gallium/u_blitter: use 2D_ARRAY for cubemap blits if possible
[mesa.git] / src / gallium / drivers / r600 / r600_blit.c
index 6947c99c2ee9ff24a5090020fa06461b685b1dc2..80aa9c057695cb4bcd6d55c7589f4e95cc0e051d 100644 (file)
  * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
  * USE OR OTHER DEALINGS IN THE SOFTWARE.
  */
+#include "r600_pipe.h"
+#include "compute_memory_pool.h"
+#include "evergreen_compute.h"
 #include "util/u_surface.h"
-#include "util/u_blitter.h"
 #include "util/u_format.h"
-#include "r600_pipe.h"
+#include "evergreend.h"
 
 enum r600_blitter_op /* bitmask */
 {
-       R600_SAVE_TEXTURES      = 1,
-       R600_SAVE_FRAMEBUFFER   = 2,
-       R600_DISABLE_RENDER_COND = 4,
+       R600_SAVE_FRAGMENT_STATE = 1,
+       R600_SAVE_TEXTURES       = 2,
+       R600_SAVE_FRAMEBUFFER    = 4,
+       R600_DISABLE_RENDER_COND = 8,
+
+       R600_CLEAR         = R600_SAVE_FRAGMENT_STATE,
 
-       R600_CLEAR         = 0,
+       R600_CLEAR_SURFACE = R600_SAVE_FRAGMENT_STATE | R600_SAVE_FRAMEBUFFER,
 
-       R600_CLEAR_SURFACE = R600_SAVE_FRAMEBUFFER,
+       R600_COPY_BUFFER   = R600_DISABLE_RENDER_COND,
 
-       R600_COPY          = R600_SAVE_FRAMEBUFFER | R600_SAVE_TEXTURES |
+       R600_COPY_TEXTURE  = R600_SAVE_FRAGMENT_STATE | R600_SAVE_FRAMEBUFFER | R600_SAVE_TEXTURES |
                             R600_DISABLE_RENDER_COND,
 
-       R600_DECOMPRESS    = R600_SAVE_FRAMEBUFFER | R600_DISABLE_RENDER_COND,
+       R600_BLIT          = R600_SAVE_FRAGMENT_STATE | R600_SAVE_FRAMEBUFFER | R600_SAVE_TEXTURES,
+
+       R600_DECOMPRESS    = R600_SAVE_FRAGMENT_STATE | R600_SAVE_FRAMEBUFFER | R600_DISABLE_RENDER_COND,
+
+       R600_COLOR_RESOLVE = R600_SAVE_FRAGMENT_STATE | R600_SAVE_FRAMEBUFFER
 };
 
 static void r600_blitter_begin(struct pipe_context *ctx, enum r600_blitter_op op)
 {
-       struct r600_pipe_context *rctx = (struct r600_pipe_context *)ctx;
+       struct r600_context *rctx = (struct r600_context *)ctx;
 
-       r600_context_queries_suspend(&rctx->ctx);
-
-       util_blitter_save_blend(rctx->blitter, rctx->states[R600_PIPE_STATE_BLEND]);
-       util_blitter_save_depth_stencil_alpha(rctx->blitter, rctx->states[R600_PIPE_STATE_DSA]);
-       if (rctx->states[R600_PIPE_STATE_STENCIL_REF]) {
-               util_blitter_save_stencil_ref(rctx->blitter, &rctx->stencil_ref);
-       }
-       util_blitter_save_rasterizer(rctx->blitter, rctx->states[R600_PIPE_STATE_RASTERIZER]);
-       util_blitter_save_fragment_shader(rctx->blitter, rctx->ps_shader);
+       util_blitter_save_vertex_buffer_slot(rctx->blitter, rctx->vertex_buffer_state.vb);
+       util_blitter_save_vertex_elements(rctx->blitter, rctx->vertex_fetch_shader.cso);
        util_blitter_save_vertex_shader(rctx->blitter, rctx->vs_shader);
-       util_blitter_save_vertex_elements(rctx->blitter, rctx->vertex_elements);
-       if (rctx->states[R600_PIPE_STATE_VIEWPORT]) {
-               util_blitter_save_viewport(rctx->blitter, &rctx->viewport);
+       util_blitter_save_geometry_shader(rctx->blitter, rctx->gs_shader);
+       util_blitter_save_tessctrl_shader(rctx->blitter, rctx->tcs_shader);
+       util_blitter_save_tesseval_shader(rctx->blitter, rctx->tes_shader);
+       util_blitter_save_so_targets(rctx->blitter, rctx->b.streamout.num_targets,
+                                    (struct pipe_stream_output_target**)rctx->b.streamout.targets);
+       util_blitter_save_rasterizer(rctx->blitter, rctx->rasterizer_state.cso);
+
+       if (op & R600_SAVE_FRAGMENT_STATE) {
+               util_blitter_save_viewport(rctx->blitter, &rctx->b.viewports.states[0]);
+               util_blitter_save_scissor(rctx->blitter, &rctx->b.scissors.states[0]);
+               util_blitter_save_fragment_shader(rctx->blitter, rctx->ps_shader);
+               util_blitter_save_blend(rctx->blitter, rctx->blend_state.cso);
+               util_blitter_save_depth_stencil_alpha(rctx->blitter, rctx->dsa_state.cso);
+               util_blitter_save_stencil_ref(rctx->blitter, &rctx->stencil_ref.pipe_state);
+                util_blitter_save_sample_mask(rctx->blitter, rctx->sample_mask.sample_mask);
        }
-       util_blitter_save_vertex_buffers(rctx->blitter,
-                                        rctx->vbuf_mgr->nr_vertex_buffers,
-                                        rctx->vbuf_mgr->vertex_buffer);
-       util_blitter_save_so_targets(rctx->blitter, rctx->ctx.num_so_targets,
-                                    (struct pipe_stream_output_target**)rctx->ctx.so_targets);
 
        if (op & R600_SAVE_FRAMEBUFFER)
-               util_blitter_save_framebuffer(rctx->blitter, &rctx->framebuffer);
+               util_blitter_save_framebuffer(rctx->blitter, &rctx->framebuffer.state);
 
        if (op & R600_SAVE_TEXTURES) {
                util_blitter_save_fragment_sampler_states(
-                       rctx->blitter, rctx->ps_samplers.n_samplers,
-                       (void**)rctx->ps_samplers.samplers);
+                       rctx->blitter, util_last_bit(rctx->samplers[PIPE_SHADER_FRAGMENT].states.enabled_mask),
+                       (void**)rctx->samplers[PIPE_SHADER_FRAGMENT].states.states);
 
                util_blitter_save_fragment_sampler_views(
-                       rctx->blitter, rctx->ps_samplers.n_views,
-                       (struct pipe_sampler_view**)rctx->ps_samplers.views);
-       }
-
-       if ((op & R600_DISABLE_RENDER_COND) && rctx->current_render_cond) {
-               rctx->saved_render_cond = rctx->current_render_cond;
-               rctx->saved_render_cond_mode = rctx->current_render_cond_mode;
-               rctx->context.render_condition(&rctx->context, NULL, 0);
+                       rctx->blitter, util_last_bit(rctx->samplers[PIPE_SHADER_FRAGMENT].views.enabled_mask),
+                       (struct pipe_sampler_view**)rctx->samplers[PIPE_SHADER_FRAGMENT].views.views);
        }
 
