radeonsi: port texture improvements from r600g
authorMarek Olšák <marek.olsak@amd.com>
Tue, 6 Aug 2013 19:30:59 +0000 (21:30 +0200)
committerMarek Olšák <marek.olsak@amd.com>
Fri, 16 Aug 2013 23:48:25 +0000 (01:48 +0200)
This started as an attempt to add support for MSAA texture transfers and
MSAA depth-stencil decompression for the DB->CB copy path.
It has gotten a bit out of control, but it's for the greater good.

Some changes do not make much sense, they are there just to make it look
like the other driver.

With a few cosmetic modifications, r600_texture.c can be shared with
a symlink.

Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
src/gallium/drivers/r600/r600_blit.c
src/gallium/drivers/r600/r600_texture.c
src/gallium/drivers/radeonsi/r600_blit.c
src/gallium/drivers/radeonsi/r600_resource.h
src/gallium/drivers/radeonsi/r600_texture.c
src/gallium/drivers/radeonsi/radeonsi_pipe.c
src/gallium/drivers/radeonsi/radeonsi_pipe.h
src/gallium/drivers/radeonsi/si_state.c

index 2230e7bbdea7bf3b168cc4017ec40461a029f85a..1c22a75582b858efcdb9303bd3bcc3be1ddd9e83 100644 (file)
@@ -171,9 +171,6 @@ void r600_blit_decompress_depth(struct pipe_context *ctx,
                                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;
                                cbsurf = ctx->create_surface(ctx,
                                                &flushed_depth_texture->resource.b.b, &surf_tmpl);
 
index 36cca17eb110ae0be0f6d4fbfcfa951f9f0a9165..742e98227e6e8057ccc48090d6094bf162c67297 100644 (file)
@@ -458,8 +458,7 @@ r600_texture_create_object(struct pipe_screen *screen,
        rtex->is_depth = util_format_has_depth(util_format_description(rtex->resource.b.b.format));
 
        rtex->surface = *surface;
-       r = r600_setup_surface(screen, rtex,
-                              pitch_in_bytes_override);
+       r = r600_setup_surface(screen, rtex, pitch_in_bytes_override);
        if (r) {
                FREE(rtex);
                return NULL;
index e533e0f2f26c1fe45def53d28a4a637ef03c53b2..396ee01ed8cedd53f04eb44c5d8d5342f074f273 100644 (file)
@@ -102,23 +102,31 @@ static void r600_blitter_end(struct pipe_context *ctx)
        r600_context_queries_resume(rctx);
 }
 
-void si_blit_uncompress_depth(struct pipe_context *ctx,
+static unsigned u_max_sample(struct pipe_resource *r)
+{
+       return r->nr_samples ? r->nr_samples - 1 : 0;
+}
+
+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_layer, unsigned last_layer,
+               unsigned first_sample, unsigned last_sample)
 {
        struct r600_context *rctx = (struct r600_context *)ctx;
-       unsigned layer, level, checked_last_layer, max_layer;
+       unsigned layer, level, sample, checked_last_layer, max_layer, max_sample;
        float depth = 1.0f;
        const struct util_format_description *desc;
-       void *custom_dsa;
+       void **custom_dsa;
        struct r600_texture *flushed_depth_texture = staging ?
                        staging : texture->flushed_depth_texture;
 
        if (!staging && !texture->dirty_level_mask)
                return;
 
+       max_sample = u_max_sample(&texture->resource.b.b);
+
        desc = util_format_description(flushed_depth_texture->resource.b.b.format);
        switch (util_format_has_depth(desc) | util_format_has_stencil(desc) << 1) {
        default:
@@ -144,30 +152,35 @@ void si_blit_uncompress_depth(struct pipe_context *ctx,
                checked_last_layer = last_layer < max_layer ? last_layer : max_layer;
 
                for (layer = first_layer; layer <= checked_last_layer; layer++) {
-                       struct pipe_surface *zsurf, *cbsurf, surf_tmpl;
+                       for (sample = first_sample; sample <= last_sample; sample++) {
+                               struct pipe_surface *zsurf, *cbsurf, surf_tmpl;
 
-                       surf_tmpl.format = texture->real_format;
-                       surf_tmpl.u.tex.level = level;
-                       surf_tmpl.u.tex.first_layer = layer;
-                       surf_tmpl.u.tex.last_layer = layer;
+                               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);
+                               zsurf = ctx->create_surface(ctx, &texture->resource.b.b, &surf_tmpl);
 
-                       surf_tmpl.format = flushed_depth_texture->real_format;
-                       cbsurf = ctx->create_surface(ctx,
-                                       (struct pipe_resource*)flushed_depth_texture, &surf_tmpl);
+                               surf_tmpl.format = flushed_depth_texture->resource.b.b.format;
+                               cbsurf = ctx->create_surface(ctx,
+                                               (struct pipe_resource*)flushed_depth_texture, &surf_tmpl);
 
-                       r600_blitter_begin(ctx, R600_DECOMPRESS);
-                       util_blitter_custom_depth_stencil(rctx->blitter, zsurf, cbsurf, ~0, custom_dsa, depth);
-                       r600_blitter_end(ctx);
+                               r600_blitter_begin(ctx, R600_DECOMPRESS);
+                               util_blitter_custom_depth_stencil(rctx->blitter, zsurf, cbsurf, 1 << sample,
+                                                                 custom_dsa[sample], depth);
+                               r600_blitter_end(ctx);
 
