src/gallium/drivers/radeonsi/si_blit.c

   1 /*
   2  * Copyright 2010 Jerome Glisse <glisse@freedesktop.org>
   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  * on the rights to use, copy, modify, merge, publish, distribute, sub
   8  * license, and/or sell copies of the Software, and to permit persons to whom
   9  * the 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 NON-INFRINGEMENT. IN NO EVENT SHALL
  18  * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
  19  * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
  20  * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
  21  * USE OR OTHER DEALINGS IN THE SOFTWARE.
  22  */
  23
  24 #include "si_pipe.h"
  25 #include "util/u_format.h"
  26 #include "util/u_surface.h"
  27
  28 enum si_blitter_op /* bitmask */
  29 {
  30         SI_SAVE_TEXTURES      = 1,
  31         SI_SAVE_FRAMEBUFFER   = 2,
  32         SI_SAVE_FRAGMENT_STATE = 4,
  33         SI_DISABLE_RENDER_COND = 8,
  34
  35         SI_CLEAR         = SI_SAVE_FRAGMENT_STATE,
  36
  37         SI_CLEAR_SURFACE = SI_SAVE_FRAMEBUFFER | SI_SAVE_FRAGMENT_STATE,
  38
  39         SI_COPY          = SI_SAVE_FRAMEBUFFER | SI_SAVE_TEXTURES |
  40                            SI_SAVE_FRAGMENT_STATE | SI_DISABLE_RENDER_COND,
  41
  42         SI_BLIT          = SI_SAVE_FRAMEBUFFER | SI_SAVE_TEXTURES |
  43                            SI_SAVE_FRAGMENT_STATE,
  44
  45         SI_DECOMPRESS    = SI_SAVE_FRAMEBUFFER | SI_SAVE_FRAGMENT_STATE |
  46                            SI_DISABLE_RENDER_COND,
  47
  48         SI_COLOR_RESOLVE = SI_SAVE_FRAMEBUFFER | SI_SAVE_FRAGMENT_STATE
  49 };
  50
  51 static void si_blitter_begin(struct pipe_context *ctx, enum si_blitter_op op)
  52 {
  53         struct si_context *sctx = (struct si_context *)ctx;
  54
  55         r600_suspend_nontimer_queries(&sctx->b);
  56
  57         util_blitter_save_vertex_buffer_slot(sctx->blitter, sctx->vertex_buffer);
  58         util_blitter_save_vertex_elements(sctx->blitter, sctx->vertex_elements);
  59         util_blitter_save_vertex_shader(sctx->blitter, sctx->vs_shader.cso);
  60         util_blitter_save_tessctrl_shader(sctx->blitter, sctx->tcs_shader.cso);
  61         util_blitter_save_tesseval_shader(sctx->blitter, sctx->tes_shader.cso);
  62         util_blitter_save_geometry_shader(sctx->blitter, sctx->gs_shader.cso);
  63         util_blitter_save_so_targets(sctx->blitter, sctx->b.streamout.num_targets,
  64                                      (struct pipe_stream_output_target**)sctx->b.streamout.targets);
  65         util_blitter_save_rasterizer(sctx->blitter, sctx->queued.named.rasterizer);
  66
  67         if (op & SI_SAVE_FRAGMENT_STATE) {
  68                 util_blitter_save_blend(sctx->blitter, sctx->queued.named.blend);
  69                 util_blitter_save_depth_stencil_alpha(sctx->blitter, sctx->queued.named.dsa);
  70                 util_blitter_save_stencil_ref(sctx->blitter, &sctx->stencil_ref.state);
  71                 util_blitter_save_fragment_shader(sctx->blitter, sctx->ps_shader.cso);
  72                 util_blitter_save_sample_mask(sctx->blitter, sctx->sample_mask.sample_mask);
  73                 util_blitter_save_viewport(sctx->blitter, &sctx->viewports.states[0]);
  74                 util_blitter_save_scissor(sctx->blitter, &sctx->scissors.states[0]);
  75         }
  76
  77         if (op & SI_SAVE_FRAMEBUFFER)
  78                 util_blitter_save_framebuffer(sctx->blitter, &sctx->framebuffer.state);
  79
  80         if (op & SI_SAVE_TEXTURES) {
  81                 util_blitter_save_fragment_sampler_states(
  82                         sctx->blitter, 2,
  83                         sctx->samplers[PIPE_SHADER_FRAGMENT].views.sampler_states);
  84
  85                 util_blitter_save_fragment_sampler_views(sctx->blitter, 2,
  86                         sctx->samplers[PIPE_SHADER_FRAGMENT].views.views);
  87         }
  88
  89         if (op & SI_DISABLE_RENDER_COND)
  90                 sctx->b.render_cond_force_off = true;
  91 }
  92
  93 static void si_blitter_end(struct pipe_context *ctx)
  94 {
  95         struct si_context *sctx = (struct si_context *)ctx;
  96
  97         sctx->b.render_cond_force_off = false;
  98         r600_resume_nontimer_queries(&sctx->b);
  99 }
 100
 101 static unsigned u_max_sample(struct pipe_resource *r)
 102 {
 103         return r->nr_samples ? r->nr_samples - 1 : 0;
 104 }
 105
 106 static void si_blit_decompress_depth(struct pipe_context *ctx,
 107                                      struct r600_texture *texture,
 108                                      struct r600_texture *staging,
 109                                      unsigned first_level, unsigned last_level,
 110                                      unsigned first_layer, unsigned last_layer,
 111                                      unsigned first_sample, unsigned last_sample)
 112 {
 113         struct si_context *sctx = (struct si_context *)ctx;
 114         unsigned layer, level, sample, checked_last_layer, max_layer, max_sample;
 115         float depth = 1.0f;
 116         const struct util_format_description *desc;
 117         struct r600_texture *flushed_depth_texture = staging ?
