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].states.saved_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) && sctx->b.current_render_cond) {
  90                 util_blitter_save_render_condition(sctx->blitter,
  91                                                    sctx->b.current_render_cond,
  92                                                    sctx->b.current_render_cond_cond,
  93                                                    sctx->b.current_render_cond_mode);
  94         }
  95 }
  96
  97 static void si_blitter_end(struct pipe_context *ctx)
  98 {
  99         struct si_context *sctx = (struct si_context *)ctx;
 100         r600_resume_nontimer_queries(&sctx->b);
 101 }
 102
 103 static unsigned u_max_sample(struct pipe_resource *r)
 104 {
 105         return r->nr_samples ? r->nr_samples - 1 : 0;
 106 }
 107
 108 static void si_blit_decompress_depth(struct pipe_context *ctx,
 109                                      struct r600_texture *texture,
 110                                      struct r600_texture *staging,
 111                                      unsigned first_level, unsigned last_level,
 112                                      unsigned first_layer, unsigned last_layer,
 113                                      unsigned first_sample, unsigned last_sample)
 114 {
 115         struct si_context *sctx = (struct si_context *)ctx;
 116         unsigned layer, level, sample, checked_last_layer, max_layer, max_sample;
 117         float depth = 1.0f;
 118         const struct util_format_description *desc;
 119         struct r600_texture *flushed_depth_texture = staging ?
 120                         staging : texture->flushed_depth_texture;
 121
 122         if (!staging && !texture->dirty_level_mask)
 123                 return;
 124
 125         max_sample = u_max_sample(&texture->resource.b.b);
 126
 127         desc = util_format_description(flushed_depth_texture->resource.b.b.format);
 128
 129         if (util_format_has_depth(desc))
 130                 sctx->dbcb_depth_copy_enabled = true;
 131         if (util_format_has_stencil(desc))
 132                 sctx->dbcb_stencil_copy_enabled = true;
 133
 134         assert(sctx->dbcb_depth_copy_enabled || sctx->dbcb_stencil_copy_enabled);
 135
 136         for (level = first_level; level <= last_level; level++) {
 137                 if (!staging && !(texture->dirty_level_mask & (1 << level)))
 138                         continue;
 139
 140                 /* The smaller the mipmap level, the less layers there are
 141                  * as far as 3D textures are concerned. */
 142                 max_layer = util_max_layer(&texture->resource.b.b, level);
 143                 checked_last_layer = last_layer < max_layer ? last_layer : max_layer;
 144
 145                 for (layer = first_layer; layer <= checked_last_layer; layer++) {
 146                         for (sample = first_sample; sample <= last_sample; sample++) {
 147                                 struct pipe_surface *zsurf, *cbsurf, surf_tmpl;
 148
 149                                 sctx->dbcb_copy_sample = sample;
 150                                 si_mark_atom_dirty(sctx, &sctx->db_render_state);
 151
 152                                 surf_tmpl.format = texture->resource.b.b.format;
 153                                 surf_tmpl.u.tex.level = level;
 154                                 surf_tmpl.u.tex.first_layer = layer;
 155                                 surf_tmpl.u.tex.last_layer = layer;
 156
 157                                 zsurf = ctx->create_surface(ctx, &texture->resource.b.b, &surf_tmpl);
 158
 159                                 surf_tmpl.format = flushed_depth_texture->resource.b.b.format;
 160                                 cbsurf = ctx->create_surface(ctx,
 161                                                 (struct pipe_resource*)flushed_depth_texture, &surf_tmpl);
 162
 163                                 si_blitter_begin(ctx, SI_DECOMPRESS);
 164                                 util_blitter_custom_depth_stencil(sctx->blitter, zsurf, cbsurf, 1 << sample,
 165                                                                   sctx->custom_dsa_flush, depth);
 166                                 si_blitter_end(ctx);
 167
 168                                 pipe_surface_reference(&zsurf, NULL);
 169                                 pipe_surface_reference(&cbsurf, NULL);
 170                         }
 171                 }
 172
 173                 /* The texture will always be dirty if some layers aren't flushed.
