r600g/compute: Fix bug in compute memory pool
[mesa.git] / src / gallium / drivers / r600 / r600_blit.c
index a88cb5e4bba6b8aa904174b37dd162ba5b848654..60dda28e1845671f15967f029c32c6e2e9ed0a97 100644 (file)
@@ -55,21 +55,19 @@ static void r600_blitter_begin(struct pipe_context *ctx, enum r600_blitter_op op
 
        r600_suspend_nontimer_queries(rctx);
 
-       util_blitter_save_vertex_buffers(rctx->blitter,
-                                        util_last_bit(rctx->vertex_buffer_state.enabled_mask),
-                                        rctx->vertex_buffer_state.vb);
-       util_blitter_save_vertex_elements(rctx->blitter, rctx->vertex_elements);
+       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_so_targets(rctx->blitter, rctx->num_so_targets,
-                                    (struct pipe_stream_output_target**)rctx->so_targets);
-       util_blitter_save_rasterizer(rctx->blitter, rctx->states[R600_PIPE_STATE_RASTERIZER]);
+       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->viewport.state);
-               util_blitter_save_scissor(rctx->blitter, &rctx->scissor);
+               util_blitter_save_scissor(rctx->blitter, &rctx->scissor.scissor);
                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->states[R600_PIPE_STATE_DSA]);
+               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);
        }
@@ -90,6 +88,7 @@ static void r600_blitter_begin(struct pipe_context *ctx, enum r600_blitter_op op
        if ((op & R600_DISABLE_RENDER_COND) && rctx->current_render_cond) {
            util_blitter_save_render_condition(rctx->blitter,
                                               rctx->current_render_cond,
+                                              rctx->current_render_cond_cond,
                                               rctx->current_render_cond_mode);
         }
 }
@@ -100,21 +99,6 @@ static void r600_blitter_end(struct pipe_context *ctx)
         r600_resume_nontimer_queries(rctx);
 }
 
-static unsigned u_max_layer(struct pipe_resource *r, unsigned level)
-{
-       switch (r->target) {
-       case PIPE_TEXTURE_CUBE:
-               return 6 - 1;
-       case PIPE_TEXTURE_3D:
-               return u_minify(r->depth0, level) - 1;
-       case PIPE_TEXTURE_1D_ARRAY:
-       case PIPE_TEXTURE_2D_ARRAY:
-               return r->array_size - 1;
-       default:
-               return 0;
-       }
-}
-
 static unsigned u_max_sample(struct pipe_resource *r)
 {
        return r->nr_samples ? r->nr_samples - 1 : 0;
@@ -143,13 +127,13 @@ void r600_blit_decompress_depth(struct pipe_context *ctx,
        /* 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->chip_class == R600 && max_sample > 0) {
+       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;
@@ -161,14 +145,13 @@ void r600_blit_decompress_depth(struct pipe_context *ctx,
        rctx->db_misc_state.copy_sample = first_sample;
        rctx->db_misc_state.atom.dirty = true;
 
-
        for (level = first_level; level <= last_level; level++) {
                if (!staging && !(texture->dirty_level_mask & (1 << level)))
                        continue;
 
                /* The smaller the mipmap level, the less layers there are
                 * as far as 3D textures are concerned. */
-               max_layer = u_max_layer(&texture->resource.b.b, level);
+               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++) {
@@ -184,15 +167,10 @@ void r600_blit_decompress_depth(struct pipe_context *ctx,
                                surf_tmpl.u.tex.level = level;
                                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, &surf_tmpl);
 
                                surf_tmpl.format = flushed_depth_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;
-                               surf_tmpl.usage = PIPE_BIND_RENDER_TARGET;
                                cbsurf = ctx->create_surface(ctx,
                                                &flushed_depth_texture->resource.b.b, &surf_tmpl);
 
@@ -220,6 +198,57 @@ void r600_blit_decompress_depth(struct pipe_context *ctx,
        rctx->db_misc_state.atom.dirty = true;
 }
 
