radeonsi: fix user fence space when MCBP is enabled
[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/format/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, false);
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_geometry_shader(v3d->blitter, v3d->prog.bind_gs);
176 util_blitter_save_so_targets(v3d->blitter, v3d->streamout.num_targets,
177 v3d->streamout.targets);
178 util_blitter_save_rasterizer(v3d->blitter, v3d->rasterizer);
179 util_blitter_save_viewport(v3d->blitter, &v3d->viewport);
180 util_blitter_save_scissor(v3d->blitter, &v3d->scissor);
181 util_blitter_save_fragment_shader(v3d->blitter, v3d->prog.bind_fs);
182 util_blitter_save_blend(v3d->blitter, v3d->blend);
183 util_blitter_save_depth_stencil_alpha(v3d->blitter, v3d->zsa);
184 util_blitter_save_stencil_ref(v3d->blitter, &v3d->stencil_ref);
185 util_blitter_save_sample_mask(v3d->blitter, v3d->sample_mask);
186 util_blitter_save_framebuffer(v3d->blitter, &v3d->framebuffer);
187 util_blitter_save_fragment_sampler_states(v3d->blitter,
188 v3d->tex[PIPE_SHADER_FRAGMENT].num_samplers,
189 (void **)v3d->tex[PIPE_SHADER_FRAGMENT].samplers);
190 util_blitter_save_fragment_sampler_views(v3d->blitter,
191 v3d->tex[PIPE_SHADER_FRAGMENT].num_textures,
192 v3d->tex[PIPE_SHADER_FRAGMENT].textures);
193 util_blitter_save_so_targets(v3d->blitter, v3d->streamout.num_targets,
194 v3d->streamout.targets);
195 }
196
197 static bool
198 v3d_render_blit(struct pipe_context *ctx, struct pipe_blit_info *info)
199 {
200 struct v3d_context *v3d = v3d_context(ctx);
201 struct v3d_resource *src = v3d_resource(info->src.resource);
202 struct pipe_resource *tiled = NULL;
203
204 if (!src->tiled) {
205 struct pipe_box box = {
206 .x = 0,
207 .y = 0,
208 .width = u_minify(info->src.resource->width0,
209 info->src.level),
210 .height = u_minify(info->src.resource->height0,
211 info->src.level),
212 .depth = 1,
213 };
214 struct pipe_resource tmpl = {
215 .target = info->src.resource->target,
216 .format = info->src.resource->format,
217 .width0 = box.width,
218 .height0 = box.height,
219 .depth0 = 1,
220 .array_size = 1,
221 };
222 tiled = ctx->screen->resource_create(ctx->screen, &tmpl);
223 if (!tiled) {
224 fprintf(stderr, "Failed to create tiled blit temp\n");
225 return false;
226 }
227 ctx->resource_copy_region(ctx,
228 tiled, 0,
229 0, 0, 0,
230 info->src.resource, info->src.level,
231 &box);
232 info->src.level = 0;
233 info->src.resource = tiled;
234 }
235
236 if (!util_blitter_is_blit_supported(v3d->blitter, info)) {
237 fprintf(stderr, "blit unsupported %s -> %s\n",
238 util_format_short_name(info->src.resource->format),
239 util_format_short_name(info->dst.resource->format));
240 return false;
241 }
242
243 v3d_blitter_save(v3d);
244 util_blitter_blit(v3d->blitter, info);
245
246 pipe_resource_reference(&tiled, NULL);
247
248 return true;
249 }
250
251 /* Implement stencil blits by reinterpreting the stencil data as an RGBA8888
252 * or R8 texture.
