v3d: Add support for handling OOM signals from the simulator.
[mesa.git] / src / gallium / drivers / v3d / v3d_blit.c
1 /*
2 * Copyright © 2015-2017 Broadcom
3 *
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
10 *
11 * The above copyright notice and this permission notice (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
13 * Software.
14 *
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
21 * IN THE SOFTWARE.
22 */
23
24 #include "util/u_format.h"
25 #include "util/u_surface.h"
26 #include "util/u_blitter.h"
27 #include "v3d_context.h"
28 #include "v3d_tiling.h"
29
30 #if 0
31 static struct pipe_surface *
32 v3d_get_blit_surface(struct pipe_context *pctx,
33 struct pipe_resource *prsc, unsigned level)
34 {
35 struct pipe_surface tmpl;
36
37 memset(&tmpl, 0, sizeof(tmpl));
38 tmpl.format = prsc->format;
39 tmpl.u.tex.level = level;
40 tmpl.u.tex.first_layer = 0;
41 tmpl.u.tex.last_layer = 0;
42
43 return pctx->create_surface(pctx, prsc, &tmpl);
44 }
45
46 static bool
47 is_tile_unaligned(unsigned size, unsigned tile_size)
48 {
49 return size & (tile_size - 1);
50 }
51
52 static bool
53 v3d_tile_blit(struct pipe_context *pctx, const struct pipe_blit_info *info)
54 {
55 struct v3d_context *v3d = v3d_context(pctx);
56 bool msaa = (info->src.resource->nr_samples > 1 ||
57 info->dst.resource->nr_samples > 1);
58 int tile_width = msaa ? 32 : 64;
59 int tile_height = msaa ? 32 : 64;
60
61 if (util_format_is_depth_or_stencil(info->dst.resource->format))
62 return false;
63
64 if (info->scissor_enable)
65 return false;
66
67 if ((info->mask & PIPE_MASK_RGBA) == 0)
68 return false;
69
70 if (info->dst.box.x != info->src.box.x ||
71 info->dst.box.y != info->src.box.y ||
72 info->dst.box.width != info->src.box.width ||
73 info->dst.box.height != info->src.box.height) {
74 return false;
75 }
76
77 int dst_surface_width = u_minify(info->dst.resource->width0,
78 info->dst.level);
79 int dst_surface_height = u_minify(info->dst.resource->height0,
80 info->dst.level);
81 if (is_tile_unaligned(info->dst.box.x, tile_width) ||
82 is_tile_unaligned(info->dst.box.y, tile_height) ||
83 (is_tile_unaligned(info->dst.box.width, tile_width) &&
84 info->dst.box.x + info->dst.box.width != dst_surface_width) ||
85 (is_tile_unaligned(info->dst.box.height, tile_height) &&
86 info->dst.box.y + info->dst.box.height != dst_surface_height)) {
87 return false;
88 }
89
90 /* VC5_PACKET_LOAD_TILE_BUFFER_GENERAL uses the
91 * VC5_PACKET_TILE_RENDERING_MODE_CONFIG's width (determined by our
92 * destination surface) to determine the stride. This may be wrong
93 * when reading from texture miplevels > 0, which are stored in
94 * POT-sized areas. For MSAA, the tile addresses are computed
95 * explicitly by the RCL, but still use the destination width to
96 * determine the stride (which could be fixed by explicitly supplying
97 * it in the ABI).
98 */
99 struct v3d_resource *rsc = v3d_resource(info->src.resource);
100
101 uint32_t stride;
102
103 if (info->src.resource->nr_samples > 1)
104 stride = align(dst_surface_width, 32) * 4 * rsc->cpp;
105 /* XXX else if (rsc->slices[info->src.level].tiling == VC5_TILING_FORMAT_T)
106 stride = align(dst_surface_width * rsc->cpp, 128); */
107 else
108 stride = align(dst_surface_width * rsc->cpp, 16);
109
110 if (stride != rsc->slices[info->src.level].stride)
111 return false;
112
113 if (info->dst.resource->format != info->src.resource->format)
114 return false;
115
116 if (false) {
117 fprintf(stderr, "RCL blit from %d,%d to %d,%d (%d,%d)\n",
118 info->src.box.x,
119 info->src.box.y,
120 info->dst.box.x,
121 info->dst.box.y,
122 info->dst.box.width,
123 info->dst.box.height);
124 }
125
126 struct pipe_surface *dst_surf =
127 v3d_get_blit_surface(pctx, info->dst.resource, info->dst.level);
128 struct pipe_surface *src_surf =
129 v3d_get_blit_surface(pctx, info->src.resource, info->src.level);
130
131 v3d_flush_jobs_reading_resource(v3d, info->src.resource);
132
133 struct v3d_job *job = v3d_get_job(v3d, dst_surf, NULL);
134 pipe_surface_reference(&job->color_read, src_surf);
135
136 /* If we're resolving from MSAA to single sample, we still need to run
137 * the engine in MSAA mode for the load.
