src/gallium/drivers/r600/r600_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 #include "r600_pipe.h"
  24 #include "util/u_surface.h"
  25 #include "util/u_blitter.h"
  26 #include "util/u_format.h"
  27 #include "evergreend.h"
  28
  29 enum r600_blitter_op /* bitmask */
  30 {
  31         R600_SAVE_FRAGMENT_STATE = 1,
  32         R600_SAVE_TEXTURES       = 2,
  33         R600_SAVE_FRAMEBUFFER    = 4,
  34         R600_DISABLE_RENDER_COND = 8,
  35
  36         R600_CLEAR         = R600_SAVE_FRAGMENT_STATE,
  37
  38         R600_CLEAR_SURFACE = R600_SAVE_FRAGMENT_STATE | R600_SAVE_FRAMEBUFFER,
  39
  40         R600_COPY_BUFFER   = R600_DISABLE_RENDER_COND,
  41
  42         R600_COPY_TEXTURE  = R600_SAVE_FRAGMENT_STATE | R600_SAVE_FRAMEBUFFER | R600_SAVE_TEXTURES |
  43                              R600_DISABLE_RENDER_COND,
  44
  45         R600_BLIT          = R600_SAVE_FRAGMENT_STATE | R600_SAVE_FRAMEBUFFER | R600_SAVE_TEXTURES |
  46                              R600_DISABLE_RENDER_COND,
  47
  48         R600_DECOMPRESS    = R600_SAVE_FRAGMENT_STATE | R600_SAVE_FRAMEBUFFER | R600_DISABLE_RENDER_COND,
  49
  50         R600_COLOR_RESOLVE = R600_SAVE_FRAGMENT_STATE | R600_SAVE_FRAMEBUFFER | R600_DISABLE_RENDER_COND
  51 };
  52
  53 static void r600_blitter_begin(struct pipe_context *ctx, enum r600_blitter_op op)
  54 {
  55         struct r600_context *rctx = (struct r600_context *)ctx;
  56
  57         r600_suspend_nontimer_queries(rctx);
  58
  59         util_blitter_save_vertex_buffer_slot(rctx->blitter, rctx->vertex_buffer_state.vb);
  60         util_blitter_save_vertex_elements(rctx->blitter, rctx->vertex_fetch_shader.cso);
  61         util_blitter_save_vertex_shader(rctx->blitter, rctx->vs_shader);
  62         util_blitter_save_so_targets(rctx->blitter, rctx->b.streamout.num_targets,
  63                                      (struct pipe_stream_output_target**)rctx->b.streamout.targets);
  64         util_blitter_save_rasterizer(rctx->blitter, rctx->rasterizer_state.cso);
  65
  66         if (op & R600_SAVE_FRAGMENT_STATE) {
  67                 util_blitter_save_viewport(rctx->blitter, &rctx->viewport.state);
  68                 util_blitter_save_scissor(rctx->blitter, &rctx->scissor.scissor);
  69                 util_blitter_save_fragment_shader(rctx->blitter, rctx->ps_shader);
  70                 util_blitter_save_blend(rctx->blitter, rctx->blend_state.cso);
  71                 util_blitter_save_depth_stencil_alpha(rctx->blitter, rctx->dsa_state.cso);
  72                 util_blitter_save_stencil_ref(rctx->blitter, &rctx->stencil_ref.pipe_state);
  73                 util_blitter_save_sample_mask(rctx->blitter, rctx->sample_mask.sample_mask);
  74         }
  75
  76         if (op & R600_SAVE_FRAMEBUFFER)
  77                 util_blitter_save_framebuffer(rctx->blitter, &rctx->framebuffer.state);
  78
  79         if (op & R600_SAVE_TEXTURES) {
  80                 util_blitter_save_fragment_sampler_states(
  81                         rctx->blitter, util_last_bit(rctx->samplers[PIPE_SHADER_FRAGMENT].states.enabled_mask),
  82                         (void**)rctx->samplers[PIPE_SHADER_FRAGMENT].states.states);
  83
  84                 util_blitter_save_fragment_sampler_views(
  85                         rctx->blitter, util_last_bit(rctx->samplers[PIPE_SHADER_FRAGMENT].views.enabled_mask),
  86                         (struct pipe_sampler_view**)rctx->samplers[PIPE_SHADER_FRAGMENT].views.views);
  87         }
  88
  89         if ((op & R600_DISABLE_RENDER_COND) && rctx->current_render_cond) {
  90            util_blitter_save_render_condition(rctx->blitter,
  91                                               rctx->current_render_cond,
  92                                               rctx->current_render_cond_cond,
  93                                               rctx->current_render_cond_mode);
  94         }
  95 }
  96
  97 static void r600_blitter_end(struct pipe_context *ctx)
  98 {
  99         struct r600_context *rctx = (struct r600_context *)ctx;
 100         r600_resume_nontimer_queries(rctx);
 101 }
 102
 103 static unsigned u_max_sample(struct pipe_resource *r)
 104 {
 105         return r->nr_samples ? r->nr_samples - 1 : 0;
 106 }
 107
 108 void r600_blit_decompress_depth(struct pipe_context *ctx,
 109                 struct r600_texture *texture,
 110                 struct r600_texture *staging,
 111                 unsigned first_level, unsigned last_level,
 112                 unsigned first_layer, unsigned last_layer,
 113                 unsigned first_sample, unsigned last_sample)
 114 {
 115         struct r600_context *rctx = (struct r600_context *)ctx;
 116         unsigned layer, level, sample, checked_last_layer, max_layer, max_sample;
 117         struct r600_texture *flushed_depth_texture = staging ?