 118                         staging : texture->flushed_depth_texture;
 119
 120         if (!staging && !texture->dirty_level_mask)
 121                 return;
 122
 123         max_sample = u_max_sample(&texture->resource.b.b);
 124
 125         desc = util_format_description(flushed_depth_texture->resource.b.b.format);
 126
 127         if (util_format_has_depth(desc))
 128                 sctx->dbcb_depth_copy_enabled = true;
 129         if (util_format_has_stencil(desc))
 130                 sctx->dbcb_stencil_copy_enabled = true;
 131
 132         assert(sctx->dbcb_depth_copy_enabled || sctx->dbcb_stencil_copy_enabled);
 133
 134         for (level = first_level; level <= last_level; level++) {
 135                 if (!staging && !(texture->dirty_level_mask & (1 << level)))
 136                         continue;
 137
 138                 /* The smaller the mipmap level, the less layers there are
 139                  * as far as 3D textures are concerned. */
 140                 max_layer = util_max_layer(&texture->resource.b.b, level);
 141                 checked_last_layer = last_layer < max_layer ? last_layer : max_layer;
 142
 143                 for (layer = first_layer; layer <= checked_last_layer; layer++) {
 144                         for (sample = first_sample; sample <= last_sample; sample++) {
 145                                 struct pipe_surface *zsurf, *cbsurf, surf_tmpl;
 146
 147                                 sctx->dbcb_copy_sample = sample;
 148                                 si_mark_atom_dirty(sctx, &sctx->db_render_state);
 149
 150                                 surf_tmpl.format = texture->resource.b.b.format;
 151                                 surf_tmpl.u.tex.level = level;
 152                                 surf_tmpl.u.tex.first_layer = layer;
 153                                 surf_tmpl.u.tex.last_layer = layer;
 154
 155                                 zsurf = ctx->create_surface(ctx, &texture->resource.b.b, &surf_tmpl);
 156
 157                                 surf_tmpl.format = flushed_depth_texture->resource.b.b.format;
 158                                 cbsurf = ctx->create_surface(ctx,
 159                                                 (struct pipe_resource*)flushed_depth_texture, &surf_tmpl);
 160
 161                                 si_blitter_begin(ctx, SI_DECOMPRESS);
 162                                 util_blitter_custom_depth_stencil(sctx->blitter, zsurf, cbsurf, 1 << sample,
 163                                                                   sctx->custom_dsa_flush, depth);
 164                                 si_blitter_end(ctx);
 165
 166                                 pipe_surface_reference(&zsurf, NULL);
 167                                 pipe_surface_reference(&cbsurf, NULL);
 168                         }
 169                 }
 170
 171                 /* The texture will always be dirty if some layers aren't flushed.
 172                  * I don't think this case can occur though. */
 173                 if (!staging &&
 174                     first_layer == 0 && last_layer == max_layer &&
 175                     first_sample == 0 && last_sample == max_sample) {
 176                         texture->dirty_level_mask &= ~(1 << level);
 177                 }
 178         }
 179
 180         sctx->dbcb_depth_copy_enabled = false;
 181         sctx->dbcb_stencil_copy_enabled = false;
 182         si_mark_atom_dirty(sctx, &sctx->db_render_state);
 183 }
 184
 185 static void si_blit_decompress_depth_in_place(struct si_context *sctx,
 186                                               struct r600_texture *texture,
 187                                               bool is_stencil_sampler,
 188                                               unsigned first_level, unsigned last_level,
 189                                               unsigned first_layer, unsigned last_layer)
 190 {
 191         struct pipe_surface *zsurf, surf_tmpl = {{0}};
 192         unsigned layer, max_layer, checked_last_layer, level;
 193         unsigned *dirty_level_mask;
 194
 195         if (is_stencil_sampler) {
 196                 sctx->db_flush_stencil_inplace = true;
 197                 dirty_level_mask = &texture->stencil_dirty_level_mask;
 198         } else {
 199                 sctx->db_flush_depth_inplace = true;
 200                 dirty_level_mask = &texture->dirty_level_mask;
 201         }
 202         si_mark_atom_dirty(sctx, &sctx->db_render_state);
 203
 204         surf_tmpl.format = texture->resource.b.b.format;
 205
 206         for (level = first_level; level <= last_level; level++) {
 207                 if (!(*dirty_level_mask & (1 << level)))
 208                         continue;
 209
 210                 surf_tmpl.u.tex.level = level;
 211
 212                 /* The smaller the mipmap level, the less layers there are
 213                  * as far as 3D textures are concerned. */
 214                 max_layer = util_max_layer(&texture->resource.b.b, level);
 215                 checked_last_layer = last_layer < max_layer ? last_layer : max_layer;
 216
 217                 for (layer = first_layer; layer <= checked_last_layer; layer++) {
 218                         surf_tmpl.u.tex.first_layer = layer;
 219                         surf_tmpl.u.tex.last_layer = layer;
 220
 221                         zsurf = sctx->b.b.create_surface(&sctx->b.b, &texture->resource.b.b, &surf_tmpl);
 222
 223                         si_blitter_begin(&sctx->b.b, SI_DECOMPRESS);
 224                         util_blitter_custom_depth_stencil(sctx->blitter, zsurf, NULL, ~0,
 225                                                           sctx->custom_dsa_flush,
 226                                                           1.0f);
 227                         si_blitter_end(&sctx->b.b);
 228
 229                         pipe_surface_reference(&zsurf, NULL);
 230                 }
 231
 232                 /* The texture will always be dirty if some layers aren't flushed.
