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 "compiler/nir/nir_builder.h"
28#include "v3d_context.h"
29#include "broadcom/common/v3d_tiling.h"
30#include "broadcom/common/v3d_tfu.h"
31
32/**
33 * The param @op_blit is used to tell if we are saving state for blitter_blit
34 * (if true) or blitter_clear (if false). If other blitter functions are used
35 * that require different state we may need something more elaborated than
36 * this.
37 */
38
39void
40v3d_blitter_save(struct v3d_context *v3d, bool op_blit)
41{
42        util_blitter_save_fragment_constant_buffer_slot(v3d->blitter,
43                                                        v3d->constbuf[PIPE_SHADER_FRAGMENT].cb);
44        util_blitter_save_vertex_buffer_slot(v3d->blitter, v3d->vertexbuf.vb);
45        util_blitter_save_vertex_elements(v3d->blitter, v3d->vtx);
46        util_blitter_save_vertex_shader(v3d->blitter, v3d->prog.bind_vs);
47        util_blitter_save_geometry_shader(v3d->blitter, v3d->prog.bind_gs);
48        util_blitter_save_so_targets(v3d->blitter, v3d->streamout.num_targets,
49                                     v3d->streamout.targets);
50        util_blitter_save_rasterizer(v3d->blitter, v3d->rasterizer);
51        util_blitter_save_viewport(v3d->blitter, &v3d->viewport);
52        util_blitter_save_fragment_shader(v3d->blitter, v3d->prog.bind_fs);
53        util_blitter_save_blend(v3d->blitter, v3d->blend);
54        util_blitter_save_depth_stencil_alpha(v3d->blitter, v3d->zsa);
55        util_blitter_save_stencil_ref(v3d->blitter, &v3d->stencil_ref);
56        util_blitter_save_sample_mask(v3d->blitter, v3d->sample_mask, 0);
57        util_blitter_save_so_targets(v3d->blitter, v3d->streamout.num_targets,
58                                     v3d->streamout.targets);
59
60        if (op_blit) {
61                util_blitter_save_scissor(v3d->blitter, &v3d->scissor);
62                util_blitter_save_framebuffer(v3d->blitter, &v3d->framebuffer);
63                util_blitter_save_fragment_sampler_states(v3d->blitter,
64                                                          v3d->tex[PIPE_SHADER_FRAGMENT].num_samplers,
65                                                          (void **)v3d->tex[PIPE_SHADER_FRAGMENT].samplers);
66                util_blitter_save_fragment_sampler_views(v3d->blitter,
67                                                         v3d->tex[PIPE_SHADER_FRAGMENT].num_textures,
68                                                         v3d->tex[PIPE_SHADER_FRAGMENT].textures);
69        }
70}
71
72static void
73v3d_render_blit(struct pipe_context *ctx, struct pipe_blit_info *info)
74{
75        struct v3d_context *v3d = v3d_context(ctx);
76        struct v3d_resource *src = v3d_resource(info->src.resource);
77        struct pipe_resource *tiled = NULL;
78
79        if (!info->mask)
80                return;
81
82        if (!src->tiled &&
83            info->src.resource->target != PIPE_TEXTURE_1D &&
84            info->src.resource->target != PIPE_TEXTURE_1D_ARRAY) {
85                struct pipe_box box = {
86                        .x = 0,
87                        .y = 0,
88                        .width = u_minify(info->src.resource->width0,
89                                           info->src.level),
90                        .height = u_minify(info->src.resource->height0,
91                                           info->src.level),
92                        .depth = 1,
93                };
94                struct pipe_resource tmpl = {
95                        .target = info->src.resource->target,
96                        .format = info->src.resource->format,
97                        .width0 = box.width,
98                        .height0 = box.height,
99                        .depth0 = 1,
100                        .array_size = 1,
101                };
102                tiled = ctx->screen->resource_create(ctx->screen, &tmpl);
103                if (!tiled) {
104                        fprintf(stderr, "Failed to create tiled blit temp\n");
105                        return;
106                }
107                ctx->resource_copy_region(ctx,
108                                          tiled, 0,
109                                          0, 0, 0,
110                                          info->src.resource, info->src.level,
111                                          &box);
112                info->src.level = 0;
113                info->src.resource = tiled;
114        }
115
116        if (!util_blitter_is_blit_supported(v3d->blitter, info)) {
117                fprintf(stderr, "blit unsupported %s -> %s\n",
118                    util_format_short_name(info->src.format),
119                    util_format_short_name(info->dst.format));
120                return;
121        }
122
123        v3d_blitter_save(v3d, true);
124        util_blitter_blit(v3d->blitter, info);
125
126        pipe_resource_reference(&tiled, NULL);
127        info->mask = 0;
128}
129
130/* Implement stencil blits by reinterpreting the stencil data as an RGBA8888
131 * or R8 texture.
