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->b);
  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_geometry_shader(rctx->blitter, rctx->gs_shader);
  63         util_blitter_save_so_targets(rctx->blitter, rctx->b.streamout.num_targets,
  64                                      (struct pipe_stream_output_target**)rctx->b.streamout.targets);
  65         util_blitter_save_rasterizer(rctx->blitter, rctx->rasterizer_state.cso);
  66
  67         if (op & R600_SAVE_FRAGMENT_STATE) {
  68                 util_blitter_save_viewport(rctx->blitter, &rctx->viewport[0].state);
  69                 util_blitter_save_scissor(rctx->blitter, &rctx->scissor[0].scissor);
  70                 util_blitter_save_fragment_shader(rctx->blitter, rctx->ps_shader);
  71                 util_blitter_save_blend(rctx->blitter, rctx->blend_state.cso);
  72                 util_blitter_save_depth_stencil_alpha(rctx->blitter, rctx->dsa_state.cso);
  73                 util_blitter_save_stencil_ref(rctx->blitter, &rctx->stencil_ref.pipe_state);
  74                 util_blitter_save_sample_mask(rctx->blitter, rctx->sample_mask.sample_mask);
  75         }
  76
  77         if (op & R600_SAVE_FRAMEBUFFER)
  78                 util_blitter_save_framebuffer(rctx->blitter, &rctx->framebuffer.state);
  79
  80         if (op & R600_SAVE_TEXTURES) {
  81                 util_blitter_save_fragment_sampler_states(
  82                         rctx->blitter, util_last_bit(rctx->samplers[PIPE_SHADER_FRAGMENT].states.enabled_mask),
  83                         (void**)rctx->samplers[PIPE_SHADER_FRAGMENT].states.states);
  84
  85                 util_blitter_save_fragment_sampler_views(
  86                         rctx->blitter, util_last_bit(rctx->samplers[PIPE_SHADER_FRAGMENT].views.enabled_mask),
  87                         (struct pipe_sampler_view**)rctx->samplers[PIPE_SHADER_FRAGMENT].views.views);
  88         }
  89
  90         if ((op & R600_DISABLE_RENDER_COND) && rctx->b.current_render_cond) {
  91            util_blitter_save_render_condition(rctx->blitter,
  92                                               rctx->b.current_render_cond,
  93                                               rctx->b.current_render_cond_cond,
  94                                               rctx->b.current_render_cond_mode);
  95         }
  96 }
  97
  98 static void r600_blitter_end(struct pipe_context *ctx)
  99 {
 100         struct r600_context *rctx = (struct r600_context *)ctx;
 101         r600_resume_nontimer_queries(&rctx->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 r600_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 r600_context *rctx = (struct r600_context *)ctx;
 117         unsigned layer, level, sample, checked_last_layer, max_layer, max_sample;
 118         struct r600_texture *flushed_depth_texture = staging ?
 119                         staging : texture->flushed_depth_texture;
 120         const struct util_format_description *desc =
 121                 util_format_description(texture->resource.b.b.format);
 122         float depth;
 123
 124         if (!staging && !texture->dirty_level_mask)
 125                 return;
 126
 127         max_sample = u_max_sample(&texture->resource.b.b);
 128
 129         /* XXX Decompressing MSAA depth textures is broken on R6xx.
 130          * There is also a hardlock if CMASK and FMASK are not present.