253 */
254 static void
255 v3d_stencil_blit(struct pipe_context *ctx, const struct pipe_blit_info *info)
256 {
257 struct v3d_context *v3d = v3d_context(ctx);
258 struct v3d_resource *src = v3d_resource(info->src.resource);
259 struct v3d_resource *dst = v3d_resource(info->dst.resource);
260 enum pipe_format src_format, dst_format;
261
262 if (src->separate_stencil) {
263 src = src->separate_stencil;
264 src_format = PIPE_FORMAT_R8_UNORM;
265 } else {
266 src_format = PIPE_FORMAT_RGBA8888_UNORM;
267 }
268
269 if (dst->separate_stencil) {
270 dst = dst->separate_stencil;
271 dst_format = PIPE_FORMAT_R8_UNORM;
272 } else {
273 dst_format = PIPE_FORMAT_RGBA8888_UNORM;
274 }
275
276 /* Initialize the surface. */
277 struct pipe_surface dst_tmpl = {
278 .u.tex = {
279 .level = info->dst.level,
280 .first_layer = info->dst.box.z,
281 .last_layer = info->dst.box.z,
282 },
283 .format = dst_format,
284 };
285 struct pipe_surface *dst_surf =
286 ctx->create_surface(ctx, &dst->base, &dst_tmpl);
287
288 /* Initialize the sampler view. */
289 struct pipe_sampler_view src_tmpl = {
290 .target = src->base.target,
291 .format = src_format,
292 .u.tex = {
293 .first_level = info->src.level,
294 .last_level = info->src.level,
295 .first_layer = 0,
296 .last_layer = (PIPE_TEXTURE_3D ?
297 u_minify(src->base.depth0,
298 info->src.level) - 1 :
299 src->base.array_size - 1),
300 },
301 .swizzle_r = PIPE_SWIZZLE_X,
302 .swizzle_g = PIPE_SWIZZLE_Y,
303 .swizzle_b = PIPE_SWIZZLE_Z,
304 .swizzle_a = PIPE_SWIZZLE_W,
305 };
306 struct pipe_sampler_view *src_view =
307 ctx->create_sampler_view(ctx, &src->base, &src_tmpl);
308
309 v3d_blitter_save(v3d);
310 util_blitter_blit_generic(v3d->blitter, dst_surf, &info->dst.box,
311 src_view, &info->src.box,
312 src->base.width0, src->base.height0,
313 PIPE_MASK_R,
314 PIPE_TEX_FILTER_NEAREST,
315 info->scissor_enable ? &info->scissor : NULL,
316 info->alpha_blend);
317
318 pipe_surface_reference(&dst_surf, NULL);
319 pipe_sampler_view_reference(&src_view, NULL);
320 }
321
322 /* Disable level 0 write, just write following mipmaps */
323 #define V3D_TFU_IOA_DIMTW (1 << 0)
324 #define V3D_TFU_IOA_FORMAT_SHIFT 3
325 #define V3D_TFU_IOA_FORMAT_LINEARTILE 3
326 #define V3D_TFU_IOA_FORMAT_UBLINEAR_1_COLUMN 4
327 #define V3D_TFU_IOA_FORMAT_UBLINEAR_2_COLUMN 5
328 #define V3D_TFU_IOA_FORMAT_UIF_NO_XOR 6
329 #define V3D_TFU_IOA_FORMAT_UIF_XOR 7
330
331 #define V3D_TFU_ICFG_NUMMM_SHIFT 5
332 #define V3D_TFU_ICFG_TTYPE_SHIFT 9
333
334 #define V3D_TFU_ICFG_OPAD_SHIFT 22
335
336 #define V3D_TFU_ICFG_FORMAT_SHIFT 18
337 #define V3D_TFU_ICFG_FORMAT_RASTER 0
338 #define V3D_TFU_ICFG_FORMAT_SAND_128 1
339 #define V3D_TFU_ICFG_FORMAT_SAND_256 2
340 #define V3D_TFU_ICFG_FORMAT_LINEARTILE 11
341 #define V3D_TFU_ICFG_FORMAT_UBLINEAR_1_COLUMN 12
342 #define V3D_TFU_ICFG_FORMAT_UBLINEAR_2_COLUMN 13
343 #define V3D_TFU_ICFG_FORMAT_UIF_NO_XOR 14
344 #define V3D_TFU_ICFG_FORMAT_UIF_XOR 15
345
346 static bool
347 v3d_tfu(struct pipe_context *pctx,
348 struct pipe_resource *pdst,
349 struct pipe_resource *psrc,
350 unsigned int src_level,
