freedreno: fix layer_stride
[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 char *buf;
105 int ret = 0;
106
107 ptrans = util_slab_alloc(&ctx->transfer_pool);
108 if (!ptrans)
109 return NULL;
110
111 /* util_slab_alloc() doesn't zero: */
112 memset(ptrans, 0, sizeof(*ptrans));
113
114 pipe_resource_reference(&ptrans->resource, prsc);
115 ptrans->level = level;
116 ptrans->usage = usage;
117 ptrans->box = *box;
118 ptrans->stride = slice->pitch * rsc->cpp;
119 ptrans->layer_stride = slice->size0;
120
121 if (usage & PIPE_TRANSFER_READ)
122 op |= DRM_FREEDRENO_PREP_READ;
123
124 if (usage & PIPE_TRANSFER_WRITE)
125 op |= DRM_FREEDRENO_PREP_WRITE;
126
127 if (usage & PIPE_TRANSFER_DISCARD_WHOLE_RESOURCE)
128 op |= DRM_FREEDRENO_PREP_NOSYNC;
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_UNSYNCHRONIZED)) {
135 ret = fd_bo_cpu_prep(rsc->bo, ctx->screen->pipe, op);
136 if ((ret == -EBUSY) && (usage & PIPE_TRANSFER_DISCARD_WHOLE_RESOURCE))
137 realloc_bo(rsc, fd_bo_size(rsc->bo));
138 else 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 return buf + slice->offset +
151 box->y / util_format_get_blockheight(format) * ptrans->stride +
152 box->x / util_format_get_blockwidth(format) * rsc->cpp +
153 box->z * slice->size0;
154
155 fail:
156 fd_resource_transfer_unmap(pctx, ptrans);
157 return NULL;
158 }
159
160 static void
161 fd_resource_destroy(struct pipe_screen *pscreen,
162 struct pipe_resource *prsc)
163 {
164 struct fd_resource *rsc = fd_resource(prsc);
165 if (rsc->bo)
166 fd_bo_del(rsc->bo);
167 FREE(rsc);
168 }
169
170 static boolean
171 fd_resource_get_handle(struct pipe_screen *pscreen,
172 struct pipe_resource *prsc,
173 struct winsys_handle *handle)
174 {
175 struct fd_resource *rsc = fd_resource(prsc);
176
177 return fd_screen_bo_get_handle(pscreen, rsc->bo,
178 rsc->slices[0].pitch * rsc->cpp, handle);
179 }
180
181
182 static const struct u_resource_vtbl fd_resource_vtbl = {
183 .resource_get_handle = fd_resource_get_handle,
184 .resource_destroy = fd_resource_destroy,
185 .transfer_map = fd_resource_transfer_map,
186 .transfer_flush_region = fd_resource_transfer_flush_region,
187 .transfer_unmap = fd_resource_transfer_unmap,
188 .transfer_inline_write = u_default_transfer_inline_write,
189 };
190
191 static uint32_t
192 setup_slices(struct fd_resource *rsc)
193 {
194 struct pipe_resource *prsc = &rsc->base.b;
195 uint32_t level, size = 0;
196 uint32_t width = prsc->width0;
197 uint32_t height = prsc->height0;
198 uint32_t depth = prsc->depth0;
199
200 for (level = 0; level <= prsc->last_level; level++) {
201 struct fd_resource_slice *slice = fd_resource_slice(rsc, level);
202 uint32_t aligned_width = align(width, 32);
203
204 slice->pitch = aligned_width;
205 slice->offset = size;
206 slice->size0 = slice->pitch * height * rsc->cpp;
207
208 size += slice->size0 * depth * prsc->array_size;
209
210 width = u_minify(width, 1);
211 height = u_minify(height, 1);
212 depth = u_minify(depth, 1);
213 }
214
215 return size;
216 }
217
218 /* 2d array and 3d textures seem to want their layers aligned to
219 * page boundaries
220 */
221 static uint32_t
222 setup_slices_array(struct fd_resource *rsc)
223 {
224 struct pipe_resource *prsc = &rsc->base.b;
225 uint32_t level, size = 0;
226 uint32_t width = prsc->width0;
227 uint32_t height = prsc->height0;
228 uint32_t depth = prsc->depth0;
229
230 for (level = 0; level <= prsc->last_level; level++) {
231 struct fd_resource_slice *slice = fd_resource_slice(rsc, level);
232 uint32_t aligned_width = align(width, 32);
233
234 slice->pitch = aligned_width;
235 slice->offset = size;
236 slice->size0 = align(slice->pitch * height * rsc->cpp, 4096);
237
238 size += slice->size0 * depth * prsc->array_size;
239
240 width = u_minify(width, 1);
241 height = u_minify(height, 1);
242 depth = u_minify(depth, 1);
243 }
244
245 return size;
246 }
247
248 /**
249 * Create a new texture object, using the given template info.
