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_vertex_shader(sctx->blitter, sctx->vs_shader);
  61         util_blitter_save_vertex_elements(sctx->blitter, sctx->vertex_elements);
  62         if (sctx->queued.named.sample_mask) {
  63                 util_blitter_save_sample_mask(sctx->blitter,
  64                                               sctx->queued.named.sample_mask->sample_mask);
  65         }
  66         if (sctx->queued.named.viewport) {
  67                 util_blitter_save_viewport(sctx->blitter, &sctx->queued.named.viewport->viewport);
  68         }
  69         if (sctx->queued.named.scissor) {
  70                 util_blitter_save_scissor(sctx->blitter, &sctx->queued.named.scissor->scissor);
  71         }
  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                                 sctx->db_render_state.dirty = true;
 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         sctx->db_render_state.dirty = true;
 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         sctx->db_render_state.dirty = true;
 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         sctx->db_render_state.dirty = true;
 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         }
 344
 345         if (buffers & PIPE_CLEAR_COLOR) {
 346                 int i;
 347
 348                 /* These buffers cannot use fast clear, make sure to disable expansion. */
 349                 for (i = 0; i < fb->nr_cbufs; i++) {
 350                         struct r600_texture *tex;
 351
 352                         /* If not clearing this buffer, skip. */
 353                         if (!(buffers & (PIPE_CLEAR_COLOR0 << i)))
 354                                 continue;
 355
 356                         if (!fb->cbufs[i])
 357                                 continue;
 358
 359                         tex = (struct r600_texture *)fb->cbufs[i]->texture;
 360                         if (tex->fmask.size == 0)
 361                                 tex->dirty_level_mask &= ~(1 << fb->cbufs[i]->u.tex.level);
 362                 }
 363         }
 364
 365         if (buffers & PIPE_CLEAR_DEPTH &&
 366             zstex && zstex->htile_buffer &&
 367             zsbuf->u.tex.level == 0 &&
 368             zsbuf->u.tex.first_layer == 0 &&
 369             zsbuf->u.tex.last_layer == util_max_layer(&zstex->resource.b.b, 0)) {
 370                 /* Need to disable EXPCLEAR temporarily if clearing
 371                  * to a new value. */
 372                 if (zstex->depth_cleared && zstex->depth_clear_value != depth) {
 373                         sctx->db_depth_disable_expclear = true;
 374                 }
 375
 376                 zstex->depth_clear_value = depth;
 377                 sctx->framebuffer.atom.dirty = true; /* updates DB_DEPTH_CLEAR */
 378                 sctx->db_depth_clear = true;
 379                 sctx->db_render_state.dirty = true;
 380         }
 381
 382         si_blitter_begin(ctx, SI_CLEAR);
 383         util_blitter_clear(sctx->blitter, fb->width, fb->height,
 384                            util_framebuffer_get_num_layers(fb),
 385                            buffers, color, depth, stencil);
 386         si_blitter_end(ctx);
 387
 388         if (sctx->db_depth_clear) {
 389                 sctx->db_depth_clear = false;
 390                 sctx->db_depth_disable_expclear = false;
 391                 zstex->depth_cleared = true;
 392                 sctx->db_render_state.dirty = true;
 393         }
 394 }
 395
 396 static void si_clear_render_target(struct pipe_context *ctx,
 397                                    struct pipe_surface *dst,
 398                                    const union pipe_color_union *color,
 399                                    unsigned dstx, unsigned dsty,
 400                                    unsigned width, unsigned height)
 401 {
 402         struct si_context *sctx = (struct si_context *)ctx;
 403
 404         si_blitter_begin(ctx, SI_CLEAR_SURFACE);
 405         util_blitter_clear_render_target(sctx->blitter, dst, color,
 406                                          dstx, dsty, width, height);
 407         si_blitter_end(ctx);
 408 }
 409
 410 static void si_clear_depth_stencil(struct pipe_context *ctx,
 411                                    struct pipe_surface *dst,
 412                                    unsigned clear_flags,
 413                                    double depth,
 414                                    unsigned stencil,
 415                                    unsigned dstx, unsigned dsty,
 416                                    unsigned width, unsigned height)
 417 {
 418         struct si_context *sctx = (struct si_context *)ctx;
 419
 420         si_blitter_begin(ctx, SI_CLEAR_SURFACE);
 421         util_blitter_clear_depth_stencil(sctx->blitter, dst, clear_flags, depth, stencil,
 422                                          dstx, dsty, width, height);
 423         si_blitter_end(ctx);
 424 }
 425
 426 /* Helper for decompressing a portion of a color or depth resource before
 427  * blitting if any decompression is needed.