+       if (op & R600_DISABLE_RENDER_COND)
+               rctx->b.render_cond_force_off = true;
 }
 
 static void r600_blitter_end(struct pipe_context *ctx)
 {
-       struct r600_pipe_context *rctx = (struct r600_pipe_context *)ctx;
-       if (rctx->saved_render_cond) {
-               rctx->context.render_condition(&rctx->context,
-                                              rctx->saved_render_cond,
-                                              rctx->saved_render_cond_mode);
-               rctx->saved_render_cond = NULL;
-       }
-       r600_context_queries_resume(&rctx->ctx);
+       struct r600_context *rctx = (struct r600_context *)ctx;
+
+       rctx->b.render_cond_force_off = false;
 }
 
-static unsigned u_num_layers(struct pipe_resource *r, unsigned level)
+static unsigned u_max_sample(struct pipe_resource *r)
 {
-       switch (r->target) {
-       case PIPE_TEXTURE_CUBE:
-               return 6;
-       case PIPE_TEXTURE_3D:
-               return u_minify(r->depth0, level);
-       case PIPE_TEXTURE_1D_ARRAY:
-               return r->array_size;
-       case PIPE_TEXTURE_2D_ARRAY:
-               return r->array_size;
-       default:
-               return 1;
-       }
+       return r->nr_samples ? r->nr_samples - 1 : 0;
 }
 
-void r600_blit_uncompress_depth(struct pipe_context *ctx, struct r600_resource_texture *texture)
+static void r600_blit_decompress_depth(struct pipe_context *ctx,
+                                      struct r600_texture *texture,
+                                      struct r600_texture *staging,
+                                      unsigned first_level, unsigned last_level,
+                                      unsigned first_layer, unsigned last_layer,
+                                      unsigned first_sample, unsigned last_sample)
 {
-       struct r600_pipe_context *rctx = (struct r600_pipe_context *)ctx;
-       unsigned layer, level;
-       float depth = 1.0f;
+       struct r600_context *rctx = (struct r600_context *)ctx;
+       unsigned layer, level, sample, checked_last_layer, max_layer, max_sample;
+       struct r600_texture *flushed_depth_texture = staging ?
+                       staging : texture->flushed_depth_texture;
+       const struct util_format_description *desc =
+               util_format_description(texture->resource.b.b.format);
+       float depth;
+
+       if (!staging && !texture->dirty_level_mask)
+               return;
+
+       max_sample = u_max_sample(&texture->resource.b.b);
 
-       if (!texture->dirty_db)
+       /* XXX Decompressing MSAA depth textures is broken on R6xx.
+        * There is also a hardlock if CMASK and FMASK are not present.
+        * Just skip this until we find out how to fix it. */
+       if (rctx->b.chip_class == R600 && max_sample > 0) {
+               texture->dirty_level_mask = 0;
                return;
+       }
 
-       if (rctx->family == CHIP_RV610 || rctx->family == CHIP_RV630 ||
-           rctx->family == CHIP_RV620 || rctx->family == CHIP_RV635)
+       if (rctx->b.family == CHIP_RV610 || rctx->b.family == CHIP_RV630 ||
+           rctx->b.family == CHIP_RV620 || rctx->b.family == CHIP_RV635)
                depth = 0.0f;
+       else
+               depth = 1.0f;
 
-       for (level = 0; level <= texture->resource.b.b.b.last_level; level++) {
-               unsigned num_layers = u_num_layers(&texture->resource.b.b.b, level);
+       /* Enable decompression in DB_RENDER_CONTROL */
+       rctx->db_misc_state.flush_depthstencil_through_cb = true;
+       rctx->db_misc_state.copy_depth = util_format_has_depth(desc);
+       rctx->db_misc_state.copy_stencil = util_format_has_stencil(desc);
+       rctx->db_misc_state.copy_sample = first_sample;
+       r600_mark_atom_dirty(rctx, &rctx->db_misc_state.atom);
 
-               for (layer = 0; layer < num_layers; layer++) {
-                       struct pipe_surface *zsurf, *cbsurf, surf_tmpl;
+       for (level = first_level; level <= last_level; level++) {
+               if (!staging && !(texture->dirty_level_mask & (1 << level)))
+                       continue;
 
