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 {
 276         struct si_context *sctx = (struct si_context *)ctx;
 277         unsigned layer, level, checked_last_layer, max_layer;
 278
 279         if (!rtex->dirty_level_mask)
 280                 return;
 281
 282         for (level = first_level; level <= last_level; level++) {
 283                 if (!(rtex->dirty_level_mask & (1 << level)))
 284                         continue;
 285
 286                 /* The smaller the mipmap level, the less layers there are
 287                  * as far as 3D textures are concerned. */
 288                 max_layer = util_max_layer(&rtex->resource.b.b, level);
 289                 checked_last_layer = last_layer < max_layer ? last_layer : max_layer;
 290
 291                 for (layer = first_layer; layer <= checked_last_layer; layer++) {
 292                         struct pipe_surface *cbsurf, surf_tmpl;
 293
 294                         surf_tmpl.format = rtex->resource.b.b.format;
 295                         surf_tmpl.u.tex.level = level;
 296                         surf_tmpl.u.tex.first_layer = layer;
 297                         surf_tmpl.u.tex.last_layer = layer;
 298                         cbsurf = ctx->create_surface(ctx, &rtex->resource.b.b, &surf_tmpl);
 299
 300                         si_blitter_begin(ctx, SI_DECOMPRESS);
 301                         util_blitter_custom_color(sctx->blitter, cbsurf,
 302                                 rtex->fmask.size ? sctx->custom_blend_decompress :
 303                                                    sctx->custom_blend_fastclear);
 304                         si_blitter_end(ctx);
 305
 306                         pipe_surface_reference(&cbsurf, NULL);
 307                 }
 308
 309                 /* The texture will always be dirty if some layers aren't flushed.
 310                  * I don't think this case occurs often though. */
 311                 if (first_layer == 0 && last_layer == max_layer) {
 312                         rtex->dirty_level_mask &= ~(1 << level);
 313                 }
 314         }
 315 }
 316
 317 void si_decompress_color_textures(struct si_context *sctx,
 318                                   struct si_textures_info *textures)
 319 {
 320         unsigned i;
 321         unsigned mask = textures->compressed_colortex_mask;
 322
 323         while (mask) {
 324                 struct pipe_sampler_view *view;
 325                 struct r600_texture *tex;
 326
 327                 i = u_bit_scan(&mask);
 328
 329                 view = textures->views.views[i];
 330                 assert(view);
 331
 332                 tex = (struct r600_texture *)view->texture;
 333                 assert(tex->cmask.size || tex->fmask.size || tex->dcc_buffer);
 334
 335                 si_blit_decompress_color(&sctx->b.b, tex,
 336                                          view->u.tex.first_level, view->u.tex.last_level,
 337                                          0, util_max_layer(&tex->resource.b.b, view->u.tex.first_level));
 338         }
 339 }
 340
 341 static void si_clear(struct pipe_context *ctx, unsigned buffers,
 342                      const union pipe_color_union *color,
 343                      double depth, unsigned stencil)
 344 {
 345         struct si_context *sctx = (struct si_context *)ctx;
 346         struct pipe_framebuffer_state *fb = &sctx->framebuffer.state;
 347         struct pipe_surface *zsbuf = fb->zsbuf;
 348         struct r600_texture *zstex =
 349                 zsbuf ? (struct r600_texture*)zsbuf->texture : NULL;
 350
 351         if (buffers & PIPE_CLEAR_COLOR) {
 352                 evergreen_do_fast_color_clear(&sctx->b, fb,
 353                                               &sctx->framebuffer.atom, &buffers,
 354                                               &sctx->framebuffer.dirty_cbufs,
 355                                               color);
 356                 if (!buffers)
 357                         return; /* all buffers have been fast cleared */
 358         }
 359
 360         if (buffers & PIPE_CLEAR_COLOR) {
 361                 int i;
 362
 363                 /* These buffers cannot use fast clear, make sure to disable expansion. */
 364                 for (i = 0; i < fb->nr_cbufs; i++) {
 365                         struct r600_texture *tex;
 366
 367                         /* If not clearing this buffer, skip. */
 368                         if (!(buffers & (PIPE_CLEAR_COLOR0 << i)))
 369                                 continue;
 370
 371                         if (!fb->cbufs[i])
 372                                 continue;
 373