 131          * Just skip this until we find out how to fix it. */
 132         if (rctx->b.chip_class == R600 && max_sample > 0) {
 133                 texture->dirty_level_mask = 0;
 134                 return;
 135         }
 136
 137         if (rctx->b.family == CHIP_RV610 || rctx->b.family == CHIP_RV630 ||
 138             rctx->b.family == CHIP_RV620 || rctx->b.family == CHIP_RV635)
 139                 depth = 0.0f;
 140         else
 141                 depth = 1.0f;
 142
 143         /* Enable decompression in DB_RENDER_CONTROL */
 144         rctx->db_misc_state.flush_depthstencil_through_cb = true;
 145         rctx->db_misc_state.copy_depth = util_format_has_depth(desc);
 146         rctx->db_misc_state.copy_stencil = util_format_has_stencil(desc);
 147         rctx->db_misc_state.copy_sample = first_sample;
 148         rctx->db_misc_state.atom.dirty = true;
 149
 150         for (level = first_level; level <= last_level; level++) {
 151                 if (!staging && !(texture->dirty_level_mask & (1 << level)))
 152                         continue;
 153
 154                 /* The smaller the mipmap level, the less layers there are
 155                  * as far as 3D textures are concerned. */
 156                 max_layer = util_max_layer(&texture->resource.b.b, level);
 157                 checked_last_layer = last_layer < max_layer ? last_layer : max_layer;
 158
 159                 for (layer = first_layer; layer <= checked_last_layer; layer++) {
 160                         for (sample = first_sample; sample <= last_sample; sample++) {
 161                                 struct pipe_surface *zsurf, *cbsurf, surf_tmpl;
 162
 163                                 if (sample != rctx->db_misc_state.copy_sample) {
 164                                         rctx->db_misc_state.copy_sample = sample;
 165                                         rctx->db_misc_state.atom.dirty = true;
 166                                 }
 167
 168                                 surf_tmpl.format = texture->resource.b.b.format;
 169                                 surf_tmpl.u.tex.level = level;
 170                                 surf_tmpl.u.tex.first_layer = layer;
 171                                 surf_tmpl.u.tex.last_layer = layer;
 172
 173                                 zsurf = ctx->create_surface(ctx, &texture->resource.b.b, &surf_tmpl);
 174
 175                                 surf_tmpl.format = flushed_depth_texture->resource.b.b.format;
 176                                 cbsurf = ctx->create_surface(ctx,
 177                                                 &flushed_depth_texture->resource.b.b, &surf_tmpl);
 178
 179                                 r600_blitter_begin(ctx, R600_DECOMPRESS);
 180                                 util_blitter_custom_depth_stencil(rctx->blitter, zsurf, cbsurf, 1 << sample,
 181                                                                   rctx->custom_dsa_flush, depth);
 182                                 r600_blitter_end(ctx);
 183
 184                                 pipe_surface_reference(&zsurf, NULL);
 185                                 pipe_surface_reference(&cbsurf, NULL);
 186                         }
 187                 }
 188
 189                 /* The texture will always be dirty if some layers or samples aren't flushed.
 190                  * I don't think this case occurs often though. */
 191                 if (!staging &&
 192                     first_layer == 0 && last_layer == max_layer &&
 193                     first_sample == 0 && last_sample == max_sample) {
 194                         texture->dirty_level_mask &= ~(1 << level);
 195                 }
 196         }
 197
 198         /* reenable compression in DB_RENDER_CONTROL */
 199         rctx->db_misc_state.flush_depthstencil_through_cb = false;
 200         rctx->db_misc_state.atom.dirty = true;
 201 }
 202
 203 static void r600_blit_decompress_depth_in_place(struct r600_context *rctx,
 204                                                 struct r600_texture *texture,
 205                                                 unsigned first_level, unsigned last_level,
 206                                                 unsigned first_layer, unsigned last_layer)
 207 {
 208         struct pipe_surface *zsurf, surf_tmpl = {{0}};
 209         unsigned layer, max_layer, checked_last_layer, level;
 210
 211         /* Enable decompression in DB_RENDER_CONTROL */
 212         rctx->db_misc_state.flush_depthstencil_in_place = true;
 213         rctx->db_misc_state.atom.dirty = true;
 214
 215         surf_tmpl.format = texture->resource.b.b.format;
 216
 217         for (level = first_level; level <= last_level; level++) {
 218                 if (!(texture->dirty_level_mask & (1 << level)))
 219                         continue;
 220
 221                 surf_tmpl.u.tex.level = level;
 222
 223                 /* The smaller the mipmap level, the less layers there are
 224                  * as far as 3D textures are concerned. */
 225                 max_layer = util_max_layer(&texture->resource.b.b, level);
 226                 checked_last_layer = last_layer < max_layer ? last_layer : max_layer;
 227
 228                 for (layer = first_layer; layer <= checked_last_layer; layer++) {
 229                         surf_tmpl.u.tex.first_layer = layer;
 230                         surf_tmpl.u.tex.last_layer = layer;
 231
 232                         zsurf = rctx->b.b.create_surface(&rctx->b.b, &texture->resource.b.b, &surf_tmpl);
 233
 234                         r600_blitter_begin(&rctx->b.b, R600_DECOMPRESS);
 235                         util_blitter_custom_depth_stencil(rctx->blitter, zsurf, NULL, ~0,
 236                                                           rctx->custom_dsa_flush, 1.0f);
 237                         r600_blitter_end(&rctx->b.b);
 238
 239                         pipe_surface_reference(&zsurf, NULL);
 240                 }
 241
 242                 /* The texture will always be dirty if some layers or samples aren't flushed.