 174                  * I don't think this case can occur though. */
 175                 if (!staging &&
 176                     first_layer == 0 && last_layer == max_layer &&
 177                     first_sample == 0 && last_sample == max_sample) {
 178                         texture->dirty_level_mask &= ~(1 << level);
 179                 }
 180         }
 181
 182         sctx->dbcb_depth_copy_enabled = false;
 183         sctx->dbcb_stencil_copy_enabled = false;
 184         si_mark_atom_dirty(sctx, &sctx->db_render_state);
 185 }
 186
 187 static void si_blit_decompress_depth_in_place(struct si_context *sctx,
 188                                               struct r600_texture *texture,
 189                                               bool is_stencil_sampler,
 190                                               unsigned first_level, unsigned last_level,
 191                                               unsigned first_layer, unsigned last_layer)
 192 {
 193         struct pipe_surface *zsurf, surf_tmpl = {{0}};
 194         unsigned layer, max_layer, checked_last_layer, level;
 195         unsigned *dirty_level_mask;
 196
 197         if (is_stencil_sampler) {
 198                 sctx->db_flush_stencil_inplace = true;
 199                 dirty_level_mask = &texture->stencil_dirty_level_mask;
 200         } else {
 201                 sctx->db_flush_depth_inplace = true;
 202                 dirty_level_mask = &texture->dirty_level_mask;
 203         }
 204         si_mark_atom_dirty(sctx, &sctx->db_render_state);
 205
 206         surf_tmpl.format = texture->resource.b.b.format;
 207
 208         for (level = first_level; level <= last_level; level++) {
 209                 if (!(*dirty_level_mask & (1 << level)))
 210                         continue;
 211
 212                 surf_tmpl.u.tex.level = level;
 213
 214                 /* The smaller the mipmap level, the less layers there are
 215                  * as far as 3D textures are concerned. */
 216                 max_layer = util_max_layer(&texture->resource.b.b, level);
 217                 checked_last_layer = last_layer < max_layer ? last_layer : max_layer;
 218
 219                 for (layer = first_layer; layer <= checked_last_layer; layer++) {
 220                         surf_tmpl.u.tex.first_layer = layer;
 221                         surf_tmpl.u.tex.last_layer = layer;
 222
 223                         zsurf = sctx->b.b.create_surface(&sctx->b.b, &texture->resource.b.b, &surf_tmpl);
 224
 225                         si_blitter_begin(&sctx->b.b, SI_DECOMPRESS);
 226                         util_blitter_custom_depth_stencil(sctx->blitter, zsurf, NULL, ~0,
 227                                                           sctx->custom_dsa_flush,
 228                                                           1.0f);
 229                         si_blitter_end(&sctx->b.b);
 230
 231                         pipe_surface_reference(&zsurf, NULL);
 232                 }
 233
 234                 /* The texture will always be dirty if some layers aren't flushed.