 233                  * I don't think this case occurs often though. */
 234                 if (first_layer == 0 && last_layer == max_layer) {
 235                         *dirty_level_mask &= ~(1 << level);
 236                 }
 237         }
 238
 239         sctx->db_flush_depth_inplace = false;
 240         sctx->db_flush_stencil_inplace = false;
 241         si_mark_atom_dirty(sctx, &sctx->db_render_state);
 242 }
 243
 244 void si_flush_depth_textures(struct si_context *sctx,
 245                              struct si_textures_info *textures)
 246 {
 247         unsigned i;
 248         unsigned mask = textures->depth_texture_mask;
 249
 250         while (mask) {
 251                 struct pipe_sampler_view *view;
 252                 struct si_sampler_view *sview;
 253                 struct r600_texture *tex;
 254
 255                 i = u_bit_scan(&mask);
 256
 257                 view = textures->views.views[i];
 258                 assert(view);
 259                 sview = (struct si_sampler_view*)view;
 260
 261                 tex = (struct r600_texture *)view->texture;
 262                 assert(tex->is_depth && !tex->is_flushing_texture);
 263
 264                 si_blit_decompress_depth_in_place(sctx, tex,
 265                                                   sview->is_stencil_sampler,
 266                                                   view->u.tex.first_level, view->u.tex.last_level,
 267                                                   0, util_max_layer(&tex->resource.b.b, view->u.tex.first_level));
 268         }
 269 }
 270
 271 static void si_blit_decompress_color(struct pipe_context *ctx,
 272                 struct r600_texture *rtex,
 273                 unsigned first_level, unsigned last_level,
 274                 unsigned first_layer, unsigned last_layer,
 275                 bool need_dcc_decompress)
 276 {
 277         struct si_context *sctx = (struct si_context *)ctx;
 278         unsigned layer, level, checked_last_layer, max_layer;
 279
 280         if (!rtex->dirty_level_mask && !need_dcc_decompress)
 281                 return;
 282
 283         for (level = first_level; level <= last_level; level++) {
 284                 void* custom_blend;
 285
 286                 if (!(rtex->dirty_level_mask & (1 << level)) && !need_dcc_decompress)
 287                         continue;
 288
 289                 if (rtex->dcc_offset && need_dcc_decompress) {
 290                         custom_blend = sctx->custom_blend_dcc_decompress;
 291                 } else if (rtex->fmask.size) {
 292                         custom_blend = sctx->custom_blend_decompress;
 293                 } else {
 294                         custom_blend = sctx->custom_blend_fastclear;
 295                 }
 296
 297                 /* The smaller the mipmap level, the less layers there are
 298                  * as far as 3D textures are concerned. */
 299                 max_layer = util_max_layer(&rtex->resource.b.b, level);
 300                 checked_last_layer = last_layer < max_layer ? last_layer : max_layer;
 301
 302                 for (layer = first_layer; layer <= checked_last_layer; layer++) {
 303                         struct pipe_surface *cbsurf, surf_tmpl;
 304
 305                         surf_tmpl.format = rtex->resource.b.b.format;
 306                         surf_tmpl.u.tex.level = level;
 307                         surf_tmpl.u.tex.first_layer = layer;
 308                         surf_tmpl.u.tex.last_layer = layer;
 309                         cbsurf = ctx->create_surface(ctx, &rtex->resource.b.b, &surf_tmpl);
 310
 311                         si_blitter_begin(ctx, SI_DECOMPRESS);
 312                         util_blitter_custom_color(sctx->blitter, cbsurf, custom_blend);
 313                         si_blitter_end(ctx);
 314
 315                         pipe_surface_reference(&cbsurf, NULL);
 316                 }
 317
 318                 /* The texture will always be dirty if some layers aren't flushed.