 118                         staging : texture->flushed_depth_texture;
 119         const struct util_format_description *desc =
 120                 util_format_description(texture->resource.b.b.format);
 121         float depth;
 122
 123         if (!staging && !texture->dirty_level_mask)
 124                 return;
 125
 126         max_sample = u_max_sample(&texture->resource.b.b);
 127
 128         /* XXX Decompressing MSAA depth textures is broken on R6xx.
 129          * There is also a hardlock if CMASK and FMASK are not present.
 130          * Just skip this until we find out how to fix it. */
 131         if (rctx->b.chip_class == R600 && max_sample > 0) {
 132                 texture->dirty_level_mask = 0;
 133                 return;
 134         }
 135
 136         if (rctx->b.family == CHIP_RV610 || rctx->b.family == CHIP_RV630 ||
 137             rctx->b.family == CHIP_RV620 || rctx->b.family == CHIP_RV635)
 138                 depth = 0.0f;
 139         else
 140                 depth = 1.0f;
 141
 142         /* Enable decompression in DB_RENDER_CONTROL */
 143         rctx->db_misc_state.flush_depthstencil_through_cb = true;
 144         rctx->db_misc_state.copy_depth = util_format_has_depth(desc);
 145         rctx->db_misc_state.copy_stencil = util_format_has_stencil(desc);
 146         rctx->db_misc_state.copy_sample = first_sample;
 147         rctx->db_misc_state.atom.dirty = true;
 148
 149         for (level = first_level; level <= last_level; level++) {
 150                 if (!staging && !(texture->dirty_level_mask & (1 << level)))
 151                         continue;
 152
 153                 /* The smaller the mipmap level, the less layers there are
 154                  * as far as 3D textures are concerned. */
 155                 max_layer = util_max_layer(&texture->resource.b.b, level);
 156                 checked_last_layer = last_layer < max_layer ? last_layer : max_layer;
 157
 158                 for (layer = first_layer; layer <= checked_last_layer; layer++) {
 159                         for (sample = first_sample; sample <= last_sample; sample++) {
 160                                 struct pipe_surface *zsurf, *cbsurf, surf_tmpl;
 161
 162                                 if (sample != rctx->db_misc_state.copy_sample) {
 163                                         rctx->db_misc_state.copy_sample = sample;
 164                                         rctx->db_misc_state.atom.dirty = true;
 165                                 }
 166
 167                                 surf_tmpl.format = texture->resource.b.b.format;
 168                                 surf_tmpl.u.tex.level = level;
 169                                 surf_tmpl.u.tex.first_layer = layer;
 170                                 surf_tmpl.u.tex.last_layer = layer;
 171
 172                                 zsurf = ctx->create_surface(ctx, &texture->resource.b.b, &surf_tmpl);
 173
 174                                 surf_tmpl.format = flushed_depth_texture->resource.b.b.format;
 175                                 cbsurf = ctx->create_surface(ctx,
 176                                                 &flushed_depth_texture->resource.b.b, &surf_tmpl);
 177
 178                                 r600_blitter_begin(ctx, R600_DECOMPRESS);
 179                                 util_blitter_custom_depth_stencil(rctx->blitter, zsurf, cbsurf, 1 << sample,
 180                                                                   rctx->custom_dsa_flush, depth);
 181                                 r600_blitter_end(ctx);
 182
 183                                 pipe_surface_reference(&zsurf, NULL);
 184                                 pipe_surface_reference(&cbsurf, NULL);
 185                         }
 186                 }
 187
 188                 /* The texture will always be dirty if some layers or samples aren't flushed.
 189                  * I don't think this case occurs often though. */
 190                 if (!staging &&
 191                     first_layer == 0 && last_layer == max_layer &&
 192                     first_sample == 0 && last_sample == max_sample) {
 193                         texture->dirty_level_mask &= ~(1 << level);
 194                 }
 195         }
 196
 197         /* reenable compression in DB_RENDER_CONTROL */
 198         rctx->db_misc_state.flush_depthstencil_through_cb = false;
 199         rctx->db_misc_state.atom.dirty = true;
 200 }
 201
 202 static void r600_blit_decompress_depth_in_place(struct r600_context *rctx,
 203                                                 struct r600_texture *texture,
 204                                                 unsigned first_level, unsigned last_level,
 205                                                 unsigned first_layer, unsigned last_layer)
 206 {
 207         struct pipe_surface *zsurf, surf_tmpl = {{0}};
 208         unsigned layer, max_layer, checked_last_layer, level;
 209
 210         /* Enable decompression in DB_RENDER_CONTROL */
 211         rctx->db_misc_state.flush_depthstencil_in_place = true;
 212         rctx->db_misc_state.atom.dirty = true;
 213
 214         surf_tmpl.format = texture->resource.b.b.format;
 215
 216         for (level = first_level; level <= last_level; level++) {
 217                 if (!(texture->dirty_level_mask & (1 << level)))
 218                         continue;
 219
 220                 surf_tmpl.u.tex.level = level;
 221
 222                 /* The smaller the mipmap level, the less layers there are
 223                  * as far as 3D textures are concerned. */
 224                 max_layer = util_max_layer(&texture->resource.b.b, level);
 225                 checked_last_layer = last_layer < max_layer ? last_layer : max_layer;
 226
 227                 for (layer = first_layer; layer <= checked_last_layer; layer++) {
 228                         surf_tmpl.u.tex.first_layer = layer;
 229                         surf_tmpl.u.tex.last_layer = layer;
 230
 231                         zsurf = rctx->b.b.create_surface(&rctx->b.b, &texture->resource.b.b, &surf_tmpl);
 232
 233                         r600_blitter_begin(&rctx->b.b, R600_DECOMPRESS);
 234                         util_blitter_custom_depth_stencil(rctx->blitter, zsurf, NULL, ~0,
 235                                                           rctx->custom_dsa_flush, 1.0f);
 236                         r600_blitter_end(&rctx->b.b);
 237
 238                         pipe_surface_reference(&zsurf, NULL);
 239                 }
 240
 241                 /* The texture will always be dirty if some layers or samples aren't flushed.