132 */
133static void
134v3d_stencil_blit(struct pipe_context *ctx, struct pipe_blit_info *info)
135{
136        struct v3d_context *v3d = v3d_context(ctx);
137        struct v3d_resource *src = v3d_resource(info->src.resource);
138        struct v3d_resource *dst = v3d_resource(info->dst.resource);
139        enum pipe_format src_format, dst_format;
140
141        if ((info->mask & PIPE_MASK_S) == 0)
142                return;
143
144        if (src->separate_stencil) {
145                src = src->separate_stencil;
146                src_format = PIPE_FORMAT_R8_UINT;
147        } else {
148                src_format = PIPE_FORMAT_RGBA8888_UINT;
149        }
150
151        if (dst->separate_stencil) {
152                dst = dst->separate_stencil;
153                dst_format = PIPE_FORMAT_R8_UINT;
154        } else {
155                dst_format = PIPE_FORMAT_RGBA8888_UINT;
156        }
157
158        /* Initialize the surface. */
159        struct pipe_surface dst_tmpl = {
160                .u.tex = {
161                        .level = info->dst.level,
162                        .first_layer = info->dst.box.z,
163                        .last_layer = info->dst.box.z,
164                },
165                .format = dst_format,
166        };
167        struct pipe_surface *dst_surf =
168                ctx->create_surface(ctx, &dst->base, &dst_tmpl);
169
170        /* Initialize the sampler view. */
171        struct pipe_sampler_view src_tmpl = {
172                .target = src->base.target,
173                .format = src_format,
174                .u.tex = {
175                        .first_level = info->src.level,
176                        .last_level = info->src.level,
177                        .first_layer = 0,
178                        .last_layer = (PIPE_TEXTURE_3D ?
179                                       u_minify(src->base.depth0,
180                                                info->src.level) - 1 :
181                                       src->base.array_size - 1),
182                },
183                .swizzle_r = PIPE_SWIZZLE_X,
184                .swizzle_g = PIPE_SWIZZLE_Y,
185                .swizzle_b = PIPE_SWIZZLE_Z,
186                .swizzle_a = PIPE_SWIZZLE_W,
187        };
188        struct pipe_sampler_view *src_view =
189                ctx->create_sampler_view(ctx, &src->base, &src_tmpl);
190
191        v3d_blitter_save(v3d, true);
192        util_blitter_blit_generic(v3d->blitter, dst_surf, &info->dst.box,
193                                  src_view, &info->src.box,
194                                  src->base.width0, src->base.height0,
195                                  PIPE_MASK_R,
196                                  PIPE_TEX_FILTER_NEAREST,
197                                  info->scissor_enable ? &info->scissor : NULL,
198                                  info->alpha_blend, false, 0);
199
200        pipe_surface_reference(&dst_surf, NULL);
201        pipe_sampler_view_reference(&src_view, NULL);
202
203        info->mask &= ~PIPE_MASK_S;
204}
205
206static bool
207v3d_tfu(struct pipe_context *pctx,
208        struct pipe_resource *pdst,
209        struct pipe_resource *psrc,
210        unsigned int src_level,
211        unsigned int base_level,
212        unsigned int last_level,
213        unsigned int src_layer,
214        unsigned int dst_layer,
215        bool for_mipmap)
216{
217        struct v3d_context *v3d = v3d_context(pctx);
218        struct v3d_screen *screen = v3d->screen;
219        struct v3d_resource *src = v3d_resource(psrc);
220        struct v3d_resource *dst = v3d_resource(pdst);
221        struct v3d_resource_slice *src_base_slice = &src->slices[src_level];
222        struct v3d_resource_slice *dst_base_slice = &dst->slices[base_level];
223        int msaa_scale = pdst->nr_samples > 1 ? 2 : 1;
224        int width = u_minify(pdst->width0, base_level) * msaa_scale;
225        int height = u_minify(pdst->height0, base_level) * msaa_scale;
226        enum pipe_format pformat;
227
228        if (psrc->format != pdst->format)
229                return false;
230        if (psrc->nr_samples != pdst->nr_samples)
231                return false;
232
233        /* Can't write to raster. */
234        if (dst_base_slice->tiling == V3D_TILING_RASTER)
235                return false;
236
237        /* When using TFU for blit, we are doing exact copies (both input and
238         * output format must be the same, no scaling, etc), so there is no
239         * pixel format conversions. Thus we can rewrite the format to use one
240         * that is TFU compatible based on its texel size.