-                       surf_tmpl.format = texture->real_format;
-                       surf_tmpl.u.tex.level = level;
+               /* The smaller the mipmap level, the less layers there are
+                * as far as 3D textures are concerned. */
+               max_layer = util_max_layer(&texture->resource.b.b, level);
+               checked_last_layer = last_layer < max_layer ? last_layer : max_layer;
+
+               for (layer = first_layer; layer <= checked_last_layer; layer++) {
+                       for (sample = first_sample; sample <= last_sample; sample++) {
+                               struct pipe_surface *zsurf, *cbsurf, surf_tmpl;
+
+                               if (sample != rctx->db_misc_state.copy_sample) {
+                                       rctx->db_misc_state.copy_sample = sample;
+                                       r600_mark_atom_dirty(rctx, &rctx->db_misc_state.atom);
+                               }
+
+                               surf_tmpl.format = texture->resource.b.b.format;
+                               surf_tmpl.u.tex.level = level;
+                               surf_tmpl.u.tex.first_layer = layer;
+                               surf_tmpl.u.tex.last_layer = layer;
+
+                               zsurf = ctx->create_surface(ctx, &texture->resource.b.b, &surf_tmpl);
+
+                               surf_tmpl.format = flushed_depth_texture->resource.b.b.format;
+                               cbsurf = ctx->create_surface(ctx,
+                                               &flushed_depth_texture->resource.b.b, &surf_tmpl);
+
+                               r600_blitter_begin(ctx, R600_DECOMPRESS);
+                               util_blitter_custom_depth_stencil(rctx->blitter, zsurf, cbsurf, 1 << sample,
+                                                                 rctx->custom_dsa_flush, depth);
+                               r600_blitter_end(ctx);
+
+                               pipe_surface_reference(&zsurf, NULL);
+                               pipe_surface_reference(&cbsurf, NULL);
+                       }
+               }
+
+               /* The texture will always be dirty if some layers or samples aren't flushed.
+                * I don't think this case occurs often though. */
+               if (!staging &&
+                   first_layer == 0 && last_layer == max_layer &&
+                   first_sample == 0 && last_sample == max_sample) {
+                       texture->dirty_level_mask &= ~(1 << level);
+               }
+       }
+
+       /* reenable compression in DB_RENDER_CONTROL */
+       rctx->db_misc_state.flush_depthstencil_through_cb = false;
+       r600_mark_atom_dirty(rctx, &rctx->db_misc_state.atom);
+}
+
+static void r600_blit_decompress_depth_in_place(struct r600_context *rctx,
+                                                struct r600_texture *texture,
+                                               bool is_stencil_sampler,
+                                                unsigned first_level, unsigned last_level,
+                                                unsigned first_layer, unsigned last_layer)
+{
+       struct pipe_surface *zsurf, surf_tmpl = {{0}};
+       unsigned layer, max_layer, checked_last_layer, level;
+       unsigned *dirty_level_mask;
+
+       /* Enable decompression in DB_RENDER_CONTROL */
+       if (is_stencil_sampler) {
+               rctx->db_misc_state.flush_stencil_inplace = true;
+               dirty_level_mask = &texture->stencil_dirty_level_mask;
+       } else {
+               rctx->db_misc_state.flush_depth_inplace = true;
+               dirty_level_mask = &texture->dirty_level_mask;
+       }
+       r600_mark_atom_dirty(rctx, &rctx->db_misc_state.atom);
+
+       surf_tmpl.format = texture->resource.b.b.format;
+
+       for (level = first_level; level <= last_level; level++) {
+               if (!(*dirty_level_mask & (1 << level)))
+                       continue;
+
+               surf_tmpl.u.tex.level = level;
+
+               /* The smaller the mipmap level, the less layers there are
+                * as far as 3D textures are concerned. */
+               max_layer = util_max_layer(&texture->resource.b.b, level);
+               checked_last_layer = last_layer < max_layer ? last_layer : max_layer;
+
+               for (layer = first_layer; layer <= checked_last_layer; layer++) {
                        surf_tmpl.u.tex.first_layer = layer;
                        surf_tmpl.u.tex.last_layer = layer;
-                       surf_tmpl.usage = PIPE_BIND_DEPTH_STENCIL;
 
-                       zsurf = ctx->create_surface(ctx, &texture->resource.b.b.b, &surf_tmpl);
+                       zsurf = rctx->b.b.create_surface(&rctx->b.b, &texture->resource.b.b, &surf_tmpl);
 
-                       surf_tmpl.format = texture->flushed_depth_texture->real_format;
-                       surf_tmpl.usage = PIPE_BIND_RENDER_TARGET;
-                       cbsurf = ctx->create_surface(ctx,
-                                       (struct pipe_resource*)texture->flushed_depth_texture, &surf_tmpl);
-
-                       r600_blitter_begin(ctx, R600_DECOMPRESS);
-                       util_blitter_custom_depth_stencil(rctx->blitter, zsurf, cbsurf, rctx->custom_dsa_flush, depth);
-                       r600_blitter_end(ctx);
+                       r600_blitter_begin(&rctx->b.b, R600_DECOMPRESS);
+                       util_blitter_custom_depth_stencil(rctx->blitter, zsurf, NULL, ~0,
+                                                         rctx->custom_dsa_flush, 1.0f);
+                       r600_blitter_end(&rctx->b.b);
 
                        pipe_surface_reference(&zsurf, NULL);
-                       pipe_surface_reference(&cbsurf, NULL);
+               }
+
+               /* The texture will always be dirty if some layers or samples aren't flushed.
+                * I don't think this case occurs often though. */
+               if (first_layer == 0 && last_layer == max_layer) {
+                       *dirty_level_mask &= ~(1 << level);
                }
        }
 
-       texture->dirty_db = FALSE;
+       /* Disable decompression in DB_RENDER_CONTROL */
+       rctx->db_misc_state.flush_depth_inplace = false;
+       rctx->db_misc_state.flush_stencil_inplace = false;
+       r600_mark_atom_dirty(rctx, &rctx->db_misc_state.atom);
 }
 
-void r600_flush_depth_textures(struct r600_pipe_context *rctx)
+void r600_decompress_depth_textures(struct r600_context *rctx,
+                              struct r600_samplerview_state *textures)
 {
-       unsigned int i;
-
-       /* FIXME: This handles fragment shader textures only. */
+       unsigned i;
+       unsigned depth_texture_mask = textures->compressed_depthtex_mask;
+
+       while (depth_texture_mask) {
+               struct pipe_sampler_view *view;
+               struct r600_pipe_sampler_view *rview;
+               struct r600_texture *tex;
+
+               i = u_bit_scan(&depth_texture_mask);
+
+               view = &textures->views[i]->base;
+               assert(view);
+               rview = (struct r600_pipe_sampler_view*)view;
+
+               tex = (struct r600_texture *)view->texture;
+               assert(tex->db_compatible);
+
+               if (r600_can_sample_zs(tex, rview->is_stencil_sampler)) {
+                       r600_blit_decompress_depth_in_place(rctx, tex,
+                                                  rview->is_stencil_sampler,
+                                                  view->u.tex.first_level, view->u.tex.last_level,
+                                                  0, util_max_layer(&tex->resource.b.b, view->u.tex.first_level));
+               } else {
+                       r600_blit_decompress_depth(&rctx->b.b, tex, NULL,
+                                                  view->u.tex.first_level, view->u.tex.last_level,
+                                                  0, util_max_layer(&tex->resource.b.b, view->u.tex.first_level),
+                                                  0, u_max_sample(&tex->resource.b.b));
+               }
+       }
+}
 
-       for (i = 0; i < rctx->ps_samplers.n_views; ++i) {
-               struct r600_pipe_sampler_view *view;
-               struct r600_resource_texture *tex;
+static void r600_blit_decompress_color(struct pipe_context *ctx,
+               struct r600_texture *rtex,
+               unsigned first_level, unsigned last_level,
+               unsigned first_layer, unsigned last_layer)
+{
+       struct r600_context *rctx = (struct r600_context *)ctx;
+       unsigned layer, level, checked_last_layer, max_layer;
 
-               view = rctx->ps_samplers.views[i];
-               if (!view) continue;
+       if (!rtex->dirty_level_mask)
+               return;
 
-               tex = (struct r600_resource_texture *)view->base.texture;
-               if (!tex->depth)
+       for (level = first_level; level <= last_level; level++) {
+               if (!(rtex->dirty_level_mask & (1 << level)))
                        continue;
 