-                       pipe_surface_reference(&zsurf, NULL);
-                       pipe_surface_reference(&cbsurf, NULL);
+                               pipe_surface_reference(&zsurf, NULL);
+                               pipe_surface_reference(&cbsurf, NULL);
+                       }
                }
 
                /* The texture will always be dirty if some layers aren't flushed.
                 * I don't think this case can occur though. */
-               if (!staging && first_layer == 0 && last_layer == max_layer) {
+               if (!staging &&
+                   first_layer == 0 && last_layer == max_layer &&
+                   first_sample == 0 && last_sample == max_sample) {
                        texture->dirty_level_mask &= ~(1 << level);
                }
        }
@@ -388,7 +401,7 @@ static void r600_compressed_to_blittable(struct pipe_resource *tex,
                                   struct texture_orig_info *orig)
 {
        struct r600_texture *rtex = (struct r600_texture*)tex;
-       unsigned pixsize = util_format_get_blocksize(rtex->real_format);
+       unsigned pixsize = util_format_get_blocksize(rtex->resource.b.b.format);
        int new_format;
        int new_height, new_width;
 
index ab5c7b7a644230360545dce710d9e5148b18d56e..da02b070bf566a6d1ce1e5398d88a636c8091d7c 100644 (file)
@@ -37,7 +37,7 @@ struct r600_transfer {
        /* Buffer transfer. */
        struct pipe_transfer            *buffer_transfer;
        unsigned                        offset;
-       struct pipe_resource            *staging;
+       struct si_resource              *staging;
 };
 
 struct r600_fmask_info {
@@ -60,11 +60,6 @@ struct r600_cmask_info {
 struct r600_texture {
        struct si_resource              resource;
 
-       /* If this resource is a depth-stencil buffer on evergreen, this contains
-        * the depth part of the format. There is a separate stencil resource
-        * for the stencil buffer below. */
-       enum pipe_format                real_format;
-
        unsigned                        size;
        unsigned                        pitch_override;
        unsigned                        is_depth;
index 1507239712637d99225033bab5beb445baa4b0a2..ee8ee14fe32f40e1b3964220942b55455b51cf14 100644 (file)
 #include "r600_resource.h"
 #include "sid.h"
 
+/* Same as resource_copy_region, except that both upsampling and downsampling are allowed. */
+static void r600_copy_region_with_blit(struct pipe_context *pipe,
+                                      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 pipe_blit_info blit;
+
+       memset(&blit, 0, sizeof(blit));
+       blit.src.resource = src;
+       blit.src.format = src->format;
+       blit.src.level = src_level;
+       blit.src.box = *src_box;
+       blit.dst.resource = dst;
+       blit.dst.format = dst->format;
+       blit.dst.level = dst_level;
+       blit.dst.box.x = dstx;
+       blit.dst.box.y = dsty;
+       blit.dst.box.z = dstz;
+       blit.dst.box.width = src_box->width;
+       blit.dst.box.height = src_box->height;
+       blit.dst.box.depth = src_box->depth;
+       blit.mask = util_format_get_mask(src->format) &
+                   util_format_get_mask(dst->format);
+       blit.filter = PIPE_TEX_FILTER_NEAREST;
+
+       if (blit.mask) {
+               pipe->blit(pipe, &blit);
+       }
+}
+
 /* Copy from a full GPU texture to a transfer's staging one. */
 static void r600_copy_to_staging_texture(struct pipe_context *ctx, struct r600_transfer *rtransfer)
 {
        struct pipe_transfer *transfer = (struct pipe_transfer*)rtransfer;
-       struct pipe_resource *texture = transfer->resource;
+       struct pipe_resource *dst = &rtransfer->staging->b.b;
+       struct pipe_resource *src = transfer->resource;
 
-       ctx->resource_copy_region(ctx, rtransfer->staging,
-                               0, 0, 0, 0, texture, transfer->level,
-                               &transfer->box);
+       if (src->nr_samples > 1) {
+               r600_copy_region_with_blit(ctx, dst, 0, 0, 0, 0,
+                                          src, transfer->level, &transfer->box);
+               return;
+       }
+
+       ctx->resource_copy_region(ctx, dst, 0, 0, 0, 0,
+                                 src, transfer->level, &transfer->box);
 }
 
 /* Copy from a transfer's staging texture to a full GPU one. */
 static void r600_copy_from_staging_texture(struct pipe_context *ctx, struct r600_transfer *rtransfer)
 {
        struct pipe_transfer *transfer = (struct pipe_transfer*)rtransfer;
-       struct pipe_resource *texture = transfer->resource;
+       struct pipe_resource *dst = transfer->resource;
+       struct pipe_resource *src = &rtransfer->staging->b.b;
        struct pipe_box sbox;
 
        u_box_3d(0, 0, 0, transfer->box.width, transfer->box.height, transfer->box.depth, &sbox);
 
-       ctx->resource_copy_region(ctx, texture, transfer->level,
+       if (dst->nr_samples > 1) {
+               r600_copy_region_with_blit(ctx, dst, transfer->level,
+                                          transfer->box.x, transfer->box.y, transfer->box.z,
+                                          src, 0, &sbox);
+               return;
+       }
+
+       ctx->resource_copy_region(ctx, dst, transfer->level,
                                  transfer->box.x, transfer->box.y, transfer->box.z,
-                                 rtransfer->staging,
-                                 0, &sbox);
+                                 src, 0, &sbox);
 }
 