 243                  * I don't think this case occurs often though. */
 244                 if (first_layer == 0 && last_layer == max_layer) {
 245                         texture->dirty_level_mask &= ~(1 << level);
 246                 }
 247         }
 248
 249         /* Disable decompression in DB_RENDER_CONTROL */
 250         rctx->db_misc_state.flush_depthstencil_in_place = false;
 251         rctx->db_misc_state.atom.dirty = true;
 252 }
 253
 254 void r600_decompress_depth_textures(struct r600_context *rctx,
 255                                struct r600_samplerview_state *textures)
 256 {
 257         unsigned i;
 258         unsigned depth_texture_mask = textures->compressed_depthtex_mask;
 259
 260         while (depth_texture_mask) {
 261                 struct pipe_sampler_view *view;
 262                 struct r600_texture *tex;
 263
 264                 i = u_bit_scan(&depth_texture_mask);
 265
 266                 view = &textures->views[i]->base;
 267                 assert(view);
 268
 269                 tex = (struct r600_texture *)view->texture;
 270                 assert(tex->is_depth && !tex->is_flushing_texture);
 271
 272                 if (rctx->b.chip_class >= EVERGREEN ||
 273                     r600_can_read_depth(tex)) {
 274                         r600_blit_decompress_depth_in_place(rctx, tex,
 275                                                    view->u.tex.first_level, view->u.tex.last_level,
 276                                                    0, util_max_layer(&tex->resource.b.b, view->u.tex.first_level));
 277                 } else {
 278                         r600_blit_decompress_depth(&rctx->b.b, tex, NULL,
 279                                                    view->u.tex.first_level, view->u.tex.last_level,
 280                                                    0, util_max_layer(&tex->resource.b.b, view->u.tex.first_level),
 281                                                    0, u_max_sample(&tex->resource.b.b));
 282                 }
 283         }
 284 }
 285
 286 static void r600_blit_decompress_color(struct pipe_context *ctx,
 287                 struct r600_texture *rtex,
 288                 unsigned first_level, unsigned last_level,
 289                 unsigned first_layer, unsigned last_layer)
 290 {
 291         struct r600_context *rctx = (struct r600_context *)ctx;
 292         unsigned layer, level, checked_last_layer, max_layer;
 293
 294         if (!rtex->dirty_level_mask)
 295                 return;
 296
 297         for (level = first_level; level <= last_level; level++) {
 298                 if (!(rtex->dirty_level_mask & (1 << level)))
 299                         continue;
 300
 301                 /* The smaller the mipmap level, the less layers there are
 302                  * as far as 3D textures are concerned. */
 303                 max_layer = util_max_layer(&rtex->resource.b.b, level);
 304                 checked_last_layer = last_layer < max_layer ? last_layer : max_layer;
 305
 306                 for (layer = first_layer; layer <= checked_last_layer; layer++) {
 307                         struct pipe_surface *cbsurf, surf_tmpl;
 308
 309                         surf_tmpl.format = rtex->resource.b.b.format;
 310                         surf_tmpl.u.tex.level = level;
 311                         surf_tmpl.u.tex.first_layer = layer;
 312                         surf_tmpl.u.tex.last_layer = layer;
 313                         cbsurf = ctx->create_surface(ctx, &rtex->resource.b.b, &surf_tmpl);
 314
 315                         r600_blitter_begin(ctx, R600_DECOMPRESS);
 316                         util_blitter_custom_color(rctx->blitter, cbsurf,
 317                                 rtex->fmask.size ? rctx->custom_blend_decompress : rctx->custom_blend_fastclear);
 318                         r600_blitter_end(ctx);
 319
 320                         pipe_surface_reference(&cbsurf, NULL);
 321                 }
 322
 323                 /* The texture will always be dirty if some layers aren't flushed.
 324                  * I don't think this case occurs often though. */
 325                 if (first_layer == 0 && last_layer == max_layer) {
 326                         rtex->dirty_level_mask &= ~(1 << level);
 327                 }
 328         }
 329 }
 330
 331 void r600_decompress_color_textures(struct r600_context *rctx,
 332                                     struct r600_samplerview_state *textures)
 333 {
 334         unsigned i;
 335         unsigned mask = textures->compressed_colortex_mask;
 336
 337         while (mask) {
 338                 struct pipe_sampler_view *view;
 339                 struct r600_texture *tex;
 340
 341                 i = u_bit_scan(&mask);
 342
 343                 view = &textures->views[i]->base;
 344                 assert(view);
 345
 346                 tex = (struct r600_texture *)view->texture;
 347                 assert(tex->cmask.size);
 348
 349                 r600_blit_decompress_color(&rctx->b.b, tex,
 350                                            view->u.tex.first_level, view->u.tex.last_level,
 351                                            0, util_max_layer(&tex->resource.b.b, view->u.tex.first_level));
 352         }
 353 }
 354
 355 /* Helper for decompressing a portion of a color or depth resource before
 356  * blitting if any decompression is needed.