-               if (tex->is_flushing_texture)
-                       continue;
+               /* The smaller the mipmap level, the less layers there are
+                * as far as 3D textures are concerned. */
+               max_layer = util_max_layer(&rtex->resource.b.b, level);
+               checked_last_layer = last_layer < max_layer ? last_layer : max_layer;
+
+               for (layer = first_layer; layer <= checked_last_layer; layer++) {
+                       struct pipe_surface *cbsurf, surf_tmpl;
+
+                       surf_tmpl.format = rtex->resource.b.b.format;
+                       surf_tmpl.u.tex.level = level;
+                       surf_tmpl.u.tex.first_layer = layer;
+                       surf_tmpl.u.tex.last_layer = layer;
+                       cbsurf = ctx->create_surface(ctx, &rtex->resource.b.b, &surf_tmpl);
+
+                       r600_blitter_begin(ctx, R600_DECOMPRESS);
+                       util_blitter_custom_color(rctx->blitter, cbsurf,
+                               rtex->fmask.size ? rctx->custom_blend_decompress : rctx->custom_blend_fastclear);
+                       r600_blitter_end(ctx);
 
-               r600_blit_uncompress_depth(&rctx->context, tex);
+                       pipe_surface_reference(&cbsurf, NULL);
+               }
+
+               /* The texture will always be dirty if some layers aren't flushed.
+                * I don't think this case occurs often though. */
+               if (first_layer == 0 && last_layer == max_layer) {
+                       rtex->dirty_level_mask &= ~(1 << level);
+               }
        }
+}
 
-       /* also check CB here */
-       for (i = 0; i < rctx->framebuffer.nr_cbufs; i++) {
-               struct r600_resource_texture *tex;
-               tex = (struct r600_resource_texture *)rctx->framebuffer.cbufs[i]->texture;
+void r600_decompress_color_textures(struct r600_context *rctx,
+                                   struct r600_samplerview_state *textures)
+{
+       unsigned i;
+       unsigned mask = textures->compressed_colortex_mask;
 
-               if (!tex->depth)
-                       continue;
+       while (mask) {
+               struct pipe_sampler_view *view;
+               struct r600_texture *tex;
 
-               if (tex->is_flushing_texture)
-                       continue;
+               i = u_bit_scan(&mask);
 
-               r600_blit_uncompress_depth(&rctx->context, tex);
+               view = &textures->views[i]->base;
+               assert(view);
+
+               tex = (struct r600_texture *)view->texture;
+               assert(tex->cmask.size);
+
+               r600_blit_decompress_color(&rctx->b.b, tex,
+                                          view->u.tex.first_level, view->u.tex.last_level,
+                                          0, util_max_layer(&tex->resource.b.b, view->u.tex.first_level));
        }
 }
 
+/* Helper for decompressing a portion of a color or depth resource before
+ * blitting if any decompression is needed.
+ * The driver doesn't decompress resources automatically while u_blitter is
+ * rendering. */
+static bool r600_decompress_subresource(struct pipe_context *ctx,
+                                       struct pipe_resource *tex,
+                                       unsigned level,
+                                       unsigned first_layer, unsigned last_layer)
+{
+       struct r600_context *rctx = (struct r600_context *)ctx;
+       struct r600_texture *rtex = (struct r600_texture*)tex;
+
+       if (rtex->db_compatible) {
+               if (r600_can_sample_zs(rtex, false)) {
+                       r600_blit_decompress_depth_in_place(rctx, rtex, false,
+                                                  level, level,
+                                                  first_layer, last_layer);
+                       if (rtex->surface.flags & RADEON_SURF_SBUFFER) {
+                               r600_blit_decompress_depth_in_place(rctx, rtex, true,
+                                                          level, level,
+                                                          first_layer, last_layer);
+                       }
+               } else {
+                       if (!r600_init_flushed_depth_texture(ctx, tex, NULL))
+                               return false; /* error */
+
+                       r600_blit_decompress_depth(ctx, rtex, NULL,
+                                                  level, level,
+                                                  first_layer, last_layer,
+                                                  0, u_max_sample(tex));
+               }
+       } else if (rtex->cmask.size) {
+               r600_blit_decompress_color(ctx, rtex, level, level,
+                                          first_layer, last_layer);
+       }
+       return true;
+}
+
 static void r600_clear(struct pipe_context *ctx, unsigned buffers,
                       const union pipe_color_union *color,
                       double depth, unsigned stencil)
 {
-       struct r600_pipe_context *rctx = (struct r600_pipe_context *)ctx;
-       struct pipe_framebuffer_state *fb = &rctx->framebuffer;
-       
+       struct r600_context *rctx = (struct r600_context *)ctx;
+       struct pipe_framebuffer_state *fb = &rctx->framebuffer.state;
+
+       if (buffers & PIPE_CLEAR_COLOR && rctx->b.chip_class >= EVERGREEN) {
+               evergreen_do_fast_color_clear(&rctx->b, fb, &rctx->framebuffer.atom,
+                                             &buffers, NULL, color);
+               if (!buffers)
+                       return; /* all buffers have been fast cleared */
+       }
+
+       if (buffers & PIPE_CLEAR_COLOR) {
+               int i;
+
+               /* These buffers cannot use fast clear, make sure to disable expansion. */
+               for (i = 0; i < fb->nr_cbufs; i++) {
+                       struct r600_texture *tex;
+
+                       /* If not clearing this buffer, skip. */
+                       if (!(buffers & (PIPE_CLEAR_COLOR0 << i)))
+                               continue;
+
+                       if (!fb->cbufs[i])
+                               continue;
+
+                       tex = (struct r600_texture *)fb->cbufs[i]->texture;
+                       if (tex->fmask.size == 0)
+                               tex->dirty_level_mask &= ~(1 << fb->cbufs[i]->u.tex.level);
+               }
+       }
+
+       /* if hyperz enabled just clear hyperz */
+       if (fb->zsbuf && (buffers & PIPE_CLEAR_DEPTH)) {
+               struct r600_texture *rtex;
+               unsigned level = fb->zsbuf->u.tex.level;
+
+               rtex = (struct r600_texture*)fb->zsbuf->texture;
+
+               /* We can't use hyperz fast clear if each slice of a texture
+                * array are clear to different value. To simplify code just
+                * disable fast clear for texture array.
+                */
+               /* Only use htile for first level */
+               if (rtex->htile_buffer && !level &&
+                   fb->zsbuf->u.tex.first_layer == 0 &&
+                   fb->zsbuf->u.tex.last_layer == util_max_layer(&rtex->resource.b.b, level)) {
+                       if (rtex->depth_clear_value != depth) {
+                               rtex->depth_clear_value = depth;
+                               r600_mark_atom_dirty(rctx, &rctx->db_state.atom);
+                       }
+                       rctx->db_misc_state.htile_clear = true;
+                       r600_mark_atom_dirty(rctx, &rctx->db_misc_state.atom);
+               }
+       }
+
        r600_blitter_begin(ctx, R600_CLEAR);
        util_blitter_clear(rctx->blitter, fb->width, fb->height,
-                          fb->nr_cbufs, buffers, fb->nr_cbufs ? fb->cbufs[0]->format : PIPE_FORMAT_NONE,
-                          color, depth, stencil);
+                          util_framebuffer_get_num_layers(fb),
+                          buffers, color, depth, stencil);
        r600_blitter_end(ctx);
+
+       /* disable fast clear */
+       if (rctx->db_misc_state.htile_clear) {
+               rctx->db_misc_state.htile_clear = false;
+               r600_mark_atom_dirty(rctx, &rctx->db_misc_state.atom);
+       }
 }
 