 235                  * I don't think this case occurs often though. */
 236                 if (first_layer == 0 && last_layer == max_layer) {
 237                         *dirty_level_mask &= ~(1 << level);
 238                 }
 239         }
 240
 241         sctx->db_flush_depth_inplace = false;
 242         sctx->db_flush_stencil_inplace = false;
 243         si_mark_atom_dirty(sctx, &sctx->db_render_state);
 244 }
 245
 246 void si_flush_depth_textures(struct si_context *sctx,
 247                              struct si_textures_info *textures)
 248 {
 249         unsigned i;
 250         unsigned mask = textures->depth_texture_mask;
 251
 252         while (mask) {
 253                 struct pipe_sampler_view *view;
 254                 struct si_sampler_view *sview;
 255                 struct r600_texture *tex;
 256
 257                 i = u_bit_scan(&mask);
 258
 259                 view = textures->views.views[i];
 260                 assert(view);
 261                 sview = (struct si_sampler_view*)view;
 262
 263                 tex = (struct r600_texture *)view->texture;
 264                 assert(tex->is_depth && !tex->is_flushing_texture);
 265
 266                 si_blit_decompress_depth_in_place(sctx, tex,
 267                                                   sview->is_stencil_sampler,
 268                                                   view->u.tex.first_level, view->u.tex.last_level,
 269                                                   0, util_max_layer(&tex->resource.b.b, view->u.tex.first_level));
 270         }
 271 }
 272
 273 static void si_blit_decompress_color(struct pipe_context *ctx,
 274                 struct r600_texture *rtex,
 275                 unsigned first_level, unsigned last_level,
 276                 unsigned first_layer, unsigned last_layer)
 277 {
 278         struct si_context *sctx = (struct si_context *)ctx;
 279         unsigned layer, level, checked_last_layer, max_layer;
 280
 281         if (!rtex->dirty_level_mask)
 282                 return;
 283
 284         for (level = first_level; level <= last_level; level++) {
 285                 if (!(rtex->dirty_level_mask & (1 << level)))
 286                         continue;
 287
 288                 /* The smaller the mipmap level, the less layers there are
 289                  * as far as 3D textures are concerned. */
 290                 max_layer = util_max_layer(&rtex->resource.b.b, level);
 291                 checked_last_layer = last_layer < max_layer ? last_layer : max_layer;
 292
 293                 for (layer = first_layer; layer <= checked_last_layer; layer++) {
 294                         struct pipe_surface *cbsurf, surf_tmpl;
 295
 296                         surf_tmpl.format = rtex->resource.b.b.format;
 297                         surf_tmpl.u.tex.level = level;
 298                         surf_tmpl.u.tex.first_layer = layer;
 299                         surf_tmpl.u.tex.last_layer = layer;
 300                         cbsurf = ctx->create_surface(ctx, &rtex->resource.b.b, &surf_tmpl);
 301
 302                         si_blitter_begin(ctx, SI_DECOMPRESS);
 303                         util_blitter_custom_color(sctx->blitter, cbsurf,
 304                                 rtex->fmask.size ? sctx->custom_blend_decompress :
 305                                                    sctx->custom_blend_fastclear);
 306                         si_blitter_end(ctx);
 307
 308                         pipe_surface_reference(&cbsurf, NULL);
 309                 }
 310
 311                 /* The texture will always be dirty if some layers aren't flushed.
 312                  * I don't think this case occurs often though. */
 313                 if (first_layer == 0 && last_layer == max_layer) {
 314                         rtex->dirty_level_mask &= ~(1 << level);
 315                 }
 316         }
 317 }
 318
 319 void si_decompress_color_textures(struct si_context *sctx,
 320                                   struct si_textures_info *textures)
 321 {
 322         unsigned i;
 323         unsigned mask = textures->compressed_colortex_mask;
 324
 325         while (mask) {
 326                 struct pipe_sampler_view *view;
 327                 struct r600_texture *tex;
 328
 329                 i = u_bit_scan(&mask);
 330
 331                 view = textures->views.views[i];
 332                 assert(view);
 333
 334                 tex = (struct r600_texture *)view->texture;
 335                 assert(tex->cmask.size || tex->fmask.size || tex->dcc_buffer);
 336
 337                 si_blit_decompress_color(&sctx->b.b, tex,
 338                                          view->u.tex.first_level, view->u.tex.last_level,
 339                                          0, util_max_layer(&tex->resource.b.b, view->u.tex.first_level));
 340         }
 341 }
 342
 343 static void si_clear(struct pipe_context *ctx, unsigned buffers,