 374                         tex = (struct r600_texture *)fb->cbufs[i]->texture;
 375                         if (tex->fmask.size == 0)
 376                                 tex->dirty_level_mask &= ~(1 << fb->cbufs[i]->u.tex.level);
 377                 }
 378         }
 379
 380         if (zstex && zstex->htile_buffer &&
 381             zsbuf->u.tex.level == 0 &&
 382             zsbuf->u.tex.first_layer == 0 &&
 383             zsbuf->u.tex.last_layer == util_max_layer(&zstex->resource.b.b, 0)) {
 384                 if (buffers & PIPE_CLEAR_DEPTH) {
 385                         /* Need to disable EXPCLEAR temporarily if clearing
 386                          * to a new value. */
 387                         if (zstex->depth_cleared && zstex->depth_clear_value != depth) {
 388                                 sctx->db_depth_disable_expclear = true;
 389                         }
 390
 391                         zstex->depth_clear_value = depth;
 392                         sctx->framebuffer.dirty_zsbuf = true;
 393                         si_mark_atom_dirty(sctx, &sctx->framebuffer.atom); /* updates DB_DEPTH_CLEAR */
 394                         sctx->db_depth_clear = true;
 395                         si_mark_atom_dirty(sctx, &sctx->db_render_state);
 396                 }
 397
 398                 if (buffers & PIPE_CLEAR_STENCIL) {
 399                         stencil &= 0xff;
 400
 401                         /* Need to disable EXPCLEAR temporarily if clearing
 402                          * to a new value. */
 403                         if (zstex->stencil_cleared && zstex->stencil_clear_value != stencil) {
 404                                 sctx->db_stencil_disable_expclear = true;
 405                         }
 406
 407                         zstex->stencil_clear_value = stencil;
 408                         sctx->framebuffer.dirty_zsbuf = true;
 409                         si_mark_atom_dirty(sctx, &sctx->framebuffer.atom); /* updates DB_STENCIL_CLEAR */
 410                         sctx->db_stencil_clear = true;
 411                         si_mark_atom_dirty(sctx, &sctx->db_render_state);
 412                 }
 413         }
 414
 415         si_blitter_begin(ctx, SI_CLEAR);
 416         util_blitter_clear(sctx->blitter, fb->width, fb->height,
 417                            util_framebuffer_get_num_layers(fb),
 418                            buffers, color, depth, stencil);
 419         si_blitter_end(ctx);
 420
 421         if (sctx->db_depth_clear) {
 422                 sctx->db_depth_clear = false;
 423                 sctx->db_depth_disable_expclear = false;
 424                 zstex->depth_cleared = true;
 425                 si_mark_atom_dirty(sctx, &sctx->db_render_state);
 426         }
 427
 428         if (sctx->db_stencil_clear) {
 429                 sctx->db_stencil_clear = false;
 430                 sctx->db_stencil_disable_expclear = false;
 431                 zstex->stencil_cleared = true;
 432                 si_mark_atom_dirty(sctx, &sctx->db_render_state);
 433         }
 434 }
 435
 436 static void si_clear_render_target(struct pipe_context *ctx,
 437                                    struct pipe_surface *dst,
 438                                    const union pipe_color_union *color,
 439                                    unsigned dstx, unsigned dsty,
 440                                    unsigned width, unsigned height)
 441 {
 442         struct si_context *sctx = (struct si_context *)ctx;
 443
 444         si_blitter_begin(ctx, SI_CLEAR_SURFACE);
 445         util_blitter_clear_render_target(sctx->blitter, dst, color,
 446                                          dstx, dsty, width, height);
 447         si_blitter_end(ctx);
 448 }
 449
 450 static void si_clear_depth_stencil(struct pipe_context *ctx,
 451                                    struct pipe_surface *dst,
 452                                    unsigned clear_flags,
 453                                    double depth,
 454                                    unsigned stencil,
 455                                    unsigned dstx, unsigned dsty,
 456                                    unsigned width, unsigned height)
 457 {
 458         struct si_context *sctx = (struct si_context *)ctx;
 459
 460         si_blitter_begin(ctx, SI_CLEAR_SURFACE);
 461         util_blitter_clear_depth_stencil(sctx->blitter, dst, clear_flags, depth, stencil,
 462                                          dstx, dsty, width, height);
 463         si_blitter_end(ctx);
 464 }
 465
 466 /* Helper for decompressing a portion of a color or depth resource before
 467  * blitting if any decompression is needed.
