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