 344                      const union pipe_color_union *color,
 345                      double depth, unsigned stencil)
 346 {
 347         struct si_context *sctx = (struct si_context *)ctx;
 348         struct pipe_framebuffer_state *fb = &sctx->framebuffer.state;
 349         struct pipe_surface *zsbuf = fb->zsbuf;
 350         struct r600_texture *zstex =
 351                 zsbuf ? (struct r600_texture*)zsbuf->texture : NULL;
 352
 353         if (buffers & PIPE_CLEAR_COLOR) {
 354                 evergreen_do_fast_color_clear(&sctx->b, fb,
 355                                               &sctx->framebuffer.atom, &buffers,
 356                                               &sctx->framebuffer.dirty_cbufs,
 357                                               color);
 358                 if (!buffers)
 359                         return; /* all buffers have been fast cleared */
 360         }
 361
 362         if (buffers & PIPE_CLEAR_COLOR) {
 363                 int i;
 364
 365                 /* These buffers cannot use fast clear, make sure to disable expansion. */
 366                 for (i = 0; i < fb->nr_cbufs; i++) {
 367                         struct r600_texture *tex;
 368
 369                         /* If not clearing this buffer, skip. */
 370                         if (!(buffers & (PIPE_CLEAR_COLOR0 << i)))
 371                                 continue;
 372
 373                         if (!fb->cbufs[i])
 374                                 continue;
 375
 376                         tex = (struct r600_texture *)fb->cbufs[i]->texture;
 377                         if (tex->fmask.size == 0)
 378                                 tex->dirty_level_mask &= ~(1 << fb->cbufs[i]->u.tex.level);
 379                 }
 380         }
 381
 382         if (buffers & PIPE_CLEAR_DEPTH &&
 383             zstex && zstex->htile_buffer &&
 384             zsbuf->u.tex.level == 0 &&
 385             zsbuf->u.tex.first_layer == 0 &&
 386             zsbuf->u.tex.last_layer == util_max_layer(&zstex->resource.b.b, 0)) {
 387                 /* Need to disable EXPCLEAR temporarily if clearing
 388                  * to a new value. */
 389                 if (zstex->depth_cleared && zstex->depth_clear_value != depth) {
 390                         sctx->db_depth_disable_expclear = true;
 391                 }
 392
 393                 zstex->depth_clear_value = depth;
 394                 sctx->framebuffer.dirty_zsbuf = true;
 395                 si_mark_atom_dirty(sctx, &sctx->framebuffer.atom); /* updates DB_DEPTH_CLEAR */
 396                 sctx->db_depth_clear = true;
 397                 si_mark_atom_dirty(sctx, &sctx->db_render_state);
 398         }
 399
 400         si_blitter_begin(ctx, SI_CLEAR);
 401         util_blitter_clear(sctx->blitter, fb->width, fb->height,
 402                            util_framebuffer_get_num_layers(fb),
 403                            buffers, color, depth, stencil);
 404         si_blitter_end(ctx);
 405
 406         if (sctx->db_depth_clear) {
 407                 sctx->db_depth_clear = false;
 408                 sctx->db_depth_disable_expclear = false;
 409                 zstex->depth_cleared = true;
 410                 si_mark_atom_dirty(sctx, &sctx->db_render_state);
 411         }
 412 }
 413
 414 static void si_clear_render_target(struct pipe_context *ctx,
 415                                    struct pipe_surface *dst,
 416                                    const union pipe_color_union *color,
 417                                    unsigned dstx, unsigned dsty,
 418                                    unsigned width, unsigned height)
 419 {
 420         struct si_context *sctx = (struct si_context *)ctx;
 421
 422         si_blitter_begin(ctx, SI_CLEAR_SURFACE);
 423         util_blitter_clear_render_target(sctx->blitter, dst, color,
 424                                          dstx, dsty, width, height);
 425         si_blitter_end(ctx);
 426 }
 427
 428 static void si_clear_depth_stencil(struct pipe_context *ctx,
 429                                    struct pipe_surface *dst,
 430                                    unsigned clear_flags,
 431                                    double depth,
 432                                    unsigned stencil,
 433                                    unsigned dstx, unsigned dsty,
 434                                    unsigned width, unsigned height)
 435 {
 436         struct si_context *sctx = (struct si_context *)ctx;
 437
 438         si_blitter_begin(ctx, SI_CLEAR_SURFACE);
 439         util_blitter_clear_depth_stencil(sctx->blitter, dst, clear_flags, depth, stencil,
 440                                          dstx, dsty, width, height);
 441         si_blitter_end(ctx);
 442 }
 443
 444 /* Helper for decompressing a portion of a color or depth resource before
 445  * blitting if any decompression is needed.