138 */
139 if (!job->msaa && info->src.resource->nr_samples > 1) {
140 job->msaa = true;
141 job->tile_width = 32;
142 job->tile_height = 32;
143 }
144
145 job->draw_min_x = info->dst.box.x;
146 job->draw_min_y = info->dst.box.y;
147 job->draw_max_x = info->dst.box.x + info->dst.box.width;
148 job->draw_max_y = info->dst.box.y + info->dst.box.height;
149 job->draw_width = dst_surf->width;
150 job->draw_height = dst_surf->height;
151
152 job->tile_width = tile_width;
153 job->tile_height = tile_height;
154 job->msaa = msaa;
155 job->needs_flush = true;
156 job->resolve |= PIPE_CLEAR_COLOR;
157
158 v3d_job_submit(v3d, job);
159
160 pipe_surface_reference(&dst_surf, NULL);
161 pipe_surface_reference(&src_surf, NULL);
162
163 return true;
164 }
165 #endif
166
167 void
168 v3d_blitter_save(struct v3d_context *v3d)
169 {
170 util_blitter_save_fragment_constant_buffer_slot(v3d->blitter,
171 v3d->constbuf[PIPE_SHADER_FRAGMENT].cb);
172 util_blitter_save_vertex_buffer_slot(v3d->blitter, v3d->vertexbuf.vb);
173 util_blitter_save_vertex_elements(v3d->blitter, v3d->vtx);
174 util_blitter_save_vertex_shader(v3d->blitter, v3d->prog.bind_vs);
175 util_blitter_save_so_targets(v3d->blitter, v3d->streamout.num_targets,
176 v3d->streamout.targets);
177 util_blitter_save_rasterizer(v3d->blitter, v3d->rasterizer);
178 util_blitter_save_viewport(v3d->blitter, &v3d->viewport);
179 util_blitter_save_scissor(v3d->blitter, &v3d->scissor);
180 util_blitter_save_fragment_shader(v3d->blitter, v3d->prog.bind_fs);
181 util_blitter_save_blend(v3d->blitter, v3d->blend);
182 util_blitter_save_depth_stencil_alpha(v3d->blitter, v3d->zsa);
183 util_blitter_save_stencil_ref(v3d->blitter, &v3d->stencil_ref);
184 util_blitter_save_sample_mask(v3d->blitter, v3d->sample_mask);
185 util_blitter_save_framebuffer(v3d->blitter, &v3d->framebuffer);
186 util_blitter_save_fragment_sampler_states(v3d->blitter,
187 v3d->tex[PIPE_SHADER_FRAGMENT].num_samplers,
188 (void **)v3d->tex[PIPE_SHADER_FRAGMENT].samplers);
189 util_blitter_save_fragment_sampler_views(v3d->blitter,
190 v3d->tex[PIPE_SHADER_FRAGMENT].num_textures,
191 v3d->tex[PIPE_SHADER_FRAGMENT].textures);
192 util_blitter_save_so_targets(v3d->blitter, v3d->streamout.num_targets,
193 v3d->streamout.targets);
194 }
195
196 static bool
197 v3d_render_blit(struct pipe_context *ctx, struct pipe_blit_info *info)
198 {
199 struct v3d_context *v3d = v3d_context(ctx);
200 struct v3d_resource *src = v3d_resource(info->src.resource);
201 struct pipe_resource *tiled = NULL;
202
203 if (!src->tiled) {
204 struct pipe_box box = {
205 .x = 0,
206 .y = 0,
207 .width = u_minify(info->src.resource->width0,
208 info->src.level),
209 .height = u_minify(info->src.resource->height0,
210 info->src.level),
211 .depth = 1,
212 };
213 struct pipe_resource tmpl = {
214 .target = info->src.resource->target,
215 .format = info->src.resource->format,
216 .width0 = box.width,
217 .height0 = box.height,
218 .depth0 = 1,
219 .array_size = 1,
220 };
221 tiled = ctx->screen->resource_create(ctx->screen, &tmpl);
222 if (!tiled) {
223 fprintf(stderr, "Failed to create tiled blit temp\n");
224 return false;
225 }
226 ctx->resource_copy_region(ctx,
227 tiled, 0,
228 0, 0, 0,
229 info->src.resource, info->src.level,
230 &box);
231 info->src.level = 0;
232 info->src.resource = tiled;
233 }
234
235 if (!util_blitter_is_blit_supported(v3d->blitter, info)) {
236 fprintf(stderr, "blit unsupported %s -> %s\n",
237 util_format_short_name(info->src.resource->format),
238 util_format_short_name(info->dst.resource->format));
239 return false;
240 }
241
242 v3d_blitter_save(v3d);
243 util_blitter_blit(v3d->blitter, info);
244
245 pipe_resource_reference(&tiled, NULL);
246
247 return true;
248 }
249
250 /* Implement stencil blits by reinterpreting the stencil data as an RGBA8888
251 * or R8 texture.