 static void r600_clear_render_target(struct pipe_context *ctx,
                                     struct pipe_surface *dst,
                                     const union pipe_color_union *color,
                                     unsigned dstx, unsigned dsty,
-                                    unsigned width, unsigned height)
+                                    unsigned width, unsigned height,
+                                    bool render_condition_enabled)
 {
-       struct r600_pipe_context *rctx = (struct r600_pipe_context *)ctx;
+       struct r600_context *rctx = (struct r600_context *)ctx;
 
-       r600_blitter_begin(ctx, R600_CLEAR_SURFACE);
+       r600_blitter_begin(ctx, R600_CLEAR_SURFACE |
+                          (render_condition_enabled ? 0 : R600_DISABLE_RENDER_COND));
        util_blitter_clear_render_target(rctx->blitter, dst, color,
                                         dstx, dsty, width, height);
        r600_blitter_end(ctx);
@@ -230,154 +491,433 @@ static void r600_clear_depth_stencil(struct pipe_context *ctx,
                                     double depth,
                                     unsigned stencil,
                                     unsigned dstx, unsigned dsty,
-                                    unsigned width, unsigned height)
+                                    unsigned width, unsigned height,
+                                    bool render_condition_enabled)
 {
-       struct r600_pipe_context *rctx = (struct r600_pipe_context *)ctx;
+       struct r600_context *rctx = (struct r600_context *)ctx;
 
-       r600_blitter_begin(ctx, R600_CLEAR_SURFACE);
+       r600_blitter_begin(ctx, R600_CLEAR_SURFACE |
+                          (render_condition_enabled ? 0 : R600_DISABLE_RENDER_COND));
        util_blitter_clear_depth_stencil(rctx->blitter, dst, clear_flags, depth, stencil,
                                         dstx, dsty, width, height);
        r600_blitter_end(ctx);
 }
 
+static void r600_copy_buffer(struct pipe_context *ctx, struct pipe_resource *dst, unsigned dstx,
+                            struct pipe_resource *src, const struct pipe_box *src_box)
+{
+       struct r600_context *rctx = (struct r600_context*)ctx;
 
+       if (rctx->screen->b.has_cp_dma) {
+               r600_cp_dma_copy_buffer(rctx, dst, dstx, src, src_box->x, src_box->width);
+       }
+       else if (rctx->screen->b.has_streamout &&
+                /* Require 4-byte alignment. */
+                dstx % 4 == 0 && src_box->x % 4 == 0 && src_box->width % 4 == 0) {
+
+               r600_blitter_begin(ctx, R600_COPY_BUFFER);
+               util_blitter_copy_buffer(rctx->blitter, dst, dstx, src, src_box->x, src_box->width);
+               r600_blitter_end(ctx);
+       } else {
+               util_resource_copy_region(ctx, dst, 0, dstx, 0, 0, src, 0, src_box);
+       }
+}
 
-/* Copy a block of pixels from one surface to another using HW. */
-static void r600_hw_copy_region(struct pipe_context *ctx,
-                               struct pipe_resource *dst,
-                               unsigned dst_level,
-                               unsigned dstx, unsigned dsty, unsigned dstz,
-                               struct pipe_resource *src,
-                               unsigned src_level,
-                               const struct pipe_box *src_box)
+/**
+ * Global buffers are not really resources, they are are actually offsets
+ * into a single global resource (r600_screen::global_pool).  The means
+ * they don't have their own buf handle, so they cannot be passed
+ * to r600_copy_buffer() and must be handled separately.
+ */
+static void r600_copy_global_buffer(struct pipe_context *ctx,
+                                   struct pipe_resource *dst, unsigned
+                                   dstx, struct pipe_resource *src,
+                                   const struct pipe_box *src_box)
 {
-       struct r600_pipe_context *rctx = (struct r600_pipe_context *)ctx;
+       struct r600_context *rctx = (struct r600_context*)ctx;
+       struct compute_memory_pool *pool = rctx->screen->global_pool;
+       struct pipe_box new_src_box = *src_box;
+
+       if (src->bind & PIPE_BIND_GLOBAL) {
+               struct r600_resource_global *rsrc =
+                       (struct r600_resource_global *)src;
+               struct compute_memory_item *item = rsrc->chunk;
+
+               if (is_item_in_pool(item)) {
+                       new_src_box.x += 4 * item->start_in_dw;
+                       src = (struct pipe_resource *)pool->bo;
+               } else {
+                       if (item->real_buffer == NULL) {
+                               item->real_buffer =
+                                       r600_compute_buffer_alloc_vram(pool->screen,
+                                                                      item->size_in_dw * 4);
+                       }
+                       src = (struct pipe_resource*)item->real_buffer;
+               }
+       }
+       if (dst->bind & PIPE_BIND_GLOBAL) {
+               struct r600_resource_global *rdst =
+                       (struct r600_resource_global *)dst;
+               struct compute_memory_item *item = rdst->chunk;
+
+               if (is_item_in_pool(item)) {
+                       dstx += 4 * item->start_in_dw;
+                       dst = (struct pipe_resource *)pool->bo;
+               } else {
+                       if (item->real_buffer == NULL) {
+                               item->real_buffer =
+                                       r600_compute_buffer_alloc_vram(pool->screen,
+                                                                      item->size_in_dw * 4);
+                       }
+                       dst = (struct pipe_resource*)item->real_buffer;
+               }
+       }
 
-       r600_blitter_begin(ctx, R600_COPY);
-       util_blitter_copy_texture(rctx->blitter, dst, dst_level, dstx, dsty, dstz,
-                                 src, src_level, src_box, TRUE);
-       r600_blitter_end(ctx);
+       r600_copy_buffer(ctx, dst, dstx, src, &new_src_box);
 }
 