 242                  * I don't think this case occurs often though. */
 243                 if (first_layer == 0 && last_layer == max_layer) {
 244                         texture->dirty_level_mask &= ~(1 << level);
 245                 }
 246         }
 247
 248         /* Disable decompression in DB_RENDER_CONTROL */
 249         rctx->db_misc_state.flush_depthstencil_in_place = false;
 250         rctx->db_misc_state.atom.dirty = true;
 251 }
 252
 253 void r600_decompress_depth_textures(struct r600_context *rctx,
 254                                struct r600_samplerview_state *textures)
 255 {
 256         unsigned i;
 257         unsigned depth_texture_mask = textures->compressed_depthtex_mask;
 258
 259         while (depth_texture_mask) {
 260                 struct pipe_sampler_view *view;
 261                 struct r600_texture *tex;
 262
 263                 i = u_bit_scan(&depth_texture_mask);
 264
 265                 view = &textures->views[i]->base;
 266                 assert(view);
 267
 268                 tex = (struct r600_texture *)view->texture;
 269                 assert(tex->is_depth && !tex->is_flushing_texture);
 270
 271                 if (rctx->b.chip_class >= EVERGREEN ||
 272                     r600_can_read_depth(tex)) {
 273                         r600_blit_decompress_depth_in_place(rctx, tex,
 274                                                    view->u.tex.first_level, view->u.tex.last_level,
 275                                                    0, util_max_layer(&tex->resource.b.b, view->u.tex.first_level));
 276                 } else {
 277                         r600_blit_decompress_depth(&rctx->b.b, tex, NULL,
 278                                                    view->u.tex.first_level, view->u.tex.last_level,
 279                                                    0, util_max_layer(&tex->resource.b.b, view->u.tex.first_level),
 280                                                    0, u_max_sample(&tex->resource.b.b));
 281                 }
 282         }
 283 }
 284
 285 static void r600_blit_decompress_color(struct pipe_context *ctx,
 286                 struct r600_texture *rtex,
 287                 unsigned first_level, unsigned last_level,
 288                 unsigned first_layer, unsigned last_layer)
 289 {
 290         struct r600_context *rctx = (struct r600_context *)ctx;
 291         unsigned layer, level, checked_last_layer, max_layer;
 292
 293         if (!rtex->dirty_level_mask)
 294                 return;
 295
 296         for (level = first_level; level <= last_level; level++) {
 297                 if (!(rtex->dirty_level_mask & (1 << level)))
 298                         continue;
 299
 300                 /* The smaller the mipmap level, the less layers there are
 301                  * as far as 3D textures are concerned. */
 302                 max_layer = util_max_layer(&rtex->resource.b.b, level);
 303                 checked_last_layer = last_layer < max_layer ? last_layer : max_layer;
 304
 305                 for (layer = first_layer; layer <= checked_last_layer; layer++) {
 306                         struct pipe_surface *cbsurf, surf_tmpl;
 307
 308                         surf_tmpl.format = rtex->resource.b.b.format;
 309                         surf_tmpl.u.tex.level = level;
 310                         surf_tmpl.u.tex.first_layer = layer;
 311                         surf_tmpl.u.tex.last_layer = layer;
 312                         cbsurf = ctx->create_surface(ctx, &rtex->resource.b.b, &surf_tmpl);
 313
 314                         r600_blitter_begin(ctx, R600_DECOMPRESS);
 315                         util_blitter_custom_color(rctx->blitter, cbsurf,
 316                                 rtex->fmask.size ? rctx->custom_blend_decompress : rctx->custom_blend_fastclear);
 317                         r600_blitter_end(ctx);
 318
 319                         pipe_surface_reference(&cbsurf, NULL);
 320                 }
 321
 322                 /* The texture will always be dirty if some layers aren't flushed.
 323                  * I don't think this case occurs often though. */
 324                 if (first_layer == 0 && last_layer == max_layer) {
 325                         rtex->dirty_level_mask &= ~(1 << level);
 326                 }
 327         }
 328 }
 329
 330 void r600_decompress_color_textures(struct r600_context *rctx,
 331                                     struct r600_samplerview_state *textures)
 332 {
 333         unsigned i;
 334         unsigned mask = textures->compressed_colortex_mask;
 335
 336         while (mask) {
 337                 struct pipe_sampler_view *view;
 338                 struct r600_texture *tex;
 339
 340                 i = u_bit_scan(&mask);
 341
 342                 view = &textures->views[i]->base;
 343                 assert(view);
 344
 345                 tex = (struct r600_texture *)view->texture;
 346                 assert(tex->cmask.size);
 347
 348                 r600_blit_decompress_color(&rctx->b.b, tex,
 349                                            view->u.tex.first_level, view->u.tex.last_level,
 350                                            0, util_max_layer(&tex->resource.b.b, view->u.tex.first_level));
 351         }
 352 }
 353
 354 /* Helper for decompressing a portion of a color or depth resource before
 355  * blitting if any decompression is needed.