+static void r600_blit_decompress_depth_in_place(struct r600_context *rctx,
+                                                struct r600_texture *texture,
+                                                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;
+
+       /* Enable decompression in DB_RENDER_CONTROL */
+       rctx->db_misc_state.flush_depthstencil_in_place = true;
+       rctx->db_misc_state.atom.dirty = true;
+
+       surf_tmpl.format = texture->resource.b.b.format;
+
+       for (level = first_level; level <= last_level; level++) {
+               if (!(texture->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;
+
+                       zsurf = rctx->b.b.create_surface(&rctx->b.b, &texture->resource.b.b, &surf_tmpl);
+
+                       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);
+               }
+
+               /* 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) {
+                       texture->dirty_level_mask &= ~(1 << level);
+               }
+       }
+
+       /* Disable decompression in DB_RENDER_CONTROL */
+       rctx->db_misc_state.flush_depthstencil_in_place = false;
+       rctx->db_misc_state.atom.dirty = true;
+}
+
 void r600_decompress_depth_textures(struct r600_context *rctx,
                               struct r600_samplerview_state *textures)
 {
@@ -238,10 +267,17 @@ void r600_decompress_depth_textures(struct r600_context *rctx,
                tex = (struct r600_texture *)view->texture;
                assert(tex->is_depth && !tex->is_flushing_texture);
 
-               r600_blit_decompress_depth(&rctx->context, tex, NULL,
-                                          view->u.tex.first_level, view->u.tex.last_level,
-                                          0, u_max_layer(&tex->resource.b.b, view->u.tex.first_level),
-                                          0, u_max_sample(&tex->resource.b.b));
+               if (rctx->b.chip_class >= EVERGREEN ||
+                   r600_can_read_depth(tex)) {
+                       r600_blit_decompress_depth_in_place(rctx, tex,
+                                                  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));
+               }
        }
 }
 
@@ -253,8 +289,6 @@ static void r600_blit_decompress_color(struct pipe_context *ctx,
        struct r600_context *rctx = (struct r600_context *)ctx;
        unsigned layer, level, checked_last_layer, max_layer;
 
-       assert(rctx->chip_class != CAYMAN);
-
        if (!rtex->dirty_level_mask)
                return;
 
@@ -264,7 +298,7 @@ static void r600_blit_decompress_color(struct pipe_context *ctx,
 
                /* The smaller the mipmap level, the less layers there are
                 * as far as 3D textures are concerned. */
-               max_layer = u_max_layer(&rtex->resource.b.b, level);
+               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++) {
@@ -274,18 +308,16 @@ static void r600_blit_decompress_color(struct pipe_context *ctx,
                        surf_tmpl.u.tex.level = level;
                        surf_tmpl.u.tex.first_layer = layer;
                        surf_tmpl.u.tex.last_layer = layer;
-                       surf_tmpl.usage = PIPE_BIND_RENDER_TARGET;
                        cbsurf = ctx->create_surface(ctx, &rtex->resource.b.b, &surf_tmpl);
 
                        r600_blitter_begin(ctx, R600_DECOMPRESS);
-                       util_blitter_custom_color(rctx->blitter, cbsurf,
-                                                 rctx->custom_blend_decompress);
+                       util_blitter_custom_color(rctx->blitter, cbsurf, rctx->custom_blend_decompress);
                        r600_blitter_end(ctx);
 
                        pipe_surface_reference(&cbsurf, NULL);
                }
 
-               /* The texture will always be dirty if some layers or samples aren't flushed.
+               /* 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);
@@ -299,13 +331,6 @@ void r600_decompress_color_textures(struct r600_context *rctx,
        unsigned i;
        unsigned mask = textures->compressed_colortex_mask;
 
-       /* Cayman cannot decompress an MSAA colorbuffer,
-        * but it can read it compressed, so skip this. */
-       assert(rctx->chip_class != CAYMAN);
-       if (rctx->chip_class == CAYMAN) {
-               return;
-       }
-
        while (mask) {
                struct pipe_sampler_view *view;
                struct r600_texture *tex;
@@ -318,9 +343,9 @@ void r600_decompress_color_textures(struct r600_context *rctx,
                tex = (struct r600_texture *)view->texture;
                assert(tex->cmask_size && tex->fmask_size);
 
-               r600_blit_decompress_color(&rctx->context, tex,
+               r600_blit_decompress_color(&rctx->b.b, tex,
                                           view->u.tex.first_level, view->u.tex.last_level,
-                                          0, u_max_layer(&tex->resource.b.b, view->u.tex.first_level));
+                                          0, util_max_layer(&tex->resource.b.b, view->u.tex.first_level));
        }
 }
 