 446  * The driver doesn't decompress resources automatically while u_blitter is
 447  * rendering. */
 448 static void si_decompress_subresource(struct pipe_context *ctx,
 449                                       struct pipe_resource *tex,
 450                                       unsigned level,
 451                                       unsigned first_layer, unsigned last_layer)
 452 {
 453         struct si_context *sctx = (struct si_context *)ctx;
 454         struct r600_texture *rtex = (struct r600_texture*)tex;
 455
 456         if (rtex->is_depth && !rtex->is_flushing_texture) {
 457                 si_blit_decompress_depth_in_place(sctx, rtex, false,
 458                                                   level, level,
 459                                                   first_layer, last_layer);
 460                 if (rtex->surface.flags & RADEON_SURF_SBUFFER)
 461                         si_blit_decompress_depth_in_place(sctx, rtex, true,
 462                                                           level, level,
 463                                                           first_layer, last_layer);
 464         } else if (rtex->fmask.size || rtex->cmask.size || rtex->dcc_buffer) {
 465                 si_blit_decompress_color(ctx, rtex, level, level,
 466                                          first_layer, last_layer);
 467         }
 468 }
 469
 470 struct texture_orig_info {
 471         unsigned format;
 472         unsigned width0;
 473         unsigned height0;
 474         unsigned npix_x;
 475         unsigned npix_y;
 476         unsigned npix0_x;
 477         unsigned npix0_y;
 478 };
 479
 480 void si_resource_copy_region(struct pipe_context *ctx,
 481                              struct pipe_resource *dst,
 482                              unsigned dst_level,
 483                              unsigned dstx, unsigned dsty, unsigned dstz,
 484                              struct pipe_resource *src,
 485                              unsigned src_level,
 486                              const struct pipe_box *src_box)
 487 {
 488         struct si_context *sctx = (struct si_context *)ctx;
 489         struct pipe_surface *dst_view, dst_templ;
 490         struct pipe_sampler_view src_templ, *src_view;
 491         unsigned dst_width, dst_height, src_width0, src_height0;
 492         unsigned src_force_level = 0;
 493         struct pipe_box sbox, dstbox;
 494
 495         /* Handle buffers first. */
 496         if (dst->target == PIPE_BUFFER && src->target == PIPE_BUFFER) {
 497                 si_copy_buffer(sctx, dst, src, dstx, src_box->x, src_box->width, false);
 498                 return;
 499         }
 500
 501         assert(u_max_sample(dst) == u_max_sample(src));
 502
 503         /* The driver doesn't decompress resources automatically while
 504          * u_blitter is rendering. */
 505         si_decompress_subresource(ctx, src, src_level,
 506                                   src_box->z, src_box->z + src_box->depth - 1);
 507
 508         dst_width = u_minify(dst->width0, dst_level);
 509         dst_height = u_minify(dst->height0, dst_level);
 510         src_width0 = src->width0;
 511         src_height0 = src->height0;
 512
 513         util_blitter_default_dst_texture(&dst_templ, dst, dst_level, dstz);
 514         util_blitter_default_src_texture(&src_templ, src, src_level);
 515
 516         if (util_format_is_compressed(src->format) ||
 517             util_format_is_compressed(dst->format)) {
 518                 unsigned blocksize = util_format_get_blocksize(src->format);
 519
 520                 if (blocksize == 8)
 521                         src_templ.format = PIPE_FORMAT_R16G16B16A16_UINT; /* 64-bit block */
 522                 else