 357  * The driver doesn't decompress resources automatically while u_blitter is
 358  * rendering. */
 359 static bool r600_decompress_subresource(struct pipe_context *ctx,
 360                                         struct pipe_resource *tex,
 361                                         unsigned level,
 362                                         unsigned first_layer, unsigned last_layer)
 363 {
 364         struct r600_context *rctx = (struct r600_context *)ctx;
 365         struct r600_texture *rtex = (struct r600_texture*)tex;
 366
 367         if (rtex->is_depth && !rtex->is_flushing_texture) {
 368                 if (rctx->b.chip_class >= EVERGREEN ||
 369                     r600_can_read_depth(rtex)) {
 370                         r600_blit_decompress_depth_in_place(rctx, rtex,
 371                                                    level, level,
 372                                                    first_layer, last_layer);
 373                 } else {
 374                         if (!r600_init_flushed_depth_texture(ctx, tex, NULL))
 375                                 return false; /* error */
 376
 377                         r600_blit_decompress_depth(ctx, rtex, NULL,
 378                                                    level, level,
 379                                                    first_layer, last_layer,
 380                                                    0, u_max_sample(tex));
 381                 }
 382         } else if (rtex->cmask.size) {
 383                 r600_blit_decompress_color(ctx, rtex, level, level,
 384                                            first_layer, last_layer);
 385         }
 386         return true;
 387 }
 388
 389 static void r600_clear(struct pipe_context *ctx, unsigned buffers,
 390                        const union pipe_color_union *color,
 391                        double depth, unsigned stencil)
 392 {
 393         struct r600_context *rctx = (struct r600_context *)ctx;
 394         struct pipe_framebuffer_state *fb = &rctx->framebuffer.state;
 395
 396         if (buffers & PIPE_CLEAR_COLOR && rctx->b.chip_class >= EVERGREEN) {
 397                 evergreen_do_fast_color_clear(&rctx->b, fb, &rctx->framebuffer.atom,
 398                                               &buffers, color);
 399         }
 400
 401         if (buffers & PIPE_CLEAR_COLOR) {
 402                 int i;
 403
 404                 /* These buffers cannot use fast clear, make sure to disable expansion. */
 405                 for (i = 0; i < fb->nr_cbufs; i++) {
 406                         struct r600_texture *tex;
 407
 408                         /* If not clearing this buffer, skip. */
 409                         if (!(buffers & (PIPE_CLEAR_COLOR0 << i)))
 410                                 continue;
 411
 412                         if (!fb->cbufs[i])
 413                                 continue;
 414
 415                         tex = (struct r600_texture *)fb->cbufs[i]->texture;
 416                         if (tex->fmask.size == 0)
 417                                 tex->dirty_level_mask &= ~(1 << fb->cbufs[i]->u.tex.level);
 418                 }
 419         }
 420
 421         /* if hyperz enabled just clear hyperz */
 422         if (fb->zsbuf && (buffers & PIPE_CLEAR_DEPTH)) {
 423                 struct r600_texture *rtex;
 424                 unsigned level = fb->zsbuf->u.tex.level;
 425
 426                 rtex = (struct r600_texture*)fb->zsbuf->texture;
 427
 428                 /* We can't use hyperz fast clear if each slice of a texture
 429                  * array are clear to different value. To simplify code just
 430                  * disable fast clear for texture array.
 431                  */
 432                 /* Only use htile for first level */
 433                 if (rtex->htile_buffer && !level && rtex->surface.array_size == 1) {
 434                         if (rtex->depth_clear_value != depth) {
 435                                 rtex->depth_clear_value = depth;
 436                                 rctx->db_state.atom.dirty = true;
 437                         }
 438                         rctx->db_misc_state.htile_clear = true;
 439                         rctx->db_misc_state.atom.dirty = true;
 440                 }
 441         }
 442
 443         r600_blitter_begin(ctx, R600_CLEAR);
 444         util_blitter_clear(rctx->blitter, fb->width, fb->height, 1,
 445                            buffers, color, depth, stencil);
 446         r600_blitter_end(ctx);
 447
 448         /* disable fast clear */
 449         if (rctx->db_misc_state.htile_clear) {
 450                 rctx->db_misc_state.htile_clear = false;
 451                 rctx->db_misc_state.atom.dirty = true;
 452         }
 453 }
 454
 455 static void r600_clear_render_target(struct pipe_context *ctx,
 456                                      struct pipe_surface *dst,