@@ -337,14 +362,21 @@ static bool r600_decompress_subresource(struct pipe_context *ctx,
        struct r600_texture *rtex = (struct r600_texture*)tex;
 
        if (rtex->is_depth && !rtex->is_flushing_texture) {
-               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 (rctx->chip_class != CAYMAN && rtex->fmask_size && rtex->cmask_size) {
+               if (rctx->b.chip_class >= EVERGREEN ||
+                   r600_can_read_depth(rtex)) {
+                       r600_blit_decompress_depth_in_place(rctx, rtex,
+                                                  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->fmask_size && rtex->cmask_size) {
                r600_blit_decompress_color(ctx, rtex, level, level,
                                           first_layer, last_layer);
        }
@@ -355,6 +387,8 @@ static boolean is_simple_msaa_resolve(const struct pipe_blit_info *info)
 {
        unsigned dst_width = u_minify(info->dst.resource->width0, info->dst.level);
        unsigned dst_height = u_minify(info->dst.resource->height0, info->dst.level);
+       struct r600_texture *dst = (struct r600_texture*)info->dst.resource;
+       unsigned dst_tile_mode = dst->surface.level[info->dst.level].mode;
 
        return info->dst.resource->format == info->src.resource->format &&
                info->dst.resource->format == info->dst.format &&
@@ -370,7 +404,64 @@ static boolean is_simple_msaa_resolve(const struct pipe_blit_info *info)
                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.height == dst_height &&
+               /* Dst must be tiled. If it's not, we have to use a temporary
+                * resource which is tiled. */
+               dst_tile_mode >= RADEON_SURF_MODE_1D;
+}
+
+static void r600_clear_buffer(struct pipe_context *ctx, struct pipe_resource *dst,
+                             unsigned offset, unsigned size, unsigned char value);
+
+static void evergreen_set_clear_color(struct pipe_surface *cbuf,
+                                     const union pipe_color_union *color)
+{
+       unsigned *clear_value = ((struct r600_texture *)cbuf->texture)->color_clear_value;
+       union util_color uc;
+
+       memset(&uc, 0, sizeof(uc));
+
+       if (util_format_is_pure_uint(cbuf->format)) {
+               util_format_write_4ui(cbuf->format, color->ui, 0, &uc, 0, 0, 0, 1, 1);
+       } else if (util_format_is_pure_sint(cbuf->format)) {
+               util_format_write_4i(cbuf->format, color->i, 0, &uc, 0, 0, 0, 1, 1);
+       } else {
+               util_pack_color(color->f, cbuf->format, &uc);
+       }
+
+       memcpy(clear_value, &uc, 2 * sizeof(uint32_t));
+}
+
+static bool can_fast_clear_color(struct pipe_context *ctx)
+{
+       struct r600_context *rctx = (struct r600_context *)ctx;
+       struct pipe_framebuffer_state *fb = &rctx->framebuffer.state;
+       int i;
+
+       if (rctx->b.chip_class < EVERGREEN) {
+               return false;
+       }
+
+       for (i = 0; i < fb->nr_cbufs; i++) {
+               struct r600_texture *tex = (struct r600_texture *)fb->cbufs[i]->texture;
+
+               if (tex->cmask_size == 0) {
+                       return false;
+               }
+
+               /* 128-bit formats are unuspported */
+               if (util_format_get_blocksizebits(fb->cbufs[i]->format) > 64) {
+                       return false;
+               }
+
+               /* the clear is allowed if all layers are bound */
+               if (fb->cbufs[i]->u.tex.first_layer != 0 ||
+                   fb->cbufs[i]->u.tex.last_layer != util_max_layer(&tex->resource.b.b, 0)) {
+                       return false;
+               }
+       }
+
+       return true;
 }
 
 static void r600_clear(struct pipe_context *ctx, unsigned buffers,
@@ -380,11 +471,58 @@ static void r600_clear(struct pipe_context *ctx, unsigned buffers,
        struct r600_context *rctx = (struct r600_context *)ctx;
        struct pipe_framebuffer_state *fb = &rctx->framebuffer.state;
 