 428  * The driver doesn't decompress resources automatically while u_blitter is
 429  * rendering. */
 430 static void si_decompress_subresource(struct pipe_context *ctx,
 431                                       struct pipe_resource *tex,
 432                                       unsigned level,
 433                                       unsigned first_layer, unsigned last_layer)
 434 {
 435         struct si_context *sctx = (struct si_context *)ctx;
 436         struct r600_texture *rtex = (struct r600_texture*)tex;
 437
 438         if (rtex->is_depth && !rtex->is_flushing_texture) {
 439                 si_blit_decompress_depth_in_place(sctx, rtex,
 440                                                   level, level,
 441                                                   first_layer, last_layer);
 442         } else if (rtex->fmask.size || rtex->cmask.size) {
 443                 si_blit_decompress_color(ctx, rtex, level, level,
 444                                          first_layer, last_layer);
 445         }
 446 }
 447
 448 struct texture_orig_info {
 449         unsigned format;
 450         unsigned width0;
 451         unsigned height0;
 452         unsigned npix_x;
 453         unsigned npix_y;
 454         unsigned npix0_x;
 455         unsigned npix0_y;
 456 };
 457
 458 static void si_compressed_to_blittable(struct pipe_resource *tex,
 459                                        unsigned level,
 460                                        struct texture_orig_info *orig)
 461 {
 462         struct r600_texture *rtex = (struct r600_texture*)tex;
 463         unsigned pixsize = util_format_get_blocksize(rtex->resource.b.b.format);
 464         int new_format;
 465         int new_height, new_width;
 466
 467         orig->format = tex->format;
 468         orig->width0 = tex->width0;
 469         orig->height0 = tex->height0;
 470         orig->npix0_x = rtex->surface.level[0].npix_x;
 471         orig->npix0_y = rtex->surface.level[0].npix_y;
 472         orig->npix_x = rtex->surface.level[level].npix_x;
 473         orig->npix_y = rtex->surface.level[level].npix_y;
 474
 475         if (pixsize == 8)
 476                 new_format = PIPE_FORMAT_R16G16B16A16_UINT; /* 64-bit block */
 477         else
 478                 new_format = PIPE_FORMAT_R32G32B32A32_UINT; /* 128-bit block */
 479
 480         new_width = util_format_get_nblocksx(tex->format, orig->width0);
 481         new_height = util_format_get_nblocksy(tex->format, orig->height0);
 482
 483         tex->width0 = new_width;
 484         tex->height0 = new_height;
 485         tex->format = new_format;
 486         rtex->surface.level[0].npix_x = util_format_get_nblocksx(orig->format, orig->npix0_x);
 487         rtex->surface.level[0].npix_y = util_format_get_nblocksy(orig->format, orig->npix0_y);
 488         rtex->surface.level[level].npix_x = util_format_get_nblocksx(orig->format, orig->npix_x);
 489         rtex->surface.level[level].npix_y = util_format_get_nblocksy(orig->format, orig->npix_y);
 490
 491         /* By dividing the dimensions by 4, we effectively decrement
 492          * last_level by 2, therefore the last 2 mipmap levels disappear and
 493          * aren't blittable. Note that the last 3 mipmap levels (4x4, 2x2,
 494          * 1x1) have equal slice sizes, which is an important assumption
 495          * for this to work.