 319                  * I don't think this case occurs often though. */
 320                 if (first_layer == 0 && last_layer == max_layer) {
 321                         rtex->dirty_level_mask &= ~(1 << level);
 322                 }
 323         }
 324 }
 325
 326 void si_decompress_color_textures(struct si_context *sctx,
 327                                   struct si_textures_info *textures)
 328 {
 329         unsigned i;
 330         unsigned mask = textures->compressed_colortex_mask;
 331
 332         while (mask) {
 333                 struct pipe_sampler_view *view;
 334                 struct r600_texture *tex;
 335
 336                 i = u_bit_scan(&mask);
 337
 338                 view = textures->views.views[i];
 339                 assert(view);
 340
 341                 tex = (struct r600_texture *)view->texture;
 342                 assert(tex->cmask.size || tex->fmask.size || tex->dcc_offset);
 343
 344                 si_blit_decompress_color(&sctx->b.b, tex,
 345                                          view->u.tex.first_level, view->u.tex.last_level,
 346                                          0, util_max_layer(&tex->resource.b.b, view->u.tex.first_level),
 347                                          false);
 348         }
 349 }
 350
 351 static void si_clear(struct pipe_context *ctx, unsigned buffers,
 352                      const union pipe_color_union *color,
 353                      double depth, unsigned stencil)
 354 {
 355         struct si_context *sctx = (struct si_context *)ctx;
 356         struct pipe_framebuffer_state *fb = &sctx->framebuffer.state;
 357         struct pipe_surface *zsbuf = fb->zsbuf;
 358         struct r600_texture *zstex =
 359                 zsbuf ? (struct r600_texture*)zsbuf->texture : NULL;
 360
 361         if (buffers & PIPE_CLEAR_COLOR) {
 362                 evergreen_do_fast_color_clear(&sctx->b, fb,
 363                                               &sctx->framebuffer.atom, &buffers,
 364                                               &sctx->framebuffer.dirty_cbufs,
 365                                               color);
 366                 if (!buffers)
 367                         return; /* all buffers have been fast cleared */
 368         }
 369
 370         if (buffers & PIPE_CLEAR_COLOR) {
 371                 int i;
 372
 373                 /* These buffers cannot use fast clear, make sure to disable expansion. */
 374                 for (i = 0; i < fb->nr_cbufs; i++) {
 375                         struct r600_texture *tex;
 376
 377                         /* If not clearing this buffer, skip. */
 378                         if (!(buffers & (PIPE_CLEAR_COLOR0 << i)))
 379                                 continue;
 380
 381                         if (!fb->cbufs[i])
 382                                 continue;
 383
 384                         tex = (struct r600_texture *)fb->cbufs[i]->texture;
 385                         if (tex->fmask.size == 0)
 386                                 tex->dirty_level_mask &= ~(1 << fb->cbufs[i]->u.tex.level);
 387                 }
 388         }
 389
 390         if (zstex && zstex->htile_buffer &&
 391             zsbuf->u.tex.level == 0 &&
 392             zsbuf->u.tex.first_layer == 0 &&
 393             zsbuf->u.tex.last_layer == util_max_layer(&zstex->resource.b.b, 0)) {
 394                 if (buffers & PIPE_CLEAR_DEPTH) {
 395                         /* Need to disable EXPCLEAR temporarily if clearing
 396                          * to a new value. */
 397                         if (zstex->depth_cleared && zstex->depth_clear_value != depth) {
 398                                 sctx->db_depth_disable_expclear = true;
 399                         }
 400
 401                         zstex->depth_clear_value = depth;
 402                         sctx->framebuffer.dirty_zsbuf = true;
 403                         si_mark_atom_dirty(sctx, &sctx->framebuffer.atom); /* updates DB_DEPTH_CLEAR */
 404                         sctx->db_depth_clear = true;
 405                         si_mark_atom_dirty(sctx, &sctx->db_render_state);
 406                 }
 407
 408                 if (buffers & PIPE_CLEAR_STENCIL) {
 409                         stencil &= 0xff;
 410
 411                         /* Need to disable EXPCLEAR temporarily if clearing
 412                          * to a new value. */
 413                         if (zstex->stencil_cleared && zstex->stencil_clear_value != stencil) {
 414                                 sctx->db_stencil_disable_expclear = true;
 415                         }
 416
 417                         zstex->stencil_clear_value = stencil;
 418                         sctx->framebuffer.dirty_zsbuf = true;
 419                         si_mark_atom_dirty(sctx, &sctx->framebuffer.atom); /* updates DB_STENCIL_CLEAR */
 420                         sctx->db_stencil_clear = true;
 421                         si_mark_atom_dirty(sctx, &sctx->db_render_state);
 422                 }
 423         }
 424
 425         si_blitter_begin(ctx, SI_CLEAR);
 426         util_blitter_clear(sctx->blitter, fb->width, fb->height,
 427                            util_framebuffer_get_num_layers(fb),