-struct texture_orig_info {
-       unsigned format;
-       unsigned width0;
-       unsigned height0;
-};
+static void r600_clear_buffer(struct pipe_context *ctx, struct pipe_resource *dst,
+                             uint64_t offset, uint64_t size, unsigned value,
+                             enum r600_coherency coher)
+{
+       struct r600_context *rctx = (struct r600_context*)ctx;
+
+       if (rctx->screen->b.has_cp_dma &&
+           rctx->b.chip_class >= EVERGREEN &&
+           offset % 4 == 0 && size % 4 == 0) {
+               evergreen_cp_dma_clear_buffer(rctx, dst, offset, size, value, coher);
+       } else if (rctx->screen->b.has_streamout && offset % 4 == 0 && size % 4 == 0) {
+               union pipe_color_union clear_value;
+               clear_value.ui[0] = value;
+
+               r600_blitter_begin(ctx, R600_DISABLE_RENDER_COND);
+               util_blitter_clear_buffer(rctx->blitter, dst, offset, size,
+                                         1, &clear_value);
+               r600_blitter_end(ctx);
+       } else {
+               uint32_t *map = r600_buffer_map_sync_with_rings(&rctx->b, r600_resource(dst),
+                                                                PIPE_TRANSFER_WRITE);
+               map += offset / 4;
+               size /= 4;
+               for (unsigned i = 0; i < size; i++)
+                       *map++ = value;
+       }
+}
 
-static void r600_compressed_to_blittable(struct pipe_resource *tex,
-                                  unsigned level,
-                                  struct texture_orig_info *orig)
+void r600_resource_copy_region(struct pipe_context *ctx,
+                              struct pipe_resource *dst,
+                              unsigned dst_level,
+                              unsigned dstx, unsigned dsty, unsigned dstz,
+                              struct pipe_resource *src,
+                              unsigned src_level,
+                              const struct pipe_box *src_box)
 {
-       struct r600_resource_texture *rtex = (struct r600_resource_texture*)tex;
-       unsigned pixsize = util_format_get_blocksize(rtex->real_format);
-       int new_format;
-       int new_height, new_width;
+       struct r600_context *rctx = (struct r600_context *)ctx;
+       struct pipe_surface *dst_view, dst_templ;
+       struct pipe_sampler_view src_templ, *src_view;
+       unsigned dst_width, dst_height, src_width0, src_height0, src_widthFL, src_heightFL;
+       unsigned src_force_level = 0;
+       struct pipe_box sbox, dstbox;
+
+       /* Handle buffers first. */
+       if (dst->target == PIPE_BUFFER && src->target == PIPE_BUFFER) {
+               if ((src->bind & PIPE_BIND_GLOBAL) ||
+                                       (dst->bind & PIPE_BIND_GLOBAL)) {
+                       r600_copy_global_buffer(ctx, dst, dstx, src, src_box);
+               } else {
+                       r600_copy_buffer(ctx, dst, dstx, src, src_box);
+               }
+               return;
+       }
 
-       orig->format = tex->format;
-       orig->width0 = tex->width0;
-       orig->height0 = tex->height0;
+       assert(u_max_sample(dst) == u_max_sample(src));
 
-       if (pixsize == 8)
-               new_format = PIPE_FORMAT_R16G16B16A16_UINT; /* 64-bit block */
-       else
-               new_format = PIPE_FORMAT_R32G32B32A32_UINT; /* 128-bit block */
+       /* The driver doesn't decompress resources automatically while
+        * u_blitter is rendering. */
+       if (!r600_decompress_subresource(ctx, src, src_level,
+                                        src_box->z, src_box->z + src_box->depth - 1)) {
+               return; /* error */
+       }
 
-       new_width = util_format_get_nblocksx(tex->format, orig->width0);
-       new_height = util_format_get_nblocksy(tex->format, orig->height0);
+       dst_width = u_minify(dst->width0, dst_level);
+        dst_height = u_minify(dst->height0, dst_level);
+       src_width0 = src->width0;
+       src_height0 = src->height0;
+        src_widthFL = u_minify(src->width0, src_level);
+        src_heightFL = u_minify(src->height0, src_level);
+
+       util_blitter_default_dst_texture(&dst_templ, dst, dst_level, dstz);
+       util_blitter_default_src_texture(rctx->blitter, &src_templ, src, src_level);
+
+       if (util_format_is_compressed(src->format) ||
+           util_format_is_compressed(dst->format)) {
+               unsigned blocksize = util_format_get_blocksize(src->format);
+
+               if (blocksize == 8)
+                       src_templ.format = PIPE_FORMAT_R16G16B16A16_UINT; /* 64-bit block */
+               else
+                       src_templ.format = PIPE_FORMAT_R32G32B32A32_UINT; /* 128-bit block */
+               dst_templ.format = src_templ.format;
+
+               dst_width = util_format_get_nblocksx(dst->format, dst_width);
+               dst_height = util_format_get_nblocksy(dst->format, dst_height);
+               src_width0 = util_format_get_nblocksx(src->format, src_width0);
+               src_height0 = util_format_get_nblocksy(src->format, src_height0);
+               src_widthFL = util_format_get_nblocksx(src->format, src_widthFL);
+               src_heightFL = util_format_get_nblocksy(src->format, src_heightFL);
+
+               dstx = util_format_get_nblocksx(dst->format, dstx);
+               dsty = util_format_get_nblocksy(dst->format, dsty);
+
+               sbox.x = util_format_get_nblocksx(src->format, src_box->x);
+               sbox.y = util_format_get_nblocksy(src->format, src_box->y);
+               sbox.z = src_box->z;
+               sbox.width = util_format_get_nblocksx(src->format, src_box->width);
+               sbox.height = util_format_get_nblocksy(src->format, src_box->height);
+               sbox.depth = src_box->depth;
+               src_box = &sbox;
+
+               src_force_level = src_level;
+       } else if (!util_blitter_is_copy_supported(rctx->blitter, dst, src)) {
+               if (util_format_is_subsampled_422(src->format)) {
+
+                       src_templ.format = PIPE_FORMAT_R8G8B8A8_UINT;
+                       dst_templ.format = PIPE_FORMAT_R8G8B8A8_UINT;
+
+                       dst_width = util_format_get_nblocksx(dst->format, dst_width);
+                       src_width0 = util_format_get_nblocksx(src->format, src_width0);
+                       src_widthFL = util_format_get_nblocksx(src->format, src_widthFL);
+
+                       dstx = util_format_get_nblocksx(dst->format, dstx);
+
+                       sbox = *src_box;
+                       sbox.x = util_format_get_nblocksx(src->format, src_box->x);
+                       sbox.width = util_format_get_nblocksx(src->format, src_box->width);
+                       src_box = &sbox;
+               } else {
+                       unsigned blocksize = util_format_get_blocksize(src->format);
+
+                       switch (blocksize) {
+                       case 1:
+                               dst_templ.format = PIPE_FORMAT_R8_UNORM;
+                               src_templ.format = PIPE_FORMAT_R8_UNORM;
+                               break;
+                       case 2:
+                               dst_templ.format = PIPE_FORMAT_R8G8_UNORM;
+                               src_templ.format = PIPE_FORMAT_R8G8_UNORM;
+                               break;
+                       case 4:
+                               dst_templ.format = PIPE_FORMAT_R8G8B8A8_UNORM;
+                               src_templ.format = PIPE_FORMAT_R8G8B8A8_UNORM;
+                               break;
+                       case 8:
+                               dst_templ.format = PIPE_FORMAT_R16G16B16A16_UINT;
+                               src_templ.format = PIPE_FORMAT_R16G16B16A16_UINT;
+                               break;
+                       case 16:
+                               dst_templ.format = PIPE_FORMAT_R32G32B32A32_UINT;
+                               src_templ.format = PIPE_FORMAT_R32G32B32A32_UINT;
+                               break;
+                       default:
+                               fprintf(stderr, "Unhandled format %s with blocksize %u\n",
+                                       util_format_short_name(src->format), blocksize);
+                               assert(0);
+                       }
+               }
+       }
+
+       dst_view = r600_create_surface_custom(ctx, dst, &dst_templ,
+                                             /* we don't care about these two for r600g */
+                                             dst->width0, dst->height0,
+                                             dst_width, dst_height);
+
+       if (rctx->b.chip_class >= EVERGREEN) {
+               src_view = evergreen_create_sampler_view_custom(ctx, src, &src_templ,
+                                                               src_width0, src_height0,
+                                                               src_force_level);
+       } else {
+               src_view = r600_create_sampler_view_custom(ctx, src, &src_templ,
+                                                          src_widthFL, src_heightFL);
+       }
+
+        u_box_3d(dstx, dsty, dstz, abs(src_box->width), abs(src_box->height),
+                 abs(src_box->depth), &dstbox);
 