351 unsigned int base_level,
352 unsigned int last_level,
353 unsigned int src_layer,
354 unsigned int dst_layer)
355 {
356 struct v3d_context *v3d = v3d_context(pctx);
357 struct v3d_screen *screen = v3d->screen;
358 struct v3d_resource *src = v3d_resource(psrc);
359 struct v3d_resource *dst = v3d_resource(pdst);
360 struct v3d_resource_slice *src_base_slice = &src->slices[src_level];
361 struct v3d_resource_slice *dst_base_slice = &dst->slices[base_level];
362 int msaa_scale = pdst->nr_samples > 1 ? 2 : 1;
363 int width = u_minify(pdst->width0, base_level) * msaa_scale;
364 int height = u_minify(pdst->height0, base_level) * msaa_scale;
365
366 if (psrc->format != pdst->format)
367 return false;
368 if (psrc->nr_samples != pdst->nr_samples)
369 return false;
370
371 uint32_t tex_format = v3d_get_tex_format(&screen->devinfo,
372 pdst->format);
373
374 if (!v3d_tfu_supports_tex_format(&screen->devinfo, tex_format))
375 return false;
376
377 if (pdst->target != PIPE_TEXTURE_2D || psrc->target != PIPE_TEXTURE_2D)
378 return false;
379
380 /* Can't write to raster. */
381 if (dst_base_slice->tiling == VC5_TILING_RASTER)
382 return false;
383
384 v3d_flush_jobs_writing_resource(v3d, psrc, V3D_FLUSH_DEFAULT, false);
385 v3d_flush_jobs_reading_resource(v3d, pdst, V3D_FLUSH_DEFAULT, false);
386
387 struct drm_v3d_submit_tfu tfu = {
388 .ios = (height << 16) | width,
389 .bo_handles = {
390 dst->bo->handle,
391 src != dst ? src->bo->handle : 0
392 },
393 .in_sync = v3d->out_sync,
394 .out_sync = v3d->out_sync,
395 };
396 uint32_t src_offset = (src->bo->offset +
397 v3d_layer_offset(psrc, src_level, src_layer));
398 tfu.iia |= src_offset;
399 if (src_base_slice->tiling == VC5_TILING_RASTER) {
400 tfu.icfg |= (V3D_TFU_ICFG_FORMAT_RASTER <<
401 V3D_TFU_ICFG_FORMAT_SHIFT);
402 } else {
403 tfu.icfg |= ((V3D_TFU_ICFG_FORMAT_LINEARTILE +
404 (src_base_slice->tiling - VC5_TILING_LINEARTILE)) <<
405 V3D_TFU_ICFG_FORMAT_SHIFT);
406 }
407
408 uint32_t dst_offset = (dst->bo->offset +
409 v3d_layer_offset(pdst, src_level, dst_layer));
410 tfu.ioa |= dst_offset;
411 if (last_level != base_level)
412 tfu.ioa |= V3D_TFU_IOA_DIMTW;
413 tfu.ioa |= ((V3D_TFU_IOA_FORMAT_LINEARTILE +
414 (dst_base_slice->tiling - VC5_TILING_LINEARTILE)) <<
415 V3D_TFU_IOA_FORMAT_SHIFT);
416
417 tfu.icfg |= tex_format << V3D_TFU_ICFG_TTYPE_SHIFT;
418 tfu.icfg |= (last_level - base_level) << V3D_TFU_ICFG_NUMMM_SHIFT;
419
420 switch (src_base_slice->tiling) {
421 case VC5_TILING_UIF_NO_XOR:
422 case VC5_TILING_UIF_XOR:
423 tfu.iis |= (src_base_slice->padded_height /
424 (2 * v3d_utile_height(src->cpp)));
425 break;
426 case VC5_TILING_RASTER:
427 tfu.iis |= src_base_slice->stride / src->cpp;
428 break;
429 case VC5_TILING_LINEARTILE:
430 case VC5_TILING_UBLINEAR_1_COLUMN:
431 case VC5_TILING_UBLINEAR_2_COLUMN:
432 break;
433 }
434
435 /* If we're writing level 0 (!IOA_DIMTW), then we need to supply the
436 * OPAD field for the destination (how many extra UIF blocks beyond
437 * those necessary to cover the height). When filling mipmaps, the
438 * miplevel 1+ tiling state is inferred.