-static unsigned r600_texture_get_offset(struct r600_texture *rtex,
-                                       unsigned level, unsigned layer)
+static unsigned r600_texture_get_offset(struct r600_texture *rtex, unsigned level,
+                                       const struct pipe_box *box)
 {
+       enum pipe_format format = rtex->resource.b.b.format;
+
        return rtex->surface.level[level].offset +
-              layer * rtex->surface.level[level].slice_size;
+              box->z * rtex->surface.level[level].slice_size +
+              box->y / util_format_get_blockheight(format) * rtex->surface.level[level].pitch_bytes +
+              box->x / util_format_get_blockwidth(format) * util_format_get_blocksize(format);
 }
 
 static int r600_init_surface(struct r600_screen *rscreen,
@@ -163,7 +214,6 @@ static int r600_init_surface(struct r600_screen *rscreen,
 
 static int r600_setup_surface(struct pipe_screen *screen,
                              struct r600_texture *rtex,
-                             unsigned array_mode,
                              unsigned pitch_in_bytes_override)
 {
        struct r600_screen *rscreen = (struct r600_screen*)screen;
@@ -229,196 +279,7 @@ static void r600_texture_destroy(struct pipe_screen *screen,
        FREE(rtex);
 }
 
-static void *si_texture_transfer_map(struct pipe_context *ctx,
-                                    struct pipe_resource *texture,
-                                    unsigned level,
-                                    unsigned usage,
-                                    const struct pipe_box *box,
-                                    struct pipe_transfer **ptransfer)
-{
-       struct r600_context *rctx = (struct r600_context *)ctx;
-       struct r600_texture *rtex = (struct r600_texture*)texture;
-       struct r600_transfer *trans;
-       boolean use_staging_texture = FALSE;
-       struct radeon_winsys_cs_handle *buf;
-       enum pipe_format format = texture->format;
-       unsigned offset = 0;
-       char *map;
-
-       /* We cannot map a tiled texture directly because the data is
-        * in a different order, therefore we do detiling using a blit.
-        *
-        * Also, use a temporary in GTT memory for read transfers, as
-        * the CPU is much happier reading out of cached system memory
-        * than uncached VRAM.
-        */
-       if (rtex->surface.level[level].mode != RADEON_SURF_MODE_LINEAR_ALIGNED &&
-           rtex->surface.level[level].mode != RADEON_SURF_MODE_LINEAR)
-               use_staging_texture = TRUE;
-
-       /* XXX: Use a staging texture for uploads if the underlying BO
-        * is busy.  No interface for checking that currently? so do
-        * it eagerly whenever the transfer doesn't require a readback
-        * and might block.
-        */
-       if ((usage & PIPE_TRANSFER_WRITE) &&
-                       !(usage & (PIPE_TRANSFER_READ |
-                                       PIPE_TRANSFER_DONTBLOCK |
-                                       PIPE_TRANSFER_UNSYNCHRONIZED)))
-               use_staging_texture = TRUE;
-
-       if (texture->flags & R600_RESOURCE_FLAG_TRANSFER)
-               use_staging_texture = FALSE;
-
-       if (use_staging_texture && (usage & PIPE_TRANSFER_MAP_DIRECTLY))
-               return NULL;
-
-       trans = CALLOC_STRUCT(r600_transfer);
-       if (trans == NULL)
-               return NULL;
-       pipe_resource_reference(&trans->transfer.resource, texture);
-       trans->transfer.level = level;
-       trans->transfer.usage = usage;
-       trans->transfer.box = *box;
-       if (rtex->is_depth) {
-               /* XXX: only readback the rectangle which is being mapped?
-               */
-               /* XXX: when discard is true, no need to read back from depth texture
-               */
-               struct r600_texture *staging_depth;
-
-               if (!r600_init_flushed_depth_texture(ctx, texture, &staging_depth)) {
-                       R600_ERR("failed to create temporary texture to hold untiled copy\n");
-                       pipe_resource_reference(&trans->transfer.resource, NULL);
-                       FREE(trans);
-                       return NULL;
-               }
-               si_blit_uncompress_depth(ctx, rtex, staging_depth,
-                                        level, level,
-                                        box->z, box->z + box->depth - 1);
-               trans->transfer.stride = staging_depth->surface.level[level].pitch_bytes;
-               trans->transfer.layer_stride = staging_depth->surface.level[level].slice_size;
-               trans->offset = r600_texture_get_offset(staging_depth, level, box->z);
-
-               trans->staging = &staging_depth->resource.b.b;
-       } else if (use_staging_texture) {
-               struct pipe_resource resource;
-               struct r600_texture *staging;
-
-               memset(&resource, 0, sizeof(resource));
-               resource.format = texture->format;
-               resource.width0 = box->width;
-               resource.height0 = box->height;
-               resource.depth0 = 1;
-               resource.array_size = 1;
-               resource.usage = PIPE_USAGE_STAGING;
-               resource.flags = R600_RESOURCE_FLAG_TRANSFER;
-
-               /* We must set the correct texture target and dimensions if needed for a 3D transfer. */
-               if (box->depth > 1 && util_max_layer(texture, level) > 0)