-       tex->width0 = new_width;
-       tex->height0 = new_height;
-       tex->format = new_format;
+       /* Copy. */
+       r600_blitter_begin(ctx, R600_COPY_TEXTURE);
+       util_blitter_blit_generic(rctx->blitter, dst_view, &dstbox,
+                                 src_view, src_box, src_width0, src_height0,
+                                 PIPE_MASK_RGBAZS, PIPE_TEX_FILTER_NEAREST, NULL,
+                                 FALSE);
+       r600_blitter_end(ctx);
+
+       pipe_surface_reference(&dst_view, NULL);
+       pipe_sampler_view_reference(&src_view, NULL);
 }
 
-static void r600_reset_blittable_to_compressed(struct pipe_resource *tex,
-                                              struct texture_orig_info *orig)
+static bool do_hardware_msaa_resolve(struct pipe_context *ctx,
+                                    const struct pipe_blit_info *info)
 {
-       tex->format = orig->format;
-       tex->width0 = orig->width0;
-       tex->height0 = orig->height0;
+       struct r600_context *rctx = (struct r600_context*)ctx;
+       struct r600_texture *dst = (struct r600_texture*)info->dst.resource;
+       unsigned dst_width = u_minify(info->dst.resource->width0, info->dst.level);
+       unsigned dst_height = u_minify(info->dst.resource->height0, info->dst.level);
+       enum pipe_format format = info->src.format;
+       unsigned sample_mask =
+               rctx->b.chip_class == CAYMAN ? ~0 :
+               ((1ull << MAX2(1, info->src.resource->nr_samples)) - 1);
+       struct pipe_resource *tmp, templ;
+       struct pipe_blit_info blit;
+
+       /* Check basic requirements for hw resolve. */
+       if (!(info->src.resource->nr_samples > 1 &&
+             info->dst.resource->nr_samples <= 1 &&
+             !util_format_is_pure_integer(format) &&
+             !util_format_is_depth_or_stencil(format) &&
+             util_max_layer(info->src.resource, 0) == 0))
+               return false;
+
+       /* Check the remaining requirements for hw resolve. */
+       if (util_max_layer(info->dst.resource, info->dst.level) == 0 &&
+           util_is_format_compatible(util_format_description(info->src.format),
+                                     util_format_description(info->dst.format)) &&
+           !info->scissor_enable &&
+           (info->mask & PIPE_MASK_RGBA) == PIPE_MASK_RGBA &&
+           dst_width == info->src.resource->width0 &&
+           dst_height == info->src.resource->height0 &&
+           info->dst.box.x == 0 &&
+           info->dst.box.y == 0 &&
+           info->dst.box.width == dst_width &&
+           info->dst.box.height == dst_height &&
+           info->dst.box.depth == 1 &&
+           info->src.box.x == 0 &&
+           info->src.box.y == 0 &&
+           info->src.box.width == dst_width &&
+           info->src.box.height == dst_height &&
+           info->src.box.depth == 1 &&
+           dst->surface.u.legacy.level[info->dst.level].mode >= RADEON_SURF_MODE_1D &&
+           (!dst->cmask.size || !dst->dirty_level_mask) /* dst cannot be fast-cleared */) {
+               r600_blitter_begin(ctx, R600_COLOR_RESOLVE |
+                                  (info->render_condition_enable ? 0 : R600_DISABLE_RENDER_COND));
+               util_blitter_custom_resolve_color(rctx->blitter,
+                                                 info->dst.resource, info->dst.level,
+                                                 info->dst.box.z,
+                                                 info->src.resource, info->src.box.z,
+                                                 sample_mask, rctx->custom_blend_resolve,
+                                                 format);
+               r600_blitter_end(ctx);
+               return true;
+       }
+
+       /* Shader-based resolve is VERY SLOW. Instead, resolve into
+        * a temporary texture and blit.
+        */
+       memset(&templ, 0, sizeof(templ));
+       templ.target = PIPE_TEXTURE_2D;
+       templ.format = info->src.resource->format;
+       templ.width0 = info->src.resource->width0;
+       templ.height0 = info->src.resource->height0;
+       templ.depth0 = 1;
+       templ.array_size = 1;
+       templ.usage = PIPE_USAGE_DEFAULT;
+       templ.flags = R600_RESOURCE_FLAG_FORCE_TILING;
+
+       tmp = ctx->screen->resource_create(ctx->screen, &templ);
+       if (!tmp)
+               return false;
+
+       /* resolve */
+       r600_blitter_begin(ctx, R600_COLOR_RESOLVE |
+                          (info->render_condition_enable ? 0 : R600_DISABLE_RENDER_COND));
+       util_blitter_custom_resolve_color(rctx->blitter, tmp, 0, 0,
+                                         info->src.resource, info->src.box.z,
+                                         sample_mask, rctx->custom_blend_resolve,
+                                         format);
+       r600_blitter_end(ctx);
+
+       /* blit */
+       blit = *info;
+       blit.src.resource = tmp;
+       blit.src.box.z = 0;
+
+       r600_blitter_begin(ctx, R600_BLIT |
+                          (info->render_condition_enable ? 0 : R600_DISABLE_RENDER_COND));
+       util_blitter_blit(rctx->blitter, &blit);
+       r600_blitter_end(ctx);
+
+       pipe_resource_reference(&tmp, NULL);
+       return true;
 }
 