250 */
251 static struct pipe_resource *
252 fd_resource_create(struct pipe_screen *pscreen,
253 const struct pipe_resource *tmpl)
254 {
255 struct fd_resource *rsc = CALLOC_STRUCT(fd_resource);
256 struct pipe_resource *prsc = &rsc->base.b;
257 uint32_t size;
258
259 DBG("target=%d, format=%s, %ux%ux%u, array_size=%u, last_level=%u, "
260 "nr_samples=%u, usage=%u, bind=%x, flags=%x",
261 tmpl->target, util_format_name(tmpl->format),
262 tmpl->width0, tmpl->height0, tmpl->depth0,
263 tmpl->array_size, tmpl->last_level, tmpl->nr_samples,
264 tmpl->usage, tmpl->bind, tmpl->flags);
265
266 if (!rsc)
267 return NULL;
268
269 *prsc = *tmpl;
270
271 pipe_reference_init(&prsc->reference, 1);
272 prsc->screen = pscreen;
273
274 rsc->base.vtbl = &fd_resource_vtbl;
275 rsc->cpp = util_format_get_blocksize(tmpl->format);
276
277 assert(rsc->cpp);
278
279 switch (tmpl->target) {
280 case PIPE_TEXTURE_3D:
281 case PIPE_TEXTURE_1D_ARRAY:
282 case PIPE_TEXTURE_2D_ARRAY:
283 size = setup_slices_array(rsc);
284 break;
285 default:
286 size = setup_slices(rsc);
287 break;
288 }
289
290 realloc_bo(rsc, size);
291 if (!rsc->bo)
292 goto fail;
293
294 return prsc;
295 fail:
296 fd_resource_destroy(pscreen, prsc);
297 return NULL;
298 }
299
300 /**
301 * Create a texture from a winsys_handle. The handle is often created in
302 * another process by first creating a pipe texture and then calling
303 * resource_get_handle.
304 */
305 static struct pipe_resource *
306 fd_resource_from_handle(struct pipe_screen *pscreen,
307 const struct pipe_resource *tmpl,
308 struct winsys_handle *handle)
309 {
310 struct fd_resource *rsc = CALLOC_STRUCT(fd_resource);
311 struct fd_resource_slice *slice = &rsc->slices[0];
312 struct pipe_resource *prsc = &rsc->base.b;
313
314 DBG("target=%d, format=%s, %ux%ux%u, array_size=%u, last_level=%u, "
315 "nr_samples=%u, usage=%u, bind=%x, flags=%x",
316 tmpl->target, util_format_name(tmpl->format),
317 tmpl->width0, tmpl->height0, tmpl->depth0,
318 tmpl->array_size, tmpl->last_level, tmpl->nr_samples,
319 tmpl->usage, tmpl->bind, tmpl->flags);
320
321 if (!rsc)
322 return NULL;
323
324 *prsc = *tmpl;
325
326 pipe_reference_init(&prsc->reference, 1);
327 prsc->screen = pscreen;
328
329 rsc->bo = fd_screen_bo_from_handle(pscreen, handle, &slice->pitch);
330 if (!rsc->bo)
331 goto fail;
332
333 rsc->base.vtbl = &fd_resource_vtbl;
334 rsc->cpp = util_format_get_blocksize(tmpl->format);
335 slice->pitch /= rsc->cpp;
336
337 assert(rsc->cpp);
338
339 return prsc;
340
341 fail:
342 fd_resource_destroy(pscreen, prsc);
343 return NULL;
344 }
345
346 static void fd_blitter_pipe_begin(struct fd_context *ctx);
347 static void fd_blitter_pipe_end(struct fd_context *ctx);
348
349 /**
350 * _copy_region using pipe (3d engine)
351 */
352 static bool
353 fd_blitter_pipe_copy_region(struct fd_context *ctx,
354 struct pipe_resource *dst,
355 unsigned dst_level,
356 unsigned dstx, unsigned dsty, unsigned dstz,
357 struct pipe_resource *src,
358 unsigned src_level,
359 const struct pipe_box *src_box)
360 {
361 /* not until we allow rendertargets to be buffers */
362 if (dst->target == PIPE_BUFFER || src->target == PIPE_BUFFER)
363 return false;
364
365 if (!util_blitter_is_copy_supported(ctx->blitter, dst, src))
366 return false;
367
368 fd_blitter_pipe_begin(ctx);
369 util_blitter_copy_texture(ctx->blitter,
370 dst, dst_level, dstx, dsty, dstz,
371 src, src_level, src_box);
372 fd_blitter_pipe_end(ctx);
373
374 return true;
375 }
376
377 /**
378 * Copy a block of pixels from one resource to another.
379 * The resource must be of the same format.
380 * Resources with nr_samples > 1 are not allowed.
381 */
382 static void
383 fd_resource_copy_region(struct pipe_context *pctx,
384 struct pipe_resource *dst,
385 unsigned dst_level,
386 unsigned dstx, unsigned dsty, unsigned dstz,
387 struct pipe_resource *src,
388 unsigned src_level,
389 const struct pipe_box *src_box)
390 {
391 struct fd_context *ctx = fd_context(pctx);
392
393 /* TODO if we have 2d core, or other DMA engine that could be used
394 * for simple copies and reasonably easily synchronized with the 3d
395 * core, this is where we'd plug it in..
