_resource_vtbl fd_resource_vtbl = {
192 .resource_get_handle = fd_resource_get_handle,
193 .resource_destroy = fd_resource_destroy,
194 .transfer_map = fd_resource_transfer_map,
195 .transfer_flush_region = fd_resource_transfer_flush_region,
196 .transfer_unmap = fd_resource_transfer_unmap,
197 .transfer_inline_write = u_default_transfer_inline_write,
198 };
199
200 static uint32_t
201 setup_slices(struct fd_resource *rsc, uint32_t alignment)
202 {
203 struct pipe_resource *prsc = &rsc->base.b;
204 uint32_t level, size = 0;
205 uint32_t width = prsc->width0;
206 uint32_t height = prsc->height0;
207 uint32_t depth = prsc->depth0;
208 /* in layer_first layout, the level (slice) contains just one
209 * layer (since in fact the layer contains the slices)
210 */
211 uint32_t layers_in_level = rsc->layer_first ? 1 : prsc->array_size;
212
213 for (level = 0; level <= prsc->last_level; level++) {
214 struct fd_resource_slice *slice = fd_resource_slice(rsc, level);
215
216 slice->pitch = align(width, 32);
217 slice->offset = size;
218 /* 1d array, 2d array, 3d textures (but not cube!) must all have the
219 * same layer size for each miplevel on a3xx. These are also the
220 * targets that have non-1 alignment.
221 */
222 if (level == 0 || layers_in_level == 1 || alignment == 1)
223 slice->size0 = align(slice->pitch * height * rsc->cpp, alignment);
224 else
225 slice->size0 = rsc->slices[0].size0;
226
227 size += slice->size0 * depth * layers_in_level;
228
229 width = u_minify(width, 1);
230 height = u_minify(height, 1);
231 depth = u_minify(depth, 1);
232 }
233
234 return size;
235 }
236
237 static uint32_t
238 slice_alignment(struct pipe_screen *pscreen, const struct pipe_resource *tmpl)
239 {
240 /* on a3xx, 2d array and 3d textures seem to want their
241 * layers aligned to page boundaries:
242 */
243 switch (tmpl->target) {
244 case PIPE_TEXTURE_3D:
245 case PIPE_TEXTURE_1D_ARRAY:
246 case PIPE_TEXTURE_2D_ARRAY:
247 return 4096;
248 default:
249 return 1;
250 }
251 }
252
253 /**
254 * Create a new texture object, using the given template info.
255 */
256 static struct pipe_resource *
257 fd_resource_create(struct pipe_screen *pscreen,
258 const struct pipe_resource *tmpl)
259 {
260 struct fd_resource *rsc = CALLOC_STRUCT(fd_resource);
261 struct pipe_resource *prsc = &rsc->base.b;
262 uint32_t size;
263
264 DBG("target=%d, format=%s, %ux%ux%u, array_size=%u, last_level=%u, "
265 "nr_samples=%u, usage=%u, bind=%x, flags=%x",
266 tmpl->target, util_format_name(tmpl->format),
267 tmpl->width0, tmpl->height0, tmpl->depth0,
268 tmpl->array_size, tmpl->last_level, tmpl->nr_samples,
269 tmpl->usage, tmpl->bind, tmpl->flags);
270
271 if (!rsc)
272 return NULL;
273
274 *prsc = *tmpl;
275
276 pipe_reference_init(&prsc->reference, 1);
277 prsc->screen = pscreen;
278
279 rsc->base.vtbl = &fd_resource_vtbl;
280 rsc->cpp = util_format_get_blocksize(tmpl->format);
281
282 assert(rsc->cpp);
283
284 if (is_a4xx(fd_screen(pscreen))) {
285 switch (tmpl->target) {
286 case PIPE_TEXTURE_3D:
287 /* TODO 3D_ARRAY? */
288 rsc->layer_first = false;
289 break;
290 default:
291 rsc->layer_first = true;
292 break;
293 }
294 }
295
296 size = setup_slices(rsc, slice_alignment(pscreen, tmpl));
297
298 if (rsc->layer_first) {
299 rsc->layer_size = align(size, 4096);
300 size = rsc->layer_size * prsc->array_size;
301 }
302
303 realloc_bo(rsc, size);
304 if (!rsc->bo)
305 goto fail;
306
307 return prsc;
308 fail:
309 fd_resource_destroy(pscreen, prsc);
310 return NULL;
311 }
312
313 /**
314 * Create a texture from a winsys_handle. The handle is often created in
315 * another process by first creating a pipe texture and then calling
316 * resource_get_handle.