 356  * The driver doesn't decompress resources automatically while u_blitter is
 357  * rendering. */
 358 static bool r600_decompress_subresource(struct pipe_context *ctx,
 359                                         struct pipe_resource *tex,
 360                                         unsigned level,
 361                                         unsigned first_layer, unsigned last_layer)
 362 {
 363         struct r600_context *rctx = (struct r600_context *)ctx;
 364         struct r600_texture *rtex = (struct r600_texture*)tex;
 365
 366         if (rtex->is_depth && !rtex->is_flushing_texture) {
 367                 if (rctx->b.chip_class >= EVERGREEN ||
 368                     r600_can_read_depth(rtex)) {
 369                         r600_blit_decompress_depth_in_place(rctx, rtex,
 370                                                    level, level,
 371                                                    first_layer, last_layer);
 372                 } else {
 373                         if (!r600_init_flushed_depth_texture(ctx, tex, NULL))
 374                                 return false; /* error */
 375
 376                         r600_blit_decompress_depth(ctx, rtex, NULL,
 377                                                    level, level,
 378                                                    first_layer, last_layer,
 379                                                    0, u_max_sample(tex));
 380                 }
 381         } else if (rtex->cmask.size) {
 382                 r600_blit_decompress_color(ctx, rtex, level, level,
 383                                            first_layer, last_layer);
 384         }
 385         return true;
 386 }
 387
 388 static boolean is_simple_msaa_resolve(const struct pipe_blit_info *info)
 389 {
 390         unsigned dst_width = u_minify(info->dst.resource->width0, info->dst.level);
 391         unsigned dst_height = u_minify(info->dst.resource->height0, info->dst.level);
 392         struct r600_texture *dst = (struct r600_texture*)info->dst.resource;
 393         unsigned dst_tile_mode = dst->surface.level[info->dst.level].mode;
 394
 395         return info->dst.resource->format == info->src.resource->format &&
 396                 info->dst.resource->format == info->dst.format &&
 397                 info->src.resource->format == info->src.format &&
 398                 !info->scissor_enable &&
 399                 info->mask == PIPE_MASK_RGBA &&
 400                 dst_width == info->src.resource->width0 &&
 401                 dst_height == info->src.resource->height0 &&
 402                 info->dst.box.x == 0 &&
 403                 info->dst.box.y == 0 &&
 404                 info->dst.box.width == dst_width &&
 405                 info->dst.box.height == dst_height &&
 406                 info->src.box.x == 0 &&
 407                 info->src.box.y == 0 &&
 408                 info->src.box.width == dst_width &&
 409                 info->src.box.height == dst_height &&
 410                 /* Dst must be tiled. If it's not, we have to use a temporary
 411                  * resource which is tiled. */
 412                 dst_tile_mode >= RADEON_SURF_MODE_1D;
 413 }
 414
 415 static void r600_clear_buffer(struct pipe_context *ctx, struct pipe_resource *dst,
 416                               unsigned offset, unsigned size, unsigned value);
 417
 418 static void evergreen_set_clear_color(struct pipe_surface *cbuf,
 419                                       const union pipe_color_union *color)
 420 {
 421         unsigned *clear_value = ((struct r600_texture *)cbuf->texture)->color_clear_value;
 422         union util_color uc;
 423
 424         memset(&uc, 0, sizeof(uc));
 425
 426         if (util_format_is_pure_uint(cbuf->format)) {
 427                 util_format_write_4ui(cbuf->format, color->ui, 0, &uc, 0, 0, 0, 1, 1);
 428         } else if (util_format_is_pure_sint(cbuf->format)) {
 429                 util_format_write_4i(cbuf->format, color->i, 0, &uc, 0, 0, 0, 1, 1);
 430         } else {
 431                 util_pack_color(color->f, cbuf->format, &uc);
 432         }
 433
 434         memcpy(clear_value, &uc, 2 * sizeof(uint32_t));
 435 }
 436
 437 static void evergreen_check_alloc_cmask(struct pipe_context *ctx,
 438                                         struct pipe_surface *cbuf)
 439 {
 440         struct r600_context *rctx = (struct r600_context *)ctx;
 441         struct r600_texture *tex = (struct r600_texture *)cbuf->texture;
 442         struct r600_surface *surf = (struct r600_surface *)cbuf;
 443
 444         if (tex->cmask_buffer)
 445                 return;
 446
 447         r600_texture_init_cmask(rctx->screen, tex);
 448
 449         /* update colorbuffer state bits */
 450         if (tex->cmask_buffer != NULL) {
 451                 uint64_t va = r600_resource_va(rctx->b.b.screen, &tex->cmask_buffer->b.b);
 452                 surf->cb_color_cmask = va >> 8;
 453                 surf->cb_color_cmask_slice = S_028C80_TILE_MAX(tex->cmask.slice_tile_max);
 454                 surf->cb_color_info |= S_028C70_FAST_CLEAR(1);
 455         }
 456 }
 457
 458 static bool can_fast_clear_color(struct pipe_context *ctx)
 459 {
 460         struct r600_context *rctx = (struct r600_context *)ctx;
 461         struct pipe_framebuffer_state *fb = &rctx->framebuffer.state;
 462         int i;
 463
 464         if (rctx->b.chip_class < EVERGREEN) {
 465                 return false;
 466         }
 467
 468         for (i = 0; i < fb->nr_cbufs; i++) {
 469                 struct r600_texture *tex = (struct r600_texture *)fb->cbufs[i]->texture;
 470
 471                 /* 128-bit formats are unuspported */
 472                 if (util_format_get_blocksizebits(fb->cbufs[i]->format) > 64) {
 473                         return false;