252 */
253 static void
254 v3d_stencil_blit(struct pipe_context *ctx, const struct pipe_blit_info *info)
255 {
256 struct v3d_context *v3d = v3d_context(ctx);
257 struct v3d_resource *src = v3d_resource(info->src.resource);
258 struct v3d_resource *dst = v3d_resource(info->dst.resource);
259 enum pipe_format src_format, dst_format;
260
261 if (src->separate_stencil) {
262 src = src->separate_stencil;
263 src_format = PIPE_FORMAT_R8_UNORM;
264 } else {
265 src_format = PIPE_FORMAT_RGBA8888_UNORM;
266 }
267
268 if (dst->separate_stencil) {
269 dst = dst->separate_stencil;
270 dst_format = PIPE_FORMAT_R8_UNORM;
271 } else {
272 dst_format = PIPE_FORMAT_RGBA8888_UNORM;
273 }
274
275 /* Initialize the surface. */
276 struct pipe_surface dst_tmpl = {
277 .u.tex = {
278 .level = info->dst.level,
279 .first_layer = info->dst.box.z,
280 .last_layer = info->dst.box.z,
281 },
282 .format = dst_format,
283 };
284 struct pipe_surface *dst_surf =
285 ctx->create_surface(ctx, &dst->base, &dst_tmpl);
286
287 /* Initialize the sampler view. */
288 struct pipe_sampler_view src_tmpl = {
289 .target = src->base.target,
290 .format = src_format,
291 .u.tex = {
292 .first_level = info->src.level,
293 .last_level = info->src.level,
294 .first_layer = 0,
295 .last_layer = (PIPE_TEXTURE_3D ?
296 u_minify(src->base.depth0,
297 info->src.level) - 1 :
298 src->base.array_size - 1),
299 },
300 .swizzle_r = PIPE_SWIZZLE_X,
301 .swizzle_g = PIPE_SWIZZLE_Y,
302 .swizzle_b = PIPE_SWIZZLE_Z,
303 .swizzle_a = PIPE_SWIZZLE_W,
304 };
305 struct pipe_sampler_view *src_view =
306 ctx->create_sampler_view(ctx, &src->base, &src_tmpl);
307
308 v3d_blitter_save(v3d);
309 util_blitter_blit_generic(v3d->blitter, dst_surf, &info->dst.box,
310 src_view, &info->src.box,
311 src->base.width0, src->base.height0,
312 PIPE_MASK_R,
313 PIPE_TEX_FILTER_NEAREST,
314 info->scissor_enable ? &info->scissor : NULL,
315 info->alpha_blend);
316
317 pipe_surface_reference(&dst_surf, NULL);
318 pipe_sampler_view_reference(&src_view, NULL);
319 }
320
321 /* Disable level 0 write, just write following mipmaps */
322 #define V3D_TFU_IOA_DIMTW (1 << 0)
323 #define V3D_TFU_IOA_FORMAT_SHIFT 3
324 #define V3D_TFU_IOA_FORMAT_LINEARTILE 3
325 #define V3D_TFU_IOA_FORMAT_UBLINEAR_1_COLUMN 4
326 #define V3D_TFU_IOA_FORMAT_UBLINEAR_2_COLUMN 5
327 #define V3D_TFU_IOA_FORMAT_UIF_NO_XOR 6
328 #define V3D_TFU_IOA_FORMAT_UIF_XOR 7
329
330 #define V3D_TFU_ICFG_NUMMM_SHIFT 5
331 #define V3D_TFU_ICFG_TTYPE_SHIFT 9
332
333 #define V3D_TFU_ICFG_OPAD_SHIFT 22
334
335 #define V3D_TFU_ICFG_FORMAT_SHIFT 18
336 #define V3D_TFU_ICFG_FORMAT_RASTER 0
337 #define V3D_TFU_ICFG_FORMAT_SAND_128 1
338 #define V3D_TFU_ICFG_FORMAT_SAND_256 2
339 #define V3D_TFU_ICFG_FORMAT_LINEARTILE 11
340 #define V3D_TFU_ICFG_FORMAT_UBLINEAR_1_COLUMN 12
341 #define V3D_TFU_ICFG_FORMAT_UBLINEAR_2_COLUMN 13
342 #define V3D_TFU_ICFG_FORMAT_UIF_NO_XOR 14
343 #define V3D_TFU_ICFG_FORMAT_UIF_XOR 15
344
345 static bool
346 v3d_tfu(struct pipe_context *pctx,
347 struct pipe_resource *pdst,
348 struct pipe_resource *psrc,