317 */
318 static struct pipe_resource *
319 fd_resource_from_handle(struct pipe_screen *pscreen,
320 const struct pipe_resource *tmpl,
321 struct winsys_handle *handle)
322 {
323 struct fd_resource *rsc = CALLOC_STRUCT(fd_resource);
324 struct fd_resource_slice *slice = &rsc->slices[0];
325 struct pipe_resource *prsc = &rsc->base.b;
326
327 DBG("target=%d, format=%s, %ux%ux%u, array_size=%u, last_level=%u, "
328 "nr_samples=%u, usage=%u, bind=%x, flags=%x",
329 tmpl->target, util_format_name(tmpl->format),
330 tmpl->width0, tmpl->height0, tmpl->depth0,
331 tmpl->array_size, tmpl->last_level, tmpl->nr_samples,
332 tmpl->usage, tmpl->bind, tmpl->flags);
333
334 if (!rsc)
335 return NULL;
336
337 *prsc = *tmpl;
338
339 pipe_reference_init(&prsc->reference, 1);
340 prsc->screen = pscreen;
341
342 rsc->bo = fd_screen_bo_from_handle(pscreen, handle, &slice->pitch);
343 if (!rsc->bo)
344 goto fail;
345
346 rsc->base.vtbl = &fd_resource_vtbl;
347 rsc->cpp = util_format_get_blocksize(tmpl->format);
348 slice->pitch /= rsc->cpp;
349
350 assert(rsc->cpp);
351
352 return prsc;
353
354 fail:
355 fd_resource_destroy(pscreen, prsc);
356 return NULL;
357 }
358
359 static void fd_blitter_pipe_begin(struct fd_context *ctx);
360 static void fd_blitter_pipe_end(struct fd_context *ctx);
361
362 /**
363 * _copy_region using pipe (3d engine)
364 */
365 static bool
366 fd_blitter_pipe_copy_region(struct fd_context *ctx,
367 struct pipe_resource *dst,
368 unsigned dst_level,
369 unsigned dstx, unsigned dsty, unsigned dstz,
370 struct pipe_resource *src,
371 unsigned src_level,
372 const struct pipe_box *src_box)
373 {
374 /* not until we allow rendertargets to be buffers */
375 if (dst->target == PIPE_BUFFER || src->target == PIPE_BUFFER)
376 return false;
377
378 if (!util_blitter_is_copy_supported(ctx->blitter, dst, src))
379 return false;
380
381 fd_blitter_pipe_begin(ctx);
382 util_blitter_copy_texture(ctx->blitter,
383 dst, dst_level, dstx, dsty, dstz,
384 src, src_level, src_box);
385 fd_blitter_pipe_end(ctx);
386
387 return true;
388 }
389
390 /**
391 * Copy a block of pixels from one resource to another.
392 * The resource must be of the same format.
393 * Resources with nr_samples > 1 are not allowed.
394 */
395 static void
396 fd_resource_copy_region(struct pipe_context *pctx,
397 struct pipe_resource *dst,
398 unsigned dst_level,
399 unsigned dstx, unsigned dsty, unsigned dstz,
400 struct pipe_resource *src,
401 unsigned src_level,
402 const struct pipe_box *src_box)
403 {
404 struct fd_context *ctx = fd_context(pctx);
405
406 /* TODO if we have 2d core, or other DMA engine that could be used
407 * for simple copies and reasonably easily synchronized with the 3d
408 * core, this is where we'd plug it in..
409 */
410
411 /* try blit on 3d pipe: */
412 if (fd_blitter_pipe_copy_region(ctx,
413 dst, dst_level, dstx, dsty, dstz,
414 src, src_level, src_box))
415 return;
416
417 /* else fallback to pure sw: */
418 util_resource_copy_region(pctx,
419 dst, dst_level, dstx, dsty, dstz,
420 src, src_level, src_box);
421 }
422
423 /**
424 * Optimal hardware path for blitting pixels.
425 * Scaling, format conversion, up- and downsampling (resolve) are allowed.