-                       resource.target = texture->target;
-               else
-                       resource.target = PIPE_TEXTURE_2D;
-
-               switch (resource.target) {
-               case PIPE_TEXTURE_1D_ARRAY:
-               case PIPE_TEXTURE_2D_ARRAY:
-               case PIPE_TEXTURE_CUBE_ARRAY:
-                       resource.array_size = box->depth;
-                       break;
-               case PIPE_TEXTURE_3D:
-                       resource.depth0 = box->depth;
-                       break;
-               default:;
-               }
-               /* Create the temporary texture. */
-               staging = (struct r600_texture*)ctx->screen->resource_create(ctx->screen, &resource);
-               if (staging == NULL) {
-                       R600_ERR("failed to create temporary texture to hold untiled copy\n");
-                       pipe_resource_reference(&trans->transfer.resource, NULL);
-                       FREE(trans);
-                       return NULL;
-               }
-
-               trans->staging = &staging->resource.b.b;
-               trans->transfer.stride = staging->surface.level[0].pitch_bytes;
-               trans->transfer.layer_stride = staging->surface.level[0].slice_size;
-               if (usage & PIPE_TRANSFER_READ) {
-                       r600_copy_to_staging_texture(ctx, trans);
-                       /* Always referenced in the blit. */
-                       radeonsi_flush(ctx, NULL, 0);
-               }
-       } else {
-               trans->transfer.stride = rtex->surface.level[level].pitch_bytes;
-               trans->transfer.layer_stride = rtex->surface.level[level].slice_size;
-               trans->offset = r600_texture_get_offset(rtex, level, box->z);
-       }
-
-       if (trans->staging) {
-               buf = si_resource(trans->staging)->cs_buf;
-       } else {
-               buf = rtex->resource.cs_buf;
-       }
-
-       if (rtex->is_depth || !trans->staging)
-               offset = trans->offset +
-                       box->y / util_format_get_blockheight(format) * trans->transfer.stride +
-                       box->x / util_format_get_blockwidth(format) * util_format_get_blocksize(format);
-
-       if (!(map = rctx->ws->buffer_map(buf, rctx->cs, usage))) {
-               pipe_resource_reference(&trans->staging, NULL);
-               pipe_resource_reference(&trans->transfer.resource, NULL);
-               FREE(trans);
-               return NULL;
-       }
-
-       *ptransfer = &trans->transfer;
-       return map + offset;
-}
-
-static void si_texture_transfer_unmap(struct pipe_context *ctx,
-                                     struct pipe_transfer* transfer)
-{
-       struct r600_transfer *rtransfer = (struct r600_transfer*)transfer;
-       struct r600_context *rctx = (struct r600_context*)ctx;
-       struct radeon_winsys_cs_handle *buf;
-       struct pipe_resource *texture = transfer->resource;
-       struct r600_texture *rtex = (struct r600_texture*)texture;
-
-       if (rtransfer->staging) {
-               buf = si_resource(rtransfer->staging)->cs_buf;
-       } else {
-               buf = si_resource(transfer->resource)->cs_buf;
-       }
-       rctx->ws->buffer_unmap(buf);
-
-       if ((transfer->usage & PIPE_TRANSFER_WRITE) && rtransfer->staging) {
-               if (rtex->is_depth) {
-                       ctx->resource_copy_region(ctx, texture, transfer->level,
-                                                 transfer->box.x, transfer->box.y, transfer->box.z,
-                                                 &si_resource(rtransfer->staging)->b.b, transfer->level,
-                                                 &transfer->box);
-               } else {
-                       r600_copy_from_staging_texture(ctx, rtransfer);
-               }
-       }
-
-       if (rtransfer->staging)
-               pipe_resource_reference((struct pipe_resource**)&rtransfer->staging, NULL);
-
-       pipe_resource_reference(&transfer->resource, NULL);
-       FREE(transfer);
-}
-
-static const struct u_resource_vtbl r600_texture_vtbl =
-{
-       r600_texture_get_handle,        /* get_handle */
-       r600_texture_destroy,           /* resource_destroy */
-       si_texture_transfer_map,        /* transfer_map */
-       u_default_transfer_flush_region,/* transfer_flush_region */
-       si_texture_transfer_unmap,      /* transfer_unmap */
-       NULL    /* transfer_inline_write */
-};
+static const struct u_resource_vtbl r600_texture_vtbl;
 
 DEBUG_GET_ONCE_BOOL_OPTION(print_texdepth, "RADEON_PRINT_TEXDEPTH", FALSE);
 
@@ -536,11 +397,8 @@ static void r600_texture_allocate_cmask(struct r600_screen *rscreen,
 static struct r600_texture *
 r600_texture_create_object(struct pipe_screen *screen,
                           const struct pipe_resource *base,
-                          unsigned array_mode,
                           unsigned pitch_in_bytes_override,
-                          unsigned max_buffer_size,
                           struct pb_buffer *buf,
-                          boolean alloc_bo,
                           struct radeon_surface *surface)
 {
        struct r600_texture *rtex;
@@ -558,13 +416,12 @@ r600_texture_create_object(struct pipe_screen *screen,
        pipe_reference_init(&resource->b.b.reference, 1);
        resource->b.b.screen = screen;
        rtex->pitch_override = pitch_in_bytes_override;
-       rtex->real_format = base->format;
 