241         */
242        if (for_mipmap) {
243                pformat = pdst->format;
244        } else {
245                switch (dst->cpp) {
246                case 16: pformat = PIPE_FORMAT_R32G32B32A32_FLOAT;   break;
247                case 8:  pformat = PIPE_FORMAT_R16G16B16A16_FLOAT;   break;
248                case 4:  pformat = PIPE_FORMAT_R32_FLOAT;            break;
249                case 2:  pformat = PIPE_FORMAT_R16_FLOAT;            break;
250                case 1:  pformat = PIPE_FORMAT_R8_UNORM;             break;
251                default: unreachable("unsupported format bit-size"); break;
252                };
253        }
254
255        uint32_t tex_format = v3d_get_tex_format(&screen->devinfo, pformat);
256
257        if (!v3d_tfu_supports_tex_format(&screen->devinfo, tex_format, for_mipmap)) {
258                assert(for_mipmap);
259                return false;
260        }
261
262        v3d_flush_jobs_writing_resource(v3d, psrc, V3D_FLUSH_DEFAULT, false);
263        v3d_flush_jobs_reading_resource(v3d, pdst, V3D_FLUSH_DEFAULT, false);
264
265        struct drm_v3d_submit_tfu tfu = {
266                .ios = (height << 16) | width,
267                .bo_handles = {
268                        dst->bo->handle,
269                        src != dst ? src->bo->handle : 0
270                },
271                .in_sync = v3d->out_sync,
272                .out_sync = v3d->out_sync,
273        };
274        uint32_t src_offset = (src->bo->offset +
275                               v3d_layer_offset(psrc, src_level, src_layer));
276        tfu.iia |= src_offset;
277        if (src_base_slice->tiling == V3D_TILING_RASTER) {
278                tfu.icfg |= (V3D33_TFU_ICFG_FORMAT_RASTER <<
279                             V3D33_TFU_ICFG_FORMAT_SHIFT);
280        } else {
281                tfu.icfg |= ((V3D33_TFU_ICFG_FORMAT_LINEARTILE +
282                              (src_base_slice->tiling - V3D_TILING_LINEARTILE)) <<
283                             V3D33_TFU_ICFG_FORMAT_SHIFT);
284        }
285
286        uint32_t dst_offset = (dst->bo->offset +
287                               v3d_layer_offset(pdst, base_level, dst_layer));
288        tfu.ioa |= dst_offset;
289        if (last_level != base_level)
290                tfu.ioa |= V3D33_TFU_IOA_DIMTW;
291        tfu.ioa |= ((V3D33_TFU_IOA_FORMAT_LINEARTILE +
292                     (dst_base_slice->tiling - V3D_TILING_LINEARTILE)) <<
293                    V3D33_TFU_IOA_FORMAT_SHIFT);
294
295        tfu.icfg |= tex_format << V3D33_TFU_ICFG_TTYPE_SHIFT;
296        tfu.icfg |= (last_level - base_level) << V3D33_TFU_ICFG_NUMMM_SHIFT;
297
298        switch (src_base_slice->tiling) {
299        case V3D_TILING_UIF_NO_XOR:
300        case V3D_TILING_UIF_XOR:
301                tfu.iis |= (src_base_slice->padded_height /
302                            (2 * v3d_utile_height(src->cpp)));
303                break;
304        case V3D_TILING_RASTER:
305                tfu.iis |= src_base_slice->stride / src->cpp;
306                break;
307        case V3D_TILING_LINEARTILE:
308        case V3D_TILING_UBLINEAR_1_COLUMN:
309        case V3D_TILING_UBLINEAR_2_COLUMN:
310                break;
311       }
312
313        /* If we're writing level 0 (!IOA_DIMTW), then we need to supply the
314         * OPAD field for the destination (how many extra UIF blocks beyond
315         * those necessary to cover the height).  When filling mipmaps, the
316         * miplevel 1+ tiling state is inferred.
