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