 468  * The driver doesn't decompress resources automatically while u_blitter is
 469  * rendering. */
 470 static void si_decompress_subresource(struct pipe_context *ctx,
 471                                       struct pipe_resource *tex,
 472                                       unsigned level,
 473                                       unsigned first_layer, unsigned last_layer)
 474 {
 475         struct si_context *sctx = (struct si_context *)ctx;
 476         struct r600_texture *rtex = (struct r600_texture*)tex;
 477
 478         if (rtex->is_depth && !rtex->is_flushing_texture) {
 479                 si_blit_decompress_depth_in_place(sctx, rtex, false,
 480                                                   level, level,
 481                                                   first_layer, last_layer);
 482                 if (rtex->surface.flags & RADEON_SURF_SBUFFER)
 483                         si_blit_decompress_depth_in_place(sctx, rtex, true,
 484                                                           level, level,
 485                                                           first_layer, last_layer);
 486         } else if (rtex->fmask.size || rtex->cmask.size || rtex->dcc_buffer) {
 487                 si_blit_decompress_color(ctx, rtex, level, level,
 488                                          first_layer, last_layer);
 489         }
 490 }
 491
 492 struct texture_orig_info {
 493         unsigned format;
 494         unsigned width0;
 495         unsigned height0;
 496         unsigned npix_x;
 497         unsigned npix_y;
 498         unsigned npix0_x;
 499         unsigned npix0_y;
 500 };
 501
 502 void si_resource_copy_region(struct pipe_context *ctx,
 503                              struct pipe_resource *dst,
 504                              unsigned dst_level,
 505                              unsigned dstx, unsigned dsty, unsigned dstz,
 506                              struct pipe_resource *src,
 507                              unsigned src_level,
 508                              const struct pipe_box *src_box)
 509 {
 510         struct si_context *sctx = (struct si_context *)ctx;
 511         struct pipe_surface *dst_view, dst_templ;
 512         struct pipe_sampler_view src_templ, *src_view;
 513         unsigned dst_width, dst_height, src_width0, src_height0;
 514         unsigned src_force_level = 0;
 515         struct pipe_box sbox, dstbox;
 516
 517         /* Handle buffers first. */
 518         if (dst->target == PIPE_BUFFER && src->target == PIPE_BUFFER) {
 519                 si_copy_buffer(sctx, dst, src, dstx, src_box->x, src_box->width, false);
 520                 return;
 521         }
 522
 523         assert(u_max_sample(dst) == u_max_sample(src));
 524
 525         /* The driver doesn't decompress resources automatically while
 526          * u_blitter is rendering. */
 527         si_decompress_subresource(ctx, src, src_level,
 528                                   src_box->z, src_box->z + src_box->depth - 1);
 529
 530         dst_width = u_minify(dst->width0, dst_level);
 531         dst_height = u_minify(dst->height0, dst_level);
 532         src_width0 = src->width0;
 533         src_height0 = src->height0;
 534
 535         util_blitter_default_dst_texture(&dst_templ, dst, dst_level, dstz);
 536         util_blitter_default_src_texture(&src_templ, src, src_level);
 537
 538         if (util_format_is_compressed(src->format) ||
 539             util_format_is_compressed(dst->format)) {
 540                 unsigned blocksize = util_format_get_blocksize(src->format);
 541
 542                 if (blocksize == 8)
 543                         src_templ.format = PIPE_FORMAT_R16G16B16A16_UINT; /* 64-bit block */
 544                 else
 545                         src_templ.format = PIPE_FORMAT_R32G32B32A32_UINT; /* 128-bit block */
 546                 dst_templ.format = src_templ.format;
 547
 548                 dst_width = util_format_get_nblocksx(dst->format, dst_width);
 549                 dst_height = util_format_get_nblocksy(dst->format, dst_height);
 550                 src_width0 = util_format_get_nblocksx(src->format, src_width0);
 551                 src_height0 = util_format_get_nblocksy(src->format, src_height0);
 552
 553                 dstx = util_format_get_nblocksx(dst->format, dstx);
 554                 dsty = util_format_get_nblocksy(dst->format, dsty);
 555
 556                 sbox.x = util_format_get_nblocksx(src->format, src_box->x);