317         */
318        if (dst_base_slice->tiling == V3D_TILING_UIF_NO_XOR ||
319            dst_base_slice->tiling == V3D_TILING_UIF_XOR) {
320                int uif_block_h = 2 * v3d_utile_height(dst->cpp);
321                int implicit_padded_height = align(height, uif_block_h);
322
323                tfu.icfg |= (((dst_base_slice->padded_height -
324                               implicit_padded_height) / uif_block_h) <<
325                             V3D33_TFU_ICFG_OPAD_SHIFT);
326        }
327
328        int ret = v3d_ioctl(screen->fd, DRM_IOCTL_V3D_SUBMIT_TFU, &tfu);
329        if (ret != 0) {
330                fprintf(stderr, "Failed to submit TFU job: %d\n", ret);
331                return false;
332        }
333
334        dst->writes++;
335
336        return true;
337}
338
339bool
340v3d_generate_mipmap(struct pipe_context *pctx,
341                    struct pipe_resource *prsc,
342                    enum pipe_format format,
343                    unsigned int base_level,
344                    unsigned int last_level,
345                    unsigned int first_layer,
346                    unsigned int last_layer)
347{
348        if (format != prsc->format)
349                return false;
350
351        /* We could maybe support looping over layers for array textures, but
352         * we definitely don't support 3D.
353         */
354        if (first_layer != last_layer)
355                return false;
356
357        return v3d_tfu(pctx,
358                       prsc, prsc,
359                       base_level,
360                       base_level, last_level,
361                       first_layer, first_layer,
362                       true);
363}
364
365static void
366v3d_tfu_blit(struct pipe_context *pctx, struct pipe_blit_info *info)
367{
368        int dst_width = u_minify(info->dst.resource->width0, info->dst.level);
369        int dst_height = u_minify(info->dst.resource->height0, info->dst.level);
370
371        if ((info->mask & PIPE_MASK_RGBA) == 0)
372                return;
373
374        if (info->scissor_enable ||
375            info->dst.box.x != 0 ||
376            info->dst.box.y != 0 ||
377            info->dst.box.width != dst_width ||
378            info->dst.box.height != dst_height ||
379            info->dst.box.depth != 1 ||
380            info->src.box.x != 0 ||
381            info->src.box.y != 0 ||
382            info->src.box.width != info->dst.box.width ||
383            info->src.box.height != info->dst.box.height ||
384            info->src.box.depth != 1) {
385                return;
386        }
387
388        if (info->dst.format != info->src.format)
389                return;
390
391        if (v3d_tfu(pctx, info->dst.resource, info->src.resource,
392                    info->src.level,
393                    info->dst.level, info->dst.level,
394                    info->src.box.z, info->dst.box.z,
395                    false)) {
396                info->mask &= ~PIPE_MASK_RGBA;
397        }
398}
399
400static struct pipe_surface *
401v3d_get_blit_surface(struct pipe_context *pctx,
402                     struct pipe_resource *prsc,
403                     enum pipe_format format,
404                     unsigned level,
405                     int16_t layer)
406{
407        struct pipe_surface tmpl;
408
409        tmpl.format = format;
410        tmpl.u.tex.level = level;
411        tmpl.u.tex.first_layer = layer;
412        tmpl.u.tex.last_layer = layer;
413
414        return pctx->create_surface(pctx, prsc, &tmpl);
415}
416
417static bool
418is_tile_unaligned(unsigned size, unsigned tile_size)
419{
420        return size & (tile_size - 1);
421}
422
423static void
424v3d_tlb_blit(struct pipe_context *pctx, struct pipe_blit_info *info)
425{
426        struct v3d_context *v3d = v3d_context(pctx);
427        struct v3d_screen *screen = v3d->screen;
428
429        if (screen->devinfo.ver < 40 || !info->mask)
430                return;
431
432        bool is_color_blit = info->mask & PIPE_MASK_RGBA;
433        bool is_depth_blit = info->mask & PIPE_MASK_Z;
434        bool is_stencil_blit = info->mask & PIPE_MASK_S;
435
436        /* We should receive either a depth/stencil blit, or color blit, but
437         * not both.