 496          *
 497          * In order to make the last 2 mipmap levels blittable, we have to
 498          * add the slice size of the last mipmap level to the texture
 499          * address, so that even though the hw thinks it reads last_level-2,
 500          * it will actually read last_level-1, and if we add the slice size*2,
 501          * it will read last_level. That's how this workaround works.
 502          */
 503         if (level > rtex->resource.b.b.last_level-2)
 504                 rtex->mipmap_shift = level - (rtex->resource.b.b.last_level-2);
 505 }
 506
 507 static void si_change_format(struct pipe_resource *tex,
 508                              unsigned level,
 509                              struct texture_orig_info *orig,
 510                              enum pipe_format format)
 511 {
 512         struct r600_texture *rtex = (struct r600_texture*)tex;
 513
 514         orig->format = tex->format;
 515         orig->width0 = tex->width0;
 516         orig->height0 = tex->height0;
 517         orig->npix0_x = rtex->surface.level[0].npix_x;
 518         orig->npix0_y = rtex->surface.level[0].npix_y;
 519         orig->npix_x = rtex->surface.level[level].npix_x;
 520         orig->npix_y = rtex->surface.level[level].npix_y;
 521
 522         tex->format = format;
 523 }
 524
 525 static void si_reset_blittable_to_orig(struct pipe_resource *tex,
 526                                        unsigned level,
 527                                        struct texture_orig_info *orig)
 528 {
 529         struct r600_texture *rtex = (struct r600_texture*)tex;
 530
 531         tex->format = orig->format;
 532         tex->width0 = orig->width0;
 533         tex->height0 = orig->height0;
 534         rtex->surface.level[0].npix_x = orig->npix0_x;
 535         rtex->surface.level[0].npix_y = orig->npix0_y;
 536         rtex->surface.level[level].npix_x = orig->npix_x;
 537         rtex->surface.level[level].npix_y = orig->npix_y;
 538         rtex->mipmap_shift = 0;
 539 }
 540
 541 void si_resource_copy_region(struct pipe_context *ctx,
 542                              struct pipe_resource *dst,
 543                              unsigned dst_level,
 544                              unsigned dstx, unsigned dsty, unsigned dstz,
 545                              struct pipe_resource *src,
 546                              unsigned src_level,
 547                              const struct pipe_box *src_box)
 548 {
 549         struct si_context *sctx = (struct si_context *)ctx;
 550         struct r600_texture *rdst = (struct r600_texture*)dst;
 551         struct pipe_surface *dst_view, dst_templ;
 552         struct pipe_sampler_view src_templ, *src_view;
 553         struct texture_orig_info orig_info[2];
 554         struct pipe_box sbox, dstbox;
 555         boolean restore_orig[2];
 556
 557         /* Fallback for buffers. */
 558         if (dst->target == PIPE_BUFFER && src->target == PIPE_BUFFER) {
 559                 si_copy_buffer(sctx, dst, src, dstx, src_box->x, src_box->width);
 560                 return;
 561         }
 562
 563         memset(orig_info, 0, sizeof(orig_info));
 564
 565         /* The driver doesn't decompress resources automatically while
 566          * u_blitter is rendering. */
 567         si_decompress_subresource(ctx, src, src_level,
 568                                   src_box->z, src_box->z + src_box->depth - 1);
 569
 570         restore_orig[0] = restore_orig[1] = FALSE;
 571
 572         if (util_format_is_compressed(src->format) &&
 573             util_format_is_compressed(dst->format)) {
 574                 si_compressed_to_blittable(src, src_level, &orig_info[0]);
 575                 restore_orig[0] = TRUE;
 576                 sbox.x = util_format_get_nblocksx(orig_info[0].format, src_box->x);
 577                 sbox.y = util_format_get_nblocksy(orig_info[0].format, src_box->y);
 578                 sbox.z = src_box->z;