 428                            buffers, color, depth, stencil);
 429         si_blitter_end(ctx);
 430
 431         if (sctx->db_depth_clear) {
 432                 sctx->db_depth_clear = false;
 433                 sctx->db_depth_disable_expclear = false;
 434                 zstex->depth_cleared = true;
 435                 si_mark_atom_dirty(sctx, &sctx->db_render_state);
 436         }
 437
 438         if (sctx->db_stencil_clear) {
 439                 sctx->db_stencil_clear = false;
 440                 sctx->db_stencil_disable_expclear = false;
 441                 zstex->stencil_cleared = true;
 442                 si_mark_atom_dirty(sctx, &sctx->db_render_state);
 443         }
 444 }
 445
 446 static void si_clear_render_target(struct pipe_context *ctx,
 447                                    struct pipe_surface *dst,
 448                                    const union pipe_color_union *color,
 449                                    unsigned dstx, unsigned dsty,
 450                                    unsigned width, unsigned height)
 451 {
 452         struct si_context *sctx = (struct si_context *)ctx;
 453
 454         si_blitter_begin(ctx, SI_CLEAR_SURFACE);
 455         util_blitter_clear_render_target(sctx->blitter, dst, color,
 456                                          dstx, dsty, width, height);
 457         si_blitter_end(ctx);
 458 }
 459
 460 static void si_clear_depth_stencil(struct pipe_context *ctx,
 461                                    struct pipe_surface *dst,
 462                                    unsigned clear_flags,
 463                                    double depth,
 464                                    unsigned stencil,
 465                                    unsigned dstx, unsigned dsty,
 466                                    unsigned width, unsigned height)
 467 {
 468         struct si_context *sctx = (struct si_context *)ctx;
 469
 470         si_blitter_begin(ctx, SI_CLEAR_SURFACE);
 471         util_blitter_clear_depth_stencil(sctx->blitter, dst, clear_flags, depth, stencil,
 472                                          dstx, dsty, width, height);
 473         si_blitter_end(ctx);
 474 }
 475
 476 /* Helper for decompressing a portion of a color or depth resource before
 477  * blitting if any decompression is needed.
 478  * The driver doesn't decompress resources automatically while u_blitter is
 479  * rendering. */
 480 static void si_decompress_subresource(struct pipe_context *ctx,
 481                                       struct pipe_resource *tex,
 482                                       unsigned level,
 483                                       unsigned first_layer, unsigned last_layer)
 484 {
 485         struct si_context *sctx = (struct si_context *)ctx;
 486         struct r600_texture *rtex = (struct r600_texture*)tex;
 487
 488         if (rtex->is_depth && !rtex->is_flushing_texture) {
 489                 si_blit_decompress_depth_in_place(sctx, rtex, false,
 490                                                   level, level,
 491                                                   first_layer, last_layer);
 492                 if (rtex->surface.flags & RADEON_SURF_SBUFFER)
 493                         si_blit_decompress_depth_in_place(sctx, rtex, true,
 494                                                           level, level,
 495                                                           first_layer, last_layer);
 496         } else if (rtex->fmask.size || rtex->cmask.size || rtex->dcc_offset) {
 497                 si_blit_decompress_color(ctx, rtex, level, level,
 498                                          first_layer, last_layer, false);
 499         }
 500 }
 501
 502 struct texture_orig_info {
 503         unsigned format;
 504         unsigned width0;
 505         unsigned height0;
 506         unsigned npix_x;
 507         unsigned npix_y;
 508         unsigned npix0_x;
 509         unsigned npix0_y;
 510 };
 511
 512 void si_resource_copy_region(struct pipe_context *ctx,
 513                              struct pipe_resource *dst,
 514                              unsigned dst_level,
 515                              unsigned dstx, unsigned dsty, unsigned dstz,
 516                              struct pipe_resource *src,
 517                              unsigned src_level,
 518                              const struct pipe_box *src_box)
 519 {
 520         struct si_context *sctx = (struct si_context *)ctx;
 521         struct pipe_surface *dst_view, dst_templ;
 522         struct pipe_sampler_view src_templ, *src_view;
 523         unsigned dst_width, dst_height, src_width0, src_height0;
 524         unsigned src_force_level = 0;
 525         struct pipe_box sbox, dstbox;
 526
 527         /* Handle buffers first. */
 528         if (dst->target == PIPE_BUFFER && src->target == PIPE_BUFFER) {
 529                 si_copy_buffer(sctx, dst, src, dstx, src_box->x, src_box->width, false);
 530                 return;
 531         }
 532
 533         assert(u_max_sample(dst) == u_max_sample(src));
 534
 535         /* The driver doesn't decompress resources automatically while
 536          * u_blitter is rendering. */
 537         si_decompress_subresource(ctx, src, src_level,