426 */
427 static void
428 fd_blit(struct pipe_context *pctx, const struct pipe_blit_info *blit_info)
429 {
430 struct fd_context *ctx = fd_context(pctx);
431 struct pipe_blit_info info = *blit_info;
432
433 if (info.src.resource->nr_samples > 1 &&
434 info.dst.resource->nr_samples <= 1 &&
435 !util_format_is_depth_or_stencil(info.src.resource->format) &&
436 !util_format_is_pure_integer(info.src.resource->format)) {
437 DBG("color resolve unimplemented");
438 return;
439 }
440
441 if (util_try_blit_via_copy_region(pctx, &info)) {
442 return; /* done */
443 }
444
445 if (info.mask & PIPE_MASK_S) {
446 DBG("cannot blit stencil, skipping");
447 info.mask &= ~PIPE_MASK_S;
448 }
449
450 if (!util_blitter_is_blit_supported(ctx->blitter, &info)) {
451 DBG("blit unsupported %s -> %s",
452 util_format_short_name(info.src.resource->format),
453 util_format_short_name(info.dst.resource->format));
454 return;
455 }
456
457 fd_blitter_pipe_begin(ctx);
458 util_blitter_blit(ctx->blitter, &info);
459 fd_blitter_pipe_end(ctx);
460 }
461
462 static void
463 fd_blitter_pipe_begin(struct fd_context *ctx)
464 {
465 util_blitter_save_vertex_buffer_slot(ctx->blitter, ctx->vtx.vertexbuf.vb);
466 util_blitter_save_vertex_elements(ctx->blitter, ctx->vtx.vtx);
467 util_blitter_save_vertex_shader(ctx->blitter, ctx->prog.vp);
468 util_blitter_save_rasterizer(ctx->blitter, ctx->rasterizer);
469 util_blitter_save_viewport(ctx->blitter, &ctx->viewport);
470 util_blitter_save_scissor(ctx->blitter, &ctx->scissor);
471 util_blitter_save_fragment_shader(ctx->blitter, ctx->prog.fp);
472 util_blitter_save_blend(ctx->blitter, ctx->blend);
473 util_blitter_save_depth_stencil_alpha(ctx->blitter, ctx->zsa);
474 util_blitter_save_stencil_ref(ctx->blitter, &ctx->stencil_ref);
475 util_blitter_save_sample_mask(ctx->blitter, ctx->sample_mask);
476 util_blitter_save_framebuffer(ctx->blitter, &ctx->framebuffer);
477 util_blitter_save_fragment_sampler_states(ctx->blitter,
478 ctx->fragtex.num_samplers,
479 (void **)ctx->fragtex.samplers);
480 util_blitter_save_fragment_sampler_views(ctx->blitter,
481 ctx->fragtex.num_textures, ctx->fragtex.textures);
482
483 fd_hw_query_set_stage(ctx, ctx->ring, FD_STAGE_BLIT);
484 }
485
486 static void
487 fd_blitter_pipe_end(struct fd_context *ctx)
488 {
489 fd_hw_query_set_stage(ctx, ctx->ring, FD_STAGE_NULL);
490 }
491
492 static void
493 fd_flush_resource(struct pipe_context *pctx, struct pipe_resource *prsc)
494 {
495 struct fd_resource *rsc = fd_resource(prsc);
496
497 if (rsc->dirty)
498 fd_context_render(pctx);
499 }
500
501 void
502 fd_resource_screen_init(struct pipe_screen *pscreen)
503 {
504 pscreen->resource_create = fd_resource_create;
505 pscreen->resource_from_handle = fd_resource_from_handle;
506 pscreen->resource_get_handle = u_resource_get_handle_vtbl;
507 pscreen->resource_destroy = u_resource_destroy_vtbl;
508 }
509
510 void
511 fd_resource_context_init(struct pipe_context *pctx)
512 {
513 pctx->transfer_map = u_transfer_map_vtbl;
514 pctx->transfer_flush_region = u_transfer_flush_region_vtbl;
515 pctx->transfer_unmap = u_transfer_unmap_vtbl;
516 pctx->transfer_inline_write = u_transfer_inline_write_vtbl;
517 pctx->create_surface = fd_create_surface;
518 pctx->surface_destroy = fd_surface_destroy;
519 pctx->resource_copy_region = fd_resource_copy_region;
520 pctx->blit = fd_blit;
521 pctx->flush_resource = fd_flush_resource;
522 }