 523                         src_templ.format = PIPE_FORMAT_R32G32B32A32_UINT; /* 128-bit block */
 524                 dst_templ.format = src_templ.format;
 525
 526                 dst_width = util_format_get_nblocksx(dst->format, dst_width);
 527                 dst_height = util_format_get_nblocksy(dst->format, dst_height);
 528                 src_width0 = util_format_get_nblocksx(src->format, src_width0);
 529                 src_height0 = util_format_get_nblocksy(src->format, src_height0);
 530
 531                 dstx = util_format_get_nblocksx(dst->format, dstx);
 532                 dsty = util_format_get_nblocksy(dst->format, dsty);
 533
 534                 sbox.x = util_format_get_nblocksx(src->format, src_box->x);
 535                 sbox.y = util_format_get_nblocksy(src->format, src_box->y);
 536                 sbox.z = src_box->z;
 537                 sbox.width = util_format_get_nblocksx(src->format, src_box->width);
 538                 sbox.height = util_format_get_nblocksy(src->format, src_box->height);
 539                 sbox.depth = src_box->depth;
 540                 src_box = &sbox;
 541
 542                 src_force_level = src_level;
 543         } else if (!util_blitter_is_copy_supported(sctx->blitter, dst, src) ||
 544                    /* also *8_SNORM has precision issues, use UNORM instead */
 545                    util_format_is_snorm8(src->format)) {
 546                 if (util_format_is_subsampled_422(src->format)) {
 547                         src_templ.format = PIPE_FORMAT_R8G8B8A8_UINT;
 548                         dst_templ.format = PIPE_FORMAT_R8G8B8A8_UINT;
 549
 550                         dst_width = util_format_get_nblocksx(dst->format, dst_width);
 551                         src_width0 = util_format_get_nblocksx(src->format, src_width0);
 552
 553                         dstx = util_format_get_nblocksx(dst->format, dstx);
 554
 555                         sbox = *src_box;
 556                         sbox.x = util_format_get_nblocksx(src->format, src_box->x);
 557                         sbox.width = util_format_get_nblocksx(src->format, src_box->width);
 558                         src_box = &sbox;
 559                 } else {
 560                         unsigned blocksize = util_format_get_blocksize(src->format);
 561
 562                         switch (blocksize) {
 563                         case 1:
 564                                 dst_templ.format = PIPE_FORMAT_R8_UNORM;
 565                                 src_templ.format = PIPE_FORMAT_R8_UNORM;
 566                                 break;
 567                         case 2:
 568                                 dst_templ.format = PIPE_FORMAT_R8G8_UNORM;
 569                                 src_templ.format = PIPE_FORMAT_R8G8_UNORM;
 570                                 break;
 571                         case 4:
 572                                 dst_templ.format = PIPE_FORMAT_R8G8B8A8_UNORM;
 573                                 src_templ.format = PIPE_FORMAT_R8G8B8A8_UNORM;
 574                                 break;
 575                         case 8:
 576                                 dst_templ.format = PIPE_FORMAT_R16G16B16A16_UINT;
 577                                 src_templ.format = PIPE_FORMAT_R16G16B16A16_UINT;
 578                                 break;
 579                         case 16:
 580                                 dst_templ.format = PIPE_FORMAT_R32G32B32A32_UINT;
 581                                 src_templ.format = PIPE_FORMAT_R32G32B32A32_UINT;
 582                                 break;
 583                         default:
 584                                 fprintf(stderr, "Unhandled format %s with blocksize %u\n",