 538                                   src_box->z, src_box->z + src_box->depth - 1);
 539
 540         dst_width = u_minify(dst->width0, dst_level);
 541         dst_height = u_minify(dst->height0, dst_level);
 542         src_width0 = src->width0;
 543         src_height0 = src->height0;
 544
 545         util_blitter_default_dst_texture(&dst_templ, dst, dst_level, dstz);
 546         util_blitter_default_src_texture(&src_templ, src, src_level);
 547
 548         if (util_format_is_compressed(src->format) ||
 549             util_format_is_compressed(dst->format)) {
 550                 unsigned blocksize = util_format_get_blocksize(src->format);
 551
 552                 if (blocksize == 8)
 553                         src_templ.format = PIPE_FORMAT_R16G16B16A16_UINT; /* 64-bit block */
 554                 else
 555                         src_templ.format = PIPE_FORMAT_R32G32B32A32_UINT; /* 128-bit block */
 556                 dst_templ.format = src_templ.format;
 557
 558                 dst_width = util_format_get_nblocksx(dst->format, dst_width);
 559                 dst_height = util_format_get_nblocksy(dst->format, dst_height);
 560                 src_width0 = util_format_get_nblocksx(src->format, src_width0);
 561                 src_height0 = util_format_get_nblocksy(src->format, src_height0);
 562
 563                 dstx = util_format_get_nblocksx(dst->format, dstx);
 564                 dsty = util_format_get_nblocksy(dst->format, dsty);
 565
 566                 sbox.x = util_format_get_nblocksx(src->format, src_box->x);
 567                 sbox.y = util_format_get_nblocksy(src->format, src_box->y);
 568                 sbox.z = src_box->z;
 569                 sbox.width = util_format_get_nblocksx(src->format, src_box->width);
 570                 sbox.height = util_format_get_nblocksy(src->format, src_box->height);
 571                 sbox.depth = src_box->depth;
 572                 src_box = &sbox;
 573
 574                 src_force_level = src_level;
 575         } else if (!util_blitter_is_copy_supported(sctx->blitter, dst, src) ||
 576                    /* also *8_SNORM has precision issues, use UNORM instead */
 577                    util_format_is_snorm8(src->format)) {
 578                 if (util_format_is_subsampled_422(src->format)) {
 579                         src_templ.format = PIPE_FORMAT_R8G8B8A8_UINT;
 580                         dst_templ.format = PIPE_FORMAT_R8G8B8A8_UINT;
 581
 582                         dst_width = util_format_get_nblocksx(dst->format, dst_width);
 583                         src_width0 = util_format_get_nblocksx(src->format, src_width0);
 584
 585                         dstx = util_format_get_nblocksx(dst->format, dstx);
 586
 587                         sbox = *src_box;
 588                         sbox.x = util_format_get_nblocksx(src->format, src_box->x);
 589                         sbox.width = util_format_get_nblocksx(src->format, src_box->width);
 590                         src_box = &sbox;
 591                 } else {
 592                         unsigned blocksize = util_format_get_blocksize(src->format);
 593
 594                         switch (blocksize) {
 595                         case 1:
 596                                 dst_templ.format = PIPE_FORMAT_R8_UNORM;
 597                                 src_templ.format = PIPE_FORMAT_R8_UNORM;
 598                                 break;
 599                         case 2:
 600                                 dst_templ.format = PIPE_FORMAT_R8G8_UNORM;
 601                                 src_templ.format = PIPE_FORMAT_R8G8_UNORM;
 602                                 break;
 603                         case 4:
 604                                 dst_templ.format = PIPE_FORMAT_R8G8B8A8_UNORM;
 605                                 src_templ.format = PIPE_FORMAT_R8G8B8A8_UNORM;
 606                                 break;
 607                         case 8:
 608                                 dst_templ.format = PIPE_FORMAT_R16G16B16A16_UINT;
 609                                 src_templ.format = PIPE_FORMAT_R16G16B16A16_UINT;
 610                                 break;
 611                         case 16:
 612                                 dst_templ.format = PIPE_FORMAT_R32G32B32A32_UINT;
 613                                 src_templ.format = PIPE_FORMAT_R32G32B32A32_UINT;
 614                                 break;
 615                         default:
 616                                 fprintf(stderr, "Unhandled format %s with blocksize %u\n",
 617                                         util_format_short_name(src->format), blocksize);
 618                                 assert(0);
 619                         }
 620                 }
 621         }
 622
 623         /* Initialize the surface. */
 624         dst_view = r600_create_surface_custom(ctx, dst, &dst_templ,
 625                                               dst_width, dst_height);
 626
 627         /* Initialize the sampler view. */
 628         src_view = si_create_sampler_view_custom(ctx, src, &src_templ,
 629                                                  src_width0, src_height0,
 630                                                  src_force_level);