+       /* fast color clear on AA framebuffers (EG+) */
+       if ((buffers & PIPE_CLEAR_COLOR) && can_fast_clear_color(ctx)) {
+               int i;
+
+               for (i = 0; i < fb->nr_cbufs; i++) {
+                       struct r600_texture *tex = (struct r600_texture *)fb->cbufs[i]->texture;
+
+                       evergreen_set_clear_color(fb->cbufs[i], color);
+                       r600_clear_buffer(ctx, fb->cbufs[i]->texture,
+                                       tex->cmask_offset, tex->cmask_size, 0);
+                       tex->dirty_level_mask |= 1 << fb->cbufs[i]->u.tex.level;
+               }
+
+               rctx->framebuffer.atom.dirty = true;
+
+               buffers &= ~PIPE_CLEAR_COLOR;
+               if (!buffers)
+                       return;
+       }
+
+       /* 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 && !level && rtex->surface.array_size == 1) {
+                       if (rtex->depth_clear != depth) {
+                               rtex->depth_clear = depth;
+                               rctx->db_state.atom.dirty = true;
+                       }
+                       rctx->db_misc_state.htile_clear = true;
+                       rctx->db_misc_state.atom.dirty = true;
+               }
+       }
+
        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);
+                          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;
+               rctx->db_misc_state.atom.dirty = true;
+       }
 }
 
 static void r600_clear_render_target(struct pipe_context *ctx,
@@ -417,23 +555,75 @@ static void r600_clear_depth_stencil(struct pipe_context *ctx,
        r600_blitter_end(ctx);
 }
 
-void r600_copy_buffer(struct pipe_context *ctx, struct
-                     pipe_resource *dst, unsigned dstx,
+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->has_streamout &&
-           /* Require dword alignment. */
-           dstx % 4 == 0 && src_box->x % 4 == 0 && src_box->width % 4 == 0) {
+       if (rctx->screen->has_cp_dma) {
+               r600_cp_dma_copy_buffer(rctx, dst, dstx, src, src_box->x, src_box->width);
+       }
+       else if (rctx->screen->has_streamout &&
+                /* Require 4-byte alignment. */
+                dstx % 4 == 0 && src_box->x % 4 == 0 && src_box->width % 4 == 0) {
+
+               /* Flush both resources. */
+               r600_flag_resource_cache_flush(rctx, src);
+               r600_flag_resource_cache_flush(rctx, dst);
+
                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);
+
+               /* Flush the dst in case the 3D engine has been prefetching the resource. */
+               r600_flag_resource_cache_flush(rctx, dst);
        } else {
                util_resource_copy_region(ctx, dst, 0, dstx, 0, 0, src, 0, src_box);
        }
 }
 
+static void r600_clear_buffer(struct pipe_context *ctx, struct pipe_resource *dst,
+                             unsigned offset, unsigned size, unsigned char value)
+{
+       struct r600_context *rctx = (struct r600_context*)ctx;
+       uint32_t v = value;
+
+       if (rctx->screen->has_cp_dma &&
+           rctx->b.chip_class >= EVERGREEN &&
+           offset % 4 == 0 && size % 4 == 0) {
+               uint32_t clear_value = v | (v << 8) | (v << 16) | (v << 24);
+
+               evergreen_cp_dma_clear_buffer(rctx, dst, offset, size, clear_value);
+       } else if (rctx->screen->has_streamout && offset % 4 == 0 && size % 4 == 0) {
+               union pipe_color_union clear_value;
+
+               clear_value.ui[0] = v | (v << 8) | (v << 16) | (v << 24);
+
+               r600_flag_resource_cache_flush(rctx, dst);
+
+               r600_blitter_begin(ctx, R600_DISABLE_RENDER_COND);
+               util_blitter_clear_buffer(rctx->blitter, dst, offset, size,
+                                         1, &clear_value);
+               r600_blitter_end(ctx);
+
+               /* Flush again in case the 3D engine has been prefetching the resource. */
+               r600_flag_resource_cache_flush(rctx, dst);
+       } else {
+               char *map = r600_buffer_mmap_sync_with_rings(rctx, r600_resource(dst),
+                                                            PIPE_TRANSFER_WRITE);
+               memset(map + offset, value, size);
+       }
+}
+
+void r600_screen_clear_buffer(struct r600_screen *rscreen, struct pipe_resource *dst,
+                             unsigned offset, unsigned size, unsigned char value)
+{
+       pipe_mutex_lock(rscreen->aux_context_lock);
+       r600_clear_buffer(rscreen->aux_context, dst, offset, size, value);
+       rscreen->aux_context->flush(rscreen->aux_context, NULL, 0);
+       pipe_mutex_unlock(rscreen->aux_context_lock);
+}
+
 static bool util_format_is_subsampled_2x1_32bpp(enum pipe_format format)
 {
        const struct util_format_description *desc = util_format_description(format);
@@ -453,12 +643,10 @@ static void r600_resource_copy_region(struct pipe_context *ctx,
                                      const struct pipe_box *src_box)
 {
        struct r600_context *rctx = (struct r600_context *)ctx;
-       struct r600_texture *rsrc = (struct r600_texture*)src;
-       struct r600_texture *rdst = (struct r600_texture*)dst;
        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;
-       struct pipe_box sbox;
+       struct pipe_box sbox, dstbox;
 