438         */
439        assert ((is_color_blit && !is_depth_blit && !is_stencil_blit) ||
440                (!is_color_blit && (is_depth_blit || is_stencil_blit)));
441
442        if (info->scissor_enable)
443                return;
444
445        if (info->src.box.x != info->dst.box.x ||
446            info->src.box.y != info->dst.box.y ||
447            info->src.box.width != info->dst.box.width ||
448            info->src.box.height != info->dst.box.height)
449                return;
450
451        if (is_color_blit &&
452            util_format_is_depth_or_stencil(info->dst.format))
453                return;
454
455        if (!v3d_rt_format_supported(&screen->devinfo, info->src.format))
456                return;
457
458        if (v3d_get_rt_format(&screen->devinfo, info->src.format) !=
459            v3d_get_rt_format(&screen->devinfo, info->dst.format))
460                return;
461
462        bool msaa = (info->src.resource->nr_samples > 1 ||
463                     info->dst.resource->nr_samples > 1);
464        bool is_msaa_resolve = (info->src.resource->nr_samples > 1 &&
465                                info->dst.resource->nr_samples < 2);
466
467        if (is_msaa_resolve &&
468            !v3d_format_supports_tlb_msaa_resolve(&screen->devinfo, info->src.format))
469                return;
470
471        v3d_flush_jobs_writing_resource(v3d, info->src.resource, V3D_FLUSH_DEFAULT, false);
472
473        struct pipe_surface *dst_surf =
474           v3d_get_blit_surface(pctx, info->dst.resource, info->dst.format, info->dst.level, info->dst.box.z);
475        struct pipe_surface *src_surf =
476           v3d_get_blit_surface(pctx, info->src.resource, info->src.format, info->src.level, info->src.box.z);
477
478        struct pipe_surface *surfaces[V3D_MAX_DRAW_BUFFERS] = { 0 };
479        if (is_color_blit)
480                surfaces[0] = dst_surf;
481
482        bool double_buffer =
483                 unlikely(V3D_DEBUG & V3D_DEBUG_DOUBLE_BUFFER) && !msaa;
484
485        uint32_t tile_width, tile_height, max_bpp;
486        v3d_get_tile_buffer_size(msaa, double_buffer,
487                                 is_color_blit ? 1 : 0, surfaces, src_surf,
488                                 &tile_width, &tile_height, &max_bpp);
489
490        int dst_surface_width = u_minify(info->dst.resource->width0,
491                                         info->dst.level);
492        int dst_surface_height = u_minify(info->dst.resource->height0,
493                                         info->dst.level);
494        if (is_tile_unaligned(info->dst.box.x, tile_width) ||
495            is_tile_unaligned(info->dst.box.y, tile_height) ||
496            (is_tile_unaligned(info->dst.box.width, tile_width) &&
497             info->dst.box.x + info->dst.box.width != dst_surface_width) ||
498            (is_tile_unaligned(info->dst.box.height, tile_height) &&
499             info->dst.box.y + info->dst.box.height != dst_surface_height)) {
500                pipe_surface_reference(&dst_surf, NULL);
501                pipe_surface_reference(&src_surf, NULL);
502                return;
503        }
504
505        struct v3d_job *job = v3d_get_job(v3d,
506                                          is_color_blit ? 1u : 0u,
507                                          surfaces,
508                                          is_color_blit ? NULL : dst_surf,
509                                          src_surf);
510        job->msaa = msaa;
511        job->double_buffer = double_buffer;
512        job->tile_width = tile_width;
513        job->tile_height = tile_height;
514        job->internal_bpp = max_bpp;
515        job->draw_min_x = info->dst.box.x;
516        job->draw_min_y = info->dst.box.y;
517        job->draw_max_x = info->dst.box.x + info->dst.box.width;
518        job->draw_max_y = info->dst.box.y + info->dst.box.height;
519        job->scissor.disabled = false;
520
521        /* The simulator complains if we do a TLB load from a source with a
522         * stride that is smaller than the destination's, so we program the
523         * 'frame region' to match the smallest dimensions of the two surfaces.
524         * This should be fine because we only get here if the src and dst boxes
525         * match, so we know the blit involves the same tiles on both surfaces.
526         */
527        job->draw_width = MIN2(dst_surf->width, src_surf->width);
528        job->draw_height = MIN2(dst_surf->height, src_surf->height);
529        job->draw_tiles_x = DIV_ROUND_UP(job->draw_width,
530                                         job->tile_width);
531        job->draw_tiles_y = DIV_ROUND_UP(job->draw_height,
532                                         job->tile_height);
533
534        job->needs_flush = true;
535        job->num_layers = info->dst.box.depth;
536
537        job->store = 0;
538        if (is_color_blit) {
539                job->store |= PIPE_CLEAR_COLOR0;
540                info->mask &= ~PIPE_MASK_RGBA;
541        }
542        if (is_depth_blit) {
543                job->store |= PIPE_CLEAR_DEPTH;
544                info->mask &= ~PIPE_MASK_Z;
545        }
546        if (is_stencil_blit){
547                job->store |= PIPE_CLEAR_STENCIL;
548                info->mask &= ~PIPE_MASK_S;
549        }
550
551        v3d41_start_binning(v3d, job);
552
553        v3d_job_submit(v3d, job);
554
555        pipe_surface_reference(&dst_surf, NULL);
556        pipe_surface_reference(&src_surf, NULL);
557}
558
559/**
560 * Creates the VS of the custom blit shader to convert YUV plane from
561 * the NV12 format with BROADCOM_SAND_COL128 modifier to UIF tiled format.