 579                 sbox.width = util_format_get_nblocksx(orig_info[0].format, src_box->width);
 580                 sbox.height = util_format_get_nblocksy(orig_info[0].format, src_box->height);
 581                 sbox.depth = src_box->depth;
 582                 src_box = &sbox;
 583
 584                 si_compressed_to_blittable(dst, dst_level, &orig_info[1]);
 585                 restore_orig[1] = TRUE;
 586                 /* translate the dst box as well */
 587                 dstx = util_format_get_nblocksx(orig_info[1].format, dstx);
 588                 dsty = util_format_get_nblocksy(orig_info[1].format, dsty);
 589         } else if (!util_blitter_is_copy_supported(sctx->blitter, dst, src)) {
 590                 if (util_format_is_subsampled_422(src->format)) {
 591                         /* XXX untested */
 592                         si_change_format(src, src_level, &orig_info[0],
 593                                          PIPE_FORMAT_R8G8B8A8_UINT);
 594                         si_change_format(dst, dst_level, &orig_info[1],
 595                                          PIPE_FORMAT_R8G8B8A8_UINT);
 596
 597                         sbox = *src_box;
 598                         sbox.x = util_format_get_nblocksx(orig_info[0].format, src_box->x);
 599                         sbox.width = util_format_get_nblocksx(orig_info[0].format, src_box->width);
 600                         src_box = &sbox;
 601                         dstx = util_format_get_nblocksx(orig_info[1].format, dstx);
 602
 603                         restore_orig[0] = TRUE;
 604                         restore_orig[1] = TRUE;
 605                 } else {
 606                         unsigned blocksize = util_format_get_blocksize(src->format);
 607
 608                         switch (blocksize) {
 609                         case 1:
 610                                 si_change_format(src, src_level, &orig_info[0],
 611                                                 PIPE_FORMAT_R8_UNORM);
 612                                 si_change_format(dst, dst_level, &orig_info[1],
 613                                                 PIPE_FORMAT_R8_UNORM);
 614                                 break;
 615                         case 2:
 616                                 si_change_format(src, src_level, &orig_info[0],
 617                                                 PIPE_FORMAT_R8G8_UNORM);
 618                                 si_change_format(dst, dst_level, &orig_info[1],
 619                                                 PIPE_FORMAT_R8G8_UNORM);
 620                                 break;
 621                         case 4:
 622                                 si_change_format(src, src_level, &orig_info[0],
 623                                                 PIPE_FORMAT_R8G8B8A8_UNORM);
 624                                 si_change_format(dst, dst_level, &orig_info[1],
 625                                                 PIPE_FORMAT_R8G8B8A8_UNORM);
 626                                 break;
 627                         case 8:
 628                                 si_change_format(src, src_level, &orig_info[0],
 629                                                 PIPE_FORMAT_R16G16B16A16_UINT);
 630                                 si_change_format(dst, dst_level, &orig_info[1],
 631                                                 PIPE_FORMAT_R16G16B16A16_UINT);
 632                                 break;
 633                         case 16:
 634                                 si_change_format(src, src_level, &orig_info[0],
 635                                                 PIPE_FORMAT_R32G32B32A32_UINT);
 636                                 si_change_format(dst, dst_level, &orig_info[1],
 637                                                 PIPE_FORMAT_R32G32B32A32_UINT);
 638                                 break;
 639                         default:
 640                                 fprintf(stderr, "Unhandled format %s with blocksize %u\n",