 557                 sbox.y = util_format_get_nblocksy(src->format, src_box->y);
 558                 sbox.z = src_box->z;
 559                 sbox.width = util_format_get_nblocksx(src->format, src_box->width);
 560                 sbox.height = util_format_get_nblocksy(src->format, src_box->height);
 561                 sbox.depth = src_box->depth;
 562                 src_box = &sbox;
 563
 564                 src_force_level = src_level;
 565         } else if (!util_blitter_is_copy_supported(sctx->blitter, dst, src) ||
 566                    /* also *8_SNORM has precision issues, use UNORM instead */
 567                    util_format_is_snorm8(src->format)) {
 568                 if (util_format_is_subsampled_422(src->format)) {
 569                         src_templ.format = PIPE_FORMAT_R8G8B8A8_UINT;
 570                         dst_templ.format = PIPE_FORMAT_R8G8B8A8_UINT;
 571
 572                         dst_width = util_format_get_nblocksx(dst->format, dst_width);
 573                         src_width0 = util_format_get_nblocksx(src->format, src_width0);
 574
 575                         dstx = util_format_get_nblocksx(dst->format, dstx);
 576
 577                         sbox = *src_box;
 578                         sbox.x = util_format_get_nblocksx(src->format, src_box->x);
 579                         sbox.width = util_format_get_nblocksx(src->format, src_box->width);
 580                         src_box = &sbox;
 581                 } else {
 582                         unsigned blocksize = util_format_get_blocksize(src->format);
 583
 584                         switch (blocksize) {
 585                         case 1:
 586                                 dst_templ.format = PIPE_FORMAT_R8_UNORM;
 587                                 src_templ.format = PIPE_FORMAT_R8_UNORM;
 588                                 break;
 589                         case 2:
 590                                 dst_templ.format = PIPE_FORMAT_R8G8_UNORM;
 591                                 src_templ.format = PIPE_FORMAT_R8G8_UNORM;
 592                                 break;
 593                         case 4:
 594                                 dst_templ.format = PIPE_FORMAT_R8G8B8A8_UNORM;
 595                                 src_templ.format = PIPE_FORMAT_R8G8B8A8_UNORM;
 596                                 break;
 597                         case 8:
 598                                 dst_templ.format = PIPE_FORMAT_R16G16B16A16_UINT;
 599                                 src_templ.format = PIPE_FORMAT_R16G16B16A16_UINT;
 600                                 break;
 601                         case 16:
 602                                 dst_templ.format = PIPE_FORMAT_R32G32B32A32_UINT;
 603                                 src_templ.format = PIPE_FORMAT_R32G32B32A32_UINT;
 604                                 break;
 605                         default:
 606                                 fprintf(stderr, "Unhandled format %s with blocksize %u\n",
 607                                         util_format_short_name(src->format), blocksize);
 608                                 assert(0);
 609                         }
 610                 }
 611         }
 612
 613         /* Initialize the surface. */
 614         dst_view = r600_create_surface_custom(ctx, dst, &dst_templ,
 615                                               dst_width, dst_height);
 616
 617         /* Initialize the sampler view. */
 618         src_view = si_create_sampler_view_custom(ctx, src, &src_templ,
 619                                                  src_width0, src_height0,
 620                                                  src_force_level);
 621
 622         u_box_3d(dstx, dsty, dstz, abs(src_box->width), abs(src_box->height),
 623                  abs(src_box->depth), &dstbox);
 624
 625         /* Copy. */
 626         si_blitter_begin(ctx, SI_COPY);
 627         util_blitter_blit_generic(sctx->blitter, dst_view, &dstbox,
 628                                   src_view, src_box, src_width0, src_height0,
 629                                   PIPE_MASK_RGBAZS, PIPE_TEX_FILTER_NEAREST, NULL,
 630                                   FALSE);
 631         si_blitter_end(ctx);
 632
 633         pipe_surface_reference(&dst_view, NULL);
 634         pipe_sampler_view_reference(&src_view, NULL);
 635 }
 636
 637 /* For MSAA integer resolving to work, we change the format to NORM using this function. */
 638 static enum pipe_format int_to_norm_format(enum pipe_format format)
 639 {
 640         switch (format) {