        /* don't include stencil-only formats which we don't support for rendering */
        rtex->is_depth = util_format_has_depth(util_format_description(rtex->resource.b.b.format));
 
        rtex->surface = *surface;
-       r = r600_setup_surface(screen, rtex, array_mode, pitch_in_bytes_override);
+       r = r600_setup_surface(screen, rtex, pitch_in_bytes_override);
        if (r) {
                FREE(rtex);
                return NULL;
@@ -582,10 +439,9 @@ r600_texture_create_object(struct pipe_screen *screen,
        }
 
        /* Now create the backing buffer. */
-       if (!buf && alloc_bo) {
+       if (!buf) {
                unsigned base_align = rtex->surface.bo_alignment;
 
-               base_align = rtex->surface.bo_alignment;
                if (!si_init_resource(rscreen, resource, rtex->size, base_align, FALSE, base->usage)) {
                        FREE(rtex);
                        return NULL;
@@ -660,15 +516,20 @@ struct pipe_resource *si_texture_create(struct pipe_screen *screen,
 
        if (!(templ->flags & R600_RESOURCE_FLAG_TRANSFER) &&
            !(templ->bind & PIPE_BIND_SCANOUT)) {
-               if (templ->flags & R600_RESOURCE_FLAG_FORCE_TILING) {
+               if (templ->flags & R600_RESOURCE_FLAG_FORCE_TILING ||
+                   templ->nr_samples > 1) {
                        array_mode = V_009910_ARRAY_2D_TILED_THIN1;
                } else if (util_format_is_compressed(templ->format)) {
                        array_mode = V_009910_ARRAY_1D_TILED_THIN1;
+               } else if (templ->usage != PIPE_USAGE_STAGING &&
+                          templ->usage != PIPE_USAGE_STREAM &&
+                          templ->target != PIPE_TEXTURE_1D &&
+                          templ->target != PIPE_TEXTURE_1D_ARRAY &&
+                          templ->height0 > 3 &&
+                          rscreen->chip_class < CIK /* XXX fix me */) {
+                       array_mode = V_009910_ARRAY_2D_TILED_THIN1;
                } else {
-                       if (rscreen->chip_class >= CIK)
-                               array_mode = V_009910_ARRAY_1D_TILED_THIN1; /* XXX fix me */
-                       else
-                               array_mode = V_009910_ARRAY_2D_TILED_THIN1;
+                       array_mode = V_009910_ARRAY_1D_TILED_THIN1;
                }
        }
 
@@ -681,8 +542,8 @@ struct pipe_resource *si_texture_create(struct pipe_screen *screen,
        if (r) {
                return NULL;
        }
-       return (struct pipe_resource *)r600_texture_create_object(screen, templ, array_mode,
-                                                                 0, 0, NULL, TRUE, &surface);
+       return (struct pipe_resource *)r600_texture_create_object(screen, templ,
+                                                                 0, NULL, &surface);
 }
 
 static struct pipe_surface *r600_create_surface(struct pipe_context *pipe,
@@ -728,7 +589,7 @@ struct pipe_resource *si_texture_from_handle(struct pipe_screen *screen,
        struct r600_screen *rscreen = (struct r600_screen*)screen;
        struct pb_buffer *buf = NULL;
        unsigned stride = 0;
-       unsigned array_mode = V_009910_ARRAY_LINEAR_ALIGNED;
+       unsigned array_mode;
        enum radeon_bo_layout micro, macro;
        struct radeon_surface surface;
        int r;
@@ -759,10 +620,12 @@ struct pipe_resource *si_texture_from_handle(struct pipe_screen *screen,
        if (r) {
                return NULL;
        }
+
        /* always set the scanout flags */
        surface.flags |= RADEON_SURF_SCANOUT;
-       return (struct pipe_resource *)r600_texture_create_object(screen, templ, array_mode,
-                                                                 stride, 0, buf, FALSE, &surface);
+
+       return (struct pipe_resource *)r600_texture_create_object(screen, templ,
+                                                                 stride, buf, &surface);
 }
 
 bool r600_init_flushed_depth_texture(struct pipe_context *ctx,
@@ -791,8 +654,6 @@ bool r600_init_flushed_depth_texture(struct pipe_context *ctx,
 
        if (staging)
                resource.flags |= R600_RESOURCE_FLAG_TRANSFER;
-       else
-               rtex->dirty_level_mask = (1 << (resource.last_level+1)) - 1;
 
        *flushed_depth_texture = (struct r600_texture *)ctx->screen->resource_create(ctx->screen, &resource);
        if (*flushed_depth_texture == NULL) {
@@ -804,8 +665,234 @@ bool r600_init_flushed_depth_texture(struct pipe_context *ctx,
        return true;
 }
 