439 */
440 if (dst_base_slice->tiling == VC5_TILING_UIF_NO_XOR ||
441 dst_base_slice->tiling == VC5_TILING_UIF_XOR) {
442 int uif_block_h = 2 * v3d_utile_height(dst->cpp);
443 int implicit_padded_height = align(height, uif_block_h);
444
445 tfu.icfg |= (((dst_base_slice->padded_height -
446 implicit_padded_height) / uif_block_h) <<
447 V3D_TFU_ICFG_OPAD_SHIFT);
448 }
449
450 int ret = v3d_ioctl(screen->fd, DRM_IOCTL_V3D_SUBMIT_TFU, &tfu);
451 if (ret != 0) {
452 fprintf(stderr, "Failed to submit TFU job: %d\n", ret);
453 return false;
454 }
455
456 dst->writes++;
457
458 return true;
459 }
460
461 bool
462 v3d_generate_mipmap(struct pipe_context *pctx,
463 struct pipe_resource *prsc,
464 enum pipe_format format,
465 unsigned int base_level,
466 unsigned int last_level,
467 unsigned int first_layer,
468 unsigned int last_layer)
469 {
470 if (format != prsc->format)
471 return false;
472
473 /* We could maybe support looping over layers for array textures, but
474 * we definitely don't support 3D.
475 */
476 if (first_layer != last_layer)
477 return false;
478
479 return v3d_tfu(pctx,
480 prsc, prsc,
481 base_level,
482 base_level, last_level,
483 first_layer, first_layer);
484 }
485
486 static bool
487 v3d_tfu_blit(struct pipe_context *pctx, const struct pipe_blit_info *info)
488 {
489 int dst_width = u_minify(info->dst.resource->width0, info->dst.level);
490 int dst_height = u_minify(info->dst.resource->height0, info->dst.level);
491
492 if ((info->mask & PIPE_MASK_RGBA) == 0)
493 return false;
494
495 if (info->scissor_enable ||
496 info->dst.box.x != 0 ||
497 info->dst.box.y != 0 ||
498 info->dst.box.width != dst_width ||
499 info->dst.box.height != dst_height ||
500 info->src.box.x != 0 ||
501 info->src.box.y != 0 ||
502 info->src.box.width != info->dst.box.width ||
503 info->src.box.height != info->dst.box.height) {
504 return false;
505 }
506
507 if (info->dst.format != info->src.format)
508 return false;
509
510 return v3d_tfu(pctx, info->dst.resource, info->src.resource,
511 info->src.level,
512 info->dst.level, info->dst.level,
513 info->src.box.z, info->dst.box.z);
514 }
515
516 /* Optimal hardware path for blitting pixels.
517 * Scaling, format conversion, up- and downsampling (resolve) are allowed.
518 */
519 void
520 v3d_blit(struct pipe_context *pctx, const struct pipe_blit_info *blit_info)
521 {
522 struct v3d_context *v3d = v3d_context(pctx);
523 struct pipe_blit_info info = *blit_info;
524
525 if (info.mask & PIPE_MASK_S) {
526 v3d_stencil_blit(pctx, blit_info);
527 info.mask &= ~PIPE_MASK_S;
528 }
529
530 if (v3d_tfu_blit(pctx, blit_info))
531 info.mask &= ~PIPE_MASK_RGBA;
532
533 if (info.mask)
534 v3d_render_blit(pctx, &info);
535
536 /* Flush our blit jobs immediately. They're unlikely to get reused by
537 * normal drawing or other blits, and without flushing we can easily
538 * run into unexpected OOMs when blits are used for a large series of
539 * texture uploads before using the textures.
540 */
541 v3d_flush_jobs_writing_resource(v3d, info.dst.resource,
542 V3D_FLUSH_DEFAULT, false);
543 }