 474                 }
 475
 476                 /* the clear is allowed if all layers are bound */
 477                 if (fb->cbufs[i]->u.tex.first_layer != 0 ||
 478                     fb->cbufs[i]->u.tex.last_layer != util_max_layer(&tex->resource.b.b, 0)) {
 479                         return false;
 480                 }
 481
 482                 /* cannot clear mipmapped textures */
 483                 if (fb->cbufs[i]->texture->last_level != 0) {
 484                         return false;
 485                 }
 486
 487                 /* only supported on tiled surfaces */
 488                 if (tex->surface.level[0].mode < RADEON_SURF_MODE_1D) {
 489                     return false;
 490                 }
 491
 492                 /* ensure CMASK is enabled */
 493                 evergreen_check_alloc_cmask(ctx, fb->cbufs[i]);
 494                 if (tex->cmask.size == 0) {
 495                         return false;
 496                 }
 497         }
 498
 499         return true;
 500 }
 501
 502 static void r600_clear(struct pipe_context *ctx, unsigned buffers,
 503                        const union pipe_color_union *color,
 504                        double depth, unsigned stencil)
 505 {
 506         struct r600_context *rctx = (struct r600_context *)ctx;
 507         struct pipe_framebuffer_state *fb = &rctx->framebuffer.state;
 508
 509         /* fast clear on colorbuffers (EG+) */
 510         if ((buffers & PIPE_CLEAR_COLOR) && can_fast_clear_color(ctx)) {
 511                 int i;
 512
 513                 for (i = 0; i < fb->nr_cbufs; i++) {
 514                         struct r600_texture *tex = (struct r600_texture *)fb->cbufs[i]->texture;
 515
 516                         evergreen_set_clear_color(fb->cbufs[i], color);
 517                         r600_clear_buffer(ctx, &tex->cmask_buffer->b.b,
 518                                         tex->cmask.offset, tex->cmask.size, 0);
 519                         tex->dirty_level_mask |= 1 << fb->cbufs[i]->u.tex.level;
 520                 }
 521
 522                 rctx->framebuffer.atom.dirty = true;
 523
 524                 buffers &= ~PIPE_CLEAR_COLOR;
 525                 if (!buffers)
 526                         return;
 527         } else if (buffers & PIPE_CLEAR_COLOR) {
 528                 int i;
 529
 530                 /* cannot use fast clear, make sure to disable expansion */
 531                 for (i = 0; i < fb->nr_cbufs; i++) {
 532                         struct r600_texture *tex = (struct r600_texture *)fb->cbufs[i]->texture;
 533                         if (tex->fmask.size == 0)
 534                             tex->dirty_level_mask &= ~(1 << fb->cbufs[i]->u.tex.level);
 535                 }
 536         }
 537
 538         /* if hyperz enabled just clear hyperz */
 539         if (fb->zsbuf && (buffers & PIPE_CLEAR_DEPTH)) {
 540                 struct r600_texture *rtex;
 541                 unsigned level = fb->zsbuf->u.tex.level;
 542
 543                 rtex = (struct r600_texture*)fb->zsbuf->texture;
 544
 545                 /* We can't use hyperz fast clear if each slice of a texture
 546                  * array are clear to different value. To simplify code just
 547                  * disable fast clear for texture array.
 548                  */
 549                 /* Only use htile for first level */
 550                 if (rtex->htile && !level && rtex->surface.array_size == 1) {
 551                         if (rtex->depth_clear != depth) {
 552                                 rtex->depth_clear = depth;
 553                                 rctx->db_state.atom.dirty = true;
 554                         }
 555                         rctx->db_misc_state.htile_clear = true;
 556                         rctx->db_misc_state.atom.dirty = true;
 557                 }
 558         }
 559
 560         r600_blitter_begin(ctx, R600_CLEAR);
 561         util_blitter_clear(rctx->blitter, fb->width, fb->height,
 562                            buffers, color, depth, stencil);
 563         r600_blitter_end(ctx);
 564
 565         /* disable fast clear */
 566         if (rctx->db_misc_state.htile_clear) {
 567                 rctx->db_misc_state.htile_clear = false;
 568                 rctx->db_misc_state.atom.dirty = true;
 569         }
 570 }
 571
 572 static void r600_clear_render_target(struct pipe_context *ctx,
 573                                      struct pipe_surface *dst,
 574                                      const union pipe_color_union *color,
 575                                      unsigned dstx, unsigned dsty,
 576                                      unsigned width, unsigned height)
 577 {
 578         struct r600_context *rctx = (struct r600_context *)ctx;
 579
 580         r600_blitter_begin(ctx, R600_CLEAR_SURFACE);
 581         util_blitter_clear_render_target(rctx->blitter, dst, color,
 582                                          dstx, dsty, width, height);
 583         r600_blitter_end(ctx);
 584 }
 585
 586 static void r600_clear_depth_stencil(struct pipe_context *ctx,
 587                                      struct pipe_surface *dst,
 588                                      unsigned clear_flags,
 589                                      double depth,
 590                                      unsigned stencil,
 591                                      unsigned dstx, unsigned dsty,
 592                                      unsigned width, unsigned height)
 593 {
 594         struct r600_context *rctx = (struct r600_context *)ctx;
 595
 596         r600_blitter_begin(ctx, R600_CLEAR_SURFACE);
 597         util_blitter_clear_depth_stencil(rctx->blitter, dst, clear_flags, depth, stencil,
 598                                          dstx, dsty, width, height);