 457                                      const union pipe_color_union *color,
 458                                      unsigned dstx, unsigned dsty,
 459                                      unsigned width, unsigned height)
 460 {
 461         struct r600_context *rctx = (struct r600_context *)ctx;
 462
 463         r600_blitter_begin(ctx, R600_CLEAR_SURFACE);
 464         util_blitter_clear_render_target(rctx->blitter, dst, color,
 465                                          dstx, dsty, width, height);
 466         r600_blitter_end(ctx);
 467 }
 468
 469 static void r600_clear_depth_stencil(struct pipe_context *ctx,
 470                                      struct pipe_surface *dst,
 471                                      unsigned clear_flags,
 472                                      double depth,
 473                                      unsigned stencil,
 474                                      unsigned dstx, unsigned dsty,
 475                                      unsigned width, unsigned height)
 476 {
 477         struct r600_context *rctx = (struct r600_context *)ctx;
 478
 479         r600_blitter_begin(ctx, R600_CLEAR_SURFACE);
 480         util_blitter_clear_depth_stencil(rctx->blitter, dst, clear_flags, depth, stencil,
 481                                          dstx, dsty, width, height);
 482         r600_blitter_end(ctx);
 483 }
 484
 485 static void r600_copy_buffer(struct pipe_context *ctx, struct pipe_resource *dst, unsigned dstx,
 486                              struct pipe_resource *src, const struct pipe_box *src_box)
 487 {
 488         struct r600_context *rctx = (struct r600_context*)ctx;
 489
 490         if (rctx->screen->b.has_cp_dma) {
 491                 r600_cp_dma_copy_buffer(rctx, dst, dstx, src, src_box->x, src_box->width);
 492         }
 493         else if (rctx->screen->b.has_streamout &&
 494                  /* Require 4-byte alignment. */
 495                  dstx % 4 == 0 && src_box->x % 4 == 0 && src_box->width % 4 == 0) {
 496
 497                 r600_blitter_begin(ctx, R600_COPY_BUFFER);
 498                 util_blitter_copy_buffer(rctx->blitter, dst, dstx, src, src_box->x, src_box->width);
 499                 r600_blitter_end(ctx);
 500         } else {
 501                 util_resource_copy_region(ctx, dst, 0, dstx, 0, 0, src, 0, src_box);
 502         }
 503 }
 504
 505 /**
 506  * Global buffers are not really resources, they are are actually offsets
 507  * into a single global resource (r600_screen::global_pool).  The means
 508  * they don't have their own cs_buf handle, so they cannot be passed
 509  * to r600_copy_buffer() and must be handled separately.
 510  *
 511  * XXX: It should be possible to implement this function using
 512  * r600_copy_buffer() by passing the memory_pool resource as both src
 513  * and dst and updating dstx and src_box to point to the correct offsets.
 514  * This would likely perform better than the current implementation.
 515  */
 516 static void r600_copy_global_buffer(struct pipe_context *ctx,
 517                                     struct pipe_resource *dst, unsigned
 518                                     dstx, struct pipe_resource *src,
 519                                     const struct pipe_box *src_box)
 520 {
 521         struct pipe_box dst_box; struct pipe_transfer *src_pxfer,
 522         *dst_pxfer;
 523
 524         u_box_1d(dstx, src_box->width, &dst_box);
 525         void *src_ptr = ctx->transfer_map(ctx, src, 0, PIPE_TRANSFER_READ,
 526                                           src_box, &src_pxfer);
 527         void *dst_ptr = ctx->transfer_map(ctx, dst, 0, PIPE_TRANSFER_WRITE,
 528                                           &dst_box, &dst_pxfer);
 529         memcpy(dst_ptr, src_ptr, src_box->width);
 530
 531         ctx->transfer_unmap(ctx, src_pxfer);
 532         ctx->transfer_unmap(ctx, dst_pxfer);
 533 }
 534
 535 static void r600_clear_buffer(struct pipe_context *ctx, struct pipe_resource *dst,
 536                               unsigned offset, unsigned size, unsigned value)
 537 {
 538         struct r600_context *rctx = (struct r600_context*)ctx;
 539
 540         if (rctx->screen->b.has_cp_dma &&
 541             rctx->b.chip_class >= EVERGREEN &&
 542             offset % 4 == 0 && size % 4 == 0) {
 543                 evergreen_cp_dma_clear_buffer(rctx, dst, offset, size, value);
 544         } else if (rctx->screen->b.has_streamout && offset % 4 == 0 && size % 4 == 0) {
 545                 union pipe_color_union clear_value;
 546                 clear_value.ui[0] = value;
 547
 548                 r600_blitter_begin(ctx, R600_DISABLE_RENDER_COND);
 549                 util_blitter_clear_buffer(rctx->blitter, dst, offset, size,
 550                                           1, &clear_value);
 551                 r600_blitter_end(ctx);
 552         } else {