 641 #define REPLACE_FORMAT_SIGN(format,sign) \
 642         case PIPE_FORMAT_##format##_##sign##INT: \
 643                 return PIPE_FORMAT_##format##_##sign##NORM
 644 #define REPLACE_FORMAT(format) \
 645                 REPLACE_FORMAT_SIGN(format, U); \
 646                 REPLACE_FORMAT_SIGN(format, S)
 647
 648         REPLACE_FORMAT_SIGN(B10G10R10A2, U);
 649         REPLACE_FORMAT(R8);
 650         REPLACE_FORMAT(R8G8);
 651         REPLACE_FORMAT(R8G8B8X8);
 652         REPLACE_FORMAT(R8G8B8A8);
 653         REPLACE_FORMAT(A8);
 654         REPLACE_FORMAT(I8);
 655         REPLACE_FORMAT(L8);
 656         REPLACE_FORMAT(L8A8);
 657         REPLACE_FORMAT(R16);
 658         REPLACE_FORMAT(R16G16);
 659         REPLACE_FORMAT(R16G16B16X16);
 660         REPLACE_FORMAT(R16G16B16A16);
 661         REPLACE_FORMAT(A16);
 662         REPLACE_FORMAT(I16);
 663         REPLACE_FORMAT(L16);
 664         REPLACE_FORMAT(L16A16);
 665
 666 #undef REPLACE_FORMAT
 667 #undef REPLACE_FORMAT_SIGN
 668         default:
 669                 return format;
 670         }
 671 }
 672
 673 static bool do_hardware_msaa_resolve(struct pipe_context *ctx,
 674                                      const struct pipe_blit_info *info)
 675 {
 676         struct si_context *sctx = (struct si_context*)ctx;
 677         struct r600_texture *dst = (struct r600_texture*)info->dst.resource;
 678         unsigned dst_width = u_minify(info->dst.resource->width0, info->dst.level);
 679         unsigned dst_height = u_minify(info->dst.resource->height0, info->dst.level);
 680         enum pipe_format format = int_to_norm_format(info->dst.format);
 681         unsigned sample_mask = ~0;
 682
 683         /* Hardware MSAA resolve doesn't work if SPI format = NORM16_ABGR and
 684          * the format is R16G16. Use R16A16, which does work.
 685          */
 686         if (format == PIPE_FORMAT_R16G16_UNORM)
 687                 format = PIPE_FORMAT_R16A16_UNORM;
 688         if (format == PIPE_FORMAT_R16G16_SNORM)
 689                 format = PIPE_FORMAT_R16A16_SNORM;
 690
 691         if (info->src.resource->nr_samples > 1 &&
 692             info->dst.resource->nr_samples <= 1 &&
 693             util_max_layer(info->src.resource, 0) == 0 &&
 694             util_max_layer(info->dst.resource, info->dst.level) == 0 &&
 695             info->dst.format == info->src.format &&
 696             !util_format_is_pure_integer(format) &&
 697             !util_format_is_depth_or_stencil(format) &&
 698             !info->scissor_enable &&
 699             (info->mask & PIPE_MASK_RGBA) == PIPE_MASK_RGBA &&
 700             dst_width == info->src.resource->width0 &&
 701             dst_height == info->src.resource->height0 &&
 702             info->dst.box.x == 0 &&
 703             info->dst.box.y == 0 &&
 704             info->dst.box.width == dst_width &&
 705             info->dst.box.height == dst_height &&
 706             info->dst.box.depth == 1 &&
 707             info->src.box.x == 0 &&
 708             info->src.box.y == 0 &&
 709             info->src.box.width == dst_width &&
 710             info->src.box.height == dst_height &&
 711             info->src.box.depth == 1 &&
 712             dst->surface.level[info->dst.level].mode >= RADEON_SURF_MODE_1D &&
 713             !(dst->surface.flags & RADEON_SURF_SCANOUT) &&
 714             (!dst->cmask.size || !dst->dirty_level_mask) && /* dst cannot be fast-cleared */
 715             !dst->dcc_buffer) {
 716                 si_blitter_begin(ctx, SI_COLOR_RESOLVE |
 717                                  (info->render_condition_enable ? 0 : SI_DISABLE_RENDER_COND));
 718                 util_blitter_custom_resolve_color(sctx->blitter,
 719                                                   info->dst.resource, info->dst.level,
 720                                                   info->dst.box.z,
 721                                                   info->src.resource, info->src.box.z,
 722                                                   sample_mask, sctx->custom_blend_resolve,
 723                                                   format);
 724                 si_blitter_end(ctx);
 725                 return true;
 726         }
 727         return false;
 728 }
 729
 730 static void si_blit(struct pipe_context *ctx,
 731                     const struct pipe_blit_info *info)
 732 {
 733         struct si_context *sctx = (struct si_context*)ctx;
 734
 735         if (do_hardware_msaa_resolve(ctx, info)) {