396 */
397
398 /* try blit on 3d pipe: */
399 if (fd_blitter_pipe_copy_region(ctx,
400 dst, dst_level, dstx, dsty, dstz,
401 src, src_level, src_box))
402 return;
403
404 /* else fallback to pure sw: */
405 util_resource_copy_region(pctx,
406 dst, dst_level, dstx, dsty, dstz,
407 src, src_level, src_box);
408 }
409
410 /**
411 * Optimal hardware path for blitting pixels.
412 * Scaling, format conversion, up- and downsampling (resolve) are allowed.
413 */
414 static void
415 fd_blit(struct pipe_context *pctx, const struct pipe_blit_info *blit_info)
416 {
417 struct fd_context *ctx = fd_context(pctx);
418 struct pipe_blit_info info = *blit_info;
419
420 if (info.src.resource->nr_samples > 1 &&
421 info.dst.resource->nr_samples <= 1 &&
422 !util_format_is_depth_or_stencil(info.src.resource->format) &&
423 !util_format_is_pure_integer(info.src.resource->format)) {
424 DBG("color resolve unimplemented");
425 return;
426 }
427
428 if (util_try_blit_via_copy_region(pctx, &info)) {
429 return; /* done */
430 }
431
432 if (info.mask & PIPE_MASK_S) {
433 DBG("cannot blit stencil, skipping");
434 info.mask &= ~PIPE_MASK_S;
435 }
436
437 if (!util_blitter_is_blit_supported(ctx->blitter, &info)) {
438 DBG("blit unsupported %s -> %s",
439 util_format_short_name(info.src.resource->format),
440 util_format_short_name(info.dst.resource->format));
441 return;
442 }
443
444 fd_blitter_pipe_begin(ctx);
445 util_blitter_blit(ctx->blitter, &info);
446 fd_blitter_pipe_end(ctx);
447 }
448
449 static void
450 fd_blitter_pipe_begin(struct fd_context *ctx)
451 {
452 util_blitter_save_vertex_buffer_slot(ctx->blitter, ctx->vtx.vertexbuf.vb);
453 util_blitter_save_vertex_elements(ctx->blitter, ctx->vtx.vtx);
454 util_blitter_save_vertex_shader(ctx->blitter, ctx->prog.vp);
455 util_blitter_save_rasterizer(ctx->blitter, ctx->rasterizer);
456 util_blitter_save_viewport(ctx->blitter, &ctx->viewport);
457 util_blitter_save_scissor(ctx->blitter, &ctx->scissor);
458 util_blitter_save_fragment_shader(ctx->blitter, ctx->prog.fp);
459 util_blitter_save_blend(ctx->blitter, ctx->blend);
460 util_blitter_save_depth_stencil_alpha(ctx->blitter, ctx->zsa);
461 util_blitter_save_stencil_ref(ctx->blitter, &ctx->stencil_ref);
462 util_blitter_save_sample_mask(ctx->blitter, ctx->sample_mask);
463 util_blitter_save_framebuffer(ctx->blitter, &ctx->framebuffer);
464 util_blitter_save_fragment_sampler_states(ctx->blitter,
465 ctx->fragtex.num_samplers,
466 (void **)ctx->fragtex.samplers);
467 util_blitter_save_fragment_sampler_views(ctx->blitter,
468 ctx->fragtex.num_textures, ctx->fragtex.textures);
469
470 fd_hw_query_set_stage(ctx, ctx->ring, FD_STAGE_BLIT);
471 }
472
473 static void
474 fd_blitter_pipe_end(struct fd_context *ctx)
475 {
476 fd_hw_query_set_stage(ctx, ctx->ring, FD_STAGE_NULL);
477 }
478
479 static void
480 fd_flush_resource(struct pipe_context *pctx, struct pipe_resource *prsc)
481 {
482 struct fd_resource *rsc = fd_resource(prsc);
483
484 if (rsc->dirty)
485 fd_context_render(pctx);
486 }
487
488 void
489 fd_resource_screen_init(struct pipe_screen *pscreen)
490 {
491 pscreen->resource_create = fd_resource_create;
492 pscreen->resource_from_handle = fd_resource_from_handle;
493 pscreen->resource_get_handle = u_resource_get_handle_vtbl;
494 pscreen->resource_destroy = u_resource_destroy_vtbl;
495 }
496
497 void
498 fd_resource_context_init(struct pipe_context *pctx)
499 {
500 pctx->transfer_map = u_transfer_map_vtbl;
501 pctx->transfer_flush_region = u_transfer_flush_region_vtbl;
502 pctx->transfer_unmap = u_transfer_unmap_vtbl;
503 pctx->transfer_inline_write = u_transfer_inline_write_vtbl;
504 pctx->create_surface = fd_create_surface;
505 pctx->surface_destroy = fd_surface_destroy;
506 pctx->resource_copy_region = fd_resource_copy_region;
507 pctx->blit = fd_blit;
508 pctx->flush_resource = fd_flush_resource;
509 }