 641                                         util_format_short_name(src->format), blocksize);
 642                                 assert(0);
 643                         }
 644                         restore_orig[0] = TRUE;
 645                         restore_orig[1] = TRUE;
 646                 }
 647         }
 648
 649         /* Initialize the surface. */
 650         util_blitter_default_dst_texture(&dst_templ, dst, dst_level, dstz);
 651         dst_view = r600_create_surface_custom(ctx, dst, &dst_templ,
 652                                               rdst->surface.level[dst_level].npix_x,
 653                                               rdst->surface.level[dst_level].npix_y);
 654
 655         /* Initialize the sampler view. */
 656         util_blitter_default_src_texture(&src_templ, src, src_level);
 657         src_view = ctx->create_sampler_view(ctx, src, &src_templ);
 658
 659         u_box_3d(dstx, dsty, dstz, abs(src_box->width), abs(src_box->height),
 660                  abs(src_box->depth), &dstbox);
 661
 662         /* Copy. */
 663         si_blitter_begin(ctx, SI_COPY);
 664         util_blitter_blit_generic(sctx->blitter, dst_view, &dstbox,
 665                                   src_view, src_box, src->width0, src->height0,
 666                                   PIPE_MASK_RGBAZS, PIPE_TEX_FILTER_NEAREST, NULL);
 667         si_blitter_end(ctx);
 668
 669         pipe_surface_reference(&dst_view, NULL);
 670         pipe_sampler_view_reference(&src_view, NULL);
 671
 672         if (restore_orig[0])
 673                 si_reset_blittable_to_orig(src, src_level, &orig_info[0]);
 674
 675         if (restore_orig[1])
 676                 si_reset_blittable_to_orig(dst, dst_level, &orig_info[1]);
 677 }
 678
 679 /* For MSAA integer resolving to work, we change the format to NORM using this function. */
 680 static enum pipe_format int_to_norm_format(enum pipe_format format)
 681 {
 682         switch (format) {
 683 #define REPLACE_FORMAT_SIGN(format,sign) \
 684         case PIPE_FORMAT_##format##_##sign##INT: \
 685                 return PIPE_FORMAT_##format##_##sign##NORM
 686 #define REPLACE_FORMAT(format) \
 687                 REPLACE_FORMAT_SIGN(format, U); \
 688                 REPLACE_FORMAT_SIGN(format, S)
 689
 690         REPLACE_FORMAT_SIGN(B10G10R10A2, U);
 691         REPLACE_FORMAT(R8);
 692         REPLACE_FORMAT(R8G8);
 693         REPLACE_FORMAT(R8G8B8X8);
 694         REPLACE_FORMAT(R8G8B8A8);
 695         REPLACE_FORMAT(A8);
 696         REPLACE_FORMAT(I8);
 697         REPLACE_FORMAT(L8);
 698         REPLACE_FORMAT(L8A8);
 699         REPLACE_FORMAT(R16);
 700         REPLACE_FORMAT(R16G16);
 701         REPLACE_FORMAT(R16G16B16X16);
 702         REPLACE_FORMAT(R16G16B16A16);
 703         REPLACE_FORMAT(A16);
 704         REPLACE_FORMAT(I16);
 705         REPLACE_FORMAT(L16);
 706         REPLACE_FORMAT(L16A16);
 707
 708 #undef REPLACE_FORMAT
 709 #undef REPLACE_FORMAT_SIGN
 710         default:
 711                 return format;
 712         }
 713 }
 714
 715 static bool do_hardware_msaa_resolve(struct pipe_context *ctx,
 716                                      const struct pipe_blit_info *info)
 717 {
 718         struct si_context *sctx = (struct si_context*)ctx;
 719         struct r600_texture *dst = (struct r600_texture*)info->dst.resource;
 720         unsigned dst_width = u_minify(info->dst.resource->width0, info->dst.level);
 721         unsigned dst_height = u_minify(info->dst.resource->height0, info->dst.level);
 722         enum pipe_format format = int_to_norm_format(info->dst.format);
 723         unsigned sample_mask = ~0;
 724
 725         if (info->src.resource->nr_samples > 1 &&
 726             info->dst.resource->nr_samples <= 1 &&
 727             util_max_layer(info->src.resource, 0) == 0 &&
 728             util_max_layer(info->dst.resource, info->dst.level) == 0 &&