+/**
+ * Initialize the pipe_resource descriptor to be of the same size as the box,
+ * which is supposed to hold a subregion of the texture "orig" at the given
+ * mipmap level.
+ */
+static void r600_init_temp_resource_from_box(struct pipe_resource *res,
+                                            struct pipe_resource *orig,
+                                            const struct pipe_box *box,
+                                            unsigned level, unsigned flags)
+{
+       memset(res, 0, sizeof(*res));
+       res->format = orig->format;
+       res->width0 = box->width;
+       res->height0 = box->height;
+       res->depth0 = 1;
+       res->array_size = 1;
+       res->usage = flags & R600_RESOURCE_FLAG_TRANSFER ? PIPE_USAGE_STAGING : PIPE_USAGE_STATIC;
+       res->flags = flags;
+
+       /* We must set the correct texture target and dimensions for a 3D box. */
+       if (box->depth > 1 && util_max_layer(orig, level) > 0)
+               res->target = orig->target;
+       else
+               res->target = PIPE_TEXTURE_2D;
+
+       switch (res->target) {
+       case PIPE_TEXTURE_1D_ARRAY:
+       case PIPE_TEXTURE_2D_ARRAY:
+       case PIPE_TEXTURE_CUBE_ARRAY:
+               res->array_size = box->depth;
+               break;
+       case PIPE_TEXTURE_3D:
+               res->depth0 = box->depth;
+               break;
+       default:;
+       }
+}
+
+static void *si_texture_transfer_map(struct pipe_context *ctx,
+                                    struct pipe_resource *texture,
+                                    unsigned level,
+                                    unsigned usage,
+                                    const struct pipe_box *box,
+                                    struct pipe_transfer **ptransfer)
+{
+       struct r600_context *rctx = (struct r600_context *)ctx;
+       struct r600_texture *rtex = (struct r600_texture*)texture;
+       struct r600_transfer *trans;
+       boolean use_staging_texture = FALSE;
+       struct radeon_winsys_cs_handle *buf;
+       unsigned offset = 0;
+       char *map;
+
+       /* We cannot map a tiled texture directly because the data is
+        * in a different order, therefore we do detiling using a blit.
+        *
+        * Also, use a temporary in GTT memory for read transfers, as
+        * the CPU is much happier reading out of cached system memory
+        * than uncached VRAM.
+        */
+       if (rtex->surface.level[level].mode != RADEON_SURF_MODE_LINEAR_ALIGNED &&
+           rtex->surface.level[level].mode != RADEON_SURF_MODE_LINEAR)
+               use_staging_texture = TRUE;
+
+       /* Use a staging texture for uploads if the underlying BO is busy. */
+       if (!(usage & PIPE_TRANSFER_READ) &&
+           (rctx->ws->cs_is_buffer_referenced(rctx->cs, rtex->resource.cs_buf, RADEON_USAGE_READWRITE) ||
+            rctx->ws->buffer_is_busy(rtex->resource.buf, RADEON_USAGE_READWRITE))) {
+               use_staging_texture = TRUE;
+       }
+
+       if (texture->flags & R600_RESOURCE_FLAG_TRANSFER) {
+               use_staging_texture = FALSE;
+       }
+
+       if (use_staging_texture && (usage & PIPE_TRANSFER_MAP_DIRECTLY)) {
+               return NULL;
+       }
+
+       trans = CALLOC_STRUCT(r600_transfer);
+       if (trans == NULL)
+               return NULL;
+       trans->transfer.resource = texture;
+       trans->transfer.level = level;
+       trans->transfer.usage = usage;
+       trans->transfer.box = *box;
+
+       if (rtex->is_depth) {
+               struct r600_texture *staging_depth;
+
+               if (rtex->resource.b.b.nr_samples > 1) {
+                       /* MSAA depth buffers need to be converted to single sample buffers.
+                        *
+                        * Mapping MSAA depth buffers can occur if ReadPixels is called
+                        * with a multisample GLX visual.
+                        *
+                        * First downsample the depth buffer to a temporary texture,
+                        * then decompress the temporary one to staging.
+                        *
+                        * Only the region being mapped is transfered.
+                        */
+                       struct pipe_resource resource;
+
+                       r600_init_temp_resource_from_box(&resource, texture, box, level, 0);
+
+                       if (!r600_init_flushed_depth_texture(ctx, &resource, &staging_depth)) {
+                               R600_ERR("failed to create temporary texture to hold untiled copy\n");
+                               FREE(trans);
+                               return NULL;
+                       }
+
+                       if (usage & PIPE_TRANSFER_READ) {
+                               struct pipe_resource *temp = ctx->screen->resource_create(ctx->screen, &resource);
+
+                               r600_copy_region_with_blit(ctx, temp, 0, 0, 0, 0, texture, level, box);
+                               r600_blit_decompress_depth(ctx, (struct r600_texture*)temp, staging_depth,
+                                                          0, 0, 0, box->depth, 0, 0);
+                               pipe_resource_reference((struct pipe_resource**)&temp, NULL);
+                       }
+               }
+               else {
+                       /* XXX: only readback the rectangle which is being mapped? */
+                       /* XXX: when discard is true, no need to read back from depth texture */
+                       if (!r600_init_flushed_depth_texture(ctx, texture, &staging_depth)) {
+                               R600_ERR("failed to create temporary texture to hold untiled copy\n");
+                               FREE(trans);
+                               return NULL;
+                       }
+
+                       r600_blit_decompress_depth(ctx, rtex, staging_depth,
+                                                  level, level,
+                                                  box->z, box->z + box->depth - 1,
+                                                  0, 0);
+
+                       offset = r600_texture_get_offset(staging_depth, level, box);
+               }
+
+               trans->transfer.stride = staging_depth->surface.level[level].pitch_bytes;