 631
 632         u_box_3d(dstx, dsty, dstz, abs(src_box->width), abs(src_box->height),
 633                  abs(src_box->depth), &dstbox);
 634
 635         /* Copy. */
 636         si_blitter_begin(ctx, SI_COPY);
 637         util_blitter_blit_generic(sctx->blitter, dst_view, &dstbox,
 638                                   src_view, src_box, src_width0, src_height0,
 639                                   PIPE_MASK_RGBAZS, PIPE_TEX_FILTER_NEAREST, NULL,
 640                                   FALSE);
 641         si_blitter_end(ctx);
 642
 643         pipe_surface_reference(&dst_view, NULL);
 644         pipe_sampler_view_reference(&src_view, NULL);
 645 }
 646
 647 /* For MSAA integer resolving to work, we change the format to NORM using this function. */
 648 static enum pipe_format int_to_norm_format(enum pipe_format format)
 649 {
 650         switch (format) {
 651 #define REPLACE_FORMAT_SIGN(format,sign) \
 652         case PIPE_FORMAT_##format##_##sign##INT: \
 653                 return PIPE_FORMAT_##format##_##sign##NORM
 654 #define REPLACE_FORMAT(format) \
 655                 REPLACE_FORMAT_SIGN(format, U); \
 656                 REPLACE_FORMAT_SIGN(format, S)
 657
 658         REPLACE_FORMAT_SIGN(B10G10R10A2, U);
 659         REPLACE_FORMAT(R8);
 660         REPLACE_FORMAT(R8G8);
 661         REPLACE_FORMAT(R8G8B8X8);
 662         REPLACE_FORMAT(R8G8B8A8);
 663         REPLACE_FORMAT(A8);
 664         REPLACE_FORMAT(I8);
 665         REPLACE_FORMAT(L8);
 666         REPLACE_FORMAT(L8A8);
 667         REPLACE_FORMAT(R16);
 668         REPLACE_FORMAT(R16G16);
 669         REPLACE_FORMAT(R16G16B16X16);
 670         REPLACE_FORMAT(R16G16B16A16);
 671         REPLACE_FORMAT(A16);
 672         REPLACE_FORMAT(I16);
 673         REPLACE_FORMAT(L16);
 674         REPLACE_FORMAT(L16A16);
 675
 676 #undef REPLACE_FORMAT
 677 #undef REPLACE_FORMAT_SIGN
 678         default:
 679                 return format;
 680         }
 681 }
 682
 683 static bool do_hardware_msaa_resolve(struct pipe_context *ctx,
 684                                      const struct pipe_blit_info *info)
 685 {
 686         struct si_context *sctx = (struct si_context*)ctx;
 687         struct r600_texture *dst = (struct r600_texture*)info->dst.resource;
 688         unsigned dst_width = u_minify(info->dst.resource->width0, info->dst.level);
 689         unsigned dst_height = u_minify(info->dst.resource->height0, info->dst.level);
 690         enum pipe_format format = int_to_norm_format(info->dst.format);
 691         unsigned sample_mask = ~0;
 692
 693         /* Hardware MSAA resolve doesn't work if SPI format = NORM16_ABGR and
 694          * the format is R16G16. Use R16A16, which does work.
 695          */
 696         if (format == PIPE_FORMAT_R16G16_UNORM)
 697                 format = PIPE_FORMAT_R16A16_UNORM;
 698         if (format == PIPE_FORMAT_R16G16_SNORM)
 699                 format = PIPE_FORMAT_R16A16_SNORM;
 700
 701         if (info->src.resource->nr_samples > 1 &&
 702             info->dst.resource->nr_samples <= 1 &&
 703             util_max_layer(info->src.resource, 0) == 0 &&
 704             util_max_layer(info->dst.resource, info->dst.level) == 0 &&
 705             info->dst.format == info->src.format &&
 706             !util_format_is_pure_integer(format) &&
 707             !util_format_is_depth_or_stencil(format) &&
 708             !info->scissor_enable &&
 709             (info->mask & PIPE_MASK_RGBA) == PIPE_MASK_RGBA &&
 710             dst_width == info->src.resource->width0 &&
 711             dst_height == info->src.resource->height0 &&
 712             info->dst.box.x == 0 &&
 713             info->dst.box.y == 0 &&
 714             info->dst.box.width == dst_width &&
 715             info->dst.box.height == dst_height &&
 716             info->dst.box.depth == 1 &&
 717             info->src.box.x == 0 &&
 718             info->src.box.y == 0 &&
 719             info->src.box.width == dst_width &&
 720             info->src.box.height == dst_height &&
 721             info->src.box.depth == 1 &&
 722             dst->surface.level[info->dst.level].mode >= RADEON_SURF_MODE_1D &&
 723             !(dst->surface.flags & RADEON_SURF_SCANOUT) &&
 724             (!dst->cmask.size || !dst->dirty_level_mask) && /* dst cannot be fast-cleared */
 725             !dst->dcc_offset) {
 726                 si_blitter_begin(ctx, SI_COLOR_RESOLVE |
 727                                  (info->render_condition_enable ? 0 : SI_DISABLE_RENDER_COND));
 728                 util_blitter_custom_resolve_color(sctx->blitter,
 729                                                   info->dst.resource, info->dst.level,
 730                                                   info->dst.box.z,
 731                                                   info->src.resource, info->src.box.z,
 732                                                   sample_mask, sctx->custom_blend_resolve,
 733                                                   format);
 734                 si_blitter_end(ctx);
 735                 return true;
 736         }
 737         return false;
 738 }
 739
 740 static void si_blit(struct pipe_context *ctx,