 599         r600_blitter_end(ctx);
 600 }
 601
 602 void r600_copy_buffer(struct pipe_context *ctx, struct pipe_resource *dst, unsigned dstx,
 603                       struct pipe_resource *src, const struct pipe_box *src_box)
 604 {
 605         struct r600_context *rctx = (struct r600_context*)ctx;
 606
 607         if (rctx->screen->has_cp_dma) {
 608                 r600_cp_dma_copy_buffer(rctx, dst, dstx, src, src_box->x, src_box->width);
 609         }
 610         else if (rctx->screen->has_streamout &&
 611                  /* Require 4-byte alignment. */
 612                  dstx % 4 == 0 && src_box->x % 4 == 0 && src_box->width % 4 == 0) {
 613
 614                 /* Flush both resources. */
 615                 r600_flag_resource_cache_flush(rctx, src);
 616                 r600_flag_resource_cache_flush(rctx, dst);
 617
 618                 r600_blitter_begin(ctx, R600_COPY_BUFFER);
 619                 util_blitter_copy_buffer(rctx->blitter, dst, dstx, src, src_box->x, src_box->width);
 620                 r600_blitter_end(ctx);
 621
 622                 /* Flush the dst in case the 3D engine has been prefetching the resource. */
 623                 r600_flag_resource_cache_flush(rctx, dst);
 624         } else {
 625                 util_resource_copy_region(ctx, dst, 0, dstx, 0, 0, src, 0, src_box);
 626         }
 627 }
 628
 629 static void r600_clear_buffer(struct pipe_context *ctx, struct pipe_resource *dst,
 630                               unsigned offset, unsigned size, unsigned value)
 631 {
 632         struct r600_context *rctx = (struct r600_context*)ctx;
 633
 634         if (rctx->screen->has_cp_dma &&
 635             rctx->b.chip_class >= EVERGREEN &&
 636             offset % 4 == 0 && size % 4 == 0) {
 637                 evergreen_cp_dma_clear_buffer(rctx, dst, offset, size, value);
 638         } else if (rctx->screen->has_streamout && offset % 4 == 0 && size % 4 == 0) {
 639                 union pipe_color_union clear_value;
 640                 clear_value.ui[0] = value;
 641
 642                 r600_flag_resource_cache_flush(rctx, dst);
 643
 644                 r600_blitter_begin(ctx, R600_DISABLE_RENDER_COND);
 645                 util_blitter_clear_buffer(rctx->blitter, dst, offset, size,
 646                                           1, &clear_value);
 647                 r600_blitter_end(ctx);
 648
 649                 /* Flush again in case the 3D engine has been prefetching the resource. */
 650                 r600_flag_resource_cache_flush(rctx, dst);
 651         } else {
 652                 uint32_t *map = r600_buffer_map_sync_with_rings(&rctx->b, r600_resource(dst),
 653                                                                  PIPE_TRANSFER_WRITE);
 654                 size /= 4;
 655                 for (unsigned i = 0; i < size; i++)
 656                         *map++ = value;
 657         }
 658 }
 659
 660 static bool util_format_is_subsampled_2x1_32bpp(enum pipe_format format)
 661 {
 662         const struct util_format_description *desc = util_format_description(format);
 663
 664         return desc->layout == UTIL_FORMAT_LAYOUT_SUBSAMPLED &&
 665                desc->block.width == 2 &&
 666                desc->block.height == 1 &&
 667                desc->block.bits == 32;
 668 }
 669
 670 static void r600_resource_copy_region(struct pipe_context *ctx,
 671                                       struct pipe_resource *dst,
 672                                       unsigned dst_level,
 673                                       unsigned dstx, unsigned dsty, unsigned dstz,
 674                                       struct pipe_resource *src,
 675                                       unsigned src_level,
 676                                       const struct pipe_box *src_box)
 677 {
 678         struct r600_context *rctx = (struct r600_context *)ctx;
 679         struct pipe_surface *dst_view, dst_templ;
 680         struct pipe_sampler_view src_templ, *src_view;
 681         unsigned dst_width, dst_height, src_width0, src_height0, src_widthFL, src_heightFL;
 682         struct pipe_box sbox, dstbox;
 683
 684         /* Handle buffers first. */
 685         if (dst->target == PIPE_BUFFER && src->target == PIPE_BUFFER) {
 686                 r600_copy_buffer(ctx, dst, dstx, src, src_box);
 687                 return;
 688         }
 689
 690         assert(u_max_sample(dst) == u_max_sample(src));
 691
 692         /* The driver doesn't decompress resources automatically while
 693          * u_blitter is rendering. */
 694         if (!r600_decompress_subresource(ctx, src, src_level,
 695                                          src_box->z, src_box->z + src_box->depth - 1)) {
 696                 return; /* error */
 697         }
 698
 699         dst_width = u_minify(dst->width0, dst_level);
 700         dst_height = u_minify(dst->height0, dst_level);
 701         src_width0 = src->width0;
 702         src_height0 = src->height0;
 703         src_widthFL = u_minify(src->width0, src_level);
 704         src_heightFL = u_minify(src->height0, src_level);
 705
 706         util_blitter_default_dst_texture(&dst_templ, dst, dst_level, dstz);
 707         util_blitter_default_src_texture(&src_templ, src, src_level);
 708
 709         if (util_format_is_compressed(src->format)) {
 710                 unsigned blocksize = util_format_get_blocksize(src->format);
 711
 712                 if (blocksize == 8)
 713                         src_templ.format = PIPE_FORMAT_R16G16B16A16_UINT; /* 64-bit block */
 714                 else
 715                         src_templ.format = PIPE_FORMAT_R32G32B32A32_UINT; /* 128-bit block */
 716                 dst_templ.format = src_templ.format;
 717