562 * This vertex shader is mostly a pass-through VS.
563 */
564static void *
565v3d_get_sand8_vs(struct pipe_context *pctx)
566{
567        struct v3d_context *v3d = v3d_context(pctx);
568        struct pipe_screen *pscreen = pctx->screen;
569
570        if (v3d->sand8_blit_vs)
571                return v3d->sand8_blit_vs;
572
573        const struct nir_shader_compiler_options *options =
574                pscreen->get_compiler_options(pscreen,
575                                              PIPE_SHADER_IR_NIR,
576                                              PIPE_SHADER_VERTEX);
577
578        nir_builder b = nir_builder_init_simple_shader(MESA_SHADER_VERTEX,
579                                                       options,
580                                                       "sand8_blit_vs");
581
582        const struct glsl_type *vec4 = glsl_vec4_type();
583        nir_variable *pos_in = nir_variable_create(b.shader,
584                                                   nir_var_shader_in,
585                                                   vec4, "pos");
586
587        nir_variable *pos_out = nir_variable_create(b.shader,
588                                                    nir_var_shader_out,
589                                                    vec4, "gl_Position");
590        pos_out->data.location = VARYING_SLOT_POS;
591        nir_store_var(&b, pos_out, nir_load_var(&b, pos_in), 0xf);
592
593        struct pipe_shader_state shader_tmpl = {
594                .type = PIPE_SHADER_IR_NIR,
595                .ir.nir = b.shader,
596        };
597
598        v3d->sand8_blit_vs = pctx->create_vs_state(pctx, &shader_tmpl);
599
600        return v3d->sand8_blit_vs;
601}
602/**
603 * Creates the FS of the custom blit shader to convert YUV plane from
604 * the NV12 format with BROADCOM_SAND_COL128 modifier to UIF tiled format.
605 * The result texture is equivalent to a chroma (cpp=2) or luma (cpp=1)
606 * plane for a NV12 format without the SAND modifier.
607 */
608static void *
609v3d_get_sand8_fs(struct pipe_context *pctx, int cpp)
610{
611        struct v3d_context *v3d = v3d_context(pctx);
612        struct pipe_screen *pscreen = pctx->screen;
613        struct pipe_shader_state **cached_shader;
614        const char *name;
615
616        if (cpp == 1) {
617                cached_shader = &v3d->sand8_blit_fs_luma;
618                name = "sand8_blit_fs_luma";
619        } else {
620                cached_shader = &v3d->sand8_blit_fs_chroma;
621                name = "sand8_blit_fs_chroma";
622        }
623
624        if (*cached_shader)
625                return *cached_shader;
626
627        const struct nir_shader_compiler_options *options =
628                pscreen->get_compiler_options(pscreen,
629                                              PIPE_SHADER_IR_NIR,
630                                              PIPE_SHADER_FRAGMENT);
631
632        nir_builder b = nir_builder_init_simple_shader(MESA_SHADER_FRAGMENT,
633                                                       options, "%s", name);
634        const struct glsl_type *vec4 = glsl_vec4_type();
635
636        const struct glsl_type *glsl_int = glsl_int_type();
637
638        nir_variable *color_out =
639                nir_variable_create(b.shader, nir_var_shader_out,
640                                    vec4, "f_color");
641        color_out->data.location = FRAG_RESULT_COLOR;
642
643        nir_variable *pos_in =
644                nir_variable_create(b.shader, nir_var_shader_in, vec4, "pos");
645        pos_in->data.location = VARYING_SLOT_POS;
646        nir_ssa_def *pos = nir_load_var(&b, pos_in);
647
648        nir_ssa_def *zero = nir_imm_int(&b, 0);
649        nir_ssa_def *one = nir_imm_int(&b, 1);
650        nir_ssa_def *two = nir_imm_int(&b, 2);
651        nir_ssa_def *six = nir_imm_int(&b, 6);
652        nir_ssa_def *seven = nir_imm_int(&b, 7);
653        nir_ssa_def *eight = nir_imm_int(&b, 8);
654
655        nir_ssa_def *x = nir_f2i32(&b, nir_channel(&b, pos, 0));
656        nir_ssa_def *y = nir_f2i32(&b, nir_channel(&b, pos, 1));
657
658        nir_variable *stride_in =
659                nir_variable_create(b.shader, nir_var_uniform, glsl_int,
660                                    "sand8_stride");
661        nir_ssa_def *stride =
662                nir_load_uniform(&b, 1, 32, zero,
663                                 .base = stride_in->data.driver_location,
664                                 .range = 1,
665                                 .dest_type = nir_type_int32);
666
667        nir_ssa_def *x_offset;
668        nir_ssa_def *y_offset;
669
670        /* UIF tiled format is composed by UIF blocks, Each block has
671         * four 64 byte microtiles. Inside each microtile pixels are stored
672         * in raster format. But microtiles have different dimensions
673         * based in the bits per pixel of the image.