+               trans->transfer.layer_stride = staging_depth->surface.level[level].slice_size;
+               trans->staging = (struct si_resource*)staging_depth;
+       } else if (use_staging_texture) {
+               struct pipe_resource resource;
+               struct r600_texture *staging;
+
+               r600_init_temp_resource_from_box(&resource, texture, box, level,
+                                                R600_RESOURCE_FLAG_TRANSFER);
+
+               /* Create the temporary texture. */
+               staging = (struct r600_texture*)ctx->screen->resource_create(ctx->screen, &resource);
+               if (staging == NULL) {
+                       R600_ERR("failed to create temporary texture to hold untiled copy\n");
+                       FREE(trans);
+                       return NULL;
+               }
+               trans->staging = &staging->resource;
+               trans->transfer.stride = staging->surface.level[0].pitch_bytes;
+               trans->transfer.layer_stride = staging->surface.level[0].slice_size;
+               if (usage & PIPE_TRANSFER_READ) {
+                       r600_copy_to_staging_texture(ctx, trans);
+               }
+       } else {
+               /* the resource is mapped directly */
+               trans->transfer.stride = rtex->surface.level[level].pitch_bytes;
+               trans->transfer.layer_stride = rtex->surface.level[level].slice_size;
+               offset = r600_texture_get_offset(rtex, level, box);
+       }
+
+       if (trans->staging) {
+               buf = trans->staging->cs_buf;
+       } else {
+               buf = rtex->resource.cs_buf;
+       }
+
+       if (!(map = rctx->ws->buffer_map(buf, rctx->cs, usage))) {
+               pipe_resource_reference((struct pipe_resource**)&trans->staging, NULL);
+               FREE(trans);
+               return NULL;
+       }
+
+       *ptransfer = &trans->transfer;
+       return map + offset;
+}
+
+static void si_texture_transfer_unmap(struct pipe_context *ctx,
+                                     struct pipe_transfer* transfer)
+{
+       struct r600_transfer *rtransfer = (struct r600_transfer*)transfer;
+       struct r600_context *rctx = (struct r600_context*)ctx;
+       struct radeon_winsys_cs_handle *buf;
+       struct pipe_resource *texture = transfer->resource;
+       struct r600_texture *rtex = (struct r600_texture*)texture;
+
+       if (rtransfer->staging) {
+               buf = rtransfer->staging->cs_buf;
+       } else {
+               buf = si_resource(transfer->resource)->cs_buf;
+       }
+       rctx->ws->buffer_unmap(buf);
+
+       if ((transfer->usage & PIPE_TRANSFER_WRITE) && rtransfer->staging) {
+               if (rtex->is_depth && rtex->resource.b.b.nr_samples <= 1) {
+                       ctx->resource_copy_region(ctx, texture, transfer->level,
+                                                 transfer->box.x, transfer->box.y, transfer->box.z,
+                                                 &rtransfer->staging->b.b, transfer->level,
+                                                 &transfer->box);
+               } else {
+                       r600_copy_from_staging_texture(ctx, rtransfer);
+               }
+       }
+
+       if (rtransfer->staging)
+               pipe_resource_reference((struct pipe_resource**)&rtransfer->staging, NULL);
+
+       FREE(transfer);
+}
+
 void si_init_surface_functions(struct r600_context *r600)
 {
        r600->context.create_surface = r600_create_surface;
        r600->context.surface_destroy = r600_surface_destroy;
 }
+
+static const struct u_resource_vtbl r600_texture_vtbl =
+{
+       r600_texture_get_handle,        /* get_handle */
+       r600_texture_destroy,           /* resource_destroy */
+       si_texture_transfer_map,        /* transfer_map */
+       u_default_transfer_flush_region,/* transfer_flush_region */
+       si_texture_transfer_unmap,      /* transfer_unmap */
+       NULL    /* transfer_inline_write */
+};
index e200d4f143ebfac5fb3176d0ad1e669cf3488c22..6424d032751edee2ebfdf9422e341775b1aafcc0 100644 (file)
@@ -185,9 +185,11 @@ static void r600_destroy_context(struct pipe_context *context)
        if (rctx->dummy_pixel_shader) {
                rctx->context.delete_fs_state(&rctx->context, rctx->dummy_pixel_shader);
        }
-       rctx->context.delete_depth_stencil_alpha_state(&rctx->context, rctx->custom_dsa_flush_depth_stencil);
-       rctx->context.delete_depth_stencil_alpha_state(&rctx->context, rctx->custom_dsa_flush_depth);
-       rctx->context.delete_depth_stencil_alpha_state(&rctx->context, rctx->custom_dsa_flush_stencil);
+       for (int i = 0; i < 8; i++) {
+               rctx->context.delete_depth_stencil_alpha_state(&rctx->context, rctx->custom_dsa_flush_depth_stencil[i]);
+               rctx->context.delete_depth_stencil_alpha_state(&rctx->context, rctx->custom_dsa_flush_depth[i]);
+               rctx->context.delete_depth_stencil_alpha_state(&rctx->context, rctx->custom_dsa_flush_stencil[i]);
+       }
        rctx->context.delete_depth_stencil_alpha_state(&rctx->context, rctx->custom_dsa_flush_inplace);
        rctx->context.delete_blend_state(&rctx->context, rctx->custom_blend_resolve);
        rctx->context.delete_blend_state(&rctx->context, rctx->custom_blend_decompress);
index 27913ed47c4d9fd2ac7a731b61fe9cb32858e7d0..e1c6f30b3d754429cd25052e92f59ac33939f89d 100644 (file)
@@ -138,9 +138,9 @@ struct r600_context {
        struct blitter_context          *blitter;
        enum radeon_family              family;
        enum chip_class                 chip_class;
-       void                            *custom_dsa_flush_depth_stencil;
-       void                            *custom_dsa_flush_depth;
-       void                            *custom_dsa_flush_stencil;
+       void                            *custom_dsa_flush_depth_stencil[8];
+       void                            *custom_dsa_flush_depth[8];
+       void                            *custom_dsa_flush_stencil[8];
        void                            *custom_dsa_flush_inplace;
        void                            *custom_blend_resolve;
        void                            *custom_blend_decompress;
@@ -227,11 +227,12 @@ struct r600_context {
 