 553                 uint32_t *map = r600_buffer_map_sync_with_rings(&rctx->b, r600_resource(dst),
 554                                                                  PIPE_TRANSFER_WRITE);
 555                 size /= 4;
 556                 for (unsigned i = 0; i < size; i++)
 557                         *map++ = value;
 558         }
 559 }
 560
 561 static bool util_format_is_subsampled_2x1_32bpp(enum pipe_format format)
 562 {
 563         const struct util_format_description *desc = util_format_description(format);
 564
 565         return desc->layout == UTIL_FORMAT_LAYOUT_SUBSAMPLED &&
 566                desc->block.width == 2 &&
 567                desc->block.height == 1 &&
 568                desc->block.bits == 32;
 569 }
 570
 571 static void r600_resource_copy_region(struct pipe_context *ctx,
 572                                       struct pipe_resource *dst,
 573                                       unsigned dst_level,
 574                                       unsigned dstx, unsigned dsty, unsigned dstz,
 575                                       struct pipe_resource *src,
 576                                       unsigned src_level,
 577                                       const struct pipe_box *src_box)
 578 {
 579         struct r600_context *rctx = (struct r600_context *)ctx;
 580         struct pipe_surface *dst_view, dst_templ;
 581         struct pipe_sampler_view src_templ, *src_view;
 582         unsigned dst_width, dst_height, src_width0, src_height0, src_widthFL, src_heightFL;
 583         unsigned src_force_level = 0;
 584         struct pipe_box sbox, dstbox;
 585
 586         /* Handle buffers first. */
 587         if (dst->target == PIPE_BUFFER && src->target == PIPE_BUFFER) {
 588                 if ((src->bind & PIPE_BIND_GLOBAL) ||
 589                                         (dst->bind & PIPE_BIND_GLOBAL)) {
 590                         r600_copy_global_buffer(ctx, dst, dstx, src, src_box);
 591                 } else {
 592                         r600_copy_buffer(ctx, dst, dstx, src, src_box);
 593                 }
 594                 return;
 595         }
 596
 597         assert(u_max_sample(dst) == u_max_sample(src));
 598
 599         /* The driver doesn't decompress resources automatically while
 600          * u_blitter is rendering. */
 601         if (!r600_decompress_subresource(ctx, src, src_level,
 602                                          src_box->z, src_box->z + src_box->depth - 1)) {
 603                 return; /* error */
 604         }
 605
 606         dst_width = u_minify(dst->width0, dst_level);
 607         dst_height = u_minify(dst->height0, dst_level);
 608         src_width0 = src->width0;
 609         src_height0 = src->height0;
 610         src_widthFL = u_minify(src->width0, src_level);
 611         src_heightFL = u_minify(src->height0, src_level);
 612
 613         util_blitter_default_dst_texture(&dst_templ, dst, dst_level, dstz);
 614         util_blitter_default_src_texture(&src_templ, src, src_level);
 615
 616         if (util_format_is_compressed(src->format)) {
 617                 unsigned blocksize = util_format_get_blocksize(src->format);
 618
 619                 if (blocksize == 8)
 620                         src_templ.format = PIPE_FORMAT_R16G16B16A16_UINT; /* 64-bit block */
 621                 else
 622                         src_templ.format = PIPE_FORMAT_R32G32B32A32_UINT; /* 128-bit block */
 623                 dst_templ.format = src_templ.format;
 624
 625                 dst_width = util_format_get_nblocksx(dst->format, dst_width);
 626                 dst_height = util_format_get_nblocksy(dst->format, dst_height);
 627                 src_width0 = util_format_get_nblocksx(src->format, src_width0);
 628                 src_height0 = util_format_get_nblocksy(src->format, src_height0);
 629                 src_widthFL = util_format_get_nblocksx(src->format, src_widthFL);
 630                 src_heightFL = util_format_get_nblocksy(src->format, src_heightFL);
 631
 632                 dstx = util_format_get_nblocksx(dst->format, dstx);
 633                 dsty = util_format_get_nblocksy(dst->format, dsty);
 634
 635                 sbox.x = util_format_get_nblocksx(src->format, src_box->x);
 636                 sbox.y = util_format_get_nblocksy(src->format, src_box->y);
 637                 sbox.z = src_box->z;
 638                 sbox.width = util_format_get_nblocksx(src->format, src_box->width);
 639                 sbox.height = util_format_get_nblocksy(src->format, src_box->height);
 640                 sbox.depth = src_box->depth;
 641                 src_box = &sbox;
 642
 643                 src_force_level = src_level;
 644         } else if (!util_blitter_is_copy_supported(rctx->blitter, dst, src)) {
 645                 if (util_format_is_subsampled_2x1_32bpp(src->format)) {
 646