674         *
675         *   8bpp microtile dimensions are 8x8
676         *  16bpp microtile dimensions are 8x4
677         *  32bpp microtile dimensions are 4x4
678         *
679         * As we are reading and writing with 32bpp to optimize
680         * the number of texture operations during the blit. We need
681         * to adjust the offsets were we read and write as data will
682         * be later read using 8bpp (luma) and 16bpp (chroma).
683         *
684         * For chroma 8x4 16bpp raster order is compatible with 4x4
685         * 32bpp. In both layouts each line has 8*2 == 4*4 == 16 bytes.
686         * But luma 8x8 8bpp raster order is not compatible
687         * with 4x4 32bpp. 8bpp has 8 bytes per line, and 32bpp has
688         * 16 bytes per line. So if we read a 8bpp texture that was
689         * written as 32bpp texture. Bytes would be misplaced.
690         *
691         * inter/intra_utile_x_offests takes care of mapping the offsets
692         * between microtiles to deal with this issue for luma planes.
693         */
694        if (cpp == 1) {
695                nir_ssa_def *intra_utile_x_offset =
696                        nir_ishl(&b, nir_iand_imm(&b, x, 1), two);
697                nir_ssa_def *inter_utile_x_offset =
698                        nir_ishl(&b, nir_iand_imm(&b, x, 60), one);
699                nir_ssa_def *stripe_offset=
700                        nir_ishl(&b,nir_imul(&b,nir_ishr_imm(&b, x, 6),
701                                             stride),
702                                 seven);
703
704                x_offset = nir_iadd(&b, stripe_offset,
705                                        nir_iadd(&b, intra_utile_x_offset,
706                                                     inter_utile_x_offset));
707                y_offset = nir_iadd(&b,
708                                    nir_ishl(&b, nir_iand_imm(&b, x, 2), six),
709                                    nir_ishl(&b, y, eight));
710        } else  {
711                nir_ssa_def *stripe_offset=
712                        nir_ishl(&b,nir_imul(&b,nir_ishr_imm(&b, x, 5),
713                                                stride),
714                                 seven);
715                x_offset = nir_iadd(&b, stripe_offset,
716                               nir_ishl(&b, nir_iand_imm(&b, x, 31), two));
717                y_offset = nir_ishl(&b, y, seven);
718        }
719        nir_ssa_def *ubo_offset = nir_iadd(&b, x_offset, y_offset);
720        nir_ssa_def *load =
721        nir_load_ubo(&b, 1, 32, one, ubo_offset,
722                    .align_mul = 4,
723                    .align_offset = 0,
724                    .range_base = 0,
725                    .range = ~0);
726
727        nir_ssa_def *output = nir_unpack_unorm_4x8(&b, load);
728
729        nir_store_var(&b, color_out,
730                      output,
731                      0xF);
732
733        struct pipe_shader_state shader_tmpl = {
734                .type = PIPE_SHADER_IR_NIR,
735                .ir.nir = b.shader,
736        };
737
738        *cached_shader = pctx->create_fs_state(pctx, &shader_tmpl);
739
740        return *cached_shader;
741}
742
743/**
744 * Turns NV12 with SAND8 format modifier from raster-order with interleaved
745 * luma and chroma 128-byte-wide-columns to tiled format for luma and chroma.
746 *
747 * This implementation is based on vc4_yuv_blit.