 741                     const struct pipe_blit_info *info)
 742 {
 743         struct si_context *sctx = (struct si_context*)ctx;
 744
 745         if (do_hardware_msaa_resolve(ctx, info)) {
 746                 return;
 747         }
 748
 749         assert(util_blitter_is_blit_supported(sctx->blitter, info));
 750
 751         /* The driver doesn't decompress resources automatically while
 752          * u_blitter is rendering. */
 753         si_decompress_subresource(ctx, info->src.resource, info->src.level,
 754                                   info->src.box.z,
 755                                   info->src.box.z + info->src.box.depth - 1);
 756
 757         if (sctx->screen->b.debug_flags & DBG_FORCE_DMA &&
 758             util_try_blit_via_copy_region(ctx, info))
 759                 return;
 760
 761         si_blitter_begin(ctx, SI_BLIT |
 762                          (info->render_condition_enable ? 0 : SI_DISABLE_RENDER_COND));
 763         util_blitter_blit(sctx->blitter, info);
 764         si_blitter_end(ctx);
 765 }
 766
 767 static void si_flush_resource(struct pipe_context *ctx,
 768                               struct pipe_resource *res)
 769 {
 770         struct r600_texture *rtex = (struct r600_texture*)res;
 771
 772         assert(res->target != PIPE_BUFFER);
 773
 774         if (!rtex->is_depth && (rtex->cmask.size || rtex->dcc_offset)) {
 775                 si_blit_decompress_color(ctx, rtex, 0, res->last_level,
 776                                          0, util_max_layer(res, 0), false);
 777         }
 778 }
 779
 780 static void si_decompress_dcc(struct pipe_context *ctx,
 781                               struct r600_texture *rtex)
 782 {
 783         if (!rtex->dcc_offset)
 784                 return;
 785
 786         si_blit_decompress_color(ctx, rtex, 0, rtex->resource.b.b.last_level,
 787                                  0, util_max_layer(&rtex->resource.b.b, 0),
 788                                  true);
 789 }
 790
 791 static void si_pipe_clear_buffer(struct pipe_context *ctx,
 792                                  struct pipe_resource *dst,
 793                                  unsigned offset, unsigned size,
 794                                  const void *clear_value_ptr,
 795                                  int clear_value_size)
 796 {
 797         struct si_context *sctx = (struct si_context*)ctx;
 798         uint32_t dword_value;
 799         unsigned i;
 800
 801         assert(offset % clear_value_size == 0);
 802         assert(size % clear_value_size == 0);
 803
 804         if (clear_value_size > 4) {
 805                 const uint32_t *u32 = clear_value_ptr;
 806                 bool clear_dword_duplicated = true;
 807
 808                 /* See if we can lower large fills to dword fills. */
 809                 for (i = 1; i < clear_value_size / 4; i++)
 810                         if (u32[0] != u32[i]) {
 811                                 clear_dword_duplicated = false;
 812                                 break;
 813                         }
 814
 815                 if (!clear_dword_duplicated) {
 816                         /* Use transform feedback for 64-bit, 96-bit, and
 817                          * 128-bit fills.
 818                          */
 819                         union pipe_color_union clear_value;
 820
 821                         memcpy(&clear_value, clear_value_ptr, clear_value_size);
 822                         si_blitter_begin(ctx, SI_DISABLE_RENDER_COND);
 823                         util_blitter_clear_buffer(sctx->blitter, dst, offset,
 824                                                   size, clear_value_size / 4,
 825                                                   &clear_value);
 826                         si_blitter_end(ctx);
 827                         return;
 828                 }
 829         }
 830
 831         /* Expand the clear value to a dword. */
 832         switch (clear_value_size) {
 833         case 1:
 834                 dword_value = *(uint8_t*)clear_value_ptr;
 835                 dword_value |= (dword_value << 8) |
 836                                (dword_value << 16) |
 837                                (dword_value << 24);
 838                 break;
 839         case 2:
 840                 dword_value = *(uint16_t*)clear_value_ptr;
 841                 dword_value |= dword_value << 16;
 842                 break;
 843         default:
 844                 dword_value = *(uint32_t*)clear_value_ptr;
 845         }
 846
 847         sctx->b.clear_buffer(ctx, dst, offset, size, dword_value, false);
 848 }
 849
 850 void si_init_blit_functions(struct si_context *sctx)
 851 {
 852         sctx->b.b.clear = si_clear;
 853         sctx->b.b.clear_buffer = si_pipe_clear_buffer;
 854         sctx->b.b.clear_render_target = si_clear_render_target;
 855         sctx->b.b.clear_depth_stencil = si_clear_depth_stencil;
 856         sctx->b.b.resource_copy_region = si_resource_copy_region;
 857         sctx->b.b.blit = si_blit;
 858         sctx->b.b.flush_resource = si_flush_resource;
 859         sctx->b.blit_decompress_depth = si_blit_decompress_depth;
 860         sctx->b.decompress_dcc = si_decompress_dcc;
 861 }