 736                 return;
 737         }
 738
 739         assert(util_blitter_is_blit_supported(sctx->blitter, info));
 740
 741         /* The driver doesn't decompress resources automatically while
 742          * u_blitter is rendering. */
 743         si_decompress_subresource(ctx, info->src.resource, info->src.level,
 744                                   info->src.box.z,
 745                                   info->src.box.z + info->src.box.depth - 1);
 746
 747         if (sctx->screen->b.debug_flags & DBG_FORCE_DMA &&
 748             util_try_blit_via_copy_region(ctx, info))
 749                 return;
 750
 751         si_blitter_begin(ctx, SI_BLIT |
 752                          (info->render_condition_enable ? 0 : SI_DISABLE_RENDER_COND));
 753         util_blitter_blit(sctx->blitter, info);
 754         si_blitter_end(ctx);
 755 }
 756
 757 static void si_flush_resource(struct pipe_context *ctx,
 758                               struct pipe_resource *res)
 759 {
 760         struct r600_texture *rtex = (struct r600_texture*)res;
 761
 762         assert(res->target != PIPE_BUFFER);
 763
 764         if (!rtex->is_depth && rtex->cmask.size) {
 765                 si_blit_decompress_color(ctx, rtex, 0, res->last_level,
 766                                          0, util_max_layer(res, 0));
 767         }
 768 }
 769
 770 static void si_pipe_clear_buffer(struct pipe_context *ctx,
 771                                  struct pipe_resource *dst,
 772                                  unsigned offset, unsigned size,
 773                                  const void *clear_value_ptr,
 774                                  int clear_value_size)
 775 {
 776         struct si_context *sctx = (struct si_context*)ctx;
 777         uint32_t dword_value;
 778         unsigned i;
 779
 780         assert(offset % clear_value_size == 0);
 781         assert(size % clear_value_size == 0);
 782
 783         if (clear_value_size > 4) {
 784                 const uint32_t *u32 = clear_value_ptr;
 785                 bool clear_dword_duplicated = true;
 786
 787                 /* See if we can lower large fills to dword fills. */
 788                 for (i = 1; i < clear_value_size / 4; i++)
 789                         if (u32[0] != u32[i]) {
 790                                 clear_dword_duplicated = false;
 791                                 break;
 792                         }
 793
 794                 if (!clear_dword_duplicated) {
 795                         /* Use transform feedback for 64-bit, 96-bit, and
 796                          * 128-bit fills.
 797                          */
 798                         union pipe_color_union clear_value;
 799
 800                         memcpy(&clear_value, clear_value_ptr, clear_value_size);
 801                         si_blitter_begin(ctx, SI_DISABLE_RENDER_COND);
 802                         util_blitter_clear_buffer(sctx->blitter, dst, offset,
 803                                                   size, clear_value_size / 4,
 804                                                   &clear_value);
 805                         si_blitter_end(ctx);
 806                         return;
 807                 }
 808         }
 809
 810         /* Expand the clear value to a dword. */
 811         switch (clear_value_size) {
 812         case 1:
 813                 dword_value = *(uint8_t*)clear_value_ptr;
 814                 dword_value |= (dword_value << 8) |
 815                                (dword_value << 16) |
 816                                (dword_value << 24);
 817                 break;
 818         case 2:
 819                 dword_value = *(uint16_t*)clear_value_ptr;
 820                 dword_value |= dword_value << 16;
 821                 break;
 822         default:
 823                 dword_value = *(uint32_t*)clear_value_ptr;
 824         }
 825
 826         sctx->b.clear_buffer(ctx, dst, offset, size, dword_value, false);
 827 }
 828
 829 void si_init_blit_functions(struct si_context *sctx)
 830 {
 831         sctx->b.b.clear = si_clear;
 832         sctx->b.b.clear_buffer = si_pipe_clear_buffer;
 833         sctx->b.b.clear_render_target = si_clear_render_target;
 834         sctx->b.b.clear_depth_stencil = si_clear_depth_stencil;
 835         sctx->b.b.resource_copy_region = si_resource_copy_region;
 836         sctx->b.b.blit = si_blit;
 837         sctx->b.b.flush_resource = si_flush_resource;
 838         sctx->b.blit_decompress_depth = si_blit_decompress_depth;
 839 }