        /* Handle buffers first. */
        if (dst->target == PIPE_BUFFER && src->target == PIPE_BUFFER) {
@@ -475,14 +663,14 @@ static void r600_resource_copy_region(struct pipe_context *ctx,
                return; /* error */
        }
 
-       dst_width = rdst->surface.level[dst_level].npix_x;
-       dst_height = rdst->surface.level[dst_level].npix_y;
+       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 = rsrc->surface.level[src_level].npix_x;
-       src_heightFL = rsrc->surface.level[src_level].npix_y;
+        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, src_box);
+       util_blitter_default_dst_texture(&dst_templ, dst, dst_level, dstz);
        util_blitter_default_src_texture(&src_templ, src, src_level);
 
        if (util_format_is_compressed(src->format)) {
@@ -494,12 +682,12 @@ static void r600_resource_copy_region(struct pipe_context *ctx,
                        src_templ.format = PIPE_FORMAT_R32G32B32A32_UINT; /* 128-bit block */
                dst_templ.format = src_templ.format;
 
-               dst_width = util_format_get_nblocksx(dst->format, rdst->surface.level[dst_level].npix_x);
-               dst_height = util_format_get_nblocksy(dst->format, rdst->surface.level[dst_level].npix_y);
-               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, rsrc->surface.level[src_level].npix_x);
-               src_heightFL = util_format_get_nblocksy(src->format, rsrc->surface.level[src_level].npix_y);
+               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);
@@ -518,9 +706,9 @@ static void r600_resource_copy_region(struct pipe_context *ctx,
                        src_templ.format = PIPE_FORMAT_R8G8B8A8_UINT;
                        dst_templ.format = PIPE_FORMAT_R8G8B8A8_UINT;
 
-                       dst_width = util_format_get_nblocksx(dst->format, rdst->surface.level[dst_level].npix_x);
-                       src_width0 = util_format_get_nblocksx(src->format, src->width0);
-                       src_widthFL = util_format_get_nblocksx(src->format, rsrc->surface.level[src_level].npix_x);
+                       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);
 
@@ -536,10 +724,22 @@ static void r600_resource_copy_region(struct pipe_context *ctx,
                                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);
@@ -550,7 +750,7 @@ static void r600_resource_copy_region(struct pipe_context *ctx,
 
        dst_view = r600_create_surface_custom(ctx, dst, &dst_templ, dst_width, dst_height);
 
-       if (rctx->chip_class >= EVERGREEN) {
+       if (rctx->b.chip_class >= EVERGREEN) {
                src_view = evergreen_create_sampler_view_custom(ctx, src, &src_templ,
                                                                src_width0, src_height0);
        } else {
@@ -558,22 +758,57 @@ static void r600_resource_copy_region(struct pipe_context *ctx,
                                                           src_widthFL, src_heightFL);
        }
 
+        u_box_3d(dstx, dsty, dstz, abs(src_box->width), abs(src_box->height),
+                 abs(src_box->depth), &dstbox);
+
        /* Copy. */
-       /* XXX Multisample texturing is unimplemented on Cayman. In the meantime,
-        * copy only the first sample (which is the only one that is uncompressed
-        * and therefore doesn't return garbage). */
        r600_blitter_begin(ctx, R600_COPY_TEXTURE);
-       util_blitter_blit_generic(rctx->blitter, dst_view, dstx, dsty,
-                                 abs(src_box->width), abs(src_box->height),
+       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,
-                                 rctx->chip_class != CAYMAN);
+                                 TRUE);
        r600_blitter_end(ctx);
 
        pipe_surface_reference(&dst_view, NULL);
        pipe_sampler_view_reference(&src_view, NULL);
 }
 