 585                                         util_format_short_name(src->format), blocksize);
 586                                 assert(0);
 587                         }
 588                 }
 589         }
 590
 591         /* Initialize the surface. */
 592         dst_view = r600_create_surface_custom(ctx, dst, &dst_templ,
 593                                               dst_width, dst_height);
 594
 595         /* Initialize the sampler view. */
 596         src_view = si_create_sampler_view_custom(ctx, src, &src_templ,
 597                                                  src_width0, src_height0,
 598                                                  src_force_level);
 599
 600         u_box_3d(dstx, dsty, dstz, abs(src_box->width), abs(src_box->height),
 601                  abs(src_box->depth), &dstbox);
 602
 603         /* Copy. */
 604         si_blitter_begin(ctx, SI_COPY);
 605         util_blitter_blit_generic(sctx->blitter, dst_view, &dstbox,
 606                                   src_view, src_box, src_width0, src_height0,
 607                                   PIPE_MASK_RGBAZS, PIPE_TEX_FILTER_NEAREST, NULL,
 608                                   FALSE);
 609         si_blitter_end(ctx);
 610
 611         pipe_surface_reference(&dst_view, NULL);
 612         pipe_sampler_view_reference(&src_view, NULL);
 613 }
 614
 615 /* For MSAA integer resolving to work, we change the format to NORM using this function. */
 616 static enum pipe_format int_to_norm_format(enum pipe_format format)
 617 {
 618         switch (format) {
 619 #define REPLACE_FORMAT_SIGN(format,sign) \
 620         case PIPE_FORMAT_##format##_##sign##INT: \
 621                 return PIPE_FORMAT_##format##_##sign##NORM
 622 #define REPLACE_FORMAT(format) \
 623                 REPLACE_FORMAT_SIGN(format, U); \
 624                 REPLACE_FORMAT_SIGN(format, S)
 625
 626         REPLACE_FORMAT_SIGN(B10G10R10A2, U);
 627         REPLACE_FORMAT(R8);
 628         REPLACE_FORMAT(R8G8);
 629         REPLACE_FORMAT(R8G8B8X8);
 630         REPLACE_FORMAT(R8G8B8A8);
 631         REPLACE_FORMAT(A8);
 632         REPLACE_FORMAT(I8);
 633         REPLACE_FORMAT(L8);
 634         REPLACE_FORMAT(L8A8);
 635         REPLACE_FORMAT(R16);
 636         REPLACE_FORMAT(R16G16);
 637         REPLACE_FORMAT(R16G16B16X16);
 638         REPLACE_FORMAT(R16G16B16A16);
 639         REPLACE_FORMAT(A16);
 640         REPLACE_FORMAT(I16);
 641         REPLACE_FORMAT(L16);
 642         REPLACE_FORMAT(L16A16);
 643
 644 #undef REPLACE_FORMAT
 645 #undef REPLACE_FORMAT_SIGN
 646         default:
 647                 return format;
 648         }
 649 }
 650
 651 static bool do_hardware_msaa_resolve(struct pipe_context *ctx,
 652                                      const struct pipe_blit_info *info)
 653 {
 654         struct si_context *sctx = (struct si_context*)ctx;
 655         struct r600_texture *dst = (struct r600_texture*)info->dst.resource;
 656         unsigned dst_width = u_minify(info->dst.resource->width0, info->dst.level);
 657         unsigned dst_height = u_minify(info->dst.resource->height0, info->dst.level);
 658         enum pipe_format format = int_to_norm_format(info->dst.format);
 659         unsigned sample_mask = ~0;
 660
 661         if (info->src.resource->nr_samples > 1 &&
 662             info->dst.resource->nr_samples <= 1 &&
 663             util_max_layer(info->src.resource, 0) == 0 &&
 664             util_max_layer(info->dst.resource, info->dst.level) == 0 &&
 665             info->dst.format == info->src.format &&
 666             !util_format_is_pure_integer(format) &&
 667             !util_format_is_depth_or_stencil(format) &&
 668             !info->scissor_enable &&
 669             (info->mask & PIPE_MASK_RGBA) == PIPE_MASK_RGBA &&