 718                 dst_width = util_format_get_nblocksx(dst->format, dst_width);
 719                 dst_height = util_format_get_nblocksy(dst->format, dst_height);
 720                 src_width0 = util_format_get_nblocksx(src->format, src_width0);
 721                 src_height0 = util_format_get_nblocksy(src->format, src_height0);
 722                 src_widthFL = util_format_get_nblocksx(src->format, src_widthFL);
 723                 src_heightFL = util_format_get_nblocksy(src->format, src_heightFL);
 724
 725                 dstx = util_format_get_nblocksx(dst->format, dstx);
 726                 dsty = util_format_get_nblocksy(dst->format, dsty);
 727
 728                 sbox.x = util_format_get_nblocksx(src->format, src_box->x);
 729                 sbox.y = util_format_get_nblocksy(src->format, src_box->y);
 730                 sbox.z = src_box->z;
 731                 sbox.width = util_format_get_nblocksx(src->format, src_box->width);
 732                 sbox.height = util_format_get_nblocksy(src->format, src_box->height);
 733                 sbox.depth = src_box->depth;
 734                 src_box = &sbox;
 735         } else if (!util_blitter_is_copy_supported(rctx->blitter, dst, src,
 736                                                    PIPE_MASK_RGBAZS)) {
 737                 if (util_format_is_subsampled_2x1_32bpp(src->format)) {
 738
 739                         src_templ.format = PIPE_FORMAT_R8G8B8A8_UINT;
 740                         dst_templ.format = PIPE_FORMAT_R8G8B8A8_UINT;
 741
 742                         dst_width = util_format_get_nblocksx(dst->format, dst_width);
 743                         src_width0 = util_format_get_nblocksx(src->format, src_width0);
 744                         src_widthFL = util_format_get_nblocksx(src->format, src_widthFL);
 745
 746                         dstx = util_format_get_nblocksx(dst->format, dstx);
 747
 748                         sbox = *src_box;
 749                         sbox.x = util_format_get_nblocksx(src->format, src_box->x);
 750                         sbox.width = util_format_get_nblocksx(src->format, src_box->width);
 751                         src_box = &sbox;
 752                 } else {
 753                         unsigned blocksize = util_format_get_blocksize(src->format);
 754
 755                         switch (blocksize) {
 756                         case 1:
 757                                 dst_templ.format = PIPE_FORMAT_R8_UNORM;
 758                                 src_templ.format = PIPE_FORMAT_R8_UNORM;
 759                                 break;
 760                         case 2:
 761                                 dst_templ.format = PIPE_FORMAT_R8G8_UNORM;
 762                                 src_templ.format = PIPE_FORMAT_R8G8_UNORM;
 763                                 break;
 764                         case 4:
 765                                 dst_templ.format = PIPE_FORMAT_R8G8B8A8_UNORM;
 766                                 src_templ.format = PIPE_FORMAT_R8G8B8A8_UNORM;
 767                                 break;
 768                         case 8:
 769                                 dst_templ.format = PIPE_FORMAT_R16G16B16A16_UINT;
 770                                 src_templ.format = PIPE_FORMAT_R16G16B16A16_UINT;
 771                                 break;
 772                         case 16:
 773                                 dst_templ.format = PIPE_FORMAT_R32G32B32A32_UINT;
 774                                 src_templ.format = PIPE_FORMAT_R32G32B32A32_UINT;
 775                                 break;
 776                         default:
 777                                 fprintf(stderr, "Unhandled format %s with blocksize %u\n",
 778                                         util_format_short_name(src->format), blocksize);
 779                                 assert(0);
 780                         }
 781                 }
 782         }
 783
 784         dst_view = r600_create_surface_custom(ctx, dst, &dst_templ, dst_width, dst_height);
 785
 786         if (rctx->b.chip_class >= EVERGREEN) {
 787                 src_view = evergreen_create_sampler_view_custom(ctx, src, &src_templ,
 788                                                                 src_width0, src_height0);
 789         } else {
 790                 src_view = r600_create_sampler_view_custom(ctx, src, &src_templ,
 791                                                            src_widthFL, src_heightFL);
 792         }
 793
 794         u_box_3d(dstx, dsty, dstz, abs(src_box->width), abs(src_box->height),
 795                  abs(src_box->depth), &dstbox);
 796
 797         /* Copy. */
 798         r600_blitter_begin(ctx, R600_COPY_TEXTURE);
 799         util_blitter_blit_generic(rctx->blitter, dst_view, &dstbox,
 800                                   src_view, src_box, src_width0, src_height0,
 801                                   PIPE_MASK_RGBAZS, PIPE_TEX_FILTER_NEAREST, NULL,
 802                                   TRUE);
 803         r600_blitter_end(ctx);
 804
 805         pipe_surface_reference(&dst_view, NULL);
 806         pipe_sampler_view_reference(&src_view, NULL);
 807 }
 808
 809 /* For MSAA integer resolving to work, we change the format to NORM using this function. */
 810 static enum pipe_format int_to_norm_format(enum pipe_format format)
 811 {
 812         switch (format) {
 813 #define REPLACE_FORMAT_SIGN(format,sign) \
 814         case PIPE_FORMAT_##format##_##sign##INT: \
 815                 return PIPE_FORMAT_##format##_##sign##NORM
 816 #define REPLACE_FORMAT(format) \
 817                 REPLACE_FORMAT_SIGN(format, U); \
 818                 REPLACE_FORMAT_SIGN(format, S)
 819
 820         REPLACE_FORMAT_SIGN(B10G10R10A2, U);
 821         REPLACE_FORMAT(R8);
 822         REPLACE_FORMAT(R8G8);