748 */
749static void
750v3d_sand8_blit(struct pipe_context *pctx, struct pipe_blit_info *info)
751{
752        struct v3d_context *v3d = v3d_context(pctx);
753        struct v3d_resource *src = v3d_resource(info->src.resource);
754        ASSERTED struct v3d_resource *dst = v3d_resource(info->dst.resource);
755
756        if (!src->sand_col128_stride)
757                return;
758        if (src->tiled)
759                return;
760        if (src->base.format != PIPE_FORMAT_R8_UNORM &&
761            src->base.format != PIPE_FORMAT_R8G8_UNORM)
762                return;
763        if (!(info->mask & PIPE_MASK_RGBA))
764                return;
765
766        assert(dst->base.format == src->base.format);
767        assert(dst->tiled);
768
769        assert(info->src.box.x == 0 && info->dst.box.x == 0);
770        assert(info->src.box.y == 0 && info->dst.box.y == 0);
771        assert(info->src.box.width == info->dst.box.width);
772        assert(info->src.box.height == info->dst.box.height);
773
774        v3d_blitter_save(v3d, true);
775
776        struct pipe_surface dst_tmpl;
777        util_blitter_default_dst_texture(&dst_tmpl, info->dst.resource,
778                                         info->dst.level, info->dst.box.z);
779        /* Although the src textures are cpp=1 or cpp=2, the dst texture
780         * uses a cpp=4 dst texture. So, all read/write texture ops will
781         * be done using 32-bit read and writes.
782         */
783        dst_tmpl.format = PIPE_FORMAT_R8G8B8A8_UNORM;
784        struct pipe_surface *dst_surf =
785                pctx->create_surface(pctx, info->dst.resource, &dst_tmpl);
786        if (!dst_surf) {
787                fprintf(stderr, "Failed to create YUV dst surface\n");
788                util_blitter_unset_running_flag(v3d->blitter);
789                return;
790        }
791
792        uint32_t sand8_stride = src->sand_col128_stride;
793
794        /* Adjust the dimensions of dst luma/chroma to match src
795         * size now we are using a cpp=4 format. Next dimension take into
796         * account the UIF microtile layouts.
797         */
798        dst_surf->width = align(dst_surf->width, 8) / 2;
799        if (src->cpp == 1)
800                dst_surf->height /= 2;
801
802        /* Set the constant buffer. */
803        struct pipe_constant_buffer cb_uniforms = {
804                .user_buffer = &sand8_stride,
805                .buffer_size = sizeof(sand8_stride),
806        };
807
808        pctx->set_constant_buffer(pctx, PIPE_SHADER_FRAGMENT, 0, false,
809                                  &cb_uniforms);
810        struct pipe_constant_buffer cb_src = {
811                .buffer = info->src.resource,
812                .buffer_offset = src->slices[info->src.level].offset,
813                .buffer_size = (src->bo->size -
814                                src->slices[info->src.level].offset),
815        };
816        pctx->set_constant_buffer(pctx, PIPE_SHADER_FRAGMENT, 2, false,
817                                  &cb_src);
818        /* Unbind the textures, to make sure we don't try to recurse into the
819         * shadow blit.
820         */
821        pctx->set_sampler_views(pctx, PIPE_SHADER_FRAGMENT, 0, 0, 0, false, NULL);
822        pctx->bind_sampler_states(pctx, PIPE_SHADER_FRAGMENT, 0, 0, NULL);
823
824        util_blitter_custom_shader(v3d->blitter, dst_surf,
825                                   v3d_get_sand8_vs(pctx),
826                                   v3d_get_sand8_fs(pctx, src->cpp));
827
828        util_blitter_restore_textures(v3d->blitter);
829        util_blitter_restore_constant_buffer_state(v3d->blitter);
830
831        /* Restore cb1 (util_blitter doesn't handle this one). */
832        struct pipe_constant_buffer cb_disabled = { 0 };
833        pctx->set_constant_buffer(pctx, PIPE_SHADER_FRAGMENT, 1, false,
834                                  &cb_disabled);
835
836        pipe_surface_reference(&dst_surf, NULL);
837
838        info->mask &= ~PIPE_MASK_RGBA;
839        return;
840}
841
842
843/* Optimal hardware path for blitting pixels.
844 * Scaling, format conversion, up- and downsampling (resolve) are allowed.
845 */
846void
847v3d_blit(struct pipe_context *pctx, const struct pipe_blit_info *blit_info)
848{
849        struct v3d_context *v3d = v3d_context(pctx);
850        struct pipe_blit_info info = *blit_info;
851
852        v3d_sand8_blit(pctx, &info);
853
854        v3d_tfu_blit(pctx, &info);
855
856        v3d_tlb_blit(pctx, &info);
857
858        v3d_stencil_blit(pctx, &info);
859
860        v3d_render_blit(pctx, &info);
861
862        /* Flush our blit jobs immediately.  They're unlikely to get reused by
863         * normal drawing or other blits, and without flushing we can easily
864         * run into unexpected OOMs when blits are used for a large series of
865         * texture uploads before using the textures.
866         */
867        v3d_flush_jobs_writing_resource(v3d, info.dst.resource,
868                                        V3D_FLUSH_DEFAULT, false);
869}
870