 670             dst_width == info->src.resource->width0 &&
 671             dst_height == info->src.resource->height0 &&
 672             info->dst.box.x == 0 &&
 673             info->dst.box.y == 0 &&
 674             info->dst.box.width == dst_width &&
 675             info->dst.box.height == dst_height &&
 676             info->dst.box.depth == 1 &&
 677             info->src.box.x == 0 &&
 678             info->src.box.y == 0 &&
 679             info->src.box.width == dst_width &&
 680             info->src.box.height == dst_height &&
 681             info->src.box.depth == 1 &&
 682             dst->surface.level[info->dst.level].mode >= RADEON_SURF_MODE_1D &&
 683             !(dst->surface.flags & RADEON_SURF_SCANOUT) &&
 684             (!dst->cmask.size || !dst->dirty_level_mask) && /* dst cannot be fast-cleared */
 685             !dst->dcc_buffer) {
 686                 si_blitter_begin(ctx, SI_COLOR_RESOLVE |
 687                                  (info->render_condition_enable ? 0 : SI_DISABLE_RENDER_COND));
 688                 util_blitter_custom_resolve_color(sctx->blitter,
 689                                                   info->dst.resource, info->dst.level,
 690                                                   info->dst.box.z,
 691                                                   info->src.resource, info->src.box.z,
 692                                                   sample_mask, sctx->custom_blend_resolve,
 693                                                   format);
 694                 si_blitter_end(ctx);
 695                 return true;
 696         }
 697         return false;
 698 }
 699
 700 static void si_blit(struct pipe_context *ctx,
 701                     const struct pipe_blit_info *info)
 702 {
 703         struct si_context *sctx = (struct si_context*)ctx;
 704
 705         if (do_hardware_msaa_resolve(ctx, info)) {
 706                 return;
 707         }
 708
 709         assert(util_blitter_is_blit_supported(sctx->blitter, info));
 710
 711         /* The driver doesn't decompress resources automatically while
 712          * u_blitter is rendering. */
 713         si_decompress_subresource(ctx, info->src.resource, info->src.level,
 714                                   info->src.box.z,
 715                                   info->src.box.z + info->src.box.depth - 1);
 716
 717         if (sctx->screen->b.debug_flags & DBG_FORCE_DMA &&
 718             util_try_blit_via_copy_region(ctx, info))
 719                 return;
 720
 721         si_blitter_begin(ctx, SI_BLIT |
 722                          (info->render_condition_enable ? 0 : SI_DISABLE_RENDER_COND));
 723         util_blitter_blit(sctx->blitter, info);
 724         si_blitter_end(ctx);
 725 }
 726
 727 static void si_flush_resource(struct pipe_context *ctx,
 728                               struct pipe_resource *res)
 729 {
 730         struct r600_texture *rtex = (struct r600_texture*)res;
 731
 732         assert(res->target != PIPE_BUFFER);
 733
 734         if (!rtex->is_depth && rtex->cmask.size) {
 735                 si_blit_decompress_color(ctx, rtex, 0, res->last_level,
 736                                          0, util_max_layer(res, 0));
 737         }
 738 }
 739
 740 void si_init_blit_functions(struct si_context *sctx)
 741 {
 742         sctx->b.b.clear = si_clear;
 743         sctx->b.b.clear_render_target = si_clear_render_target;
 744         sctx->b.b.clear_depth_stencil = si_clear_depth_stencil;
 745         sctx->b.b.resource_copy_region = si_resource_copy_region;
 746         sctx->b.b.blit = si_blit;
 747         sctx->b.b.flush_resource = si_flush_resource;
 748         sctx->b.blit_decompress_depth = si_blit_decompress_depth;
 749 }