 647                         src_templ.format = PIPE_FORMAT_R8G8B8A8_UINT;
 648                         dst_templ.format = PIPE_FORMAT_R8G8B8A8_UINT;
 649
 650                         dst_width = util_format_get_nblocksx(dst->format, dst_width);
 651                         src_width0 = util_format_get_nblocksx(src->format, src_width0);
 652                         src_widthFL = util_format_get_nblocksx(src->format, src_widthFL);
 653
 654                         dstx = util_format_get_nblocksx(dst->format, dstx);
 655
 656                         sbox = *src_box;
 657                         sbox.x = util_format_get_nblocksx(src->format, src_box->x);
 658                         sbox.width = util_format_get_nblocksx(src->format, src_box->width);
 659                         src_box = &sbox;
 660                 } else {
 661                         unsigned blocksize = util_format_get_blocksize(src->format);
 662
 663                         switch (blocksize) {
 664                         case 1:
 665                                 dst_templ.format = PIPE_FORMAT_R8_UNORM;
 666                                 src_templ.format = PIPE_FORMAT_R8_UNORM;
 667                                 break;
 668                         case 2:
 669                                 dst_templ.format = PIPE_FORMAT_R8G8_UNORM;
 670                                 src_templ.format = PIPE_FORMAT_R8G8_UNORM;
 671                                 break;
 672                         case 4:
 673                                 dst_templ.format = PIPE_FORMAT_R8G8B8A8_UNORM;
 674                                 src_templ.format = PIPE_FORMAT_R8G8B8A8_UNORM;
 675                                 break;
 676                         case 8:
 677                                 dst_templ.format = PIPE_FORMAT_R16G16B16A16_UINT;
 678                                 src_templ.format = PIPE_FORMAT_R16G16B16A16_UINT;
 679                                 break;
 680                         case 16:
 681                                 dst_templ.format = PIPE_FORMAT_R32G32B32A32_UINT;
 682                                 src_templ.format = PIPE_FORMAT_R32G32B32A32_UINT;
 683                                 break;
 684                         default:
 685                                 fprintf(stderr, "Unhandled format %s with blocksize %u\n",
 686                                         util_format_short_name(src->format), blocksize);
 687                                 assert(0);
 688                         }
 689                 }
 690         }
 691
 692         dst_view = r600_create_surface_custom(ctx, dst, &dst_templ, dst_width, dst_height);
 693
 694         if (rctx->b.chip_class >= EVERGREEN) {
 695                 src_view = evergreen_create_sampler_view_custom(ctx, src, &src_templ,
 696                                                                 src_width0, src_height0,
 697                                                                 src_force_level);
 698         } else {
 699                 src_view = r600_create_sampler_view_custom(ctx, src, &src_templ,
 700                                                            src_widthFL, src_heightFL);
 701         }
 702
 703         u_box_3d(dstx, dsty, dstz, abs(src_box->width), abs(src_box->height),
 704                  abs(src_box->depth), &dstbox);
 705
 706         /* Copy. */
 707         r600_blitter_begin(ctx, R600_COPY_TEXTURE);
 708         util_blitter_blit_generic(rctx->blitter, dst_view, &dstbox,
 709                                   src_view, src_box, src_width0, src_height0,
 710                                   PIPE_MASK_RGBAZS, PIPE_TEX_FILTER_NEAREST, NULL);
 711         r600_blitter_end(ctx);
 712
 713         pipe_surface_reference(&dst_view, NULL);
 714         pipe_sampler_view_reference(&src_view, NULL);
 715 }
 716
 717 /* For MSAA integer resolving to work, we change the format to NORM using this function. */
 718 static enum pipe_format int_to_norm_format(enum pipe_format format)
 719 {
 720         switch (format) {
 721 #define REPLACE_FORMAT_SIGN(format,sign) \
 722         case PIPE_FORMAT_##format##_##sign##INT: \
 723                 return PIPE_FORMAT_##format##_##sign##NORM
 724 #define REPLACE_FORMAT(format) \
 725                 REPLACE_FORMAT_SIGN(format, U); \
 726                 REPLACE_FORMAT_SIGN(format, S)
 727
 728         REPLACE_FORMAT_SIGN(B10G10R10A2, U);
 729         REPLACE_FORMAT(R8);
 730         REPLACE_FORMAT(R8G8);
 731         REPLACE_FORMAT(R8G8B8X8);
 732         REPLACE_FORMAT(R8G8B8A8);
 733         REPLACE_FORMAT(A8);
 734         REPLACE_FORMAT(I8);
 735         REPLACE_FORMAT(L8);
 736         REPLACE_FORMAT(L8A8);
 737         REPLACE_FORMAT(R16);
 738         REPLACE_FORMAT(R16G16);
 739         REPLACE_FORMAT(R16G16B16X16);
 740         REPLACE_FORMAT(R16G16B16A16);
 741         REPLACE_FORMAT(A16);