 823         REPLACE_FORMAT(R8G8B8X8);
 824         REPLACE_FORMAT(R8G8B8A8);
 825         REPLACE_FORMAT(A8);
 826         REPLACE_FORMAT(I8);
 827         REPLACE_FORMAT(L8);
 828         REPLACE_FORMAT(L8A8);
 829         REPLACE_FORMAT(R16);
 830         REPLACE_FORMAT(R16G16);
 831         REPLACE_FORMAT(R16G16B16X16);
 832         REPLACE_FORMAT(R16G16B16A16);
 833         REPLACE_FORMAT(A16);
 834         REPLACE_FORMAT(I16);
 835         REPLACE_FORMAT(L16);
 836         REPLACE_FORMAT(L16A16);
 837
 838 #undef REPLACE_FORMAT
 839 #undef REPLACE_FORMAT_SIGN
 840         default:
 841                 return format;
 842         }
 843 }
 844
 845 static void r600_msaa_color_resolve(struct pipe_context *ctx,
 846                               const struct pipe_blit_info *info)
 847 {
 848         struct r600_context *rctx = (struct r600_context *)ctx;
 849         struct pipe_screen *screen = ctx->screen;
 850         struct pipe_resource *tmp, templ;
 851         struct pipe_blit_info blit;
 852         unsigned sample_mask =
 853                 rctx->b.chip_class == CAYMAN ? ~0 :
 854                 ((1ull << MAX2(1, info->src.resource->nr_samples)) - 1);
 855
 856         assert(info->src.level == 0);
 857         assert(info->src.box.depth == 1);
 858         assert(info->dst.box.depth == 1);
 859
 860         if (is_simple_msaa_resolve(info)) {
 861                 r600_blitter_begin(ctx, R600_COLOR_RESOLVE);
 862                 util_blitter_custom_resolve_color(rctx->blitter,
 863                                                   info->dst.resource, info->dst.level,
 864                                                   info->dst.box.z,
 865                                                   info->src.resource, info->src.box.z,
 866                                                   sample_mask, rctx->custom_blend_resolve,
 867                                                   int_to_norm_format(info->dst.format));
 868                 r600_blitter_end(ctx);
 869                 return;
 870         }
 871
 872         /* resolve into a temporary texture, then blit */
 873         templ.target = PIPE_TEXTURE_2D;
 874         templ.format = info->src.resource->format;
 875         templ.width0 = info->src.resource->width0;
 876         templ.height0 = info->src.resource->height0;
 877         templ.depth0 = 1;
 878         templ.array_size = 1;
 879         templ.last_level = 0;
 880         templ.nr_samples = 0;
 881         templ.usage = PIPE_USAGE_STATIC;
 882         templ.bind = PIPE_BIND_RENDER_TARGET | PIPE_BIND_SAMPLER_VIEW;
 883         templ.flags = R600_RESOURCE_FLAG_FORCE_TILING; /* dst must not have a linear layout */
 884
 885         tmp = screen->resource_create(screen, &templ);
 886
 887         /* resolve */
 888         r600_blitter_begin(ctx, R600_COLOR_RESOLVE);
 889         util_blitter_custom_resolve_color(rctx->blitter,
 890                                           tmp, 0, 0,
 891                                           info->src.resource, info->src.box.z,
 892                                           sample_mask, rctx->custom_blend_resolve,
 893                                           int_to_norm_format(tmp->format));
 894         r600_blitter_end(ctx);
 895
 896         /* blit */
 897         blit = *info;
 898         blit.src.resource = tmp;
 899         blit.src.box.z = 0;
 900
 901         r600_blitter_begin(ctx, R600_BLIT);
 902         util_blitter_blit(rctx->blitter, &blit);
 903         r600_blitter_end(ctx);
 904
 905         pipe_resource_reference(&tmp, NULL);
 906 }
 907
 908 static void r600_blit(struct pipe_context *ctx,
 909                       const struct pipe_blit_info *info)
 910 {
 911         struct r600_context *rctx = (struct r600_context*)ctx;
 912
 913         assert(util_blitter_is_blit_supported(rctx->blitter, info));
 914
 915         if (info->src.resource->nr_samples > 1 &&
 916             info->dst.resource->nr_samples <= 1 &&
 917             !util_format_is_depth_or_stencil(info->src.resource->format) &&
 918             !util_format_is_pure_integer(int_to_norm_format(info->src.resource->format))) {
 919                 r600_msaa_color_resolve(ctx, info);
 920                 return;
 921         }
 922
 923         /* The driver doesn't decompress resources automatically while
 924          * u_blitter is rendering. */
 925         if (!r600_decompress_subresource(ctx, info->src.resource, info->src.level,
 926                                          info->src.box.z,
 927                                          info->src.box.z + info->src.box.depth - 1)) {
 928                 return; /* error */
 929         }
 930
 931         r600_blitter_begin(ctx, R600_BLIT);
 932         util_blitter_blit(rctx->blitter, info);
 933         r600_blitter_end(ctx);
 934 }
 935
 936 static void r600_flush_resource(struct pipe_context *ctx,
 937                                 struct pipe_resource *res)
 938 {
 939         struct r600_texture *rtex = (struct r600_texture*)res;
 940
 941         assert(res->target != PIPE_BUFFER);
 942
 943         if (!rtex->is_depth && rtex->cmask.size) {
 944                 r600_blit_decompress_color(ctx, rtex, 0, res->last_level,
 945                                            0, res->array_size - 1);
 946         }
 947 }
 948
 949 void r600_init_blit_functions(struct r600_context *rctx)
 950 {
 951         rctx->b.b.clear = r600_clear;
 952         rctx->b.b.clear_render_target = r600_clear_render_target;
 953         rctx->b.b.clear_depth_stencil = r600_clear_depth_stencil;
 954         rctx->b.b.resource_copy_region = r600_resource_copy_region;
 955         rctx->b.b.blit = r600_blit;
 956         rctx->b.b.flush_resource = r600_flush_resource;
 957         rctx->b.clear_buffer = r600_clear_buffer;
 958 }