 /* r600_blit.c */
 void si_init_blit_functions(struct r600_context *rctx);
-void si_blit_uncompress_depth(struct pipe_context *ctx,
+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_layer, unsigned last_layer,
+               unsigned first_sample, unsigned last_sample);
 void si_flush_depth_textures(struct r600_context *rctx,
                             struct r600_textures_info *textures);
 void r600_decompress_color_textures(struct r600_context *rctx,
index 77a4339fb379ca23e45832d9c94127443d6600fa..9e4550fc9403f4a849eb796b6687fd98dc74bb90 100644 (file)
@@ -788,7 +788,7 @@ static void si_delete_dsa_state(struct pipe_context *ctx, void *state)
 }
 
 static void *si_create_db_flush_dsa(struct r600_context *rctx, bool copy_depth,
-                                   bool copy_stencil)
+                                   bool copy_stencil, int sample)
 {
        struct pipe_depth_stencil_alpha_state dsa;
         struct si_state_dsa *state;
@@ -800,7 +800,8 @@ static void *si_create_db_flush_dsa(struct r600_context *rctx, bool copy_depth,
                si_pm4_set_reg(&state->pm4, R_028000_DB_RENDER_CONTROL,
                               S_028000_DEPTH_COPY(copy_depth) |
                               S_028000_STENCIL_COPY(copy_stencil) |
-                              S_028000_COPY_CENTROID(1));
+                              S_028000_COPY_CENTROID(1) |
+                              S_028000_COPY_SAMPLE(sample));
        } else {
                si_pm4_set_reg(&state->pm4, R_028000_DB_RENDER_CONTROL,
                               S_028000_DEPTH_COMPRESS_DISABLE(1) |
@@ -1942,10 +1943,10 @@ static void si_db(struct r600_context *rctx, struct si_pm4_state *pm4,
        level = surf->base.u.tex.level;
        rtex = (struct r600_texture*)surf->base.texture;
 
-       format = si_translate_dbformat(rtex->real_format);
+       format = si_translate_dbformat(rtex->resource.b.b.format);
 
        if (format == V_028040_Z_INVALID) {
-               R600_ERR("Invalid DB format: %d, disabling DB.\n", rtex->real_format);
+               R600_ERR("Invalid DB format: %d, disabling DB.\n", rtex->resource.b.b.format);
        }
        assert(format != V_028040_Z_INVALID);
 
@@ -3182,6 +3183,8 @@ static void *si_create_blend_custom(struct r600_context *rctx, unsigned mode)
 
 void si_init_state_functions(struct r600_context *rctx)
 {
+       int i;
+
        rctx->context.create_blend_state = si_create_blend_state;
        rctx->context.bind_blend_state = si_bind_blend_state;
        rctx->context.delete_blend_state = si_delete_blend_state;
@@ -3195,10 +3198,12 @@ void si_init_state_functions(struct r600_context *rctx)
        rctx->context.bind_depth_stencil_alpha_state = si_bind_dsa_state;
        rctx->context.delete_depth_stencil_alpha_state = si_delete_dsa_state;
 
-       rctx->custom_dsa_flush_depth_stencil = si_create_db_flush_dsa(rctx, true, true);
-       rctx->custom_dsa_flush_depth = si_create_db_flush_dsa(rctx, true, false);
-       rctx->custom_dsa_flush_stencil = si_create_db_flush_dsa(rctx, false, true);
-       rctx->custom_dsa_flush_inplace = si_create_db_flush_dsa(rctx, false, false);
+       for (i = 0; i < 8; i++) {
+               rctx->custom_dsa_flush_depth_stencil[i] = si_create_db_flush_dsa(rctx, true, true, i);
+               rctx->custom_dsa_flush_depth[i] = si_create_db_flush_dsa(rctx, true, false, i);
+               rctx->custom_dsa_flush_stencil[i] = si_create_db_flush_dsa(rctx, false, true, i);
+       }
+       rctx->custom_dsa_flush_inplace = si_create_db_flush_dsa(rctx, false, false, 0);
        rctx->custom_blend_resolve = si_create_blend_custom(rctx, V_028808_CB_RESOLVE);
        rctx->custom_blend_decompress = si_create_blend_custom(rctx, V_028808_CB_FMASK_DECOMPRESS);