349 unsigned int src_level,
350 unsigned int base_level,
351 unsigned int last_level,
352 unsigned int src_layer,
353 unsigned int dst_layer)
354 {
355 struct v3d_context *v3d = v3d_context(pctx);
356 struct v3d_screen *screen = v3d->screen;
357 struct v3d_resource *src = v3d_resource(psrc);
358 struct v3d_resource *dst = v3d_resource(pdst);
359 struct v3d_resource_slice *src_base_slice = &src->slices[src_level];
360 struct v3d_resource_slice *dst_base_slice = &dst->slices[base_level];
361 int msaa_scale = pdst->nr_samples > 1 ? 2 : 1;
362 int width = u_minify(pdst->width0, base_level) * msaa_scale;
363 int height = u_minify(pdst->height0, base_level) * msaa_scale;
364
365 if (psrc->format != pdst->format)
366 return false;
367 if (psrc->nr_samples != pdst->nr_samples)
368 return false;
369
370 uint32_t tex_format = v3d_get_tex_format(&screen->devinfo,
371 pdst->format);
372
373 if (!v3d_tfu_supports_tex_format(&screen->devinfo, tex_format))
374 return false;
375
376 if (pdst->target != PIPE_TEXTURE_2D || psrc->target != PIPE_TEXTURE_2D)
377 return false;
378
379 /* Can't write to raster. */
380 if (dst_base_slice->tiling == VC5_TILING_RASTER)
381 return false;
382
383 v3d_flush_jobs_writing_resource(v3d, psrc);
384 v3d_flush_jobs_reading_resource(v3d, pdst);
385
386 struct drm_v3d_submit_tfu tfu = {
387 .ios = (height << 16) | width,
388 .bo_handles = {
389 dst->bo->handle,
390 src != dst ? src->bo->handle : 0
391 },
392 .in_sync = v3d->out_sync,
393 .out_sync = v3d->out_sync,
394 };
395 uint32_t src_offset = (src->bo->offset +
396 v3d_layer_offset(psrc, src_level, src_layer));
397 tfu.iia |= src_offset;
398 if (src_base_slice->tiling == VC5_TILING_RASTER) {
399 tfu.icfg |= (V3D_TFU_ICFG_FORMAT_RASTER <<
400 V3D_TFU_ICFG_FORMAT_SHIFT);
401 } else {
402 tfu.icfg |= ((V3D_TFU_ICFG_FORMAT_LINEARTILE +
403 (src_base_slice->tiling - VC5_TILING_LINEARTILE)) <<
404 V3D_TFU_ICFG_FORMAT_SHIFT);
405 }
406
407 uint32_t dst_offset = (dst->bo->offset +
408 v3d_layer_offset(pdst, src_level, dst_layer));
409 tfu.ioa |= dst_offset;
410 if (last_level != base_level)
411 tfu.ioa |= V3D_TFU_IOA_DIMTW;
412 tfu.ioa |= ((V3D_TFU_IOA_FORMAT_LINEARTILE +
413 (dst_base_slice->tiling - VC5_TILING_LINEARTILE)) <<
414 V3D_TFU_IOA_FORMAT_SHIFT);
415
416 tfu.icfg |= tex_format << V3D_TFU_ICFG_TTYPE_SHIFT;
417 tfu.icfg |= (last_level - base_level) << V3D_TFU_ICFG_NUMMM_SHIFT;
418
419 switch (src_base_slice->tiling) {
420 case VC5_TILING_UIF_NO_XOR:
421 case VC5_TILING_UIF_XOR:
422 tfu.iis |= (src_base_slice->padded_height /
423 (2 * v3d_utile_height(src->cpp)));
424 break;
425 case VC5_TILING_RASTER:
426 tfu.iis |= src_base_slice->stride / src->cpp;
427 break;
428 case VC5_TILING_LINEARTILE:
429 case VC5_TILING_UBLINEAR_1_COLUMN:
430 case VC5_TILING_UBLINEAR_2_COLUMN:
431 break;
432 }
433
434 /* If we're writing level 0 (!IOA_DIMTW), then we need to supply the
435 * OPAD field for the destination (how many extra UIF blocks beyond
436 * those necessary to cover the height). When filling mipmaps, the
437 * miplevel 1+ tiling state is inferred.