 729             info->dst.format == info->src.format &&
 730             !util_format_is_pure_integer(format) &&
 731             !util_format_is_depth_or_stencil(format) &&
 732             !info->scissor_enable &&
 733             (info->mask & PIPE_MASK_RGBA) == PIPE_MASK_RGBA &&
 734             dst_width == info->src.resource->width0 &&
 735             dst_height == info->src.resource->height0 &&
 736             info->dst.box.x == 0 &&
 737             info->dst.box.y == 0 &&
 738             info->dst.box.width == dst_width &&
 739             info->dst.box.height == dst_height &&
 740             info->dst.box.depth == 1 &&
 741             info->src.box.x == 0 &&
 742             info->src.box.y == 0 &&
 743             info->src.box.width == dst_width &&
 744             info->src.box.height == dst_height &&
 745             info->src.box.depth == 1 &&
 746             dst->surface.level[info->dst.level].mode >= RADEON_SURF_MODE_1D &&
 747             !(dst->surface.flags & RADEON_SURF_SCANOUT) &&
 748             (!dst->cmask.size || !dst->dirty_level_mask) /* dst cannot be fast-cleared */) {
 749                 si_blitter_begin(ctx, SI_COLOR_RESOLVE |
 750                                  (info->render_condition_enable ? 0 : SI_DISABLE_RENDER_COND));
 751                 util_blitter_custom_resolve_color(sctx->blitter,
 752                                                   info->dst.resource, info->dst.level,
 753                                                   info->dst.box.z,
 754                                                   info->src.resource, info->src.box.z,
 755                                                   sample_mask, sctx->custom_blend_resolve,
 756                                                   format);
 757                 si_blitter_end(ctx);
 758                 return true;
 759         }
 760         return false;
 761 }
 762
 763 static void si_blit(struct pipe_context *ctx,
 764                     const struct pipe_blit_info *info)
 765 {
 766         struct si_context *sctx = (struct si_context*)ctx;
 767
 768         if (do_hardware_msaa_resolve(ctx, info)) {
 769                 return;
 770         }
 771
 772         assert(util_blitter_is_blit_supported(sctx->blitter, info));
 773
 774         /* The driver doesn't decompress resources automatically while
 775          * u_blitter is rendering. */
 776         si_decompress_subresource(ctx, info->src.resource, info->src.level,
 777                                   info->src.box.z,
 778                                   info->src.box.z + info->src.box.depth - 1);
 779
 780         if (sctx->screen->b.debug_flags & DBG_FORCE_DMA &&
 781             util_try_blit_via_copy_region(ctx, info))
 782                 return;
 783
 784         si_blitter_begin(ctx, SI_BLIT |
 785                          (info->render_condition_enable ? 0 : SI_DISABLE_RENDER_COND));
 786         util_blitter_blit(sctx->blitter, info);
 787         si_blitter_end(ctx);
 788 }
 789
 790 static void si_flush_resource(struct pipe_context *ctx,
 791                               struct pipe_resource *res)
 792 {
 793         struct r600_texture *rtex = (struct r600_texture*)res;
 794
 795         assert(res->target != PIPE_BUFFER);
 796
 797         if (!rtex->is_depth && rtex->cmask.size) {
 798                 si_blit_decompress_color(ctx, rtex, 0, res->last_level,
 799                                          0, util_max_layer(res, 0));
 800         }
 801 }
 802
 803 void si_init_blit_functions(struct si_context *sctx)
 804 {
 805         sctx->b.b.clear = si_clear;
 806         sctx->b.b.clear_render_target = si_clear_render_target;
 807         sctx->b.b.clear_depth_stencil = si_clear_depth_stencil;
 808         sctx->b.b.resource_copy_region = si_resource_copy_region;
 809         sctx->b.b.blit = si_blit;
 810         sctx->b.b.flush_resource = si_flush_resource;
 811         sctx->b.blit_decompress_depth = si_blit_decompress_depth;
 812 }