 742         REPLACE_FORMAT(I16);
 743         REPLACE_FORMAT(L16);
 744         REPLACE_FORMAT(L16A16);
 745
 746 #undef REPLACE_FORMAT
 747 #undef REPLACE_FORMAT_SIGN
 748         default:
 749                 return format;
 750         }
 751 }
 752
 753 static bool do_hardware_msaa_resolve(struct pipe_context *ctx,
 754                                      const struct pipe_blit_info *info)
 755 {
 756         struct r600_context *rctx = (struct r600_context*)ctx;
 757         struct r600_texture *dst = (struct r600_texture*)info->dst.resource;
 758         unsigned dst_width = u_minify(info->dst.resource->width0, info->dst.level);
 759         unsigned dst_height = u_minify(info->dst.resource->height0, info->dst.level);
 760         enum pipe_format format = int_to_norm_format(info->dst.format);
 761         unsigned sample_mask =
 762                 rctx->b.chip_class == CAYMAN ? ~0 :
 763                 ((1ull << MAX2(1, info->src.resource->nr_samples)) - 1);
 764
 765         if (info->src.resource->nr_samples > 1 &&
 766             info->dst.resource->nr_samples <= 1 &&
 767             util_max_layer(info->src.resource, 0) == 0 &&
 768             util_max_layer(info->dst.resource, info->dst.level) == 0 &&
 769             info->dst.format == info->src.format &&
 770             !util_format_is_pure_integer(format) &&
 771             !util_format_is_depth_or_stencil(format) &&
 772             !info->scissor_enable &&
 773             (info->mask & PIPE_MASK_RGBA) == PIPE_MASK_RGBA &&
 774             dst_width == info->src.resource->width0 &&
 775             dst_height == info->src.resource->height0 &&
 776             info->dst.box.x == 0 &&
 777             info->dst.box.y == 0 &&
 778             info->dst.box.width == dst_width &&
 779             info->dst.box.height == dst_height &&
 780             info->dst.box.depth == 1 &&
 781             info->src.box.x == 0 &&
 782             info->src.box.y == 0 &&
 783             info->src.box.width == dst_width &&
 784             info->src.box.height == dst_height &&
 785             info->src.box.depth == 1 &&
 786             dst->surface.level[info->dst.level].mode >= RADEON_SURF_MODE_1D) {
 787                 r600_blitter_begin(ctx, R600_COLOR_RESOLVE);
 788                 util_blitter_custom_resolve_color(rctx->blitter,
 789                                                   info->dst.resource, info->dst.level,
 790                                                   info->dst.box.z,
 791                                                   info->src.resource, info->src.box.z,
 792                                                   sample_mask, rctx->custom_blend_resolve,
 793                                                   format);
 794                 r600_blitter_end(ctx);
 795                 return true;
 796         }
 797         return false;
 798 }
 799
 800 static void r600_blit(struct pipe_context *ctx,
 801                       const struct pipe_blit_info *info)
 802 {
 803         struct r600_context *rctx = (struct r600_context*)ctx;
 804
 805         if (do_hardware_msaa_resolve(ctx, info)) {
 806                 return;
 807         }
 808
 809         assert(util_blitter_is_blit_supported(rctx->blitter, info));
 810
 811         /* The driver doesn't decompress resources automatically while
 812          * u_blitter is rendering. */
 813         if (!r600_decompress_subresource(ctx, info->src.resource, info->src.level,
 814                                          info->src.box.z,
 815                                          info->src.box.z + info->src.box.depth - 1)) {
 816                 return; /* error */
 817         }
 818
 819         r600_blitter_begin(ctx, R600_BLIT);
 820         util_blitter_blit(rctx->blitter, info);
 821         r600_blitter_end(ctx);
 822 }
 823
 824 static void r600_flush_resource(struct pipe_context *ctx,
 825                                 struct pipe_resource *res)
 826 {
 827         struct r600_texture *rtex = (struct r600_texture*)res;
 828
 829         assert(res->target != PIPE_BUFFER);
 830
 831         if (!rtex->is_depth && rtex->cmask.size) {
 832                 r600_blit_decompress_color(ctx, rtex, 0, res->last_level,
 833                                            0, res->array_size - 1);
 834         }
 835 }
 836
 837 void r600_init_blit_functions(struct r600_context *rctx)
 838 {
 839         rctx->b.b.clear = r600_clear;
 840         rctx->b.b.clear_render_target = r600_clear_render_target;
 841         rctx->b.b.clear_depth_stencil = r600_clear_depth_stencil;
 842         rctx->b.b.resource_copy_region = r600_resource_copy_region;
 843         rctx->b.b.blit = r600_blit;
 844         rctx->b.b.flush_resource = r600_flush_resource;
 845         rctx->b.clear_buffer = r600_clear_buffer;
 846         rctx->b.blit_decompress_depth = r600_blit_decompress_depth;
 847 }