438 */
439 if (dst_base_slice->tiling == VC5_TILING_UIF_NO_XOR ||
440 dst_base_slice->tiling == VC5_TILING_UIF_XOR) {
441 int uif_block_h = 2 * v3d_utile_height(dst->cpp);
442 int implicit_padded_height = align(height, uif_block_h);
443
444 tfu.icfg |= (((dst_base_slice->padded_height -
445 implicit_padded_height) / uif_block_h) <<
446 V3D_TFU_ICFG_OPAD_SHIFT);
447 }
448
449 int ret = v3d_ioctl(screen->fd, DRM_IOCTL_V3D_SUBMIT_TFU, &tfu);
450 if (ret != 0) {
451 fprintf(stderr, "Failed to submit TFU job: %d\n", ret);
452 return false;
453 }
454
455 dst->writes++;
456
457 return true;
458 }
459
460 boolean
461 v3d_generate_mipmap(struct pipe_context *pctx,
462 struct pipe_resource *prsc,
463 enum pipe_format format,
464 unsigned int base_level,
465 unsigned int last_level,
466 unsigned int first_layer,
467 unsigned int last_layer)
468 {
469 if (format != prsc->format)
470 return false;
471
472 /* We could maybe support looping over layers for array textures, but
473 * we definitely don't support 3D.
474 */
475 if (first_layer != last_layer)
476 return false;
477
478 return v3d_tfu(pctx,
479 prsc, prsc,
480 base_level,
481 base_level, last_level,
482 first_layer, first_layer);
483 }
484
485 static bool
486 v3d_tfu_blit(struct pipe_context *pctx, const struct pipe_blit_info *info)
487 {
488 int dst_width = u_minify(info->dst.resource->width0, info->dst.level);
489 int dst_height = u_minify(info->dst.resource->height0, info->dst.level);
490
491 if ((info->mask & PIPE_MASK_RGBA) == 0)
492 return false;
493
494 if (info->dst.box.x != 0 ||
495 info->dst.box.y != 0 ||
496 info->dst.box.width != dst_width ||
497 info->dst.box.height != dst_height ||
498 info->src.box.x != 0 ||
499 info->src.box.y != 0 ||
500 info->src.box.width != info->dst.box.width ||
501 info->src.box.height != info->dst.box.height) {
502 return false;
503 }
504
505 if (info->dst.format != info->src.format)
506 return false;
507
508 return v3d_tfu(pctx, info->dst.resource, info->src.resource,
509 info->src.level,
510 info->dst.level, info->dst.level,
511 info->src.box.z, info->dst.box.z);
512 }
513
514 /* Optimal hardware path for blitting pixels.
515 * Scaling, format conversion, up- and downsampling (resolve) are allowed.
516 */
517 void
518 v3d_blit(struct pipe_context *pctx, const struct pipe_blit_info *blit_info)
519 {
520 struct v3d_context *v3d = v3d_context(pctx);
521 struct pipe_blit_info info = *blit_info;
522
523 if (info.mask & PIPE_MASK_S) {
524 v3d_stencil_blit(pctx, blit_info);
525 info.mask &= ~PIPE_MASK_S;
526 }
527
528 if (v3d_tfu_blit(pctx, blit_info))
529 info.mask &= ~PIPE_MASK_RGBA;
530
531 if (info.mask)
532 v3d_render_blit(pctx, &info);
533
534 /* Flush our blit jobs immediately. They're unlikely to get reused by
535 * normal drawing or other blits, and without flushing we can easily
536 * run into unexpected OOMs when blits are used for a large series of
537 * texture uploads before using the textures.
538 */
539 v3d_flush_jobs_writing_resource(v3d, info.dst.resource);
540 }