+/* For MSAA integer resolving to work, we change the format to NORM using this function. */
+static enum pipe_format int_to_norm_format(enum pipe_format format)
+{
+       switch (format) {
+#define REPLACE_FORMAT_SIGN(format,sign) \
+       case PIPE_FORMAT_##format##_##sign##INT: \
+               return PIPE_FORMAT_##format##_##sign##NORM
+#define REPLACE_FORMAT(format) \
+               REPLACE_FORMAT_SIGN(format, U); \
+               REPLACE_FORMAT_SIGN(format, S)
+
+       REPLACE_FORMAT_SIGN(B10G10R10A2, U);
+       REPLACE_FORMAT(R8);
+       REPLACE_FORMAT(R8G8);
+       REPLACE_FORMAT(R8G8B8X8);
+       REPLACE_FORMAT(R8G8B8A8);
+       REPLACE_FORMAT(A8);
+       REPLACE_FORMAT(I8);
+       REPLACE_FORMAT(L8);
+       REPLACE_FORMAT(L8A8);
+       REPLACE_FORMAT(R16);
+       REPLACE_FORMAT(R16G16);
+       REPLACE_FORMAT(R16G16B16X16);
+       REPLACE_FORMAT(R16G16B16A16);
+       REPLACE_FORMAT(A16);
+       REPLACE_FORMAT(I16);
+       REPLACE_FORMAT(L16);
+       REPLACE_FORMAT(L16A16);
+
+#undef REPLACE_FORMAT
+#undef REPLACE_FORMAT_SIGN
+       default:
+               return format;
+       }
+}
+
 static void r600_msaa_color_resolve(struct pipe_context *ctx,
                              const struct pipe_blit_info *info)
 {
@@ -582,7 +817,7 @@ static void r600_msaa_color_resolve(struct pipe_context *ctx,
        struct pipe_resource *tmp, templ;
        struct pipe_blit_info blit;
        unsigned sample_mask =
-               rctx->chip_class == CAYMAN ? ~0 :
+               rctx->b.chip_class == CAYMAN ? ~0 :
                ((1ull << MAX2(1, info->src.resource->nr_samples)) - 1);
 
        assert(info->src.level == 0);
@@ -595,7 +830,8 @@ static void r600_msaa_color_resolve(struct pipe_context *ctx,
                                                  info->dst.resource, info->dst.level,
                                                  info->dst.box.z,
                                                  info->src.resource, info->src.box.z,
-                                                 sample_mask, rctx->custom_blend_resolve);
+                                                 sample_mask, rctx->custom_blend_resolve,
+                                                  int_to_norm_format(info->dst.format));
                r600_blitter_end(ctx);
                return;
        }
@@ -611,7 +847,7 @@ static void r600_msaa_color_resolve(struct pipe_context *ctx,
        templ.nr_samples = 0;
        templ.usage = PIPE_USAGE_STATIC;
        templ.bind = PIPE_BIND_RENDER_TARGET | PIPE_BIND_SAMPLER_VIEW;
-       templ.flags = 0;
+       templ.flags = R600_RESOURCE_FLAG_FORCE_TILING; /* dst must not have a linear layout */
 
        tmp = screen->resource_create(screen, &templ);
 
@@ -620,7 +856,8 @@ static void r600_msaa_color_resolve(struct pipe_context *ctx,
        util_blitter_custom_resolve_color(rctx->blitter,
                                          tmp, 0, 0,
                                          info->src.resource, info->src.box.z,
-                                         sample_mask, rctx->custom_blend_resolve);
+                                         sample_mask, rctx->custom_blend_resolve,
+                                          int_to_norm_format(tmp->format));
        r600_blitter_end(ctx);
 
        /* blit */
@@ -645,7 +882,7 @@ static void r600_blit(struct pipe_context *ctx,
        if (info->src.resource->nr_samples > 1 &&
            info->dst.resource->nr_samples <= 1 &&
            !util_format_is_depth_or_stencil(info->src.resource->format) &&
-           !util_format_is_pure_integer(info->src.resource->format)) {
+           !util_format_is_pure_integer(int_to_norm_format(info->src.resource->format))) {
                r600_msaa_color_resolve(ctx, info);
                return;
        }
@@ -665,9 +902,9 @@ static void r600_blit(struct pipe_context *ctx,
 
 void r600_init_blit_functions(struct r600_context *rctx)
 {
-       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;
-       rctx->context.blit = r600_blit;
+       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;
 }