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