-static void r600_resource_copy_region(struct pipe_context *ctx,
-                                     struct pipe_resource *dst,
-                                     unsigned dst_level,
-                                     unsigned dstx, unsigned dsty, unsigned dstz,
-                                     struct pipe_resource *src,
-                                     unsigned src_level,
-                                     const struct pipe_box *src_box)
+static void r600_blit(struct pipe_context *ctx,
+                      const struct pipe_blit_info *info)
 {
-       struct r600_resource_texture *rsrc = (struct r600_resource_texture*)src;
-       struct texture_orig_info orig_info[2];
-       struct pipe_box sbox;
-       const struct pipe_box *psbox;
-       boolean restore_orig[2];
+       struct r600_context *rctx = (struct r600_context*)ctx;
+       struct r600_texture *rdst = (struct r600_texture *)info->dst.resource;
 
-       memset(orig_info, 0, sizeof(orig_info));
-
-       /* Fallback for buffers. */
-       if (dst->target == PIPE_BUFFER && src->target == PIPE_BUFFER) {
-               util_resource_copy_region(ctx, dst, dst_level, dstx, dsty, dstz,
-                                          src, src_level, src_box);
+       if (do_hardware_msaa_resolve(ctx, info)) {
                return;
        }
 
-       if (rsrc->depth && !rsrc->is_flushing_texture)
-               r600_texture_depth_flush(ctx, src, FALSE);
+       /* Using SDMA for copying to a linear texture in GTT is much faster.
+        * This improves DRI PRIME performance.
+        *
+        * resource_copy_region can't do this yet, because dma_copy calls it
+        * on failure (recursion).
+        */
+       if (rdst->surface.u.legacy.level[info->dst.level].mode ==
+           RADEON_SURF_MODE_LINEAR_ALIGNED &&
+           rctx->b.dma_copy &&
+           util_can_blit_via_copy_region(info, false)) {
+               rctx->b.dma_copy(ctx, info->dst.resource, info->dst.level,
+                                info->dst.box.x, info->dst.box.y,
+                                info->dst.box.z,
+                                info->src.resource, info->src.level,
+                                &info->src.box);
+               return;
+       }
 
-       restore_orig[0] = restore_orig[1] = FALSE;
+       assert(util_blitter_is_blit_supported(rctx->blitter, info));
 
-       if (util_format_is_compressed(src->format)) {
-               r600_compressed_to_blittable(src, src_level, &orig_info[0]);
-               restore_orig[0] = TRUE;
-               sbox.x = util_format_get_nblocksx(orig_info[0].format, src_box->x);
-               sbox.y = util_format_get_nblocksy(orig_info[0].format, src_box->y);
-               sbox.z = src_box->z;
-               sbox.width = util_format_get_nblocksx(orig_info[0].format, src_box->width);
-               sbox.height = util_format_get_nblocksy(orig_info[0].format, src_box->height);
-               sbox.depth = src_box->depth;
-               psbox=&sbox;
-       } else
-               psbox=src_box;
-
-       if (util_format_is_compressed(dst->format)) {
-               r600_compressed_to_blittable(dst, dst_level, &orig_info[1]);
-               restore_orig[1] = TRUE;
-               /* translate the dst box as well */
-               dstx = util_format_get_nblocksx(orig_info[1].format, dstx);
-               dsty = util_format_get_nblocksy(orig_info[1].format, dsty);
+       /* The driver doesn't decompress resources automatically while
+        * u_blitter is rendering. */
+       if (!r600_decompress_subresource(ctx, info->src.resource, info->src.level,
+                                        info->src.box.z,
+                                        info->src.box.z + info->src.box.depth - 1)) {
+               return; /* error */
        }
 
-       r600_hw_copy_region(ctx, dst, dst_level, dstx, dsty, dstz,
-                           src, src_level, psbox);
-
-       if (restore_orig[0])
-               r600_reset_blittable_to_compressed(src, &orig_info[0]);
+       if (rctx->screen->b.debug_flags & DBG_FORCE_DMA &&
+           util_try_blit_via_copy_region(ctx, info))
+               return;
 
-       if (restore_orig[1])
-               r600_reset_blittable_to_compressed(dst, &orig_info[1]);
+       r600_blitter_begin(ctx, R600_BLIT |
+                          (info->render_condition_enable ? 0 : R600_DISABLE_RENDER_COND));
+       util_blitter_blit(rctx->blitter, info);
+       r600_blitter_end(ctx);
 }
 
-void r600_init_blit_functions(struct r600_pipe_context *rctx)
+static void r600_flush_resource(struct pipe_context *ctx,
+                               struct pipe_resource *res)
 {
-       rctx->context.clear = r600_clear;
-       rctx->context.clear_render_target = r600_clear_render_target;
-       rctx->context.clear_depth_stencil = r600_clear_depth_stencil;
-       rctx->context.resource_copy_region = r600_resource_copy_region;
+       struct r600_texture *rtex = (struct r600_texture*)res;
+
+       assert(res->target != PIPE_BUFFER);
+
+       if (!rtex->is_depth && rtex->cmask.size) {
+               r600_blit_decompress_color(ctx, rtex, 0, res->last_level,
+                                          0, util_max_layer(res, 0));
+       }
 }
 
-void r600_blit_push_depth(struct pipe_context *ctx, struct r600_resource_texture *texture)
+void r600_init_blit_functions(struct r600_context *rctx)
 {
-       struct pipe_box sbox;
-
-       sbox.x = sbox.y = sbox.z = 0;
-       sbox.width = texture->resource.b.b.b.width0;
-       sbox.height = texture->resource.b.b.b.height0;
-       /* XXX that might be wrong */
-       sbox.depth = 1;
-
-       r600_hw_copy_region(ctx, (struct pipe_resource *)texture, 0,
-                           0, 0, 0,
-                           (struct pipe_resource *)texture->flushed_depth_texture, 0,
-                           &sbox);
+       rctx->b.b.clear = r600_clear;
+       rctx->b.b.clear_render_target = r600_clear_render_target;
+       rctx->b.b.clear_depth_stencil = r600_clear_depth_stencil;
+       rctx->b.b.resource_copy_region = r600_resource_copy_region;
+       rctx->b.b.blit = r600_blit;
+       rctx->b.b.flush_resource = r600_flush_resource;
+       rctx->b.clear_buffer = r600_clear_buffer;
+       rctx->b.blit_decompress_depth = r600_blit_decompress_depth;
 }